Acc

epte

d A

rtic

le

This article has been accepted for publication and undergone full peer review but has not

been through the copyediting, typesetting, pagination and proofreading process, which may

lead to differences between this version and the Version of Record. Please cite this article as

doi: 10.1111/gcb.12909

This article is protected by copyright. All rights reserved.

Received Date : 14-Nov-2014

Accepted Date : 16-Feb-2015

Article type : Primary Research Articles

Twenty-five years of change in southern African passerine diversity: non-

climatic factors of change

Guillaume Péron (1)*

Res Altwegg (1,2)

(1) Department of Statistical Sciences, University of Cape Town, Rondebosch 7701, Cape

Town, South Africa

(2) African Climate and Development Initiative, University of Cape Town, Rondebosch

7701, South Africa

* corresponding author: peron_guillaume@yahoo.fr

Running title: Mapping change in southern African passerine diversity

Key-words: Biodiversity; Occupancy models; Citizen Science; Conservation planning;

Species richness; Bush Encroachment; Pesticide Accumulation; Climate Change; South

Africa

Abstract

We analysed more than 25 years of change in passerine bird distribution in South Africa,

Swaziland and Lesotho, in order to show that species distributions can be influenced by

processes that are at least in part independent of the local strength and direction of climate

change: land use and ecological succession. We used occupancy models that separate

species’ detection from species’ occupancy probability, fitted to citizen science data from

both phases of the Southern African Bird Atlas Project (1987-1996 and 2007-2013).

Temporal trends in species’ occupancy probability were interpreted in terms of local

Acc

epte

d A

rtic

le

This article is protected by copyright. All rights reserved.

extinction/colonisation and temporal trends in detection probability were interpreted in terms

of change in abundance. We found for the first time at this scale that, as predicted in the

context of bush encroachment, closed savanna specialists increased where open savanna

specialists decreased. In addition, the trend in the abundance of species a priori thought to be

favoured by agricultural conversion was negatively correlated with human population

density, which is in line with hypotheses explaining the decline in farmland birds in the

Northern Hemisphere. In addition to climate, vegetation cover and the intensity and time

since agricultural conversion constitute important predictors of biodiversity changes in the

region. Their inclusion will improve the reliability of predictive models of species

distribution.

Introduction

Human encroachment upon natural landscapes imposes huge pressure on ecosystems

(Vitousek, 1997; Foley et al., 2005). Much effort is therefore currently going into predicting

the consequences of anthropogenic global change on biodiversity and particularly on species

distributions. Here we focus on factors of variation with direct effect on animal species

distribution but limited or indirect links to climate: land use and ecological succession.

Ecological succession can follow land use change (e.g., farmland abandonment; Sirami et al.,

2006), or happen in conjunction with climate change (e.g., if climate change induces an

ecological regime shift; Scheffer and Carpenter, 2003), and is known to drastically modify

vegetation cover at the continental scale and within a few decades (Chazdon et al., 2007).

Land use change can be even faster and affect large tracts of land over mere months or years

(Achard et al., 2002). By contrast, we may expect climate change and its effect on species’

distributions to be more gradual (see however Pearson & Dawson, 2003; Thuiller et al.,

Acc

epte

d A

rtic

le

This article is protected by copyright. All rights reserved.

2004). The role of land use and ecological succession may, thereby, explain discrepancies

between range shifts and climate change, such as the decoupling between the poleward shift

in the wintering range of North American birds and the concomitant increase in winter

temperatures (La Sorte & Jetz, 2012).

We analysed more than 25 years of change in passerine bird distributions in South

Africa and neighboring countries. We confronted these observed changes with three

theoretical predictions. First, due to a combination of change in farming practices and

atmospheric CO2 concentration, woody vegetation has been encroaching upon grasslands in

the region (Wigley et al., 2010; Buitenwerf et al., 2012; Parr et al., 2012). This phenomenon

is occurring both within the Grassland Biome (which is becoming more heterogeneous) and

at the transition with the Savanna Biome (which is shifting towards the centre of the

Grassland Biome). We predicted (prediction 1) that bird species adapted to mosaic

landscapes of grassland and savanna should decline locally and see their distribution shift as

such mosaics change into increasingly pure savanna, whereas in the same locations the

abundance of woodland species should increase as more habitat becomes available for them

(Harrison et al., 1997; Seymour & Dean, 2009; Sirami & Monadjem, 2012). Second, the

region underwent an accelerating anthropization in the last century (Niedertscheider et al.,

2012). Arable lands have been converted to irrigated agriculture, dams have been built and

commercial forests planted, all of which is expected to favour some bird species at the

expense of others (Hockey et al., 2011). Based on what is currently happening in the

Northern Hemisphere, pesticide accumulation in anthropized areas could also cause declines

in the species that utilize these landscapes (Donald et al., 2001; Hart et al., 2006; Mineau &

Whiteside, 2013; Hallmann et al., 2014). We translated these two mechanisms into

predictions about expected patterns of change in South African bird distribution as follows:

prediction 2: species adapted to cultivated land should increase in range overall; and

Acc

epte

d A

rtic

le

This article is protected by copyright. All rights reserved.

prediction 3: their abundance should not increase monotonically with the intensity of land

use; rather, their abundance should decrease after a threshold of land use intensity or time

since agricultural conversion. To address our three predictions, we estimated spatially-

explicit, species-specific rates of distribution change (colonization and extinction

probabilities), and population trend (change in abundance), for 191 widespread South African

passerine bird species. We used occupancy models that accommodate imperfect sampling,

i.e., the fact that sampling effort is not independent from environmental factors, and that the

observed biodiversity is a subset of the total biodiversity.

Material and Methods

Citizen-science and occupancy models

By harnessing the sampling effort of thousands of participants, citizen-science projects can

sample the spatial distribution of large suites of species at large spatial scales and fine

resolution, making it possible to study the correlation between species’ attributes and their

range dynamics. However, citizen-science databases come with inherent problems

(Dickinson et al., 2010; Bonney et al., 2014). First, sampling effort is necessarily biased

towards areas where citizen scientists live, have access to, or prefer to go. It is for example

conceivable that as awareness of climate change increased among data contributors, sampling

effort increased along purported colonisation fronts for temperature-limited species. Second,

some species always go undetected during surveys and observed species richness is a fraction

of true species richness. Species’ detection probability can in particular depend on their local

population density. Both issues (spatially-biased sampling effort and imperfect detection) are

hereafter collectively referred to as imperfect sampling.

The Southern African Bird Atlas Project or SABAP (http://sabap2.adu.org.za/;

Harrison et al., 1997) provides extensive information about bird distributions in South Africa,

Acc

epte

d A

rtic

le

This article is protected by copyright. All rights reserved.

Lesotho and Swaziland (South Africa hereafter), as well as Namibia, Botswana and

Zimbabwe. We used the data from between Jan. 1, 1987 and Dec. 31, 1991 (first phase or

SABAP1) and between June 1, 2007 and Oct. 5, 2013 (second phase or SABAP2, which is

still ongoing after our cut-off date). During the first phase, South Africa was divided into

2025 quarter degree grid cells (each cell ca. 25 × 25 km). During the second phase, each

quarter degree grid cell was further divided into 5’×5’grid cells. For this study we aggregated

the data of the second phase at the quarter degree resolution in order to be able to compare

both phases. Birders were asked to submit lists of all species that they saw or heard during

two hour or longer visits to the grid cells. By doing so, they also provided information about

the species they did not see or hear. In addition, birders were allowed to add further sightings

that they made in a grid cell up to five days after the initial list was compiled. We restricted

our analysis to the 191 passerine species reported in more than 2% of SABAP checklists

(excluding Palearctic migrants and introduced species).

To accommodate imperfect sampling in SABAP data, we designed occupancy models.

Occupancy models are state-space models that use detections / non-detection data, replicated

across space and/or time, to separate species’ occupancy probability from detection

probability (MacKenzie et al., 2006; Bled et al., 2013; Péron & Altwegg, 2015). Briefly, we

denote the probability that a given grid cell is occupied by a given species and p the

probability that this species is detected in the grid cell if it occupies it. For any given visit by

a SABAP contributor to the focal grid cell, the data for the focal species is if the

species is detected, otherwise, and in a single-season, single-species occupancy model

the probability of recording observation bit y is:

Acc

epte

d A

rtic

le

This article is protected by copyright. All rights reserved.

In other words, if the species is not detected, it is either because it was absent or

because it was present but escaped detection, which creates uncertainty about species

distribution. The magnitude of this uncertainty increases when the number of visits to the

focal grid cell (sampling effort) decreases. Temporal change in occupancy probability was

interpreted in terms of local extinction/colonization dynamics (state process). In other words,

a cell in which occupancy increased was likely to be colonized if not previously occupied,

and conversely a cell in which occupancy probability decreased was likely to lose the species

if previously occupied. Temporal change in detection probability (observation process) was

interpreted in terms of change in abundance (Robertson et al., 1995; Royle & Nichols, 2003;

Huntley et al., 2012; Péron & Altwegg, 2015). In our approach detection probabilities were

therefore not treated as nuisance parameters; rather, they carried information about relative

species’ abundance, under the rationale that a species is more likely to be detected if it is

locally common than locally rare. Furthermore for SABAP2 we used a removal design model

structure (MacKenzie et al., 2006) to separate, for each contribution, the observations made

during the initial visit from any further observation made during the following five-day

period.

We used a set of eight spatially-explicit covariates (four climate variables, two

vegetation cover variables, human population density, and density of the road network;

Appendix S1) to estimate spatial variation in occupancy and detection probabilities,

separately during the two phases of the SABAP. Our approach is to use correlations to

constrain the parameters of an interpolative model of species’ distribution, but we do not use

these correlations between species’ occurrence and environmental covariates to test for a

functional relationship, i.e., we do not claim to document projectable climate envelopes

Acc

epte

d A

rtic

le

This article is protected by copyright. All rights reserved.

(Beale et al., 2008). We did not use a multi-season parameterisation of occupancy models

(MacKenzie et al., 2003) because the data were sparse across large regions and with only two

“seasons” (i.e., two SABAP phases), the added complexity of the multi-season

parameterisation did not bring any clear benefits for our purpose. We used a parametric

bootstrap with 150 replicates to compute a standard error for all the spatially-explicit metrics

used in this study.

Spatial predictions with occupancy models: a few caveats

Like many other species distribution modelling efforts, we used the available detection/non-

detection data to extrapolate and interpolate to areas that were not sampled by SABAP

contributors. We did that while allowing for the fact that species can go undetected in places

where they actually occur, i.e., non-detection does not always mean absence (Altwegg et al.,

2008). Unfortunately, when detection probability is low, relatively dense non-detection data

(many non-detection events for the same set of covariate values) may be required to conclude

that a species is truly absent rather than present but not detected. This can obviously limit the

power of the inference compared to an analysis that would ignore the risk of bias associated

with imperfect sampling. Furthermore, the parameters of occupancy models have a finite

domain of definition, i.e., where a species is absent, its detection probability is not defined,

and conversely where a species has a zero detection probability it is not possible to directly

estimate whether it is present. All of the above can lead to uninformative estimates; typically,

occupancy probability can locally reach the boundary one where detection probability is

locally at boundary zero, meaning the species is estimated as occurring in the area, but

impossible to detect. Unfortunately, these boundary values do not necessarily have wide

confidence intervals (as estimated by parametric bootstrap), which could have been used to

identify problematic values. We dealt with this problem by working with the product of

occupancy and detection probabilities, which we term “recording probability”. Recording

Acc

epte

d A

rtic

le

This article is protected by copyright. All rights reserved.

probability is the probability that a species occurs in a grid cell and is recorded during any

visit by a SABAP contributor and we interpret it as an index of abundance. As an alternative

to the use of recording probability, we selected a cut-off value of 33% occupancy probability.

Below this cutoff value an area was considered as not belonging to the area of regular

presence of the species. We defined “cut-off detection probability” as the spatially-explicit

variable with value zero if the species has an occupancy probability below 33%, and equal to

the species’ detection probability otherwise. A preliminary analysis of the sensitivity to the

choice of cut-off value (between 10 and 50%) found no qualitative effect on the conclusions

in this article. Earlier studies have used reporting rates, i.e. the proportion of checklists that

report a species, as a measure of that species’ abundance (Robertson et al., 1995; Huntley et

al., 2012). Compared to this earlier approach, our two metrics (recording probability and cut-

off detection probability) present the advantage that they account for variation in sampling

effort and detection probability more naturally, and allow the separate modelling of species’

presence and abundance.

Predictions: non-climatic factors of change in passerine distribution

To test prediction 1, we categorized species into four classes of habitat preference, based on

information in Roberts Birds of Southern Africa, 7th

edition (Hockey et al., 2005). This

textbook provides habitat association information using categorical variables ranging from 0

to 4 (0: the species never uses the habitat; 4: primary habitat for the species). Category 1

represented specialists of closed savanna and woodlands (in Roberts textbook: ranked as 4 for

their use of savanna, forest or thicket, and 2 or lower for their use of grasslands). Category 4

represented specialists of open savanna never occurring in closed-canopy habitats (in Roberts

textbook: ranked as 4 for their use of grasslands and lower than 2 for their use of savanna,

woodland or thicket). Intermediate categories 2 and 3 respectively ranked higher and lower

for their use of wooded habitat than grassland habitat, without being as clear-cut as categories

Acc

epte

d A

rtic

le

This article is protected by copyright. All rights reserved.

1 and 4. We expected that Category 1 would increase throughout the Savanna Biome and at

the border with the Grassland Biome, and that Category 4 would decrease where Category 1

increased. We tested that prediction using a mixed regressive spatial autoregressive model

(MRSAR) of average recording probability in Category 1 species against average recording

probability in Category 4 species. MRSARs (Ord, 1975) estimate the fixed effect of spatially-

explicit explanatory covariates while accounting for spatial auto-correlation in the dependent

variable using a weight matrix. They provide a measure of spatial auto-correlation and of the

remaining spatial variance. We defined our weight matrix with weight 1 for grid cells that

were neighbours and weight 0 otherwise. All probabilities were logit-transformed before

analysis in the MRSAR. Sampling uncertainty in both the predicted and explanatory variables

was carried over into the MRSAR using a parametric bootstrap.

To test predictions 2 and 3 we categorized species into four classes of sensitivity to

anthropization. We used the information provided in the Roberts textbook (level of use of

agricultural areas ranked from 0: no use, to 4: intensive use), combined with ancillary

information from Hockey et al. (2011) and personal observations. The four classes went from

Category 1, species that are rarely or never recorded in anthropized areas, to Category 4,

species that thrive in cultivated areas or are otherwise positively affected by anthropization.

Although the effect of anthropization is complex, we predicted that the semi-desert Karoo

Biomes should experience increases in farmland species because of the development of

irrigation (prediction 2). By contrast, in the areas that were converted to agriculture more than

two centuries ago, we expected the same trend as in Europe and North America: a decline in

farmland birds (prediction 3). We tested that prediction using a MRSAR of cut-off detection

probability in category 4 species against log-transformed human population density. The

detail of the categories is presented in Appendix Table S1.

Acc

epte

d A

rtic

le

This article is protected by copyright. All rights reserved.

Results

Preliminary considerations

In all but a few species, occupancy probability estimates were lower throughout during

SABAP1 than SABAP2 (see “General trends” in the Discussion section). Fig. 1 illustrates

this phenomenon in two widespread species. Overall across species, the humid south-east

exhibited the most pronounced increase in occupancy probabilities, whereas the Savanna

Biome appeared as the most stable. Lastly, in the west, most species decreased in detection

probability, whereas in the east most species increased in detection probability.

Prediction 1: Closed-savanna species take over open-savanna species

The average grid cell-specific changes in recording probability among open and closed

savanna specialists were negatively correlated with each other (MRSAR: slope parameter –

0.22 ± bootstrap SE 0.09; Fig. 2A vs. 2D). In particular, closed-savanna specialists increased

in recording probability in the cells containing the transitions between the Coastal Forest,

Savanna, and Grassland Biomes along the Indian Ocean coast and the Great Escarpment (see

Appendix Fig. S1 for geographical references). Open-savanna specialists exhibited a strong

increase in recording probability in the Grassland Biome, and comparatively little change in

the northern savannas (Limpopo province).

Predictions 2 and 3: Correlates of land use intensification

Given the general trend for species to increase in occupancy probability within their range,

the relative stability of species categorized as favored by anthropization (Fig. 3D) suggests

that the effects of anthropization operated before the start of the first atlas period.

Nevertheless, we observed on average an increase in cut-off detection probability in the

Grassy Karoo and Sweet Grassland Bioregions for species categorized as favored by

anthropization (Fig. S3D; see Appendix Fig. S1 for geographical references). By contrast, the

Acc

epte

d A

rtic

le

This article is protected by copyright. All rights reserved.

same species declined in the areas with most intense human population density (MRSAR:

slope parameter –0.016 ± bootstrap SE 0.008; Fig. S3D). Counter-intuitively, species

categorized as incompatible with anthropization showed moderate increase throughout (Fig.

3A & S3A).

Discussion

Using data from a large-scale citizen-science project, we documented large changes in the

distributions of passerine birds across South Africa. Our results suggest that bush

encroachment and land-use changes are two important factors affecting species range

dynamics in the region. In the following, we first discuss the methodological implications

regarding occupancy models for crowdsourced data analysis in general and SABAP data in

particular, and second the implications for our understanding of large-scale ecological

processes.

General trends and methodological experience sharing

The reported overall increase in passerine occupancy and detection probability over the study

period corroborates preliminary analyses of SABAP reporting rates for passerine birds

(Loftie-Eaton, 2014). In contrast, longer-lived taxa with heavier body mass than passerines

are often showing rapid declines in the study region (Broms et al., 2014; Hofmeyr et al.,

2014; Loftie-Eaton, 2014). Increases in occupancy probability for passerines could come

from the change in protocol between SABAP1 and SABAP2. SABAP2 data are collected at a

finer spatial resolution than SABAP1 data, and we aggregated them at the quarter degree

resolution to compare them to SABAP1 data. The finer resolution of the SABAP2 protocol

may however have enticed contributors to visit areas that they ignored during SABAP1, and

discover species in those areas that were left undetected during SABAP1. This mechanism

would not be adequately captured by occupancy models because, in these models, detection

Acc

epte

d A

rtic

le

This article is protected by copyright. All rights reserved.

probability is defined conditional on species’ occurrence in the sampled area.

Downscaling/upscaling methods (e.g., Azaele et al., 2012) could in the future prove useful to

address this change in the spatial resolution of the data collection protocol between two

sampling period.

Alternatively, the observed increase in occupancy probability could also be explained

by changes in observer behaviour and skills. It is possible that, for example via the

dissemination of ornithological knowledge, spatial heterogeneity in the probability that

observers identified and reported all the species they saw decreased between SABAP1 and 2.

Such homogenization of observers’ skill level could potentially be addressed using time-

varying observer random effects. Time-invariant observer random effects are readily fitted to

citizen science data, including to SABAP data (Bled et al., 2013). Data permitting, they could

be modified into time-varying random effects to accommodate changes in observers’ skill

levels.

On the other hand, if the reported increase in occupancy probability is not spurious

and caused by the aforementionned two sources of bias, it suggests that species have been

consolidating their range, i.e, that within ecologically-homogeneous regions, their

distribution was patchy during SABAP1 and gaps were filled between SABAP1 and

SABAP2.

Prediction 1: Closed-savanna species take over open-savanna species

The result that closed-savanna species increased where open-savanna species decreased is

robust to caveats mentioned in the previous paragraph because it comes from the comparison

of metrics affected by the same type of bias. We thus support the hypothesis that bush

encroachment is a major driver of change in bird distribution in the study region (Seymour &

Dean, 2009; Sirami & Monadjem, 2012; Hofmeyr et al., 2014). Our study is the first to find

evidence of this spatial scale for the effect of bush encroachment on avian assemblages. In a

Acc

epte

d A

rtic

le

This article is protected by copyright. All rights reserved.

species like common fiscal Lanius collaris (Fig. 1A), which requires dense shrubs

interspersed with open areas, bush encroachment meant extinction from the northernmost

cells (Fig. 1A.i against 1A.iv), probably linked to the closing of small grassy patches there,

but increased abundance over the rest the range, probably linked to the increased availability

of bush patches within agricultural landscapes and grasslands (Fig. 1A.ii against 1A.v).

Common fiscal is one of the most familiar and easy-to-detect members of the region’s

avifauna, so these elements are robust to the observer skill caveat mentioned under “General

trends” above.

Prediction 2: Correlates of land use intensification

For the same reason as for the bush encroachment effect, the results pertaining to the

correlates of land use intensification are considered robust to the caveats mentioned under

“General trends” above. As expected, the effects of land use were however less

straightforward to identify than the effect of bush encroachment because land use affects

species’ habitat and demographic performance in various ways. An example of counter-

intuitive results is the increase, within the Savanna Biome and in particular along the Great

Escarpment, in the abundance of the species that were a priori categorized as incompatible

with anthropization (Fig. S3A). This result could be explained in part by the fact that we

excluded all the rare species from the analysis. It suggests that those species that were

common enough to be included in our analysis, but still qualified as incompatible with

anthropization, have been benefiting from bush encroachment, particularly in the ecotone

areas. By contrast, the abundance of red-billed quelea Quelea quelea (Fig. 1B), a grassland

species which benefits from cereal cultivation and therefore classifies as a human

commensal, increased most markedly in the agricultural landscapes of the centre of the study

region, in the Grassy Karoo and Sweet Grassland Bioregions (Fig. 1B). Lastly, species that

were a priori considered as benefiting from anthropization were found to decline in detection

Acc

epte

d A

rtic

le

This article is protected by copyright. All rights reserved.

probability in the regions with highest human population density (i.e., heaviest human

footprint). This result supports the hypothesis of a threshold in land use intensity, after which

human commensals begin to decline. This effect has recently been reported in most countries

with intensive agricultures (e.g., Mineau and Whiteside 2013, Hallmann et al. 2014).

Conclusion

Our two main results are 1) that open-savanna specialists decreased throughout South Africa

where closed-savanna specialists increased, and 2) that species a priori classified as favoured

by anthropization decreased in abundance where human population density was highest.

These two results are robust to caveats mentioned under “General trends” above because

they compare metrics affected by the same type of bias. The first result constitutes evidence

for a major effect of bush encroachment on avian assemblages in the region, thereby

corroborating earlier evidence at the more local scale and with fewer species (Seymour &

Dean, 2009; Sirami & Monadjem, 2012). The second result suggests a threshold in land use

intensity after which even human commensals are negatively affected. We conclude that in

the Grassland and Savanna Biome, land use change and ecological succession appear as

faster, more direct drivers of animal distribution changes than climate variables themselves.

This conclusion would probably be challenged in more arid biomes, where surface water

availability (Wichmann et al., 2003) and thermic stress (Smit et al., 2013), both directly

linked to climate, are likely to play a greater role than changes in vegetation structure. With

regard to strategies for species distribution modelling, our study supports the consensus that,

at the sub-continental to regional scale, land use and vegetation structure variables cannot be

ignored, whereas at the continental scale climate differences may overrule land use

differences (Pearson & Dawson, 2003; Thuiller et al., 2004). Occupancy models represented

an investment in computer time which we deemed justified due to the variation in sampling

effort by citizen scientists across the study region (South Africa, Lesotho and Swaziland), and

Acc

epte

d A

rtic

le

This article is protected by copyright. All rights reserved.

especially since the objective was to compare the dynamics of different species and groups of

species with very different detection probabilities. Occupancy models allowed us to account

for species-specific biases. However, some specificities of the SABAP dataset remain

challenging for standard occupancy models and additional methodological development may

therefore be required in the future (see “General trends” above). Nevertheless, our two main

results already exemplify the role of factors of variation other than climate change and offer

important insight into ongoing dynamics of avian assemblage composition driven by

ecological succession and land use change.

Acknowledgements

We thank the volunteer atlasers who collected the data and the Animal Demography Unit at

University of Cape Town for making the data available to us. C. Beale, F. Bled, G.

Duckworth, D. Guo, B. Huntley, S. Mecenero, and Statistics South Africa helped with

covariate data access and formatting. This work was funded by SANBI and by the National

Research Foundation of South Africa (Grant 85802). The NRF accepts no liability for

opinions, findings and conclusions or recommendations expressed in this publication.

Literature cited

Achard F, Eva HD, Stibig H-J, Mayaux P, Gallego J, Richards T, Malingreau J-P (2002)

Determination of deforestation rates of the world’s humid tropical forests. Science, 297,

999–1002.

Altwegg R, Wheeler M, Erni B (2008) Climate and the range dynamics of species with

imperfect detection. Biology letters, 4, 581–4.

Azaele S, Cornell S, Kunin W (2012) Downscaling species occupancy from coarse spatial

scales. Ecological Applications, 22, 1004–1014.

Acc

epte

d A

rtic

le

This article is protected by copyright. All rights reserved.

Beale CM, Lennon JJ, Gimona A (2008) Opening the climate envelope reveals no macroscale

associations with climate in European birds. Proceedings of the National Academy of

Sciences of the United States of America, 105, 14908–14912.

Bled F, Nichols JD, Altwegg R (2013) Dynamic occupancy models for analyzing species’

range dynamics across large geographic scales. Ecology and evolution, 3, 4896–4909.

Bonney R, Shirk JL, Phillips TB, Wiggins A, Ballard HL, Miller-Rushing AJ, Parrish JK

(2014) Next Steps for Citizen Science. Science, 343, 1436–1437.

Broms KM, Johnson DS, Altwegg R, Conquest LL (2014) Spatial occupancy models applied

to atlas data show Southern Ground Hornbills strongly depend on protected areas.

Ecological Applications, 24, 363–374.

Buitenwerf R, Bond WJ, Stevens N, Trollope WSW (2012) Increased tree densities in South

African savannas: >50 years of data suggests CO2 as a driver. Global Change Biology,

18, 675–684.

Chazdon RL, Letcher SG, van Breugel M, Martínez-Ramos M, Bongers F, Finegan B (2007)

Rates of change in tree communities of secondary Neotropical forests following major

disturbances. Philosophical transactions of the Royal Society of London. Series B,

Biological sciences, 362, 273–89.

Dickinson JL, Zuckerberg B, Bonter DN (2010) Citizen Science as an Ecological Research

Tool: Challenges and Benefits. Annual Review of Ecology, Evolution, and Systematics,

41, 149–172.

Donald P, Green RE, Heath MF (2001) Agricultural intensification and the collapse of

Europe’s farmland bird populations. Proceedings Of The Royal Society B-Biological

Sciences, 268, 25–9.

Foley JA, Defries R, Asner GP et al. (2005) Global consequences of land use. Science, 309,

570–4.

Hallmann CA, Foppen RPB, van Turnhout CAM, de Kroon H, Jongejans E (2014) Declines

in insectivorous birds are associated with high neonicotinoid concentrations. Nature,

511, 341–343.

Harrison JA, Allan DG, Underhill LG, Herremans M, Tree AJ, Parker V, Brown CJ (eds.)

(1997) The Atlas of Southern African Birds. BirdLife South Africa, Johannesburg.

Hart JD, Milsom TP, Fisher G, Wilkins V, Moreby SJ, Murray AWA, Robertson PA (2006)

The relationship between yellowhammer breeding performance, arthropod abundance

and insecticide applications on arable farmland. Journal of Applied Ecology, 43, 81–91.

Hockey PAR, Dean WRJ, Ryan PG (2005) Roberts Birds of Southern Africa, 7th Edition, 7th

edn. The Trustees of the John Voelcker Bird Book Fund, Cape Town.

Acc

epte

d A

rtic

le

This article is protected by copyright. All rights reserved.

Hockey PAR, Sirami C, Ridley AR, Midgley GF, Babiker HA (2011) Interrogating recent

range changes in South African birds: confounding signals from land use and climate

change present a challenge for attribution. Diversity and Distributions, 17, 254–261.

Hofmeyr SD, Symes CT, Underhill LG (2014) Secretarybird Sagittarius serpentarius

Population Trends and Ecology: Insights from South African Citizen Science Data (ed

Slotow R). PloS one, 9, e96772.

Huntley B, Altwegg R, Barnard P, Collingham YC, Hole DG (2012) Modelling relationships

between species spatial abundance patterns and climate. Global Ecology and

Biogeography, 21, 668–681.

Loftie-Eaton M (2014) Geographic Range Dynamics of South Africa’s Bird Species. MSc

thesis, University of Cape Town.

MacKenzie DI, Nichols JD, Hines JE, Knutson MG, Franklin AB (2003) Estimating site

occupancy, colonization, and local extinction when a species is detected imperfectly.

Ecology, 84, 2200–2207.

MacKenzie DI, Nichols JD, Royle JA, Pollock KH, Bailey LL, Hines JE (2006) Occupancy

estimation and modelling. Academic Press, San Diego, CA.

Mineau P, Whiteside M (2013) Pesticide acute toxicity is a better correlate of U.S. grassland

bird declines than agricultural intensification. (ed Clarke RH). PloS one, 8, e57457.

Niedertscheider M, Gingrich S, Erb K-H (2012) Changes in land use in South Africa between

1961 and 2006: an integrated socio-ecological analysis based on the human

appropriation of net primary production framework. Regional Environmental Change,

12, 715–727.

Ord K (1975) Estimation methods for models of spatial interaction. Journal of the American

Statistical Association, 349, 120–126.

Parr CL, Gray EF, Bond WJ (2012) Cascading biodiversity and functional consequences of a

global change-induced biome switch. Diversity and Distributions, 18, 493–503.

Pearson RG, Dawson TP (2003) Predicting the impacts of climate change on the distribution

of species: are bioclimate envelope models useful? Global Ecology and Biogeography,

12, 361–371.

Péron G, Altwegg R (2015) The abundant centre syndrome and species distributions: insights

from closely related species pairs in southern Africa. Global Ecology and Biogeography,

24, 215–225.

Robertson A, Simmons RE, Jarvis AM, Brown CJ (1995) Can bird atlas data be used to

estimate population size? A case study using Namibian endemics. Biological

Conservation, 71, 87–95.

Royle JA, Nichols JD (2003) Estimating abundance from repeated presence-absence data or

point counts. Ecology, 84, 777–790.

Acc

epte

d A

rtic

le

This article is protected by copyright. All rights reserved.

Scheffer M, Carpenter SR (2003) Catastrophic regime shifts in ecosystems: linking theory to

observation. Trends in Ecology & Evolution, 18, 648–656.

Seymour CL, Dean WRJ (2009) The influence of changes in habitat structure on the species

composition of bird assemblages in the southern Kalahari. Austral Ecology, 35, 581–

592.

Sirami C, Monadjem A (2012) Changes in bird communities in Swaziland savannas between

1998 and 2008 owing to shrub encroachment. Diversity and Distributions, 18, 390–400.

Sirami C, Brotons L, Martin J-L (2007) Vegetation and songbird response to land

abandonment: from landscape to census plot. Diversity & Distributions, 13, 42–52.

Smit B, Harding CT, Hockey PAR, McKechnie AE (2013) Adaptive thermoregulation during

summer in two populations of an arid-zone passerine. Ecology, 94, 1142–54.

La Sorte FA, Jetz W (2012) Tracking of climatic niche boundaries under recent climate

change. The Journal of animal ecology, 81, 914–25.

Thuiller W, Araújo MB, Lavorel S (2004) Do we need land-cover data to model species

distributions in Europe? Journal of Biogeography, 31, 353–361.

Vitousek PM (1997) Human Domination of Earth’s Ecosystems. Science, 277, 494–499.

Wichmann MC, Jeltsch F, Dean WRJ, Moloney KA, Wissel C (2003) Implication of climate

change for the persistence of raptors in arid savanna. 1, 186–202.

Wigley BJ, Bond WJ, Hoffman MT (2010) Thicket expansion in a South African savanna

under divergent land use: local vs. global drivers? Global Change Biology, 16, 964–976.

Figure legends

Fig. 1. Examples of occupancy models of species distribution using citizen science data, for

two widespread species with contrasting habitat requirements: (A) common fiscal (a.k.a.

fiscal shrike or Southern fiscal) Lanius collaris, and (B) red-billed quelea Quelea quelea. (i to

iii): distribution during SABAP1 (1987-1991). (iv to vi): distribution during SABAP2 (2007-

2013). (i and iv): Estimated occupancy probability. (ii and v) Estimated detection probability.

(iii and vi ) Reporting rate, i.e. the proportion of checklists contributed by citizen scientists in

which the species was reported.

Acc

epte

d A

rtic

le

This article is protected by copyright. All rights reserved.

Fig. 2. Summed change in recording probability (the product of occupancy and detection

probabilities) for four categories of species with varying preference for closed or open

savanna habitats. n is the number of species in each category. The corresponding map of

detection probability is provided in Fig. S2.

Fig. 3. Average change in recording probability (the product of occupancy and detection

probabilities) for four categories of species with varying sensitivity to anthropisation. n is the

number of species in each category. The corresponding map of detection probability is

provided in Fig. S3.

Acc

epte

d A

rtic

le

This article is protected by copyright. All rights reserved.

Acc

epte

d A

rtic

le

This article is protected by copyright. All rights reserved.