Drought resilience of maize-legume agroforestry systems in ...

Drought resilience of maize-legume agroforestry systems in Malawi

By

Amber Catherine Kerr

A dissertation submitted in partial satisfaction of the

requirements for the degree of

Doctor of Philosophy

in

Energy and Resources

in the

Graduate Division

of the

University of California, Berkeley

Committee in charge:

Professor Margaret S. Torn, Chair Professor John Harte

Professor Lynn Huntsinger

Fall 2012

1

Abstract

Drought resilience of maize-legume agroforestry systems in Malawi

by

Amber Catherine Kerr

Doctor of Philosophy in Energy and Resources

University of California, Berkeley

Professor Margaret Torn, Chair

Agroforestry – the practice of growing trees and crops on the same land – has recently been suggested as a potential tool for adaptation to climate change. There are many mechanisms by which trees on farmland could help to ameliorate climatic stresses: moderation of microclimate, increased soil water-holding capacity, increased infiltration, reduced runoff, complementary use of water resources, and diversification of production risk. However, these benefits remain largely speculative, with very few controlled experiments to demonstrate them. Furthermore, little is known about how agroforestry systems themselves might be adversely affected by climatic stress, especially during the vulnerable stage of seedling establishment. This dissertation attempts to address some of these questions with regard to “fertilizer tree” agroforestry systems in Malawi. In these systems, maize is intercropped with fast-growing leguminous trees, and biomass from the trees is regularly incorporated into the soil to increase crop yields. Though fertilizer trees are demonstrably effective under normal climatic conditions in southern Africa, their drought tolerance has not been systematically examined. For this three-year project (2008-2011), a novel and low-cost rain exclusion shelter design was used to impose an artificial drought on several Gliricidia sepium and Tephrosia candida fertilizer tree systems at Makoka Agricultural Research Station (15º31’S, 35º13’E) in southern Malawi. In the first experiment, a mature Gliricidia-maize intercropping system was subjected to complete rain exclusion from maize anthesis to maize harvest during the 2009-2010 and 2010-2011 growing seasons. Monoculture maize was used as a control. The drought dramatically decreased maize yields in both years (by 61% and 39%, respectively), with approximately similar decreases in both Gliricidia and monoculture plots. This implies that Gliricidia does not protect maize from drought conditions, but neither does it exacerbate the effects of drought by competing with maize for water. Whether in the presence or absence of drought, maize yields in Gliricidia plots were higher than in monoculture plots. Gliricidia had no effect on soil moisture and had a slight cooling effect on microclimate at the end of the growing season. In the second experiment, a similar drought – compounded with the additional stress of late planting – was imposed upon newly established fertilizer tree systems of three types: Gliricidia-

2

maize intercropping, Tephrosia-maize relay intercropping, and Tephrosia improved fallows. Seedling survival, growth, and biomass production were monitored for signs of adverse effects. Results varied by species: Gliricidia growth was retarded by drought and late planting, but Gliricidia survival remained near 100%, and its biomass at the end of the growing season was not affected. By contrast, Tephrosia maintained constant height growth under these stresses, but its survival decreased, which translated into lower maize yields the following year. However, all three fertilizer tree systems demonstrated good overall drought resilience and conferred maize yield benefits even under suboptimal conditions. A modeling experiment was conducted to compliment the above field experiments. Output from global and regional climate models was applied to long-term data sets on weather and maize yield at this location. The results suggested that for the next several decades, current precipitation variability in southern Malawi will likely pose a greater risk to maize production than will long-term precipitation trends caused by global climate change. However, higher maximum temperatures have the potential to reduce maize yields within the next several decades. The field experiments described herein suggest that fertilizer tree systems can improve maize production even under conditions of drought stress. Current climate variability and future climate change should pose no obstacle to their adoption in Malawi. Although this project did not find evidence that fertilizer trees directly protect the maize crop from drought, their beneficial effects on food security and farm income may indirectly help to cushion farmers against climate variability. The performance of agroforestry systems under future climate is a topic that demands much more research, both descriptive and theoretical (including controlled experiments, simulation models, and meta-analyses). Furthermore, climate manipulation experiments in the developing world are an underutilized but potentially important tool for understanding the effect of climate change on subsistence agricultural systems and developing climate-resilient alternatives. It is hoped that the methods and outcomes of this dissertation represent a small step toward closing both of those knowledge gaps.

i

Dedication

To my grandmother, Catherine “Kay” Kerr (1911-2010).

Principled, stubborn, courageous, and tireless, she worked to preserve the health and beauty

of the San Francisco Bay for future generations. Many are grateful for her inspiring legacy, but none more so than her granddaughter.

iii

Table of contents Chapter 1. Role of agroforestry in climate adaptation .............................................................. 1

Abstract ....................................................................................................................................... 1 1.1 Reducing vulnerability of agriculture to climate change ................................................... 1

1.1.1 Current approaches to agricultural adaptation ............................................................ 2 1.1.2 Agroecological perspectives on climate resilience ..................................................... 2

1.1.2.1 Insights from traditional knowledge .................................................................... 2 1.1.2.2 Ecosystem mimicry .............................................................................................. 3

1.2 Definitions and examples of agroforestry .......................................................................... 3 1.2.1 Definition and scope ................................................................................................... 3 1.2.2 Theoretical bases of agroforestry ................................................................................ 3 1.2.3 Common elements of agroforestry systems ................................................................ 5 1.2.4 How might agroforestry play a role in climate adaptation? ........................................ 5

1.3 Climate modulation by agroforestry systems: direct effects .............................................. 5 1.3.1 Microclimate ............................................................................................................... 6

1.3.1.1 Air and soil temperature ....................................................................................... 6 1.3.1.2 Evaporative demand ............................................................................................. 6

1.3.2 Soil properties ............................................................................................................. 6 1.3.2.1 Soil water-holding capacity ................................................................................. 6 1.3.2.2 Changes in infiltration and runoff ........................................................................ 6 1.3.2.3 Water use competition or complementarity ......................................................... 7 1.3.2.4 Hydraulic lift ........................................................................................................ 7 1.3.2.5 Soil nutrient status................................................................................................ 7

1.3.3 Physical protection against storm damage .................................................................. 8 1.4 Climate modulation by agroforestry systems: indirect effects .......................................... 8

1.4.1 Biodiversity conservation ........................................................................................... 8 1.4.2 Weeds, pests and diseases ........................................................................................... 9 1.4.3 Risk management by diversification of production .................................................... 9 1.4.4 Enhancing farm productivity and income ................................................................... 9

1.5 Impacts of climate change on agroforestry systems ........................................................ 10 1.5.1 Precipitation .............................................................................................................. 10 1.5.2 Temperature .............................................................................................................. 10 1.5.3 CO2 fertilization ........................................................................................................ 11 1.5.4 Extreme events and natural disasters ........................................................................ 11

1.6 Synergies and conflicts between adaptation and mitigation ............................................ 12 1.6.1 Previous work on climate mitigation aspects of agroforestry ................................... 12 1.6.2 Potential synergies .................................................................................................... 12 1.6.3 Potential conflicts ...................................................................................................... 13

1.7 Need for future work ........................................................................................................ 14 1.7.1 Experimental climate manipulations in the field ...................................................... 14 1.7.2 Simulation modeling ................................................................................................. 14 1.7.3 Meta-analyses of agroforestry climate response across space and time ................... 15

1.8 Conclusions ...................................................................................................................... 15 1.8.1 Acknowledgements ................................................................................................... 15

iv

Chapter 2. Introduction to Malawi field experiments ............................................................. 17 Abstract ..................................................................................................................................... 17 2.1 Motivations and goals of this project ............................................................................... 17

2.1.1 Protecting food security in the developing world under future climate .................... 18 2.1.1.1 Identifying climate-resilient production systems in vulnerable areas ................ 18 2.1.1.2 Meeting the need for more climate manipulation experiments ......................... 18

2.1.2 Understanding response of agroforestry systems to climate change ........................ 18 2.1.2.1 Gathering empirical evidence on climate response of agroforestry systems ..... 19 2.1.2.2 Elucidating mechanisms and interactions .......................................................... 19

2.2 Background on Malawian agriculture .............................................................................. 19 2.2.1 Current statistics ........................................................................................................ 20

2.2.1.1 Demographic and economic statistics ................................................................ 20 2.2.1.2 Agricultural statistics ......................................................................................... 21

2.2.2 History of agriculture in Malawi ............................................................................... 22 2.2.2.1 Pre-colonial introduction of maize ..................................................................... 22 2.2.2.2 The agricultural estate system ............................................................................ 22

2.2.3 Challenges to agricultural production in Malawi ...................................................... 23 2.2.3.1 Poor soil fertility and lack of access to inputs ................................................... 23 2.2.3.2 High rainfall variability ...................................................................................... 24 2.2.3.3 Population density and small farm size ............................................................. 24 2.2.3.4 Economic and social challenges ........................................................................ 24

2.2.4 Fertilizer subsidy: history, successes, costs .............................................................. 25 2.2.4.1 Starter Pack Scheme and Targeted Input Programme, 1998-99 to 2004-05 ...... 25 2.2.4.2 Farm Input Subsidy Programme (2005-06 to present) ...................................... 25

2.3 Agroforestry research in southern Africa ........................................................................ 26 2.3.1 History and goals of SADC-ICRAF ......................................................................... 26 2.3.2 Experiments on fertilizer trees .................................................................................. 26

2.3.2.1 Tree species used ............................................................................................... 26 2.3.2.2 Types of fertilizer tree systems .......................................................................... 27

2.3.3 Adoption and livelihood impacts of fertilizer trees .................................................. 29 2.4 Gaps filled by the current project ..................................................................................... 30

2.4.1 Examining crop and tree response to controlled drought ......................................... 30 2.4.2 Comparing mature trees and seedlings ..................................................................... 30

2.5 Outline of experimental methods ..................................................................................... 30 2.5.1 Makoka Agricultural Research Station ..................................................................... 30 2.5.2 Philosophy of three-year field experiment ................................................................ 31 2.5.3 Allocation of field tasks ............................................................................................ 31 2.5.4 Allocation of field tasks ............................................................................................ 32 2.5.5 Scope, implications, and limitations of this work ..................................................... 33 2.5.6 Acknowledgements ................................................................................................... 34

Chapter 3. Design, construction, and performance of rain exclusion shelters ...................... 35

Abstract ..................................................................................................................................... 35 3.1 Introduction and motivation ............................................................................................. 35

3.1.1 Intrinsic challenges of manipulating precipitation .................................................... 36 3.1.1.1 Intercepting target amount, versus achieving a realistic temporal pattern ......... 36

v

3.1.1.2 Difficulty in controlling all aspects of plant-water relations ............................. 37 3.1.1.3 Unintended consequences .................................................................................. 37

3.1.2 Existing rainout shelter designs ................................................................................ 37 3.1.2.1 Shelters with fixed or removable roofs .............................................................. 37 3.1.2.2 Fractional interception with slats or gutters ....................................................... 38 3.1.2.3 Below-canopy interception of throughfall ......................................................... 38 3.1.2.4 Watering experiments ........................................................................................ 38

3.1.3 Goals of this project .................................................................................................. 38 3.2 Methods............................................................................................................................ 39

3.2.1 Design of rain shelters .............................................................................................. 39 3.2.2 Evaluating rain shelter performance ......................................................................... 44

3.2.2.1 Rainfall interception ........................................................................................... 44 3.2.2.2 Soil moisture ...................................................................................................... 44 3.2.2.3 Leaf water content .............................................................................................. 45 3.2.2.4 Litter decomposition rates .................................................................................. 45 3.2.2.5 Light interception ............................................................................................... 45 3.2.2.6 Pan evaporation .................................................................................................. 46 3.2.2.7 Soil temperature ................................................................................................. 46 3.2.2.8 Air temperature .................................................................................................. 46

3.3 Performance of rain shelters ............................................................................................ 46 3.3.1 Intended effects ......................................................................................................... 46

3.3.1.1 Total rainfall ....................................................................................................... 46 3.3.1.2 Litter decomposition .......................................................................................... 47 3.3.1.3 Gliricidia leaf water content ............................................................................... 47

3.3.2 Unintended effects .................................................................................................... 48 3.3.2.1 Air temperature .................................................................................................. 48 3.3.2.2 Pan evaporation .................................................................................................. 48 3.3.2.3 Soil temperature ................................................................................................. 48 3.3.2.4 Incident radiation: significant effect .................................................................. 49 3.3.2.5 Dust on leaves .................................................................................................... 49

3.3.3 Durability .................................................................................................................. 50 3.3.3.1 Roof durability under normal conditions ........................................................... 50 3.3.3.2 Violent storms: weathered successfully ............................................................. 50 3.3.3.3 Termites: a major problem ................................................................................. 50

3.4 Discussion ........................................................................................................................ 51 3.4.1 Possible improvements to existing design ................................................................ 51

3.4.1.1 Incident radiation ............................................................................................... 51 3.4.1.2 Surface and belowground water flow ................................................................ 51 3.4.1.3 Attack by termites .............................................................................................. 52 3.4.1.4 Insufficient roof height for trees ........................................................................ 52 3.4.1.5 Control for confounding effects by watering under shelters ............................. 52

3.4.2 Future applications .................................................................................................... 53 3.4.2.1 Other crops and other locations ......................................................................... 53 3.4.2.2 Introduce heating or other variables .................................................................. 53


vi

Chapter 4. Effects of drought on an established Gliricidia-maize intercropping system ..... 55 Abstract ..................................................................................................................................... 55 4.1 Introduction and motivation ............................................................................................. 55

4.1.1 Fertilizer trees and climatic stress ............................................................................. 56 4.1.2 Previous work at this field site .................................................................................. 56

4.1.2.1 Maize yield and soil nutrient dynamics ............................................................. 57 4.1.2.2 Rooting depth and water use patterns ................................................................ 57 4.1.2.3 Soil organic matter ............................................................................................. 57

4.1.3 Goals and hypotheses of this experiment .................................................................. 58 4.1.3.1 Effect of drought on maize yield in monoculture and Gliricidia intercropping systems 58 4.1.3.2 Interaction of inorganic fertilizer with drought and cropping system ............... 58 4.1.3.3 Influence of Gliricidia on microclimate and soil moisture ................................ 58 4.1.3.4 Maize yield variability of monoculture and agroforestry plots across different years 59

4.2 Methods............................................................................................................................ 60 4.2.1 Modification of existing experimental layout ........................................................... 62 4.2.2 Rain manipulations ................................................................................................... 63 4.2.3 Land preparation and Gliricidia pruning ................................................................... 64

4.2.3.1 Land preparation ................................................................................................ 64 4.2.3.2 Pruning and weighing of Gliricidia biomass ..................................................... 64

4.2.4 Maize cultivation and harvest ................................................................................... 65 4.2.4.1 Planting and thinning ......................................................................................... 65 4.2.4.2 Weeding and fertilizing ...................................................................................... 65 4.2.4.3 Harvest ............................................................................................................... 66

4.2.5 Microclimatic and soil measurements ...................................................................... 67 4.2.6 Data analysis ............................................................................................................. 67

4.3 Results .............................................................................................................................. 67 4.3.1 Gliricidia biomass production ................................................................................... 68

4.3.1.1 Total Gliricidia biomass ..................................................................................... 68 4.3.2 Maize performance ................................................................................................... 69

4.3.2.1 Total grain production ........................................................................................ 69 4.3.2.2 Maize reproductive characteristics .................................................................... 72

4.3.3 Soil and microclimate ............................................................................................... 74 4.3.3.1 Soil moisture ...................................................................................................... 74 4.3.3.2 Air temperature .................................................................................................. 74 4.3.3.3 Soil temperature ................................................................................................. 76 4.3.3.4 Pan evaporation .................................................................................................. 76

4.3.4 Historical maize yield, 1993-2006 ............................................................................ 76 4.4 Discussion ........................................................................................................................ 78

4.4.1 Review of hypotheses ............................................................................................... 78 4.4.2 Implications for use of Gliricidia in drought conditions ........................................... 79

4.4.2.1 Gliricidia can convey yield benefits even in drought conditions ....................... 79 4.4.2.2 Gliricidia may help to moderate soil temperature and air temperature ............. 79 4.4.2.3 There is not enough evidence to promote Gliricidia for drought protection ..... 80

4.4.3 Unanswered questions and future work .................................................................... 80

vii

4.4.3.1 Effect of tree species, location and soil type ...................................................... 80 4.4.3.2 Effect of drought timing and magnitude ............................................................ 81 4.4.3.3 Interactions of drought, temperature, and other global changes ........................ 81 4.4.3.4 Comparison of Gliricidia to other organic inputs .............................................. 81


Chapter 5. Effects of drought and planting time on establishment of Gliricidia and Tephrosia seedlings ..................................................................................................................... 83

5.1 Introduction and motivation ............................................................................................. 83 5.1.1 Previous work on seedling establishment in fertilizer tree systems ......................... 84

5.1.1.1 General principles of fertilizer tree establishment ............................................. 84 5.1.1.2 Studies on fertilizer tree seedlings ..................................................................... 85 5.1.1.3 Knowledge gaps for fertilizer tree seedlings ..................................................... 86

5.1.2 Goals and hypotheses of this experiment .................................................................. 87 5.1.2.1 Comparing drought response of young versus mature Gliricidia ...................... 87 5.1.2.2 Comparing drought response of Gliricidia versus Tephrosia ............................ 87 5.1.2.3 Comparing Tephrosia relay intercropping versus improved fallows ................. 87 5.1.2.4 Comparing trees planted on time versus trees planted late ................................ 88 5.1.2.5 Assessing fertilizer tree performance under optimal conditions ........................ 88

5.2 Methods............................................................................................................................ 88 5.2.1 Field establishment and experimental layout ............................................................ 88

5.2.1.1 History and background of field site .................................................................. 88 5.2.1.2 Initial layout of the experiment .......................................................................... 89 5.2.1.3 Subsequent modifications in layout ................................................................... 92

5.2.2 Seedling establishment .............................................................................................. 92 5.2.2.1 Gliricidia ............................................................................................................ 92 5.2.2.2 Tephrosia ............................................................................................................ 93

5.2.3 Field and seedling management ................................................................................ 94 5.2.3.1 General management ......................................................................................... 94 5.2.3.2 Gliricidia intercropping ...................................................................................... 94 5.2.3.3 Tephrosia relay intercropping ............................................................................ 94 5.2.3.4 Tephrosia improved fallow ................................................................................ 96 5.2.3.5 Key management dates ...................................................................................... 96

5.2.4 Rain manipulation, soil moisture, and microclimate ................................................ 97 5.2.4.1 Rain manipulation schedule ............................................................................... 97 5.2.4.2 Soil moisture measurements .............................................................................. 97 5.2.4.3 Measurements of microclimate .......................................................................... 97

5.2.5 Seedling survival, height, and allometry ................................................................... 98 5.2.5.1 Seedling presence-absence ................................................................................. 98 5.2.5.2 Seedling height ................................................................................................... 98 5.2.5.3 Seedling allometry ............................................................................................. 99

5.2.6 Data analysis ............................................................................................................. 99 5.3 Results ............................................................................................................................ 100

5.3.1 Effectiveness of rain shelters .................................................................................. 100 5.3.2 Effect of optimally managed Gliricidia and Tephrosia systems on maize yield .... 101

viii

5.3.3 Effects of drought and late tree planting on maize yield ........................................ 102 5.3.3.1 Assumptions about causation ........................................................................... 102 5.3.3.2 Results .............................................................................................................. 102

5.3.4 Effects of drought and late tree planting on tree biomass production .................... 104 5.3.4.1 Production of leaves and wood ........................................................................ 105 5.3.4.2 Production of litterfall and reproductive biomass in Tephrosia ....................... 107 5.3.4.3 Connecting tree biomass production to maize yield ........................................ 108

5.3.5 Effects on seedling germination, survival, height and phenology .......................... 110 5.3.5.1 Germination of Tephrosia seeds ...................................................................... 110 5.3.5.2 Seedling survival .............................................................................................. 110 5.3.5.3 Seedling height ................................................................................................. 115 5.3.5.4 Summary of data on individual seedlings ........................................................ 117

5.3.6 Allometry ................................................................................................................ 117 5.3.6.1 Model development ......................................................................................... 117 5.3.6.2 Comparison of measured and modeled plot-level biomass ............................. 118

5.3.7 Microclimatic effects .............................................................................................. 119 5.3.7.1 Average air temperature ................................................................................... 120 5.3.7.2 Air temperature in excess of damage thresholds ............................................. 120

5.4 Discussion ...................................................................................................................... 121 5.4.1 Hypotheses revisited ............................................................................................... 121 5.4.2 Future work needed ................................................................................................. 123

5.5 Conclusions .................................................................................................................... 124 5.5.1 Acknowledgements ................................................................................................. 125

Chapter 6. Current and future climate in southern Malawi: implications for maize production .................................................................................................................................. 127

Abstract ................................................................................................................................... 127 6.1 Introduction .................................................................................................................... 127

6.1.1 Southern Africa’s vulnerability ............................................................................... 127 6.1.2 Climate change in southern Africa ......................................................................... 128 6.1.3 Climatic influences on maize in southern Africa .................................................... 128

6.1.3.1 Precipitation ..................................................................................................... 129 6.1.3.2 Temperature ..................................................................................................... 129

6.1.4 Proposed adaptation measures ................................................................................ 130 6.1.5 A case study in southern Malawi ............................................................................ 130

6.2 Methods.......................................................................................................................... 131 6.2.1 Analysis of historical climate .................................................................................. 131

6.2.1.1 Quality control of data ..................................................................................... 131 6.2.1.2 Analysis of historical temperature ................................................................... 132 6.2.1.3 Analysis of historical rainfall ........................................................................... 132 6.2.1.4 Calculation of growing degree-days ................................................................ 133 6.2.1.5 Conversion of daily temperature to hourly temperature .................................. 134

6.2.2 Analysis of future climate ....................................................................................... 136 6.2.2.1 Selection of GCMs and time periods ............................................................... 136 6.2.2.2 Application of predicted changes to historical data ......................................... 136 6.2.2.3 MarkSim daily weather generator .................................................................... 137

ix

6.2.3 Evaluating implications for maize production ........................................................ 138 6.2.3.1 Effects of rainfall changes on maize growth .................................................... 138 6.2.3.2 Effects of temperature changes on maize growth ............................................ 139

6.3 Results ............................................................................................................................ 140 6.3.1 Historical trends over time ...................................................................................... 140

6.3.1.1 Temperature trends .......................................................................................... 140 6.3.1.2 Rainfall trends .................................................................................................. 140

6.3.2 Accuracy of MarkSim baseline climate .................................................................. 143 6.3.2.1 Daily temperature ............................................................................................. 143 6.3.2.2 Daily rainfall .................................................................................................... 143

6.3.3 Simulations of future climate .................................................................................. 144 6.3.3.1 Application of predicted changes to historical data ......................................... 145 6.3.3.2 Output from MarkSim using BCCR:BCM2 .................................................... 146

6.3.4 Impacts on maize production .................................................................................. 146 6.3.4.1 Rainfall ............................................................................................................. 146 6.3.4.2 Temperature ..................................................................................................... 147

6.4 Discussion ...................................................................................................................... 148 6.4.1 Comments on methods ............................................................................................ 148 6.4.2 Comments on results ............................................................................................... 149 6.4.3 Drawbacks of the present study .............................................................................. 150 6.4.4 Future work ............................................................................................................. 150

6.5 Conclusions .................................................................................................................... 150 Chapter 7. Conclusions and policy recommendations .......................................................... 153

Abstract ................................................................................................................................... 153 7.1 Summary of findings from field experiments ................................................................ 153

7.1.1 Low-cost rain manipulation experiments can inform climate adaptation ............... 153 7.1.1.1 Simple rain shelter designs can aid agricultural research in the developing world 154 7.1.1.2 Site-specific and species-specific information is valuable for adaptation planning 154

7.1.2 Mature Gliricidia-maize intercropping systems perform well under drought ........ 154 7.1.2.1 Complementarity of water use during drought makes Gliricidia a “no-regrets” option 154 7.1.2.2 Gliricidia may confer modest drought protection under some circumstances . 154

7.1.3 Fertilizer tree seedlings are more resilient to drought than to late planting ............ 154 7.1.3.1 Gliricidia seedlings are highly resistant to environmental stresses ................. 154 7.1.3.2 Tephrosia seedlings suffer from late planting but partly catch up during dry season 155 7.1.3.3 Young fertilizer tree systems benefit maize yield even in suboptimal conditions 155

7.1.4 Existing climate variability may matter more than future climate change ............. 155 7.1.4.1 Predicted rainfall changes in Malawi fall within scope of current variability . 155 7.1.4.2 Temperature increase may threaten maize production more than changes in rainfall 155

x

7.1.4.3 Climate and crop yield predictions from GCMs should be calibrated to local baselines 155

7.2 Additional observations on fertilizer tree systems in Malawi ....................................... 156 7.2.1 Labor requirements pose a major obstacle to Gliricidia intercropping .................. 156

7.2.1.1 Gliricidia has exacting labor requirements and interferes with other field tasks 156 7.2.1.2 Labor requirements of Gliricidia are greatest in the establishment phase ....... 156 7.2.1.3 The scope and consequences of Gliricidia labor requirements need further study 157

7.2.2 Fuelwood from fertilizer trees is less desirable than fuelwood from forests .......... 157 7.2.2.1 Quantity is no substitute for quality ................................................................. 157 7.2.2.2 Small branches are less useful than large logs ................................................. 158

7.2.3 Fertilizer trees interfere with traditional food intercrops ........................................ 158 7.2.3.1 Intercropping is already used by nearly all farmers in southern Malawi ......... 158 7.2.3.2 Food intercrops fill important needs that fertilizer trees do not ....................... 159

7.2.4 Fertilizer trees may directly contribute to biodiversity conservation ..................... 159 7.2.4.1 Agroforestry can affect biodiversity in many ways, but effects are poorly documented ..................................................................................................................... 159 7.2.4.2 Fertilizer trees may directly provide habitat for displaced miombo fauna ...... 159

7.3 Comparing fertilizer trees with other options ................................................................ 161 7.3.1 Annual legumes and grain legumes ........................................................................ 161 7.3.2 Inorganic fertilizer .................................................................................................. 161

7.4 Climate adaptation options for Malawian agriculture ................................................... 162 7.4.1 Technical approaches .............................................................................................. 162

7.4.1.1 Improved maize cultivars ................................................................................. 162 7.4.1.2 Crop diversification ......................................................................................... 162 7.4.1.3 Irrigation and micro-irrigation ......................................................................... 162 7.4.1.4 Weather forecasting ......................................................................................... 163

7.4.2 Structural approaches .............................................................................................. 163 7.4.2.1 Improving social safety nets in case of crop failure ........................................ 163 7.4.2.2 Strengthening national infrastructure ............................................................... 163 7.4.2.3 Economic diversification ................................................................................. 163 7.4.2.4 Poverty reduction ............................................................................................. 163

7.4.3 Agroforestry in context ........................................................................................... 164 7.5 Knowledge gaps and future research priorities .............................................................. 164

7.5.1 Controlled climate manipulations in tropical subsistence agriculture .................... 164 7.5.2 Systematic assessment of agroforestry’s role in climate adaptation ....................... 164

7.6 Summary and conclusions ............................................................................................. 165 References .................................................................................................................................. 167

xi

List of figures Figure 2-1. The Majoni family (Makoka, Malawi, January 2010) fertilizing a field of

intercropped maize and vegetables typical of a smallholder farm. ....................................... 22

Figure 2-2. A close-up of the field in Figure 2-1. Clockwise from lower left: maize; groundnut; pumpkin; common bean. ....................................................................................................... 22

Figure 2-3. Smallholder maize cultivation on a steep slope (Makoka, Malawi, December 2009)................................................................................................................................................ 23

Figure 2-4. Felling of trees in miombo forest for use as fuelwood (Makoka, Malawi, October 2010). .................................................................................................................................... 23

Figure 2-5. Temporal schematics (annual or multi-year) of three different types of fertilizer tree systems intercropped with maize. ......................................................................................... 28

Figure 2-6. Google Earth satellite image of Makoka Agricultural Research Station. Label (1) shows the location of MZ12 (15º31’08.90” S, 35º13’25.44” E), while label (2) shows the location of Nkula Field (15º31’13.55” S, 35º13’49.43” E). Image date is 1973, but field station layout remains similar today.. ................................................................................... 32

Figure 3-1. An aerial view of several completed rain shelters (the smaller design). .................... 40

Figure 3-2. Landscape context for the 18 shelters at the seedling experiment (the larger shelter design). .................................................................................................................................. 40

Figure 3-3. Blueprints for the smaller shelter design: (a) flat side; (b) peaked side. .................... 41

Figure 3-4. Blueprints for the larger shelter design: (a) flat side; (b) peaked side. ...................... 42

Figure 3-5. Bolts (secured underneath with wing-nuts) fastening a shelter roof. ......................... 43

Figure 3-6. Furled roof allowing throughfall. ............................................................................... 43

Figure 3-7. Trenches to divert rainwater flowing from the roofs of (a) the smaller shelters at the long-term Gliricidia trial (with field assistants); (b) the larger shelters at the seedling trial.44

Figure 3-8. Map of average rainfall reduction under shelter roof. ................................................ 46

Figure 3-9. Soil moisture differences between control and shelter plots after four weeks of complete rain exclusion. ....................................................................................................... 47

Figure 3-10. Decomposition of fresh Gliricidia biomass in litterbags after 28 days. .................. 47

Figure 3-11. Shelters had no significant effect on Gliricidia leaf moisture content. .................... 47

Figure 3-12. Difference between air temperature in ambient plots versus rain shelter plots over a 5-day period (8-13 May 2009). ............................................................................................. 48

Figure 3-13. Three 24-hour trials of pan evaporation (April-May 2009) showed no difference between control and shelter plots. ......................................................................................... 49

xii

Figure 3-14. The shelters significantly reduced PAR (on three measurements days with in May 2009 with varying sky conditions). ....................................................................................... 49

Figure 3-15. The shading effect of the shelter roof can be seen in the right half of the photo. .... 49

Figure 3-16. Roof of shelter reinforced with 2 mm wire (before installation of clear polyethylene sheets). .................................................................................................................................. 50

Figure 3-17. Vertical poles (belowground and 30 cm aboveground) were wrapped in black polyethylene to discourage termites. ..................................................................................... 51

Figure 3-18. Termite damage to rain shelters. (a) Shelter collapse due to termite attack on vertical poles. (b) Vertical pole left hanging after being eaten by termites from bottom up. 52

Figure 4-1. Close-up of a Gliricidia-maize intercropping plot (March 2010). ............................. 60

Figure 4-2. Looking across MZ12 from Block 1 (in the foreground) to Block 3. (February 2010.)............................................................................................................................................... 60

Figure 4-3. Seasonal progression of the Gliricidia-maize intercropping system at MZ12, showing (a) Gliricidia pruning at end of dry season, (b) new growth at beginning of wet season; (c) and (d) maize maturation; (e) senesced maize ready for harvest; and (f) dry-season Gliricidia fallow. ................................................................................................................... 61

Figure 4-4. Field layout of MZ12. Shaded plots are used in the current experiment. G = Gliricidia (0 = sole maize; 1 = Gliricidia intercrop); N = nitrogen fertilizer (0 = none; 0.5 = 46 kg N / ha-yr; 1 = 92 kg N / ha-yr); R = rain (0 = ambient, 1 = drought). ........................ 63

Figure 4-5. (a) Layout of a Gliricidia intercropping plot in MZ12. Layout of sole maize plots is identical except without trees. Plots are contiguous along their 6.1 m edges. (b) Key to plot diagram above. ...................................................................................................................... 63

Figure 4-6. Pruning Gliricidia shortly before maize planting (24 November 2009). The trees were previously pruned in October and will be pruned again in January. ............................ 65

Figure 4-7. Burying freshly pruned Gliricidia biomass under a ridge, on which maize will be planted (24 November 2009). ............................................................................................... 65

Figure 4-8. Harvesting maize in the field (28 April 2010). .......................................................... 67

Figure 4-9. A sampling of maize cobs of various sizes from different plots at MZ12 (22 April 2009). .................................................................................................................................... 67

Figure 4-10. Effect of drought on maize yields in monoculture and Gliricidia plots across three years: (a) 2008-2009, when the drought manipulation was not successful; (b) 2009-2010; (c) 2010-2011. Error bars are ± 1 SEM. See Table 4-5 for significance of differences. ........... 70

Figure 4-11. Gravimetric soil moisture (0-20 cm) as a function of drought treatment and cropping system. Trends are shown for (a) 2008-2009; (b) 2009-2010; (c) 2010-2011. Error bars are ± 1 SEM. .................................................................................................................. 73

xiii

Figure 4-12. Air temperature in maize monoculture plots and Gliricidia plots over several representative days at the end of the growing season (12-14 May 2009). Each line represents an average of five plots of that type. .................................................................... 75

Figure 4-13. Mid-afternoon soil temperature (0-5 cm depth) as a function of cropping type and drought treatment at two time points in March 2010. Error bars represent ± 1 SEM. ......... 75

Figure 4-14. Effect of Gliricidia on pan evaporation (in mm per 24 hours) over three days at the end of the 2009 growing season. ........................................................................................... 76

Figure 4-15. Response ratio of fertilizer as a function of growing season rainfall (based on MZ12 maize yield data, 1993-2006). (N1=92 kg/ha; N0.5=46 kg/ha). ............................................. 77

Figure 4-16. Response ratio of Gliricidia as a function of growing season rainfall (based on MZ12 maize yield data, 1993-2006). .................................................................................... 77

Figure 5-1. Layout of Nkula Field (as seen from the upper right of Figure 5-2) in April 2009. The far right rain shelter is Plot 1-6. Construction materials and a watchman’s hut are in the foreground. ............................................................................................................................ 90

Figure 5-2. Map of experimental layout at Nkula Field. .............................................................. 90

Figure 5-3. Plot maps of the three types of agroforestry systems used in this experiment. Sole maize plots were configured identically to intercropping plots but without the trees. ......... 91

Figure 5-4. Propagation of Gliricidia seedlings in a raised nursery bed (November 2009). A machete is being used to prune the roots between each seedling row. ................................. 93

Figure 5-5. Gliricidia seedlings ten days after being transplanted to the field (January 2009). ... 93

Figure 5-6. Tephrosia germinating in the field. ............................................................................ 93

Figure 5-7. Six-week-old Tephrosia relay intercrop. .................................................................... 93

Figure 5-8. Three-month-old Tephrosia improved fallow. ........................................................... 93

Figure 5-9. Gliricidia seedling resprouting from roots (at arrow) after an attempt was made to remove it at 10 months of age. (The seedlings behind it are of the same age but uncut.) .... 95

Figure 5-10. Two-year-old Gliricidia pruned to a 30-cm stump. In the background are unpruned one-year-old Gliricidia seedlings and a two-year old Tephrosia improved fallow awaiting pruning. ................................................................................................................................. 95

Figure 5-11. Collecting litterfall in a quadrat in a one-year-old Tephrosia improved fallow. ..... 95

Figure 5-12. Separating and weighing leaves and branches in a two-year-old Tephrosia improved fallow. ................................................................................................................................... 95

Figure 5-13. The author with a 22-month-old Tephrosia improved fallow. ................................. 98

Figure 5-14. Flowers on 22-month-old Gliricidia. ....................................................................... 98

Figure 5-15. Seeds on 10-month-old Tephrosia. .......................................................................... 98

xiv

Figure 5-16. Gravimetric soil moisture (0-20 cm) immediately before (28 Mar 2009) and six weeks after (11 May 2009) the 2009 drought treatment. Error bars are ± 1 SEM. The drought treatment, which excluded <4% of annual rainfall, had no effect on soil moisture.............................................................................................................................................. 100

Figure 5-17. Gravimetric soil moisture (0-5 cm), 7 May 2010. The drought treatment (Rain =1), which excluded 33% of annual rainfall, reduced soil moisture (p = 0.002) except in plot 4-5 (indicated by X). .................................................................................................................. 100

Figure 5-18. Unpruned Gliricidia, 13 months after planting, overshadowing stunted 7-week-old maize (February 2010). ....................................................................................................... 107

Figure 5-19. 10-month old Tephrosia improved fallow plot in October 2009, showing the effects of late planting (left half) and early planting (right half). A harvested maize plot is in the foreground. .......................................................................................................................... 107

Figure 5-20. Correlations between fertilizer tree biomass yield in November 2010 and maize yield the following year (in May 2011). Shown are (a) all three agroforestry systems and both rain levels; (b) plots with ambient rainfall only; (c) drought plots only; (d) Gliricidia intercropping plots only; (e) Tephrosia relay intercropping plots only; (f) Tephrosia improved fallow plots only. ................................................................................................ 109

Figure 5-21. Survivorship of Gliricidia seedlings in (a) 2008-2009 and (b) 2009-2010. The relatively low survival of late-planted seedlings in 2008-2009 is an artifact due to inconsistent planting technique. Error bars represent ±1 SEM (but are not visible on most points). ................................................................................................................................ 111

Figure 5-22. Survivorship of Tephrosia seedlings in (a) 2008-2009 and (b) 2009-2010. RI = relay intercropping; IF = improved fallow. Error bars represent ± 1 SEM. ................................ 111

Figure 5-23. Relationship between seedling survival and biomass production in Tephrosia relay intercropping plots in (a) 2008-2009 and (b) 2009-2010. Correlations are calculated separately for early- and late-planted seedlings. ................................................................. 114

Figure 5-24. Relationship between seedling survival (in 2008-2009) and biomass production (in 2009-2010) in Tephrosia improved fallow plots. Correlations are calculated separately for early- and late-planted seedlings. ........................................................................................ 114

Figure 5-25. Height trends of Gliricidia seedlings in 2009-2010: effects of (a) planting time and (b) drought. Error bars represent ± 1 SEM. (2008-2009 heights are not reported due to methodological problems with the treatments.) .................................................................. 115

Figure 5-26. Height trends of Tephrosia: (a) 2008-2009 relay intercropping; (b) 2008-2009 improved fallows; (c) 2009-2010 relay intercropping; (d) 2009-2010 improved fallows. Error bars are ± 1 SEM. ...................................................................................................... 116

Figure 5-27. Allometric models for Gliricidia and Tephrosia seedlings: (a) Gliricidia quadratic model; (b) Tephrosia exponential model; (c) Gliricidia plot-by-plot comparison of actual versus predicted biomass (2008-2009 data); (d) Tephrosia plot-by-plot comparison of actual versus predicted biomass (2008-2009 data). In figures (c) and (d), the 1:1 line is solid while the linear model fit line is dashed. ...................................................................................... 119

xv

Figure 5-28. Comparison of air temperature over a typical three-day period (4-6 April 2010) in different cropping systems at Nkula Field. Each line represents the average of two plots of that type. .............................................................................................................................. 120

Figure 6-1. Location of Makoka Agricultural Research Station in Malawi (15º31’S, 35º13’E). 131

Figure 6-2. Annual distribution of rainfall at Makoka, Malawi, 1971-2008. ............................. 131

Figure 6-3. Average maximum and minimum temperatures at Makoka, Malawi, 1971-2000. . 132

Figure 6-4. Maize wilting, Makoka, Malawi, 1/2010. ................................................................ 132

Figure 6-5. Typical fit of (a) sine-exponential model and (b) triangle model to temperature data from Makoka (illustrated by seven days in May 2009). ..................................................... 135

Figure 6-6. Output from three GCMs for 2040-2069 for grid cell containing Makoka, Malawi. Shown are monthly average changes in (a) precipitation (mm/day); (b) daily max temperature (ºC); (c) daily min temperature (ºC). .............................................................. 137

Figure 6-7. Relationship between rainfall and yield of unfertilized maize, Makoka, Malawi (1993-2006). ........................................................................................................................ 139

Figure 6-8. Trends in average maximum and minimum temperatures at Makoka, Malawi, 1971-2000..................................................................................................................................... 140

Figure 6-9. Trends in Makoka rainfall over time, 1971-2008. (a) Total rainfall, 1 July – 30 June. (b) Number of rain days (>2 mm rain). (c) Number of heavy rain days (>20 mm rain). (d) Coefficient of variation of daily rainfall, including zero values. (e) True start date of growing season. (f) End date of growing season (constrained to be after 1 February). (g) Length of growing season. .................................................................................................. 142

Figure 6-10. Relationship between the NINO3 index and Nov-Apr rainfall at Makoka, Malawi (1971-2000). ........................................................................................................................ 143

Figure 6-11. Comparison of BCCR:BCM2 baseline monthly temperature output (1975-2010) to Makoka historical data (1971-2008). .................................................................................. 143

Figure 6-12. Comparison of BCCR:BCM2 baseline monthly precipitation output (1975-2010) to Makoka historical data (1971-2008). .................................................................................. 144

Figure 6-13. Impacts of temperature increases on maize yield in 2040-2069, assuming a 1% decline in yield for every GDD>30. “Ideal” assumes no temperatures >30 ºC; “historical” represents data from 1971-2000; “T addition” represents addition of monthly temperature changes from BCCR:BCM2 to historical data; and “T simulation” represents daily temperature data from MarkSim with output from BCCR:BCM2. .................................... 147

Figure 7-1. Stirring nsima (maize porridge) on a typical cooking fire, fueled by two large logs.............................................................................................................................................. 158

Figure 7-2. Examples of animals observed utilizing Tephrosia and Gliricidia trees at Nkula Field. (a) frog; (b) tortoise beetle; (c) chameleon; (d) pin-tailed whydah; (e) ladybug; (f) black-eyed bulbul (nest). ..................................................................................................... 160

xvi

List of tables Table 2-1. Basic statistics on Malawi’s geography and economy. ............................................... 20

Table 2-2. Basic demographic data for Malawi, compared to regional and world averages. ....... 21

Table 2-3. Some characteristics of fertilizer tree species used in Southern Africa. ..................... 27

Table 2-4. Characteristics of Makoka Agricultural Research Station, Malawi. ........................... 31

Table 4-1. Characteristics of MZ12 topsoil (Oxic Haplustalf), 0-20 cm, Makoka Agricultural Research Station (from Makumba et al., 2009). ................................................................... 56

Table 4-2. Summary of drought experiment at MZ12 (written by the author for an informational sign for visitors mounted at the entrance to the field). ......................................................... 59

Table 4-3. Dates for key management activities at MZ12 drought experiment, 2008-2011. (For multi-day tasks, the start date is noted.) ................................................................................ 66

Table 4-4. Inputs of Gliricidia biomass to maize plots at MZ12. Leaves were returned to the plot while wood was removed. Values are kg/ha dry weight except as otherwise noted. Values are mean ± 1 SEM. No treatment effects were significant except for block (see text). ....... 68

Table 4-5. Response of maize yields (kg/ha) at MZ12 to cropping system, rain manipulation, and addition of inorganic fertilizer. Within a given year and a given variable (outlined in black boxes), means with different letters are significantly different (p < 0.05). .......................... 71

Table 4-6. Response of maize reproductive characteristics to cropping system, rain manipulation, and addition of inorganic fertilizer. Within a given year and a given variable (outlined in black boxes), means with different letters are significantly different (p < 0.05). n.s. = not significant. ............................................................................................................................. 72

Table 5-1. Topsoil (0-20 cm) properties, MZ21 (adjacent to Nkula Field), Makoka Agricultural Research Station (from Makumba et al., 2009). ................................................................... 89

Table 5-2. Dates for key management activities at Nkula Field drought experiment, 2008-2011................................................................................................................................................ 96

Table 5-3. Dates of destructive (biomass) and non-destructive seedling measurements at Nkula Field. ..................................................................................................................................... 99

Table 5-4. Maize yields in sole maize and three different agroforestry systems (established in 2008) under ambient rainfall and early planting. Yields are given in kg ha-1 and as a percentage of the control (sole maize in that year). Means not sharing the same letter at are significantly different at p = 0.05. ....................................................................................... 101

Table 5-5. Response of maize yields (kg/ha) at Nkula Field to cropping system, rain manipulation, and timing of planting of agroforestry trees. Within a given year and a given variable (outlined in black boxes), means with different letters are significantly different (p < 0.05). ................................................................................................................................ 103

xvii

Table 5-6. Tree biomass production (kg/ha) at Nkula Field as affected by cropping system, rain manipulation, and timing of tree planting. Within each outlined black box (indicating a given year, a given biomass type, and a given variable), means with different letters are significantly different (p < 0.05). Gray-shaded cells in 2008-2009 indicate that the trees were cut prematurely to reconfigure the experimental design (Section 5.2.1.3), rather than as part of their standard management. ..................................................................................... 106

Table 5-7. Production of litterfall (November 2009) and mature reproductive biomass (October 2010) by Tephrosia. Within each outlined black box (indicating a given year, a given biomass type, and a given variable), means with different letters are significantly different (p < 0.05). ............................................................................................................................ 108

Table 5-8. Survivorship of Tephrosia seedlings in relay intercropping and improved fallow systems, 2008-2009 and 2009-2010. Data are reported only for newly established seedlings (<1 year old), not for seedlings in their second year. Survivorship is based on the number of confirmed germinants at the first survey. Within a given year and a given variable (outlined in black boxes), means with different letters are significantly different (p < 0.05). ........... 113

Table 5-9. Summary of treatment effects (drought and late planting) on individual seedling responses. ............................................................................................................................ 117

Table 5-10. Average air temperatures and number of hours exceeding temperature thresholds at Nkula Field (30 March to 19 April 2010). No error estimates are calculated, as each row represents the average of only two plots of that type. ......................................................... 121

Table 6-1. Accuracy of several interpolation methods in recreating 28 days of half-hourly temperature data at Makoka. Models were evaluated for GDD with lower and upper thresholds of 10 and 30 ºC, respectively, and for GDD over 30 ºC. RMSE = root mean squared error. ...................................................................................................................... 134

Table 6-2. Locations of closest grid cells to Makoka in chosen GCMs. For BCCR, Makoka is located between two cells (longitudes 36.56 and 33.75); values were linearly interpolated between these cells. ............................................................................................................. 136

Table 6-3. Comparison of Makoka historical climate (1971-2008) with climate simulated by MarkSim using BCCR:CM3 (1975-2010). Temperature units are ºC; rainfall units are mm. Parentheses show standard deviations. Significance values are from t-tests between historical and simulated annual values. .............................................................................. 144

Table 6-4. Summary of current and future climate parameters relevant to maize production at Makoka, Malawi. Italic decimal numbers in parentheses (252.9) are standard errors, while fractions in parentheses (5/37) are the number of years that meet the criterion divided by the number of years in the sample. ........................................................................................... 145

xviii

List of acronyms and abbreviations BCCR:BCM2 Bjerknes Centre for Climate Research: Bergen Climate Model 2 CAN Calcium ammonium nitrate CIFOR Centre for International Forestry Research CV Coefficient of variation CSIRO:MK3 Commonwealth Scientific and Industrial Research Organisation:

Climate Model Mark 3 CGIAR Consultative Group on International Agricultural Research AICc Corrected Akaike Information Criterion DSSAT Decision Support System for Agrotechnology Transfer DD Degree-days DW Dry weight ENSO El Niño / Southern Oscillation FEWSNET Famine Early Warning System Network FISP Food Input Subsidy Programme FW Fresh weight GCM General circulation model GDP Gross domestic product GDD Growing degree-days INM:CM3 Institute for Numerical Mathematics: Climate Model 3 IPCC Intergovernmental Panel on Climate Change ICRAF International Centre for Research in Agroforestry LER Land equivalence ratio LSM Least squares mean MZ12 Maize experiment 12 (Makoka) MK Malawi kwacha MarkSim Markov Chain Rainfall Simulator MEI Multivariate ENSO index Ndfa Nitrogen derived from atmospheric fixation POM Particulate organic matter PES Payment for ecosystem services PPFD Photosynthetic photon flux density PAR Photosynthetically active radiation PPP Purchasing power parity RMSE Root mean squared error SOM Soil organic matter SADC Southern African Development Community SRES Special Report on Emissions Scenarios SEM Standard error of the mean SSA Sub-Saharan Africa TKW Thousand kernel weight UNDP United Nations Development Programme WUE Water use efficiency WaNuLCAS Water, Nutrient, and Light Capture in Agroforestry Systems

xix

Acknowledgements To thank everyone adequately is an impossible task, and this list does not come close. No words, and no rank order, can do justice to those who have made my work and my life possible.

First of all, I would like to thank Soroptimist International for awarding me the Founder Region Fellowship in 2010-2011. This generous $10,000 fellowship covered many of my research and living expenses. It has been a pleasure to get to know the board members of Soroptimist Founder Region, and I am highly honored to be their grantee.

I am grateful for additional support from several other funding sources: the Energy and Resources Group block grant and a UC Berkeley Graduate Division travel grant. Thank you also to the Charles A. and Anne Morrow Lindbergh Foundation for their vote of confidence in selecting my project for a Certificate of Merit in 2010.

My colleagues at the World Agroforestry Center in Malawi helped me in innumerable and invaluable ways. Without their warm welcome, ongoing participation, and financial and in-kind support, my work would have been impossible. My research supervisor, Festus Akinnifesi, has been the single most important source of advice and support for my fieldwork. He has always been swift to answer my questions and enthusiastic in his advocacy of my projects. The magnitude of his contributions cannot be overstated: this thesis would not exist without him. Festus would have been a member of my dissertation committee but for bureaucratic barriers.

During the project’s third year, when I was in California, Simon Mng’omba skilfully managed the logistics from Chitedze. On the ground at Makoka, the project’s daily tasks were faithfully overseen by Chikumbutsa Kwakwala. No words of thanks to Simon and Chiku can suffice; they have devoted many months of work to the success of this project.

Other ICRAF colleagues to whom I am especially grateful include Konisaga Mwafongo for teaching me the nuts and bolts of Gliricidia intercropping; Gudeta Sileshi for many stimulating theoretical discussions; Maxwell Ntungama for dozens of cheerful hours spent driving me up and down Malawi; Fannie Gondwe and Lorraine Itaye for expert administrative assistance; and Wezi Hara for data entry. Zikomo kwambiri!

The physical tasks for my fieldwork were carried out by many dozens of people. The most indispensable was Damson Singo, Makoka’s carpenter. He built and maintained the 36 rain shelters that formed the heart of my experiments, with assistance from Chimwemwe Mlongoti, Jose Kapesi, and Alfred Mwase. Mr. Singo also contributed to the design of the rain shelters. For sheer number of hours, the most important people in this project were the watchmen: Kingston Juma, Asida Machaka, Tiyes Malalo, and Justin Tambala. They (and other watchmen) spent, cumulatively, about 30,000 hours guarding my experimental fields.

The success of my experiments was due in large part to their location at Makoka Agricultural Research Station, so I want to express my gratitude to the management staff at Makoka – Godfrey Ching’oma, Rose Mkandawire, and Chimwemwe Kwangwani – for welcoming me and kindly granting me permission to use Makoka’s land and buildings. Thank you to Harry Kalombozi, Makoka’s meterologist, for recording the daily rainfall data.

xx

I also want to thank several outstanding businesses in Zomba who made my work easier and more pleasant: Agricultural Trading Company (ATC) who cordially helped me with many purchases large and small; Tasty Bites, where I sat with my laptop for hundreds of hours; and All Seasons Internet Cafe, whose staff stayed past closing time countless nights for my sake.

On a personal note, my time in Malawi was made immeasurably more rewarding by the daily companionship of my roommates and friends: Jacintha Kamponda, Katja Dietrich, and Kondwani Makoko. Jacintha helped me learn Chichewa, showed me how to cook nsima and tie a chitenje, and made our house feel like a home; for these things and a thousand more I will always be grateful. Katja saved my work-life balance by taking me on many adventures, and Kondwani provided good cheer and conversation after long days in the field.

My ability to carry out this dissertation project was built upon the foundation of academic and personal support I have received as a graduate student at UC Berkeley. Again, I cannot hope to adequately thank all those who have helped me, but I will try.

Margaret Torn, my dissertation advisor, has been a beacon of enthusiasm and bright ideas since my first semester at Berkeley. Her mentorship has led me to grapple with tricky scientific questions, break out of my academic comfort zones, and expand my analytical toolkit. Her warmth and encouragement have bolstered my confidence as I tackled problems I didn’t think I could solve. She has also steered me toward many opportunities to hone my research and teaching skills. Margaret is one of my core role models, and I cannot thank her enough for her guidance and patience over the past nine years.

John Harte, my academic advisor, has been cheering for me since before I set foot on campus. I am deeply grateful for his unparalleled patience, eager brainstorming, keen insights, and genuine care for his students. Despite being one of the most brillant people I know, John always makes me feel smarter when I talk to him. I also owe heartfelt thanks to the whole Harte Lab group for their feedback on my work over the years.

Meeting with Lynn Huntsinger, my third dissertation committee member, has always been a pleasure; I greatly admire her breadth of knowledge and am buoyed by her optimism. Many thanks to Lynn for keeping me realistic, motivated, and well-rounded.

My qualifying exam committee members, in addition to Margaret and Lynn, deserve special acknowledgement: Dan Kammen, Carol Shennan, and Todd Dawson. I had nightmares about my QE for a full three years before I took it, but they made it most enjoyable. Their influence extended far beyond the exam itself: Dan kept me in the loop on energy and agricultural policy in Africa; Carol shared practical advice on working in Malawi; and Todd answered countless questions on field methods. If not for his 2011 sabbatical in South Africa, Todd would have been on my dissertation committee as well.

I would also like to thank the other professors who have had a major influence on me during my time at Berkeley: Alex Farrell, Isha Ray, and Louise Fortmann. Isha and Louise instilled in me a profound respect for the role of social sciences in natural resource management; I only wish I had the talent to become more of a social scientist myself. I miss Alex’s shrewd critiques and boundless energy; his death will motivate me to work harder for the rest of my career.

xxi

My nine years at Berkeley were immensely enriched by my fellow graduate students in the Energy and Resources Group. ERG soon came to feel like home, especially thanks to my classmates who became personal friends (including Adam Smith, Barbara Haya, Malini Ranganathan, Adam Brandt, Elizabeth Stoltzfus, Danielle S. Christianson, and Anna Kantenbacher). I also got invaluable feedback on my research from fellow ERGies (including Arne Jacobsen, Rob Bailis, Rebecca Ghanadan, Cristina Castanha, Andy Jones, and Andrew Crane-Droesch). There are too many others to mention by name. Thank you all!

Special thanks are due to all the ERG staff for their tireless and cheerful efforts in making our department run smoothly, especially Sandra Dovali, Donna Bridges, and Kay Burns. Kay went far beyond the call of duty to make sure that I would be able to graduate on time and without undue financial burden; I cannot thank her enough.

At the end of the whole process, Daphne Szutu made my life much easier by spending weeks typing up tens of thousands of data points from my field notebooks. I owe Daphne my gratitude for her meticulous and diligent work, and I could not have met the filing deadline without her.

Thank you to several fellow scholars and good friends from outside Berkeley: Janice Chyou, Jessica Chevalier, and Zachary Mason. Jessica and Zach contributed directly to the writing of my thesis by spending dozens of hours in joint study with me. Janice and I, who now each have a few more letters after our names than when we first met, have been by each others’ sides through all kinds of weather.

A few acknowledgements are due to non-human entities, as well. Thanks to my faithful iBook G4, Clara, for stoically dealing with of massive amounts of data, daily jarring on dusty minibuses, and long weeks without rebooting. Thanks also to her successor, MacBook Madiba, on whom this thesis was written. And thanks to my Trek 1500 road bike, Will Scarlet, for getting me to and from campus faster than I thought possible.

I owe a debt of gratitude to caffeine and chocolate, two of the most useful substances ever discovered. I estimate that the completion of my doctorate has required the consumption of approximately 2000 cups of tea and coffee, as well as 75 kg of chocolate – much of which was in the form of Tim Tams and Violet Crumbles mailed across the world by my devoted mother. Thank you also to Edvard Greig, Antonin Dvořák, Camille Saint-Saëns, and other brilliant composers whose music soothed my mind as I grappled with overtaxing concepts.

This list would be incomplete without the names of those who are dearest to me. It is no overstatement to say that I could not have achieved this goal without them.

My husband Jeremy Manson is the most central of all my reasons for being joyful. A survivor of the Ph.D. process himself, he has been a shoulder to lean on, a cheerleader, a technical consultant, a stand-up comic, a counselor, and the dearest friend I could imagine. Whether visiting me in Malawi, sending me care packages full of Economist magazines and solar flashlights, or making roasted garlic soup to fortify me for a big deadline, Jeremy has kept a constant smile on my face. Of everyone mentioned here, Jeremy has paid the greatest cost: namely, two years of an absent partner – more than anyone should have to endure. But, to my

xxii

deepest gratitude, he did. I could not have borne my most difficult experiences in the field without the knowledge that Jeremy would be there for me, come what may.

The academic accomplishments of my brothers Rex Kerr and Cliff Kerr have inspired me, and their encouragement over the past nine years has been invaluable. Rex, who has a quintessential scientist’s mind, asked helpful questions despite the fact that his expertise is in an unrelated field. Cliff and I, although he is an undisclosed (but significant) number of years younger than me, had a race to finish our dissertations, and he won by a landslide. That’s all right. I don’t mind being the least brilliant one in the family.

My university education would not have been possible without financial support from my late grandparents Al and Wenetta Childs and my aunt Carol Childs. Even so, their financial contributions pale in comparison to their joyful encouragement over the years. Carol is one of my most faithful correspondents, wherever in the world she and I may be.

My parents Sandy and April Kerr have devoted a vast portion of their lives to teaching their children, and I will never be able to thank them enough for this. From my earliest days, I remember Mom teaching me the names of plants, Dad looking up words in the dictionary with me, and both encouraging me to ask questions constantly – and not to be satisfied with simple answers. How incredibly lucky I am to have parents like them.

My late grandparents Clark and Kay Kerr will always be a part of me. Through their devotion to education, their tales of travels across continents, and their commitment to making the world a better place for humankind, they set a shining example for me from my childhood through my graduate school career. My grandmother’s indomitable spirit stood her in good stead in her work as a champion for the environment and the citizens of the San Francisco Bay Area. Throughout my career I will try to carry on that spirit. It is to her that this thesis is dedicated.

Berkeley, California August 31, 2012

1

Chapter 1. Role of agroforestry in climate adaptation

Abstract

Agriculture is one of the economic sectors most vulnerable to negative impacts from climate change, particularly among subsistence farmers in the tropics. Agroforestry – the practice of growing trees on farmland – has recently attracted interest for its potential role in reducing this vulnerability; however, empirical data are sparse. This chapter reviews relevant hypotheses and available evidence for the climate adaptation benefits of agroforestry. The goal of this review is to provide a theoretical context for the field experiments described in Chapters 3 through 6.

Existing work has shown beneficial effects of agroforestry trees on microclimate (air and soil temperature, and evaporative demand); these local effects could help to ameliorate temperature increases caused by global warming. Trees can also play an important role in aiding soil infiltration and reducing runoff, which may become increasingly important with more extreme precipitation events. However, complementarity of water use between trees and crops may become a greater problem under future conditions of water stress. Apart from these biophysical effects, economic effects of risk diversification and income generation may make agroforestry systems an important tool for helping farmers cope with climate-related stresses.

A separate but related question is whether agroforestry systems themselves will be harmed by climate change, and if so, how this will affect their potential use for climate adaptation. The little work that has been done indicates that agroforestry trees generally fare well under drought, increased temperature, and elevated CO2, but much more research is needed.

Although there are potential synergies between climate adaptation and climate mitigation in agroforestry systems, it should not be assumed that these goals always coincide. Maximizing carbon storage may preclude the productive use of trees and tree products, and restoring soil fertility may cause N2O emissions. Agroforestry also poses difficulties for carbon accounting.

Opportunities abound for productive research in this area, including field experiments, improved simulation models, and meta-analyses of existing data. There is a great need for concrete information on how agroforestry systems will respond to global climate change.

1.1 Reducing vulnerability of agriculture to climate change

Global climate change poses a threat to agricultural production in many world regions, most particularly developing countries in the tropics (Lal et al., 2005; Cline, 2007; Easterling et al., 2007). Despite uncertainties in the extent and timing of future climate change, it is clear that significant changes will continue well into the 21st century, regardless of mitigation efforts. Thus, the need for climate change adaptation is now unavoidable (IPCC, 2007). Research on

2

adaptation is urgently needed in the agricultural sector, and the scope of the challenge demands international action (World Bank, 2007).

1.1.1 Current approaches to agricultural adaptation

Climate change can affect agricultural productivity through a variety of physical factors, including temperature, precipitation, CO2 fertilization, storm intensity, and sea level rise. The physical changes may also engender biological changes (such as changes in the balance of competition from pests, pathogens, and weeds) and socioeconomic changes (such as migration, public health burdens, and government resources) that can indirectly affect agricultural production. These phenomena are reviewed by Easterling et al. (2007).

A variety of solutions have been proposed to cope with these effects, such as changing crop species or varieties, using irrigation, expanding crop insurance, and promoting alternate livelihoods. However, it is increasingly realized that part of the reason for the climatic vulnerability of current agricultural systems may lie in the structure of the systems themselves, and that a return to more traditional and diverse agroecosystems may help to reduce this vulnerability. The following section further explores this idea.

1.1.2 Agroecological perspectives on climate resilience

In recent years, a broad consensus has emerged that more diverse ecosystems are generally more capable of resisting and recovering from the effects of environmental disturbance (Hooper et al., 2005; Balvanera et al., 2006), with the caveat that the processes involved can be highly complex. Many authors have asserted that this relationship between stability and diversity (at least functional diversity) also applies to agricultural ecosystems (Lin et al., 2008; Lin, 2011). Tomich et al. (2011) review principles of agroecology in the context of current and future global changes, emphasizing the importance of considering agricultural systems not in isolation but as components of social-ecological systems.

Though space does not permit a thorough discussion of the literature on agroecological resilience, the following sections touch upon several different perspectives on this issue: traditional knowledge and ecosystem mimicry. Both of these may shed light on the central question of this chapter: how can agroforestry systems help farmers adapt to variable and changing climates?

1.1.2.1 Insights from traditional knowledge

There exists a rich literature on indigenous practices for coping with climatic disturbances. Within these diverse practices, several consistent themes emerge.

First of all, subsistence producers seek to create systems that can survive climatic disturbance. A comparison of villages in Sweden and Tanzania (Tengö and Belfrage, 2004) revealed several universal strategies: diversification of agroecosystems both in space and in time; protection of natural ecosystems and wild species; use of natural indicators to predict climate variability; and careful maintenance of key ecosystem services such as nutrient cycling and pest control. In an unpredictable climate, risk cannot be avoided, but can be effectively managed by keen attention to the environment (Roe et al., 1998).

3

If the subsistence system itself fails, community members must fall back upon other options. A common response to drought-induced food shortage is the use of wild foods: in Sierra Leone, bush yams and palm hearts (Leach, 1994); in Kenya, native fruit (Eriksen, 2005); in Botswana, bushmeat and mopane caterpillars (Dube and Sekhwela, 2007). Natural ecosystems are often less vulnerable than agroecosystems to climate variability (as discussed in the next section).

1.1.2.2 Ecosystem mimicry

The concept of “ecosystem mimicry” (Ewel, 1999; Pate and Dawson, 1999) states that agroecosystems can achieve greater productivity and resilience if they share similar structures and functions to a natural ecosystem. These benefits are assumed to arise from the responses of diverse species in different functional groups engaging in complementary resource capture.

Agroforestry systems, though much more simplified than natural ecosystems, are more complex than monocultures, and as such, might be expected to exhibit more resilience to climate-related stresses (van Noordwijk and Ong, 1999). However, if inappropriate species combinations are used, the opposite may be true (Ong and Leakey, 1999). The following sections will review the available evidence for these ideas.

1.2 Definitions and examples of agroforestry

1.2.1 Definition and scope

Agroforestry is the integration of trees into farm landscapes. It can take a variety of forms and resists strict definition (Wojtkowski, 1998). Agroforestry has existed as long as agriculture itself; farmers have always valued trees for their multiple uses. It has only been a topic of academic interest, however, since the 1970s (Huxley, 1999), and many biological and socioeconomic questions remain to be answered (Sanchez, 1995).

Agroforestry is already widely used in both tropical and temperate regions. It has garnered increased attention in recent years for its potential to help increase food production (Garrity et al., 2010) while protecting the natural environment, for example by sequestering carbon (Makundi and Sathaye, 2004), conserving soil and water resources (Young, 1997), and protecting biodiversity (Schroth et al., 2004).

1.2.2 Theoretical bases of agroforestry

There are many ways in which the components of an agroforestry system can be integrated in space and time. In some systems, trees are grown in perpetual association with crops. Examples include trees as windbreaks, isolated trees in fields, and trees providing cover for a shade-loving crop (such as coffee or cocoa). In other systems, the spatial arrangement changes over time: for example, in the taungya systems of southeast Asia, high-value timber trees are planted amongst annual crops and are permitted to eventually overshadow the crop. And in other systems, such as improved fallows, the trees and the annual crop are never physically adjacent.

The central biophysical hypothesis of agroforestry (Cannell et al., 1996) is that, in order for the system to outperform a monoculture, the trees must acquire nutrients, water, or sunlight that

4

would not have been acquired by the crop. In its most basic form, the effect of agroforestry trees on crop yield can be described as follows:

I = F - C Equation 1-1

where I = increase in crop yield; F = fertility-enhancing effect of trees; C = competitive effect of trees.

This can be further refined as:

I = Fnoncomp – Ccomp,nonrecyc Equation 1-2

where I = increase in crop yield; Fnoncomp = fertility-enhancing effect of trees that does not depend on resources obtained competitively from the crop; Ccomp,nonrecyc = competition effect of trees due to the appropriation of resources that are not ultimately recycled back to the crop.

In some cases, crop production is limited by unfavorable environmental conditions rather than by resource limitation per se. In this case, agroforestry trees can benefit the crop indirectly via effects on microclimate, soil and water conservation, and pest control.

Whatever the reason for the beneficial effect of the trees, it can be described by the land equivalence ratio, or LER (Wojtkowski, 1998):

LER = (Yab/Ya) + (Yba/Yb) Equation 1-3

where Ya = yield (units of value per area) of species a as a monoculture; Yab = yield of species a when intercropped with species b; Yb = yield of species b as a monoculture; Yba = yield of species b when intercropped with species a.

If the two species have no effect on each other, then LER = 1. By contrast, LER > 1 indicates that complementary or facilitative effects dominate, while LER < 1 indicates that competitive effects dominate. (Note that this formulation assumes the same planting density whether the species is grown as a monoculture or an intercrop.)

These biophysical descriptions of complementarity do not include the more complicated issue of economic complementarity. Apart from their direct effect on crop yields, agroforestry trees can also provide the farmer with fuelwood, timber, fodder, fruit, nuts, honey, and medicine (all of which can generate cash income), as well as carbon sequestration and biodiversity conservation. Depending on the timing and the value of these other outputs, this economic complementarity can justify the use of an agroforestry system even when the crop does not directly benefit.

5

1.2.3 Common elements of agroforestry systems

Agroforestry is practiced in both tropical and temperate climates. In temperate climates, especially in developed countries, the emphasis is usually on large-scale static systems (such as riparian buffers) to confer environmental benefits. By contrast, in tropical developing countries, agroforestry systems are usually more complex in space and time (e.g., Kumar and Nair, 2004) and are practiced on a smaller scale, with emphasis on food production and income generation.

Despite this diversity of systems and locations, all agroforestry systems share several common elements as compared to crop monocultures:

• greater diversity of species and growth forms;

• perennial species as a central component;

• greater potential for competition and complementarity;

• greater complexity in use and management.

With these commonalities in mind, let us now examine how the benefits of agroforestry systems may be enhanced or reduced in the context of global climate change.

1.2.4 How might agroforestry play a role in climate adaptation?

Fortunately, there are many reasons to expect that agroforestry could help farmers maintain the productivity of their land under warmer, drier, more variable climates. In the most general sense, agroforestry may buffer against adverse climate conditions by:

• physically protecting crops (e.g. from heat, flooding, soil erosion)

• accessing different resources when resources become scarce

• diversifying sources of food and income generation

• increasing overall agricultural productivity (and/or income)

Each of these will be discussed more detail below in the context of climate adaptation. These benefits of agroforestry systems have already been well described in the literature, though not usually with reference to climate adaptation. The exception is a review paper by Rao, Verchot, and Laarman (2007). This chapter attempts to build upon and expand their work.

1.3 Climate modulation by agroforestry systems: direct effects

In theory, the presence of trees can either ameliorate or exacerbate undesirable climatic effects; thus, the word “modulation” is used here rather than “moderation.”

6

1.3.1 Microclimate

1.3.1.1 Air and soil temperature

Shade from mature trees reduces air and soil temperature (and, as a result, can increase soil moisture). If these benefits outweigh competition for light and water, crop productivity may be enhanced. This could be especially important where temperatures begin to exceed the maximum physiological tolerance for a staple crop, as is expected to happen in many locations throughout the tropics (Sanchez, 2005).

An example of this effect is the agroforestry parklands in the Sahel in which mature Acacia albida trees are interspersed throughout millet and sorghum fields, approximately doubling the yield of the crop in their vicinity (Sanchez, 1995). Acacia albida is a useful agroforestry species because of its phenology: it sheds its leaves during the growing season, when the crop’s light requirements are highest. During the early growing season, when high soil temperatures can impede crop establishment, even a leafless Acacia albida canopy can reduce soil temperatures in the vicinity by up to 10 ºC (Vandenbeldt and Williams, 1992).

Microclimatic benefits have been demonstrated in a variety of agroforestry systems: for example, shade trees on coffee farms in Mexico help to moderate both maximum and minimum air temperatures (Lin, 2007), while shade trees on Indonesian cacao farms can reduce air temperature by 4 ºC (Steffan-Dewenter et al., 2007).

1.3.1.2 Evaporative demand

The presence of trees can reduce evaporative demand in crop canopies not only by affecting air and soil temperature, but also by increasing local humidity via transpiration and by reducing wind speed. This has been demonstrated for the two systems mentioned above: shaded Gliricidia-cacao agroforests in Indonesia had 12% higher relative humidity than cacao monocultures (Steffan-Dewenter et al., 2007), and shaded coffee agroforests in Mexico had 32% lower evaporative demand than unshaded systems (Lin, 2010). Other examples are thoroughly reviewed by Stigter (2010a).

1.3.2 Soil properties

1.3.2.1 Soil water-holding capacity

Several types of agroforestry systems have been shown to substantially increase soil organic matter after only a few years (Albrecht et al., 2004). This is more obviously a mitigation benefit than an adaptation mechanism. However, soil carbon (and improved soil structure) can increase soil water-holding capacity, thus enabling the soil to retain moisture for longer after a precipitation event. Phiri et al. (2003) directly demonstrated this benefit in Sesbania improved fallows in Zambia. Regardless of changes in total precipitation, soil water-holding capacity could become especially important if precipitation tends to fall in fewer, larger events.

1.3.2.2 Changes in infiltration and runoff

Agroforestry practices can increase soil infiltration rates through several mechanisms: improved soil structure and porosity, channels left by dead tree roots (Chirwa et al., 2003b), and changes in

7

small-scale soil topography (Lin and Richards, 2007). The resulting infiltration increases have the advantage of both increasing recharge of soil water and reducing runoff that could potentially lead to erosion (Cannavo et al., 2011). Via effects on infiltration, agroforestry trees can also help to protect soils that are prone to waterlogging (F. Akinnifesi, pers. comm.).

However, there is a downside to this effect: depending on the species of trees and the local hydrology, an increase in tree cover can actually decrease the surface water supply (Hayward, 2005), causing problems for those who depend on water from streams and rivers. This contrasts with the common argument that planting more trees on a watershed or regional scale will increase rainfall. In any case, this effect would not usually arise at an individual farm scale.

1.3.2.3 Water use competition or complementarity

If trees are able to access a deeper layer of the soil water profile than crops, they may be able to reliably produce useful biomass – even in drought conditions – while not interfering with crop growth. Several recent studies have suggested that shown that agroforestry systems can increase water use efficiency (WUE) under a variety of conditions (Rao et al., 2007; Sileshi et al., 2011). Even if the trees do not increase yields of an annual crop, they may provide the farmer with an important auxiliary source of food or income.

However, it can be difficult to design agroforestry systems with this degree of complementarity in water use (Ong et al., 2002). Even in agroforestry systems that generally currently show good complementarity in water use, water competition may arise as a result of climate variability (Rao et al., 2007) or future climate change (Cannavo et al., 2011). When trees and crops directly compete for water, productivity of one or both components will be diminished, depending on which is the more effective competitor.

Therefore, agroforestry systems in which trees and crops use overlapping water supplies may become increasingly unsuitable in places where climate change reduces water availability. Ong and Leakey (1999) argue that many agroforestry species in semi-arid Africa are likely to provoke water competition even under current climatic conditions.

1.3.2.4 Hydraulic lift

In some cases, trees can bring water up from depth and release it into the surface layers of the soil in a process called hydraulic lift (Caldwell et al., 1998). This effect has been demonstrated in the agroforestry species Cajanus cajan, though no such effect could be found for Sesbania sesban (Sekiya and Yano, 2004). Under dry conditions, a tree is unlikely to release water into surface soils, thus its net effect on nearby shallow-rooted species will likely still be neutral or negative (Ludwig et al., 2003; Ludwig et al., 2004). Hydraulic lift may in theory be able to facilitate crop growth, but no examples could be found of this occurring in the field.

1.3.2.5 Soil nutrient status

Soil nutrient replenishment is the primary goal of many agroforestry systems worldwide. Although nitrogen-fixing perennials are especially well suited for this purpose (Mafongoya et al., 2004), agroforestry can confer soil fertility benefits even when N fixation does not occur or where N is not a limiting nutrient. Relevant mechanisms include the capture and return of

8

nutrients that would have escaped the crop rooting zone and extraction of nutrients from deep soil horizons where annual crops cannot reach (Cannell et al., 1996).

The contribution of nutrient replenishment to climate change adaptation is less obvious than the contribution of microclimatic effects. However, nutrient availability to plants is in part determined by soil moisture, so an increase in soil nutrients may partly compensate for a decrease in soil moisture. For example, a recent study in Malawi (Snapp et al., 2010) showed that nitrogen inputs from short-rotation perennials (pigeonpea and Tephrosia vogelii), as well as inorganic nitrogen addition, helped to reduce variability in maize yield across a range of locations and rainfall patterns.

1.3.3 Physical protection against storm damage

The effect of global warming on the intensity and frequency of storms is still a subject of much debate, but current evidence suggests that intensity of precipitation events is likely to increase and damaging wind speeds may become more frequent (Meehl et al., 2007). The function of agroforestry systems as physical barriers to storm damage is already well-known (Stigter, 2010b), and this role could become increasingly important under future climate.

One example of this protective function is erosion control. Agroforestry practices such as hillside terraces and riparian buffers can be highly effective in preventing erosion and protecting water quality (Rocheleau et al., 1988). If precipitation events increase in intensity, effective erosion control will become even more necessary, especially in areas where steep hillsides are cultivated. An example of this was seen after a 2005 hurricane in Chiapas, Mexico, where structurally complex coffee agroforests suffered less landslide damage than comparable coffee monocultures (Philpott et al., 2008).

1.4 Climate modulation by agroforestry systems: indirect effects

The previous section discussed physical mechanisms by which agroforestry systems might influence the microenvironment experienced by crops, thus potentially moderating (or exacerbating) the local effects of global climate change. However, there are many other climate-relevant functions of agroforestry that do not depend on direct physical interaction between trees and crops.

1.4.1 Biodiversity conservation

It has been suggested (Hannah, 2004) that agroforestry can play an important role in ecosystem adaptation to climate change by creating habitat corridors through which species can migrate. This can be the case even when trees are sparse or isolated (Harvey et al., 2004), although larger and more mature agroforestry trees are likely to have greater biodiversity value than small or heavily pruned trees (Garbach et al., 2011). Although biodiversity conservation does not necessarily have a direct benefit to farmers, it could be accounted for in ecosystem service payment schemes (see Section 1.4.4).

9

1.4.2 Weeds, pests and diseases

It is possible that increased biodiversity in agroforestry systems could help to reduce climate-related pest outbreaks, though this has not been demonstrated so far. In fact, the opposite effect may materialize. Previous work in sub-Saharan Africa has demonstrated that agroforestry trees themselves are vulnerable to a variety of pests, including aphids, beetles, termites, and cutworms (Sileshi et al., 2008), and in some cases these pests can spread to and damage the associated food crop (Vanlauwe and Giller, 2006). It is not clear under which conditions pest-reducing or pest-enhancing effects would be expected to dominate; this is a complex question for which empirical evidence is almost wholly lacking thus far.

Climate change may also affect the balance of competition between crops and weeds (Ziska and Runion, 2007). Weed reduction is a significant benefit of some agroforestry systems (Hauser et al., 2006), and this function may become increasingly important in cases where the competitiveness of weeds is enhanced due to altered climate or increased atmospheric CO2.

1.4.3 Risk management by diversification of production

This may be the most important of all the potential adaptation functions of agroforestry, but it is not always straightforward. In general, a diverse agroecosystem can buffer against risk if its constituent species respond differently to disturbances (van Noordwijk and Ong, 1999). In the case of agroforestry, the perennial species is indeed likely to have a different response function; however, it usually does not produce a staple food, so for food security to be enhanced by this diversity effect alone, the farm must be embedded within a functioning market system.

Fortunately, that is usually the case, and, as mentioned above, sale of tree products has long been an important coping strategy in cases of crop failure (Eriksen et al., 2005; Dube and Sekhwela, 2007). A recent study by Nguyen et al. (2012) found that multipurpose agroforestry trees in Vietnam are less vulnerable to climatic disturbances than rice, the staple crop, and that trees help to ensure food and income security in years of unfavorable climate.

1.4.4 Enhancing farm productivity and income

An overall increase in farm productivity (which should be considered separately from the above discussion of stabilizing productivity) can potentially provide savings to fall back upon in years when unfavorable temperature or precipitation result in poor yields. As nearly all agroforestry systems share the goal of increasing farm productivity (for example, by replenishing soil nutrients as described in Section 1.3.2.5), this could potentially be a very widespread benefit. However, for the buffering effect of enhanced farm income to be fully realized, adequate physical and financial infrastructure is necessary.

Thorlakson (2011) reviews this topic in the context of agroforestry systems in Western Kenya and concludes that “For extremely poor households, improving general well-being will be the most effective way to enhance resilience to future hazards associated with climate change.”

Agroforestry can contribute to farm income not only through increased crop yields and salable tree products, but also via payment for ecosystem services (PES). Examples of ecosystem services provided by agroforestry systems include carbon sequestration, biodiversity

10

conservation, and erosion control. A thorough discussion is outside the scope of this paper, but a recent review by van Noordwijk et al. (2011) concluded that agroforestry provides many opportunities for smallholder farmers to benefit from PES while simultaneously increasing their own resilience to climate change.

1.5 Impacts of climate change on agroforestry systems

The above effects of agroforestry systems on the climate resilience of agricultural production will not necessarily materialize if climate change causes agroforestry systems themselves to function differently. Although trees can be negatively affected by the same climatic factors as annual crops, in some cases the performance of agroforestry systems might actually improve as a result of climate change. This topic has received very little attention thus far (Neufeldt et al., 2012). The following section reviews several potentially important avenues by which climate change may affect the success of agroforestry systems.

1.5.1 Precipitation

Tree seedlings can be especially sensitive to water availability during their germination and establishment phases. In any agroforestry system, but especially those that require frequent establishment of seedlings, water stress may kill enough young trees to render the system ineffective. This phenomenon, though undoubtedly common, is not well documented in the literature; one exception is a paper by Kwesiga et al. (1999) describing establishment failure of Sesbania sesban due to sporadic rainfall in Eastern Zambia.

Mature trees, as compared to newly-established seedlings or annual crops, are generally more resilient to fluctuations in water supply, which is one major reason that agroforestry systems may prove beneficial in future climate. However, very few attempts have been made to test this assumption empirically. At the time of writing, Schwendenmann et al. (2010) had conducted the only experiment in which drought stress was imposed on an agroforestry system. Their study on a cacao-Gliricidia intercrop in Indonesia indicated that Gliricidia suffered no notable harm from a 13-month drought and showed no evidence of water competition with cacao.

Drought resilience of agroforestry trees is likely to depend not only upon the tree species and the location, but also upon the other components of the system; therefore, responses to water stress at the single-species level cannot necessarily be extrapolated to agroforestry systems. Much more work is needed on this complex topic, especially on agroforestry species that are used where droughts are common or likely to become more so.

1.5.2 Temperature

The only available study examining the effect of increased temperature on agroforestry species (Esmail and Oelbermann, 2011) found that seedlings of Cedrela odorata and Gliricidia sepium showed favorable growth responses to a 2 ºC temperature increase. However, this study has limited scope for extrapolation, as it was conducted in a growth chamber on single-species microcosms maintained at constant soil moisture. Effects of increased temperatures in the field are likely to be more complicated, including water stress, fire risk, and pest and disease burden.

11

In general, agroforestry trees would be expected to show less vulnerability than annual crops to direct harm from high temperatures. As relatively slow-growing perennials, they are inherently less sensitive to heat, and they have more reserves and a longer time span in which to recover from temperature-induced injury. Nonetheless, it will be useful to establish upper limits for the physiological heat tolerance of agroforestry species. For annual crops, such limits are generally well known and depend upon the phenological stage of crop development.

As an important caveat, warmer temperatures may pose a significant risk to agroforestry species whose primary function is to produce edible or marketable reproductive biomass (flowers, fruits, or seeds). Successful pollination and fruit development often depends upon favorable climate conditions during a short time window, and excessive heat or violent storms during that time may mean the failure of a year’s harvest. It is also worth noting that temperate agroforestry systems may be adversely affected by a lack of adequate chill for fruit production (Neufeldt et al., 2012).

1.5.3 CO2 fertilization

No experiments using carbon dioxide enrichment have been performed on agroforestry species in the field, though several have been performed in growth chambers. The question of how CO2 affects tree growth is particularly interesting for nitrogen-fixing species, which generally show more favorable growth responses to CO2 fertilization than non-fixing species (Thomas et al., 2007) and may increase their N fixation rates as a result.

Several experiments have upheld these expectations. Tissue et al. (1997) found that elevated CO2 stimulated growth (by 84%) and N fixation (by 25%) of 70-day-old Gliricidia seedlings, even with limited soil N (contrasting with earlier results by the same group, in which limited soil N curtailed the CO2 enhancement effect). Tischler et al. (2000) found similar effects for two Sesbania species. By contrast, Esmail and Oelbermann (2011), in the study on Cedrela odorata and Gliricidia sepium mentioned above, found that CO2 enrichment decreased foliar nitrogen content and did not enhance growth in either species.

In agroforestry systems, elevated CO2 may be important for its effects not only on tree growth, but also on the C:N ratios of tree biomass, which may in turn have noticeable effects on litter decomposition and soil nutrient availability. Other belowground effects are also possible: CO2 enrichment can result in increased carbon allocation to roots (Thomas et al., 2007), which in turn can affect the balance of competition between perennials and annuals (Bond and Midgley, 2000). This phenomenon could in theory appear in many if not most agroforestry systems, but whether it would significantly influence their function is yet unknown.

Similarly to precipitation and temperature effects, CO2 effects on agroforestry systems can only be fully understood when studied in species mixtures under field conditions. Unfortunately, practical considerations make this difficult; in fact, to date, no CO2 enrichment experiments have been carried out in any type of tropical cropland (Schlenker and Lobell, 2010).

1.5.4 Extreme events and natural disasters

Although agroforestry trees may help to buffer the effects of extreme climate events (as described in Section 1.3.3), trees themselves are vulnerable to these events. Storms, drought, fire,

12

or pest outbreaks can damage or kill even mature agroforestry trees, causing a significant loss of investment for the farmer (Osborne, 2010). Ideally, an agroforestry system would at least be more disaster-resistant than the systems it replaces, and if so, it can be considered a useful adaptation to climate-related disasters (Kalame et al., 2011). However, even so, the destruction of an agroforestry system may represent a greater loss of investment than the destruction of annual crops.

If severe or repeated climate disturbances necessitate temporary or permanent migration, investment in trees will be not only unlikely but unwise. This effect has already been seen in some parts of the Sudan and Sahel regions of Africa due to current climate variability and change (Ziervogel et al., 2006).

1.6 Synergies and conflicts between adaptation and mitigation

Several recent papers have mentioned the potential for synergy between climate adaptation and climate mitigation – i.e., carbon storage – in agroforestry systems (Verchot et al., 2007; Syampungani et al., 2010; Kalame et al., 2011). However, in each case, the details of the purported climate adaptation have been discussed briefly and broadly, without invoking empirical evidence. Although it is appealing to assume that the goal of carbon storage would usually align with the potential climate adaptation benefits of agroforestry described above, this question deserves closer examination and will be explored briefly below.

1.6.1 Previous work on climate mitigation aspects of agroforestry

Agroforestry systems usually store more carbon than the monocultures they replace, but the magnitude and timing of C accumulation varies greatly from one system to another. Some relevant principals and examples are reviewed by Albrecht et al. (2004).

Even similar systems in similar locations can exhibit different trends: a 2009 study at Msekera Research Station in Zambia (Kaonga and Bayliss-Smith) showed that leguminous agroforestry trees contributed substantially to soil organic carbon stocks after only a few years, while a study on similar species on farms in Malawi failed to find any effect of the trees on soil C storage, even over a decadal time span (Snapp et al., 2010). Differences in soil type and management practices may contribute to these seeming incongruities.

A global assessment by Makundi and Sathaye (2004) estimated that agroforestry represents between 6% and 21% of the total C sequestration potential of tropical forestry achievable by the year 2030, much of it at a low or negative cost. However, they acknowledged the difficulty of monitoring and verifying changes in C stocks on smallholder farms.

Several authors (Palm et al., 2010; Kim, 2012) have noted that it is important to consider the effects of agroforestry systems on non-CO2 gases. Especially in nitrogen-fixing systems, N2O emissions can substantially counteract the climate mitigation effects of carbon sequestration.

1.6.2 Potential synergies

If Thorlakson (2011) is correct that increasing farm income is the single most important way to help resource-poor farmers cope with the risks of climate change, then carbon payments might

13

be one of the key processes by which agroforestry systems could aid climate adaptation. Major opportunities and obstacles relating to carbon payments for agroforestry are reviewed by Anderson and Zerriffi (2012).

A potentially important biophysical synergy is the fact that increases in soil organic matter (a mitigation goal) tend to improve soil water-holding capacity (an adaptation goal). This was demonstrated by Phiri et al. (2003) in improved fallows in Zambia, but many more examples are needed. Inputs from agroforestry trees may affect different soil organic matter (SOM) fractions differently, and the effect may depend on edaphic factors such as soil clay content (Beedy et al., 2010).

1.6.3 Potential conflicts

An effective way to sequester large amounts of carbon quickly in an ecosystem is to establish fast-growing trees and preserve them untouched. However, this is at odds with the multi-purpose nature of most agroforestry systems: the trees are meant to be used. Most of the uses of agroforestry trees - whether as green manure, fodder, firewood, timber, fruit, or something else – involve appropriation of tree biomass and the release of its carbon. An exception might be trees that are planted to provide shade or physical barriers, but even in these cases the trees are often pruned, and tree density must be limited to avoid interfering with crop growth. Thus, maximizing carbon storage is unlikely to simultaneously maximize an agroforestry system’s direct utility to a farmer (for example, Wise and Cacho, 2005).

Another effective way to use trees for carbon sequestration, rather than leaving them untouched, is to ensure that their timber is harvested for durable products. However, this is also at odds with the needs of smallholders: fuelwood is generally needed in much larger volumes than building materials, and resource-poor farmers may have difficulty waiting for a decade or more to see a return on their investment in timber trees (Osborne, 2011).

In the case of agroforestry systems designed to restore soil fertility, the amount of nitrogen fixed by the system may correlate directly with its N2O emissions, thus placing a farm-level benefit in conflict with a detriment to the global climate. In nitrogen-fixing agroforestry systems, the global warming potential of N2O emissions can exceed that of sequestered CO2 and should not be overlooked (Kim, 2012). However, the net effect of such an agroforestry system of climate mitigation effect depends on whether it is compared to an unfertilized system or to a system with other N inputs (such as inorganic fertilizer). Although N inputs from agroforestry are often assumed to have the same N2O emissions factor as inorganic N (e.g., Palm et al., 2010), this assumption may be invalid due to differences in input quality and timing, as well as due to effects of the trees themselves on soil texture and microclimate.

A recent review by Anderson and Zerriffi (2012) argues that development and carbon goals of agroforestry projects often conflict with each other, and that it is not wise to maximize the latter at the expense of the former. Furthermore, although it is generally smallholders who most stand to benefit from agroforestry, smallholders are least attractive from a C mitigation standpoint due to high transaction costs. The authors conclude that “many of the tensions described may be difficult to resolve because the underlying approach of carbon mitigation and development projects are quite different.”

14

1.7 Need for future work

A recurring theme throughout this chapter has been the notable lack of empirical research on the role of agroforestry systems in climate adaptation. Given the widespread current use of agroforestry systems, and their potentially increasing future role in meeting challenges of food production and environmental protection, their performance under future climate deserves much more study. Following is a brief summary that highlights some of the biggest gaps.

1.7.1 Experimental climate manipulations in the field

To date, there has been only one published example of a climate manipulation on an agroforestry system (Schwendenmann et al., 2010). Although the stature and spatial complexity of most agroforestry systems makes them less amenable to in situ climate manipulations than annual crops, this difficulty is by no means insurmountable, as quite a few such manipulations have been carried out in natural ecosystems of greater complexity and stature (for example, Nepstad et al., 2007).

One reason for the lack of climate manipulation studies on agroforestry systems is that the vast majority of field studies on the effects of climate change – both ecological (Martin et al., 2012) and agricultural (Schlenker and Lobell, 2010) – have been located in temperate developed countries, in accordance with the concentration of financial resources and expertise. Agroforestry systems are of greatest importance to smallholders in the tropical developing world and have not attracted the same level of research attention.

Growth chamber studies on agroforestry species, such as those conducted by Tissue et al. (1997) and Esmail and Oelbermann (2011), can help to establish baseline physical responses of a given species to environmental fluctuations. However, because of the importance of species interactions in agroforestry systems, they are not an adequate substitute for field studies.

1.7.2 Simulation modeling

The current state-of-the-art simulation models for intercropping systems lag far behind those for monocultures, and this is especially true for agroforestry systems (Neufeldt et al., 2012). This is likely due to both the greater complexity and the lesser economic importance of intercrops as compared to monocrops.

WaNuLCAS (Water, Nutrient, and Light Capture in Agroforestry Systems) is a model for simulating above- and below-ground interactions in a two-species system containing a tree and an annual crop (van Noordwijk and Lusiana, 1999). Although WaNuLCAS is probably the most widely used and most intensively developed agroforestry simulation model, neither it nor any other current model is suitable for accurate long-term simulations of climate change effects on agroforestry systems (Neufeldt et al., 2012). Although it has potential to do so, it would first need to be modified to account for changes in phenology and the effects of long-term drought (van Noordwijk, pers. comm.).

15

1.7.3 Meta-analyses of agroforestry climate response across space and time

Despite the lack of dedicated studies on the climate response of agroforestry systems, a wealth of information can potentially be obtained from examining existing data across different years and locations. At the time of writing, the only available meta-analysis on the climate response of agroforestry systems was a study by Sileshi et al. (2011) examining long-term rain-use efficiency in three different fertilizer tree systems across southern Africa. This study provides a good example of a question that could be asked about other systems in other regions.

Many other questions are amenable to meta-analysis: for example, under what conditions do agroforestry systems show reduced yield variability as compared to monocultures? Are there thresholds of environmental stress beyond which a given agroforestry system fails to convey yield benefits? What is the climate envelope of a given agroforestry type, and how will that envelope shift with future climate change?

1.8 Conclusions

The role of agroforestry in climate change adaptation is an important but almost wholly unexplored question. The literature reviewed here suggests that, on balance, agroforestry has the potential to reduce the vulnerability of smallholder agriculture to climate change. However, this suggestion is based mostly on theory or on extrapolation of results from monoculture systems; empirical, field-based evidence is currently sparse.

Given the inevitability of anthropogenic climate change, and the potential vulnerability of smallholder farmers, agroforestry’s role in climate adaptation needs to be examined systematically across systems and locations. At present, it seems premature to generalize that agroforestry systems are as a rule more suitable for future climate than are comparable monocultures, or that synergies between adaptation and mitigation are inevitable. Promotion of agroforestry technologies without regard to their performance in future climatic conditions could be detrimental to the farmers they are intended to benefit.

There are many opportunities for productive research in this area, including field experiments, improved simulation models, and meta-analyses of existing data. It is time to move past optimistic generalizations about agroforestry and time to gather concrete information on how agroforestry systems will actually respond to global climate change.

1.8.1 Acknowledgements

Portions of this chapter were previously submitted as a prospectus for the author’s doctoral qualifying exam (November 2007). The prospectus material was greatly improved by comments and suggestions from the exam committee members (Dan Kammen, chair; Margaret Torn; Lynn Huntsinger; Todd Dawson; and Carol Shennan, UC Santa Cruz).

Grateful acknowledgement is given to Louis Verchot (CIFOR, Bogor, Indonesia) and Meine van Noordwijk (ICRAF, Bogor, Indonesia) for several informative discussions on this topic, both in person and by correspondence.

17

Chapter 2. Introduction to Malawi field experiments

Abstract

This chapter describes the motivation, geographical and historical context, experimental approach, and intended outcomes of the experiments presented in this dissertation. The overarching motivation of the work is twofold: (1) to contribute to the goal of agricultural climate adaptation in the developing world and (2) to provide a detailed and mechanistic example of how an agroforestry system responds to a climatic disturbance.

This particular system and location – fertilizer tree agroforestry systems in Malawi – was chosen based on evidence that subsistence agriculture in Malawi may be particularly vulnerable to climate change and that agroforestry systems might confer resilience to climate change; however; both of these ideas are as yet unproven.

Malawian agriculture faces many challenges, including low soil fertility, high population density and small farm size, and structural economic challenges that impede development. Although agricultural productivity has been recently boosted by a fertilizer subsidy, the long-term sustainability of smallholder agriculture is still at risk. Agroforestry research in Malawi, spearheaded by SADC-ICRAF, has focused on alleviating some of these problems. Much work has been done on fertilizer trees (fast-growing nitrogen-fixing trees that are intercropped with maize) with impressive effects on maize yields but limited adoption thus far.

This project attempts to fill several gaps in the fertilizer tree literature by (1) examining the response of fertilizer trees to a controlled drought, thus testing their suitability for conditions of erratic rainfall; and (2) comparing the drought response of seedlings and mature trees.

The project was carried out at Makoka Agricultural Research Station in southern Malawi with the assistance of many local field workers and several ICRAF colleagues. Although the scope of a three-year field experiment is necessarily limited, it is hoped that these results will contribute both to agroforestry theory and to practical climate adaptation knowledge for Malawian smallholders.

2.1 Motivations and goals of this project

The central question of this dissertation is as follows:

How and why will drought affect the performance of fertilizer tree agroforestry systems in Malawi?

To answer this question, a novel design for rain exclusion shelters (Chapter 3) was developed and deployed in both a mature fertilizer tree system (Chapter 4) and several newly established fertilizer tree systems (Chapter 5). The results are considered in the context of predicted climate change in Malawi and its effects on agricultural production (Chapter 6).

18

As a prelude to the experimental results, this chapter discusses the motivation for the research, provides background on the location and the field site, outlines the experimental approach, and presents the intended outcomes.

2.1.1 Protecting food security in the developing world under future climate

It is widely reported that anthropogenic climate change poses major risks to the agricultural sector, particularly in tropical developing countries (Cline, 2007). Agricultural climate adaptation deserves a prominent position on the global development agenda (World Bank, 2007), and this project attempts to inform climate adaptation planning in several ways.

2.1.1.1 Identifying climate-resilient production systems in vulnerable areas

To meet the challenges of agricultural production under global change, Tomich et al. (2011) call for “not a single path but many paths of sustainable intensification (and in some cases deintensification)... based on a wide range of systems that are appropriate to a very large number of specific agroecological and socioeconomic contexts.” The lack of a one-size-fits-all solution means that much work is needed to identify agricultural systems suitable for future climatic conditions at vulnerable locations.

The choice of this particular system and location – fertilizer tree agroforestry systems in Malawi – was based on published evidence that subsistence agriculture in southern Africa may be particularly vulnerable to climate change (Chapter 6) and that agroforestry systems might have greater climate resilience than comparable monocultures (Chapter 1). However, these ideas were approached as questions to be answered rather than assumptions to be built upon.

2.1.1.2 Meeting the need for more climate manipulation experiments

Due to their expense and logistical challenges, field-based climate manipulation experiments are rare even in the developed world. In the developing world, they are essentially non-existent. For example, there have been no CO2 enrichment experiments in croplands anywhere in the tropics (Schlenker and Lobell, 2010). Despite the difficulties involved, climate manipulations provide an essential tool in understanding the response of ecosystems to environmental change, as they are able to investigate cause and effect in a way that simulation modeling or statistical modeling are not (Shen and Harte, 2000).

This project intends to increase the accessibility of such experiments by developing a low-cost design for rain manipulation shelters suitable for use in agricultural systems in the developing world. At the time of writing, no such design could be found in the existing literature.

2.1.2 Understanding response of agroforestry systems to climate change

As detailed in Chapter 1, there is an almost complete lack of experimental data on how agroforestry systems respond to changes in climate. Even indirect approaches to this question – simulation modeling, meta-analysis, and systematic review of relevant theory – have rarely been employed. Despite the lack of data, many researchers are optimistic about agroforestry’s potential contributions to climate adaptation. This dissertation aims to address several aspects of this knowledge gap.

19

2.1.2.1 Gathering empirical evidence on climate response of agroforestry systems

To date, there has been only one climate manipulation experiment on an agroforestry system anywhere in the world (Schwendenmann et al., 2010), in a Gliricidia-cacao system in Sulawesi, Indonesia. Verchot et al. (2007) have identified studies of climate response across agroforestry systems and locations as a high priority for future research.

Thus, it is intended that this dissertation will make a substantial contribution to the currently underexplored topic of how agroforestry systems perform under climatic stress. As the only climate manipulation experiment on an agroforestry system in sub-Saharan Africa, the only one involving an annual crop, and the only one to examine agroforestry trees at different life stages, this project represents a substantial step toward filling some of the gaps discussed in Chapter 1. It is hoped that the results will not only provide practical information for agricultural climate adaptation in Malawi, but also set an example for similar experiments in the developing world.

2.1.2.2 Elucidating mechanisms and interactions

Beyond simply recording the response of an agroforestry system to a climatic disturbance, it is useful to investigate the underlying reasons for that response. For example, the following questions could provide useful mechanistic information about most agroforestry systems:

• Has a change in one environmental parameter (e.g., rainfall) affected other parameters (e.g., soil moisture and soil temperature)?

• What are the effects of the disturbance on the survival, growth, biomass allocation, and phenology of each species?

• How do the species’ responses vary over time? • Are there below-ground as well as above-ground effects? • Has the balance of competition changed in the system?

The practical goal of asking such questions is to inform appropriate management practices and species selection in agroforestry systems. If an agroforestry system does not perform well under climatic stress, how might its resilience be improved? If the system does perform well, what elements of its performance can be applied to other systems in other locations?

To the extent possible, this dissertation attempts to address these types of mechanistic questions for the particular case of fertilizer trees in southern Malawi under a drought manipulation. (Specific questions and hypotheses are detailed in Chapters 4 and 5.) It is hoped that this approach can help to extend the relevance of the results beyond their immediate context.

2.2 Background on Malawian agriculture

Before describing fertilizer tree agroforestry systems in Malawi, it will be helpful to briefly outline the broader context of Malawian agriculture. The agricultural statistics can best be understood in light of some essential facts on the country’s population and economy.

20

2.2.1 Current statistics

2.2.1.1 Demographic and economic statistics

Malawi is among the world’s least developed nations; its 2011 Human Development Index rank was 171 out of 187 countries (UNDP, 2012). Basic socioeconomic indicators for Malawi are summarized in Table 2-1 and Table 2-2.

Malawi is one of the least urbanized countries in Africa, with only about 20% of the population living in cities; most of the remainder are engaged in agriculture. A large fraction of Malawians are food-insecure, and nearly half of young children experience stunted growth due to chronic malnutrition (National Statistical Office, 2011).

Official statistics for HIV prevalence in Malawi range from 11% to 13% (PRB, 2012), among the highest in the world, though this rate has stabilized and begun to decline. The gradual reversal in HIV infection rates has caused Malawi’s life expectancy, once on a downward trend, to begin increasing again (from 44 years in 2004 to 53 years in 2011).

Malawi’s internal economy is mostly non-monetized. Employment is difficult to measure, as approximately 80% of the population is employed in subsistence agriculture. Poor infrastructure, lack of information, and government disincentives have discouraged the development of internal trade (Dorward and Kydd, 2004), and this poses a barrier to poverty reduction (Ellis et al., 2003).

Table 2-1. Basic statistics on Malawi’s geography and economy.

Geography Economy

Land area 118,480 km2 (of which 24,400 km2 is Lake Malawi) Currency

(2008-2011)

Malawi kwacha (US$1 = MK 142 to MK 159) Annual rainfall 75 cm (SW) to 180 cm (N)

Average annual temperature 16°C (plateau) to 25°C (lowland) GDP (PPP) $14.1 billion (2011)

Shared borders Zambia (NW), Mozambique (SW and SE), Tanzania (NE) Per capita

GDP (PPP) $900 (2011)

Colonial history Independence from Britain, 1964 GDP growth 5.5% (2011)

Government Democratic republic, as of 1994 Foreign debt $1.35 billion

Languages English (official), Chichewa (official), regional languages Exports $922 million (2011)

(tobacco, tea, sugar)

Religions Christianity (83%), Islam (13%), other (2%), none (2%)

Imports $1.69 billion (2011) (food, petroleum, consumer goods) % urban pop 20%; growing at 1% per year

Major cities Blantyre (856,000); Lilongwe (capital, 821,000) Income

distribution Unequal (Gini coefficient = 0.39)

Source: All data from Central Intelligence Agency (2012).

21

Table 2-2. Basic demographic data for Malawi, compared to regional and world averages.

Malawi East Africa average 1 World average

Population 15,883,000 – –

Population growth rate 2.8% 2.7% 1.2%

Birth rate 1 43 38 20

Death rate 1 15 11 8

Infant mortality rate 2 66 60 41

Total fertility rate 5.7 5.1 2.4

Life expectancy (years) 53 57 70

Population density/km2 134 54 52

HIV infection rate 12.9% 5.2% 0.8%

Source: All data from Population Reference Bureau (Ellis et al., 2003). 1 PRB classifies Malawi as part of east Africa rather than part of southern Africa. 2 Per 1,000 population per year. 3 Per 1,000 live births.

2.2.1.2 Agricultural statistics

Export agriculture. Agriculture accounts for more than 90% of Malawi’s foreign exchange (Government of Malawi, 2006). Malawi is the most tobacco-dependent economy in the world; tobacco represents about 70% of the country’s export revenues (Conroy et al., 2006). There are several reasons for the continuing dominance of tobacco: it is well-suited to Malawi’s climate, has high labor and low land requirements, and has accumulated considerable infrastructure and political support.

However, in light of projected long-term declines in world tobacco demand, diversification seems increasingly necessary. Malawi’s other agricultural exports include tea, sugar, coffee, cotton, paprika, and macadamia nuts, but none of these are currently in a position to replace the export earnings of tobacco.

Crops for export are generally grown on large agricultural estates (see Section 2.2.2.2). Irrigated land is used only for high-value crops: it occupies only about 0.5% of Malawi’s cultivated area, but it accounts for 10% of all agricultural exports (Mkandawire, 1999).

Subsistence agriculture. Maize provides two-thirds of calories consumed in Malawi (Byerlee and Heisey, 1997), making Malawi one of the world’s most maize-dependent countries. Although much of Malawi is climatically suitable for maize, the crop’s high nutrient demands have gradually degraded soil fertility in areas where neither fertilizer nor fallow are viable options (Young, 2005). High-yielding maize varieties have been developed for the region, but many farmers do not use them (Byerlee and Jewell, 1997), and they often do not perform well in the absence of external inputs.

22

Figure 2-1. The Majoni family (Makoka, Malawi, January 2010) fertilizing a field of intercropped maize and vegetables typical of a smallholder farm.

Figure 2-2. A close-up of the field in Figure 2-1. Clockwise from lower left: maize; groundnut; pumpkin; common bean.

Maize and tobacco are often intercropped in Malawi, either with each other or with other food crops (Figure 2-1 and Figure 2-2). Intercropping conveys several advantages for Malawian smallholders (Shaxson and Tauer, 1992): it allows sequential decision-making (with regard to rainfall, availability of inputs, and other variables) as the growing season progresses, and it allows staggered harvesting as different crops are ready at different times. The latter is especially important when labor is limited. Shaxson and Tauer (1992) report that in a survey of six villages in southern Malawi, 43 crops were recorded, 95% of farms used intercropping, and 85% of maize was intercropped.

2.2.2 History of agriculture in Malawi

2.2.2.1 Pre-colonial introduction of maize

Maize (Zea mays) is native to Central America and was introduced to southern Africa by Portuguese traders perhaps in the 17th century (Miracle, 1966); the details of its introduction and dissemination are lost to history. Its name in Chichewa, Malawi’s dominant native language, is chimanga, which literally means “from the coast” (McCann, 2005). The high yields of maize per unit of land and labor gave it great appeal, and despite its relatively poor storage characteristics and its intolerance of drought and nutrient deficiencies, it came to dominate or displace native crops such as sorghum and millet. Its prevalence in Malawi is exemplified by the Chichewa saying “chimanga ndi moyo,” or “maize is life” (McCann, 2005).

2.2.2.2 The agricultural estate system

Under the leadership of Hastings Kamuzu Banda, Malawi’s first president (from 1966 to 1994), Malawi’s agricultural estates – chiefly tobacco, tea, sugar, cotton, and coffee – were well supported, but conditions did not improve for ordinary Malawians (Conroy et al., 2006). In fact,

23

smallholders were legally prohibited from growing these high-value cash crops until the transition to democracy in 1994. Most smallholders still do not grow any cash crops, and estates produce the vast majority of agricultural exports (Mkandawire, 1999).

The agricultural estate system has had broader social effects as well. Poor Malawians often forgo working on their own land to work as laborers for wealthy farmers or estate owners. This agricultural wage labor, or ganyu, is a crucial source of livelihood for much of the population (Dorward and Kydd, 2004). Unequal land distribution has contributed to Malawi’s highly unequal distribution of income.

2.2.3 Challenges to agricultural production in Malawi

Despite Malawi’s overwhelming dependence on agriculture – both in the formal and informal sectors – Malawi faces a variety of challenges to profitable and sustainable agricultural production. These have been reviewed by Malawi’s Ministry of Agriculture and Food Security (2006) and are briefly outlined here.

2.2.3.1 Poor soil fertility and lack of access to inputs

Intensive cultivation and inadequate nutrient replenishment have already degraded soil fertility in many parts of Malawi, and the trend is likely to continue in the near future. The Malawi Government (2001) estimates the country’s average annual rate of soil erosion at 29 tons per hectare, equal in value to 8% of Malawi’s GDP. This problem is worsened by high rates of deforestation (Figure 2-4) and by the widespread practice of cultivating steep slopes (Figure 2-3) due to shortages of available land (see Section 2.2.3.3).

Figure 2-3. Smallholder maize cultivation on a steep slope (Makoka, Malawi, December 2009).

Figure 2-4. Felling of trees in miombo forest for use as fuelwood (Makoka, Malawi, October 2010).

24

As of 2006, only 72% of Malawian farmers were using any kind of soil fertility management practices; in some districts, the figure was as low as 20% (Government of Malawi, 2007). The recent reinstatement of the fertilizer subsidy may improve these statistics; in 2008-2009, 65% of Malawian families received a fertilizer coupon (Dorward and Chirwa, 2011). However, inorganic fertilizer is not a complete solution; its full benefits can only be realized in combination with organic inputs and conservation tillage (Government of Malawi, 2006).

2.2.3.2 High rainfall variability

Rainfall in southern Africa is characterized by high interannual variability, in part due to the pronounced drying effect of El Niño (Cane et al., 1994; Hachigonta and Reason, 2006). For example, the entire region suffered from severe droughts in 1991/1992, affecting over 100 million people and necessitating large amounts of food aid. Conversely, much of Mozambique experienced devastating floods in 2000, and Malawi was hit by flooding in the following year (Devereux, 2002a), resulting in several-fold reductions in maize yield. Chapter 6 reviews predictions of climate change in Malawi and how it may affect food production.

2.2.3.3 Population density and small farm size

The average household in Malawi currently has less than one hectare of land, and in most villages, plots are subdivided with each successive generation (Ellis et al., 2003). Lack of available land is exacerbated by unequal land distribution (Smale and Heisey, 1997). In Malawi’s more densely populated southern region, the average landholding is approximately half a hectare, close to the minimum limit of what is necessary for food self-sufficiency.

The scope for agricultural extensification in Malawi is limited: according to the Malawi Government (2001), 31% of Malawi’s land area is arable, but the area actually under agriculture is 49%. In other words, 18% of Malawi’s land surface is already inappropriately cultivated. Young (2005) describes Malawi’s situation as follows: “Virtually all land that can be sustainably cultivated is being farmed, as well as substantial areas where continued cultivation will inevitably lead to land degradation.”

2.2.3.4 Economic and social challenges

Biophysical conditions are not the sole, or even usually the most important, causes of food insecurity (DeRose et al., 1998; Sen, 1999). Poverty undermines the power to purchase food and the inputs needed to produce it. In much of rural sub-Saharan Africa, poverty is perpetuated by lack of access to markets, lack of available credit, inadequate and deteriorating infrastructure, poor quality of services such as education and health care, and unfavorable global market conditions (Jayne et al., 2006). Lack of transport infrastructure is a particular problem in Malawi (Government of Malawi, 2006); this is due not only to inadequate investment but also due to the fact that Malawi is landlocked and mountainous.

In Malawi, as in most of southern Africa, a large and increasing minority of the working-age population is infected with the AIDS virus. de Waal and Whiteside (2003) suggest that the AIDS epidemic is leading to the emergence of “new variant famine,” in which communities are more easily harmed by, and less able to recover from, biophysical or economic setbacks to food production. Many argue that AIDS increases the dependency ratio and reduces available labor;

25

others (Jayne et al., 2006) dispute this. Combating the AIDS epidemic is a centerpiece of government policy in Malawi and is frequently mentioned in agricultural planning.

2.2.4 Fertilizer subsidy: history, successes, costs

Malawi has an erratic history of subsidy programs for fertilizer and other agricultural inputs. A full chronology of these programs appears rather bewildering: for example, fertilizer price subsidies were reduced in 1984-86, increased from 1987 to 1993, reduced again in 1995-96 and increased in 1998-2000 (Akinnifesi and Kwesiga, 2002). These oscillations were largely a result of structural adjustment programs in the 1980s and 1990s intended to reduce Malawi’s massive external debt (Conroy et al., 2006).

Recently, Malawi has implemented several input subsidy schemes that have generally been regarded as successful; these are briefly described below.

2.2.4.1 Starter Pack Scheme and Targeted Input Programme, 1998-99 to 2004-05

The Starter Pack Scheme began in 1998 and continued with some modifications and gaps until 2005. It provided each farm family with enough high-quality inputs to plant approximately 0.1 hectares: 5 kg of urea fertilizer, 2 kg of hybrid maize seed, and 2 kg groundnut or pigeonpea seed (Akinnifesi and Kwesiga, 2002). This policy was “not intended to be a social protection program, but rather to kick-start agricultural development” (Dorward and Chirwa, 2011).

2.2.4.2 Farm Input Subsidy Programme (2005-06 to present)

The Farm Input Subsidy Programme (FISP; previously known as the Malawi Agricultural Input Subsidy Programme) built upon the success of the Starter Pack Scheme, but expanded its reach and goals in hopes of achieving food security at a national level. By many accounts, it has succeeded. Denning et al. (2009) refer to this program as “one of the most ambitious and successful assaults on hunger in the history of the African continent.”

Under the current scheme, about 60% of Malawian farm households (those with low or moderate incomes) receive a coupon to purchase a 50 kg bag of fertilizer at a price of 500 Malawi kwacha (about US$3 at current exchange rates), which represents a discount of approximately 90%. It appears that, even when accounting for years of favorable rainfall, this intervention has increased national maize yield by about 30% (Dorward and Chirwa, 2011).

Despite its overall success, FISP has suffered from notable implementation challenges. An unknown number of the coupons are sold to non-eligible households, undermining the program’s goals of provisioning the poorest farmers. The financial burden for the Malawian government is also considerable: in 2008/09, the total cost of the subsidy was US$242.3 million, which represented 16% of Malawi’s national budget and 6.6% of GDP (Dorward and Chirwa, 2011). Fertilizer prices were unusually high that year, and under normal conditions, the program appears to have satisfactory economic returns (Dorward and Chirwa, 2011).

Does the fertilizer subsidy program obviate the need for soil fertility restoration through agroforestry? The consensus seems to be that agroforestry and other organic inputs complement rather than duplicate the use of inorganic inputs (Government of Malawi, 2006). Even with the

26

subsidy, many households are unable to obtain enough fertilizer to apply at optimal rates. Furthermore, fertilizer use efficiency is generally higher when fertilizer is used in conjunction with organic inputs (Akinnifesi et al., 2007; Dorward and Chirwa, 2011). The existence of an effective fertilizer subsidy program reduces the short-term urgency of agroecological soil fertility restoration, but does not supplant its long-term benefits.

2.3 Agroforestry research in southern Africa

2.3.1 History and goals of SADC-ICRAF

Agroforestry research in southern Africa has been conducted by a variety of governmental and non-governmental organizations. The most prominent of these is SADC-ICRAF (the Southern Africa Development Community regional office of ICRAF). ICRAF, the International Centre for Research on Agroforestry, is now called the World Agroforestry Centre but is still frequently referred to by its former acronym. For this project, the author benefitted from official affiliation with ICRAF-Malawi as a Ph.D. researcher.

Since its inception in 1987, SADC-ICRAF has pursued several main goals for the benefit of smallholder farmers (Kwesiga et al., 2003):

• Soil fertility improvement and food security (through the use of fertilizer trees and soil and water conservation)

• Income generation and poverty alleviation (through improved crop yields and the sale of tree products such as timber, fruit, and medicine)

• Conservation of natural resources and biodiversity (through reduced pressure on forests and domestication of indigenous trees)

All these goals are interconnected, but this project focuses primarily on the first goal of restoring soil fertility through the use of fertilizer trees. As mentioned in Section 2.2.4, there are ways to achieve this goal that do not involve agroforestry; this effort should be seen as one of a variety of complementary methods to promote food security on Malawian farms.

2.3.2 Experiments on fertilizer trees

Due to constraints on farm size, soil fertility, and agricultural inputs, agroforestry research in Malawi has devoted much attention to “fertilizer trees” (Kwesiga et al., 2003) – leguminous trees that are grown to provide nitrogen to the maize crop. A variety of different tree species have been used in a variety of configurations.

2.3.2.1 Tree species used

The agroforestry species that has generally produced the greatest increase in maize yields is Sesbania sesban, a fast-growing leguminous tree endemic to Africa (Kwesiga et al., 1999; Ikerra et al., 2001). It has two main drawbacks, however: first, it cannot be directly seeded onto field plots, but must be first cultivated as seedlings in a nursery; and second, it cannot be coppiced, so it is not suitable for hedgerow intercropping.

27

Other leguminous trees that have produced good results include Tephrosia vogelii (Mafongoya et al., 2003), which is also known locally as a useful fish poison (ICRAF, 2006), Tephrosia candida, which is similar to T. vogelii but faster-growing and less toxic (Mafongoya et al., 2003), and Gliricidia sepium (Chirwa et al., 2007), which if unpruned grows to a medium-sized tree with lilac flowers favored by honeybees. Gliricidia coppices well and is suitable for use in hedgerows, while the coppicing ability of Tephrosia has very limited.

Leucaena leucocephala, a species used for soil restoration and cattle forage throughout the world, has also been tested in southern Africa (Mafongoya, 2006), but has usually not performed as well as the above species. Native mature Acacia albida trees, mentioned in Chapter 1 for their microclimatic benefits in the Sahel, have also long been valued on southern African farms (Saka et al., 1994), though their use in this region has not been well studied.

Cajanus cajan, or pigeon pea, is a perennial shrub that produces an edible seed and is widely used throughout the region, especially in Malawi (Chirwa et al., 2007; Sirrine et al., 2010). It is primarily valued for its use as a supplementary food crop, and is not generally considered a “fertilizer tree,” but can also provide some of the same soil fertility benefits as the tree species described above.

These trees (summarized in Table 2-3) can be used in a variety of designs, although not every tree can be used with every design. Following are descriptions of three of the most common system designs for fertilizer trees in southern Africa.

Table 2-3. Some characteristics of fertilizer tree species used in Southern Africa.

Species Sesbania sesban Tephrosia vogelii

Tephrosia candida

Gliricidia sepium

Extent of current use Limited Moderate Moderate Limited

Native to Africa Tropical Africa Tropical Africa C. America

Propagation Transplants Direct seeding Direct seeding Transplants

Coppicing No No No Yes

Auxiliary uses Fodder Fish poison, insecticide Fuelwood Fodder, honey,

timber

Growth form Small tree Shrub Shrub / small tree Medium tree

Max rooting depth (m) >4 m no data available no data available 5.6 m

Sources: (Akinnifesi et al., 2004; Cook et al., 2005; ICRAF, 2006)

2.3.2.2 Types of fertilizer tree systems

Refer to Figure 2-5 for conceptual diagrams of the systems described in this section.

28

(a) Improved fallows

(b) Relay intercropping

(c) Hedgerow intercropping

Figure 2-5. Temporal schematics (annual or multi-year) of three different types of fertilizer tree systems intercropped with maize.

29

Improved fallows. Improved fallows, like natural fallows, involve a cessation of cropping on the land for one or more years in order to restore soil fertility. However, unlike in natural fallows, the fallowed land is intentionally planted with a nitrogen-fixing tree species. Usually the fallows are allowed to grow for two years, to be followed by two or three years of continuous maize monoculture. Improved fallows can result in such drastic yield increases that, even taking into account the fallow years, they can increase net maize yields by 50-100% or more (Kwesiga et al., 1999). They also have the additional benefits of providing firewood, timber, fodder, and weed suppression. However, they are not generally suitable for small farms. Opinions on their potential for broad application range from optimistic (Sanchez, 1999) to pessimistic (Hauser et al., 2006).

Relay intercropping. This system is mostly used in Malawi, where high population density makes it infeasible for farmers to take land out of production. In relay intercropping, tree seedlings are sown with the maize, grow up underneath it, and finish maturing after the maize senesces with the end of the rainy season in early fall (Kwesiga et al., 2003). The trees are allowed to continue growing for a total of 8-9 months; they are then cut and their leafy biomass is incorporated into the soil shortly before the next round of maize planting. The soil fertility benefits and wood production are usually of lesser magnitude than in improved fallows (Ikerra et al., 2001) but can still be significant.

Hedgerow intercropping. Hedges of the desired tree species are cultivated between rows of the maize crop in perpetuity (for up to several decades, depending on the species). Aboveground biomass is pruned once or several times a year and added to the plot (Chirwa et al., 2007); some species, e.g., Gliricidia, can be lopped off at the ground and will resprout vigorously (Hauser et al., 2005). Names for this system vary somewhat; Kwesiga et al. (2003) call it simply “intercropping,” as distinct from hedgerow intercropping, which they consider synonymous with alley cropping. Terminology notwithstanding, its main distinguishing feature is the severe pruning of the hedgerows.

2.3.3 Adoption and livelihood impacts of fertilizer trees

The benefits of fertilizer trees in southern Africa have been reviewed by Akinnifesi et al. (2008). In summary, they can contribute significantly to farm livelihoods, not only through greatly improved maize yields but also through sale of the auxiliary products described above (Quinion et al., 2010). However, constraints of knowledge, labor, and germplasm availability have limited their use thus far (Ajayi et al., 2003). Despite these challenges, Ajayi et al. (2011) are optimistic about their future prospects for further adoption.

Although fertilizer trees are often adopted by a large fraction of the farmers who receive extension assistance to test them (Keil et al., 2005), they are not yet widespread in southern Africa overall, with adopters numbering roughly 400,000 in 2008 (Akinnifesi et al.). A recent initiative in Malawi, the Agroforestry Food Security Programme (funded by Irish Aid), may have added as many as 147,000 new users of fertilizer tree systems from 2007 to 2011 (Ajayi et al., 2011). However, accurate numbers are difficult to come by, as farmers may cease to use the technology in the absence of continued support.

30

Agronomically, there exists great adoption potential for all three of these systems in southern Africa (Kwesiga et al., 2003); however, as with any new agroforestry technology, adoption is slowed by the complexity of the practice, the lack of information and materials, and farmers’ concerns about risk and uncertainty (Mercer, 2004).

2.4 Gaps filled by the current project

As described above, a great deal of research has already been done on the performance of fertilizer tree systems in Malawi. This project intends to fill several specific gaps within the substantial body of existing literature.

2.4.1 Examining crop and tree response to controlled drought

Although several previous studies have examined soil water dynamics in fertilizer tree systems (Phiri et al., 2003; Sekiya and Yano, 2004; Chirwa et al., 2007; Sileshi et al., 2011), none have attempted a drought manipulation. This project intends to go beyond these observational studies to more definitively establish cause-and-effect relationships between water stress and performance of fertilizer tree systems.

2.4.2 Comparing mature trees and seedlings

The topic of seedling establishment is not emphasized in the fertilizer tree literature. Although a variety of obstacles to successful establishment have been recognized, including drought, disease, fire, insects, and browsing (Sileshi et al., 2008), these effects are usually not quantified. When a fertilizer tree system is compromised by poor establishment, this tends to be mentioned in passing (e.g., Kwesiga et al., 1999) rather than being the center of inquiry.

The present study attempts to address this lack of information by examining the effects of drought and late planting on seedlings of two different fertilizer tree species (Tephrosia candida and Gliricidia sepium). It also presents an opportunity to compare the drought response of newly planted Gliricidia seedlings with that of mature Gliricidia trees.

2.5 Outline of experimental methods

The following section outlines the overarching approach and methodological context of this project, while the details of each experiment are described in Chapters 3 through 6.

2.5.1 Makoka Agricultural Research Station

In empirical agroforestry research, there are tradeoffs between on-farm trials (with the cooperation and participation of farmers) versus trials on agricultural research stations. The former have the advantage of realistic conditions, but they also have greater logistical difficulty and uncontrolled variables. The latter are easier to set up, monitor, and control, but may not reflect the conditions experienced by farmers on the ground. As this study was primarily biophysical, and as precise control and measurement were considered essential, it was conducted at Makoka Agricultural Research Station (Figure 2-6).

31

Makoka Agricultural Research Station (located approximately 20 km southeast of Zomba, the former capital of Malawi) was established by a joint effort of the British government and the Malawian government in 1967. Since then, it has played an important national role in varietal and management trials of both subsistence crops and cash crops, including cassava, cotton, pigeonpea, and maize. SADC-ICRAF has conducted agroforestry trials at Makoka since the early 1990s, and the presence of established agroforestry trials was a major reason for choosing to carry out the current project there. Makoka’s location and climate are summarized in Table 2-4, Figure 6-2, and Figure 6-3.

Table 2-4. Characteristics of Makoka Agricultural Research Station, Malawi.

Latitude / longitude 15º30’ S, 35º15’ E

Average annual temperature 10º-21ºC win; 19º-34ºC sum

Average annual rainfall 1024 mm (range 560-1600)

Elevation 1030 m

Soil texture 52% sand, 37% clay

Soil type Oxic Haplustalf

Reference (Ikerra et al., 2001; Chirwa et al., 2007)

2.5.2 Philosophy of three-year field experiment

Three years is a short time in an ecosystem context. Ecosystem climate manipulations (e.g., de Valpine and Harte, 2001; Suttle et al., 2007) often take several years, if not longer, to reach a new equilibrium in terms of species composition and ecosystem function. Often, the change in species composition is, itself, an essential factor causing the change in function. Thus, allowing time for equilibration of species composition change is essential for getting useful data from a natural ecosystem. Agricultural systems do not have this feature, so a brief manipulation is more relevant in an agroecosystem than it would be in a natural ecosystem.

Although three years is nevertheless a shorter time span than would be ideal for an agricultural experiment, especially in a location with considerable interannual variability, it represents a necessary compromise between theoretical and practical considerations. Accordingly, some caution must be used in generalizing these results. It should be noted that all three years of this project were characterized by average and adequate rainfall, though the timing of rainfall in 2009-2010 led to some drought stress at the beginning of the growing season. Different results might have been obtained in years that were much wetter or drier.

2.5.3 Allocation of field tasks

These experiments would have been impossible without the help of a large number of people at Makoka Research Station. The participation of various contributors to different elements of the experiments is described below.

32

Although three years is nevertheless a shorter time span than would be ideal for an agricultural experiment, especially in a location with considerable interannual variability, it represents a necessary compromise between theoretical and practical considerations. Accordingly, some caution must be used in generalizing these results. It should be noted that all three years of this project were characterized by average and adequate rainfall, though the timing of rainfall in 2009-2010 led to some drought stress at the beginning of the growing season. Different results might have been obtained in years that were much wetter or drier.

2.5.4 Allocation of field tasks

These experiments would have been impossible without the help of a large number of people at Makoka Research Station. The participation of various contributors to different elements of the experiments is described below.

Figure 2-6. Google Earth satellite image of Makoka Agricultural Research Station. Label (1) shows the location of MZ12 (15º31’08.90” S, 35º13’25.44” E), while label (2) shows the location of Nkula Field (15º31’13.55” S, 35º13’49.43” E). Image date is 1973, but field station layout remains similar today. Image is from http://gismap.ciat.cgiar.org/MarkSimGCM. Inset map shows location within Malawi.

!1

2

33

For all routine tasks (land preparation, maize planting, weeding, fertilizing, tree pruning, and maize harvest), teams of field workers were hired from the local community. Depending on the urgency and difficulty of the task, these teams generally numbered between four and twenty people. Field workers included both men and women, all of whom were farmers and many of whom had participated in other ICRAF projects at Makoka.

The author was present for, and participated in, all field maintenance tasks. Communication regarding field tasks was conducted in a mixture of basic Chichewa (on the part of the author) and basic English (translated for the non-English-speaking team members by one or more English-speakers). All workers were paid the same prevailing daily wage (MK 250 per day).

The rain shelters (described in Chapter 3) were conceived and designed by the author with input from Makoka’s carpenter, Damson Singo. The author procured all materials, determined the precise location of each shelter, and oversaw construction and maintenance, but the actual construction was carried out by Mr. Singo and his assistants.

Both experiments were continuously overseen by day and night watchmen, who ensured that the crops and the structures were not damaged by either humans or animals, and who notified the author when any problems arose.

All environmental and population data (including soil samples, light and temperature readings, pan evaporation, litterbags, seedling survival and seedling heights, maize phenology, maize height, and maize plant counts) were collected and analyzed by the author. Data from maize and tree biomass harvests, due to the labor-intensity of these processes, were collected by a small team of workers (including ICRAF employees Chiku Kwakwala and Konisaga Mwafongo as well as several experienced casual laborers). However, harvest data collection was overseen and verified by the author.

During the experiment’s third year (2010-2011), the author was only present in Malawi through the end of October. From November 2010 until May 2011, all field maintenance tasks and data collection were overseen by Chiku Kwakwala (based at Makoka) and Simon Mng’omba (based at Chitedze Agricultural Research Station near Lilongwe). To facilitate consistency of methods between years, the author provided written protocols for field tasks, including detailed data sheets for all harvest activities. Frequent communication by email helped to resolve any contingencies.

The contributions of all these individuals are gratefully acknowledged. This dissertation project is a team effort, and the hours invested by the author are a small fraction of those invested by the community at Makoka.

2.5.5 Scope, implications, and limitations of this work

In light of the complexity of the factors influencing food security in Malawi, it may seem relatively trivial to study the performance of a particular agricultural technology in isolation from its social and economic context. The biophysical performance of an agricultural system is only one aspect of its overall effectiveness (Giampetro, 2004).

34

Nevertheless, a successful agricultural system cannot exist without successful technologies. As Sayer and Campbell (2004, p. 57) write, “High-technology research on the components of agricultural systems is still vital, but it has to be constantly reviewed to ensure that it is correctly placed in the context of changing local biophysical and socio-economic conditions.” Thus, this thesis attempts to acknowledge the inherent limitations of the project and to present its findings in a way that is accessible across disciplines.

At a broad level, this research project is expected to provide useful information to governments, NGOs and multilateral organizations interested in climate adaptation in the developing world. In addition, this project is likely to be of interest to the agroforestry research and modeling community as only the second example of a climate manipulation experiment on an agroforestry system.

The most important goal of this project is to benefit smallholder farmers in southern Africa, many of whom face major challenges in meeting their food security needs under current climate variability and are especially vulnerable to future climate change. This dissertation will have achieved its goal if it can make even a small contribution to the urgent task of enabling sustainable agricultural development in a changing climate.


Portions of this chapter were previously submitted as a prospectus for the author’s doctoral qualifying exam (November 2007). The prospectus material was greatly improved by comments and suggestions from the exam committee members (Dan Kammen, chair; Margaret Torn; Lynn Huntsinger; Todd Dawson; and Carol Shennan, UC Santa Cruz).

Other portions of this chapter were previously submitted as a final paper entitled “Disappearing forests in Malawi: Causes and solutions” in fulfillment of the requirements of EEP 153 (May 2005), taught by Robin Marsh. Her feedback and guidance are gratefully acknowledged.

35

Chapter 3. Design, construction, and performance of rain exclusion shelters

Abstract

To date, there have been few climate manipulation experiments in developing countries, in large part due to prohibitive costs. This chapter describes rain exclusion shelters designed for use on intercropping systems of maize and leguminous trees at Makoka, Malawi. The stature of these systems (2-3 m) makes rain manipulation challenging: the crops are too short for below-canopy exclusion, but too tall to easily be covered by other designs described in the literature.

The goal of this project was to create a novel rain shelter design that used low-cost, locally available materials, employed local labor and expertise, and incorporated crops up to 3 m in height, while not interfering with photosynthesis or microclimate. The final design was an open-sided structure with a pitched roof. Materials used included eucalyptus poles, cured pine timber, and clear polyethylene greenhouse sheeting. The shelters were built efficiently by local carpenters using local materials.

The shelters effectively diverted rain (excluding a 1-m buffer inside the shelter to eliminate edge effects). After four weeks of complete rain exclusion, gravimetric soil moisture was 39% lower than in control plots at 0-20 cm depths (9% vs. 15%) and 29% lower at 20-40 cm depths (12% vs. 17%). Unintentional effects of the shelters were minimal; a slight moderating effect on air T was observed, with a net increase of ~0.1°C, while no significant effect was seen on pan evaporation. The shelters blocked up to 33% of PAR (photosynthetically active radiation); further investigation is needed to determine whether the changed light conditions affected plant growth, and, if so, to revise the design accordingly.

The shelters achieved their goals of low cost and acceptable durability while performing well in terms of rain interception and soil moisture reduction. This design or similar designs may prove useful in tropical agricultural research, especially in developing nations where research budgets are limited but understanding climate change impacts is a high priority.

3.1 Introduction and motivation

Over the past several decades, experiments examining the impact of climate change on terrestrial ecosystems have become increasingly sophisticated. A number of recent experiments have successfully manipulated multiple global change factors in a field setting – such as temperature, precipitation, atmospheric CO2 concentration, nitrogen deposition, and species composition. The basic principles of such field experiments are reviewed by Shen and Harte (2000). Well-known multifactoral climate manipulation experiments include Jasper Ridge in California (Zavaleta et al., 2003), SoyFACE in Illinois (Bernacchi et al., 2007), CLIMAITE in Denmark (Carter et al., 2011), and BACE in Massachusetts (Hoeppner and Dukes, 2012).

36

In accordance with the availability of funding and expertise, the vast majority of climate manipulation experiments (and ecological field studies in general; Martin et al., 2012) have been carried out in temperate developed countries. Tropical ecosystems, especially subsistence-focused agroecosystems, have been neglected by comparison. In light of predictions that agriculture in many parts of the developing world will suffer significant negative consequences as a result of climate change (e.g., Lobell et al., 2008; Schlenker and Lobell, 2010), there is a pressing need to gather more empirical data on how tropical agroecosystems will respond to climatic stress.

Climate adaptation plans for agriculture in developing countries have so far mostly relied upon process-based crop simulation models (e.g., Jones and Thornton, 2003), socioeconomic surveys (e.g., Thorlakson, 2011), and proxies such as historical climate data (e.g., Burke et al., 2009). Controlled field experiments in which one or more climate variables are manipulated are very rare in the developing world. However, climate manipulation experiments can often address questions of causality and multi-factor interactions more directly than can modeling or proxy approaches.

Affordable designs for climate manipulation infrastructure, employing local labor and locally obtainable materials, could allow experimental approaches to be more widely used in developing countries. Such experiments could generate a wealth of locally relevant, practical information about crop response to climate change in vulnerable regions.

The goal of this study was to develop and deploy inexpensive rain manipulation infrastructure appropriate for use in agricultural systems in sub-Saharan Africa. When this project was conceived in October 2007, no similar studies could be found in the scholarly literature; this gap persists at the time of writing (August 2012). All previously described rain shelter designs (section 3.1.2) were outside the budgetary scope of a typical developing-country research station, and furthermore, most would not have been appropriate for maize or other field crops of similar stature (approximately 2 m high). Therefore, the successful design and deployment of an affordable climate manipulation experiment had potential to serve as a useful example for future efforts.

Our variable of interest for this study was precipitation, due to concern over current – and possibly increasing – variability in growing-season precipitation over southern Africa (Tadross et al., 2009). However, it should be emphasized that low-cost methods to manipulate other variables such as temperature and CO2 are also needed.

3.1.1 Intrinsic challenges of manipulating precipitation

Even without budgetary constraints, there are many potential difficulties in successfully executing a rain exclusion experiment. Design flaws can hamper the intended interception of rainfall and reduction in soil moisture; furthermore, rain manipulation infrastructure can have unintended consequences on light, microclimate, and other biophysical variables.

3.1.1.1 Intercepting target amount, versus achieving a realistic temporal pattern

Climate change may affect the frequency, duration, clustering, and/or intensity of rainfall (Dore, 2005), any of which can be manipulated in an experimental setting with or without changing

37

total rainfall (Knapp et al., 2002). The effects of a change in total rainfall can depend greatly on the timing of that change (Jentsch et al., 2007), and one of the greatest challenges in rain manipulation is ensuring that plots maintain a meaningful temporal pattern of rainfall.

A variety of approaches have been used for the manipulation of rainfall amount and/or timing. Existing rain shelter designs (reviewed in section 3.1.2) include fixed-interception slats, variable-interception slats, a retractable roof that deploys during rain events, or a complete exclosure with watering supplementation. The appropriate design depends on the question at hand; for example, the latter design (complete exclosure plus watering) permits full control over any aspect of precipitation timing or amount, whereas the former design (fixed-interception slats) controls only the fraction of precipitation intercepted.

3.1.1.2 Difficulty in controlling all aspects of plant-water relations

Real-life drought events are often associated with other climatic conditions, such as high temperature, low humidity, and high solar radiation. Altering rainfall in isolation may fail to capture important additive and interactive effects of these correlated variables.

Furthermore, care must be taken to control surface and below-ground water flow in drought plots without introducing confounding effects such as root damage or soil disturbance. This becomes a greater challenge when working with deep-rooted perennial plants (Hanson, 2000).

3.1.1.3 Unintended consequences

Rain interception infrastructure can inadvertently modify many other aspects of the immediate environment, such as radiation, humidity, and wind speed (Hanson, 2000). In addition, the process of constructing and operating the infrastructure may cause soil disturbance and compaction, leaching of chemicals from the building materials, and physical damage to plants.

3.1.2 Existing rainout shelter designs

The basic principles of rain manipulation have been reviewed by Hanson (2000), while Miranda et al. (2011) summarize some notable recent experiments. Following is a brief overview of designs described in the literature.

3.1.2.1 Shelters with fixed or removable roofs

Recent studies using rainout shelters with solid, non-removable roofs include Fay et al. (2000), in a tallgrass prairie in Kansas, USA, and English et al. (2005), in a semi-arid grassland in Arizona, USA. With this type of shelter, irrigation must be applied under the roof to prevent the complete demise of the experimental plants. Both of the aforementioned experiments used large shelters (9 m × 14 m and 9 m × 18 m, respectively) and applied several different irrigation regimes to examine the effects of precipitation variability.

A different approach is to use shelters with solid roofs that can be removed when not needed, permitting some rain to enter the plots. To study drought effects on pigeonpea in India, Nam et al. (2001) used shelters with similar dimensions to the above, but with manually removable roofs constructed from clear polyethylene sheets. Zhu et al. (2011) used a somewhat different approach for a drought study on maize in China: the crop was grown in a series of 1.6-m deep

38

troughs, which were manually covered by mobile rolling roofs during rain events. The CLIMAITE experiment in Danish heathland (Carter et al., 2011) uses a more sophisticated design with clear plastic curtains that are automatically deployed by a rain sensor.

3.1.2.2 Fractional interception with slats or gutters

Yahdjian and Sala (2002) describe a basic rainout shelter design to intercept variable amounts of precipitation using transparent acrylic V-shaped gutters of varying widths. However, their shelters – designed for an Argentinian shrubland – are much smaller (2 m × 1.9 m) and of shorter stature (0.5 m high) than the others mentioned above. More recently, Miranda et al. (2011) describe a similar design (2 m × 3 m; 1.1 m high) for a desert shrubland in Spain: opaque polycarbonate gutters, controlled by rain sensors, tilt horizontally to intercept rain or vertically to allow sunlight into the plot. West et al. (2012), in the South African fynbos, use a different approach: clear polycarbonate louvers over the plots (4 m × 4 m) can be closed to exclude all rain or opened to admit all rain. In some plots, the louvers are kept permanently open to control for non-rainfall effects of the shelters.

3.1.2.3 Below-canopy interception of throughfall

For vegetation taller than a few meters, enclosing the canopy is difficult. An alternative approach is to intercept precipitation below the canopy, which has been done successfully in several field experiments. Nepstad et al. (2007) used nearly 6,000 plastic panels to exclude 60% of the throughfall in a 1 ha plot of rainforest in the Brazilian Amazon. More recently, Schwendenmann et al. (2010) used clear polyethylene sheeting mounted on 0.5 m × 5 m bamboo frames to intercept throughfall in a cacao agroforest (Theobroma cacao with an overstory of Gliricidia sepium) in Sulawesi, Indonesia. The work of Schwendenmann et al. is the only other rainfall manipulation experiment on an agroforestry system that could be found in the literature at the time of writing (August 2012).

3.1.2.4 Watering experiments

Watering experiments, rather than drought experiments, are especially useful in locations where increased precipitation is more likely (or of greater interest) than decreased precipitation. Two recent examples, both from California grasslands, are Zavaleta et al. (2003) and Suttle et al. (2007). To avoid unintended effects from the chemical composition of the water, watering treatments are usually done with rainwater that is collected on-site and then applied to the plot. It is possible to combine watering and drought treatments in the same experiment, in which case rainwater collected from the latter treatment can be applied to the former treatment.

3.1.3 Goals of this project

This project had two main goals, a methodological goal and an investigative goal. The methodological goal was to design, deploy, and evaluate low-cost rain shelters suitable for use on tall field crops (e.g., maize) in sub-Saharan Africa. All of the existing designs described above were impractical for this application in terms of cost, materials, and/or stature; therefore, a novel design was needed.

39

The investigative goal (described in Chapters 4 and 5) was to evaluate the effects of a severe drought on maize-legume agroforestry systems. This chapter only reports results relevant to the performance of the rain shelters; crop response to drought is reported in Chapters 4 and 5.

3.2 Methods

3.2.1 Design of rain shelters

The final design was an open-sided structure with a pitched roof (Figure 3-1 and Figure 3-2). Following is a complete list of materials used and their sources:

• Hardwood (eucalyptus) poles for vertical supports (local merchants) • Cured softwood (pine) timbers for horizontal supports (Wico Sawmill, Zomba, Malawi) • Greenhouse sheeting (clear polyethylene sheets) (Polypack Ltd., Blantyre, Malawi) • 3-inch and 4-inch nails (Macsteel, Zomba, Malawi) • 2 mm galvanized steel wire (Macsteel, Zomba, Malawi) • Black polyethylene sheeting for termite protection (local merchants)

The blueprints of the final design are shown in Figure 3-3 (the smaller design) and Figure 3-4 (the larger design). To prevent stress at attachment points, greenhouse sheeting was attached to the roof pressed between two timbers: the lower timber was part of the roof frame, while the upper timber was fully detachable. Figure 3-5 shows the method of roof attachment.

40

Figure 3-1. An aerial view of several completed rain shelters (the smaller design).

Figure 3-2. Landscape context for the 18 shelters at the seedling experiment (the larger shelter design).

41

(a)

(b)

Figure 3-3. Blueprints for the smaller shelter design: (a) flat side; (b) peaked side.

42

(a)

(b)

Figure 3-4. Blueprints for the larger shelter design: (a) flat side; (b) peaked side.

43

The initial shelter design used 5” bolts, each with two washers and a wing-nut, to attach the upper timbers to the lower timbers (Figure 3-5). The purpose of the bolts was to facilitate quick removal of the roofs. Unfortunately, at $0.89 per bolt assembly, this hardware proved prohibitively expensive ($34 for each small shelter and $60 for each large shelter, which represented 10-15% of the total cost).

In place of bolts, 4” nails were used to fasten the upper timbers. Although the bolts were functionally superior, the nails cost only a fraction as much and functioned satisfactorily as long as the roofs were not removed frequently. Removal of the upper timber allowed the sheeting to be rolled (Figure 3-6), letting rainfall pass through when desired. However, without bolts, this feature was too time-consuming, and was not used except for demonstration.

Figure 3-5. Bolts (secured underneath with wing-nuts) fastening a shelter roof.

The total cost of this design - including materials, transport, and labor - was approximately US$300 per shelter. Gutters on the eaves of the shelters (excluded from the final design) would have added an additional US$100. In lieu of gutters, trenches were dug at the soil surface (Figure 3-7) to collect and divert the runoff.

Figure 3-6. Furled roof allowing throughfall.

44

3.2.2 Evaluating rain shelter performance

The methods described below pertain to the rain shelters at the long-term Gliricidia trial (MZ12). Similar but less intensive measurements were performed on the slightly larger shelters at the seedling trial (Nkula) with similar results. Selected aspects of shelter performance at Nkula are described in Chapter 5.

3.2.2.1 Rainfall interception

Rainfall was measured underneath selected shelters using ground-level rain gauges at (a) each corner, (b) 1 m inset from each corner, (c) in the center of the plot, and (d) 1 m outside the plot (see diagram in Figure 3-8). During each growing season, the rain gauges were deployed soon after the shelter roofs were installed, and the gauges remained in place until maize harvest and roof removal. Gauges were measured and emptied after one major or several minor rain events. Any rainwater collected from a rain gauge was returned to the area below the gauge.

3.2.2.2 Soil moisture

Soil moisture was measured gravimetrically using core samples obtained with an Edelman soil auger. Each soil core was divided into a 0-20 cm sample and a 20-40 cm sample. Measurements

(a)

(b)

Figure 3-7. Trenches to divert rainwater flowing from the roofs of (a) the smaller shelters at the long-term Gliricidia trial (with field assistants); (b) the larger shelters at the seedling trial.

45

at 40-60 cm depth could not be obtained in all plots and were therefore not included in the analysis.

Soil moisture measurements were performed approximately once per month during the growing season. Due to the physical difficulty of collecting the samples and the damage caused to the plots, in each round of sampling only one soil sample was taken per plot (near the center).

Upon collection, soil samples were immediately enclosed in paper envelopes and their fresh weight (FW) was recorded, after which they were dried in a 140 ºC oven until they reached a constant dry weight (DW). Gravimetric moisture content was calculated as (FW - DW) / (FW).

3.2.2.3 Leaf water content

Near the end of the 2008-2009 growing season, three fully mature leaves were collected from each of the six Gliricidia trees nearest to the center of each agroforestry plot (see Chapter 4 for plot diagrams). To ensure no residual effects of dew, leaves were collected in the late afternoon. Gravimetric moisture content of the leaves was calculated as described above.

3.2.2.4 Litter decomposition rates

Litterbags with dimensions of 20 cm × 20 cm were constructed from 1 mm fabric mesh. In March 2009, a sufficient amount of fresh Gliricidia leaf biomass was collected (from Gliricidia trees nearby but outside the plots) to fill each litterbag with approximately 120 grams of fresh biomass. The exact weight of fresh biomass varied by bag; each bag was sewn shut and its fresh weight was recorded. Throughout the process of filling litterbags, subsamples of the Gliricidia biomass was taken to determine its average moisture content.

In each Gliricidia plot, a litterbag containing fresh biomass was buried at the bottom of a central ridge (see Chapter 2 for a description of ridges) at a depth of approximately 20 cm. Litterbags were left in place for 28 days, after which they were retrieved and the contents were removed. The remains of the Gliricidia leaves were cleaned of any residual dirt, placed in envelopes, and dried in a 140 ºC oven until they reached a constant dry weight (DW2). This weight was then compared to the estimated dry weight of the fresh biomass (DW1) that had been initially placed into that particular litterbag. The decomposed fraction was calculated as (DW1 - DW2) / (DW1).

3.2.2.5 Light interception

PPFD (photosynthetic photon flux density; µmol m-2 s-1) in each plot was measured using a handheld FieldScout quantum flux meter (model 3415F; Spectrum Technologies, Plainfield, IL, USA). Measurements were taken several times per growing season at various times of day. Although readings could not be taken simultaneously in each plot, all the plots in each experiment could be completed in approximately 15 minutes, during which time the sun’s angle did not change appreciably. Variability in PPFD readings due to shifting cloud cover was assumed to be randomly distributed over all plots and thus was assumed not to bias the results.

Because midday PPFD values at this latitude (15º31’S) on sunny days often exceeded the sensor’s maximum value (1999 µmol m-2 s-1), measurements had to be taken in the early- to mid-

46

morning or the mid- to late afternoon (or on cloudy days). The inability to sample at all times of day may have affected measurement of PPFD reduction in unknown ways.

3.2.2.6 Pan evaporation

Aluminum pans of 20 cm diameter were filled with a standard volume of water (150 mL) and deployed in the center of selected plots. After 24 hours ± 15 minutes in situ, the volume of water from each pan was measured and converted to height (mm). Any remaining water was discarded outside the plots. These measurements were performed several times in the latter half of each of the two growing seasons.

3.2.2.7 Soil temperature

Soil temperature was measured near the center of each plot using a handheld temperature probe (Checktemp 1, model HI 98509, Hanna Instruments, Smithfield, RI). The probe was inserted to 5 cm depth (the approximate maximum penetration depth of the probe in the dry clay soils) and allowed to calibrate for one minute before the temperature reading was taken. Measurements were obtained several times in the latter half of the 2008-2009 growing season.

3.2.2.8 Air temperature

Air temperature was measured at 1 m above ground level in selected plots (both shelter and non-shelter plots) at 15-minute intervals over a 10-day period in April and May 2009. Measurements were taken with HOBO Pendant Temperature data loggers (model UA-001-8K; Onset Computer Corporation, Bourne, MA, USA).

3.3 Performance of rain shelters

Three aspects of the rain shelters’ performance are reported below: (1) their intended effects on rainfall, soil moisture, and related variables; (2) their unintended effects on PAR, air temperature, and other microclimatic factors; and (3) their durability under field conditions.

3.3.1 Intended effects

3.3.1.1 Total rainfall

The shelters effectively diverted rain, as shown in Figure 3-8 (which represents rain interception averaged across all measured rain shelters and all rain events). At a distance of 1 m inside the shelter, on average 90% of rainfall was excluded; therefore, 1 m was selected as an appropriate buffer size to minimize edge effects in the plots.

After four weeks of complete rain exclusion (February – March 2009), soil moisture decreased

Figure 3-8. Map of average rainfall reduction under shelter roof.

47

by 39% at 0-20 cm depths and by 29% at 20-40 cm depths (Figure 3-9). This reduction in soil moisture is approximately representative of measurements taken at other times and at the seedling experiment (refer to Chapter 4 for more detailed soil moisture measurements).

3.3.1.2 Litter decomposition

The decomposition rate of Gliricidia leaves was significantly reduced underneath the rain shelters (Figure 3-10). Over the 28-day incubation, decomposition in shelter plots was approximately 23% below that of control plots.

3.3.1.3 Gliricidia leaf water content

Leaf water content of Gliricidia was unchanged by the presence of rain shelters (Figure 3-11). This suggests that the shelters did not inflect significant water stress on the trees, not even trees at the center of the plot. This result concurs with the fact that the shelters were not designed to affect water movement at Gliricidia’s rooting depth, which can exceed 20 meters.

Figure 3-9. Soil moisture differences between control and shelter plots after four weeks of complete rain exclusion.

Figure 3-10. Decomposition of fresh Gliricidia biomass in litterbags after 28 days.

Figure 3-11. Shelters had no significant effect on Gliricidia leaf moisture content.

48

3.3.2 Unintended effects


The overall effect of the shelters on air temperature was minimal (Figure 3-12), although they did appear to have a slight moderating effect, resulting in cooler daytime temperatures and warmer nighttime temperatures than in the non-shelter plots. Averaged over day and night, the net effect was an of increase of ~0.1°C.


The shelters had no effect on pan evaporation (Figure 3-13). Evaporation rate is chiefly a function of temperature, relative humidity, and wind speed. The absence of an effect on air temperature was noted above (section 3.3.2.1); the lack of differences in pan evaporation rates implies that the shelters probably also had negligible effects on wind speed and ambient humidity in the crop canopy.


A significant positive effect of the shelters on soil temperature was observed (Figure 4-13). However, in the absence of an effect on air temperature, the difference in soil temperature may be due to decreased soil moisture, which could in turn affect latent heat transfer, soil heat capacity, or other soil thermal properties (as opposed to being a direct effect of the shelters).

Figure 3-12. Difference between air temperature in ambient plots versus rain shelter plots over a 5-day period (8-13 May 2009).

49

3.3.2.4 Incident radiation: significant effect

The shelter roofs caused a reduction in PPFD of approximately 33% (Figure 3-14 and Figure 3-15). This reduction was consistent across different days and times of measurement. Light is not generally thought to be a limiting factor for plant growth in this environment, and even under the shelter roofs, midday PPFD generally exceeded the level of 650 µmol m-2 s-1 that is considered to be more than adequate for optimal maize growth (Tollenaar, 1999). However, for future applications, it would be desirable to directly test whether this reduction in radiation affects the photosynthesis rate of maize or trees (for example, by performing instantaneous measurements with a handheld photosynthesis sensor in the field).

3.3.2.5 Dust on leaves

Lack of rainfall within the drought plots caused the leaves of both the Gliricidia and the maize to accumulate an abnormally thick layer of dust and mold. This may have impeded photosynthesis and contributed to the negative yield effect seen in maize. However, because the deep-rooted Gliricidia experienced no negative impact on biomass production despite experiencing dust accumulation, it seems likely that the magnitude of the impact is minor. In future studies, it may be preferable to control for this unintended effect by briefly opening the roofs every several weeks to admit sufficient rain to clean the leaves.

Figure 3-13. Three 24-hour trials of pan evaporation (April-May 2009) showed no difference between control and shelter plots.

Figure 3-14. The shelters significantly reduced PAR (on three measurements days with in May 2009 with varying sky conditions).

Figure 3-15. The shading effect of the shelter roof can be seen in the right half of the photo.

50

3.3.3 Durability

The shelters proved sufficiently durable for their intended use, especially considering their low-cost construction. However, their first two years of use revealed several design flaws, namely, inadequate support for the polyethylene sheets and insufficient protection against termites. These problems are discussed in detail below.

3.3.3.1 Roof durability under normal conditions

All polyethylene sheets and nearly all roof timbers survived two growing seasons unscathed, with only a few cracked timbers needing replacement the second year. However, during the first growing season (2008-09), the sheets on several roofs began to sag and collect water, sometimes causing slow water leaks into the plot. In December 2009, at the onset of the second growing season, the shelters were reinforced with 2 mm galvanized steel wire to prevent the polyethylene sheets from sagging (Figure 3-16) at an additional cost of US $30 per shelter (labor plus materials). This retrofit was successful, and no sagging was subsequently observed.

3.3.3.2 Violent storms: weathered successfully

The shelters proved remarkably resilient to violent weather events, including a rainstorm in early 2010 accompanied by high wind gusts that tore a corrugated iron roof off of a cotton ginnery several hundred meters away. The shelters, on which the polyethylene roofs had already been installed, survived the storm with no significant damage. This may be because the polyethylene was clamped to the roof with long planks of wood rather than at individual attachment points, which would have experienced intense stress in high winds.

At the end of the dry season (September – November), high temperatures in southern Malawi create intense updrafts that give rise to dust-devils (kamvulumvulu in Chichewa). These

whirlwinds are strong enough to lift maize stover, leaves, papers, and plastic bags many hundreds of meters into the air. Kamvulumvulu were observed directly striking the roofless shelters on several occasions and doing no damage. It is fortunate that these whirlwinds are a phenomenon only of the dry season and thus pose no danger to the polyethylene roofs.

3.3.3.3 Termites: a major problem

The most serious threat to the integrity of the shelters was termites. Despite efforts to discourage termite attack by wrapping the belowground portions of the vertical poles in black polyethylene (Figure 3-17),

Figure 3-16. Roof of shelter reinforced with 2 mm wire (before installation of clear polyethylene sheets).

51

termites inflicted major damage on the shelters by the end of the second year. Originating in the soil, the termites consumed the vertical poles from the bottom up (Figure 3-18b), in several cases leading to the complete collapse of a shelter (Figure 3-18a).

The intensity of termite activity varied greatly in both time and in space. Termites are most active during the dry season; thus, most of the termite damage occurred when the shelters were not in use, allowing time for repairs before the onset of the next growing season. The two experimental fields in this project, approximately 1 km apart, suffered very different amounts of termite damage: at MZ12, approximately 10% of the vertical poles needed replacement at the beginning of the third growing season, whereas at Nkula, the figure was nearly 25%.

3.4 Discussion

The overall performance of the shelters was satisfactory. Rain was almost completely excluded; soil moisture was significantly reduced down to 40 cm depth, and there were few microclimatic artifacts. These successes were notable in light of the shelters’ simple design and low-cost construction. However, several aspects of their design could be improved for future applications.

3.4.1 Possible improvements to existing design

3.4.1.1 Incident radiation

With the existing design, it was not feasible to remove the roofs between rainstorms, as the process requires several hours of labor per roof. A slightly higher-budget project could overcome this problem by fastening the roofs using bolts with wing-nuts (as shown in Figure 3-5). Another way to minimize effects on PAR could be to use a similar but more translucent roof material; other similar studies have reported PAR reductions of closer to 25% (English et al., 2005) or even as little as 8% (Huxman et al., 2004) using polyethylene plastic sheeting.

3.4.1.2 Surface and belowground water flow

Roof-mounted gutters, though expensive (they would have increased the cost of these shelters by 30%), would confer a major improvement in managing water flow. A cheaper alternative that would still be likely to reduce infiltration near the plot would be to line the ground-level trenches with black polythene plastic. Controlling surface and belowground water flow would become even more important if studying deeper-rooted crops.

Figure 3-17. Vertical poles (belowground and 30 cm aboveground) were wrapped in black polyethylene to discourage termites.

52

3.4.1.3 Attack by termites

(a)

(b)

Figure 3-18. Termite damage to rain shelters. (a) Shelter collapse due to termite attack on vertical poles. (b) Vertical pole left hanging after being eaten by termites from bottom up.

The above approach to termite management – making modest efforts to discourage termite attack, and repairing any damage that does arise – may be acceptable for some short-term research projects. If the shelters are expected to last longer than three years, more aggressive measures may be needed. Possible alternatives include treating the vertical poles with insecticide before construction; setting the poles in concrete rather than directly in the soil; or using non-wood materials (e.g., plastic or metal) for the vertical poles. The latter solution, however, would greatly increase construction cost, undermining a major goal of this project.

3.4.1.4 Insufficient roof height for trees

In several cases, the maximum height of the trees in the second year equaled or exceeded that of the roof, and tree height growth was impeded by the roof. However, in such cases, the tree roots almost certainly extended outside the influence of the shelter, and thus the shelter is not likely to be effective. Rather than attempting to build taller shelters (see section 3.4.2.1), a more prudent solution might be to limit the duration of the shelters’ use to the period during which tree seedlings can easily be accommodated under the roofs.

3.4.1.5 Control for confounding effects by watering under shelters

Applying a watering treatment equivalent to ambient rainfall under some of the shelters could reveal whether the shelters affected maize yield through any processes other than water availability – such as reduction in solar radiation, dust accumulation, or trampling or soil compaction. A watering treatment was not used in the present study due to budgetary constraints (it would have necessitated at least a dozen additional dozen rain shelters).

53

3.4.2 Future applications

3.4.2.1 Other crops and other locations

Observations during these shelters’ construction and use suggest that the maximum viable height for this design is probably not much more than 3 m at the roof peak. Further height extensions might compromise the strength of the structure while also making roof access impractical. However, a 3 m roof height could accommodate many other widely grown tropical crops of similar stature to maize (for example, sorghum, millet, cassava, and coffee, as well as short-statured varieties of sugarcane and plantain).

Where project budgets permit, it would still be preferable to construct more durable shelters with superior mechanisms for roof removal and management of water flow. However, in reality, such high-cost shelters are usually impractical for research institutions in developing countries. On the rare occasions that such experiments have been carried out in low-or middle-income countries, the large majority of the funding has come from institutions in high-income countries (e.g., Nepstad et al., 2007; Schwendenmann et al., 2010). Thus, affordable shelter designs could provide more autonomy to researchers in many locations in many parts of Africa, Asia, and Latin America where ecosystem climate manipulations have never before been done in a field setting.

3.4.2.2 Introduce heating or other variables

Drought and high temperatures can have multiplicative effects in reducing maize yield (Paulsen, 1994), while increased atmospheric CO2 may mitigate these effects somewhat (Lobell et al., 2011). Controlled field experiments examining the interactions of these variables in vulnerable regions could be valuable in climate adaptation planning.

The manipulation of temperature and CO2 in the field can be difficult and expensive, which contradicts this study’s goals of affordability and accessibility. However, at least one of these variables lends itself to a low-tech approach. Passive greenhouses can increase air temperature by 2-6 ºC (Shen and Harte, 2000) with minimal artifacts if the greenhouses are not fully enclosed. The present rain shelter design could be modified to manipulate air temperature by partially enclosing the sides of the shelters. Side enclosures could be used with or without roofs, allowing the effects of temperature and drought to be studied separately.

3.5 Conclusions

There is a pressing need for rain manipulation experiments in sub-Saharan Africa, and the results of this project show that such experiments can be accomplished with reasonable labor requirements, readily available materials, and a modest budget. The shelters used in this experiment are, to the best of the author’s knowledge, the only above-canopy rain shelters to use wooden rather than metal frames, and, therefore, are likely the most economical of any comparable design reported in the literature.

The design was built efficiently by local carpenters using local materials, and the shelters achieved their goals of low cost and sufficient durability. The rain interception and soil moisture reduction of the shelters was also satisfactory. Microclimatic artifacts were minor; however, further tests are needed to confirm that PAR reduction does not plant growth. This design has

54

many potential uses in tropical agricultural research, especially in developing nations where research budgets may be limited.


This project would not have been possible without the expert assistance of Makoka carpenter Mr. Damson Singo, who helped to design the shelters and oversaw all construction. Mr. Singo was also instrumental in troubleshooting design flaws, repairing ongoing wear and tear, and installing and removing the shelter roofs each rainy season. Mr. Chimwemwe Mlongoti, Mr. Alfred Mwase, and Mr. Jose Kapesi also contributed an immense amount of time, labor, and experience to the construction of the shelters. The author is deeply grateful to them all.

55

Chapter 4. Effects of drought on an established Gliricidia-maize intercropping system

Abstract

Fertilizer tree systems such as Gliricidia intercropping show promise for boosting maize yields and improving food security in southern Africa, but little is known about their performance under drought conditions. This three-year experiment (2008-2011) used rain exclusion shelters to test the effects of drought on a mature Gliricidia-maize intercropping system at a long-term field site (MZ12) at Makoka Agricultural Research Station in southern Malawi. It was hypothesized that Gliricidia might improve the drought resilience of maize production via effects on soil moisture and microclimate.

The drought manipulation was carried out successfully in two out of the three years, reducing overall maize yields by 61% and 39% in 2009-2010 and 2010-2011, respectively. Yield reductions in Gliricidia plots were proportional to those in maize monoculture plots, suggesting that Gliricidia’s effects on soil quality continued to confer yield benefits under drought conditions and did not interact with the drought treatment either positively or negatively. Gliricidia had no effect on soil moisture in any part of the growing season, though it did moderate soil temperature and air temperature at the end of the growing season.

These results, along with long-term yield data (1993-2006) from this field site, indicate that the yield benefits of Gliricidia are greatest under conditions of adequate rainfall but that Gliricidia still outperforms monoculture maize even in drought conditions. Although these findings are not sufficient to support the hypothesis that Gliricidia increases drought resilience, they nevertheless suggest that Gliricidia intercropping will remain a useful tool for improving maize yields in adverse and uncertain climatic conditions.

It is important not to generalize these results to all fertilizer tree systems, as other tree species in other environmental conditions may respond differently. There is considerable need for future research to optimize these soil fertility technologies for current climate variability and future climate change.


Malawi’s Southern Region is characterized by high population densities, small and declining farm sizes, and insufficient levels of soil nutrient replenishment, representing an extreme example of southern Africa’s food security challenges. The primary goal of the fertilizer tree systems described in Chapter 2 is to help address these challenges. However, their performance under adverse climatic conditions is largely unexamined. Understanding the climate response of fertilizer tree systems may help farmers cope with current climate variability and is likely to become even more important under future climate change.

56

This experiment described in this chapter aims to contribute to this understanding by imposing an experimental drought on a mature Gliricidia-maize intercropping system. The experiment’s goals were twofold: first, to characterize the impacts of the drought on maize yields in the presence and absence of Gliricidia, and second, to elucidate relevant mechanisms by which Gliricidia might affect the drought-tolerance of maize production.

4.1.1 Fertilizer trees and climatic stress

Knowledge about the effectiveness of fertilizer tree systems under various climatic stresses is mostly anecdotal at present. Ikerra et al. (2001) described high interannual variability in the biomass production of Sesbania fallows in southern Malawi, but did not elaborate on the reasons. In eastern Zambia, Kwesiga et al. (1999) observed negative effects of drought on Sesbania seedling establishment in eastern Zambia, while Phiri et al. (2003) noted beneficial effects of mature Sesbania fallows on soil moisture. However, none of these results were translated into practical advice to help farmers cope with climate variability.

At the time of writing, the only known review of fertilizer tree performance as a function of climate was by Sileshi et al. (2011), who examined the rain use efficiency (RUE, defined as yield per unit rainfall) of three Leucaena-maize intercropping systems in Zambia and Nigeria. They found that the RUE of fertilizer tree systems exceeded that of unfertilized maize monocultures, but the advantage was generally greater in wet years than in dry years; furthermore, the RUE of fertilized maize monocultures was equal to or greater than that of fertilizer tree systems. The authors called for further work on the effect of water availability on performance of fertilizer trees of different species at different locations.

4.1.2 Previous work at this field site

The current experiment was conducted at a long-term Gliricidia-maize intercropping trial at Makoka Agricultural Research Station (described in Chapter 2) near Zomba in southern Malawi. This trial, designated MZ12, was established in 1992 by ICRAF researchers (see Section 4.5.1) to evaluate the potential of Gliricidia to increase maize yield. Table 4-1 describes baseline soil properties at the site, and Figure 4-3 shows the appearance of the field throughout the growing season.

MZ12 has been maintained continuously for the past two decades and has been used for a variety of different studies, including interactions of Gliricidia inputs with inorganic fertilizer and long-term effects on Gliricidia on soil organic matter stocks. Before discussing the current experiment, it will be helpful to summarize the results of previous work at MZ12 to provide context on how this Gliricidia-maize intercropping system performs under normal conditions.

Table 4-1. Characteristics of MZ12 topsoil (Oxic Haplustalf), 0-20 cm, Makoka Agricultural Research Station (from Makumba et al., 2009).

Clay (%) 42

Sand (%) 46

Silt (%) 12

Bulk density (g/cm3) 1.42

pH in water (1:2.5) 5.9

Organic carbon (g/kg soil) 8.8

P-‐Olsen (mg/kg soil) 26.0

Exchangeable K (mmol (+)/kg soil) 3.0

Exchangeable Ca (mmol (+)/kg soil) 44.0

Exchangeable Mg (mmol (+)/kg soil) 16.0

57

4.1.2.1 Maize yield and soil nutrient dynamics

Past results have shown that Gliricidia has a greatly beneficial effect on maize yields at MZ12. Makumba et al. (2006) and Akinnifesi et al. (2007) reported that after eleven years of intercropping at MZ12, maize yields in unfertilized Gliricidia plots were more than three times higher than maize yields in unfertilized monoculture plots (3800 kg/ha versus 1100 kg/ha).

The highest yields were seen in plots combining Gliricidia and inorganic N and P fertilizer at 46 kg/ha and 32 kg/ha, respectively (each half the recommended dose). Akinnifesi et al. (2007) described this as a synergistic effect of organic and inorganic inputs – possibly due to improved soil texture and reduced leaching losses in Gliricidia plots – though an earlier study by Makumba et al. (2005) found that the combination of Gliricidia and inorganic fertilizer had only an additive effect.

Phosphorus addition at this site has produced mixed results: Makumba et al. (2006) found that inorganic P fertilizer did not affect maize yields. However, Akinnifesi et al. (2007) found that optimum grain yield at MZ12 was achieved with 32 kg/ha P addition, while Mweta et al. (2007a; 2007b) described an optimum P rate of 20 kg/ha. Mweta et al. further reported that Gliricidia recycles P from depths below the maize rooting zone and that Gliricidia prunings help to solubilize soil P (perhaps via effects on soil organic matter), thus providing two pathways to increase P availability to the maize crop.

Although P is an important limiting factor to crop growth in many parts of sub-Saharan Africa (Sanchez, 1995), the lack of a dramatic response to P addition at MZ12 led to the phosphorus treatment being discontinued in 2008.

4.1.2.2 Rooting depth and water use patterns

Although rooting depth and water use of Gliricidia have not been studied at MZ12, extensive studies have been conducted at a similar trial (MZ21) less than 1 km away. The results have generally indicated good complementarity of water use in maize and Gliricidia.

For example, Makumba et al. (2009) found that Gliricidia had relatively low rooting density (0.38 cm/cm3) in the top 0-20 cm compared to maize (1.02 cm/cm3) and that most Gliricidia fine roots were located at depths greater than 40 cm. Chirwa et al. (2007) found no evidence of water competition between maize and Gliricidia during the growing season, though soil in Gliricidia plots was somewhat drier than soil in monoculture plots by the end of the dry season. A simulation using the WaNuLCAS model (Chirwa et al., 2006a) did not replicate these results, implying instead that Gliricidia would deplete soil water during the growing season.

4.1.2.3 Soil organic matter

Results on soil organic matter dynamics at MZ12 have been subtle and somewhat controversial. Makumba et al. (2007) reported that, ten years after Gliricidia establishment at MZ12, soil carbon storage in the 0-20 cm layer was 11 Mg/ha C, or 1.6 times that of sole maize. They estimated that trunks and coarse roots of Gliricidia stored an additional 17 Mg/ha C. Although they observed that CO2 efflux was also increased in Gliricidia plots (up to three-fold higher than in monoculture plots), they concluded that Gliricidia-maize intercropping could increase net C sequestration compared to sole maize. However, Kim (2012) disputed these results on the

58

grounds that Makumba et al. failed to use an accurate baseline for soil C and that they neglected to include possible effects of CH4 and N2O.

Beedy et al. (2010) found that Gliricidia at MZ12 had different effects on different fractions of soil organic matter: compared to maize monoculture, Gliricidia plots showed a 12% overall increase in SOM and a 40% increase in particulate organic matter (POM). They noted that the latter effect, along with an observed increase in CEC (cation exchange capacity), might magnify the effects of Gliricidia on maize yields beyond what could be expected from its nutrient inputs alone (perhaps via effects on soil structure).

4.1.3 Goals and hypotheses of this experiment

The current experiment (conducted over three growing seasons from 2008 to 2011) aimed to build upon previous work at this location by examining the response of the Gliricidia-maize intercropping system to an artificial drought. Individual goals and hypotheses are listed below.

4.1.3.1 Effect of drought on maize yield in monoculture and Gliricidia intercropping systems

Goal. The foremost goal of this experiment was to quantify the impact of a major drought on maize yield in the presence and absence of Gliricidia. As described above, Gliricidia plots at this location have already been shown to have higher maize yields than monoculture plots, so the effects of drought needed to be considered in the context of this baseline.

Hypothesis 1. Drought will have a proportionally greater negative effect on maize yield in maize monoculture plots than in Gliricidia intercropping plots.

Hypothesis 2. The negative effect of drought on maize yield will manifest as a decrease in thousand-kernel weight, due to the imposition of the drought during anthesis and grain-filling.

4.1.3.2 Interaction of inorganic fertilizer with drought and cropping system

Goal. The primary function of Gliricidia intercropping systems is to provide nitrogen to the maize crop, a role that can also be fulfilled by inorganic fertilizer. However, Gliricidia may have effects on soil properties and on microclimate that fertilizer does not, and under drought conditions, these ancillary effects may become more relevant than the supply of nutrients. Thus, this experiment aimed to separate the nitrogen-provisioning effect of the trees from their other effects by including inorganic nitrogen fertilizer as a variable.

Hypothesis 3. Relative impacts of drought on maize yield will be similar or more severe in fertilized monoculture plots as compared to unfertilized monoculture plots; in other words, nitrogen fertilizer alone will not convey protection against drought effects.

4.1.3.3 Influence of Gliricidia on microclimate and soil moisture

Goal. Previous work has demonstrated the potential of other types of fertilizer trees in sub-Saharan Africa to reduce soil temperature (Acacia albida; Vandenbeldt and Williams, 1992) and increase soil water-holding capacity (Sesbania sesban; Phiri et al., 2003). Furthermore, Chirwa et al. (2007) and others have shown that the rooting zone of Gliricidia does not tend to overlap with that of maize, implying that Gliricidia is unlikely to cause depletion of soil water at shallow depths. Thus, this experiment sought to determine whether Gliricidia might exert

59

influences on microclimate or soil moisture that could help to protect maize from drought conditions.

Hypothesis 4. After the imposition of the drought, Gliricidia plots will maintain higher soil moisture in the maize rooting zone (0-20 cm) than will maize monoculture plots.

Hypothesis 5. Gliricidia plots will have lower air temperature, soil temperature, and pan evaporation rates than maize monoculture plots.

4.1.3.4 Maize yield variability of monoculture and agroforestry plots across different years

Goal. To supplement the results of the rain manipulation experiment, this project aimed to examine the relative effects of interannual variability on Gliricidia plots and maize monoculture plots using the long-term record of maize yield data (1993-2006) for MZ12. If Gliricidia trees protect maize from drought via the mechanisms described above, they may confer proportionally greater benefits during years of natural drought, leading to an overall buffering of yield variability compared to maize monoculture.

Hypothesis 6. In years of significantly below-average growing season precipitation (<600 mm from 1 November to 30 April), the maize yield advantage of Gliricidia plots will exceed that of their yield advantage in normal years.

Hypothesis 7. Interannual variability of maize yield from 1993-2006 will be lower in Gliricidia plots than in maize monoculture plots.

Hypothesis 7 is informed by a meta-analysis (Sileshi et al., 2011) in which Leucaena-maize intercropping systems in Zambia and Nigeria were seen to have lower interannual yield variability than monoculture unfertilized maize. Similar results were reported by Snapp et al. (2010) for other green manure systems in Malawi.

The experiment’s goals and approaches are briefly outlined in Table 4-2 below. The following section describes the methods in detail.

Table 4-2. Summary of drought experiment at MZ12 (written by the author for an informational sign for visitors mounted at the entrance to the field).

Effects of drought and nutrient addition on sole maize and on a Gliricidia-‐maize intercropping system

Year of establishment: 1992

Phase 1: Nitrogen and phosphorus treatments (1992 -‐ 2007): Completed

Phase 2: Drought treatment (2008 -‐ present)

Problem: Future climate change in Malawi is likely to shorten the growing season and reduce total rainfall.

Objectives: Determine response of maize and maize-‐Gliricidia intercrop to an artificial drought

60

imposed by rain shelters.

Treatments:

A. Two production systems 1. Sole maize 2. Maize + Gliricidia

B. Three inorganic nitrogen levels (CAN, calcium ammonium nitrate) 1. 0 kg N / ha 2. 46 kg N / ha 3. 92 kg N / ha (full recommended dose)

C. Two precipitation levels 1. Full ambient precipitation 2. Reduced precipitation in mid-‐ to late growing season (exact amount varies by year)

4.2 Methods

The core innovation of this experiment was a series of open-sided rain shelters that were deployed above selected plots in MZ12 (Figure 4-2). Details of rain shelter design and performance are described in Chapter 3, while methods specific to this experiment are described below. Figure 4-1 depicts a typical plot, while Figure 4-3 and Table 4-3 provide an overview of field management practices.

Figure 4-1. Close-up of a Gliricidia-maize intercropping plot (March 2010).

Figure 4-2. Looking across MZ12 from Block 1 (in the foreground) to Block 3. (February 2010.)

61

(a) 16 October 2009

(b) 16 December 2009

(c) 24 January 2010

(d) 16 February 2010

(e) 20 April 2010

(f) 28 September 2010

Figure 4-3. Seasonal progression of the Gliricidia-maize intercropping system at MZ12, showing (a) Gliricidia pruning at end of dry season, (b) new growth at beginning of wet season; (c) and (d) maize maturation; (e) senesced maize ready for harvest; and (f) dry-season Gliricidia fallow.

62

4.2.1 Modification of existing experimental layout

At the initiation of this experiment, the field layout at MZ12 consisted of a total of 81 plots laid out in three strips separated by 1-m wide paths (Figure 4-4) with each strip representing one complete block. The three original treatment factors, applied in a full factorial design, were as follows:

(1) three cropping systems (sole maize plus two different Gliricidia pruning regimes), (2) three levels of inorganic N addition (0, 46, and 92 kg/ha), (3) three levels of inorganic P addition (0, 20, and 40 kg/ha).

This gave rise to a total of 3 × 3 × 3 = 27 plots per block, for a total of 3 × 27 = 81 plots.

In the present experiment, P was not a factor of interest (due to the low response rate described in Section 4.1.2.1), nor were the different Gliricidia pruning regimes (which had been standardized to a single regime many years previously). This elimination of factors provided the flexibility to impose an additional factor, rain manipulation, on the existing experimental design while still maintaining three replicates per treatment. The new design consisted of three factors as follows:

(1) cropping system (sole maize and Gliricidia); (2) inorganic N addition (0, 46, and 92 kg/ha); (3) rainfall (ambient and reduced).

This required a total of 2 × 3 × 2 = 12 plots per block, leaving 15 plots per block unused for this experiment. The unused plots were maintained and harvested as normal, but were generally not used for data collection.

Two complications were encountered when imposing the new experimental design upon the old one. First, because there was no space between plots within a block (Figure 4-4), the presence of a rain shelter in one plot was likely to affect the adjacent plots. Thus, the plots on either side of a rain shelter had to be excluded from use in the current experiment.

Second, because the P treatment had only been discontinued in the previous year, possible residual effects of P could not be ruled out. However, there was not a sufficient number of plots to select plots of only one former P level. As a compromise, plots were selected so that each treatment combination in the new experiment (i.e., a specific combination of cropping type, N level, and rainfall) included exactly one plot of each P level (0, 20, and 40 kg/ha) for a total of three replicates per treatment.

63

Figure 4-4. Field layout of MZ12. Shaded plots are used in the current experiment. G = Gliricidia (0 = sole maize; 1 = Gliricidia intercrop); N = nitrogen fertilizer (0 = none; 0.5 = 46 kg N / ha-yr; 1 = 92 kg N / ha-yr); R = rain (0 = ambient, 1 = drought).

(a)

(b)

Figure 4-5. (a) Layout of a Gliricidia intercropping plot in MZ12. Layout of sole maize plots is identical except without trees. Plots are contiguous along their 6.1 m edges. (b) Key to plot diagram above.

4.2.2 Rain manipulations

The design of the rain structures permitted only complete rain exclusion (see Chapter 3); their rain interception could only be controlled by roof installation and removal. However, as roof installation was difficult and time-consuming, the decision was made to impose a single long-duration drought coinciding with maize anthesis and grain-filling, when maize is known to be highly vulnerable to drought (Edmeades et al., 2000).

In the first year of the experiment (2008-2009), unforeseen construction delays resulted in the drought being imposed later than intended and reducing its effectiveness. In the second two years (2009-2010 and 2010-2011), the drought was initiated at the onset of anthesis and maintained until maize harvest. Table 4-3 lists relevant dates for each year.

The 2008-2009 rain manipulation (7 Mar to 15 May) intercepted 172 mm of the 1070 mm annual total rainfall (16.1%). Subsequent years achieved considerably higher interception: the

64

2009-2010 manipulation (27 Jan to 10 May) intercepted 669 mm of the 999 mm annual total (67.0%), while the 2010-2011 manipulation (5 Feb to 6 Jun) intercepted 365 mm of the 1166 mm annual total (31.3%).

It is worth noting that these markedly different interception fractions were achieved by roof installation dates that differed by only a few weeks; this underscores the brevity of the growing season in southern Malawi (Figure 6-2) and the potentially important influence of interannual rainfall variability on crop production.

4.2.3 Land preparation and Gliricidia pruning

4.2.3.1 Land preparation

Land preparation was conducted towards the end of the dry season (in September or October of each year) in accordance with prevalent local practices. In maize monoculture plots, this process consisted of using hand hoes to fell all the weeds in the plot, after which the weed biomass and the maize stover from the previous harvest (see Section 4.2.4.3) were distributed evenly amongst the plot and buried underneath each ridge. Due to the labor-intensive nature of this process, it is usually carried out well before the rainy season begins. In Gliricidia plots, an additional step was needed: cutting and incorporating Gliricidia leaf biomass along with the weeds and maize stover as described above. This process is described in the next subsection.

4.2.3.2 Pruning and weighing of Gliricidia biomass

Gliricidia trees were pruned three times per year: once at the end of the dry season (late September or early October), once shortly before maize planting (in mid-November), and once about four weeks after maize planting (late December). Pruning dates for the three years of this experiment are given in Table 4-3. Although optimal pruning times for Gliricidia are a matter of some debate (Makumba et al., 2005; Chirwa et al., 2006b), this regime was chosen to correspond with previous management practices at MZ12.

At the time of the first pruning, Gliricidia aboveground biomass consisted of numerous vertical trunks generally 2-3 m high and 2-4 cm in diameter (Figure 4-3f). This biomass was separated into two compartments: leaves (plus small branches) and wood. The wood was removed from the field for use as fuelwood. At subsequent prunings (Figure 4-6), regrowth was considerably less and consisted only of leaves and small branches; all this biomass was returned to the plot.

In each case, pruning was carried out by using machetes to cut regrowth from the stump approximately 30 cm above ground level. Leaves and branches from each plot were weighed in the field, and a subsample was taken for moisture content. Biomass was then evenly distributed in the plot and buried underneath each ridge using hand hoes (Figure 4-7). At the third pruning, when maize seedlings were already present, the Gliricidia biomass was carefully buried in the side of each ridge to avoid disturbing the maize roots.

65

Figure 4-6. Pruning Gliricidia shortly before maize planting (24 November 2009). The trees were previously pruned in October and will be pruned again in January.

Figure 4-7. Burying freshly pruned Gliricidia biomass under a ridge, on which maize will be planted (24 November 2009).

4.2.4 Maize cultivation and harvest

4.2.4.1 Planting and thinning

Maize hybrid variety DeKalb 8053 (Monsanto) was planted in each plot once the rains had stabilized (generally late November or early December of each year; see Table 4-3 for dates). This maize variety was chosen because it had been used at MZ12 in recent previous years. Various local conventions exist for the spacing of maize plants within rows. To be consistent with prior methodology at MZ12, maize was planted 30 cm apart (with two seeds per planting station, approximately 5 cm deep). Figure 4-5 depicts the location of maize and Gliricidia plants within a plot.

Approximately four weeks after germination, maize was thinned to one individual per planting station, and empty planting stations were filled in by transplanting these extra maize seedlings (to the extent possible). This reflected local practice and was consistent with previous years’ field management at MZ12.

4.2.4.2 Weeding and fertilizing

All plots were weeded with hand hoes twice during the growing season, at approximately four and seven weeks after maize planting. This represented local best practice and minimized the influence of weeds on crop growth for the duration of the experiment.

In fertilized plots, calcium ammonium nitrate (CAN) was applied twice during the growing season (usually shortly after weeding; dates are given in Table 4-3). The fertilizer was delivered in individual doses by punching a small hole (10 cm deep) in the soil approximately

66

10 cm from the base of each maize seedling and dropping a measured scoop of fertilizer into the hole. This practice, intended to reduce fertilizer waste via leaching or runoff, is common in Malawi due to the cost of fertilizer relative to labor.

4.2.4.3 Harvest

Processing of maize plants. Maize was harvested once the vegetative body of the maize plants had fully senesced, generally in late April or early May. First, the maize plants in each plot were counted; they were then cut at ground level with a machete and the cobs were removed from the husks (Figure 4-8). Maize stover from each plot was collected and weighed with a hanging scale in the field; a subsample was taken to determine moisture content. After weighing, maize stover was left lying in the furrows in the field until land preparation in September (see Section 4.2.3.1).

Processing of maize grain. The number of cobs from each plot were counted, after which the grain was shelled by hand (removed from the rachis) and the rachides and grain were weighed separately. Subsamples were taken to determine moisture content of both these yield components. For maize grain, the moisture content subsample consisted of a known number of kernels (either 200, 500, or 1000 kernels per plot, varying by year) so that the TKW (thousand kernel weight) could be determined.

Table 4-3. Dates for key management activities at MZ12 drought experiment, 2008-2011. (For multi-day tasks, the start date is noted.)

2008-‐2009 2009-‐2010 2010-‐2011

Maize Planting 26 Nov 2008 25 Nov 2009 6 Dec 2010

Harvest 20 Apr 2009 27 Apr 2010 19 May 2011

Rain exclusion

Begin 7 Mar 2009 27 Jan 2010 5 Feb 2011

End 15 May 2009 10 May 2010 a 6 Jun 2011

Total interception 172 mm (16%) 669 mm (67%) 365 mm (31%)

Gliricidia pruning

First pruning 29 Sep 2008 a 9 Oct 2009 1 Oct 2010

Second pruning 8 Nov 2008 23 Nov 2009 14 Nov 2010

Third pruning 22 Dec 2008 2 Jan 2010 28 Dec 2010

Field management

First fertilizing 23 Dec 2008 4 Jan 2010 18 Dec 2010

Second fertilizing 4 Jan 2009 none b 25 Jan 2010

First weeding 18 Dec 2008 31 Dec 2009 4 Jan 2011

Second weeding 14 Jan 2009 a 2 Feb 2010 a 26 Jan 2011 a Date is approximate. b In 2009-‐2010, fertilizer was applied in one dose instead of two doses.

67

Figure 4-8. Harvesting maize in the field (28 April 2010).

Figure 4-9. A sampling of maize cobs of various sizes from different plots at MZ12 (22 April 2009).

4.2.5 Microclimatic and soil measurements

Refer to Chapter 3 for details on the methodology of soil moisture, soil and air temperature, and pan evaporation measurements. (The same data sets served a dual purpose for evaluating the efficacy of the rain shelters and evaluating the environmental effects of Gliricidia).

4.2.6 Data analysis

Data on Gliricidia biomass production, maize yield, and maize plant characteristics were averaged at the plot level before analysis. Most microclimatic and soil data (soil moisture, soil temperature, air temperature, and pan evaporation) were collected as only a single reading per plot and thus averaging was unnecessary. Treatment effects were assessed using fixed-effects ANOVAs with the variables and interactions of interest; more details are given in the relevant Results subsections. Unless otherwise noted, error bars represent ± 1 SEM. Statistical analyses were carried out with Microsoft Excel for Mac 14.1.4 (Microsoft Corporation, Seattle, WA, USA) and JMP 10.0.0 (SAS Institute, Cary, NC, USA).

4.3 Results

This section first briefly mentions Gliricidia biomass production, then describes the effect of the drought and other experimental treatments (2008-2011) on maize production and presents the environmental effects of the Gliricidia trees. The section concludes by reporting historical yield trends from MZ12 from 1993-2006.

68

4.3.1 Gliricidia biomass production

As Gliricidia are deep-rooted nitrogen-fixing trees, mature Gliricidia would not be expected to show a noticeable growth response to surface-level rain exclusion nor to nitrogen fertilization. Nevertheless, it was methodologically prudent to test these assumptions.

4.3.1.1 Total Gliricidia biomass

Leaf and wood production by Gliricidia trees over the course of the experiment is given in Table 4-4. The three prunings (October, November, and December) each produced roughly equal amounts of leaf biomass (each approximately 500 to 1000 kg/ha), while the October pruning also produced substantial amounts of wood (approximately 5000 kg/ha).

Assuming a Gliricidia leaf N content of 3.9% by dry weight (Vanlauwe et al., 2001), this implies that Gliricidia biomass represents an annual input of slightly over 100 kg N/ha. The fraction of this N that is derived from the atmosphere (%Ndfa) versus internally recycled is unknown; values for other fertilizer tree species are generally in the range of 40-80%, but tend to be highly dependent upon site and management practices (Mafongoya et al., 2004).

Gliricidia biomass production was tested for treatment effects (rain exclusion and addition of N fertilizer) with a fixed-effects ANOVA, but no significant treatment effects were found for any of the biomass harvests over the three years of the experiment.

However, a block effect was seen: in the second and third prunings, Block 3 tended to have lower Gliricidia biomass yields than Blocks 1 and 2. This effect was significant for the third pruning in 2008-2009 (p = 0.0001), for the second and third prunings in 2009-2010 (p = 0.0437 and 0.0015), and for the third pruning in 2010-2011 (p = 0.0021). This effect may have been due to shade from the adjacent forest (seen in the background of Figure 4-2).

Although survival of the trees was not formally monitored, no mortality of any Gliricidia trees was observed over the course of the experiment.

Table 4-4. Inputs of Gliricidia biomass to maize plots at MZ12. Leaves were returned to the plot while wood was removed. Values are kg/ha dry weight except as otherwise noted. Values are mean ± 1 SEM. No treatment effects were significant except for block (see text).

2008 -‐ 2009 2009 -‐ 2010 2010-‐ 2011

First pruning (leaves) n/a A 1202.7 ± 95.7 663.3 ± 52.6

First pruning (wood) n/a A 4755.4 ± 312.5 4995.34 ± 208.3

Second pruning 629.4 ± 44.9 472.4 ± 31.8 1033.1 ± 56.6

Third pruning 614.0 ± 25.1 1044.37 ± 35.4 935.0 ± 35.1

Annual total n/a 2719.5 2631.37

Annual total, kg N / ha B n/a 106.1 102.6

A Pruning was carried out but data were not collected. B Assuming 3.9% N by dry weight (Vanlauwe et al., 2001).

69

4.3.2 Maize performance

This section reports two different aspects of maize performance: (1) total grain production and (2) reproductive characteristics. Vegetative growth characteristics (stand survival, stover biomass, and plant height) were also examined, but the former parameter showed no relationship with any of the treatment variables, while the latter two parameters showed similar trends to grain yield. Thus, details of maize vegetative growth are not reported here.

4.3.2.1 Total grain production

Overall maize yields at MZ12 during the three years of the experiment (2008-2009, 2009-2010, and 2010-2011) were generally similar to the historical averages for MZ12, with grain production in most plots ranging between 1000 and 5000 kg/ha. Treatment effects were significant in all years; they are reported in Table 4-5 and discussed in detail below.

Cropping system effects. In all three years, maize yields in Gliricidia plots significantly exceeded those in monoculture plots. Averaged across all fertilizer and rain treatments, Gliricidia response ratios (ratios of yield in Gliricidia plots to yield in sole maize plots) were 1.47, 1.93, and 1.87 for the 2009, 2010, and 2011 harvests respectively.

Fertilizer effects. Addition of calcium ammonium nitrate fertilizer significantly increased maize yields in 2008-2009 and 2010-2011. (The effect of fertilizer did not attain significance in 2009-2010, though trends were in the expected direction.) However, in no case was the effect of a half-dose of fertilizer (46 kg/ha N) distinguishable from the effect of a full dose of fertilizer (92 kg/ha N). This may imply that 92 kg/ha is in excess of crop demand, but it may also be due to incorrectly measuring the half-doses of fertilizer (which were delivered with bottle caps that were sometimes filled above the rim).

In all three years, a significant or near-significant interaction was seen between Gliricidia and fertilizer. Plots with neither Gliricidia nor fertilizer had very low grain yields (approximately 1100 kg/ha in each of the three years), while plots with one or more nitrogen sources (Gliricidia, fertilizer, or Gliricidia + fertilizer) had much higher yields (ranging from 2600 to 4900 kg/ha) and were generally indistinguishable from each other. This suggests that these nitrogen inputs are substitutive rather than additive. No evidence was seen for the synergistic interaction reported by Akinnifesi et al. (2007).

Rain manipulation effects. In the first year (2008-2009), the artificial drought did not significantly affect grain yield (p = 0.8946) due to its late timing. In the second and third years, drought plots had dramatically lower grain yield than ambient plots (a 60.8% reduction at the 2010 harvest and a 38.9% reduction at the 2011 harvest), though these effects did not reach the threshold of statistical significance (p = 0.1818 and 0.1330, respectively) due to high interplot variability and the low number of replicates.

The potential interaction of Gliricidia and drought on maize yield was of particular interest in this experiment (as described in Section 4.1.3.1). Accordingly, this interaction is reported in Table 4-5. However, the interaction was not statistically significant in any of the three years, implying that drought had similar effects in Gliricidia plots and in sole maize plots.

70

In 2009-2010, although the Gliricidia-drought interaction did not achieve rigorous statistical significance (p = 0.0703), the large differences in the means suggest that drought did in fact cause a proportionally greater reduction in maize yield in Gliricidia plots (a 73.2% decrease, compared to a 48.8% decrease for sole maize). However, the opposite trend was seen in 2010-2011, when yield decreases due to drought were 45.2% and 35.3% in sole maize and Gliricidia plots, respectively. These interactions are further explored in Figure 4-10.

Block effects. Block 3 had significantly lower maize yields than Blocks 1 and 2 in 2008-2009 and 2009-2010. Yields in Block 3 were approximately half that of yields in Blocks 1 and 2.

To determine whether edaphic differences between the blocks might be causing the differences in maize yield, a pot experiment was carried out. Two soil samples were taken from each of the 81 plots in the experiment and were used to fill (81 × 2 =) 162 1-litre black polyethylene bags, in each of which a maize seedling was grown for 60 days. At the time of destructive harvest, maize seedlings grown in potted soil from each of the three blocks exhibited no block-related differences in height, aboveground biomass, or belowground biomass (data not shown). Thus, the difference in maize yields in Block 3 may have been due to shading or other influences from the adjacent forest. For unknown reasons, this effect did not reappear in 2010-2011.

(a)

(b)

(c)

Figure 4-10. Effect of drought on maize yields in monoculture and Gliricidia plots across three years: (a) 2008-2009, when the drought manipulation was not successful; (b) 2009-2010; (c) 2010-2011. Error bars are ± 1 SEM. See Table 4-5 for significance of differences.

71

Table 4-5. Response of maize yields (kg/ha) at MZ12 to cropping system, rain manipulation, and addition of inorganic fertilizer. Within a given year and a given variable (outlined in black boxes), means with different letters are significantly different (p < 0.05).

Variable Level 2008-‐2009 yield (kg/ha)

2009-‐2010 yield (kg/ha)

2010-‐2011 yield (kg/ha)

Cropping system

Maize (M) 2403.15 A 1422.31 A 2268.57 A

Gliricidia (G) 3539.42 B 2746.18 B 4240.88 B

p > F 0.0022 0.0022 <0.0001

Inorganic fertilizer

0 kg N/ha (0) 1867.75 A 1713.78 A 2572.87 A

46 kg N/ha (0.5) 3234.98 B 2092.50 A 3268.77 A,B

92 kg N/ha (1) 3809.12 B 2446.45 A 3922.52 B

p > F 0.0167 0.4651 0.0434

Rainfall

Ambient (0) 3126.95 A 2994.20 A 4041.00 A

Drought (1) 2815.61 A 1174.29 A 2468.44 A

p > F 0.8946 0.1818 0.1330

Block

Block 1 3778.66 A 2875.12 A 3575.80 A

Block 2 3374.68 A 2353.00 A 3146.44 A

Block 3 1760.51 B 1024.61 B 3041.93 A

p > F 0.0101 0.0119 0.5411

Whole model p > F 0.0066 0.0022 <0.0001

Interactions of interest

(values are LSMs)

Crop × Fertilizer p = 0.0253 M,0: 1127.51 A M,0.5: 3573.13 B M,1: 4930.94 B G,0: 4226.74 B G,0.5: 4511.58 B G,1: 4302.06 B

Crop × Fertilizer p = 0.1017 M,0: 1088.30 A M,0.5: 2570.90 B M,1: 2980.35 B G,0: 3921.01 B G,0.5: 3195.85 B G,1: 3494.29 B

Crop × Fertilizer p = 0.0357 M,0: 1132.01 A M,0.5: 2800.14 B M,1: 3836.78 B,C G,0: 4655.89 C G,0.5: 4379.54 C G,1: 4650.42 C

Crop × Rain p = 0.5482 M,0: 908.41 A M,1: 1346.61 A G,0: 4324.98 B G,1: 4128.50 B

Crop × Rain p = 0.0703 M,0: 1204.33 A M,1: 972.27 A G,0: 4938.27 B G,1: 2903.75 A

Crop × Rain p = 0.5588 M,0: 1563.35 A M,1: 700.66 A G,0: 5333.12 B G,1: 3978.67 B

72

4.3.2.2 Maize reproductive characteristics

Although maize grain yield is the most important result from a practical perspective, from a theoretical perspective it is useful to understand the underlying processes that affect maize yield. Yield can be reduced by adverse conditions during any of the following growth stages: vegetative growth, cob formation, fertilization, grain-filling, and grain drying.

This experiment quantified the success of all five of these developmental stages. Problems arising during vegetative growth (causing stand loss) and grain drying (causing rotten grain) were found to be minor and not related to any of the treatment effects. Accordingly, this discussion focuses on the middle three stages: cob formation (number of cobs per plant), fertilization (number of kernels per cob), and grain-filling (thousand kernel weight). Figure 4-9 displays a sample of cobs from MZ12 illustrating the results of variation in these processes.

Table 4-6 summarizes treatment effects on maize reproductive characteristics. (Data for 2008-2009 are not reported due to the minimal effects of the drought treatment in that year.)

Drought. In 2009-2010 and 2010-2011, drought had a major impact on all three of these stages of reproductive development. In drought plots, plants had fewer cobs, cobs had fewer grains, and grains were smaller. (The first of these effects was unexpected because the drought was imposed at anthesis, after cob formation had begun. It may be an artifact due to the production of small cobs that were overlooked in the counting process.) The most highly significant difference (p <0.0001) was seen for thousand kernel weight (TKW). This suggested that water stress from the drought treatment successfully affected the process of grain-filling, as intended.

Table 4-6. Response of maize reproductive characteristics to cropping system, rain manipulation, and addition of inorganic fertilizer. Within a given year and a given variable (outlined in black boxes), means with different letters are significantly different (p < 0.05). n.s. = not significant.

2009-‐2010 2010-‐2011

Variable Level Cobs per plant

Grains per cob TKW (g) Cobs per

plant Grains per cob TKW (g)

Cropping system

Maize (M)

n.s.

144.35 A 276.03 A 0.7652 A 155.25 A 384.06 A

Gliricidia (G) 217.98 B 320.75 B 0.8954 B 245.88 B 427.90 B

p > F 0.0198 0.0009 0.0078 <0.0001 0.0091

Inorganic fertilizer

0 kg N/ha

n.s. n.s.

274.16 A 0.7064 A 163.14 A

n.s. 46 kg N/ha 308.42 B 0.8788 B 199.96 A,B

92 kg N/ha 312.59 B 0.9057 B 238.61 B

p > F 0.0287 0.0023 0.0124

Rainfall

Ambient 0.7760 A 225.13 A 329.90 A 0.8868 A 227.44 A 440.27 A

Drought 0.5798 B 137.21 B 266.88 B 0.7739 B 173.70 B 371.68 B

p > F 0.0173 0.0063 <0.0001 0.0192 0.0092 0.0001

73

(a)

(b)

(c)

Figure 4-11. Gravimetric soil moisture (0-20 cm) as a function of drought treatment and cropping system. Trends are shown for (a) 2008-2009; (b) 2009-2010; (c) 2010-2011. Error bars are ± 1 SEM.

74

Other factors. Cropping system and fertilization level also affected the success of cob formation, fertilization, and grain-filling. The presence of Gliricidia conferred beneficial effects on all three of these processes in both years (with the exception of cobs per plant in 2009-2010). Nitrogen fertilizer had similar but weaker effects, suggesting that Gliricidia may be benefitting these reproductive processes through other mechanisms in addition to the alleviation of nutrient stress.

4.3.3 Soil and microclimate

The results above indicate that Gliricidia confers a similar proportional benefit to maize yield under both ambient and drought conditions, thus providing no evidence that the trees are directly protecting maize from drought stress under these conditions. However, it is still useful to explore possible effects of Gliricidia on soil moisture and microclimate, as these effects (or lack thereof) might become relevant to maize production under other circumstances.

4.3.3.1 Soil moisture

Soil moisture was measured several times during the latter part of each growing season, shortly before and during the deployment of the rain shelters. Clear effects of the drought treatment can be seen (Figure 4-11). (The data reported here are for 0-20 cm depth; samples from 20-40 cm depth were also collected, but they showed very similar trends to the 0-20 cm samples.)

In all three years, the drought treatment significantly decreased soil moisture within one month of its initiation. This effect was most pronounced in 2009-2010, as would be expected due to the higher interception fraction in that year (Section 4.2.2). However, the drought treatment also significantly (p = 0.0060) decreased soil moisture in 2008-2009, with a decrease of greater magnitude than that seen 2010-2011. This indicates that the lack of drought effects on maize yield in 2008-2009 was due to a mismatch in timing rather than an insufficient amount of rain interception per se.

As Figure 4-11 shows, Gliricidia had no impact on 0-20 cm soil moisture throughout the growing season (nor on 20-40 cm soil moisture; data not shown). Soil moisture levels in Gliricidia plots were consistently indistinguishable from those in sole maize plots.

The level of nitrogen fertilizer also had no effect on soil moisture (data not shown). Significant block effects were seen, with soil moisture decreasing from Block 1 to Block 3, but this effect was inconsistent over time (it occurred only in March 2009, January 2010, and March 2010).


Air temperature was measured in 10 plots over an eight-day period at the end of the 2009 growing season (9-16 May 2009). The primary purpose of this dataset was to provide information on the microclimatic effects of the rain shelters (Chapter 3); maize had already been harvested by this time, reducing the relevance of these data to maize production.

Figure 4-12 shows air temperature over three typical days during this time period. Throughout most of the day, temperature was indistinguishable in maize and Gliricidia plots; however, daily maxima were approximately 2 ºC cooler in Gliricidia plots. Averaged over eight 24-hour periods, Gliricidia plots were 0.508 ± 0.034 ºC cooler than sole maize plots (mean ± SEM). In

75

addition, maize plots spent an average of 1.09 hours per day over 30 ºC, while in Gliricidia plots the average was only 0.31 hours per day.

Figure 4-12. Air temperature in maize monoculture plots and Gliricidia plots over several representative days at the end of the growing season (12-14 May 2009). Each line represents an average of five plots of that type.

Figure 4-13. Mid-afternoon soil temperature (0-5 cm depth) as a function of cropping type and drought treatment at two time points in March 2010. Error bars represent ± 1 SEM.

Despite these relatively clear effects, it should be noted that the effect of Gliricidia on air temperature likely varies a great deal throughout the growing season due to the pruning regime of intercropped Gliricidia (Figure 4-3). These results are not likely to be representative of the earlier part of the growing season when Gliricidia is pruned to a stump.

76


Soil temperature was monitored on two occasions (mid-afternoon on 15 March and 20 March 2010). Similarly to air temperature, soil temperature was significantly cooled by the presence of Gliricidia, with an average decrease of approximately 5 ºC (Figure 4-13). The effect of drought can also be seen: soil temperatures in drought plots were approximately 3 ºC higher than in ambient plots, though Gliricidia appeared to slightly moderate this effect. The higher soil temperatures in drought plots are most likely due to a lack of evaporative cooling from the soil, as the rain shelters themselves had no detectable microclimatic effects (Chapter 3).


Gliricidia significantly reduced evaporation rates at the soil surface level (Figure 4-14) on two out of the three days on which data were collected. Variations in ambient temperature and wind speed may account for the inconsistency of this effect from day to day.

4.3.4 Historical maize yield, 1993-2006

Before turning to a discussion of the current results, it will be informative to view them in the context of historical data. As mentioned in Section 4.1.2, MZ12 is a long-term field trial that has been used for various investigations on Gliricidia management, nutrient inputs, and soil characteristics. Data on maize yield are available from 1993 to 2006.

These data are also discussed in Chapter 6, where it is shown that over this 14-year period the relationship between rainfall and maize production at MZ12 is modestly negative (Figure 6-7). Over the years of this dataset, excess rain (leading to waterlogging and nutrient leaching) appears to be a greater problem than insufficient rain. With this in mind, the data were re-examined to determine how interannual variations in rainfall modulate the effects of Gliricidia and fertilizer on maize yield.

Response ratio as a function of rainfall. For each year, the response ratio (the ratio of maize yield in plots with a given input and without that input) was calculated and plotted against that year’s growing-season rainfall (1 Nov - 30 Apr). Results are shown in Figure 4-15 and Figure 4-16. In both Gliricidia plots and fertilized plots, the response ratio increases with increasing rainfall, indicating that both of these inputs convey greater benefits to maize yield in wet years than in dry years.

Due to the limited number of data points, no attempt was made to model these relationships non-linearly; however, for nitrogen fertilizer and possibly also for Gliricidia, the positive

Figure 4-14. Effect of Gliricidia on pan evaporation (in mm per 24 hours) over three days at the end of the 2009 growing season.

77

relationship would be expected to reverse once rainfall reaches a sufficient level to cause excessive nutrient leaching.

Figure 4-15. Response ratio of fertilizer as a function of growing season rainfall (based on MZ12 maize yield data, 1993-2006). (N1=92 kg/ha; N0.5=46 kg/ha).

Figure 4-16. Response ratio of Gliricidia as a function of growing season rainfall (based on MZ12 maize yield data, 1993-2006).

Interannual variability in yield. The 1993-2006 maize yield data were also examined for interannual variability as a function of cropping system. Maize yield for each system was averaged for each of the 14 years, and then coefficients of variation (CVs) and their standard errors (Sokal and Rohlf, 1995, p. 138) were calculated for each of these datasets with n = 14. Results are given below.

Including all fertilizer treatments:

Sole maize: CV = 51.78% ± 10.15% Gliricidia: CV = 28.24% ± 5.54%

Including only unfertilized plots:

Sole maize: CV = 33.47% ± 6.56% Gliricidia: CV = 28.34% ± 5.56%

It can be seen that fertilizer has no impact on the variability of yields in Gliricidia plots, and that in the absence of fertilizer, sole maize plots and Gliricidia plots are similarly variable. However, fertilized sole maize plots show significantly greater interannual variability than do Gliricidia plots. The lack of interannual variability in unfertilized sole maize plots may be due to the fact that severe nitrogen limitations mute the effects of other environmental disturbances.

78

4.4 Discussion

4.4.1 Review of hypotheses

It is now appropriate to revisit the seven hypotheses described in Section 4.1.3 and evaluate the extent to which they were supported or refuted.

Hypothesis 1. Drought will have a proportionally greater negative effect on maize yield in maize monoculture plots than in Gliricidia intercropping plots.

There was no statistically significant Gliricidia-drought interaction either of the years of successful drought manipulation. The non-significant trends differed by year: in 2010-2011, maize yield reduction due to drought was modestly greater in monoculture plots than in Gliricidia plots. However, in 2009-2010, a near-significant interaction effect (p = 0.0703; Table 4-5) suggested that drought actually caused proportionally greater yield reductions in Gliricidia plots than in sole maize plots. Thus, Hypothesis 1 was not supported.

Hypothesis 2. The negative effect of drought on maize yield will manifest as a decrease in thousand kernel weight, due to the imposition of the drought during anthesis and grain-filling.

Hypothesis 2 was strongly supported (p = 0.0001; Table 4-6). However, impacts on grain-filling did not appear to completely explain the effects of this drought on maize yields, as drought also negatively affected the number of grains per cob (suggesting that pollination efficacy was reduced) and the number of cobs per plant.

Hypothesis 3. Relative impacts of drought on maize yield will be similar or even more severe in fertilized monoculture plots as compared to unfertilized monoculture plots; in other words, nitrogen fertilizer alone will not convey any protection against drought effects.

Hypothesis 3 was supported, as maize yield showed no significant interactions between N fertilizer level and drought treatment.

Hypothesis 4. After the imposition of the drought, Gliricidia plots will maintain higher soil moisture in the maize rooting zone (0-20 cm) than will maize monoculture plots.

Hypothesis 4 was not supported. There was no evidence for an effect of Gliricidia on soil moisture at any depth at any time, with or without the rain manipulation (Figure 4-11).

Hypothesis 5. Gliricidia plots will have lower air temperature, soil temperature, and pan evaporation rates than maize monoculture plots.

All three components of Hypothesis 5 (air temperature, soil temperature, and pan evaporation) were supported by the available data. However, it is important to note that microclimatic data were collected near the end of the growing season, potentially making these results less relevant to maize production.

Hypothesis 6. In years of significantly below-average growing season precipitation (<600 mm from 1 November to 30 April), the maize yield advantage of Gliricidia plots will exceed that of their yield advantage in normal years.

79

The historical record (1993-2006) provided three years in which growing-season rainfall was below 600 mm. In these years, the response ratio of Gliricidia plots was poorer on average than their response ratio in normal years (Figure 4-16), in one case even remaining below 1.0 (indicating that Gliricidia plots had lower yields than monoculture plots). Overall, there was a clear trend of increasing response ratio with increasing rainfall (in concordance with Sileshi et al., 2011). Thus, Hypothesis 6 was not supported.

Hypothesis 7. Interannual variability of maize yield from 1993-2006 will be lower in Gliricidia plots than in maize monoculture plots.

The extent to which this hypothesis was supported depended on whether fertilized plots or unfertilized plots were considered. Considering only unfertilized treatments, interannual variability was identical in Gliricidia and sole maize plots. Considering all treatments, interannual variability was significantly lower in Gliricidia plots. This conflicts with the finding by Snapp et al. (2010) that inorganic fertilizer reduces interannual variability of maize yields in Malawi.

However, interannual variability alone may not be a useful metric of cropping system reliability, because low variability can indicate consistently poor yields (as is the case with unfertilized sole maize). Thus, the status of Hypothesis 7 remains ambiguous.

4.4.2 Implications for use of Gliricidia in drought conditions

Although the results from a three-year field experiment in a single location must be interpreted tentatively, several practical recommendations arise from the findings above.

4.4.2.1 Gliricidia can convey yield benefits even in drought conditions

In one of the two drought years (2010-2011), Gliricidia drought plots yielded significantly more maize grain than did monoculture drought plots. Yields of Gliricidia drought plots also exceeded yields of monoculture drought plots in 2009-2010, though the difference did not attain statistical significance. Thus, it appears that Gliricidia can increase maize yields even under conditions of prolonged drought.

This is a more modest claim than saying that Gliricidia helps to directly protect maize from the effects of drought. No evidence was found for a direct protective effect; rather, over the two successful years of drought treatment in this experiment, the proportional negative impact of drought on Gliricidia plots was equal to or perhaps even greater than the impact of drought on monoculture plots. This finding was corroborated by long-term maize yield data (1993-2006) indicating that Gliricidia conveys greater yield benefits under normal conditions than under drought conditions.

Nevertheless, for a farmer choosing between cropping systems, the most important consideration may simply be that maize-Gliricidia systems outperform maize monocultures under both normal conditions and drought conditions.

4.4.2.2 Gliricidia may help to moderate soil temperature and air temperature

Toward the end of the growing season, when Gliricidia trees in the intercropping system have accumulated substantial aboveground biomass (Figure 4-3e), their moderating effects on

80

microclimate appear to be substantial, with reductions of about 2 ºC for daily maximum air temperatures and reductions of 5 ºC for soil surface temperatures.

This effect could be helpful in regions where maize production is potentially threatened by temperature increases (Chapter 6), but it is unclear whether the timing of these microclimatic effects is appropriate to protect maize during its most vulnerable phenological stages. Further research is needed to determine the microclimatic effects of Gliricidia intercropping systems throughout the growing season.

4.4.2.3 There is not enough evidence to promote Gliricidia for drought protection

Despite recent optimism that agroforestry systems will, in general, help farmers adapt to warmer and drier climates (Verchot et al., 2007; Syampungani et al., 2010), evidence to support this idea has been mostly speculative rather than concrete. The current study has empirically tested whether Gliricidia intercropping can increase the drought resilience of maize production, and has found little evidence to suggest that it does. Soil moisture would have been the most obvious pathway by which Gliricidia might have exerted a protective effect; however, Gliricidia trees were shown to have no effect on soil moisture throughout the growing season.

Although the results of this experiment do not support the drought-protection hypothesis, by no means do they represent the most negative possible outcome. In the presence of belowground competition for water, fertilizer trees can devastate maize yields. Smith et al. (2004) cite research in semi-arid Kenya in which alley cropping with Senna spectabilis and Leucaena leucocephala reduced maize yields by 39% and 95%, respectively. Careful selection and management of fertilizer tree species is essential to avoid water competition (Smith et al., 2004). The current study indicates that Gliricidia poses no threat to maize production under drought conditions and can therefore be a good choice in locations with uncertain rainfall.

Furthermore, any agricultural technology that boosts food production and farm income in normal years can, in theory, help to reduce farmers’ vulnerability to the impacts of environmental disturbances by providing them with additional resources (an idea discussed in further detail in Chapter 1). In this light, the beneficial effects of Gliricidia on maize yield can be framed as a benefit for climate resilience. That said, its benefits (both in drought years and in non-drought years) should be compared to the benefits of other soil fertility management practices, including annual legumes and inorganic fertilizer.

4.4.3 Unanswered questions and future work

This experiment represents only a first step toward characterizing the drought resilience of fertilizer tree systems; many other questions remain to be answered.

4.4.3.1 Effect of tree species, location and soil type

It remains to be seen to what extent the satisfactory drought performance of Gliricidia intercropping can be generalized to other fertilizer tree systems. However, given the variety of tree species used throughout the tropics, and the potentially crucial influences of climate and soil type, broad generalization is probably unwise. (The failed Leucaena system in Kenya mentioned above is an example of an inappropriate combination of species, site, and management practices.)

81

To some extent, this question can be addressed by meta-analysis of existing trials (as was done by Sileshi et al., 2011). The extensive research in fertilizer tree systems (Akinnifesi et al., 2010) conducted in southern Africa over the past several decades – both on-farm and off-farm – represents a wealth of information waiting to be tapped for this goal. However, meta-analysis has its limitations and cannot fully substitute for manipulation experiments or simulation models (as discussed in Chapter 1).

4.4.3.2 Effect of drought timing and magnitude

The temporal pattern of rain exclusion in this study was partly determined by the configuration of the rain shelters. Results may have differed with a different pattern of rain exclusion (for example, a constant 30% or 60% interception throughout the growing season rather than a single 100% interception during anthesis and grain-filling). In light of the subtle and uncertain changes in rainfall predicted for southern Africa (see Chapter 6), it will be useful to design rain exclusion experiments for this region that mimic current interannual variability as well as (or instead of) future change.

4.4.3.3 Interactions of drought, temperature, and other global changes

This experiment manipulated only a single climatic variable, rainfall. However, future climate change may manifest as simultaneous changes in temperature, CO2 concentration, and a variety of rainfall characteristics (perhaps including average annual total, onset and cessation of rainy season, maximum dry spell length, and intensity of single events). This daunting combination of factors can perhaps be most thoroughly explored with simulation models. More field experiments (not only on agroforestry systems but on a variety of other agricultural systems) are needed to provide a basis for such modeling.

4.4.3.4 Comparison of Gliricidia to other organic inputs

In most experiments on fertilizer tree systems in southern Africa, unfertilized maize is used as a control. Fertilized maize is sometimes used as an additional control. However, it is relatively rare to include annual legume intercrops as a control. Nevertheless, this is a potentially useful comparison, because annual legumes (especially grain legumes such as groundnuts and beans) are commonly used by and easily accessible to farmers in the region. Although perennial legumes generally convey greater benefits to maize yield (Sileshi et al., 2009), they have been met with a lack of enthusiasm due to farmers’ strong preference for annual grain legumes (Bezner-Kerr et al., 2007).

Thus, when evaluating the drought resilience of Gliricidia or any other fertilizer tree system, it would be prudent to compare it to a system involving an annual legume. In theory, perennial crops have several advantages over annual crops in drought conditions, but this needs to be verified in the southern African context.

4.5 Conclusions

This chapter has shown that Gliricidia intercropping systems increase maize yield under both normal rainfall and prolonged drought conditions. However, similarly to inorganic fertilizer, the yield benefits of Gliricidia are greater under conditions of adequate rainfall. Gliricidia trees

82

intercropped with maize do not appear to influence soil moisture either positively or negatively, and thus their effect on the drought resilience of maize is relatively neutral.

Although these results do not warrant the promotion of Gliricidia specifically for adaptation to climate variability or change, they do suggest that Gliricidia intercropping (unlike some other fertilizer tree systems) is characterized by a high level of complementarity in water use, i.e., the trees are able to meet their water requirements without depriving the crop of needed water. This implies that Gliricidia intercropping is likely to remain useful even in adverse and uncertain climatic conditions. Much remains to be learned about the performance of other fertilizer trees under drought conditions and under other types of climatic stresses.


The author is grateful to the ICRAF researchers who established and managed the long-term Gliricidia intercropping experiment at MZ12: Prof. Jumanne Maghembe (from 1991 to 1999) and Prof. Festus Akinnifesi (from 2000 to 2011). Dr. Wilkson Makumba co-managed the trial from 1999 to 2002. They were assisted by field staff Thomson Chilunga, Konisaga Mwafongo, and Aubrey Semu. The work of these individuals provided an essential foundation for the success of the current project. It was a privilege to be granted access to such a well-characterized and long-term field site.

During the 2010-2011 growing season, when the author was not in Malawi, Simon Mng’omba and Chikumbutsa Kwakwala managed the experiment and collected the data. This project could not have been completed without them; many thanks are due them for their painstaking efforts.

83

Chapter 5. Effects of drought and planting time on establishment of Gliricidia and Tephrosia seedlings

Abstract Little is known about the effects of adverse climatic conditions on the performance of tree seedlings in agroforestry systems. This experiment explores the effects of two potentially interacting stressors (drought and late planting) on the survival and growth of seedlings in three different fertilizer tree agroforestry systems in southern Malawi: Gliricidia-maize intercropping, Tephrosia-maize relay intercropping, and Tephrosia improved fallows. The experiment was conducted over three growing seasons (2008-2011), one of which (2009-2010) included an artificial drought created with rain exclusion shelters.

The effects of drought and late planting varied considerably depending on the species and the variable in question. The survival of Gliricidia seedlings remained close to 100% throughout the experiment, but height growth of Gliricidia was impaired by both drought and late planting. Tephrosia seedlings showed no height response to drought, but their survival was reduced by a combination of drought and late planting. Both species were able to largely recover from these stresses by the beginning of the next growing season; accordingly, effects on seedling biomass production and subsequent maize yields were generally minimal.

These results suggest that fertilizer tree systems are useful for soil fertility restoration even when impeded by drought and poor management practices at the seedling establishment stage. However, the species-specific differences seen here suggest that generalizations about these systems should be done with care. More work is needed on the resilience of fertilizer tree seedlings to climatic stresses in other contexts.


An organism’s vulnerability to unfavorable climatic conditions often depends on the stage of its life cycle. In trees, young seedlings are usually the most vulnerable. Seedlings’ small stature limits their access to resources (water, light, and nutrients), makes them vulnerable to temperature fluctuations, and limits the carbohydrate reserves that they can use to recover from injury. Furthermore, young seedlings may not have fully developed protective mechanisms such as waxy cuticles or the ability to produce heat-shock proteins. Even in highly K-selected species, death rates of young seedlings are as a rule much higher than those of mature trees.

Thus, studying the effect of a climatic disturbance on mature trees provides an incomplete picture of possible effects on the system over the long term. All forest systems must regenerate – whether every year or on the scale of centuries – and the ultimate effect of a change in climate can only be predicted if its effect on each stage of the trees’ life cycle is understood.

84

In this context, the experiment described in the previous chapter represents only a snapshot of how Gliricidia-maize intercropping systems respond to drought. To more fully assess their suitability under drought conditions, it is important to know how well Gliricidia systems tolerate drought before the trees are fully established. This provided the motivation for establishing a second drought experiment focusing on Gliricidia seedling survival and establishment. The second experiment also provided a opportunity to compare Gliricidia with Tephrosia candida, another “fertilizer tree” with a shorter life cycle.

Before considering the details of this experiment, it will be helpful to briefly review the existing literature on fertilizer tree seedling establishment and to identify knowledge gaps.

5.1.1 Previous work on seedling establishment in fertilizer tree systems

Seedling establishment practices in agroforestry systems vary a great deal according to the type of system and the tree species used (Rocheleau et al., 1988). This section focuses on fertilizer tree agroforestry systems, which have some commonalities with each other despite their important differences.

5.1.1.1 General principles of fertilizer tree establishment

Several characteristics and constraints are common to seedling establishment in most fertilizer tree systems. First, trees are usually planted in the field alongside a food crop (such as maize or millet); accordingly, each seedling must be able to grow in close proximity to the food crop but not disturb nor be disturbed by it. This makes it infeasible to create buffers or catchment zones around each seedling.

Second, as fertilizer tree systems are almost invariably used in subsistence agriculture, the methods of their propagation must be suitable for users with limited resources. Practices such as grafting, in-field irrigation, and application of synthetic pesticides are likely to be impractical for the primary users of fertilizer tree technologies.

Third, fertilizer tree systems are rarely integrated with free-ranging livestock. Protecting tree seedlings from browsing livestock is a major theme of the seedling establishment literature (e.g., Love et al., 2009), but this problem generally does not arise in fertilizer tree systems.

Within this broad outline, fertilizer tree systems vary considerably in their methods of seedling establishment. One important variable is the frequency with which trees need to be established. For example, in Tephrosia candida relay intercropping systems (described later in this chapter), trees are planted from seed every year and destroyed at the end of the year. By contrast, in Gliricidia intercropping systems, trees are only planted once and maintained for decades.

Another important variable in fertilizer tree establishment is how the trees are propagated. In increasing order of investment, trees can be direct-sown as seed in the field, propagated from cuttings or stakes, or grown in a nursery and then transplanted as bare-root or potted seedlings. Some fertilizer tree species only thrive under one of these methods; for example, Tephrosia species are almost invariably planted in the field as seed, while Sesbania sesban is generally transplanted in pots. Others, such as Gliricidia sepium, are suited to a variety of methods.

85

Development and testing of standard practices for seedling establishment in fertilizer tree systems has largely occurred outside the formal literature. Most published studies on fertilizer trees take seedling establishment practices as a given rather than making them the focus of inquiry. Only a few studies have explicitly focused on the seedling stage of fertilizer tree systems; these are reviewed in the following section.

5.1.1.2 Studies on fertilizer tree seedlings

Although a variety of tree species can be used as fertilizer trees, here we will focus on several species commonly used in southern Africa: Gliricidia sepium, Sesbania sesban, Tephrosia candida, and T. vogelii. Another important species is Acacia albida (Faidherbia albida), but as A. albida individuals on farmland are generally self-sown wild trees (Rocheleau et al., 1988; Saka et al., 1994), we will not consider their propagation here. (Perennial grain legumes such as Crotalaria grahamiana and Cajanus cajan are also widely used, but as soil fertility improvement is not their primary goal, their management considerations differ somewhat.)

Gliricidia. The most comprehensive study on Gliricidia propagation methods was conducted by Chintu et al. (2004) in a maize intercropping system in Zambia. They found that potted and bare-root seedlings had equally high rates of survival (90-100%) and biomass production after two years, while stem cuttings fared worst (30% survival) and direct-seeding was intermediate (56%). However, they concluded that farmers might still prefer direct seeding for cost reasons.

Several studies in Central American pastures (Zahawi, 2005; Zahawi and Holl, 2009) concluded that Gliricidia can be effectively propagated as stakes. They reported that stakes accumulated much more biomass in the first year than did seedlings, while the establishment costs of stakes were one-half to one-tenth that of seedlings. The first study (2005) found stake survival rates of nearly 100%, though the second study (2009) found survival rates of 15-85%.

Gliricidia is used for many purposes other than fertilizer tree systems; for example, its high seedling survival rate, coupled with its drought tolerance and rapid growth, makes it a good candidate for reforesting marginal lands in the subhumid tropics (Foroughbakhch et al., 2006).

Sesbania. Like Gliricidia, Sesbania can be propagated as potted seedlings, bare-root seedlings, stem cuttings, or by direct seeding. However, the success rates of non-seedling propagation methods are quite low. Kwesiga et al. (1999) concluded that seedling-based methods are the only economically viable means of establishing Sesbania in the field, while other authors have argued that direct seeding (Owuor et al., 2001) or stem cuttings (Oduol and Akunda, 1988) may be practical in some circumstances.

Regardless of the propagation method, Sesbania tends to be vulnerable to pests (Sileshi et al., 2008; Reubens et al., 2009) and to require inoculation with specific rhizobia (Mafongoya et al., 2003), so management of Sesbania seedlings can be difficult despite the species’ high potential for biomass production. Ikerra et al. (2001) found that Sesbania biomass production was not strongly dependent on planting density (there was little difference between 7400 versus 14800 trees per hectare), suggesting that variations in mortality rate may be mitigated by density-dependent growth responses of the surviving trees.

86

Tephrosia. The two Tephrosia species commonly used in fertilizer tree systems (T. candida and T. vogelii) are invariably propagated by direct seeding. This characteristic increases their ease of use and makes them more attractive to farmers (Keil et al., 2005; Mafongoya and Kuntashula, 2005). Between the two species, T. candida is generally superior in its biomass production, its weed suppression, and its effect on maize yield (Mafongoya et al., 2003).

Although germination rates can be improved by pre-treating Tephrosia seed in boiling water (Babayemi et al., 2003), generally no seed treatment is necessary. Farmers have found that they can compensate for lower germination rates by planting three Tephrosia seeds per hole rather than one (Ngugi, 2002).

5.1.1.3 Knowledge gaps for fertilizer tree seedlings

As previously mentioned, the development of standard practices for establishment of fertilizer trees is generally not well-documented in the peer-reviewed literature. There is also little information about how deviations from standard practices and ideal conditions may affect the success of seedling establishment and thus the success of the systems themselves. Several examples are described below.

Labor availability often creates disparities between researcher-managed trials versus on-farm adoption of fertilizer tree systems. In contrast to households with adequate labor, labor-limited households tend to find fertilizer trees less profitable (Sirrine et al., 2010) and are less likely to adopt them (Ajayi et al., 2003). This is partly because fertilizer trees tend to have high labor requirements at the onset of the rains, when there are many other demands on farmers’ time (Mafongoya, 2006; Sirrine et al., 2010). As a result, farmers may end up planting fertilizer tree seeds or seedlings later in the season than is recommended. However, the effects of this delayed timing have not been studied (F. Akinnifesi, personal communication, 2008).

The effects of unfavorable climatic conditions on establishment of fertilizer trees have only been reported in the context of interannual variability, and even so, such reports are rare. Ikerra et al. (2001) noted that Sesbania biomass production fluctuated greatly over three years of a field trial at Makoka, Malawi, but did not specify the patterns or causes. Kwesiga et al. (1999) described the outcome of on-farm trials of fertilizer trees in eastern Zambia in 1994-95, a year with poor rainfall: “We estimate that 60% survival of the fallow species in the first three months is required for satisfactory biomass production at the end of two years. In 1994-95, 82% of the surveyed farmers for Tephrosia, 63% for pigeonpea, and 48% for Sesbania achieved this level of survival.” Three-quarters of these farmers identified drought as a major impediment to seedling establishment.

Despite hints that climate variability is an important factor in these systems’ success, only one controlled experiment looking at climate effects on fertilizer tree seedlings – or indeed any type of agroforestry seedling – could be found at the time of writing. That experiment (Esmail and Oelbermann, 2011) found that a 2-3ºC temperature increase boosted growth of Cedrela odorata and Gliricidia sepium seedlings, while a CO2 concentration of 800 ppm lowered leaf N content but conferred no growth benefit (in contrast to the findings of Tissue et al., 1997). However, the 2011 study was conducted in growth chambers, and the authors urged the use of field trials – especially with soil moisture included as a variable – to provide a more complete picture of seedling response to climate change.

87

This climate-related knowledge gap persists not just for agroforestry seedling establishment, but for all aspects of agroforestry systems (Rao et al., 2007; Verchot et al., 2007). Neufeldt et al. (2012) concluded that “for most tree species grown in agroforestry systems, virtually no information on climate responses is available,” in part due to the difficulty of conducting controlled climate experiments on trees as compared to annual crops.

In summary, very few studies have explicitly examined how departures from ideal climate or management practices affect the efficacy of fertilizer tree systems. Furthermore, the process of seedling establishment in these systems has usually been taken for granted. The experiment described in this chapter aims to address some of these gaps.

5.1.2 Goals and hypotheses of this experiment

5.1.2.1 Comparing drought response of young versus mature Gliricidia

Goal. The foremost goal of this experiment was to examine the drought response of newly established Gliricidia-maize intercropping systems in contrast to the drought response of the long-established systems described in the previous chapter. In the previous chapter, surface-level rain exclusion did not affect survival or growth of mature Gliricidia, which was to be expected due to their rooting depth. However, it was not known whether a similar perturbation would affect younger Gliricidia and, if so, how this might affect maize yield.

Hypothesis 1. Survival, growth, and biomass production of newly established Gliricidia seedlings will be negatively affected by a prolonged drought.

Hypothesis 2. Negative effects of drought on seedling performance will translate into negative effects on maize yield in the following year.

5.1.2.2 Comparing drought response of Gliricidia versus Tephrosia

Goal. The second goal was to compare the drought response of Gliricidia intercropping systems with that of a different species, Tephrosia candida. Though the two species are similar in many ways (they are both fast-growing, short-lived tropical leguminous trees), they have different establishment methods (seedlings versus direct sowing) and are grown in different field configurations. Gliricidia and Tephrosia are both widely available in Malawi, and better knowledge of their drought resistance may help target their appropriate promotion.

Hypothesis 3. The magnitude of drought effects on seedling survival, growth, and biomass production will differ between Gliricidia and Tephrosia. (No assumption was made about which species would be more greatly affected.)

5.1.2.3 Comparing Tephrosia relay intercropping versus improved fallows

Goal. The third goal was to compare two different Tephrosia candida intercropping systems: relay intercropping (RI) and improved fallows (IF). In the former system, Tephrosia is annually undersown in the maize rows; in the latter system, it is grown without maize in a monoculture fallow for two years. This comparison was intended to clarify whether the presence of maize has a beneficial or detrimental effect on Tephrosia seedling performance and whether that effect might depend on water availability.

88

Hypothesis 4. Due to competition with the maize crop for water and light, Tephrosia seedlings in relay intercropping systems will have lower survival and growth rates than those in improved fallows.

Hypothesis 5. The performance gap between Tephrosia relay intercropping and Tephrosia improved fallows will become more pronounced under drought conditions.

5.1.2.4 Comparing trees planted on time versus trees planted late

Goal. The fourth goal was to measure the effect of delayed planting of seeds or seedlings on each of the three systems described above: Gliricidia intercropping, Tephrosia relay intercropping, and Tephrosia improved fallows. This one-month delay was intended to simulate the conditions that a labor-constrained farmer might face.

Hypothesis 6. Delaying planting by one month will reduce survival, growth, and biomass production of Gliricidia and Tephrosia, with corresponding reductions in maize yield.

Hypothesis 7. Late-planted seedlings will be more negatively affected by drought than will seedlings planted on time.

5.1.2.5 Assessing fertilizer tree performance under optimal conditions

Goal. The fifth goal, auxiliary to the other goals, was to provide data on the performance of these fertilizer tree systems under optimal conditions. Although such data are already reported in the literature, it was expected that more baseline data would nonetheless be useful to affirm or refute the potential of fertilizer tree systems to improve maize yields in this location.

Hypothesis 8. Gliricidia intercropping will have no effect on maize yield for the first two years, after which it will have a strong positive effect. Tephrosia relay intercropping will have no effect in the first year, after which it will have a modest positive effect. Tephrosia improved fallows will have a strong positive effect in the third year (the first post-fallow maize crop).

5.2 Methods

5.2.1 Field establishment and experimental layout

Due to limited land availability at Makoka Agricultural Research Station, the experiment’s size and layout were constrained. The following design thus reflects some practical compromises.

5.2.1.1 History and background of field site

The experiment was conducted at Makoka Agricultural Research Station (15º31' S, 35º13' E, 1030 m elevation, 20 km southwest of Zomba, Malawi). A complete site description is provided in Chapter 2. The plot, known as “Nkula Field” by Makoka staff, is located at the far eastern end of the research station, adjacent to Nsamira village. Nkula Field is a one-hectare strip on a gentle (<3%) slope, the upper two-thirds of which is under continuous use for soybean and sweet potato varietal trials. The lower one-third (0.35 ha), where the seedling experiment was established, had been left as natural fallow for an unknown period of time.

89

Nkula Field is directly adjacent to MZ21, an experimental site established in 1995 by Paxie Chirwa and Wilkson Makumba for studies of Gliricidia and pigeonpea intercropping (Chirwa et al., 2003a; Chirwa et al., 2006b; Chirwa et al., 2007; Makumba et al., 2007; Makumba et al., 2009). Their measurements of soil physical and chemical properties at this site are reproduced in Table 5-1. An overview of the field is shown in Figure 5-1.

The seedling experiment was established in November 2008, shortly prior to the onset of the rainy season. First, a tractor was used to plow, harrow, and ridge the field along its entire length; second, plots were demarcated and paths between the plots were leveled by hand with hoes; third, the ridges were refined by hand to correspond with the spacing typically used in farmers’ fields (ridges 30 cm high and 75 cm apart). Heavy weed growth after the first rainfall suggested that the land had previously been fallowed for several years.

5.2.1.2 Initial layout of the experiment

The experiment was laid out in a randomized complete block design with four replicates. Treatments included the following variables, each applied at the whole-plot level, in full factorial combination:

Cropping system: (1) Sole maize (control); (2) maize-Gliricidia intercrop; (3) maize-Tephrosia relay intercrop; (4) Tephrosia improved fallow.

Rainfall: (1) Ambient (control); (2) rainfall exclusion (see Section 5.2.4.1 for details).

This design implies (4 replicates × 4 cropping systems × 2 rainfall levels =) 32 plots. However, four additional Gliricidia plots were added (one in each block; two with rain shelters and two without). This helped to equalize the number of Gliricidia individuals with the number of Tephrosia individuals, because Gliricidia is less densely planted than Tephrosia within a plot. Thus, the total number of plots was (32 + 4 =) 36.

The irregular shape of the blocks (Figure 5-2) was due to the limited field space available. Each block was chosen to encompass a portion of the slope conditions (ranging from upslope in Block 1 to downslope in Block 4) and the landscape matrix (ranging from sweet potato cropping near Block 1 to miombo forest in Block 4).

Table 5-1. Topsoil (0-20 cm) properties, MZ21 (adjacent to Nkula Field), Makoka Agricultural Research Station (from Makumba et al., 2009).

Clay (%) 38

Sand (%) 54

Silt (%) 8

Bulk density (g/cm3) 1.55

pH in water (1:2.5) 5.6

Organic carbon (g/kg soil) 9.3

Olsen extractable P (mg/kg soil) 10.3

Exchangeable K (mmol (+)/kg soil) 3.7

Exchangeable Ca (mmol (+)/kg soil) 17.3

Exchangeable Mg (mmol (+)/kg soil) 4.2

90

Figure 5-1. Layout of Nkula Field (as seen from the upper right of Figure 5-2) in April 2009. The far right rain shelter is Plot 1-6. Construction materials and a watchman’s hut are in the foreground.

Figure 5-2. Map of experimental layout at Nkula Field.

91

Figure 5-3. Plot maps of the three types of agroforestry systems used in this experiment. Sole maize plots were configured identically to intercropping plots but without the trees.

For seedling plots, planting time (early vs. late planting) was applied as a split-plot treatment. The axis of division was oriented uphill-downhill (rather than across a contour) in order to avoid systematic differences in water flow or soil moisture between the two halves. Each planting time treatment was applied to equal numbers of left and right sides (designated respectively as

92

“Village” and “Station” sides in Figure 5-2) for each type of plot. Figure 5-3 shows the detailed layout of each plot type.

5.2.1.3 Subsequent modifications in layout

In the first year of the experiment (2008-2009), after the tree seedlings were planted, the rainfall exclusion was implemented belatedly and did not achieve the desired effects (see Section 5.2.4.1). For the agroforestry systems in which trees are maintained for two or more years (Gliricidia-maize intercropping and Tephrosia improved fallow), this meant missing the only opportunity to observe the effects of drought on newly planted seedlings. Thus, a decision was made to remove half of the seedlings in each of these plots in order to start again with newly planted seedlings in the second year (2009-2010).

Because the plots had already been divided on a left / right basis for the planting time treatment (early versus late planting), the only remaining axis of division was uphill / downhill. To equalize possible differences in water flow or other environmental variables, removal and replanting was done in an equal number of uphill halves and downhill halves for each type of plot. As there were nine trees per row (Figure 5-3) the division resulted in split plots of slightly different sizes, with the uphill half containing rows of five trees and the downhill half containing rows of four trees.

Note that this split-plot treatment (2008-09 establishment versus 2009-10 establishment) did not apply to sole maize plots or to Tephrosia-maize relay intercropping plots, because those systems are by definition re-established on an annual basis.

5.2.2 Seedling establishment

Gliricidia and Tephrosia were managed according to advice from ICRAF research staff (F. Akinnifesi and G. Sileshi, pers. comm., 2008) and from employees at the ICRAF nursery at Makoka. All seedlings were propagated from seed obtained from seed orchards at Makoka. Once in the field, seedlings were grown without any irrigation, fertilizer, or pesticide. Species-specific details are given below.

5.2.2.1 Gliricidia

Gliricidia seeds were sown in a raised nursery bed (Figure 5-4) on 25 October 2008 and 22 October 2009, approximately two months before transplanting to the field. During this time, seedlings were shaded, watered, and root-pruned according to standard ICRAF extension procedures (Akinnifesi et al., 2006).

Gliricidia seedlings were removed from the nursery bed and transported to the field by wheelbarrow during times of minimal evaporative demand (early in the morning, late in the evening, or during cloudy or rainy days). They were planted in pre-prepared holes 20 cm deep and 90 cm apart in every other maize row (Figure 5-5). Planting dates are listed in Table 5-2. Seedlings not used at the earlier planting date were maintained in the nursery bed until the later planting date. All batches of Gliricidia seedlings were planted by the author except for the late-planted 2008-2009 seedlings, some of which were planted by nursery staff.

93

Figure 5-4. Propagation of Gliricidia seedlings in a raised nursery bed (November 2009). A machete is being used to prune the roots between each seedling row.

Figure 5-5. Gliricidia seedlings ten days after being transplanted to the field (January 2009).

Figure 5-6. Tephrosia germinating in the field.

Figure 5-7. Six-week-old Tephrosia relay intercrop.

Figure 5-8. Three-month-old Tephrosia improved fallow.

5.2.2.2 Tephrosia

Tephrosia seeds were sown directly in the field (Figure 5-6) at or shortly after the time of maize sowing (Table 5-2). Seeds were planted three to a hole, approximately 2 cm deep, at planting stations 90 cm apart. In relay intercropping plots, seeds were planted on the side of each ridge,

94

halfway between two maize plants (Figure 5-7). In improved fallow plots, seeds were planted on the top of each ridge, and no maize was present (Figure 5-8).

Several weeks after germination, Tephrosia seedlings were thinned to two per planting station. Missing seedlings were filled in by transplanting from other nearby planting stations. Transplanting rules varied from year to year: in 2008-2009, planting stations with only one seedling received a transplant to bring the total number to two, whereas in 2009-2010, planting stations with only one germinant were not supplemented by transplanting, and in 2010-2011, no transplanting was carried out. This change in protocol was motivated by the observation that transplanted Tephrosia seedlings did not thrive.

5.2.3 Field and seedling management

5.2.3.1 General management

Maize (hybrid variety DeKalb 8053) was planted each year in all plots except Tephrosia improved fallows, according to the standard local practices described in Chapter 2. Weeding was done by hand twice during each growing season (at approximately three and eight weeks after maize planting). No inorganic fertilizer or pesticide was applied to the plots. Maize was harvested using the same procedures described in Chapter 4.

5.2.3.2 Gliricidia intercropping

As described in Section 5.2.1.3, the Gliricidia seedlings in half of each plot were cut down at approximately 10 months of age, at the beginning of the 2009-2010 growing season. Their biomass was returned to the plot, and new seedlings were planted in their place. Regrowth did occur from the roots and buried branches of the old seedlings (Figure 5-9), but any regrowth was periodically cut and buried throughout the subsequent growing season.

The remaining Gliricidia seedlings were left untouched until, at nearly two years of age (October 2010), they were pruned to a stump approximately 30 cm high (Figure 5-10). This practice encourages vigorous resprouting in Gliricidia. At the time of pruning, leaves (including twigs) were separated from large branches; each type of biomass was weighed in the field, and a subsample was taken to determine moisture content. Leaves were then buried underneath each ridge of soil, while branches were removed from the plot for use as firewood. Below-ground biomass was not measured. Gliricidia were pruned and incorporated again on 24 December 2010, but biomass was not measured.

5.2.3.3 Tephrosia relay intercropping

In each of the three growing seasons, Tephrosia seedlings in relay intercropping plots were left undisturbed until approximately 10 months of age (October or November), at which point they were cut at ground level. Tephrosia candida is a non-coppicing species, so this practice invariably kills the trees. Leaves and branches were separated, weighed, and incorporated into the plot according to the procedure described above for Gliricidia.

95

Figure 5-9. Gliricidia seedling resprouting from roots (at arrow) after an attempt was made to remove it at 10 months of age. (The seedlings behind it are of the same age but uncut.)

Figure 5-10. Two-year-old Gliricidia pruned to a 30-cm stump. In the background are unpruned one-year-old Gliricidia seedlings and a two-year old Tephrosia improved fallow awaiting pruning.

Figure 5-11. Collecting litterfall in a quadrat in a one-year-old Tephrosia improved fallow.

Figure 5-12. Separating and weighing leaves and branches in a two-year-old Tephrosia improved fallow.

96

In November 2009, when the first year’s Tephrosia seedlings were ready to be incorporated, it was observed that a substantial portion of their above-ground biomass had already been shed as litterfall. Thus, an attempt was made to quantify litterfall. (Litterfall from Gliricidia was not measured, as the quantity of Gliricidia litter remaining in the plots at the end of the dry season appeared to be very small.) Litter was collected from two randomly selected quadrats per sampled Tephrosia subplot. Each quadrat was defined by a seedling at its corner, forming a square 90 cm per side (Figure 5-11). Tephrosia litter was sorted from that of other species and dried at 70 ºC to constant weight. Litterfall measurements were not repeated in October 2010, due to time constraints.

5.2.3.4 Tephrosia improved fallow

Tephrosia seedlings in improved fallow plots were left to grow on the same schedule as Gliricidia, with the exception that the seedlings planted in 2008-2009 were cut at ground level and killed at the beginning of the third growing season (October 2010), when they were slightly less than two years of age (Figure 5-12). Their biomass was measured and incorporated as described for Tephrosia relay intercropping (including litterfall in November 2009).

5.2.3.5 Key management dates

The following table (Table 5-2) specifies the dates of key management activities at Nkula Field for the three years of the experiment (2008-2009, 2009-2010, and 2010-2011).

Table 5-2. Dates for key management activities at Nkula Field drought experiment, 2008-2011.

2008-‐2009 2009-‐2010 b 2010-‐2011

Maize Planting 5-‐6 Dec 2008 22 Dec 2009 c 5 Dec 2010

Harvest 1 May 2009 7 May 2010 17 May 2011

Rain exclusion Begin 26-‐28 Mar 2009 25-‐28 Feb 2010 n/a

End 16 May 2009 13 May 2010 n/a

Gliricidia Early planting 5-‐7 Jan 2009 29 Dec 2009 n/a

Late planting 10-‐11 Feb 2009 31 Jan -‐ 1 Feb 2010 n/a

Tephrosia a Early planting 30 Dec 2008 23 Dec 2009 c 4 Dec 2010 (RI)

Late planting 4-‐5 Feb 2009 3 Feb 2010 5 Jan 2011 (RI) a Tephrosia seeds were planted at the same time in relay intercrop plots as in improved fallow plots. b In 2009-‐2010, seedlings were replanted in only half of each Gliricidia plot and each Tephrosia improved fallow plot. In the other half of each of these plots, seedlings planted in 2008-‐2009 were retained. c These dates represented the third and final attempt to plant seed. Previous attempts (on 26-‐27 Nov 2009 and 12-‐14 Dec 2009) failed due to lack of subsequent rain.

97

5.2.4 Rain manipulation, soil moisture, and microclimate

The drought treatment was implemented with open-sided rain shelters installed in half of the plots (see Figure 5-2 for field layout and Figure 5-3 for plot layout). Details of rain shelter design, construction, and performance are described in Chapter 3. Unlike at MZ12, at Nkula Field the assumption was made that a rain shelter in one plot would not interfere with adjacent plots (due to the two-meter wide paths between plots).

Surface water flow was controlled by a 30-cm ridge on the uphill side of each plot and by 30-cm deep channels on all four sides of each plot (see Chapter 3, Figure 3-7b). These features were maintained and repaired throughout the growing season.

5.2.4.1 Rain manipulation schedule

As at MZ12, the rain manipulation schedule (see Table 5-2 for dates) was chosen to emulate a “worst-case scenario” for maize production: a prolonged dry spell beginning at anthesis, when maize is generally most vulnerable, and persisting throughout the majority of the remaining rainy season. The schedule was focused on the vulnerability of maize rather than that of seedlings because it was unknown what timing of drought might be most detrimental to the agroforestry seedlings under study.

In the first year (2008-2009), unforeseen delays in construction delayed deployment of the rain shelter roofs until seven weeks after anthesis, at which point the artificial drought was too late to affect that year’s maize yield. However, the rain manipulations were carried out nonetheless in order to observe possible effects on seedling performance. The 2008-2009 manipulations achieved an interception of approximately 40 mm, 4% of that year’s annual total (1070 mm).

In the second year (2009-2010), the shelters were already in place at the beginning of the growing season, and the roofs were deployed from the first sign of maize anthesis in late February until harvest in early May. The 2009-2010 manipulations achieved an interception of approximately 327 mm, 33% of that year’s annual total (999 mm).

In the third year (2010-2011), use of the rain shelters was discontinued for three reasons: (1) some of the shelters had been heavily damaged by termites (see Chapter 3); (2) labor and funding to maintain the shelters was limited; (3) cessation of the drought treatment permitted observation of any indirect effects of drought on maize yield the subsequent year.

5.2.4.2 Soil moisture measurements

Soil samples were collected several times throughout the 2008-2009 and 2009-2010 growing seasons using either an Edelmann soil auger or a bulk density corer. Collection depth varied from 5 to 30 cm. Due to the labor-intensive and destructive nature of sample collection, each plot was sampled at only one point (from a furrow near the plot center) at each sampling date. Gravimetric soil moisture was measured by drying samples to constant weight in a 70 ºC oven.

5.2.4.3 Measurements of microclimate

Air temperature was monitored over a three-week period in March and April 2010 according to the methods described in Chapter 3 (Section 3.2.2.8). PAR was also periodically recorded in

98

each plot, but as those data were collected solely to evaluate the unintended effects of the rain shelters, they are not reported here.

5.2.5 Seedling survival, height, and allometry

Throughout the first two growing seasons (2008-2009 and 2009-2010), a series of non-destructive seedling surveys were carried out. The goals of these surveys were twofold: (1) to provide more detailed temporal data than would be possible with a single destructive harvest, and (2) to provide information on how germination, survival, and growth contributed to a given biomass outcome. The measurements are summarized in Table 5-3 and described below.

5.2.5.1 Seedling presence-absence

During the first few weeks after germination or transplanting to the field, the seedlings’ size was generally was too small and/or too uniform for height measurements to provide useful data. Thus, early in the growing season, only presence or absence was recorded. For Tephrosia, these data indicated germination success (because exactly three seeds were sown at each planting station); for Gliricidia, they indicated post-transplant mortality.

5.2.5.2 Seedling height

Several times per growing season (see Table 5-3 for dates), each seedling’s height was measured from ground level to the top of the apical meristem (or the tallest apical meristem if more than one was present). Only live trees were measured, and deaths or absences were noted, allowing the height data to double as survival data. These measurements were performed only on seedlings in their first year, as both species in their second year were generally too tall to be easily measured (Figure 5-13 and Figure 5-14) and their mortality was negligible.

Figure 5-13. The author with a 22-month-old Tephrosia improved fallow.

Figure 5-14. Flowers on 22-month-old Gliricidia.

Figure 5-15. Seeds on 10-month-old Tephrosia.

99

At the end of the experiment’s second year (October 2010), two additional parameters were recorded along with the final height measurements of the one-year-old trees: the number of main branches and the presence of reproductive biomass (buds, flowers, or seeds; Figure 5-14 and Figure 5-15). It is considered desirable to minimize allocation to reproductive biomass in fertilizer tree systems, as this potentially reduces the total amount of green biomass that will be returned to the plot (Mafongoya et al., 2003).

5.2.5.3 Seedling allometry

The height and aboveground biomass of selected individual seedlings was measured at several times during the first two years of the experiment. Because these measurements were destructive, they were performed as part of the plot-level biomass harvests in November 2009 and October 2010. Additional data were collected periodically from seedlings that were culled because they were growing in the wrong position. The goal of these data was to develop allometric equations to non-destructively estimate aboveground biomass of Gliricidia and Tephrosia stands.

Table 5-3. Dates of destructive (biomass) and non-destructive seedling measurements at Nkula Field.

SEEDLINGS PLANTED 2008-‐2009 SEEDLINGS PLANTED 2009-‐2010 Early-‐planted Late-‐planted Early-‐planted Late-‐planted

Date Glir T-‐RI T-‐IF Glir T-‐RI T-‐IF Glir T-‐RI T-‐IF Glir T-‐RI T-‐IF

22-‐Jan-‐09 P/A P/A P/A -‐ -‐ -‐ -‐ -‐ -‐ -‐ -‐ -‐ 05-‐Mar-‐09 P/A P/A P/A P/A P/A P/A -‐ -‐ -‐ -‐ -‐ -‐ 05-‐Apr-‐09 Ht Ht Ht P/A P/A P/A -‐ -‐ -‐ -‐ -‐ -‐ 13-‐May-‐09 -‐ -‐ -‐ Ht Ht Ht -‐ -‐ -‐ -‐ -‐ -‐ 22-‐Oct-‐09 Ht Ht Ht Ht Ht Ht -‐ -‐ -‐ -‐ -‐ -‐ 01-‐Nov-‐09 -‐ Lt Lt -‐ Lt Lt -‐ -‐ -‐ -‐ -‐ -‐ 10-‐Nov-‐09 Bm b Bm Bm b Bm b Bm Bm b -‐ -‐ -‐ -‐ -‐ -‐ 17-‐Feb-‐10 -‐ -‐ -‐ -‐ -‐ -‐ Ht -‐ -‐ Ht -‐ -‐ 28-‐Feb-‐10 -‐ -‐ -‐ -‐ -‐ -‐ -‐ Ht Ht -‐ P/A P/A 04-‐Apr-‐10 -‐ -‐ -‐ -‐ -‐ -‐ Ht -‐ -‐ Ht -‐ -‐ 11-‐May-‐10 -‐ -‐ -‐ -‐ -‐ -‐ Ht Ht Ht Ht Ht Ht 29-‐Sep-‐10 -‐ -‐ -‐ -‐ -‐ -‐ Ht Ht Ht Ht Ht Ht 22-‐Oct-‐10 Bm -‐ Bm Bm -‐ Bm Bm Bm -‐ Bm Bm -‐

Abbreviations: Glir = Gliricidia; T-‐RI = Tephrosia relay intercropping; T-‐IF = Tephrosia improved fallow; P/A = presence/absence; Ht = height; Lt = litterfall; Bm = biomass.

a For datasets covering several days, this indicates the start date. b Biomass was only measured in half of each plot; the other half was left to grow another year.

5.2.6 Data analysis

Statistical analyses were carried out with Microsoft Excel for Mac 14.1.4 (Microsoft Corporation, Seattle, WA, USA) and JMP 10.0.0 (SAS Institute, Cary, NC, USA). All data were

100

averaged at the plot level (either whole-plot or split-plot) and analyzed using fixed-effects ANOVAs examining the variables and interactions of interest (see relevant Results sections for details). Seedling data, though they were collected at the individual level, were aggregated at the plot level and thus were analyzed as above.

5.3 Results

This section will first describe the efficacy of the drought manipulations, then report treatment effects (drought and planting time) on seedling biomass yield and maize yield at the plot level, and finally examine responses at the level of the individual seedlings.

5.3.1 Effectiveness of rain shelters

As expected, the 2008-2009 drought manipulation did not achieve the desired effect of reducing soil moisture. Figure 5-16 shows identical gravimetric soil moisture at 0-20 cm depths in ambient and drought plots at the beginning (28 March 2009) and at the end (mid-May 2009) of the manipulation. Samples from 20-40 cm and 40-60 cm also revealed no difference between the ambient and drought treatments (data not shown).

The 2009-2010 drought manipulation, however, did significantly reduce soil moisture (Figure 5-17). Due to time and labor constraints, soil moisture was quantified only once, in May 2010 (ambient = 7.62% ± 2.07% (SEM); drought = 6.37% ± 2.16%; p = 0.002). It was assumed that Nkula Field followed similar trends to MZ12 (which had a near-identical shelter design, manipulation schedule, soil type, and experimental layout). Corroborating this assumption, drought effects on soil appearance and plant water stress were visually apparent during the 2009-2010 rain manipulation at Nkula.

Figure 5-16. Gravimetric soil moisture (0-20 cm) immediately before (28 Mar 2009) and six weeks after (11 May 2009) the 2009 drought treatment. Error bars are ± 1 SEM. The drought treatment, which excluded <4% of annual rainfall, had no effect on soil moisture.

Figure 5-17. Gravimetric soil moisture (0-5 cm), 7 May 2010. The drought treatment (Rain =1), which excluded 33% of annual rainfall, reduced soil moisture (p = 0.002) except in plot 4-5 (indicated by X).

101

One of the drought plots (Plot 4-5; see map in Figure 5-2) was located at a low point in the corner of the field and could not be effectively drained by the system of channels and ridges that served the other plots. Soil moisture in this plot remained high throughout the growing season (Figure 5-17), with a value in May 2010 of >7 standard deviations above the mean of the other drought plots. Thus, Plot 4-5 was excluded from any analysis of drought effects.

5.3.2 Effect of optimally managed Gliricidia and Tephrosia systems on maize yield

To provide context for the impacts of drought and late planting, it will first be helpful to examine how the three newly established agroforestry systems affected maize yield under optimal conditions (ambient rainfall and early planting).

Table 5-4. Maize yields in sole maize and three different agroforestry systems (established in 2008) under ambient rainfall and early planting. Yields are given in kg ha-1 and as a percentage of the control (sole maize in that year). Means not sharing the same letter at are significantly different at p = 0.05.

These baseline results present several notable aspects. First, yields of sole maize were less than 0.6 tons per hectare in all three years. This is very low even compared to the national average of 1.6 tons per hectare for unfertilized hybrid maize (Smale and Heisey, 1997).

Second, yields of sole maize varied considerably between the three years, with the 2009 yield being less than one-fifth that of the 2010 yield. This is unlikely to be due only to weather, as all three years had near-average growing season rainfall. Crop damage from termites and monkeys may be partly responsible for the difference. A heavy weed burden during the first year post-fallow may also have interfered with crop growth.

Third, Gliricidia intercropping severely depressed maize yields in 2010, the second year after establishment. The trees were not pruned during their second year (in accordance with Akinnifesi et al., 2006), during which time they grew to overshadow the maize (Figure 5-18), reducing yields nearly to zero.

Fourth, Tephrosia improved fallows, although they displaced maize cropping in the first two years, brought about the highest yield of any cropping system in the third year (1.7 tons per hectare, which was 6.2 times the yield of the control plots). Tephrosia relay intercropping, on the other hand, had only a modest beneficial effect on maize yield in the second and third years (with yields 1.3 and 1.7 times that of the control plots, respectively).

Cropping system

2009 maize yield 2010 maize yield 2011 maize yield Mean % of ctrl Mean % of ctrl Mean % of ctrl

Sole maize 95.5 A 524.3 A 268.5 A

Gliricidia 126.3 A 132.2% 45.2 B 8.6% 1400.5 B,C 521.5%

Tephrosia RI 136.8 A 143.2% 711.5 C 135.7% 468.9 A,B 174.6%

Tephrosia IF n/a n/a n/a n/a 1669.9 C 621.9%

102

These baseline results indicate that the three agroforestry systems studied (with the exception of Gliricidia in the second year) work as intended to improve maize yields under normal conditions (i.e., timely planting and adequate rainfall). The question remains, though, how the impediments of drought and late tree planting might confound these benefits.

5.3.3 Effects of drought and late tree planting on maize yield

For this experiment, as for the users of fertilizer tree systems, maize yield is the ultimate variable of interest. Thus, we will focus first on how the experimental treatments affected maize yield before taking a closer look at how they affected the trees themselves.

5.3.3.1 Assumptions about causation

Drought. In theory, drought could affect maize yield both directly (through its impact on maize plants) and indirectly (through its impact on tree seedlings or soil properties). Because the only effective drought manipulation was carried out from February to May 2010, the following assumptions were made about the effects (if any) of the drought treatment on maize yields:

• Effects on 2009 maize yields are likely due to non-drought artifacts of the rain shelters.

• Effects on 2010 maize yields are likely due to direct water stress imposed on the maize.

• Effects on 2011 maize yields are likely due to effects on tree growth in the previous year (and/or effects on soil properties).

Timing of tree planting. Late tree planting was implemented as planned in all three years of the experiment. However, it was not expected to have an effect on the first maize harvest, because no tree biomass had yet been incorporated. Thus, the split-plot treatment of tree planting time was ignored in the May 2009 maize harvest. It was assumed that the subsequent maize two harvests (2010 and 2011) could be affected by timing of tree planting in the previous year.

5.3.3.2 Results

For each of the three years of the experiment, analysis of variance was used to describe the response of maize yield to each of the independent variables (cropping system, rain exclusion, and tree planting time) and their first-order interactions. The half-plots in which seedlings were re-established in 2009 (see Section 5.2.1.3) were not included in this analysis. Thus, for Gliricidia intercropping and Tephrosia improved fallows, this three-year sequence represents the first and second years of seedling establishment (2009 and 2010) followed by the first year after biomass incorporation (2011). Results are summarized in Table 5-5 below.

103

Table 5-5. Response of maize yields (kg/ha) at Nkula Field to cropping system, rain manipulation, and timing of planting of agroforestry trees. Within a given year and a given variable (outlined in black boxes), means with different letters are significantly different (p < 0.05).

Variable Level 2008-‐2009 yield (kg/ha)

2009-‐2010 yield (kg/ha)

2010-‐2011 yield (kg/ha)

Cropping system

Maize (M) 95.49 A 524.26 A 246.87 A

Gliricidia (G) 126.28 A 154.20 B 1108.56 B

Tephrosia RI (R) 136.80 A 711.75 C 450.64 A

Tephrosia IF (F) n/a n/a 1123.89 B

p > F 0.6875 <0.0001 0.0001

Rainfall

Ambient (0) 108.13 A 399.41 A 1043.80 A

Drought (1) 132.85 A 281.17 B 833.47 B

p > F 0.2896 0.0197 0.0385

Timing of planting

Early (E) n/a 311.76 A 1112.40 A

Late (L) n/a 442.67 A 772.66 A

p > F n/a 0.2061 0.0698

Block

Block 1 112.48 A 403.48 A 1368.72 A

Block 2 105.63 A 324.28 A 1039.92 A,B

Block 3 137.12 A 424.98 A 787.14 B

Block 4 126.12 A 454.16 A 512.98 C

p > F 0.4372 0.2627 <0.0001

Whole model p > F 0.1344 <0.0001 <0.0001

Interactions of interest

(values are LSMs)

Crop × Rain p > F = 0.0392 M,0 = 117.17 A,B,C M,1 = 68.84 C G,0 = 92.83 B,C G,1 = 154.75 A R,0 = 108.40 A,B,C R,1 = 152.54 A,B

Crop × Timing p > F = 0.0025 G,E = 64.44 C G,L = 269.86 B R,E = 811.22 A R,L = 746.81 A

Crop × Rain p > F = 0.1018 M,0 = 268.53 C M,1 = 225.21 C G,0 = 1592.48 A,B G,1 = 1549.99 A,B R,0 = 890.73 C R,1 = 782.19 C F,0 = 1907.26 A F,1 = 1265.88 B,C

Crop × Rain p > F = 0.0145 G,0 = 167.15 C G,1 = 175.97 C R,0 = 779.01 A R,1 = 566.48 B

104

Drought. As anticipated, the rain manipulation in 2009-2010 did have a significant effect on maize yield (a 29.6% reduction, p = 0.0197). However, in 2010-2011, when no rain manipulation was carried out, drought plots still had significantly lower maize yield (a 20.1% reduction, p = 0.0385) than ambient plots. This lagged effect may be due to the drought’s negative effects on tree seedlings in the previous year.

The components of this lagged drought effect are indicated by the 2010-2011 Crop × Rain interaction in Table 5-5. Although the overall interaction is marginally non-significant (p = 0.1018), it can be seen that the effect of the 2009-2010 drought on the following year’s maize yields was proportionally greatest in Tephrosia improved fallow plots (a 34% reduction, significant at p < 0.05), while the other three cropping systems showed modest and non-significant maize yield reductions due to drought (16%, 12%, and 3% respectively in Tephrosia relay intercropping, sole maize, and Gliricidia intercropping).

The presence of a significant (p = 0.0392) Crop × Rain interaction in 2008-2009 is unexpected, because the drought treatment intercepted only 4% of annual rainfall and did not appear to affect soil moisture (Section 5.3.1). This significant interaction arises from the fact that the 2008-2009 drought treatment actually appeared to increase maize yield in Gliricidia and Tephrosia plots. In sole maize plots, the opposite was true. No obvious explanation for this trend is apparent.

Timing of tree planting. This effect was not evaluated in the first year. In the second year, a significant (p = 0.0025) Crop × Timing interaction could be seen: late tree planting slightly decreased yields in Tephrosia relay intercropping plots, whereas it greatly (319%, p < 0.05) increased maize yields in Gliricidia plots. Figure 5-18 shows why: the larger and healthier the unpruned Gliricidia trees were, the more they overshadowed and stunted the maize.

In the third year (2010-2011), after trees had been cut and the cuttings incorporated in Gliricidia and Tephrosia improved fallow plots, late tree planting had a consistent and negative effect on maize yield across all three agroforestry systems (a 31% reduction, p = 0.0698).

Block and management effects. There were no significant differences between blocks except in the third year, when a gradient of decreasing maize yield (p < 0.0001) could clearly be seen from Block 1 (uphill) to Block 4 (downhill). The cause is unclear, but it may be because the experiment was not as closely tended during the third year (during which the author was mostly absent) as during the first two years. This could have led to greater pre-harvest losses due to encroachment by vervet monkeys and other animals from the miombo forest fragment that bordered Block 4.

5.3.4 Effects of drought and late tree planting on tree biomass production

Two years of data are available on tree biomass production (November 2009 and October 2010). In the 2010-2011 growing season, the experiment was terminated after maize harvest and did not extend to tree biomass harvest in October/November 2011.

In November 2009, half of the seedlings in each Gliricidia plot and each Tephrosia improved fallow plot were removed in order to plant new seedlings for the 2009-2010 rain manipulation

105

(Section 5.2.1.3). This harvest was not a standard management practice (the seedlings would normally have been left another year before cutting). Thus, the 2008-2009 seedling biomass data for these two systems should not be directly compared to the 2009-2010 data, which represent the harvest of the other half of each plot after a full two years of growth. Nevertheless, the 2008-2009 data are included here to give a more complete picture.

5.3.4.1 Production of leaves and wood

As described in Section 5.2.3, the aboveground biomass of seedlings was separated into leaves (which were incorporated into the plot) and wood (which was removed from the plot). Here, results are reported separately for leaves and for total biomass (i.e., leaves + wood), but trends are generally similar regardless of the biomass type considered.

Table 5-6 shows that seedling biomass varied significantly (p <0.0001) by cropping system. Tephrosia improved fallows produced by far the most biomass (even in November 2009, when all systems were the same age), while Gliricidia intercropping and Tephrosia relay intercropping were statistically indistinguishable from each other in both years.

Drought. Unlike maize yield, tree biomass yield was not significantly affected by the drought treatment in 2009-2010, nor were there any significant interaction effects between drought and other variables.

Timing of tree planting. Late planting greatly decreased total seedling biomass production in both 2008-2009 (p = 0.0277) and 2009-2010 (p = 0.0200). However, there were significant interactions between planting time and cropping system. Biomass production in Tephrosia improved fallows was profoundly impaired by late planting (with a reduction of 59% in 2008-2009 and 42% in 2009-2010). This effect is visually apparent in Figure 5-19. It should be noted that the near-halving of biomass production of Tephrosia improved fallows in 2009-2010 originates from a four-week difference in planting time almost two years previously.

Although late planting also appeared to decrease biomass production in Tephrosia relay intercropping (by 48% in 2008-2009 and 13% in 2009-2010), the differences were not statistically significant. Late planting seemed to have no effect, or even a slight positive effect, on Gliricidia biomass production.

Block and management effects. Differences between blocks were marginally significant (p = 0.0578) in 2009-2010, with the same trend of decreasing productivity from Block 1 to Block 4 that was seen in the following year’s maize yields (Section 5.3.3.2). Possible explanations include weed burden, pests, or allelopathy from the scrubland and forest adjacent to Block 4; alternatively, Block 1 could have received fertilizer runoff from the adjacent field uphill.

106

Table 5-6. Tree biomass production (kg/ha) at Nkula Field as affected by cropping system, rain manipulation, and timing of tree planting. Within each outlined black box (indicating a given year, a given biomass type, and a given variable), means with different letters are significantly different (p < 0.05). Gray-shaded cells in 2008-2009 indicate that the trees were cut prematurely to reconfigure the experimental design (Section 5.2.1.3), rather than as part of their standard management.

Variable Level 2008-‐2009 biomass (kg/ha) 2009-‐2010 biomass (kg/ha)

Leaves Total Leaves Total

Cropping system

Gliricidia (G) 166.86 A 166.86 A 344.98 A 4163.21 A

Tephrosia RI (R) 1149.98 A 1828.54 A 461.97 A 1254.21 A

Tephrosia IF (F) 2458.88 B 4968.02 B 3133.03 B 29227.55 B

p > F <0.0001 0.0002 <0.0001 <0.0001

Rainfall

Ambient (0) 990.22 A 1744.45 A 1066.93 A 11519.81 A

Drought (1) 1220.01 A 2317.23 A 1346.21 A 10031.62 A

p > F 0.2322 0.2336 0.3328 0.3762

Timing of planting

Early (E) 1593.58 A 3658.41 A 1604.91 A 14454.18 A

Late (L) 608.14 B 1845.96 B 771.31 A 6532.43 B

p > F 0.0236 0.0277 0.1421 0.0200

Block

Block 1 1376.91 A 2785.47 A 1759.66 A 15636.76 A

Block 2 959.80 A 1706.80 A 1097.25 A,B 9243.53 A,B

Block 3 1062.56 A 1703.60 A 1101.97 A,B 9874.97 A,B

Block 4 988.03 A 1862.53 A 792.12 B 8066.9 B

p > F 0.6477 0.4670 0.0669 0.0578

Whole model p > F <0.0001 <0.0001 <0.0001 <0.0001

Interactions of interest (values are

LSMs)

Crop × Timing p > F = 0.0159 F,E = 3513.3 A F,L = 1782.8 B G,E = 365.7 D G,L = 425.6 C,D R,E = 1507.7 B,C R,L = 706.1 B,C,D

Crop × Timing p > F = 0.0040 F,E = 7413.1 A F,L = 3004.2 B G,E = 674.6 B G,L = 1041.6 B R,E = 2887.0 B R,L = 1492.1 B

Crop × Timing p > F = 0.0207 F,E = 4085.8 A F,L = 2394.8 B G,E = 830.4 C G,L = 1047.2 C R,E = 1225.5 B,C R,L = 948.3 C

Crop × Timing p > F = 0.0039 F,E = 47330.1 A F,L = 27272.1 B G,E = 10691.7 C G,L = 8076.8 C R,E = 7027.9 C R,L = 6110.3 C

107

Figure 5-18. Unpruned Gliricidia, 13 months after planting, overshadowing stunted 7-week-old maize (February 2010).

Figure 5-19. 10-month old Tephrosia improved fallow plot in October 2009, showing the effects of late planting (left half) and early planting (right half). A harvested maize plot is in the foreground.

5.3.4.2 Production of litterfall and reproductive biomass in Tephrosia

Two often-overlooked components of biomass production in Tephrosia are litterfall (Figure 5-11) and seed pods (Figure 5-15), both of which are most significant at the end of the dry season. These biomass types are not typically quantified in field trials. Accordingly, they are not included in the data in Table 5-6 (except for reproductive biomass in October 2010, which is a component of total biomass), but an attempt was made to quantify them separately. Due to time constraints, only one of these biomass types was measured per year (litterfall in November 2009, and seed pods in October 2010).

Table 5-7 shows that both litterfall and seeds can account for a large fraction of Tephrosia biomass and that they can be affected by the same factors that affect total biomass production.

Litterfall. Total litterfall production, in kilograms per hectare, was much greater (p < 0.0001) in Tephrosia improved fallows than in Tephrosia relay intercropping systems of the same age; however, the ratio of litterfall to green leaves was approximately the same in both systems (16.3% versus 21.4% respectively; n.s.). The drought treatment was not successful in 2008-2009 and thus was not expected to affect litterfall. However, late planting had a strong negative effect on both the total amount of litterfall (p <0.0001) and the fraction of leaves dropped as litter (p = 0.0066), indicating that leaf senescence was delayed in late-planted trees. Failing to quantify litterfall could thus underestimate the biomass contribution of early-planted trees.

Reproductive biomass. In October 2010, Tephrosia improved fallows produced much more reproductive biomass – both by absolute and proportional metrics – than did Tephrosia relay intercropping systems. This can largely be attributed to the fact that the fallows were two years

108

old in October 2010, while the intercropping system was only one year old. Drought appeared to have a significant negative effect on reproductive biomass, but this may be an artifact of the rain shelter roofs, which tended to constrain the canopy of the two-year-old fallows.

Table 5-7. Production of litterfall (November 2009) and mature reproductive biomass (October 2010) by Tephrosia. Within each outlined black box (indicating a given year, a given biomass type, and a given variable), means with different letters are significantly different (p < 0.05).

Variable Level Litterfall (Nov 2009) Reproductive biomass (Oct 2010)

Total (kg/ha)

% of leaf biomass

Total (kg/ha)

% of total biomass

Cropping system

Tephrosia RI (R) 658.05 A 21.44% A 21.46 A 0.76% B

Tephrosia IF (F) 225.55 B 16.33% A 3827.34 B 13.92% A

p > F <0.0001 0.3435 <0.0001 <0.0001

Rainfall

Ambient (0) 455.50 A 20.41% A 2649.48 A 9.42% A

Drought (1) 428.11 A 17.36% A 1199.42 B 5.26% A

p > F 0.6124 0.2686 0.0149 0.0002

Timing of planting

Early (E) 694.04 A 23.01% A 2342.43 A 6.66% A

Late (L) 189.56 B 14.77% B 1506.47 A 8.02% A

p > F <0.0001 0.0066 0.1360 0.2348

5.3.4.3 Connecting tree biomass production to maize yield

In evaluating the performance of fertilizer tree systems, it is essential to know whether tree biomass production in one year is related to maize yield in the following year. In addressing this question, it can be assumed that one kilogram of Gliricidia biomass is roughly equivalent to one kilogram of Tephrosia biomass. Published values of Gliricidia leaf N content range from 1.8% (Mafongoya et al., 1998) to 3.9% (Vanlauwe et al., 2001), while leaf N content of Tephrosia candida is reported as between 3.1% and 3.3% (Mafongoya et al., 2003).

Figure 5-20 examines the plot-by-plot correlation between total tree biomass production in 2010 and maize yield in 2011, classified by drought treatment (left column) and cropping system (right column). A positive relationship is obvious in each case, indicating that these systems are functioning as intended to improve soil fertility and thus boost maize yields.

However, the R2 values varied widely (between 0.2 and 0.7), suggesting that maize yield is affected by factors other than tree inputs. Correlations were weakest in drought plots and in Tephrosia relay intercropping plots. It should be noted that some correlations are substantially enhanced by a single Tephrosia improved fallow plot that yielded 85 tons of biomass (leaves + wood) per hectare and produced 7400 kg/ha of maize the following year.

109

(a)

(d)

(b)

(e)

(c)

(f)

Figure 5-20. Correlations between fertilizer tree biomass yield in November 2010 and maize yield the following year (in May 2011). Shown are (a) all three agroforestry systems and both rain levels; (b) plots with ambient rainfall only; (c) drought plots only; (d) Gliricidia intercropping plots only; (e) Tephrosia relay intercropping plots only; (f) Tephrosia improved fallow plots only.

110

Because only leaf biomass is returned to the plot, correlations were also performed between leaf biomass and maize yield (data not shown). These correlations were similar to but weaker than the correlations for total tree biomass, so total tree biomass was chosen as the most useful variable for correlation. A possible reason for the strength of the correlation is that total tree biomass may be related to the magnitude of belowground inputs (such as root exudates and fine root decomposition), which were not measured but which may be relevant to soil nutrient status and hence to maize yields.

5.3.5 Effects on seedling germination, survival, height and phenology

Although users of fertilizer tree systems are ultimately concerned with tree biomass production and its subsequent effect on maize yields, it can be useful to understand how these outcomes are determined by processes at the scale of individual seedlings. Such process-based understanding can inform management practices and help to create more efficient and resilient agroforestry systems. The following sections explore how individual seedlings respond to the conditions imposed in this experiment.

5.3.5.1 Germination of Tephrosia seeds

The success of Tephrosia germination was not expected to be affected by the drought treatment in either 2008-2009 or 2009-2010, because in both years the seeds were planted at least three weeks before the treatment was initiated (Table 5-2). The 2008-2009 data showed a consistent rate of germination across both cropping systems and planting times (mean ± SEM):

Relay intercropping: early planting (73.2% ± 2.4% germination success) was identical to late planting (73.0% ± 1.8% germination success).

Improved fallow: early planting (71.5% ± 1.2% germination success) was identical to late planting (74.4% ± 1.2% germination success).

Germination data collected in 2009-2010 were incomplete and had to be discarded, but in light of the above results, it is unlikely that any treatment effects occurred. The consistent 70-75% germination rate (achieved without any seed pre-treatment) suggests that, at least in this location, three seeds per planting station is appropriate to achieve the target of two germinants.

5.3.5.2 Seedling survival

Seedling survival was recorded at several points throughout the 2008-2009 and 2009-2010 growing seasons (Table 5-3). For Gliricidia seedlings, the starting population was defined as the number of transplanted individuals; for Tephrosia, the starting population was defined as was the number of successful germinants as recorded in the surveys described above.

Gliricidia. The survival of Gliricidia seedlings is shown in Figure 5-21. Survival remained at or near 100% in all treatments in both growing seasons, with the exception of late-planted seedlings in 2008-2009, which suffered approximately 18% mortality by October 2009 (nine months after planting).

111

(a)

(b)

Figure 5-21. Survivorship of Gliricidia seedlings in (a) 2008-2009 and (b) 2009-2010. The relatively low survival of late-planted seedlings in 2008-2009 is an artifact due to inconsistent planting technique. Error bars represent ±1 SEM (but are not visible on most points).

(a)

(b)

Figure 5-22. Survivorship of Tephrosia seedlings in (a) 2008-2009 and (b) 2009-2010. RI = relay intercropping; IF = improved fallow. Error bars represent ± 1 SEM.

112

The survival rate of late-planted Gliricidia in 2008-2009 should not be directly compared to the other survival rates, as this batch of seedlings was planted inconsistently from the others. Logistical problems led to some of the late-planted 2008-2009 individuals being drawn from an immature population of seedlings and planted by untrained workers under time constraints, whereas the other batches were drawn from an adequately mature population and planted by the author. Mortality in the former batch arose from poor planting technique and subsequent damage during field management tasks, due to the seedlings’ small size and low visibility. Thus, the excess mortality represents a management effect rather than a late planting effect per se. (An 80-85% survival rate is typical for transplanted Gliricidia managed by farmers.)

It can be seen in Figure 5-21 that in some cases Gliricidia survival actually increased from one survey date to the next. This was not due to calculation errors but rather due to the remarkable regeneration capacity of Gliricidia. For example, a seedling that was broken at ground level and showed no living aboveground biomass would be recorded as “dead” in one survey, whereas by the next survey it might have begun vigorously resprouting (Figure 5-9).

Tephrosia. Survivorship of Tephrosia was lower than that of Gliricidia and showed more complicated patterns (Table 5-8 and Figure 5-22). Most notably, survivorship differed significantly between relay intercropping plots and improved fallow plots, but the direction of the effect reversed from one year to the next, with improved fallow plots showing better survival in 2008-2009 (91% versus 79% in October 2009) and relay intercropping showing better survival in 2009-2010 (86% versus 66% in September 2010). This may be related to the timing of field management tasks: weeding of newly established improved fallow plots was several weeks later than optimal in 2009-2010, possibly leading to significant competition from weeds.

Considered separately, neither the rain exclusion nor the timing of planting significantly affected Tephrosia survival in in either year, but in 2009-2010, a significant Rain × Timing interaction effect appeared for both April and September survival. At both of these survey points, it appears as though late planting only became detrimental to seedling survival in the presence of drought. This can be explained by the timing of the artificial drought: the late-planted Tephrosia had less than one month to establish before the drought began.

The significant block effect in 2009-10 shows a gradient of increasing seedling survival from Block 1 to Block 4, which is the opposite of the block effect observed for maize yield and seedling biomass yield (both of which decreased from Block 1 to Block 4). No causal explanation is apparent for this trend.

113

Table 5-8. Survivorship of Tephrosia seedlings in relay intercropping and improved fallow systems, 2008-2009 and 2009-2010. Data are reported only for newly established seedlings (<1 year old), not for seedlings in their second year. Survivorship is based on the number of confirmed germinants at the first survey. Within a given year and a given variable (outlined in black boxes), means with different letters are significantly different (p < 0.05).

Variable Level 2008-‐2009 survival 2009-‐2010 survival

04 Apr 22 Oct 11 May 29 Sep

Cropping system

Tephrosia RI (R) 94.64% A 79.74% A 93.05% A 85.97% A

Tephrosia IF (F) 96.42% A 91.35% B 83.32% B 66.24% B

p > F 0.4930 0.0042 0.0005 <0.0001

Rainfall

Ambient (0) 97.32% A 89.06% A 86.63% A 73.87% A

Drought (1) 95.98% A 91.18% A 89.74% A 78.34% A

p > F 0.3920 0.4179 0.2563 0.3338

Timing of planting

Early (E) 96.43% A 93.08% A 89.25% A 84.64% A

Late (L) 96.87% A 87.16% A 87.12% A 67.57% A

p > F 0.3147 0.3362 0.0656 0.4171

Block

Block 1 94.87% A 86.61% A 81.69% A 66.42% A

Block 2 96.87% A 92.41% A 84.05% A 72.45% A,B

Block 3 98.21% A 89.73% A 91.24% B 81.60% B

Block 4 96.65% A 91.74% A 94.86% B 83.95% B

p > F 0.5047 0.3991 0.0053 0.0419

Whole model p > F 0.6034 0.0332 0.0002 0.0001

Interactions of interest (values are

LSMs)

None None Rain × Timing p > F = 0.0019 0,E = 76.47% A 0,L = 84.79% A,B 1,E = 89.03% B 1,L = 77.47% A

Rain × Timing p > F = 0.0164 0,E = 66.83% A 0,L = 61.53% A 1,E = 83.83% B 1,L = 54.24% A

114

The relationship between plot-level seedling survival and biomass production in Tephrosia is examined in Figure 5-23 and Figure 5-24. (This relationship was not examined for Gliricidia because of its nearly 100% survival rate.) In relay intercropping plots (Figure 5-23), the relationship is weakly positive in early-planted plots, and there is no relationship in late-planted plots. In improved fallow plots (Figure 5-24), no relationship is apparent (though only one year of data could be used, because the improved fallows planted in 2009-2010 were not yet harvested at the termination of the experiment in May 2011).

Thus, at least at the levels of survival seen in this experiment (60% - 90%), it appears that surviving Tephrosia seedlings generally compensate for any missing seedlings in terms of biomass production. Seedling height may provide more useful information about seedling biomass production than does seedling survival, as discussed in the next section.

(a)

(b)

Figure 5-23. Relationship between seedling survival and biomass production in Tephrosia relay intercropping plots in (a) 2008-2009 and (b) 2009-2010. Correlations are calculated separately for early- and late-planted seedlings.

Figure 5-24. Relationship between seedling survival (in 2008-2009) and biomass production (in 2009-2010) in Tephrosia improved fallow plots. Correlations are calculated separately for early- and late-planted seedlings.

115

5.3.5.3 Seedling height

Several times during each growing season, seedling height measurements were carried out as a proxy for health and biomass production. The goal of repeating this non-destructive measurement was to assess if the seedlings were initially affected by drought or late planting but later recovered. Results are shown in Figure 5-25 (Gliricidia) and Figure 5-26 (Tephrosia).

Gliricidia. Due to the methodological problems with the drought and late planting treatments in 2008-2009, height data for Gliricidia were only analyzed for 2009-2010. In this year, all groups of Gliricidia seedlings began with the same average height (45 cm) in mid-February, which was shortly after the second batch were transplanted from the nursery and shortly before rain manipulations commenced. Subsequently, height growth in stressed seedlings clearly diverged from height growth in non-stressed seedlings. For both stress treatments (drought and late planting), final height at the end of the dry season was approximately 80 cm, in contrast to 110 cm in non-stress treatments.

(a)

(b)

Figure 5-25. Height trends of Gliricidia seedlings in 2009-2010: effects of (a) planting time and (b) drought. Error bars represent ± 1 SEM. (2008-2009 heights are not reported due to methodological problems with the treatments.)

Tephrosia. Height trends in Tephrosia seedlings were analyzed separately for relay intercropping plots and improved fallow plots; however, trends were very similar in both cropping systems (Figure 5-26). Late planting had a significant negative impact on height in all treatments at all time points. This is to be expected because, unlike Gliricidia seedlings (all of which were sown in the nursery simultaneously), late-planted Tephrosia seedlings were sown from seed approximately four weeks later than their early-planted counterparts. The resulting initial gap in average height persisted throughout the remainder of the growing season.

Rain exclusion, on the other hand, had no impact on Tephrosia seedling height in any treatment at any time point. This is somewhat surprising in light of the survivorship results reported in

116

Table 5-8: namely, that artificial drought significantly reduced survivorship of late-planted Tephrosia seedlings. The reverse phenomenon was seen in Gliricidia: drought reduced growth but had no effect on survivorship. These results are further explored in Section 5.4.1.

(a)

(b)

(c)

(d)

Figure 5-26. Height trends of Tephrosia: (a) 2008-2009 relay intercropping; (b) 2008-2009 improved fallows; (c) 2009-2010 relay intercropping; (d) 2009-2010 improved fallows. Error bars are ± 1 SEM.

117

5.3.5.4 Summary of data on individual seedlings

The above results suggest a variety of species- and management-specific responses, with no universal response to either treatment effect. Seedling germination and survival were generally not effected by either type of stress, though the interaction between the two stresses decreased survivorship of Tephrosia seedlings. Height growth of Gliricidia seedlings was greatly impacted both drought and late planting, but height growth of Tephrosia seedlings was only impacted by late planting. These results are briefly summarized in Table 5-9.

Table 5-9. Summary of treatment effects (drought and late planting) on individual seedling responses.

Gliricidia Tephrosia

Survival Growth Germination Survival Growth

Drought No effect Negative n/a No effect No effect

Late planting No effect Negative No effect No effect Negative

Interaction No effect No effect No effect Negative No effect

5.3.6 Allometry

As a complement to the above data on seedling height, simple allometric models were developed in order to determine whether seedling height was a useful proxy for aboveground biomass in these systems. These models were based on height and weight measurements from 103 individual seedlings (37 Gliricidia and 76 Tephrosia) with fresh weights ranging from 15 g to 1665 g. (Fresh weight was used due to the convenience of weighing the individual seedlings in the field.) All seedlings were less than one year old at the time of destructive measurement.

5.3.6.1 Model development

Potential models were selected from a subset of the library of non-linear models in JMP 10.0.0 (up to 4th-degree polynomials and up to 4-parameter exponentials and biexponentials). Criteria for model evaluation included information content (using the corrected Akaike Information Criterion, AICc, which seeks to optimize the tradeoff between model fit and number of parameters) and realistic behavior for height values near zero.

The most successful models for each species are given in Equation 5-1 (Gliricidia) and Equation 5-2 (Tephrosia). These models are depicted in Figure 5-27a and Figure 5-27b, respectively. For Gliricidia, a simple quadratic equation provided a good approximation of biomass, whereas for Tephrosia, only an exponential formulation could adequately predict biomass of both large and small seedlings.

BGlir = (0.51197 · h) + (0.020243 · h2) Equation 5-1

BTeph = -90.18 + (53.43 × e 0.01654 · h ) Equation 5-2

118

where BGlir = aboveground biomass of Gliricidia (g FW); BTeph = aboveground biomass of Tephrosia (g FW); h = height of seedling apical meristem (cm).

A drawback of Equation 5-2 is that it yields a negative biomass value for seedling heights below 31.65 cm. This suggests that it may not be possible to use a single model to successfully predict the biomass of both large and small Tephrosia seedlings using only height as an input.

It should be noted that the coefficients of both of the above models were calculated for seedling fresh weight (FW) rather than dry weight (DW). In the current study, this could be corrected by multiplying the resulting biomass estimates by the average percentage dry weight for each species under these conditions: 36.94% for Gliricidia and 42.99% for Tephrosia. To improve upon these results, drying the seedlings before weighing would be advisable.

5.3.6.2 Comparison of measured and modeled plot-level biomass

The above allometric models were applied to individual seedling height measurements taken in October 2009, shortly before the destructive harvest that preceded land preparation for the 2009-2010 growing season. Model predictions for biomass of individual seedlings were then aggregated at the plot level and compared to the actual plot-level biomass values from the November 2009 destructive harvest.

Correlations between predicted and actual biomass were very strong (Gliricidia: R2 = 0.93, slope = 1.06; Tephrosia: R2 = 0.86, slope = 1.08). These correlations are depicted in Figure 5-27c and Figure 5-27d. The linear fit lines are nearly indistinguishable from the 1:1 line, indicating that the above models are useful for predicting plot-level biomass of seedlings of this size at this location. Although these plot-level biomass predictions utilized the height of every seedling within the plot, it would be desirable to be able to accurately predict plot biomass using a relatively small subsample of individual seedlings; this is left for future work.

It should be noted that these simple models are not expected to apply to Gliricidia and Tephrosia seedlings in their second or subsequent years. Allometric modeling of two-year old fertilizer trees is made more accurate by the use of D10 measurements (stem diameter at 10 cm height) and also benefits from characterization of leaves and wood as separate biomass compartments (Kaonga and Bayliss-Smith, 2010).

119

(a)

(b)

(c)

(d)

Figure 5-27. Allometric models for Gliricidia and Tephrosia seedlings: (a) Gliricidia quadratic model; (b) Tephrosia exponential model; (c) Gliricidia plot-by-plot comparison of actual versus predicted biomass (2008-2009 data); (d) Tephrosia plot-by-plot comparison of actual versus predicted biomass (2008-2009 data). In figures (c) and (d), the 1:1 line is solid while the linear model fit line is dashed.

5.3.7 Microclimatic effects

In light of the possible microclimatic benefits of agroforestry described in Chapter 1, an attempt was made to evaluate the effect of these fertilizer tree systems on air temperature at 1 m above ground level. Half-hourly temperature records were collected with Hobo Pendant Temp data loggers (Onset Corporation, Bourne, MA, USA) from 29 March to 20 April 2010. One logger was used in each of eight plots (equating to two replicates per plot type). In cropping systems that had both one- and two-year old trees in place at that time (Gliricidia intercropping and Tephrosia improved fallows), subplots with two-year-old trees were selected.

120

5.3.7.1 Average air temperature

The results were not suitable for hypothesis testing due to the limited number of data loggers available. However, they strongly suggested that the system with the strongest microclimatic effect was Tephrosia improved fallows (Figure 5-28), which had an average air temperature approximately 3 ºC cooler than sole maize plots. This is unsurprising in light of the considerable aboveground biomass and height attained by two-year-old improved fallows (Figure 5-13). However, as maize is not present simultaneously with the trees in an improved fallow, this microclimatic benefit is not relevant to maize production. Gliricidia appeared to cause a very modest decrease in air temperature (about 1 ºC cooler than sole maize plots), while Tephrosia relay intercropping appeared to have no effect (Table 5-10).

Figure 5-28. Comparison of air temperature over a typical three-day period (4-6 April 2010) in different cropping systems at Nkula Field. Each line represents the average of two plots of that type.

5.3.7.2 Air temperature in excess of damage thresholds

Each system also was evaluated for the number of hours spent above certain thresholds: 30 ºC (a temperature which, if chronically exceeded, can depress maize yields) and 37 ºC (a temperature at which acute heat stress can damage a maize crop) (Southworth et al., 2000; Lobell et al., 2011). Again, only Tephrosia improved fallows appeared to have a noticeable benefit in this regard, essentially eliminating any time spent over 37 ºC (Table 5-10).

Indications of subtle air temperature benefits in Gliricidia deserve more investigation, but they should be considered in light of the fact that these Gliricidia plots were at a stage in the cropping cycle where the trees have attained maximum aboveground biomass. Thus, their effects on air temperature would likely be reduced or absent at other times of year.

121

Table 5-10. Average air temperatures and number of hours exceeding temperature thresholds at Nkula Field (30 March to 19 April 2010). No error estimates are calculated, as each row represents the average of only two plots of that type.

Cropping system Average T (°C) Deviation from overall average

Average hours per day >30 °C

Average hours per day >37 °C

Maize 24.46 °C 0.92 °C 6.35 hours 0.55 hours

Gliricidia 23.94 °C -‐0.02 °C 5.19 hours 0.24 hours

Tephrosia RI 24.42 °C 0.84 °C 6.08 hours 0.39 hours

Tephrosia IF 22.98 °C -‐1.74 °C 2.10 hours 0.04 hours

5.4 Discussion

5.4.1 Hypotheses revisited

It will be helpful to frame the discussion by revisiting the eight hypotheses outlined in the Introduction. Some hypotheses were supported, while others were not supported.

Hypothesis 1. Survival, growth, and biomass production of newly established Gliricidia seedlings will be negatively affected by a prolonged drought.

Only one component of this hypothesis was supported: drought significantly reduced height growth of Gliricidia during the 2009-2010 growing season (Section 5.3.5.3). Survival of Gliricidia remained close to 100% in both ambient and drought treatments.

The effects of drought on Gliricidia biomass production are not fully known, because Gliricidia seedlings in their first year that were affected by the successful drought treatment in 2009-2010 were not ready for biomass harvest until October 2011, after the experiment had concluded. For the Gliricidia seedlings that were in their second year when the drought was imposed, there was no apparent effect of drought on biomass production.

Hypothesis 2. Negative effects of drought on seedling performance will translate into negative effects on maize yield in the following year.

There is some evidence to support this hypothesis, as the drought treatment significantly reduced maize yields in 2010-2011 even though the rain manipulation had been terminated at the end of the previous year. However, upon closer examination (Table 5-5), the effect varied by cropping system and was only significant in Tephrosia improved fallow plots. This corresponds with the observation that tree biomass production in October 2010 (five months after the end of the successful drought treatment) was not significantly different between drought and non-drought plots. It may be that plot-to-plot variability obscured the effect of drought on biomass production in Tephrosia improved fallows but that this effect was large enough to affect the following year’s maize production.

122

Hypothesis 3. The magnitude of drought effects on seedling survival, growth, and biomass production will differ between Gliricidia and Tephrosia.

Survival and growth. This hypothesis was supported, though the details were not precisely predicted (Table 5-9). Drought had no effect on Gliricidia seedling survival, but it had a strong negative impact on height growth. The reverse was true for Tephrosia: drought had no impact on height growth, but it did reduce survival (in late-planted seedlings only).

These findings are somewhat surprising in light of a report by Sileshi et al. (2008) on perceptions of fertilizer tree mortality in Zambia: farmers listed drought as the single most important cause of mortality in both Gliricidia and Tephrosia (more important than insects, fire, or disease). This study indicated that drought was not a significant cause of seedling mortality in Gliricidia, at least under researcher-managed field conditions.

The contrasting growth responses in Gliricidia and Tephrosia imply that the species may have different physiological strategies for coping with drought: Gliricidia, being a longer-lived and less prolifically seeding species, may down-regulate its growth in order to increase its chances of survival during water stress. By contrast, the optimal strategy for Tephrosia may be to continue growth under drought conditions with the goal of reaching reproductive maturity (which it can achieve in 9-10 months) and producing seeds that will germinate in a more favorable year.

Biomass. For seedlings that were in their first year when the drought was imposed, no data on final biomass production are available. For seedlings that were in their second year (or for Tephrosia seedlings in relay intercropping plots, which have only a one-year cycle), drought did not appear to affect final biomass production, with the possible exception of Tephrosia improved fallows as mentioned above.

Hypothesis 4. Tephrosia seedlings in relay intercropping systems will have lower survival and growth rates than those in improved fallows.

Growth. This hypothesis was supported with respect to biomass production; the 2008-2009 partial harvest showed that one-year-old Tephrosia improved fallows produced approximately twice as much biomass as Tephrosia relay intercrops of the same age (Table 5-6). However, heights of seedlings in the two systems did not differ significantly (Figure 5-26). This suggests that perhaps a separate allometric model is needed for Tephrosia seedlings in each system.

Survival. Although survival differences between seedlings in the two Tephrosia systems were significant in both years, the direction of the effect depended on the year. In 2009-2010, relay intercrops had higher survival rates than improved fallows, whereas the reverse was true in 2008-2009 (Table 5-8). This reversal may have been due to variations in management practices. On the basis of these data, it cannot be concluded that either system has an advantage regarding seedling survival.

Hypothesis 5. The performance gap between Tephrosia relay intercropping and Tephrosia improved fallows will become more pronounced under drought conditions.

This hypothesis was not supported; in fact, the opposite was found. Drought only appeared to affect seedling survival in Tephrosia improved fallows (in combination with late planting), not in

123

Tephrosia relay intercropping. Drought had no significant effect on height growth or final biomass production in either Tephrosia system.

Hypothesis 6. Delaying planting by one month will reduce survival, growth, and biomass production of Gliricidia and Tephrosia, with corresponding reductions in maize yield.

This hypothesis was not supported with respect to survival (no effect was seen in either species), but it was strongly supported with respect to height growth (Figure 5-25 and Figure 5-26) for both species. Late planting also translated into negative effects on biomass production, but the effect was only statistically significant for Tephrosia improved fallows (Table 5-6). These effects on tree biomass production appeared to cause reduced maize yields in the following year (p = 0.0698; Table 5-5).

Hypothesis 7. Late-planted seedlings will be more negatively affected by drought than will seedlings planted on time.

This hypothesis was only supported with respect to seedling survival in Tephrosia improved fallows (Table 5-8). There was no evidence for an interaction effect of drought and late planting on seedling survival in the other two fertilizer tree systems, nor was there evidence for an interaction effect on height growth or biomass production of seedlings in any system. This is perhaps somewhat surprising given that the drought was implemented shortly after the late-planted seedlings were established. It suggests that both Gliricidia and Tephrosia are quite drought-resilient even early in their establishment.

Hypothesis 8. Gliricidia intercropping will have no effect on maize yield for the first two years, after which it will have a strong positive effect. Tephrosia relay intercropping will have no effect in the first year, after which it will have a modest positive effect. Tephrosia improved fallows will have a strong positive effect in the third year (the first post-fallow maize crop).

Hypothesis 8 was fully supported with the exception of maize yields in second-year Gliricidia plots. Rather than having no effect, Gliricidia had a profound negative impact in the second year, essentially eliminating maize production for that year. Recommendations vary regarding whether or not to prune Gliricidia in the second year (Akinnifesi et al., 2006; Ajayi et al., 2009). Results from this experiment suggest that in light of negative effects on maize yield, combined with the apparent resilience of Gliricidia to even the most severe physical damage (Figure 5-9), Gliricidia intercropping systems can benefit from pruning beginning in the second year rather than the third year.

It is encouraging that all three fertilizer tree systems were seen to confer major benefits on maize yield, not only under optimal conditions (Table 5-4) but also, to a lesser extent, under suboptimal conditions of drought and late planting. For users of fertilizer trees, maize yield is the key outcome, and any variations in seedling survival or growth are only relevant to the extent that they affect that outcome.

5.4.2 Future work needed

At the time of writing, this experiment represents the first example of an empirical investigation of the response of fertilizer tree seedlings to climatic stress. There are many opportunities for

124

similar experiments looking at other combinations of factors: different magnitudes and temporal patterns of drought; additional global change factors such as temperature and CO2; different locations with varying soil types; different tree species and cultivars. The importance of characterizing the response of seedlings separately from the response of mature trees should not be overlooked.

A potential shortcoming of this experiment was the lack of characterization of belowground biomass production by fertilizer trees. Belowground biomass is generally a neglected topic in agroforestry systems due to the difficult and labor-intensive methods involved; however, a more detailed understanding of belowground dynamics will likely be helpful in understanding the response of these systems to drought and other environmental factors.

Biomass inputs from fertilizer trees could be better understood with more accurate quantification of litterfall (in this study, especially important in Tephrosia) and estimates of the fraction of nitrogen derived from atmospheric fixation (%Ndfa), which can vary greatly not only according to species but also according to site and management practices (Mafongoya et al., 2004). Optimizing the performance of fertilizer tree systems in consideration of belowground biomass, litterfall, and %Ndfa may result in different management decisions than if only aboveground vegetative biomass is considered.

5.5 Conclusions

Although the conclusions of a three-year experiment in a single location must necessarily be tentative, several themes emerge from the above results. First, the three fertilizer tree systems studied (Gliricidia intercropping, Tephrosia relay intercropping, and Tephrosia improved fallows) all functioned as intended to increase maize production under normal climatic conditions. This study underscores the efficacy of their use for soil fertility restoration in maize subsistence cropping in southern Africa.

Second, all the fertilizer tree systems studied exhibited considerable resilience to the stresses of drought and late planting. Late planting had a somewhat greater impact than drought, and the two factors negatively interacted to decrease Tephrosia survival; nevertheless, the systems continued to provide benefits to maize yields even under these adverse conditions.

Third, and more generally, these results underscore that species-specific differences can be important even among similar species and similar systems. It is important to use caution when generalizing about (for example) the effect of climate change on agroforestry systems, or even the effect of drought on fertilizer trees. More work is needed on other systems under other climatic conditions.

Studies on the physiological responses of fertilizer tree seedlings to climatic stress can make an important contribution to efforts to improve maize yield (and, more broadly, food security) in southern Africa under current climate variability and future climate change. A detailed understanding of how these systems respond to variations in climate can further elucidate the role they can play in the management of production risk in subsistence farming.

125


Festus Akinnifesi provided essential insight into the design and hypotheses of this experiment. In particular, he is responsible for the inclusion of planting time as a key variable. Simon Mng’omba and Chikumbutsa Kwakwala managed the experiment and collected the data during the 2010-2011 growing season when the author was not present in Malawi. This work could not have been completed without their important contributions.

127

Chapter 6. Current and future climate in southern Malawi: implications for maize production

Abstract

Maize in southern Africa is frequently described as vulnerable to current climate variability and future climate change. Recent studies suggest that changes in future rainfall will be subtle, whereas future temperature increases may strike a major blow to maize production. This chapter evaluates the validity of these general trends at Makoka Agricultural Research Station (15º31S, 35º13E), where the field experiments described in this dissertation were performed. Makoka provides an interesting case study in this regard because it represents a densely populated, maize-dependent area of Malawi’s Southern Highlands; it also has the rare benefit of being described by a long-term meteorological data set (1971-2008).

These long-term data were used to evaluate the existence of long-term trends in temperature and rainfall; no trends were seen except for a weak ENSO teleconnection and a modest warming trend consistent with the global background rate. Two methods were then applied to investigate future climate: (1) predicted monthly changes in temperature and rainfall from the BCCR:BCM2 global circulation model for 2040-2069 were directly added to the daily historical data; (2) MarkSim, a Markov chain daily weather simulator, was run with output from BCCR:BCM2 for 2040-2069.

Neither approach revealed notable changes in precipitation patterns. The former approach generated mild warming that depressed maize yield by several percentage points, while the latter approach generated dramatic warming that cut maize yield by one-third, enough to plunge the average farm family into food insecurity. The lack of an accurate baseline for the latter approach suggests that the former approach may be more credible with regards to temperature.

The high geographic heterogeneity of landscapes in Malawi suggests taking a customized approach to questions of climate adaptation, rather than applying outputs from climate models at a large scale. The simple analysis described herein may be a useful first step to climate adaptation planning for other maize-growing areas in the region.

6.1 Introduction

6.1.1 Southern Africa’s vulnerability

Southern Africa has historically been vulnerable to climate-related shocks for several reasons. The amount and timing of precipitation is highly variable in many parts of the region; the majority of southern Africans are subsistence farmers reliant on rainfed systems; and widespread poverty constrains coping options. Hunger and famine in southern Africa, despite their complex underlying causes, are often instigated by adverse weather events.

128

Despite concerted recent efforts to improve seasonal forecasts (O'Brien and Vogel, 2003) and predict impending famine (for example, USAID’s Famine Early Warning System), climate-related food insecurity is still common in Southern Africa. The most recent food crisis, in 2002, was catalyzed by a series of droughts and floods that spanned five countries. These events threatened 14 million people and killed thousands, with Malawi bearing the brunt of the impact. The human toll surpassed that of the previous severe drought in 1991-92, in part due to poorer preparation by governments and civil society (Devereux, 2002a).

In the past several decades, two additional factors have arisen that may further destabilize Southern Africa’s food security. The first is HIV/AIDS, which has hit Southern Africa the hardest of any world region. The damage wrought by AIDS on the working-age population, and the resulting increase in dependency ratio, may undermine food production and increase vulnerability to stresses, leading to “new variant famine” (de Waal and Whiteside, 2003). The second factor is anthropogenic climate change.

6.1.2 Climate change in southern Africa

The most recent IPCC assessment report predicted that Southern Africa will experience a net reduction in rainfall and a greater-than-average increase in temperature (Boko et al., 2007; Christensen et al., 2007). This, combined with the region’s generally warm and semi-arid climate, high baseline climate variability, dependence on rainfed subsistence agriculture, widespread poverty, and high biodiversity, classifies it as a vulnerable region according to many authors (e.g., Bauer and Scholz, 2010). Other studies have produced less pessimistic rainfall forecasts (Shongwe et al., 2009; Tadross et al., 2009) but have still stressed the need for forewarning and preparation.

A number of recent studies have examined 20th-century southern African rainfall data for signs of climate change already underway. These historical analyses include South Africa (Fauchereau et al., 2003), Zimbabwe (Mazvimavi, 2010), Zambia (Stern and Cooper, 2011), Botswana (Batisani and Yarnal, 2010), and the region encompassing Zimbabwe, Zambia, Malawi, and Mozambique (Tadross et al., 2009). With the exception of Botswana, none of these countries appeared to show long-term trends in total rainfall; however, several authors noted significant trends toward increased interannual variability (Fauchereau et al., 2003) and delayed growing season onset (Tadross et al., 2009) at some locations.

Complicating these assessments is the fact that southern Africa experiences rainfall cycles with periods of years to decades. These cycles are driven by El Niño (Cane et al., 1994), the North Atlantic Oscillation (Stige et al., 2006), and Indian Ocean temperatures (Hoerling et al., 2006).

Despite the modest evidence for existing climatic change in southern Africa, and the difficulty in untangling trends from long-term oscillations, a solid consensus exists (Christensen et al., 2007) that southern Africa’s future climate will be warmer with less reliable rainfall and that the sustainability of rainfed subsistence agriculture in the region warrants great concern.

6.1.3 Climatic influences on maize in southern Africa

Maize is the staple crop throughout most of southern Africa. Although it is less resilient to environmental stresses than many of the native crops it displaced, its central role in the region’s

129

history and culture is unassailable (McCann, 2005). Some authors (e.g., Kurukulasuriya and Mendelsohn, 2006) have suggested that many African farmers will need to switch away from maize in order to successfully adapt to warmer, drier climates; however, given its current dominance, protecting maize in southern Africa from the effects of climate change remains an important goal for the foreseeable future.

A number of recent statistical modeling studies have predicted grim outcomes for maize in Southern Africa under medium- to long-term future climate (Jones and Thornton, 2003; Lobell et al., 2008; Schlenker and Lobell, 2010). Lobell et al. (2008) noted that, out of all major crops and regions in the developing world, southern African maize is likely to suffer among the greatest losses due to climate change and thus deserves high priority for adaptation efforts.

Climate change in southern Africa is expected to influence maize yields via several distinct but related mechanisms: precipitation and temperature.

6.1.3.1 Precipitation

The most obvious type of precipitation change is an overall decrease in rainfall. Recent modeling studies (Tadross et al., 2005; Shongwe et al., 2009; Tadross et al., 2009) predict that most parts of southern Africa will see little or no change in total annual rainfall under future climate. However, these same studies suggest that changes are likely in the timing of precipitation (e.g., later onset, earlier cessation, and longer dry periods). Maize is more drought-sensitive during certain parts of its life cycle, particularly during anthesis (Edmeades et al., 2000), so the exact timing of drought is crucial. Compensating for these changes in timing could make the difference between maintaining good yields and suffering poor yields under future climate (Crespo et al., 2011).

Flooding and waterlogging can also be major problems for maize in southern Africa (e.g., Sileshi et al., 2011) and have been implicated in recent famines (Devereux, 2002a). Even if total precipitation does not change, a greater number of heavy rain events could increase flood risk.

6.1.3.2 Temperature

Higher temperatures can negatively affect maize in three ways: (1) acute heat stress, which directly disrupts the plant’s metabolism; (2) increased soil evaporation, leading to water stress; (3) accelerated phenology, shortening the time period in which the maize crop can intercept radiation and thus decreasing its cumulative photosynthesis (Muchow, 2000). The threshold for the first phenomenon is generally considered to be below 40 ºC (Maiti and Wesche-Ebeling, 1998), whereas the threshold for the second and third phenomena are dependent upon other environmental factors. Lobell et al. (2011) reported declines in African maize yield when daily maximum temperatures exceeded 30 ºC, though studies in North America have reported this threshold to be 33 or 34 ºC (Southworth et al., 2000).

Heat stress is thought to have multiplicative effects with photoinhibition and with drought stress (Paulsen, 1994), which often occur simultaneously with high temperatures in the field.

130

6.1.4 Proposed adaptation measures

Various solutions have been proposed to protect smallholder maize production in sub-Saharan Africa (SSA) from the effects of climate change. These include adjusting maize planting dates (Crespo et al., 2011); adopting more heat- and drought-tolerant maize varieties (Lobell et al., 2011); switching to alternate crops that are currently used in other African regions, e.g., the Sahel (Kurukulasuriya and Mendelsohn, 2006); growing a wider variety of crops, including perennials, as an insurance policy against uncertain climate (Thorlakson, 2011); using trees to improve soil quality and microclimate (Rao et al., 2007); and deploying irrigation on a large scale (Kurukulasuriya et al., 2006).

These options vary widely in their ease of implementation (Lobell et al., 2008), and their success hinges upon understanding the conditions faced by local communities (Thornton et al., 2011). Potentially relevant factors include the baseline climate and its anticipated changes, current crop varieties and cultivation practices, average farm size, and proximity to markets. This paper aims to inform climate adaptation at a local scale by examining the possible impacts of climate change on subsistence maize production in densely populated southern Malawi.

6.1.5 A case study in southern Malawi

Southern Malawi provides an interesting case study within southern Africa for several reasons. In many ways, its geographic and socioeconomic statistics are typical of the region: most residents are smallholder farmers cultivating maize; it has a unimodal rainy season of variable onset and duration; and, as a landlocked country in the heart of southern Africa, it is insulated from oceanic influences. In other ways, southern Malawi is atypically vulnerable: household income is low and farm sizes are especially small (Sirrine et al., 2010), constraining coping options for adverse climatic conditions.

This analysis focuses on Makoka Agricultural Research Station (15º31' S, 35º13' E, 1030 m elevation, Figure 6-1) 20 km southwest of Zomba, Malawi. Established in 1970, this government research station is one of the most complete sources of meteorological data in southern Malawi, where weather stations are few and far between. In addition, data from controlled maize trials are available from the site for some years, providing an opportunity to correlate climate conditions with maize yield.

Using daily data on precipitation and temperature (maximum and minimum) at Makoka from 1971-2008, in conjunction with output from several GCMs (General Circulation Models), this study intends to answer the following questions:

1. Has maximum or minimum temperature increased at Makoka in the past 40 years? 2. Has the amount, variability, or timing of rainfall changed at Makoka in the past 40 years?

3. How are future temperature increases likely to affect maize yield at Makoka? How important is acute heat stress versus changes in phenology?

4. How are future changes in precipitation likely to affect maize yield at Makoka? Will they be more or less important than changes in temperature?

5. In light of these findings, what adaptation options might be best suited for this region?

131

6. Do these local results agree with large-scale assessments for the region? If not, how can the differences be explained, and what are the implications for climate adaptation planning?

6.2 Methods

Daily meteorological data were obtained via the meteorological office at Makoka Agricultural Research Station (P/Bag 3, Thondwe, Malawi). Half-hourly temperature records were collected by the author during field visits in 2009 and 2010 using Hobo Pendant Temp data loggers (Onset Corporation, Bourne, MA, USA). Statistical analyses were carried out with Microsoft Excel for Mac 14.1.4 (Microsoft Corporation, Seattle, WA, USA) and JMP 8.0.2 (SAS Institute, Cary, NC, USA). Unless otherwise noted, error bars represent ±1 SEM.

6.2.1 Analysis of historical climate

To provide a baseline against which to evaluate future changes in climate, temperature and precipitation data from 1971 to 2008 were analyzed for historical averages and possible long-term trends. Because the rainy season in Malawi occurs between November and April (Figure 6-2), the agronomic year is defined as being from 1 July to 30 June.

6.2.1.1 Quality control of data

Temperature and precipitation data were examined for missing or extreme values (defined as temperature <0 ºC or >50 ºC and precipitation >200 mm per day. No extreme values were found. Missing values were found only in temperature data and were dealt with in one of two ways: (1) If four or fewer consecutive days were missing, their values were linearly interpolated from the days immediately adjacent. (2) If more than four consecutive days were missing, the entire month was discarded. Years with missing months were excluded from any analysis requiring annual averages.

!

Figure 6-1. Location of Makoka Agricultural Research Station in Malawi (15º31’S, 35º13’E).

Figure 6-2. Annual distribution of rainfall at Makoka, Malawi, 1971-2008.

132

6.2.1.2 Analysis of historical temperature

Temperature data were available for 29 agronomic years (1971-1972 through 1999-2000). Simple linear regressions were used to evaluate trends over time in daily maximum temperature, daily minimum temperature, and number of days above 30 ºC. Trends in annual averages and in individual growing season months (November through April) were examined separately. For reference, average maximum and minimum temperatures for each month (across all 29 years of data) are shown in Figure 6-3.

6.2.1.3 Analysis of historical rainfall

Rainfall data were available for 37 agronomic years (1971-1972 through 2007-2008). Simple linear regressions were used to evaluate trends over time in the following parameters.

Summary statistics. The following agronomically relevant parameters were calculated for each year: total annual rainfall, total growing season rainfall (1 November through 30 April), number of rain days (>2 mm), number of heavy rain days (>20 mm), coefficient of variation of rainfall, and maximum dry spell length (with no rain >2 mm). Unless otherwise specified, statistics were calculated for the growing season rather than for the 12-month year.

Growing season start and end dates. Farmers in southern Africa decide when to plant maize by considering not only the day of the year, but also the recent rainfall and expected future rainfall (Crespo et al., 2011). This study applied the definitions of Tadross et al. (2009), which are specific to maize in southern Africa:

• Planting date, constrained to occur no earlier than 1 August, was defined as the final day of a dekad (10-day period) in which >25 mm of rain had fallen.

• True planting date was defined similarly but with the caveat that the subsequent 20 days must not contain 10 consecutive dry days (where dry is defined as <2 mm rainfall).

• Growing season cessation, constrained to occur no earlier than 1 February, was defined as the final day of three consecutive dekads of <20 mm each.

Figure 6-3. Average maximum and minimum temperatures at Makoka, Malawi, 1971-2000.

!

Figure 6-4. Maize wilting, Makoka, Malawi, 1/2010.

133

Growing season length was defined as the number of days between true planting date and cessation date. If true planting date differed from planting date, the season was considered to have a false start. A false start can be disastrous for farmers, as they may invest considerable resources (seed, fertilizer, and labor) into a crop that fails. This occurred throughout southern Malawi in 2009-2010 (Figure 6-4), leading to widespread food insecurity (Figure 4; FEWSNET, 2010).

Correlation with ENSO. To explore the relationship between El Niño / Southern Oscillation (ENSO) and rainfall at this location, historical values were obtained for two ENSO metrics: (1) the multivariate ENSO index (MEI) (Wolter, 2012) and (2) the NINO3 index (NOAA Climate Prediction Center, 2011). Bimonthly MEI values were averaged over each of the 12-month agronomic years for which rainfall data were available, whereas monthly NINO3 values were averaged over February and March of each year (following the methods of Cane et al., 1994). Total growing-season rainfall was then regressed against the averaged MEI and NINO3 values.

6.2.1.4 Calculation of growing degree-days

The concept of growing degree-days (GDD) is frequently used to describe crop phenology. Most maize cultivars require at least 1400 Celsius GDD to reach maturity (Maiti and Wesche-Ebeling, 1998). If this total is not attained during the growing season – because temperatures during the growing season are too hot or too cold, because the growing season is truncated by lack of rain, or both – the maize crop may yield poorly or may fail altogether.

GDD can be calculated in a variety of ways. The simplest definition (Tollenaar and Dwyer, 1999) states that the GDD accumulated on a given day are equal to the daily mean temperature (defined as the average of Tmax and Tmin) minus a baseline, Tbase, below which no development takes place (Equation 6-1):

GDD =Tmax + Tmin

2!

"#

$

%&'Tbase

However, by averaging Tmax and Tmin over a day, this approach ignores fractions of a day spent below Tbase or above an optimal temperature, Topt. To correct for this omission, it is necessary to interpolate from daily maxima and minima to hourly (or smaller) time-steps. Regardless of which interpolation method is used (see section 6.2.1.5), GDD can then be calculated as follows (for time-steps of size t across the number of days in the growing season, N):

GDDbase,opt = DDtt=1

N! where DD =

0 if Tt < TbaseT "Tbase if Tbase # Tt # ToptTopt "Tbase if Tt > Topt

$

%&&

'&&

(

)&&

*&&

Equation 6-2 was used to calculate GDD throughout this study. This approach (from Lobell et al., 2011) assumes that development proceeds equally above Topt as at Topt. Other approaches (UC IPM, 2003) assume that development slows or stops above Topt, but the former approach is most commonly used for maize.

Equation 6-2

Equation 6-1

134

In each agronomic year with available temperature data at Makoka, GDD were summed over the growing season (the interval between the true planting date and the growing season cessation date for that year; Section 6.2.1.3). The calculations assumed a Tbase of 10 ºC and a Topt of 30º C (Tollenaar and Dwyer, 1999). Degree-days spent above Topt were also calculated.

6.2.1.5 Conversion of daily temperature to hourly temperature

Temperature data, especially at small or remote meteorological stations, are often collected only as daily maxima and minima. For many phenological applications, these daily values must be interpolated to hourly or smaller time steps, but doing so is not straightforward.

An early approach (Lindsey and Newman, 1956) models the daily temperature pattern as a triangle; subsequent refinements include sine approximation (Snyder, 1985), sine-exponential approximation (Parton and Logan, 1981), and a double-sine square root model (Cesaraccio et al., 2001). No single method has proven universally optimal: Wann et al. (1985) recommended Parton and Logan’s sine-exponential model, whereas Cesaraccio et al. (2001) reported that their own model was superior, and Roltsch et al. (1999) concluded that the simple triangle model outperformed its more complex counterparts.

The suitability of a model for a given location can be determined by comparing its output to hourly temperature data from that location, ideally spanning multiple years. For Makoka, the only available temperature data with fine temporal resolution (in this case, half-hourly) were from May 2009 and March-April 2010 (a total of 28 days). Using these data, several interpolation methods were evaluated for accuracy (Table 6-1). The sine-exponential model required parameterization1 and thus was the most time-consuming to use, but it also provided the most accurate GDD estimates by a considerable margin.

Table 6-1. Accuracy of several interpolation methods in recreating 28 days of half-hourly temperature data at Makoka. Models were evaluated for GDD with lower and upper thresholds of 10 and 30 ºC, respectively, and for GDD over 30 ºC. RMSE = root mean squared error.

Data data

Equation 6-‐1 Triangle model

Sine model Sine-‐exp model

Total GDD10,30 274.1 330.1 325.8 329.0 313.2

Daily mean error 21.12% 19.55% 16.84% 12.58%

Daily RSME 22.64% 21.00% 17.64% 13.59%

Total GDD>30 0.42 n/a 0.58 1.42 1.44

Daily mean error n/a 60.0% 320.0% 333.3%

Daily RSME n/a 117.1% 399.9% 419.1%

1 Parameter values fitted from the 28 days of Makoka half-hourly data were α = 1 hr 0 min; β = -1 hr 36 min; and γ = 3.22; these were in reasonable agreement with the values reported by Parton and Logan (1981). See Wann et al. (1985) for a description of the model.

135

All four of the interpolation methods overestimated GDD to a larger extent than reported in the literature (e.g., Parton and Logan, 1981; Wann et al., 1985; Roltsch et al., 1999). This appears to be because the daily maximum temperature usually occurred not as a gradual rise but as a sharp, short-lived spike (Figure 6-5), despite being averaged across 12 data loggers spread over several hectares. It is not known how this pattern might change with a longer data record.

Furthermore, all the interpolation methods greatly overestimated time spent near the maximum temperature (Table 6-1). The sine and sine-exponential models overpredicted time spent above 30 ºC by more than a factor of three. The triangle model, despite its mediocre performance overall, was the most accurate in predicting the duration of temperatures near the maximum (Figure 6-5). (The approach in Equation 6-1 cannot address this question, as it uses only average temperature.)

Thus, the sine-exponential model was selected for calculations of GDD10,30, and GDD>30, with the latter divided by a factor of 3.2 to account for the observed overestimation. Implications of this model choice will be explored in the Discussion.

(a)

(b)

Figure 6-5. Typical fit of (a) sine-exponential model and (b) triangle model to temperature data from Makoka (illustrated by seven days in May 2009).

136

6.2.2 Analysis of future climate

The overarching goal of this portion of the study was to obtain simulations of future climate for this location in Malawi and to evaluate the suitability of the future climate for growing maize. Future climate data were obtained in the same format as the historical climate data (i.e., daily precipitation and daily maximum and minimum temperatures) to facilitate a direct comparison using the same analyses as described in Section 6.2.1.

Two complementary approaches were used: (1) GCM predictions were obtained for changes in the above meteorological variables, and these changes were applied directly to the historical daily data; (2) GCM output was used to run the MarkSim weather generator, and the resulting simulated daily data were subjected to the same analyses as the historical data. The first approach is likely to represent local conditions more faithfully, whereas the second is likely to apply global climate forcings more realistically. Each method is described in detail below.

6.2.2.1 Selection of GCMs and time periods

To employ the two complementary approaches listed above, it was convenient to select GCMs with output available both through the IPCC Data Distribution Centre (ipcc-data.org) and the MarkSim daily weather generator (gismap.ciat.cgiar.org/MarkSimGCM). The three GCMs that met this criteria were BCCR:BCM2 (Norway), CSIRO:MK3 (Australia), and INM:CM3 (Russia). Because each GCM yielded quite different predictions for the grid cell encompassing Makoka (Figure 6-6), results from each GCM were analyzed separately and then compared. Grid cell locations for each model are listed in Table 6-2.

Table 6-2. Locations of closest grid cells to Makoka in chosen GCMs. For BCCR, Makoka is located between two cells (longitudes 36.56 and 33.75); values were linearly interpolated between these cells.

Makoka BCCR:BCM2 CSIRO:MK3 INM:CM3 Latitude -‐15.31 -‐15.34 -‐15.85 -‐16.00

Longitude 35.13 35.16 35.63 35.00 Distance n/a 27.1 km 48.4 km 75.6 km

The time period 2040-2069 was selected for analysis (in accordance with Jones and Thornton, 2003). Earlier predictions may not demonstrate notable departures from present-day climate, whereas predictions further in the future may be rendered increasingly obsolete by agricultural and economic development in Malawi in the intervening decades. All simulations used the SRES A1B scenario, an emissions scenario assuming a “balanced” energy future and moderate population growth (Nakicenovic et al., 2000).

6.2.2.2 Application of predicted changes to historical data

137

Temperature. Monthly predictions were obtained from each GCM for changes in daily maximum temperature and changes in daily minimum temperature during 2040-2069 (Figure 6-6). Predicted changes for the appropriate month were then added to each of the historical maximum and minimum daily temperature values (from 1971-2000), and the resulting data sets were analyzed as described in Sections 0 and 6.2.1.4 above.

Rainfall. There is little value in simply adding predicted changes in precipitation amounts to existing daily precipitation data, because assumptions must be made about how (if at all) climate change will alter the timing, frequency, and variance of precipitation (Wilks, 1992). Thus, no attempt was made to create future daily precipitation data directly from the historical data. Instead, monthly changes in average precipitation (in mm/day) predicted by the GCMs were summed to estimate total precipitation changes over the growing season (November – April) and over the agronomic year.

6.2.2.3 MarkSim daily weather generator

MarkSim is a third-order Markov rainfall generator that has been developed and refined since 1993 (Jones et al., 2009). It uses approximately 120 parameters related to rainfall, temperature, and radiation complied from 10,000 calibration stations worldwide. MarkSim employs several different downscaling methods (including stochastic downscaling and weather typing) to apply predictions from several different GCMs to an 0.5º × 0.5º global grid. Available online (CGIAR, 2011), MarkSim can easily be used to generate daily datasets of future weather at a user-specified latitude and longitude.

Comparison of baseline output to Makoka

(a)

(b)

(c)

Figure 6-6. Output from three GCMs for 2040-2069 for grid cell containing Makoka, Malawi. Shown are monthly average changes in (a) precipitation (mm/day); (b) daily max temperature (ºC); (c) daily min temperature (ºC).

138

historical data. Before analyzing MarkSim’s future weather data, it was prudent to evaluate the accuracy of MarkSim’s current (baseline) weather data for this location. The available years for online MarkSim output are limited to only 1975 (a pre-climate-change baseline from the WorldClim database) and 2010 through 2095 (for output from various GCMs); it is not possible to select individual years in the 20th century.

Thus, to approximate the timespans 1971-2008 (for Makoka historical precipitation data) and 1971-2000 (for Makoka historical temperature data), MarkSim output for 1975 and 2010 was combined as follows. For precipitation, 22 years of 1975 data and 15 years of 2010 data were used (a total of 37 years), and for temperature, 17 years of 1975 data and 7 years of 2010 data were used (for a total of 24 years). In each case, the mean year of these simulated data sets (1989 and 1985, respectively) was the same as for the historical data sets.

These simulated data sets of daily temperature and precipitation were analyzed similarly to the historical data (as described in Section 6.2.1 above), with the exception of interannual trends, which could not be assessed in the simulated data. Descriptive statistics of the simulated and historical data sets were compared.

Generation of future daily climate data for Makoka’s location. The above comparisons indicated that BCCR:BCM2 (Ottera et al., 2009) yielded moderate estimates of future temperature and precipitation changes (more pronounced than CSIRO:MK3 but less pronounced than INM:CM3). Thus, BCCR:BCM2 alone was chosen for this analysis.

Using Makoka’s latitude and longitude, MarkSim was run using output from BCCR:BCM2 for each year from 2040 to 2069. Though this 29-year span was not identical in length to the Makoka historical data sets (which are 37 and 24 years long for temperature and precipitation, respectively), it was chosen to correspond with the time slice available for GCM output from the IPCC Data Distribution Centre. This facilitated direct comparison with the approach in section 6.2.2.2.

These 30-year data sets of daily temperature and precipitation were then subjected to the same analyses as the historical data (described in sections 0 through 6.2.1.5), including growing season length, GDD accumulation, and number of degree-days over 30 ºC.

6.2.3 Evaluating implications for maize production

The ultimate goal of this study was to translate predicted climate changes at Makoka into possible effects on the maize crop. For future work, the most comprehensive approach would be to use a crop growth simulation model with a daily time-step (such as CERES-Maize in the DSSAT model family). However, this preliminary analysis was limited to several quantifiable climate parameters that are known to affect maize growth and yield.

6.2.3.1 Effects of rainfall changes on maize growth

Total growing season rainfall. Rainfall timing and interactions of rainfall with other factors confound the identification of simple thresholds above or below which rainfall is unfavorable for maize growth. Thirteen years of historical yield data from Makoka (Figure 6-7) do not reveal a lower limit for sufficient growing season rainfall; however, they suggest that rainfall in excess of

139

1200 mm may impair maize yield at this location (F. Akinnifesi, pers. comm., 2010). Thus, upper and lower thresholds of concern were defined as >1200 mm and <450 mm, respectively. (The latter amount was chosen to be considerably lower than the lowest rainfall recorded in the 13 years of historical yield data.) Future refinements could use rainfall indices weighted by the water requirements of each maize growth stage (Tadross et al., 2007).

False start of growing season. The percentage of years in which the growing season had a false start (section 6.2.1.3) was calculated both for present climate and for simulated future climates.

Growing season length. Most maize varieties require 120-140 days to reach physiological maturity; the exact duration may depend upon GDD and other factors. The percentage of years in which the actual growing season was less than 120 days (section 6.2.1.3) was calculated both for present climate and for simulated future climates.

6.2.3.2 Effects of temperature changes on maize growth

GDD in growing season. Once the start and end dates of the growing season had been determined according to the precipitation record in the corresponding year, the GDD8,30 were summed according to the method in section 6.2.1.4. In both the historical data and the simulated future data, a GDD sum of less than 1400 Celsius-days was designated as the threshold of risk for poor crop development.

Degree-days above 30 ºC. Lobell et al. (2011), based on a regression analysis of maize yield and climate at 123 sites in SSA, reported that each degree-day spent above 30 ºC reduced yield by 1% under optimal conditions and 1.7% under drought conditions. Because of the difficulty in defining “drought conditions” for any particular site, the 1% decline per DD>30 was used as a conservative assumption.

To evaluate possible socioeconomic implications of this effect, the above factor for yield decline was applied to the maize production of a typical subsistence farm in southern Malawi. A present-day farm was assumed to have 0.7 hectares of cultivated maize (Akinnifesi et al., 2010), average maize yields of 1.1 tons per hectare (Sauer and Tchale, 2009), a household size of five (Snapp et al., 2002), and average per capita maize consumption of 148 kg per person-year (Smale and Heisey, 1997)2. This implies, reasonably, that the average present-day farm would meet its occupants’ needs with 74

2 Maize accounts for two-thirds of all calories in the Malawian diet, making Malawi the most maize-dependent nation in the world (Smale and Heisey, 1997).

Figure 6-7. Relationship between rainfall and yield of unfertilized maize, Makoka, Malawi (1993-2006).

140

person-days of maize left over. The above yield was adjusted by a factor of (-0.01 × GDD>30), and the household maize budget was recalculated.

Number of days above 37 ºC. Previous work (Southworth et al., 2000) has reported that maize crops can be quickly and severely damaged when maximum temperatures exceed 37 ºC. Days during the growing season in which the maximum temperature exceeded 37 ºC were tallied for both historical data and simulated future data.

6.3 Results

6.3.1 Historical trends over time

6.3.1.1 Temperature trends

Both maximum and minimum temperatures showed a slight but steady increase over time (Figure 6-8). Maximum temperatures showed a statistically significant increase of 0.03 ºC per year (p < 0.01), while minimum temperatures showed a non-significant increase of 0.01 ºC per year (p < 0.2). These rates of increase were similar to the global average rate of surface temperature increase over this time period, which was approximately 0.02 ºC per year (Trenberth et al., 2007). At Makoka, contrary to the observed global pattern, maximum temperatures increased more than minimum temperatures.

6.3.1.2 Rainfall trends

None of the investigated rainfall parameters showed a monotonic trend over time (Figure 6-9). Although considerable year-to-year variability was present, there was no evidence for long-term change in the mean or the variability of any parameter studied. A longer data set might be able to reveal multi-year or decadal cycles, but analysis of cyclical trends was not attempted with this 37-year data set.

Figure 6-8. Trends in average maximum and minimum temperatures at Makoka, Malawi, 1971-2000.

141

(a)

(b)

(c)

(d)

142

(e)

(f)

(g)

Figure 6-9. Trends in Makoka rainfall over time, 1971-2008. (a) Total rainfall, 1 July – 30 June. (b) Number of rain days (>2 mm rain). (c) Number of heavy rain days (>20 mm rain). (d) Coefficient of variation of daily rainfall, including zero values. (e) True start date of growing season. (f) End date of growing season (constrained to be after 1 February). (g) Length of growing season. Note: x-axis labels denote the calendar year in which the agronomic year ends.

143

The correlation of ENSO with rainfall (either whole-year or growing-season) varied depending on the ENSO index used. MEI (Multivariate ENSO Index) had no relationship with rainfall (p > 0.95; data not shown). However, NINO3 values averaged over February and March did show a negative relationship with rainfall at Makoka (p < 0.05; Figure 6-10). Although the correlation was not as strong as reported by Cane et al. (1994), it was in the expected direction.

6.3.2 Accuracy of MarkSim baseline climate

The accuracy of MarkSim in replicating Makoka’s current climate varied by parameter, with rainfall simulations generally more accurate than temperature simulations.

6.3.2.1 Daily temperature

Monthly means. Simulated current temperatures (Figure 6-11) were an average of 4 ºC higher than historical temperatures, with the greatest overestimate seen from November to April.

Extremes. The simulation significantly overestimated the highest daily maximum temperatures (Table 6-3). Historically, the highest recorded temperature in any given year was generally in the low to mid-30 degrees C, whereas in the simulated data, the highest temperature in each year was invariably in the high 30s or low 40s.The simulation also significantly underestimated the lowest minimum temperatures.

6.3.2.2 Daily rainfall

Monthly and annual totals. On a monthly total basis, simulated current rainfall (Figure 6-12) was similar to historically measured rainfall. The most notable exception was the month of April, for which simulated rainfall was two to three times actual rainfall. Annual total was also moderately overestimated (Table 6-3).

Figure 6-10. Relationship between the NINO3 index and Nov-Apr rainfall at Makoka, Malawi (1971-2000).

Figure 6-11. Comparison of BCCR:BCM2 baseline monthly temperature output (1975-2010) to Makoka historical data (1971-2008).

144

Frequency and intensity. The simulation correctly estimated the number of rain days per year, but it overestimated the number of heavy rain days (>20 mm) and also overestimated the intensity of the heaviest rain event in a given year (Table 6-3).

Growing season start and end dates. Historical data showed an average growing season start date of 19 November and an average end date of 16 April, with an average length of 149.7 days. The simulated data had a similar average start date (22 November) but a later end date (6 May) and a longer growing season overall (166.2 days; p > 0.01).

These departures of the baseline MarkSim model output from the historical data should be noted when evaluating the results of the MarkSim future climate simulations below.

Table 6-3. Comparison of Makoka historical climate (1971-2008) with climate simulated by MarkSim using BCCR:CM3 (1975-2010). Temperature units are ºC; rainfall units are mm. Parentheses show standard deviations. Significance values are from t-tests between historical and simulated annual values.

Temperature Rainfall

Annual Tmax

Annual Tmin

Highest Tmax

Lowest Tmin

Annual mean

Rain days

Rain days >20 mm

Heaviest event

Historical data

26.02 (0.57)

15.73 (0.26)

33.93 (1.05)

8.04 (0.73)

983.5 (252.9)

56.8 (11.6)

15.3 (4.9)

78.16 (22.9)

MarkSim BCCR:CM3

30.45 (0.47)

19.42 (0.40)

38.53 (1.34)

3.93 (1.98)

1088.3 (225.1)

57.2 (12.1)

16.7 (3.9)

89.8 (29.1)

Significance p < 0.001 p < 0.001 p < 0.001 p < 0.001 p < 0.01 n.s. p < 0.05 p < 0.05

6.3.3 Simulations of future climate

Results of future climate simulations are summarized in Table 6-4, and the most relevant aspects of these results are highlighted in the following sections.

Figure 6-12. Comparison of BCCR:BCM2 baseline monthly precipitation output (1975-2010) to Makoka historical data (1971-2008).

145

Table 6-4. Summary of current and future climate parameters relevant to maize production at Makoka, Malawi. Italic decimal numbers in parentheses (252.9) are standard errors, while fractions in parentheses (5/37) are the number of years that meet the criterion divided by the number of years in the sample.

Historical data

BCCR:BCM2 baseline (1975-‐

2010)

Addition of BCCR changes (2040-‐69) to historical data

MarkSim simulation with BCCR (2040-‐69)

Total annual rainfall 983.5 mm (252.9)

1088.3 mm (225.1) 953.1 mm 1025.2 mm

(286.6)

Total Nov-‐Apr rainfall 936.2 (249.6)

1033.0 (229.5) 909.4 mm 944.8 mm

(270.3)

% of years with Nov-‐Apr rainfall <450 mm 0% 0% -‐-‐-‐ 0%

% of years with Nov-‐Apr rainfall <1200 mm

13.5% (5/37)

13.5% (5/37) -‐-‐-‐ 20.7%

(6/29)

% of years with false start to growing season

27.0% (10/37)

35.1% (13/37) -‐-‐-‐ 27.6%

(8/29)

% of years with growing season <120 days

24.3% (9/37)

13.5% (5/37) -‐-‐-‐ 34.5%

(10/29)

Average GDD10,30 per growing season

1778.4 (403.7)

1733.8 (436.0)

1919.3 (438.9)

1501.3 (559.1)

% of years with growing season <1400 GDD10,30

20.8% (5/24)

29.2% (7/24)

16.7% (4/24)

33.3% (8/24)

Average GDD>30 per growing season

0.53 (0.45)

16.25 (4.83)

2.99 (1.90)

30.42 (14.09)

Yield loss due to GDD>30 (average; maximum)

0.53%; 1.61%

16.25%; 26.67%

2.99%; 6.98%

30.42%; 59.26%

Number of days per year >37 ºC

0.042 (1/24)

2.17 (52/24)

0.042 (1/24)

8.67 (208/24)

Annual average maximum temperature

26.02 ºC (0.57)

30.45 ºC (0.47) 27.69 ºC 30.50 ºC

(0.50)

Annual average minimum temperature

15.73 ºC (0.26)

19.42 ºC (0.40) 17.69 ºC 19.48 ºC

(0.45)

6.3.3.1 Application of predicted changes to historical data

Temperature. Monthly BCCR:BCM2 output for Makoka in 2040-2069 predicted increases of 1.67 and 1.96 ºC (averaged across all months) for maximum and minimum temperatures,

146

respectively. These adjustments increased annual mean maximum temperature from 26.02 to 27.69 and annual mean minimum temperature from 15.73 to 17.69. In the absence of new daily rainfall data, GDD were calculated with the same growing season start and end dates as the original historical data. Thus, with this method, average GDD per growing season inevitably increased (from 1778.4 to 1919.3 GDD).

Rainfall. Monthly BCCR:BCM2 output for Makoka 2040-2069 predicted only modest rainfall changes (as can be seen in Figure 6-6), decreasing total annual rainfall by 3.1% and November – April rainfall by 2.9% for new totals of 953.1 and 909.4 mm, respectively. There were no years in which this minor adjustment crossed the threshold of a wet (>1200 mm) or dry (<450 mm) growing season. Daily rainfall values were not calculated with this method.

6.3.3.2 Output from MarkSim using BCCR:BCM2

Temperature. The annual average maximum temperature (30.50 ºC) from a MarkSim simulation using BCCR:BCM2 output for 2040-2069 was no warmer than the annual average maximum temperature (30.45 ºC) from the 1975-2010 baseline simulation. The lack of change between the baseline simulation (which used both WorldClim and BCCR:BCM2 data) and the future simulation (which used only BCCR:BCM2 data) can be explained if WorldClim estimates higher temperatures than BCCR:BCM2 for a given time period, thus artificially increasing the baseline temperatures. Rather than being a notable result, this outcome suggests that merging these two data sources obscured the warming trend generated by BCCR:BCM2.

Despite the lack of difference in annual average temperature, a large difference was seen in the frequency of extreme temperatures. The 2040-2069 BCCR:BCM2 simulation showed an average of 30.42 GDD>30 per growing season (as compared to 16.25 GDD>30 for the baseline) and an average of 8.67 days per growing season in which temperatures exceeded 37 ºC (as compared to 2.17 days for the baseline).

Rainfall. No notable differences were seen in total rainfall between the baseline simulation and the 2040-2069 BCCR:BCM2 simulation. There was also no evidence for an increased frequency of “false starts” to the growing season. However, the 2040-2069 simulation had an incidence of abnormally short growing seasons (<120 days) more than twice that of the baseline simulation (34.5% as compared to 13.5%). As the total rainfall amount did not change, this may be due to an increased prevalence of long dry spells.

6.3.4 Impacts on maize production

6.3.4.1 Rainfall

Annual and growing season totals. There was no evidence that future climate change – regardless of the method of forecasting – will harm maize production as a result of insufficient average rainfall. In no case did rainfall drop below the 450 mm (November-April) tentatively identified as a risk threshold. In each data set, a sizeable minority of years did have potentially excessive amounts of rainfall (>1200 mm, November-April), but the slight increase in the 2040-2069 simulation (as compared to the baseline simulation or the historical data) could be due to chance.

147

Timing. No changes were apparent in the probability of a false start to the growing season. There was some evidence for an increased probability of truncated (<120 day) growing seasons in 2040-2069, but this was not corroborated by an increased probability of growing seasons with insufficient (<1400) GDD, suggesting that slightly warmer temperatures may help to compensate for slightly shorter growing seasons.

6.3.4.2 Temperature

Monthly and annual averages. Due to the discrepancies between the baseline simulation and the historical data, little can be said about the interpretation of simulated future average temperature. It appears that simulated temperatures are >2 ºC too high on average.

Extremes. The 2040-2069 simulation indicated that days >37 ºC would become common during maize growth (8.67 days per growing season, as compared to 2.17 for the baseline simulation and 0.04 for the addition approach). This analysis does not attempt to quantify the impacts of acute heat stress on maize yield, but it is inevitable that 9 days of >37 ºC would be detrimental. Note, however, that the addition approach suggests negligible risk from this phenomenon.

The number of GDD>30 per growing season increased sharply between the baseline simulation (16.25) and the 2040-2069 simulation (30.42). If one accepts Lobell’s (2011) assertion that each additional GDD>30 leads to at least a 1% loss of maize yield, this translates into losses of 16% and 30%, respectively. However, it should be noted that the addition approach gives very different results than the simulation approach in this regard; the addition approach produced an average of only 2.99 GDD>30 per growing season.

Maize yield decline. Figure 6-13 translates increased GDD>30 into effects on the maize production of an average smallholder farm in southern Malawi (see section 6.2.3.2 for details). In the current climate, temperatures above 30 ºC have little or no impact on maize yield; the same is true for the 2040-2069 addition approach. Only the 2040-2069 MarkSim simulation suggests severe yield declines that could push families deep into food insecurity.

Figure 6-13. Impacts of temperature increases on maize yield in 2040-2069, assuming a 1% decline in yield for every GDD>30. “Ideal” assumes no temperatures >30 ºC; “historical” represents data from 1971-2000; “T addition” represents addition of monthly temperature changes from BCCR:BCM2 to historical data; and “T simulation” represents daily temperature data from MarkSim with output from BCCR:BCM2.

148

6.4 Discussion

Despite the geographic specificity of this case study, aspects of the methods and results can be translated more generally to studies on climate change and crop production in SSA.

6.4.1 Comments on methods

Interpolation of daily maximum and minimum temperatures. All of the interpolation models tested (section 6.2.1.5) yielded a surprisingly poor fit to the empirical half-hourly temperature data, in sharp disagreement with the literature. It is noteworthy that none of these interpolation models (Lindsey and Newman, 1956; Parton and Logan, 1981; Snyder, 1985; Wann et al., 1985; Roltsch et al., 1999; Cesaraccio et al., 2001) were formulated under tropical conditions. Given the indispensability of such models for translating GCM output into agricultural impacts (e.g., Lobell et al., 2011), more attention should be devoted to fine-tuning them for the tropics and perhaps specifically for sub-Saharan Africa.

The interpolation models performed most poorly when temperatures approached their maxima. This study made a rough attempt to ameliorate the problem by dividing by the modeled GDD>30 by its estimated average error, a factor of 3.2. However, robust and accurate methods of estimating time spent near the maximum temperature can be crucial for judging the harm inflicted by a warmer climate. If short-lived spikes in temperature (such as those seen in the Makoka half-hourly data) are common in this region, a different model formulation (not a smooth sine wave) may be necessary to adequately capture their dynamics.

Forecasting with simple addition versus MarkSim simulations. The patterns and implications of future climate appeared quite different depending on the method used. Adding changes from GCMs to historical data sets suggest mild effects, whereas using absolute estimates from a GCM suggest drastic changes with serious negative impacts on maize production. Which of these is likely to be more accurate?

The differences between empirical current climate at Makoka and simulated current climate at Makoka (with WorldClim and BCCR:BCM2) are greater than the predicted future climate changes as generated directly by BCCR:BCM2 (compare Figure 6-6 with Figure 6-11). This suggests that the simple addition approach might give a more accurate estimate of Makoka’s future climate, at least for temperature.

Rainfall patterns with realistic timing and variability cannot be generated with simple addition. However, a Markov chain model parameterized by historical data for a particular location (in this case, Makoka) would likely generate more accurate daily rainfall data than would a Markov chain model driven by GCM output at a 0.5º grid cell resolution. The former approach might be preferable if time permits.

Southern Malawi has highly variable topography: it is studded with small hills (Figure 6-4), punctuated by steep mountains, and intersected by large lakes and rivers. A single GCM grid cell can contain tremendous variation in average rainfall and temperature. In these types of heterogeneous landscapes, working with historical data rather than just downscaled GCM output may be particularly important.

149

6.4.2 Comments on results

It is now appropriate to revisit the six questions asked in the Introduction:

1. Has maximum or minimum temperature increased at Makoka in the past 40 years? Yes, temperatures have increased slightly, in approximate agreement with the global average increase (Trenberth et al., 2007).

2. Has the amount, variability, or timing of rainfall changed at Makoka in the past 40 years? No, there is no evidence that any rainfall parameters at Makoka have changed over the historical record (1971-2008). This corroborates the majority of similar studies in southern Africa (Fauchereau et al., 2003; Shongwe et al., 2009; Tadross et al., 2009; Batisani and Yarnal, 2010; Mazvimavi, 2010).

3. How are future temperature increases likely to affect maize yield at Makoka? How important is acute heat stress versus changes in phenology? The conclusion depends on the forecasting method used. The simple addition approach suggests a mild negative impact of temperatures >30 ºC and virtually no impact from acute heat stress. On the other hand, output directly from the BCCR:BCM2 model for 2040-2069 suggests extensive periods of temperatures >30 ºC, possibly decreasing yields by an average of 30%, and much more common episodes of acute heat stress (>37 ºC).

4. How are future changes in precipitation likely to affect maize yield at Makoka? Will they be more or less important than changes in temperature? Neither forecasting approach predicted significant changes in average annual precipitation for Makoka. In most years (both current and simulated future), rainfall is more than adequate for maize production; in fact, excess rain may be a more common problem than insufficient rain (Sileshi et al., 2011). There was no evidence of increased risk for a false start to the growing season, though the probability of a truncated growing season <120 days appeared to increase between the baseline simulation and the 2040-2069 simulation. Direct effects of precipitation will likely be intrinsically less important than temperature.

5. In light of these findings, what adaptation options might be best suited for this region? In this author’s opinion, the best strategy to ensure food security at Makoka is to focus on coping with current climate variability and improving yields under normal conditions. The potential for improving maize yields with existing technologies (Sileshi et al., 2010) far outstrips any projected losses from climate change over the short- to medium term. If climate adaptation measures are employed, heat-tolerant varieties of maize (or other crops) may prove more useful than drought-tolerant ones. Integrated soil fertility management can improve soil structure to mitigate waterlogging.

6. Do these local results agree with large-scale assessments for the region? If not, how can the differences be explained, and what are the implications for climate adaptation? These results agree with recent studies (e.g., Tadross et al., 2009) indicating that short- to medium-term climate changes over Southern Africa are likely to be subtle, especially for precipitation. In fact, this local analysis indicates a less alarming situation for maize production than do other recent reports (Lobell et al., 2008; Schlenker and Lobell, 2010). This could be due to the simplicity of the present analysis; it could also be because Makoka is somewhat buffered by its generally ample rainfall and moderate temperatures.

150

6.4.3 Drawbacks of the present study

This is study only intended to serve as a rough sketch of a particular location, so the potential areas for improvement are numerous. Perhaps the most apparent is the creation of a baseline climate simulation for 1975-2010 by combining output from WorldClim and BCCR:CM3. As evidenced by the lack of a temperature increase between the baseline simulation and the future simulation (BCCR:CM3, 2040-2069), the two data sources appear to have sufficiently different outputs that they cannot be feasibly merged. A more appropriate solution might be to use several dozen duplicate years of 2010 BCCR:CM3 output as the baseline for future BCCR:CM3 data. The lack of availability of earlier BCCR:CM3 daily data is unfortunate, but the difference between BCCR:CM3 in 1989 and BCCR:CM3 in 2010 may well be less than the difference between WorldClim and BCCR:CM3 in any given year.

Another evident drawback was the relatively short time-series of half-hourly temperature data from Makoka (28 days in May 2009 and March-April 2010). For accurate parameterization of an interpolation model, it would be ideal to have at least one year, if not several years, of half-hourly data. Unfortunately, long-term temperature data sets at this temporal resolution are not easy to procure in SSA. This represents an unmet need for climate modeling and downscaling.

6.4.4 Future work

A useful extension to this project would be to generate daily MarkSim output from other GCMs (perhaps CSIRO:MK3 and INM:CM3) and compare the results to those from BCCR:CM3. As the compilation and analysis of these daily data sets is a time-consuming process, this study only compared the predicted monthly changes generated by these three GCMs (Figure 6-6).

Other authors (e.g., Tadross et al., 2007) have noted possible increases in dry spell length due to future climate change. The current analysis only includes dry spells if they attain the definition of “growing season end” (three dekads of <20 mm each), but shorter dry spells can still be detrimental to crop growth, especially in combination with temperature stress. It could be informative to examine the occurrence of dry spells in historical daily climate data as compared with GCM-driven future climate data, provided that the GCM is capable of generating baseline climate data that accurately replicate dry spells in the historical record.

Many of the questions in this paper could be explored in more detail with a process-based crop model such as CERES-Maize. For example, a crop model would allow quantification of damage from acute heat stress, independent estimates of yield decreases due to accelerated phenology, a more nuanced investigation of the interactions between temperature and drought, and concrete evaluation of climate change adaptation options such as heat-tolerant maize varieties and improved soil infiltration.

6.5 Conclusions

The historical record (1971-2008) shows that dramatic climate changes have not yet become apparent at Makoka, Malawi. Temperatures are slowly increasing in accordance with the global average, while rainfall appears to have no discernable trend over this time period (save for a modest correlation with the NINO3 ENSO index). Current climate is generally quite suitable for

151

growing maize, the region’s staple food, despite occasional crop failures due to poorly-timed rain. Will the next four decades be better, worse, or about the same?

At this particular location in Malawi, dire warnings about plummeting maize yields due to climate change may prove unnecessary. The conclusions, however, very much depend upon the approach used. Taking long-term historical temperature data sets and adding temperature increments predicted for 2040-2069 by the GCM BCCR:BCM2 does not provoke great concern. Expected declines in maize yield due to higher temperatures would be only a few percent at most. However, if one takes for granted the absolute values of the 2040-2069 BCCR:BCM2 output for the grid cell containing this location, the situation looks much bleaker, with temperature-related yield losses of 30% or more.

Given Malawi’s considerable geographic heterogeneity, it seems reasonable to assume that the former approach – based on historical data – is at least as accurate as the absolute values from the GCM. However, simple addition cannot answer questions about future precipitation patterns. In this location, the effects of rainfall changes do not seem likely to equal the effects of temperature changes, but this is only a preliminary investigation. A process-based crop model could help to elucidate the effects of dry spells and temperature-rainfall interactions.

Many maize-growing locations in southern Africa could find value in undertaking a similar analysis using local historical data. The region’s variable climate and topography do not facilitate a “one size fits all” approach. Efficient choices for climate adaptation may demand knowledge on the scale of a district or a village.

153

Chapter 7. Conclusions and policy recommendations Abstract

This dissertation has shown that Gliricidia and Tephrosia intercropping are effective tools to increase maize yields even under drought conditions. Mature Gliricidia trees may help to protect maize from drought damage, and seedlings of both species are resilient to drought impacts, even under suboptimal conditions of late planting. These findings were achieved with the use of low-cost rain exclusion shelters, which may prove useful in other applications.

A number of informal observations deserve further investigation. First, despite views to the contrary in the literature, Gliricidia was found to have formidable labor requirements that are likely to discourage its widespread adoption. Second, fuelwood produced by fertilizer trees was observed to be of lower quality than that harvested from natural forests, implying that fuelwood from these agroforestry systems is of lesser value than previously believed. Third, anecdotes from farmers indicated that displacement of traditional food intercrops is a disincentive to the adoption of fertilizer trees. Fourth, more study is needed on how fertilizer trees may directly benefit biodiversity by providing habitat for miombo forest macrofauna.

Though these results indicate that fertilizer trees are a useful option in current and future climate, they need to be considered in the context of other nutrient management and climate adaptation options for Malawi. Fertilizer trees are unlikely to replace the use of annual legumes and inorganic fertilizer, nor will they substitute for broader-based adaptation efforts such as economic diversification and improved social safety nets for crop failure.

There are numerous opportunities for important future research in this area. Climate manipulations on subsistence agricultural systems in the tropics are almost completely lacking, but the need for such experiments is increasing as climate change continues to threaten agricultural production. Furthermore, little is known about how agroforestry systems will respond to changes in climate, either from a practical or a theoretical standpoint. Pursuit of these research topics will help to promote sustainable development and food security in vulnerable regions under current climate variability and future climate change.

7.1 Summary of findings from field experiments

Following is a brief review of the most important findings from Chapters 3 through 6.

7.1.1 Low-cost rain manipulation experiments can inform climate adaptation

Chapter 3 argued that affordable rain shelter designs, including the one described herein, can help to fill an important gap in agricultural research.

154

7.1.1.1 Simple rain shelter designs can aid agricultural research in the developing world

The shelter design in this dissertation is a novel contribution to the rain manipulation literature, most of which focuses on resource-intensive designs not accessible to developing-country researchers. Although the simple rain shelters described in Chapter 3 were not especially durable nor flexible in their use, and they manipulated only a single variable (precipitation), these may be worthwhile tradeoffs given the affordability and accessibility of the design.

7.1.1.2 Site-specific and species-specific information is valuable for adaptation planning

Manipulation experiments play a valuable role in understanding the effects of environmental change (Shen and Harte, 2000), and the concentration of climate manipulation experiments in developed temperate countries underscores that there is a serious lack of data relevant to climate adaptation in the tropics. Most information on the impacts of climate change on tropical agriculture exists in the form of statistical models that, while a useful tool, are unable to fully account for novel combinations of temperature, water availability, soil type, and crop variety. Conducting more climate manipulation experiments on cropping systems of particular interest will help to fill this gap.

7.1.2 Mature Gliricidia-maize intercropping systems perform well under drought

Chapter 4 showed that, despite the potential for increased tree-crop water competition in a severe drought, Gliricidia trees continued to deliver maize yield benefits.

7.1.2.1 Complementarity of water use during drought makes Gliricidia a “no-regrets” option

Gliricidia has been chosen as an effective intercropping species because its deep rooting patterns complement those of maize (Makumba et al., 2009). The results from the present drought experiment in a mature Gliricidia-maize intercropping system upheld previous findings that Gliricidia and maize do not compete for soil water: the presence of Gliricidia continued to increase maize yields even under drought conditions.

7.1.2.2 Gliricidia may confer modest drought protection under some circumstances

Beyond simply a lack of competition for water, there is some evidence that Gliricidia may help to directly protect the maize crop from the influence of drought, perhaps through influences on air temperature (though not, apparently, on soil moisture). It was observed in 2010-2011 that the impact of the drought on maize yield was proportionally less in Gliricidia plots than in monoculture plots; however, in 2009-2010, the opposite effect was seen, so more evidence is needed to draw conclusions in this regard.

7.1.3 Fertilizer tree seedlings are more resilient to drought than to late planting

Chapter 5 demonstrated that seedlings of both Gliricidia and Tephrosia show considerable resilience to both drought and late planting, though effects differed somewhat by species.

7.1.3.1 Gliricidia seedlings are highly resistant to environmental stresses

Neither drought nor late planting had any significant effect on Gliricidia seedling survival, which remained close to 100% throughout the entire experiment. Although height growth of

155

Gliricidia was initially impeded by both of these stress factors, the disadvantaged seedlings later caught up in terms of biomass production.

7.1.3.2 Tephrosia seedlings suffer from late planting but partly catch up during dry season

Tephrosia seedlings did suffer a significant negative impact from late planting (more so than from drought, but, like Gliricidia, they also largely recovered from these losses by the end of the dry season. The details of the effects differed somewhat between relay intercropping systems and improved fallow systems.

7.1.3.3 Young fertilizer tree systems benefit maize yield even in suboptimal conditions

All three systems tested at Nkula Field performed well in terms of their effect on maize yield, even under drought conditions and late planting. The greatest yield benefits were seen in Tephrosia improved fallow systems, but this calculation depends on the number of years over which the yield effect is averaged. Unlike Tephrosia improved fallows and Gliricidia intercropping, both of which depressed or negated maize yield in the first two years, Tephrosia relay intercropping increased maize yield every year after establishment.

7.1.4 Existing climate variability may matter more than future climate change

Chapter 6 examined the current state of knowledge on predicted future climate and maize yields in Malawi and concluded that the challenges of existing climate variability will predominate over the effects of global climate change for the next several decades.

7.1.4.1 Predicted rainfall changes in Malawi fall within scope of current variability

Climate forecasts for Malawi do not generally indicate a major change in rainfall patterns, though there may be some truncation of the beginning of the growing season (Tadross et al., 2009). At Makoka Agricultural Research Station, the focal point of this study, rainfall is generally adequate for maize production, so future climate change is not likely to make conditions any worse on average for the next several decades. However, increasing rainfall variability and dry spell length, combined with an increase in temperature, may lead to more periods of extreme water stress. Implementing options to cope with current climate variability will help to prepare agricultural systems for future climate change (Cooper et al., 2008).

7.1.4.2 Temperature increase may threaten maize production more than changes in rainfall

Unlike precipitation trends, temperature trends at Makoka are already clearly changing and are likely to continue to do so. Warmer future temperatures may pose a risk to maize production, as daytime maximum temperatures during the growing season are already close to the threshold of 30 ºC beyond which maize yield is depressed (Lobell et al., 2011). This suggests a possible reason to seek microclimatic benefits from agroforestry trees.

7.1.4.3 Climate and crop yield predictions from GCMs should be calibrated to local baselines

The output of several GCMs (General Circulation Models) describing future climate at this location in Malawi would have borne little relevance to reality if considered in absolute terms rather than relative terms. The inaccuracy of baseline GCM output for the location’s current climate considerably exceeded the magnitude of predicted future change. This is perhaps unsurprising considering the mismatch between the scale of GCM grid cells and the scale of the

156

dramatic topography in southern Malawi, where rainfall can vary substantially even from one village to the next. Thus, when predicting and preparing for future climate change in this region, it will be essential to make use of local climate data sets at the smallest scale possible.

7.2 Additional observations on fertilizer tree systems in Malawi

Beyond the formal results presented in Chapters 3 through 6, this project generated a number of informal observations that may be worth following up in future work. None of the following observations on fertilizer trees were pursued quantitatively, and therefore any conclusions must be tentative at best. However, they may help to place the central findings of this thesis in the broader context of the strengths and weaknesses of fertilizer tree systems in actual practice.

7.2.1 Labor requirements pose a major obstacle to Gliricidia intercropping

The author fully participated in all the management tasks for each of the agroforestry systems during the first two years of the experiment. Detailed records were not kept of the time spent on each of these tasks; however, it was readily apparent the different cropping systems presented profound differences in labor requirements.

7.2.1.1 Gliricidia has exacting labor requirements and interferes with other field tasks

Gliricidia – both as seedlings and as mature trees – were enormously more labor-intensive than sole maize or Tephrosia. Even an accurate accounting of hours would not convey the extent of the difficulty of Gliricidia management: much of the work was physically demanding (slashing large branches with a machete) or exacting and tedious (burying pruned branches underneath ridges with young maize plants).

Maize-focused tasks (planting, thinning, weeding, fertilizing, and harvesting) were also made more difficult and time-consuming by the presence of the Gliricidia trees. In the newly-established Gliricidia plots, the seedlings were often accidentally trampled or hoed; in the mature Gliricidia plots, tripping on the gnarled stumps caused inconvenience or even injury. One of the field assistants recounted an incident several years earlier in which he had been hospitalized due to an infected cut on his leg from a Gliricidia stump.

These problems were much less apparent in Tephrosia systems. In Tephrosia improved fallows, maize and trees are grown sequentially rather than simultaneously, eliminating the difficulty of maintaining both components in close proximity to each other. Moreover, Tephrosia improved fallows are only cut once in a five-year cycle, in contrast to Gliricidia, which are cut three times per year. In Tephrosia relay intercropping, trees grow on the sides of the ridges rather in the furrows (Figure 5-7), making them less of an obstacle to field management tasks. They are cut only once per year, and their stems remain small enough to make cutting relatively easy.

7.2.1.2 Labor requirements of Gliricidia are greatest in the establishment phase

Gliricidia also has the disadvantage of a very labor-intensive establishment process. As described in Chapter 5, the recommended method for establishing Gliricidia seedlings is to grow them in a raised nursery bed with frequent watering and root-pruning before transplanting them to the field. The author spent approximately 40 hours carrying out these tasks for the 324 seedlings planted in 2009-2010. At a planting density of 7,400 seedlings per hectare, this

157

equates to 900 hours of labor to establish one hectare of Gliricidia. Approximately half of this labor demand falls at the onset of the rains during a period of a few weeks when farmers have many demands on their time. Even assuming a farm size of 0.5 ha and a time window of four weeks, this equates to almost 50 hours per week demanded by Gliricidia establishment.

Tephrosia, by contrast, is direct-sown in the field, and can be planted at the rate of about 27 hours per hectare in the author’s experience (assuming a density of 14,800 planting stations per hectare). This is less than one-thirtieth the time required to plant Gliricidia over the same area.

7.2.1.3 The scope and consequences of Gliricidia labor requirements need further study

In light of this vast difference in labor demand for establishment, and the much lower labor demand for tree management and pruning, it seems unsurprising that Malawian farmers use Tephrosia much more widely than Gliricidia. However, this explanation does not precisely concur with the published literature.

Akinnifesi et al. (2008) acknowledge that Gliricidia has seen minimal adoption in Malawi, but they attribute this to lack of available germplasm and to the depression of maize yield that can occur in the first two years, rather than to the system’s high labor requirements. Ajayi et al. (2009), in a survey of smallholder farmers in Zambia, found that Gliricidia had comparable labor inputs to sole maize, and it yielded higher returns to labor than did Tephrosia or Sesbania improved fallows. They attributed unfavorable opinions about Gliricidia labor demands to the fact that these demands might conflict with other field tasks, though their data did not indicate that such a conflict was inevitable.

These discrepancies suggest that farmers’ experience with Gliricidia labor requirements is an important topic for further investigation, especially in light of tradeoffs between promoting Gliricidia and other promoting other soil fertility technologies. Vanlauwe and Giller (2006) argue that, historically, green manure crops have only been spontaneously adopted amongst smallholders if the cropping system imposes no extra labor cost. This hypothesis was echoed by the aforementioned field assistant, who offered this succinct explanation for why he did not use Gliricidia on his own farm: “It is very hard work.”

7.2.2 Fuelwood from fertilizer trees is less desirable than fuelwood from forests

7.2.2.1 Quantity is no substitute for quality

Fertilizer tree systems can produce a substantial amount of fuelwood. For example, early-planted Tephrosia improved fallows at Nkula Field produced 43 tons of wood per hectare in October 2010 after two years of growth. Even averaged over the five-year fallow cycle, that equates to more than 8 tons per hectare per year – in theory, more than adequate to supply fuelwood in perpetuity to a typical six-person household on an 0.5-ha farm (assuming 10 kg per capita per week; Abbot and Homewood, 1999).

However, it can be misleading to simply report the total wood biomass produced by agroforestry plots without acknowledging its quality as fuelwood. As described by Brouwer et al. (1997), Malawian households tend to strongly prefer split-wood and large branches as fuel. Small branches and twigs are generally only used if no other options are available.

158

Unfortunately, wood produced by Gliricidia intercropping and Tephrosia relay intercropping is uniformly of small diameter (<3 cm). Tephrosia improved fallows can produce some larger trunks of up to 15 cm diameter, but nonetheless, the majority of the wood produced by the improved fallow system does not fall into the “desirable” categories defined by Brouwer et al.

7.2.2.2 Small branches are less useful than large logs

There are several reasons why small-diameter branches are less desirable: (1) they tend to be less dense (especially Gliricidia, which have a pithy or hollow core) than large logs, and thus produce a cooler flame; (2) they cause the fire to demand much more frequent tending. In rural Malawian households, the fire is often left burning for much or all of the day while the head of the household tends to various tasks inside or outside the house. This approach is only possible with large-diameter logs (shown in Figure 7-1).

Many authors (for example, Akinnifesi et al., 2010) have mentioned production of fuelwood by fertilizer tree systems as a labor-saving and forest-conserving ancillary benefit. However, the quality of wood produced by fertilizer trees is overlooked in the literature. Poor wood quality may help to explain why such systems have not been more widely adopted, even in areas suffering from fuelwood scarcity.

During midday breaks from work at Nkula Field, the field assistants regularly ventured into the adjacent forest (Figure 2-4) to collect large branches to bring home for cooking lunch. This practice continued even after they had been invited to take the cut Tephrosia branches piled around the field. The Tephrosia branches remained unused until the workers were

specifically asked to remove them. This preference may not be universal, but it certainly deserves more attention than it has received in the agroforestry literature.

7.2.3 Fertilizer trees interfere with traditional food intercrops

7.2.3.1 Intercropping is already used by nearly all farmers in southern Malawi

Although fertilizer tree experiments usually employ monoculture maize as a control, monoculture maize is not the norm in southern Malawi. A survey of smallholder farms in six villages near Zomba (only a few kilometers from the site of the present study) indicated that 95% of farms were intercropped to some degree (Shaxson and Tauer, 1992). Although 84% of the total farm area in the survey was planted to maize, 85% of that maize was grown with one or more intercrops (for example, pigeonpeas, cowpeas, sorghum, or pumpkins). This practice is illustrated in Figure 2-1 and Figure 2-2.

Intercropping maize with fertilizer trees usually precludes the use of traditional food intercrops, because the introduction of trees into a maize polyculture risks creating conflicts of physical space, resource use, and management tasks. Thus, farmers are usually advised to plant no other species along with fertilizer trees and maize. That is not to say that integrating a food intercrop

Figure 7-1. Stirring nsima (maize porridge) on a typical cooking fire, fueled by two large logs.

159

in these systems is impossible: Chirwa et al. (2003a) successfully incorporated pigeonpea into a Gliricidia-maize intercropping system. However, the topic has received little research attention.

7.2.3.2 Food intercrops fill important needs that fertilizer trees do not

A well-managed fertilizer tree system would be expected to increase maize yields to such an extent that total food production exceeds that of a traditionally intercropped maize system. However, this simple calculus ignores three important factors: (1) timing of harvest; (2) diversity of nutrition sources, and (3) sale of cash crops.

First, some food intercrops (such as pumpkins and sweet potatoes) mature several months earlier or later than the maize crop, helping to alleviate periods of hunger. Second, intercrops may provide nutritional benefits that maize cannot: for example, legumes are a key source of protein for rural Malawians (Snapp et al., 2010). And third, grain legumes such as pigeonpea can provide an important source of cash income to poor households (Sirrine et al., 2010).

In light of this, it is perhaps unsurprising that farmers usually prefer edible legumes to fertilizer trees when the options are directly compared in on-farm trials (for example, Bezner-Kerr et al., 2007; Kamanga et al., 2010), even if the edible legumes convey less benefit to maize yield. Historically, efforts to promote fertilizer trees have not acknowledged the logic nor even the existence of this preference. Devising ways to integrate edible legumes or other food intercrops could make fertilizer tree systems more appealing to smallholder farmers.

7.2.4 Fertilizer trees may directly contribute to biodiversity conservation

7.2.4.1 Agroforestry can affect biodiversity in many ways, but effects are poorly documented

The biodiversity benefits of agroforestry in southern Africa have been mentioned by several authors (Chirwa et al., 2008; Akinnifesi et al., 2010; Syampungani et al., 2010), but in different contexts. Chirwa et al. discuss the role of agroforestry in preserving indigenous tree species; Akinnifesi et al. review the beneficial effects of fertilizer trees on soil fauna; and Syampungani et al. assert that the timber and non-timber outputs of agroforestry systems can help reduce pressure on natural forests.

This variety of perspectives belies the fact that little is actually known about the significance of agroforestry’s effects on any of these aspects of biodiversity. With the exception of several detailed studies on soil macroinvertebrates under improved fallows in Zambia (for example, Sileshi and Mafongoya, 2007), most of the literature connecting agroforestry and biodiversity in southern Africa is based on assertion rather than data. (The assumption that fertilizer tree systems lead to “land-sparing” of natural forests is especially tenuous in light of the issue of fuelwood quality discussed in Section 7.2.2 above.) The role of agroforestry in biodiversity conservation in southern Africa seems to be wide open to future research, and one aspect in particular appears to have received no attention thus far: the ability of fertilizer trees to serve as habitat or migration corridors for miombo fauna.

7.2.4.2 Fertilizer trees may directly provide habitat for displaced miombo fauna

Miombo, a type of open woodland dominated by Brachystegia trees, is the prevalent forest type across much of southern Africa. Much miombo forest has already been lost to land clearing for

160

agriculture and fuelwood, and human exploitation of these forests poses an ongoing risk to their biodiversity (Sileshi et al., 2007).

Might fertilizer trees play a direct role in mitigating this risk by providing habitat or migration corridors for displaced miombo fauna? A tree-maize intercrop, though very different from a natural forest, might nonetheless provide a more hospitable substitute than a field of annual crops. Other types of agroforestry systems have been shown to support a greater diversity of macrofauna than comparable monoculture systems (for example, Steffan-Dewenter et al., 2007), and it has been suggested that this effect can help facilitate the movement of forest species in response to habitat fragmentation and climate change (Harvey et al., 2004).

No attempt has yet been made to quantify such an effect for fertilizer trees, but the author’s informal observations at Nkula Field indicated that macrofauna – in this case, insects, reptiles, amphibians, and birds – were more diverse and abundant in agroforestry plots than in sole maize plots (Figure 7-2). Birds provided the most notable examples. Numerous woodland birds (such as the pin-tailed whydah, Vidua macroura, and the scarlet-chested sunbird, Chalcomitra senegalensis) were observed perching, foraging, and courting on the trees; a black-eyed bulbul (Pycnonotus barbatus) successfully nested in a Tephrosia improved fallow.

(a)

(b)

(c)

(d)

(e)

(f)

Figure 7-2. Examples of animals observed utilizing Tephrosia and Gliricidia trees at Nkula Field. (a) frog; (b) tortoise beetle; (c) chameleon; (d) pin-tailed whydah; (e) ladybug; (f) black-eyed bulbul (nest). Miombo is the richest bird habitat in southern Africa, both in terms of population density and species diversity (Newman, 2002). If fertilizer tree systems can contribute to the maintenance of this diversity, that will be a meaningful reason to encourage their further dissemination.

161

7.3 Comparing fertilizer trees with other options

Although this dissertation has focused solely on fertilizer tree systems from a biophysical perspective, such a narrow focus is not appropriate from a policy perspective. Policymakers should consider fertilizer trees in the context of other options for improving soil fertility: namely, other types of legumes and inorganic fertilizer.

7.3.1 Annual legumes and grain legumes

It is uncommon to include annual legumes as a comparison in agroforestry experiments, and as a result, information on the benefits of annual versus perennial legumes is for the most part parallel rather than integrated. One exception is a study by Snapp et al. (2002) comparing the effects of pigeonpea, groundnut, and Tephrosia vogelii on Malawian farms; they observed no maize yield benefit from any of these legumes over a two-year period. Other studies have produced mixed results: Chirwa et al. (2003a) found that Gliricidia conveyed large benefits to maize yield while pigeonpea conferred none, and several other authors have suggested that the effects of grain legumes on maize yield are positive and comparable to those of Tephrosia relay intercropping (Kamanga et al., 2010; Snapp et al., 2010).

Regardless of the magnitude of soil fertility effects, it is clear that grain legumes (all of which are annuals except for pigeonpea) are in much more widespread use in Malawi than are fertilizer trees. This being the case, further research is needed on the relative soil fertility impacts of each type of legume and in what cases the two options can be integrated (as discussed in Section 7.2.3).

The reasons for the predominance of grain legumes deserve further consideration. Bezner-Kerr et al. (2007) emphasize that Malawian farmers are skeptical about the use of legumes for the sole purpose of soil fertility improvement; rather, they prefer legumes that produce food and other co-benefits. This observation is echoed by Sirrine et al. (2010). The driving forces are not just agronomic, but also social and cultural: for example, in some parts of Malawi, trees are considered the property of men while food crops are assumed to be in the domain of women. These traditions seem to have contributed to a strong preference for pigeonpea and groundnut over Tephrosia among the female farmers interviewed by Bezner-Kerr et al. (2007).

7.3.2 Inorganic fertilizer

The current subsidy program (described in Chapter 2) has made inorganic fertilizer a much more affordable option for Malawian smallholders. Addition of inorganic fertilizer usually improves maize yields beyond what can be achieved with agroforestry practices alone (Ajayi et al., 2009; Kamanga et al., 2010). Although inorganic fertilizer cannot be expected to have the same beneficial effects on soil structure and microclimate as fertilizer trees, it can nevertheless reduce yield variability in maize in Malawi (Snapp et al., 2010), though not to the same extent as pigeonpea and Tephrosia.

On the other hand, even with the subsidy program, fertilizer is a cash expense. Losing this investment as a result of unfavorable rains can be a major hardship for farmers. This occurred during the failed start to the rainy season of December 2009 – January 2010 (FEWSNET, 2010), when many farmers had to replant maize two or three times. In light of the potential

162

synergies between chemical fertilizer and agroforestry inputs (Akinnifesi et al., 2007), increasing farmers’ access to both inorganic and organic nutrient sources may be the best way to ensure long-term food security.

7.4 Climate adaptation options for Malawian agriculture

It is important to keep in mind that agroforestry is only one of many tools to improve livelihoods of Malawian farmers under current and future climate conditions. Denning et al. (2009) list the following policy priorities for climate adaptation in Malawian agriculture: water harvesting, improved water use efficiency, and expanded irrigation; conservation farming (including agroforestry); heat- and drought tolerant maize varieties; diversification to high-value crops, value-added products, and off-farm employment; weather forecasting; and crop insurance. Several of these options are discussed in more detail below.

7.4.1 Technical approaches

7.4.1.1 Improved maize cultivars

In developing maize cultivars that are suitable for future climates in sub-Saharan Africa, resistance to high temperatures may be more important than resistance to drought (Lobell et al., 2011). A successful tropical maize variety must cope well not only with temperature extremes, but also with the phenological acceleration (and concomitant loss of interception of solar radiation) caused by warm days (Muchow, 2000). However, nutrient inputs will remain an important issue: improved maize cultivars may not provide any benefits over local varieties in the absence of adequate fertilizer (Snapp et al., 2010).

7.4.1.2 Crop diversification

Cash crops. Diversification away from tobacco is a necessary goal even in the absence of climate change. Perennial cash crops such as coffee, tea, and macadamia would potentially be more drought-tolerant than tobacco, though their heat tolerance may not be adequate in a warmer climate.

Staple crops. It has been suggested that shifting away from maize toward more drought-tolerant staple crops, such as cassava and sorghum, would improve the drought resilience of agriculture in southern Africa. However, Malawi already produces substantial quantities of root crops (chiefly cassava and sweet potato), nearly equal in tonnage to maize (Mkandawire, 1999) before correcting for moisture content. The supposedly overwhelming preference of Malawians for maize is often overstated (Devereux, 2002b). More work is needed on whether changes in mixtures of staple crops are possible in Malawi in the near- to medium-term, and if so, to what extent they can contribute to climate resilience.

7.4.1.3 Irrigation and micro-irrigation

As mentioned in Chapter 2, only a tiny fraction (approximately 0.5%) of Malawi’s arable land is irrigated. Although the country generally has adequate surface water supplies –in Lake Malawi, in several other large lakes, and in numerous rivers – the country’s mountainous topography and generally poor infrastructure have impeded the use of irrigation in all but the most profitable large estates. Increasing the amount of irrigated cropland is a policy focus of

163

the current government (Mpaka, 2010), though the geographic focus is restricted to the shores of Lake Malawi, where the logistics of large-scale irrigation are relatively straightforward. Making small-scale irrigation accessible to farmers in Malawi’s more remote and hilly areas remains a major challenge on which little work has been done.

7.4.1.4 Weather forecasting

Detailed foreknowledge of the amount and timing of rainfall in a given season would allow farmers to optimize their choices of when and what to plant. Although such detailed predictions are currently impossible, precipitation patterns in southern Africa tend to be reliably influenced by global oceanic and atmospheric processes (Hoerling et al., 2006; Stige et al., 2006) and are thus amenable to coarse-scale predictions. Timely information delivered to farmers regarding the probability of rainfall distribution in the current growing season can determine the success or failure of a harvest. Several recent publications discuss this issue in detail in the southern African context (O'Brien and Vogel, 2003; Vogel and O'Brien, 2006).

7.4.2 Structural approaches

7.4.2.1 Improving social safety nets in case of crop failure

Better management of Malawi’s national grain reserve could go a long way toward buffering the effects of climate variability on food security; meeting food shortages from national grain reserves costs three to four times less than importing grain (Devereux, 2002b). Access to crop insurance, taken for granted in most developed countries, is currently non-existent in Malawi; insurance schemes accessible to smallholders could help to reduce the impacts of climate disturbances both on small and large scales.

7.4.2.2 Strengthening national infrastructure

Improving transport, food storage, information technology, and other infrastructure is essential for the development of the agricultural sector and for poverty reduction (Government of Malawi, 2006), which in turn will increase options for coping with climate impacts.

7.4.2.3 Economic diversification

Due to its geographical isolation (Young, 2005) and its high ratio of unskilled to skilled workers, Malawi is not currently well-situated to be a manufacturing or service hub. A better educational system is a key step to expanding Malawians’ economic opportunities, but this will necessarily be a slow process (Conroy et al., 2006). Agriculture will remain the key livelihood activity for most Malawians for the foreseeable future, underscoring the importance of identifying and promoting agricultural systems that are sustainable in the face of climate variability and change.

7.4.2.4 Poverty reduction

Households with more financial resources are better situated to cope with any kind of external stress, including climatic ones (Thorlakson, 2011). Furthermore, households with a greater complement of skills, labor, and capital have a wider variety of income-generating options available to them during climatically unfavorable periods (Eriksen et al., 2005). Thus, most

164

schemes that successfully reduces rural poverty is also likely to mitigate the negative impacts of climate change on rural communities in Malawi.

7.4.3 Agroforestry in context

Agroforestry in general, and fertilizer tree systems in particular, can in theory contribute to several of the above types of climate adaptation activities: in particular, crop diversification, as well as the broader goals of income generation and poverty reduction. Many of the solutions mentioned above will take many years or decades to implement, during which time agroforestry can play a role in stabilizing food production under conditions of current climate variability. Although agroforestry will not reshape Malawi’s national economy to the extent necessary to prosper in future climate, it is for the most part a widely accessible “no-regrets” option to be considered along with other methods of enhancing farm livelihoods.

7.5 Knowledge gaps and future research priorities

This section presents a brief synthesis and summary of the research needs discussed in the previous six chapters.

7.5.1 Controlled climate manipulations in tropical subsistence agriculture

Most of the current evidence for agricultural impacts of climate change in the tropics comes from statistical simulation models, not from experiments in the field. However, manipulation experiments are the most rigorous way to link causes and effects, as well as the most effective way to reliably create novel combinations of temperature, precipitation, soil type and cropping system. In light of the potentially profound impacts of climate change on subsistence agriculture in sub-Saharan Africa, much more concrete information is needed on how particular cropping systems will respond to particular changes and how these impacts can be ameliorated (Thornton et al., 2011).

7.5.2 Systematic assessment of agroforestry’s role in climate adaptation

In recent years, several authors have suggested that agroforestry systems can make important contributions to climate adaptation but have failed to provide concrete evidence for this claim (for example, Syampungani et al., 2010). The majority of papers addressing the topic of agroforestry for climate adaptation have tended to focus purely on its economic effects rather than its biophysical effects (Kalame et al., 2011; Thorlakson, 2011).

Several recent studies openly call for more work in this area: “Research into the contributions of agroforestry in buffering against climate variability is not well advanced” (Verchot et al., 2007); “Few studies examine the specific role agroforestry techniques can play in helping farmers reduce their vulnerability to climate change” (Thorlakson, 2011); and “Very little research has been done on the impacts of climate change on agroforestry systems” (Neufeldt et al., 2012).

Multiple different approaches (including climate manipulation experiments, simulation models, and meta-analyses) can be applied to shed light on this topic. The variety of agroforestry systems around the world and the multiple dimensions of climate change (changes in CO2,

165

temperature, precipitation, extreme events, and temporal variations thereof) imply that the scope of possible research is vast.

7.6 Summary and conclusions

Motivated by the expected negative impact of future climate change on smallholder agricultural production in southern Africa, this dissertation set out to investigate the role of agroforestry technologies – specifically fertilizer trees – in helping farmers cope with rainfall variability.

Experimental rain manipulation indicated that fertilizer trees perform well under drought conditions. In a mature Gliricidia-maize intercropping system, maize yields remained significantly higher in agroforestry plots than in maize monoculture plots, even when a severe drought was imposed for two years. There was some evidence that the Gliricidia trees helped protect maize from drought stress via effects on microclimate and soil properties, though further research is needed.

Newly-established Gliricidia and Tephrosia systems also proved resilient to drought. Though the seedlings suffered temporary setbacks in survival and growth, they largely regained lost biomass after the drought was lifted, and they conferred dramatic benefits on maize yield even under drought conditions. For Tephrosia, timely planting accentuated these benefits.

These results suggest that fertilizer trees in southern Africa are a useful tool for soil fertility improvement even under adverse climatic conditions. Increased water stress as a result of future climate change does not appear to contraindicate the use of Gliricidia and Tephrosia intercropping systems; on the contrary, they may help stabilize maize yield in the face of climate variability.

It is important, however, not to overgeneralize these findings. A variety of other organic and inorganic technologies are available for soil fertility improvement in southern Africa; drought resilience is only one of many criteria for prudent selection. Furthermore, the response of agroforestry systems to climate-induced stresses is likely to be highly dependent upon the tree and crop species, the location, and the exact nature of the stress. Much more research is needed to accurately delineate the broader role of agroforestry in climate change adaptation.

In Malawi, a primary goal of climate adaptation must be to identify livelihood-enhancing agricultural systems that cope well with current climate variability while also buffering against future climate change. Gliricidia-maize and Tephrosia-maize intercropping may be examples of such systems. Many more such options are needed to ensure that Malawi achieves sustainable agricultural production and food security in the coming decades.

167

References Abbot, J. I. O. and K. Homewood (1999). "A history of change: causes of miombo woodland decline in a protected area in Malawi." Journal of Applied Ecology 36 (3): 422-433.

Ajayi, O. C., F. K. Akinnifesi, G. Sileshi and W. Kanjipite (2009). "Labour inputs and financial profitability of conventional and agroforestry-based soil fertility management practices in Zambia." Agrekon 48 (3): 276-292.

Ajayi, O. C., S. Franzel, E. Kuntashula and F. Kwesiga (2003). "Adoption of improved fallow technology for soil fertility management in Zambia: Empirical studies and emerging issues." Agroforestry Systems 59: 317-326.

Ajayi, O. C., F. Place, F. K. Akinnifesi and G. W. Sileshi (2011). "Agricultural success from Africa: the case of fertilizer tree systems in southern Africa (Malawi, Tanzania, Mozambique, Zambia, and Zimbabwe)." International Journal of Agricultural Sustainability 9 (1): 129-136.

Akinnifesi, F. K., O. C. Ajayi, G. Sileshi, P. W. Chirwa and J. Chianu (2010). "Fertiliser trees for sustainable food security in the maize-based production systems of East and Southern Africa. A review." Agronomy for Sustainable Development 30 (3): 615-629.

Akinnifesi, F. K., P. W. Chirwa, O. C. Ajayi, G. Sileshi, P. Matakala, F. R. Kwesiga, . . . W. Makumba (2008). "Contributions of agroforestry research to livelihood of smallholder farmers in Southern Africa: 1. Taking stock of the adaptation, adoption and impact of fertilizer tree options." Agricultural Journal 3 (1): 58-75.

Akinnifesi, F. K. and F. R. Kwesiga (2002). "Can agroforestry make a difference as an exit strategy for the Starter Pack Scheme in Malawi?" Malawi Journal of Agricultural Sciences 1 (1): 53-65.

Akinnifesi, F. K., W. Makumba, G. Sileshi, O. C. Ajayi and D. Mweta (2007). "Synergistic effect of inorganic N and P fertilizers and organic inputs from Gliricidia sepium on productivity of intercropped maize in Southern Malawi." Plant and Soil 294: 203-217.

Akinnifesi, F. K., J. Nyirongo, T. M. Cullen, P. Matakala, G. Sileshi, O. C. Ajayi and W. Makumba (2006). Gliricidia-Maize Intercropping: An Extension Trainers' Guide. Lilongwe, Malawi: World Agroforestry Centre (ICRAF).

Akinnifesi, F. K., E. C. Rowe, S. J. Livesley, F. R. Kwesiga, B. Vanlauwe and J. Alegre (2004). "Tree root architecture." Chapter 4 in Below-ground Interactions in Tropical Agroecosystems: Concepts and Models with Multiple Plant Components. M. van Noordwijk, G. Cadisch and C. K. Ong. Wallingford, UK, CAB International: 61-81.

Albrecht, A., G. Cadisch, E. Blanchart, S. M. Sitompul and B. Vanlauwe (2004). "Below-ground inputs: relationships with soil quality, soil C storage and soil structure." Chapter 11 in Below-ground Interactions in Tropical Agroecosystems: Concepts and Models with Multiple Plant Components. M. van Noordwijk, G. Cadisch and C. K. Ong. Wallingford, UK, CAB International: 193-207.

168

Anderson, E. K. and H. Zerriffi (2012). "Seeing the trees for the carbon: agroforestry for development and carbon mitigation." Climatic Change In press: 17 pp.

Babayemi, O. J., I. O. Daniel, M. A. Bamikole, A. Ogungbesan and B. O. Oduguwa (2003). "Preliminary studies on Tephrosia species: effect of seed treatments on germination." Nigerian Journal of Animal Production 30 (1/2): 209-216.

Balvanera, P., A. B. Pfisterer, N. Buchmann, J.-S. He, D. Raffaelli and B. Schmid (2006). "Quantifying the evidence for biodiversity effects on ecosystem functioning and services." Ecology Letters 9: 1146-1156.

Batisani, N. and B. Yarnal (2010). "Rainfall variability and trends in semi-arid Botswana: implications for climate change adaptation policy." Applied Geography 30 (2010): 483-489.

Bauer, S. and I. Scholz (2010). "Adaptation to climate change in Southern Africa: New boundaries for sustainable development?" Climate and Development 2 (2): 83-93.

Beedy, T. L., S. S. Snapp, F. K. Akinnifesi and G. W. Sileshi (2010). "Impact of Gliricidia sepium intercropping on soil organic matter fractions in a maize-based cropping system." Agriculture, Ecosystems & Environment 138 (3-4): 139-146.

Bernacchi, C. J., B. A. Kimball, D. R. Quarles, S. P. Long and D. R. Ort (2007). "Decreases in stomatal conductance of soybean under open-air elevation of CO2 are closely coupled with decreases in ecosystem evapotranspiration." Plant Physiology 143 (1): 134-144.

Bezner-Kerr, R., S. Snapp, M. Chirwa, L. Shumba and R. Msachi (2007). "Participatory research on legume diversification with Malawian smallholder farmers for improved human nutrition and soil fertility." Experimental Agriculture 43: 437-453.

Boko, M., I. Niang, C. Vogel, A. Githeko, M. Medany, B. Osman-Elasha, . . . P. Yanda (2007). "Africa." Chapter 9 in Climate Change 2007: Impacts, Adaptation and Vulnerability. Contribution of Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. M. L. Parry, O. F. Canziani, J. P. Palutikof, P. J. van der Linden and C. E. Hanson. Cambridge, UK, Cambridge University Press: 433-467.

Bond, W. J. and G. F. Midgley (2000). "A proposed CO2-controlled mechanism of woody plant invasion in grasslands and savannas." Global Change Biology 6 (8): 865-869.

Brouwer, I. D., J. C. Hoorweg and M. J. Van Liere (1997). "When Households Run Out of Fuel: Responses of Rural Households to Decreasing Fuelwood Availability, Ntcheu District, Malawi." World Development 25 (2): 255-266.

Burke, M. B., D. B. Lobell and L. Guarino (2009). "Shifts in African crop climates by 2050, and the implications for crop improvement and genetic resources conservation." Global Environmental Change 19: 317-325.

Byerlee, D. and P. W. Heisey (1997). "Evolution of the African maize economy." Chapter 2 in Africa's Emerging Maize Revolution. D. Byerlee and C. K. Eicher. Boulder, CO, Lynne Rienner Publishers: 9-22.

169

Byerlee, D. and D. Jewell (1997). "The technological foundation of the revolution." Chapter 9 in Africa's Emerging Maize Revolution. D. Byerlee and C. K. Eicher. Boulder, CO, Lynne Reinner Publishers: 127-143.

Caldwell, M. M., T. E. Dawson and J. H. Richards (1998). "Hydraulic lift: consequences of water efflux from the roots of plants." Oecologia 113: 151-161.

Cane, M. A., G. Eshel and R. W. Buckland (1994). "Forecasting Zimbabwean maize yield using eastern equatorial Pacific sea surface temperature." Nature 370: 204-205.

Cannavo, P., J. Sansoulet, J.-M. Harmand, P. Siles, E. Dreyer and P. Vaast (2011). "Agroforestry associating coffee and Inga densiflora results in complementarity for water uptake and decreases deep drainage in Costa Rica." Agriculture, Ecosystems & Environment 140: 1-13.

Cannell, M. G. R., M. van Noordwijk and C. K. Ong (1996). "The central agroforestry hypothesis: the trees must acquire resources that the crop would not otherwise acquire." Agroforestry Systems 33: 1-5.

Carter, M. S., P. Ambus, K. R. Albert, K. S. Larsen, M. Andersson, A. Priemé, . . . C. Beier (2011). "Effects of elevated atmospheric CO2, prolonged summer drought and temperature increase on N2O and CH4 fluxes in a temperate heathland." Soil Biology and Biochemistry 43: 1660-1670.

Cesaraccio, C., D. Spano, P. Duce and R. L. Snyder (2001). "An improved model for determining degree-day values from daily temperature data." International Journal of Biometeorology 45: 161-169.

CGIAR (2011). "MarkSim DSSAT weather file generator." Retrieved 20 January 2012, from http://gismap.ciat.cgiar.org/MarkSimGCM/.

Chintu, R., P. L. Mafongoya, T. S. Chirwa, E. Kuntashula, D. Phiri and J. Matibini (2004). "Propagation and management of Gliricidia sepium planted fallows in sub-humid eastern Zambia." Experimental Agriculture 40: 341-352.

Chirwa, P. W., F. K. Akinnifesi, G. Sileshi, S. Syampungani, F. K. Kalaba and O. C. Ajayi (2008). "Opportunity for conserving and utilizing agrobiodiversity through agroforestry in Southern Africa." Biodiversity (Ottawa) 9 (1-2): 45-48.

Chirwa, P. W., C. R. Black and C. K. Ong (2006a). "The potential of site specific measurements (SSM) and modelling in agroforestry research and development: the case of the WaNuLCAS model simulation in a Gliricidia + maize mixed cropping system in southern Malawi." International Precision Forestry Symposium, Stellenbosch, South Africa, Stellenbosch University.

Chirwa, P. W., C. R. Black, C. K. Ong and J. Maghembe (2006b). "Nitrogen dynamics in cropping systems in southern Malawi containing Gliricidia sepium, pigeonpea and maize." Agroforestry Systems 67 (1): 93-106.

170

Chirwa, P. W., C. R. Black, C. K. Ong and J. A. Maghembe (2003a). "Tree and crop productivity in gliricidia/maize/pigeonpea cropping systems in southern Malawi." Agroforestry Systems 59 (3): 265-277.

Chirwa, P. W., P. L. Mafongoya and R. Chintu (2003b). "Mixed planted-fallows using coppicing and non-coppicing tree species for degraded Acrisols in eastern Zambia." Agroforestry Systems 59: 243-251.

Chirwa, P. W., C. K. Ong, J. Maghembe and C. R. Black (2007). "Soil water dynamics in cropping systems containing Gliricidia sepium, pigeonpea and maize in southern Malawi." Agroforestry Systems 69: 29-43.

Christensen, J. H., B. Hewitson, A. Busuioc, A. Chen, X. Gao, I. Held, . . . P. Whetton (2007). "Regional Climate Projections." Chapter 11 in Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. S. Solomon, D. Qin, M. Manninget al. Cambridge, UK, Cambridge University Press: 847-940.

CIA (2012). "The World Factbook: Malawi." Retrieved 3 August 2012, from https://www.cia.gov/library/publications/the-world-factbook/geos/mi.html.

Cline, W. R. (2007). Global Warming and Agriculture: Impact Estimates by Country. Washington D.C.: Center for Global Development / Peterson Institute for International Economics.

Conroy, A. C., M. J. Blackie, A. Whiteside, J. C. Malewezi and J. D. Sachs (2006). Poverty, AIDS and Hunger: Breaking the Poverty Trap in Malawi. Hampshire, UK: Palgrave Macmillan.

Cook, B. G., B. C. Pengelly, S. D. Brown, J. L. Donnelly, D. A. Eagles, M. A. Franco, . . . R. Schultze-Kraft (2005). "Tropical Forages: an interactive selection tool." Retrieved 11/09/2007, from http://www.tropicalforages.info/.

Cooper, P. J. M., J. Dimes, K. P. C. Rao, B. Shapiro, B. Shiferaw and S. Twomlow (2008). "Coping better with current climating variability in the rain-fed farming systems of sub-Saharan Africa: an essential first step in adapting to future climate change?" Agriculture, Ecosystems & Environment 126: 24-35.

Crespo, O., S. Hachigonta and M. Tadross (2011). "Sensitivity of southern African maize yields to the definition of sowing dekad in a changing climate." Climatic Change 106: 267-283.

de Valpine, P. and J. Harte (2001). "Plant responses to experimental warming in a montane meadow." Ecology 82 (3): 637-648.

de Waal, A. and A. Whiteside (2003). "New variant famine: AIDS and food crisis in southern Africa." Lancet 362: 1234-1237.

171

Denning, G., P. Kabambe, P. Sanchez, A. Malik, R. Flor, R. Harawa, . . . J. Sachs (2009). "Input Subsidies to Improve Smallholder Maize Productivity in Malawi: Toward an African Green Revolution." PLoS Biology 7 (1): e1000023.

DeRose, L., E. Messer and S. Millman (1998). Who's Hungry? And How Do We Know? Food Shortage, Poverty, and Deprivation. Tokyo: United Nations University Press.

Devereux, S. (2002a). "The Malawi famine of 2002." IDS Bulletin 33 (4): 70-78.

Devereux, S. (2002b). Malawi’s Food Crisis of 2001/02: Causes, Consequences and Policy Lessons. Lilongwe, Malawi, ActionAid.

Dore, M. H. I. (2005). "Climate change and changes in global precipitation patterns: What do we know?" Environment International 31 (8): 1167-1181.

Dorward, A. and E. Chirwa (2011). "The Malawi Agricultural Input Subsidy Programme: 2005-6 to 2008-9." International Journal of Agricultural Sustainability 9 (1): In press.

Dorward, A. and J. Kydd (2004). "The Malawi 2002 food crisis: the rural development challenge." Journal of Modern African Studies 42 (3): 343-361.

Dube, O. P. and M. B. M. Sekhwela (2007). Community Coping Strategies in Semiarid Limpopo Basin Part of Botswana: Enhancing Adaptation Capacity to Climate Change. AIACC Working Paper. Washington, D.C., International START Secretariat: 38.

Easterling, W., P. Aggarwal, P. Batima, K. Brander, L. Erda, M. Howden, . . . F. Tubiello (2007). "Food, fibre and forest products." 5 in Climate Change 2007: Impacts, Adaptation and Vulnerability. Contribution of Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. M. L. Parry, O. F. Canziani, J. P. Palutikof, P. J. van der Linden and C. E. Hanson. Cambridge, UK, Cambridge University Press: 273-313.

Edmeades, G. O., M. Bänziger and J.-M. Ribaut (2000). "Maize improvement for drought-limited environments." Chapter 6 in Physiological Bases for Maize Improvement. M. E. Otegui and G. A. Slafer. Binghamton, NY, Haworth Press: 75-111.

Ellis, F., M. Kutengule and A. Nyasulu (2003). "Livelihoods and rural poverty reduction in Malawi." World Development 31 (9): 1495-1510.

English, N. B., J. F. Weltzin, A. Fravolini, L. Thomas and D. G. Williams (2005). "The influence of soil texture and vegetation on soil moisture under rainout shelters in a semi-desert grassland." Journal of Arid Environments 63: 324-343.

Eriksen, S. (2005). "The role of indigenous plants in household adaptation to climate change: the Kenyan experience." Chapter 23 in Climate Change and Africa. P. S. Low. Cambridge, UK, Cambridge University Press: 258-259.

Eriksen, S. H., K. Brown and P. M. Kelly (2005). "The dynamics of vulnerability: locating coping strategies in Kenya and Tanzania." The Geographical Journal 171 (4): 287-305.

172

Esmail, S. and M. Oelbermann (2011). "The impact of climate change on the growth of tropical agroforestry tree seedlings." Agroforestry Systems 83 (2): 235-244.

Ewel, J. J. (1999). "Natural systems as models for the design of sustainable systems of land use." Agroforestry Systems 45: 1-21.

Fauchereau, N., S. Trzaska, M. Rouault and Y. Richard (2003). "Rainfall variability and changes in southern Africa during the 20th century in the global warming context." Natural Hazards 29 (2): 139-154.

Fay, P. A., J. D. Carlisle, A. K. Knapp, J. M. Blair and S. L. Collins (2000). "Altering rainfall timing and quantity in a mesic grassland ecosystem: design and performance of rainfall manipulation shelters." Ecosystems 3: 308-319.

FEWSNET (2010). "Malawi Food Security Update, March 2010." Retrieved 28 March 2010, from www.fews.net/docs/Publications/Malawi_FSU_March_2010_final.pdf.

Foroughbakhch, R., M. A. Alvarado-Vazquez, J. L. Hernandez-Pinero, A. Rocha-Estrada, M. A. Guzman-Lucio and E. J. Trevino-Garza (2006). "Establishment, growth and biomass production of 10 tree woody species introduced for reforestation and ecological restoration in northeastern Mexico." Forest Ecology and Management 235 (1-3): 194-201.

Garbach, K., A. Martinez-Salinas, V. T. Eviner, M. Lubell and F. DeClerck (2011). "Live fence complexity and the conservation of avian biodiversity in a pasture-dominated agricultural ecosystem." 96th Annual Meeting of the Ecological Society of America (August 7-12, 2011), Austin, TX.

Garrity, D. P., F. K. Akinnifesi, O. C. Ajayi, G. W. Sileshi, J. G. Mowo, A. Kalinganire, . . . J. Bayala (2010). "Evergreen Agriculture: a robust approach to sustainable food security in Africa." Food Security 2 (3): 197-214.

Giampetro, M. (2004). "Integrated assessment of agroecosystems and multi-criteria analysis: basic definitions and challenges." Chapter 5 in Multi-scale integrated analysis of agroecosytems. Boca Raton, FL, CRC Press: 93-126.

Government of Malawi (2001). "Country presentation: Malawi." Third United Nations Conference on the Least Developed Countries, 14-20 May 2001, Brussels, Belgium, United Nations.

Government of Malawi (2006). Strategic Plan: 2007 - 2012. Lilongwe, Malawi, Ministry of Agriculture and Food Security: 56 pages.

Government of Malawi (2007). Monitoring and evaluation system for food security and nutrition policies of Malawi: 3rd M&E report of the FNSP working group. Lilongwe, Malawi, Ministry of Agriculture and Food Security: 147 pp.

Hachigonta, S. and C. J. C. Reason (2006). "Interannual variability in dry and wet spell characteristics over Zambia." Climate Research 32: 49-62.

173

Hannah, L. (2004). "Agroforestry and climate change-integrated conservation strategies." Chapter 20 in Agroforestry and Biodiversity Conservation in Tropical Landscapes. G. Schroth, G. A. B. da Fonseca, C. A. Harveyet al. Washington, DC, Island Press: 473-486.

Hanson, P. J. (2000). "Large-scale water manipulations." Chapter 23 in Methods in Ecosystem Science. O. E. Sala, R. B. Jackson, H. A. Mooney and R. W. Howarth. New York, Springer-Verlag: 341-352.

Harvey, C. A., N. I. J. Tucker and A. Estrada (2004). "Live fences, isolated trees, and windbreaks: tools for conserving biodiversity in fragmented tropical landscapes." Chapter 11 in Agroforestry and Biodiversity Conservation in Tropical Landscapes. G. Schroth, G. A. B. da Fonseca, C. A. Harvey et al. Washington, DC, Island Press: 261-289.

Hauser, S., C. Nolte and R. J. Carksy (2006). "What role can planted fallows play in the humid and sub-humid zone of West and Central Africa?" Nutrient Cycling in Agroecosystems 76: 297-318.

Hauser, S., L. Norgrove, B. Duguma and E. Asaah (2005). "Soil water regime under rotational fallow and alternating hedgerows on an Ultisol in southern Cameroon." Agroforestry Systems 64: 73-82.

Hayward, B. (2005). From the mountain to the tap: how land use and water management can work for the rural poor. Hayle, UK, DFID: 54 pp.

Hoeppner, S. S. and J. S. Dukes (2012). "Interactive responses of old-field plant growth and composition to warming and precipitation." Global Change Biology 18 (5): 1754-1768.

Hoerling, M., J. Hurrell, J. Eischeid and A. Phillips (2006). "Detection and attribution of twentieth-century northern and southern African rainfall change." Journal of Climate 19: 3989-4008.

Hooper, D. U., F. S. Chapin, J. J. Ewel, A. Hector, P. Inchausti, S. Lavorel, . . . D. A. Wardle (2005). "Effects of biodiversity on ecosystem functioning: a consensus of current knowledge." Ecological Monographs 75 (1): 3-35.

Huxley, P. (1999). Tropical Agroforestry. Oxford, UK: Blackwell Science.

Huxman, T. E., J. M. Cable, D. D. Ignace, J. A. Eilts, N. B. English, J. Weltzin and D. G. Williams (2004). "Response of net ecosystem gas exchange to a simulated precipitation pulse in a semi-arid grassland: the role of native versus non-native grasses and soil texture." Oecologia 141 (2): 295-305.

ICRAF (2006). "AgroforestryTree Database: A tree species reference and selection guide." Retrieved 11/09/2007, 2007, from http://www.worldagroforestrycentre.org/sea/Products/AFDbases/AF/index.asp.

Ikerra, S. T., J. A. Maghembe, P. C. Smithson and R. J. Buresh (2001). "Dry-season sesbania fallows and their influence on nitrogen availability and maize yields in Malawi." Agroforestry Systems 52: 13-21.

174

IPCC (2007). Summary for Policymakers, Climate Change 2007: Impacts, Adaptation and Vulnerability: Working Group II Contribution to the Intergovernmental Panel on Climate Change Fourth Assessment Report. Geneva, IPCC Secretariat: 23.

Jayne, T. S., D. Mather and E. Mghenyi (2006). Smallholder farming under increasingly difficult circumstances: policy and public investment priorities for Africa. MSU International Development Working Paper. East Lansing, Michigan, Michigan State University: 29.

Jentsch, A., J. Kreyling and C. Beierkuhnlein (2007). "A new generation of climate-change experiments: events, not trends." Frontiers in Ecology and the Environment 5 (7): 365-374.

Jones, P. G. and P. K. Thornton (2003). "The potential impacts of climate change on maize production in Africa and Latin America in 2055." Global Environmental Change 13: 51-59.

Jones, P. G., P. K. Thornton and J. Heinke (2009). Generating characteristic daily weather data using downscaled climate model data from the IPCC's Fourth Assessment. Nairobi, International Livestock Research Institute: 24 pp.

Kalame, F. B., R. Aidoo, J. Nkem, O. C. Ajayi, M. Kanninen, O. Luukkanen and M. Idinoba (2011). "Modified taungya system in Ghana: a win–win practice for forestry and adaptation to climate change?" Environmental Science & Policy 14: 519-530.

Kamanga, B. C. G., S. R. Waddington, M. J. Robertson and K. E. Giller (2010). "Risk analysis of maize-legume crop combinations with smallholder farmers varying in resource endowment in central Malawi." Experimental Agriculture 46 (1): 1-21.

Kaonga, M. L. and T. P. Bayliss-Smith (2009). "Carbon pools in tree biomass and the soil in improved fallows in eastern Zambia." Agroforestry Systems 76 (1): 37-51.

Kaonga, M. L. and T. P. Bayliss-Smith (2010). "Allometric models for estimation of aboveground carbon stocks in improved fallows in eastern Zambia." Agroforestry Systems 78 (3): 217-232.

Keil, A., M. Zeller and S. Franzel (2005). "Improved tree fallows in smallholder maize production in Zambia: do initial testers adopt the technology?" Agroforestry Systems 64: 225-236.

Kim, D.-G. (2012). "Estimation of net gain of soil carbon in a nitrogen-fixing tree and crop intercropping system in sub-Saharan Africa: results from re-examining a study." Agroforestry Systems In press: 10 pp.

Knapp, A. K., P. A. Fay, J. M. Blair, S. L. Collins, M. D. Smith, J. D. Carlisle, . . . J. K. McCarron (2002). "Rainfall variability, carbon cycling, and plant species diversity in a mesic grassland." Science 298 (5601): 2202-2205.

Kumar, B. M. and P. K. R. Nair (2004). "The enigma of tropical homegardens." Agroforestry Systems 61: 135-152.

175

Kurukulasuriya, P. and R. Mendelsohn (2006). Crop selection: adapting to climate change in Africa. CEEPA Discussion Paper No. 26: Centre for Environmental Economics and Policy in Africa, University of Pretoria.

Kurukulasuriya, P., R. Mendelsohn, R. Hassan, J. Benhin, T. Deressa, M. Diop, . . . A. Dinar (2006). "Will African agriculture survive climate change?" The World Bank Economic Review 20 (3): 367-388.

Kwesiga, F., F. K. Akinnifesi, P. L. Mafongoya, M. H. McDermott and A. Agumya (2003). "Agroforestry research and development in southern Africa during the 1990s: Review and challenges ahead." Agroforestry Systems 59: 173-186.

Kwesiga, F. R., S. Franzel, F. Place, D. Phiri and C. P. Simwanza (1999). "Sesbania sesban improved fallows in eastern Zambia: Their inception, development and farmer enthusiasm." Agroforestry Systems 47: 49-66.

Lal, R., N. Uphoff, B. A. Stewart and D. O. Hansen, Eds. (2005). Climate Change and Global Food Security. Boca Raton, FL, CRC Press.

Leach, M. (1994). Rainforest Relations: Gender and Resource Use Among the Mende of Gola, Sierra Leone. Washington, D.C.: Smithsonian Institution Press.

Lin, B. B. (2007). "Agroforestry management as an adaptive strategy against potential microclimate extremes in coffee agriculture." Agricultural and Forest Meteorology 144: 85-94.

Lin, B. B. (2010). "The role of agroforestry in reducing water loss through soil evaporation and crop transpiration in coffee agroecosystems." Agricultural and Forest Meteorology 150: 510-518.

Lin, B. B. (2011). "Resilience in agriculture through crop diversification: adaptive management for environmental change." BioScience 61 (3): 183-193.

Lin, B. B., I. Perfecto and J. Vandermeer (2008). "Synergies between agricultural intensification and climate change could create surprising vulnerabilities for crops." BioScience 58 (9): 847-854.

Lin, B. B. and P. L. Richards (2007). "Relationships of random roughness and soil depression storage on coffee farms of varying shade levels." Ecological Society of America Annual Meeting (August 5-10, 2007), San Jose, CA.

Lindsey, A. A. and J. E. Newman (1956). "Use of official weather data in spring time - temperature analysis of an Indiana phenological record." Ecology 37 (4): 812-823.

Lobell, D. B., M. Bänziger, C. Magorokosho and B. Vivek (2011). "Nonlinear heat effects on African maize as evidenced by historical yield trials." Nature Climate Change 1: 42-45.

Lobell, D. B., M. B. Burke, C. Tebaldi, M. D. Mastrandrea, W. P. Falcon and R. L. Naylor (2008). "Prioritizing climate change adaptation needs for food security in 2030." Science 319: 607-610.

176

Love, B., E. Bork and D. Spaner (2009). "Tree seedling establishment in living fences: a low-cost agroforestry management practice for the tropics." Agroforestry Systems 77 (1): 1-8.

Ludwig, F., T. E. Dawson, H. Kroon, F. Berendse and H. H. T. Prins (2003). "Hydraulic lift in Acacia tortilis trees on an East African savanna." Oecologia 134: 293-300.

Ludwig, F., T. E. Dawson, H. H. T. Prins, F. Berendse and H. de Kroon (2004). "Below-ground competition between trees and grasses may overwhelm the facilitative effects of hydraulic lift." Ecology Letters 7: 623-631.

Mafongoya, P., K. E. Giller, D. Odee, S. Gathumbi, S. K. Ndufa and S. M. Sitompul (2004). "Benefiting from N2-fixation and manging rhizobia." Chapter 13 in Below-ground Interactions in Tropical Agroecosystems: Concepts and Models with Multiple Plant Components. M. van Noordwijk, G. Cadisch and C. K. Ong. Wallingford, UK, CAB International: 227-242.

Mafongoya, P. L. (2006). "Appropriate technologies to replenish soil fertility in southern Africa." Nutrient Cycling in Agroecosystems 76: 137-151.

Mafongoya, P. L., R. Chintu, T. S. Chirwa, J. Matibini and S. Chikale (2003). "Tephrosia species and provenances for improved fallows in southern Africa." Agroforestry Systems 59: 279-288.

Mafongoya, P. L. and E. Kuntashula (2005). "Participatory evaluation of Tephrosia species and provenances for soil fertility improvement and other uses using farmer criteria in eastern Zambia." Experimental Agriculture 41 (1): 69-80.

Mafongoya, P. L., P. K. R. Nair and B. H. Dzowela (1998). "Mineralization of nitrogen from decomposing leaves of multipurpose trees as affected by their chemical composition." Biology and Fertility of Soils 27 (2): 143-148.

Maiti, R. K. and P. Wesche-Ebeling (1998). "Influences of the environment." Chapter 6 in Maize Science. R. K. Maiti and P. Wesche-Ebeling. Enfield, NH, Science Publishers: 233-294.

Makumba, W., F. K. Akinnifesi, B. Janssen and O. Oenema (2007). "Long-term impact of a gliricidia-maize intercropping system on carbon sequestration in southern Malawi." Agriculture, Ecosystems & Environment 118 (1-4): 237-243.

Makumba, W., F. K. Akinnifesi and B. H. Janssen (2009). "Spatial rooting patterns of gliricidia, pigeon pea and maize intercrops and effect on profile soil N and P distribution in southern Malawi." African Journal of Agricultural Research 4 (4): 278-288.

Makumba, W., B. Janssen, O. Oenema and F. K. Akinnifesi (2005). "Influence of time of application on the performance of gliricidia prunings as a source of N for maize." Experimental Agriculture 42 (1): 51-63.

Makumba, W., B. Janssen, O. Oenema, F. K. Akinnifesi, D. Mweta and F. Kwesiga (2006). "The long-term effects of a gliricidia-maize intercropping system in Southern Malawi, on gliricidia and maize yields, and soil properties." Agriculture, Ecosystems & Environment 116 (1-2): 85-92.

177

Makundi, W. R. and J. A. Sathaye (2004). "GHG mitigation potential and cost in tropical forestry - relative role for agroforestry." Enviroment, Development, and Sustainability 6: 235-260.

Martin, L. J., B. Blossey and E. Ellis (2012). "Mapping where ecologists work: biases in the global distribution of terrestrial ecological observations." Frontiers in Ecology and the Environment 10 (4): 195-201.

Mazvimavi, D. (2010). "Investigating changes over time of annual rainfall in Zimbabwe." Hydrology and Earth System Sciences 14: 2671-2679.

McCann, J. C. (2005). Maize and Grace: Africa's Encounter with a New World Crop, 1500-2000. Harvard, MA: Cambridge University Press.

Meehl, G. A., T. F. Stocker, W. D. Collins, P. Friedlingstein, A. T. Gaye, J. M. Gregory, . . . Z.-C. Zhao (2007). "Global Climate Projections." Chapter 10 in Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. S. Solomon, D. Qin, M. Manninget al. Cambridge, UK, Cambridge University Press.

Mercer, D. E. (2004). "Adoption of agroforestry innovations in the tropics: a review." Agroforestry Systems 61-2 (1): 311-328.

Miracle, M. (1966). Maize in Tropical Africa. Madison, WI: University of Wisconsin Press.

Miranda, J. D., C. Armas, F. M. Padilla and F. I. Pugnaire (2011). "Climatic change and rainfall patterns: Effects on semi-arid plant communities of the Iberian Southeast." Journal of Arid Environments 75 (12): 1302-1309.

Mkandawire, T. (1999). Agriculture, Employment and Poverty in Malawi (ILO/SAMAT Policy Paper No. 9). Harare, Zimbabwe, ILO/SAMAT: 34 pp.

Mpaka, C. (2010). "Malawi: Green Belt Initiative Taking Shape." Inter Press Service News Agency. Retrieved 3 August 2012, from http://www.ipsnews.net/2010/01/malawi-green-belt-initiative-taking-shape/.

Muchow, R. C. (2000). "Improving maize grain yield potential in the tropics." Chapter 4 in Physiological Bases for Maize Improvement. M. E. Otegui and G. A. Slafer. Binghamton, NY, Haworth Press: 47-58.

Mweta, D. E., F. K. Akinnifesi, J. D. K. Saka, W. Makumba and N. Chokotho (2007a). "Green manure from prunings and mineral fertilizer affect phosphorus adsorption and uptake by maize crop in a gliricidia-maize intercropping." Scientific Research and Essay 2 (10): 446-453.

Mweta, D. E., F. K. Akinnifesi, J. D. K. Saka, W. Makumba and N. Chokotho (2007b). "Use of pruning and mineral fertilizer affects soil phosphorus availability and fractionation in a gliricidia/maize intercropping system." African Journal of Agricultural Research 2 (10): 521-527.

178

Nakicenovic, N., J. Alcamo, G. Davis, B. de Vries, J. Fenhann, S. Gaffin, . . . Z. Dadi (2000). IPCC Special Report on Emissions Scenarios. Cambridge, UK: Cambridge University Press.

Nam, N. H., Y. S. Chauhan and C. Johansen (2001). "Effect of timing of drought stress on growth and grain yield of extra-short-duration pigeonpea lines." Journal of Agricultural Science 136: 179-189.

National Statistical Office (2011). Malawi Demographic and Health Survey 2010: Preliminary Report. Zomba, Malawi, National Statistical Office of Malawi and ICF Macro: 36 pages.

Nepstad, D. C., I. M. Tohve, D. Ray, P. Moutinho and G. Cardinot (2007). "Mortality of large trees and lianas following experimental drought in an Amazon forest." Ecology 88 (9): 2259-2269.

Neufeldt, H., I. K. Dawson, E. Luedeling, O. C. Ajayi, T. Beedy, A. Gebrekirstos, . . . J. C. Weber (2012). Climate change vulnerability of agroforestry. ICRAF Working Paper No. 143. Nairobi, World Agroforestry Centre.

Newman, K. (2002). Birds of Southern Afirca. 8th edition. Cape Town: Struik Publishers.

Ngugi, D. (2002). Agroforestry in Malawi and Zambia: summary report of a CTA/MAFE study visit. Waginengen, The Netherlands, Technical Centre for Agricultural and Rural Cooperation (CTA): 32.

Nguyen, Q., M. Hoang, I. Öborn and M. van Noordwijk (2012). "Multipurpose agroforestry as a climate change resiliency option for farmers: an example of local adaptation in Vietnam." Climatic Change In press: 17 pp.

NOAA Climate Prediction Center (2011). "Monthly atmospheric and sea surface temperature indices." Retrieved 22 January 2012, from http://www.cpc.ncep.noaa.gov/data/indices/.

O'Brien, K. and C. Vogel, Eds. (2003). Coping with climate variability: the use of seasonal climate forecasts in southern Africa. Hampshire, UK, Ashgate.

Oduol, P. and E. Akunda (1988). "Vegetative propagation of Sesbania sesban by cuttings." Agroforestry Systems 6 (1): 283-288.

Ong, C. K. and R. R. B. Leakey (1999). "Why tree-crop interactions in agroforestry appear at odds with tree-grass interactions in tropical savannahs." Agroforestry Systems 45: 109-129.

Ong, C. K., J. Wilson, J. D. Deans, J. Mulayta, T. Raussen and N. Wajja-Musukwe (2002). "Tree-crop interactions: manipulation of water use and root function." Agricultural Water Management 53: 171-186.

Osborne, T. M. (2010). Carbon capital: The political ecology of carbon forestry and development in Chiapas, Mexico (Ph.D. thesis). Berkeley, CA, Energy and Resources Group, University of California, Berkeley.

179

Osborne, T. M. (2011). "Carbon forestry and agrarian change: access and land control in a Mexican rainforest." Journal of Peasant Studies 38 (4): 859-883.

Ottera, O. H., M. Bentsen, I. Bethke and N. G. Kvamsto (2009). "Simulated pre-industrial climate in Bergen Climate Model (version 2): model description and large-scale circulation features." Geoscientific Model Development 2 (2): 197-212.

Owuor, B. O., S. Gudu and A. Niang (2001). "Direct seeding of Sesbania sesban for green manure in agroforestry systems - a short communication." Agroforestry Systems 52 (1): 23-25.

Palm, C. A., S. M. Smukler, C. C. Sullivan, P. K. Mutuo, G. I. Nyadzi and M. G. Walsh (2010). "Identifying potential synergies and trade-offs for meeting food security and climate change objectives in sub-Saharan Africa." Proceedings of the National Academy of Sciences of the United States of America 107 (46): 19661-19666.

Parton, W. J. and J. A. Logan (1981). "A model for diurnal variation in soil and air temperature." Agricultural Meteorology 23: 205-216.

Pate, J. S. and T. E. Dawson (1999). "Assessing the performance of woody plants in uptake and utilisation of carbon, water and nutrients: Implications for designing agricultural mimic systems." Agroforestry Systems 45 (1-3): 245-275.

Paulsen, G. M. (1994). "High temperature responses of crop plants." Chapter 14A in Physiology and Determination of Crop Yield. K. J. Boote, J. M. Bennett, T. R. Sinclair and G. M. Paulsen. Madison, WI, American Society of Agronomy, Crop Science Society of America, Soil Science Society of America: 365-398.

Philpott, S. M., B. B. Lin, S. Jha and S. J. Brines (2008). "A multi-scale assessment of hurricane impacts on agricultural landscapes based on land use and topographic features." Agriculture, Ecosystems & Environment 128 (1-2): 12-20.

Phiri, E., H. Verplancke, F. Kwesiga and P. Mafongoya (2003). "Water balance and maize yield following improved sesbania fallow in eastern Zambia." Agroforestry Systems 59: 197-205.

PRB (2012). "World Population Data Sheet 2012." Retrieved 3 August 2012, from http://www.prb.org/Publications/Datasheets/2012/world-population-data-sheet.aspx.

Quinion, A., P. W. Chirwa, F. K. Akinnifesi and O. C. Ajayi (2010). "Do agroforestry technologies improve the livelihoods of the resource poor farmers? Evidence from Kasungu and Machinga districts of Malawi." Agroforestry Systems 80 (3): 457-465.

Rao, K. P. C., L. V. Verchot and J. Laarman (2007). "Adaptation to climate change through sustainable management and development of agroforestry systems." Journal of SAT Agricultural Research 4 (1): 1-30.

Reubens, B., J. Poesen, J. Nyssen, Y. Leduc, A. Z. Abraha, S. Tewoldeberhan, . . . B. Muys (2009). "Establishment and management of woody seedlings in gullies in a semi-arid environment (Tigray, Ethiopia)." Plant and Soil 324 (1-2): 131-156.

180

Rocheleau, D., F. Weber and A. Field-Juma (1988). Agroforestry in Dryland Africa. Nairobi: International Council for Research in Agroforestry.

Roe, E., L. Huntsinger and K. Labnow (1998). "High reliability pastoralism." Journal of Arid Environments 39: 39-55.

Roltsch, W. J., F. G. Zalom, A. J. Strawn, J. F. Strand and M. J. Pitcairn (1999). "Evaluation of several degree-day estimation methods in California climates." International Journal of Biometeorology 42 (4): 169-176.

Saka, A. R., W. T. Bunderson, O. A. Itimu, H. S. K. Phombeya and Y. Mbekeani (1994). "The effects of Acacia albida on soils and maize grain yields under smallholder farm conditions in Malawi." Forest Ecology and Management 64 (2-3): 217-230.

Sanchez, P. A. (1995). "Science in agroforestry." Agroforestry Systems 30: 5-55.

Sanchez, P. A. (1999). "Improved fallows come of age in the tropics." Agroforestry Systems 47: 3-12.

Sanchez, P. A. (2005). "Reducing hunger in tropical Africa while coping with climate change." Chapter 1 in Climate Change and Global Food Security. R. Lal, N. Uphoff, B. A. Stewart and D. O. Hansen. Boca Raton, FL, CRC Press: 3-19.

Sauer, J. and H. Tchale (2009). "The Economics of Soil Fertility Management in Malawi." Review of Agricultural Economics 31 (3): 535-560.

Sayer, J. A. and B. M. Campbell (2004). The Science of Sustainable Development: Local Livelihoods and the Global Environment. Cambridge, UK: Cambridge University Press.

Schlenker, W. and D. B. Lobell (2010). "Robust negative impacts of climate change on African agriculture." Environmental Research Letters 5 (2010): 14010.

Schroth, G., G. A. B. da Fonseca, C. A. Harvey, C. Gascon, H. L. Vasconcelos and A.-M. N. Izac, Eds. (2004). Agroforestry and Biodiversity Conservation in Tropical Landscapes. Washington, DC, Island Press.

Schwendenmann, L., E. Veldkamp, G. Moser, D. Hölscher, M. Köhler, Y. Clough, . . . O. van Straaten (2010). "Effects of an experimental drought on the functioning of a cacao agroforestry system, Sulawesi, Indonesia." Global Change Biology 16: 1515-1530.

Sekiya, N. and K. Yano (2004). "Do pigeon pea and sesbania supply groundwater to intercropped maize through hydraulic lift? - Hydrogen stable isotope investigation of xylem waters." Field Crops Research 86: 167-173.

Sen, A. (1999). "Famines and other crises." Chapter 7 in Development as Freedom. New York, Anchor Books: 160-188.

Shaxson, L. and L. W. Tauer (1992). "Intercropping and diversity: an economic analysis of cropping patterns on smallholder farms in Malawi." Experimental Agriculture 28: 211-228.

181

Shen, K. P. and J. Harte (2000). "Ecosystem climate manipulations." Chapter 24 in Methods in Ecosystem Science. O. E. Sala, R. B. Jackson, H. A. Mooney and R. W. Howarth. New York, Springer-Verlag: 341-352.

Shongwe, M. E., G. J. van Oldenborgh and B. J. J. M. van den Hurk (2009). "Projected changes in mean and extreme precipitation in Africa under global warming. Part I: Southern Africa." Journal of Climate 22 (13): 3819-3837.

Sileshi, G., F. K. Akinnifesi, O. C. Ajayi, S. Chakeredza, M. Kaonga and P. Matakala. (2007). "Contribution of agroforestry to ecosystem services in the miombo eco-region of eastern and southern Africa." African Journal of Environmental Science and Technology 1 (4): 68-80.

Sileshi, G., F. K. Akinnifesi, O. C. Ajayi and F. Place (2009). Evidence for impact of green fertilizers on maize production in sub-Saharan Africa: a meta-analysis. ICRAF Occasional Paper No. 10. Nairobi, World Agroforestry Centre: 36 pp.

Sileshi, G., F. K. Akinnifesi, L. K. Debusho, T. Beedy, O. C. Ajayi and S. Mng'omba (2010). "Variation in maize yield gaps with plant nutrient inputs, soil type and climate across sub-Saharan Africa." Field Crops Research 116 (1-2): 1-13.

Sileshi, G. and P. Mafongoya (2007). "Quantity and quality of organic inputs from coppicing leguminous trees influence abundance of soil macrofauna in maize crops in eastern Zambia." Biology and Fertility of Soils 43: 333-340.

Sileshi, G. W., F. K. Akinnifesi, O. C. Ajayi and B. Muys (2011). "Integration of legume trees in maize-based cropping systems improves rain use efficiency and yield stability under rain-fed agriculture." Agricultural Water Management 98 (9): 1364-1372.

Sileshi, G. W., E. Kuntashula, P. Matakala and P. O. Nkunika (2008). "Farmers' perceptions of tree mortality, pests and pest management practices in agroforestry in Malawi, Mozambique and Zambia." Agroforestry Systems 72 (2): 87-101.

Sirrine, D., C. Shennan, S. Snapp, G. Kanyama-Phiri, B. Kamanga and J. R. Sirrine (2010). "Improving recommendations resulting from on-farm research: agroforestry, risk, profitability and vulnerability in southern Malawi." International Journal of Agricultural Sustainability 8 (4): 290-304.

Smale, M. and P. W. Heisey (1997). "Maize technology and productvity in Malawi." Chapter 5 in Africa's Emerging Maize Revolution. D. Byerlee and C. K. Eicher. Boulder, CO, Lynne Reinner Publishers: 63-79.

Smith, M., S. S. O. Burgess, D. Suprayogo, B. Lusiana and Widianto (2004). "Uptake, partitioning, and redistribution of water by roots in mixed-species agroecosystems." Chapter 9 in Below-ground Interactions in Tropical Agroecosystems: Concepts and Models with Multiple Plant Components. M. van Noordwijk, G. Cadisch and C. K. Ong. Wallingford, UK, CAB International: 157-170.

182

Snapp, S. S., M. J. Blackie, R. A. Gilbert, R. Bezner-Kerr and G. Y. Kanyama-Phiri (2010). "Biodiversity can support a greener revolution in Africa." Proceedings of the National Academy of Sciences of the United States of America 107 (48): 20840-20845.

Snapp, S. S., D. D. Rohrbach, F. Simtowe and H. A. Freeman (2002). "Sustainable soil management options for Malawi: can smallholder farmers grow more legumes?" Agriculture Ecosystems & Environment 91 (1-3): 159-174.

Snyder, R. L. (1985). "Hand calculating degree-days." Agricultural and Forest Meteorology 35: 353-358.

Sokal, R. R. and F. J. Rohlf (1995). Biometry. 3rd edition. New York: W.H. Freeman.

Southworth, J., J. C. Randolph, M. Habeck, O. C. Doering, R. A. Pfeifer, D. G. Rao and J. J. Johnston (2000). "Consequences of future climate change and changing climate variability on maize yields in the midwestern United States." Agriculture, Ecosystems & Environment 82 (1-3): 139-158.

Steffan-Dewenter, I., M. Kessler, J. Barkmann, M. M. Bos, D. Buchori, S. Erasmi, . . . T. Tscharntke (2007). "Tradeoffs between income, biodiversity, and ecosystem functioning during tropical rainforest conversion and agroforestry intensification." Proceedings of the National Academy of Sciences of the United States of America 104 (12): 4973-4978.

Stern, R. D. and P. J. M. Cooper (2011). "Assessing climate risk and climate change using rainfall data - a case study from Zambia." Experimental Agriculture 47 (2): 241-266.

Stige, L. C., J. Stave, K.-S. Chan, L. Cianelli, N. Pettorelli, M. Glantz, . . . N. Stenseth (2006). "The effect of climate variation on agro-pastoral production in Africa." Proceedings of the National Academy of Sciences of the United States of America 103 (9): 3049-3053.

Stigter, K., Ed. (2010a). Applied Agrometeorology. Berlin, Springer.

Stigter, K. (2010b). "Combating Disasters by Using Agroforestry." Chapter III.5.1 in Applied Agrometeorology. K. Stigter. Berlin, Springer: 653-656.

Suttle, K. B., M. A. Thomsen and M. E. Power (2007). "Species interactions reverse grassland responses to changing climate." Science 315: 640-643.

Syampungani, S., P. W. Chirwa, F. K. Akinnifesi and O. C. Ajayi (2010). "The potential of using agroforestry as a win-win solution to climate change mitigation and adaptation and meeting food security challenges in southern Africa." Agricultural Journal 5 (2): 80-88.

Tadross, M., P. Suarez, A. Lotsch, S. Hachigonta, M. Mdoka, L. Unganai, . . . M. Muchinda (2007). Changes in growing-season rainfall characteristics and downscaled scenarios of change over southern Africa: implications for growing maize. IPCC Regional Expert Meeting on Regional Impacts, Adaptation, Vulnerability, and Mitigation. Nadi, Fiji: 193-204.

183

Tadross, M., P. Suarez, A. Lotsch, S. Hachigonta, M. Mdoka, L. Unganai, . . . M. Muchinda (2009). "Growing-season rainfall and scenarios of future change in southeast Africa: implications for cultivating maize." Climate Research 40: 147-169.

Tadross, M. A., B. C. Hewitson and M. T. Usman (2005). "The interannual variability of the onset of the maize growing season over South Africa and Zimbabwe." Journal of Climate 18: 3356-3372.

Tengö, M. and K. Belfrage (2004). "Local Management Practices for Dealing with Change and Uncertainty: A Cross-scale Comparison of Cases in Sweden and Tanzania." Ecology and Society 9 (3): Article 4.

Thomas, R. B., S. J. Van Bloem and W. H. Schlesinger (2007). "Climate change and symbiotic nitrogen fixation in agroecosystems." Chapter 4 in Agroecosystems in a Changing Climate. P. C. D. Newton, R. A. Carran, G. R. Edwards and P. A. Niklaus. Boca Raton, FL, CRC Press: 85-116.

Thorlakson, T. (2011). Reducing subsistence farmers' vulnerability to climate change: the potential contributions of agroforestry in western Kenya. Occasional Paper 16, World Agroforestry Centre, Nairobi: 62 pp.

Thornton, P. K., P. G. Jones, P. J. Ericksen and A. J. Challinor (2011). "Agriculture and food systems in sub-Saharan Africa in a 4ºC+ world." Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences 369 (1934): 117-136.

Tischler, C. R., H. W. Polley, H. B. Johnson and R. E. Pennington (2000). "Seedling response to elevated CO2 in five epigeal species." International Journal of Plant Science 161 (5): 779-783.

Tissue, D. T., J. P. Megonigal and R. B. Thomas (1997). "Nitrogenase activity and N2 fixation are stimulated by elevated CO2 in a tropical N2-fixing tree." Oecologia 109 (1): 28-33.

Tollenaar, M. (1999). "Duration of the grain-filling period in maize is not affected by photoperiod and incident PPFD during the vegetative phase." Field Crops Research 62 (1): 15-21.

Tollenaar, M. and L. M. Dwyer (1999). "Physiology of maize." Pages 169-204 in Crop Yield: Physiology and Processes. D. L. Smith and C. Hamel. Berlin, Springer-Verlag.

Tomich, T. P., S. Brodt, H. Ferris, R. Galt, W. R. Horwath, E. Kebreab, . . . L. Yang (2011). "Agroecology: A Review from a Global-Change Perspective." Annual Review of Environment and Resources 36 (1): 193-222.

Trenberth, K. E., P. D. Jones, P. Ambenje, R. Bojariu, D. Easterling, A. Klein Tank, . . . P. Zhai (2007). "Observations: Surface and Atmospheric Climate Change." Chapter 3 in Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. S. Solomon, D. Qin, M. Manninget al. Cambridge, UK, Cambridge University Press.

184

UC IPM (2003). "About degree-days." Retrieved 14 January 2012, from http://www.ipm.ucdavis.edu/WEATHER/ddconcepts.html.

UNDP (2012). "International Human Development Indicators." Retrieved 3 August 2012, from http://hdr.undp.org/en/data/trends/.

van Noordwijk, M., M. H. Hoang, H. Neufeldt, I. Öborn and T. Yatich, Eds. (2011). How trees and people can co-adapt to climate change: reducing vulnerability through multifunctional agroforestry landscapes. Nairobi, World Agroforestry Center.

van Noordwijk, M. and B. Lusiana (1999). "WaNuLCAS, a model of water, nutrient and light capture in agroforestry systems." Agroforestry Systems 43: 217-242.

van Noordwijk, M. and C. K. Ong (1999). "Can the ecosystem mimic hypotheses be applied to farms in African savannahs?" Agroforestry Systems 45 (1-3): 131-158.

Vandenbeldt, R. J. and J. H. Williams (1992). "The effect of soil surface temperature on the growth of millet in relation to the effect of Faidherbia albida trees." Agricultural and Forest Meteorology 60 (1-2): 93-100.

Vanlauwe, B., K. Aihou, S. Aman, B. Tossah, J. Diels, N. Sanginga and R. Merckx (2001). "Leaf quality of selected hedgerow species at two canopy ages in the derived savanna zone of West Africa." Agroforestry Systems 53 (1): 21-30.

Vanlauwe, B. and K. E. Giller (2006). "Popular myths around soil fertility management in sub-Saharan Africa." Agriculture, Ecosystems & Environment 116 (1-2): 34-46.

Verchot, L. V., M. van Noordwijk, S. Kandji, T. Tomich, C. Ong, A. Albrecht, . . . C. Palm (2007). "Climate change: linking adaptation and mitigation through agroforestry." Mitigation and Adaptation Strategies for Global Change 12 (5): 901-918.

Vogel, C. and K. O'Brien (2006). "Who can eat information? Examining the effectiveness of seasonal climate forecasts and regional climate-risk management strategies." Climate Research 33: 111-122.

Wann, M., D. Yen and H. J. Gold (1985). "Evaluation and calibration of three models for daily cycle of air temperature." Agricultural and Forest Meteorology 34 (2-3): 121-128.

West, A. G., T. E. Dawson, T. L. Aston, W. J. Bond, G. F. Midgley and E. C. February (2012). "Diverse functional responses to drought in a Mediterranean-type shrubland in South Africa." New Phytologist 195 (2): 396-407.

Wilks, D. S. (1992). "Adapting stochastic weather generation algorithms for climate change studies." Climatic Change 22: 67-84.

Wise, R. and O. Cacho (2005). "A bioeconomic analysis of carbon sequestration in farm forestry: a simulation study of Gliricidia sepium." Agroforestry Systems 64: 237-250.

185

Wojtkowski, P. (1998). The Theory and Practice of Agroforestry Design. Enfield, NH: Science Publishers, Inc.

Wolter, K. (2012). "Multivariate ENSO index." Retrieved 17 January 2012, from http://www.esrl.noaa.gov/psd/enso/mei/table.html.

World Bank (2007). World Development Report 2008: Agriculture for Development. Washington D.C., The World Bank: 365 pp.

Yahdjian, L. and O. E. Sala (2002). "A rainout shelter design for intercepting different amounts of rainfall." Oecologia 133: 95-101.

Young, A. (1997). Agroforestry for Soil Management. 2nd ed. Wallingford, UK: CAB International.

Young, A. (2005). "Poverty, hunger, and population policy: linking Cairo with Johannesburg." The Geographical Journal 171 (1): 83-95.

Zahawi, R. A. (2005). "Establishment and growth of living fence species: An overlooked tool for the restoration of degraded areas in the tropics." Restoration Ecology 13 (1): 92-102.

Zahawi, R. A. and K. D. Holl (2009). "Comparing the performance of tree stakes and seedlings to restore abandoned tropical pastures." Restoration Ecology 17 (6): 854-864.

Zavaleta, E. S., M. R. Shaw, N. R. Chiariello, B. D. Thomas, E. E. Cleland, C. B. Field and H. A. Mooney (2003). "Grassland responses to three years of elevated temperature, CO(2), precipitation, and N deposition." Ecological Monographs 73 (4): 585-604.

Zhu, J.-j., X.-p. Wang, C.-x. Sun, X.-m. Zhu, M. Li, G.-d. Zhang, . . . Z.-l. Wang (2011). "Mapping of QTL Associated with Drought Tolerance in a Semi-Automobile Rain Shelter in Maize (Zea mays L.)." Agricultural Sciences in China 10 (7): 987-996.

Ziervogel, G., A. Nyong, B. Osman, C. Conde, S. Cortés and T. Downing (2006). Climate Variability and Change: Implications for Household Food Security. AIACC Working Paper. Washington, D.C., International START Secretariat: 32.

Ziska, L. H. and G. B. Runion (2007). "Future weed, pest, and disease problems for plants." Chapter 11 in Agroecosystems in a Changing Climate. P. C. D. Newton, R. A. Carran, G. R. Edwards and P. A. Niklaus. Boca Raton, FL, CRC Press: 261-287.

Drought resilience of maize-legume agroforestry systems in ...

Documents

Transcript of Drought resilience of maize-legume agroforestry systems in ...