Delaying conservation actions for improved knowledge: how long should we wait

L E T T E RDelaying conservation actions for improved

knowledge: how long should we wait?

Hedley S. Grantham,1* Kerrie A.

Wilson,1 Atte Moilanen,2 Tony

Rebelo3 and Hugh P.

Possingham1

1The Ecology Centre, University

of Queensland, St Lucia,

Brisbane, Qld 4072, Australia2Department of Biological and

Environmental Sciences, PO Box

65 (Biocentre III), FIN-00014,

University of Helsinki, Helsinki,

Finland3South African National

Biodiversity Institute, Private

Bag X101, Pretoria 0001, South

Africa

*Correspondence: E-mail:

[email protected]

Abstract

Decisions about where conservation actions are implemented are based on incomplete

knowledge about biodiversity. The Protea Atlas is a comprehensive database, containing

information collated over a decade. Using this data set in a series of retrospective

simulations, we compared the outcome from different scenarios of information gain, and

habitat protection and loss, over a 20-year period. We assumed that there was no

information on proteas at the beginning of the simulation but knowledge improved each

year. Our aim was to find out how much time we should spend collecting data before

protecting habitat when there is ongoing loss of habitat. We found that, in this case,

surveying for more than 2 years rarely increased the effectiveness of conservation

decisions in terms of representation of proteas in protected areas and retention within

the landscape. If the delay is too long, it can sometimes be more effective just using a

readily available habitat map. These results reveal the opportunity costs of delaying

conservation action to improve knowledge.

Keywords

Biodiversity data, dynamic conservation planning, resource allocation, uncertainty.

Ecology Letters (2009) 12: 293–301

I N T R O D U C T I O N

Data on biodiversity, even of well-known taxa, are grossly

incomplete and often biased taxonomically, temporally and

geographically (Reddy & Davalos 2003). Nevertheless, given

ongoing habitat loss and degradation we must use the best

information that is available to make choices about where to

invest limited funds for biodiversity conservation (Bode

et al. 2008). Protected areas are the foundation of many

conservation strategies and their location is often deter-

mined by sparse biological data and coarse surrogates for

biodiversity such as habitat types (Ferrier 2002).

Investing in data increases knowledge but surveying takes

time and money (Gardner et al. 2008; Grantham et al. 2008;

Perhans et al. 2008). Surveying takes time because of the

logistics of reaching survey sites, limitations in the number

of experts and limitations on funding (Gaston & Rodrigues

2003; Hill et al. 2005). Surveying can be expensive because

of costs such as wages and fuel (Hill et al. 2005). Even well-

funded programs can take decades to complete. For

example, the Biological Survey of the state of South

Australia has the goal to complete systematic surveys across

the state to provide a broad baseline inventory of vegetation,

plants and animals. This programme started in 1971 and is

due to be completed in 2015.

The effectiveness of different biodiversity data, how they

represent other biodiversity features and how they influence

the design of protected areas has been explored extensively

(e.g. Freitag & van Jaarsveld 1998; Gaston & Rodrigues

2003; Gladstone & Davis 2003; Grand et al. 2007). These

studies take a subset of the data and determine its

effectiveness and efficiency for achieving conservation

objectives assuming the whole system is static, that is, data

collection is instantaneous and there is no habitat loss. In

reality, however, implementation of new protected areas and

surveying is time-consuming and landscapes are dynamic

(Meir et al. 2004). Recent developments in conservation

planning techniques have started to consider these realistic

constraints when designing new protected areas (Costello &

Polasky 2004; Moilanen & Cabeza 2007; Murdoch et al.

2007; Wilson et al. 2007; Underwood et al. 2008). Grantham

et al. (2008) compared different investments in survey data

and used these to simulate the implementation of new

protected areas and habitat loss and over a 10-year period.

They discovered that the effectiveness of different survey

investments was subject to diminishing returns. However,

Ecology Letters, (2009) 12: 293–301 doi: 10.1111/j.1461-0248.2009.01287.x

� 2009 Blackwell Publishing Ltd/CNRS

this and other previous work on the value of improved

knowledge have ignored the time required to obtain data. If

too much time is spent collecting data before making

decisions, we may stand to miss opportunities for their

protection, particularly if there is ongoing loss of habitat.

Conversely, not enough data may lead to poor decisions

(Possingham et al. 2007). In the context of dynamic and

ever-changing world, would the resource spent on learning

be better spent on doing? Presumably a trade-off exists,

although to our knowledge this has not yet been quantified.

We assess the value of delaying conservation actions in

order to invest in data collection, while accounting for the

constraints imposed by a finite budget and ongoing habitat

loss. Surveys were simulated using one of the most extensive

point locality databases in the world, The Protea Atlas

(Forshaw 1998). We applied land-use simulations over a

20-year period to the Fynbos biome in South Africa. We

assumed that there was no information on the distribution

of Proteas at the beginning of the simulation but there was

an existing habitat map and our aim was to find which

combination of investment in biological survey and conser-

vation action produced the best outcome in terms of

protection of Proteas.

M A T E R I A L S A N D M E T H O D S

Study area

Our study area was the Fynbos biome, South Africa. We

divided our landscape into 80 773 planning units, each

1 km2. We excluded any planning units that were partly

outside the region. Using data on vegetation cover from

Cowling et al. (1999), we categorized planning units as

unvegetated and not containing protea habitat if their

centres had no native vegetation. We then excluded these

from the analyses. These exclusions left us with 53 385

planning units. Of these, we classified 18 457 as protected

if their centres were within a protected area (IUCN Type

I-III) based on data from Reyers et al. (2007). The

remaining 34 928 planning units were available for

conservation management but also vulnerable to habitat

loss.

Data

We used data in The Protea Atlas (Forshaw 1998) to

measure the advantages of spending more or less time on

data collection before conservation actions are made. Within

our study area this database contained over 40 000 plots

with 220 000 occurrence records for 381 taxa of species in

the family Proteaceae. Subspecies of protea was treated the

same as species. We removed all records of proteas that

were cultivated or hybrids.

We first developed our evaluation data set, which we

assumed represented the complete distribution of proteas.

To model the distribution of proteas with more than 20

records we applied the maximum entropy method using the

machine learning software Maxent version 1.8.6 (Phillips

et al. 2006; Pearson et al. 2007). The model finds the

probability distribution of maximum entropy, which are

values that are most spread out, or closest to uniform,

subject to constraints imposed by the information available

regarding the surveyed distribution of proteas and environ-

mental conditions across our study area. The models had the

same resolution as the planning units and were based on 25

biological, physical and climatic predictor variables. We

created presence ⁄ absence distribution models of each

protea from using the cumulative probability, which in

Maxent is an indicator of habitat suitability (Phillips et al.

2006). We assume a protea is present if the cumulative

probability was over 20 (S. Phillips, personal communica-

tion) and the planning unit was not classified as cleared. For

proteas with less than 20 records, we did not use

distribution models, recording them as present within

uncleared planning units containing actual records. We

randomly checked 25% of the models to ensure their

predictive accuracy by measuring the area under the curve

(ROC > 0.95) (Elith et al. 2006) and compared several

modelled distributions to published distribution maps. We

used 334 modelled proteas and 47 unmodelled proteas as

our evaluation data set. For the purposes of this evaluation

of the value of data we assume that these data represent the

�truth� about the complete distribution of each protea

although in reality there are likely to be many omission and

commission errors as common with this type of data

(Rondinini et al. 2006).

Scenarios

To compare the outcomes of surveying for different lengths

of time we considered different scenarios using land-use

simulations over a 20-year period (Fig. 1.). There were three

components to the land-use simulations: (a) survey of

proteas and development of a decision support database, (b)

implementation of new protected areas and (c) habitat

destruction.

We compared six different survey durations: 1, 2, 4, 6, 8

and 10 years. We calculated the number of plots surveyed

per year based on the rate of collection of data for The

Protea Atlas by volunteers over a period of around 10–

12 years. About 40 000 plots were surveyed in our study

region. We therefore assumed that surveyors covered 3000

plots each year. As a test of sensitivity of our results, we

increased and decreased these survey rates to 1000 and 6000

plots surveyed each year. For each scenario we calculated

the number of plots surveyed over the entire survey period

294 H. S. Grantham et al. Letter


using the survey rate multiplied by the duration of surveying.

We assumed that all plots were available for surveying

during simulations. For each scenario we created new

distribution maps for each protea based on our evaluation

data set. We randomly selected surveyed plots each year

from the entire data set. For each different survey duration,

we then assembled partial data sets on the distribution of

proteas. In the same way as for our evaluation data set, we

modelled the distribution of proteas with more than 20

records and listed others as present in uncleared planning

units where they had been observed. We used these data to

build a decision-support database to select new conservation

areas.

For each simulation we modelled a 10-year period of new

protected areas. We did this to standardize between

comparisons. The 10-year period started after the final

survey period. For example, if there was a 4-year period of

surveying, the 10 year period of establishing new protected

areas extended from years 5 to 14, inclusive. The rate of new

protected areas was 2000 km2 year)1 based on the average

annual rate of establishment of statutory and non-statutory

protected areas over the last 20 years (Cowling & Pressey

2003), with non-statutory areas contributing 50% to the rate.

For each of the 10 years following protea data collection, we

allocated 2000 km2 of conservation action, then simulated

loss of habitat in areas that had not been protected. In years

when no protection occurred, we simulated only habitat

loss.

To select new protected areas, we used an algorithm that

accounts for likely effects of habitat loss (described in

Moilanen & Cabeza 2007). The algorithm minimized

expected short-term loss of conservation value, measured

as a function of retention across the entire landscape. In

this, the planning unit added next is the site k that has

greatest DV STL(k),

DV STLðkÞ¼ 1

ck

XNS

j¼1

Vjðrjð1;Sþk;U�kÞÞ�Vjðrjð1;S ;U ÞÞ� �

;

in which Ns is the number of species (proteas; indexed by j ),

ck is the cost of adding site k to the set of protected areas.

We used a convex square root benefit function, Vj(Æ), to

translate representation of species j to conservation value.

Using such a function skews selection towards rare proteas

(Arponen et al. 2005). rj(n, S, U) is the expected retention n

time steps after the present time. This measure of retention

accounts for representation in selected sites S as well as

expected representation from the set of unprotected sites

remaining in the landscape, U:

rjðn; S ; U Þ ¼X

i2S

aijþX

i2U

ð1� liÞnaij ;

where li is the yearly estimated loss rate of site i and aij is the

amount of species j in site i. The expected addition in

the retention of proteas after the addition of site k into the

protected area network works out to be the probability that

the site would be lost multiplied by the representation of

species in the site [1–(1–lk)n ]akj (Moilanen & Cabeza 2007).

The more advanced algorithm for sequential planning

described by Moilanen & Cabeza (2007), the sequence

ordering algorithm, was not used due to computational

limitations with systems having tens of thousands of sites.

We compared this to using a maximum gain algorithm

(described in Moilanen & Cabeza 2007), which selected

planning units to maximize short-term gain of conservation

value inside protected areas. This algorithm iteratively

selected the planning unit k that gave the highest marginal

increase in conservation value, DV STG(k), relative to existing

protected areas and previously selected notional protected

areas. The marginal gain criterion used in this algorithm is:

DV STGðkÞ ¼ 1

ck

XNS

j¼1

Vj ðRjðS þ kÞÞ �VjðRjðSÞÞ� �

;

in which Rj (S) is the representation (summed amount of

occurrences) of species j inside the set of areas S, which in

this case includes already protected or notionally selected

areas.

Each year of the whole simulation we simulated spatially-

explicit stochastic �habitat loss�. The loss rate was based on

the average rate of clearing of native vegetation in South

Africa between 1988 and 1993, estimated to be around 2%

Figure 1 Study design for each scenario. We ran simulations for

20 years. Survey periods for new data on proteas varied between 1

and 10 years. We assumed that, once surveys were complete,

building the decision-support database was instantaneous. The

number of years doing nothing after new protected areas were

implemented was inversely related to the number of years

surveying proteas, varying from 0 to 10 years.

Letter Delaying conservation 295


per annum (Biggs & Scholes 2002). More recent and

spatially refined data were not available. This rate was

assigned to vegetation types (described in Mucina &

Rutherford 2006) based on their relative vulnerability to

clearing, obtained from Reyers et al. (2007). We normalized

vulnerability values and assigned an annual clearing rate for

each vegetation type as 2%*(normalized vulnerability value).

We then assigned each planning unit an average clearing rate

based on the vegetation types it contained. This value

became the probability of a planning unit being cleared

altogether in any 1 year. In the simulations that incorporated

existing protected areas, we assigned zero clearing proba-

bilities to planning units that were already protected. We

repeated each simulation 20 times after testing that this level

of repetition produced stable solutions with stochastic

simulated clearing. As a sensitivity test we halved and

doubled rates for both protection and habitat loss.

Often conservation planning is based solely on a habitat

map developed using easily obtainable data. We, therefore,

also included scenarios that involved no surveying of

proteas and consequent complete reliance of planning on an

existing habitat classification. The classification was based

on climate, geology and topography (described in Cowling

& Heijnis 2001) and we assumed it was freely available. To

compare the benefits of using survey data over the habitat

map alone we also ran simulations that involved different

time periods before implementation of new protected areas

based solely on the habitat map.

Scenario evaluation

We evaluated the outcome of each scenario with two

metrics: representation of proteas in protected areas and

retention of proteas in the landscape (Pressey et al. 2004),

both based on the full protea data set. We measured

representation in protected areas by first calculating the

proportion of a proteas distribution within actual or

notional protected areas at the end of the simulation. We

then measured the mean representation for each protea

across the 20 replicate simulations. Last, we used the

lower quartile of the 381 protea mean values to focus our

measures of representation on the proteas with least

protection from clearing. For retention we used a similar

approach but instead of measuring the proportion of a

proteas distribution within actual or notional protected

areas, we measured the proportion of a proteas distribu-

tion still remaining in uncleared planning units at the end

of the simulation, regardless of the conservation status of

those planning units. We ran each scenario with and

without the existing protected areas. For scenarios that

ignored the existing protected areas, we re-classified all

formally protected planning units as available for protec-

tion and vulnerable to loss and estimated their average

clearing rates according to vegetation types they contained,

as above.

R E S U L T S

When measuring representation, we found that any survey

data, regardless of the collection rate or period of survey,

was more effective at representing or retaining proteas than

the habitat map (Fig. 2a). We that found 1–2 years of

surveying resulted in protected areas with very good

Figure 2 The effectiveness of different survey periods for pro-

tecting proteas at the end of 20-year simulations. Existing

protected areas in the study region were initially included as

protected before the simulations were commenced. We measured

both (a) representation of proteas in protected areas and (b)

retention of proteas in the landscape. We based both measure-

ments on the full protea distributional data using lower quartiles

across all 381 proteas and 20 replicate simulations. For scenarios

with no survey data (zero survey period), we used a habitat map to

guide selection of notional protected areas.



representation of proteas. There were similar values in

representation between 1 and 10 years of surveying despite

a progressive increase in information on the distribution of

proteas (Table 1). We found that around 1 year of surveying

was the optimal period when measuring the retention of

proteas in the landscape. Surveying for any longer than this

presented significant opportunity costs due to ongoing

habitat loss (Fig. 2b). Basing conservation decisions on the

habitat map was better than delaying decisions for longer

than 4 years so that surveys could be completed. All

differences in retention values, however, were relatively

small (Fig. 2b).

We found, predictably, that representation and retention

values were higher when the initial protected areas were

included (Fig. 3). When initial protected areas were not

included in the analysis we found the difference in

representation was large between zero survey (the habitat

map) and all survey periods. There was however, only a

small difference in representation between the different

survey periods (Fig. 3a). When measuring retention, we

found that around 1–2 years was the optimal survey period

with and without the existing protected areas (Fig. 3b). We

also found that the habitat map produced better retention

values than any survey period longer than 4 years (Fig. 3b).

All of our results were sensitive to rates of protection

(Fig. 4). Increasing the protection rate predictably increased

representation across all survey periods. The habitat map

performed worse than any amount of survey except when

the protection rate was high. In that case, representation was

higher with zero surveys than with any amount of survey

(Fig. 4a). The difference in retention values were small

between a short survey period (1–2 years) and the habitat

map for all protection rates (Fig. 4b), but longer survey

periods reduced retention for all rates of protection.

Applying double protection rates meant that all planning

units were allocated either to protection or clearing. Within

these simulations, a potential maximum of 40 000 km2 of

new protected areas could have been implemented. How-

ever, there were too few planning units to realize this

maximum, with between 30 000 and 35 000 km2 allocated

to protection.

Table 1 Description of plots for the different scenarios

Years of

surveying

Survey

rate

Total number

of plots

Number

of records

1 1000 1000 4853

2 1000 2000 9743

4 1000 4000 19 978

6 1000 6000 29 454

8 1000 8000 39 643

10 1000 10 000 49 571

1 3000 3000 14 865

2 3000 6000 29 454

4 3000 12 000 58 959

6 3000 18 000 88 894

8 3000 24 000 118 798

10 3000 30 000 148 310

1 6000 6000 29 454

2 6000 12 000 58 959

4 6000 24 000 118 798

6 6000 36 000 178 012

8 6000 43 863* 216 857

*This should have been 48 000 but there were only 43 863 plots in

our data set.

Figure 3 The effectiveness of different survey periods for pro-

tecting proteas at the end of 20-year simulations with and without

existing protected areas. We measured both (a) representation of

proteas in protected areas and (b) retention of proteas in the

landscape. We based both measurements on the full protea

distributional data using lower quartiles across all 381 proteas and

20 replicate simulations. Survey rate was 3000 plots per annum. For

scenarios with no survey data (zero survey period), we used a

habitat map to guide selection of notional protected areas.



Different habitat loss rates influenced representation

values particularly for long survey periods of between 6 and

10 years (Fig. 4c). When loss rate was high, retention values

were similar between different survey periods and for the

habitat map (Fig. 4d). When loss rate was low, there was a

relatively large difference between different survey periods.

One year of surveying resulted in higher retention compared

to the 10-year survey period. After a survey period of

around 3 years the retention values were less than those

from the habitat map.

We found very similar results between using the

minimum loss algorithm and the maximum gain algorithm.

Additionally, we found very similar results when a habitat

map was used simultaneously with survey data to select

protected areas. When we compared different time periods

before implementing using the habitat map compared to

survey data we found that all survey periods had a higher

representation and retention value than the habitat map

(Fig. 5). The difference in values was larger for represen-

tation but less for retention.

D I S C U S S I O N

We found that, under a variety of circumstances, only 1–

2 years of protea survey provided the best conservation

outcomes for both representation of proteas in protected

areas and retention in the landscape. Rarely did surveying

longer than 2 years increase the representation of proteas in

protected areas or retention in the landscape. This was

despite a substantial increase in the knowledge of protea

distributions. For example, in 1 year of surveying at a rate of

3000 plots per year, there were 14 863 occurrence records

but for 10 years of surveying there were 148 310 occurrence

records (Table 1).

These results challenge those of Balmford & Gaston

(1999) who found that biodiversity data are generally good

value. This is because we assume that time spent surveying

is time spent not taking action. While more data are always

useful, our study suggests that delaying conservation to

collect data when there is ongoing habitat loss might not

necessarily be the best strategy. Other studies have

similarly found that time is an important factor when

determining the costs and benefits of collecting data (e.g.

Gerber et al. 2005).

Why was the best strategy to only survey proteas for 1–

2 years before starting conservation action? It is likely that

there were diminishing returns on investing in more data

and our scenarios reached a threshold at which knowledge

of the distribution of proteas was not the limiting factor in

making effective decisions but rather the loss of habitat

while surveying proceeded (Grantham et al. 2008). Longer

survey periods meant that there was more information on

proteas but this did not necessarily translate into more

effective conservation decisions because of the potential for

habitat that contains proteas to be destroyed. For repre-

sentation of proteas in protected areas, we found that with a

higher protection rate there was little difference in the

effectiveness of different lengths of surveys but, for a lower

protection rate, there was a larger difference. When

protection rates were low and surveying proceeded over a

Figure 4 The effectiveness of different sur-

vey periods for protecting proteas at the end

of 20-year simulations in relation to different

rates of protection and habitat loss. Existing

protected areas in the study region were

initially included as protected before the

simulations were commenced. We measured

both (a) representation of proteas in pro-

tected areas and (b) retention of proteas in

the landscape. We based both measurements

on distributional data using lower quartile

across all 381 proteas and 20 replicate

simulations. We doubled and halved rates

of protection (a, b) and habitat loss (c, d).

Survey rate was 3000 plots per annum. For

scenarios with no survey data (zero survey

period), we used a habitat map to guide

selection of notional protected areas.



longer period, protea habitat was more vulnerable to

clearing. We also found that the rate of habitat loss had a

notable influence on retention. When habitat loss rates were

low, intuitively, the length of surveying mattered less

because fewer occurrences of proteas were lost while

surveying proceeded. When habitat loss rates were high and

survey periods long, more protea occurrences were exposed

to clearing, resulting in lower retention of proteas in the

landscape.

The habitat map did not perform particularly well in

representing proteas in protected areas compared to survey

data. The situation was similar for retention except when the

survey period was long or protection rate was high. Using a

habitat map alone outperformed actual data collection when

measuring retention after a survey period of around 4 years.

When the protected rate was high, the habitat map almost

certainly did well due to the more rapid protection of protea

habitat and the prevention of habitat loss. The general poor

performance of the habitat map as a surrogate for proteas

was probably due to it being developed using broad-scale

environmental variables. These variables probably did not

differentiate the fine-scale variation in habitat, local ende-

mism and rapid spatial turnover typical of Proteaceae in this

biome (McDonald et al. 1996).

We used two measures comparing the performance of

different amounts of data collection, representation in

protected areas and retention in the landscape. Represen-

tation in the protected areas identified in this analysis is an

informative measure of the relative effectiveness of conser-

vation decisions. Representation in protected areas will

remain for many species an important determinant of their

persistence in the long term. Often a protected area may be

the only intact habitat in the landscape. For example, in the

region surrounding Cape Town, nearly all vegetation has

been cleared except for within the protected areas.

Retention, however, is a more comprehensive indicator of

the likely persistence of species as a reduction in a proteas

distribution regardless of whether they are protected is likely

to reduce its long-term persistence.

We made several assumptions in our land-use model and

in our survey rates and patterns. First, we used past

protection rates to estimate future ones. This region is a

priority for global conservation (Myers et al. 2000). Hence,

in the last decade there has been a concerted conservation

effort and an increase in the protection rate, and it seems

likely that this rate will continue. Similarly the habitat loss

rate was based on past loss of native vegetation and is an

uncertain indicator of future rates. Given the complexities

of accurately predicting patterns of future land-use including

both protection and habitat loss (Lambin et al. 2001),

improving upon this information is beyond the scope of this

study. We also assumed that protection could occur

throughout the region. In reality there are many complex

impediments to successful implementation but many of

these considerations were also outside the scope of this

study (for a good discussion of implementation impedi-

ments, see Knight et al. 2006). We based our survey rates on

the collection of survey data for The Protea Atlas and

explored the sensitivity of these rates on the results.

Surprisingly, there was little difference between survey

rates. We assumed surveying occurred in a random pattern.

This may have overestimated effectiveness to some extent

Figure 5 The effectiveness of different survey periods compared

with a habitat map for protecting proteas at the end of 20-year

simulations. Existing protected areas in the study region

were initially included as protected before the simulations were

commenced. Different periods before implementation were

evaluated comparing survey data to the habitat map to guide the

selection of new protected areas. Time before implementation was

the time between the start of the simulation and the implemen-

tation of 10 years of protection. We measured both (a) represen-

tation of proteas in protected areas and (b) retention of proteas in

the landscape. We based both measurements on distributional data

using lower quartiles across all 381 proteas and 20 replicate

simulations. Survey rate was 3000 plots per annum.



as random surveying might better detect the variation in

protea distributions than patchy and biased data collection,

which is more likely to occur in reality.

Data on the distribution of species have many other uses

than locating protected areas. For example, The Protea Atlas

is used for assessing IUCN Red Data List status, setting

management priorities for rarer species and designing

management plans for species, so that more sampling may

be required than only for conservation planning. Some of

this increased cost may therefore be cost-effective for

conservation management.

It is difficult to predict how general our results might be

for other regions until similar studies are repeated in other

regions. The Fynbos is a region characterized by high

species and beta diversity (McDonald et al. 1996). If only a

few years of surveying results in effective conservation

decisions on where to located protected areas in this region,

it is likely that other regions with similar or lower diversity

might correspondingly require only a relatively short survey

period. However, as demonstrated with the application of

our sensitivity analysis, landscape dynamics has an important

influence on the required length of surveying. Landscape

variables such as the extent of existing protected areas and

habitat loss rates should also be considered.

There has been a lack of critical testing of the value of data

investment and more research required on how long to

collect data before implementing conservation actions

(Gerber et al. 2005). For simplicity we simulated surveys

occurring before the implementation of new protected areas.

Realistically, however, conservation organizations would

collect data and implement conservation actions simulta-

neously. While it is unknown how this type of scenario would

alter our results, it is likely that it would lead to more efficient

outcomes, in terms of representation and retention, due to a

reduction in the area vulnerable to habitat loss. Furthermore,

we only included one type of data. Increasingly it is

recognized that, for effective conservation planning, multi-

faceted data sets are required, including those describing

implementation opportunities and constraints (Knight et al.

2006). In addition, protected areas are just one conservation

action and normally conservation organizations would apply

a suite of actions. New research is revealing novel methods

of optimizing multiple actions in multiple places and

incorporating into the planning process the constraints of

people, time and money (Murdoch et al. 2007; Wilson et al.

2007; Underwood et al. 2008). When all these factors are

considered, the trade-off between acting and learning is more

complex than we have indicated here. Any given action in

time and space is not only traded-off with all other actions

but also against the value of delaying an action to learn about

where, when and how actions should occur. Further research

on the trade-offs between the costs and benefits of different

actions, different types of data and different allocations

between learning and doing should ultimately lead to more

effective conservation decisions.

A C K N O W L E D G E M E N T S

We are grateful to Matt Watts for his help with data

processing. We thank Mandy Lombard, Belinda Reyers and

the South African Biodiversity Institute (SANBI) for data.

H.S.G. was supported by a University of Queensland

postgraduate award and an Environmental Futures Network

(ARC) travel grant. A.M. was supported by the Academy of

Finland, project 1206883. H.P.P. and K.A.W were sup-

ported by the Australian Research Council and the

Commonwealth Environmental Research Fund. We thank

Bob Pressey for useful comments that improved this

manuscript.

R E F E R E N C E S

Arponen, A., Heikkinen, R.K., Thomas, C.D. & Moilanen, A.

(2005). The Value of Biodiversity in Reserve Selection:

Representation, Species Weighting, and Benefit Functions.

Conservation Biology, 19, 2009–2014.

Balmford, A. & Gaston, K.J. (1999). Why biodiversity surveys are

good value. Nature, 398, 204–205.

Biggs, R. & Scholes, R.J. (2002). Land-cover changes in South

Africa 1911–1993. S. Afr. J. Sci., 98, 420–424.

Bode, M., Wilson, K.A., Brooks, T.M., Turner, W.R., Mittermeier,

R.A., McBride, M.F. et al. (2008). Cost-effective global conser-

vation spending is robust to taxonomic group. Proc. Natl Acad.

Sci. USA, 105, 6498–6501.

Costello, C. & Polasky, S. (2004). Dynamic reserve site selection.

Resour. Energy Econ., 26, 157–174.

Cowling, R.M. & Heijnis, C.E. (2001). The identification of Broad

Habitat Units as biodiversity entities for systematic conservation

planning in the Cape Floristic Region. S. Afr. J. Bot., 67, 15–38.

Cowling, R.M. & Pressey, R.L. (2003). Introduction to systematic

conservation planning in the Cape Floristic Region. Biol. Conserv.,

112, 1–13.

Cowling, R.M., Pressey, R.L., Lombard, A.T., Heijnis, C.E., Rich-

ardson, D.M. & Cole, N. (1999). Framework for a conservation

plan for the Cape Floristic Region. University of Cape Town

Institute for Plant Conservation.

Elith, J., Graham, C.H., Anderson, R.P., Dudik, M., Ferrier, S.,

Guisan, A., Hijmans, R.J., Huettmann, F., Leathwick, J.R.,

Lehmann, A., Li, J., Lohmann, L.G., Loiselle, B.A., Manion, G.,

Moritz, C., Nakamura, M., Nakazawa, Y., McC, J., Overton, M.,

Townsend Peterson, A., Phillips, S.J., Richardson, K., Scachetti-

Pereira, R., Schapire, R.E., Soberon, J., Williams, S., Wisz, M.S.

and Zimmermann, N.E. (2006). Novel methods improve pre-

diction of species’ distributions from occurrence data. Ecography,

29, 129–151.

Ferrier, S. (2002). Mapping spatial pattern in biodiversity for

regional conservation planning: where to from here? Syst. Biol.,

51, 331–363.

Forshaw, N. (1998). Protea Atlas Project. Available from http://

protea.worldonline.co.za accessed January 19, 2006.



Freitag, S. & van Jaarsveld, A.S. (1998). Sensitivity of selection

procedures for priority conservation areas to survey extent,

survey intensity and taxonomic knowledge. Proc. R. Soc. Lond., B,

Biol. Sci., 265, 1475–1482.

Gardner, T.A., Barlow, J., Araujo, I.S., Avila-Pires, T.C., Bonaldo,

A.B., Costa, J.E. et al. (2008). The cost-effectiveness of biodi-

versity surveys in tropical forests. Ecol. Lett., 11, 139–150.

Gaston, K.J. & Rodrigues, A.S.L. (2003). Reserve selection in re-

gions with poor biological data. Conserv. Biol., 17, 188–195.

Gerber, L.R., Beger, M., McCarthy, M.A. & Possingham, H.P.

(2005). A theory for optimal monitoring of marine reserves.

Ecol. Lett., 8, 829–837.

Gladstone, W. & Davis, J. (2003). Reduced survey intensity and its

consequences for marine reserve selection. Biodivers. Conserv., 12,

1525–1536.

Grand, J., Cummings, M.P., Rebelo, T.G., Ricketts, T.H. & Neel,

M.C. (2007). Biased data reduce efficiency and effectiveness of

conservation reserve networks. Ecol. Lett., 10, 364–374.

Grantham, H.S., Moilanen, A., Wilson, K.A., Pressey, R.L., Rebelo,

T.G. & Possingham, H.P. (2008). Diminishing return on

investment for biodiversity data in conservation planning. Con-

serv. Lett., 1, 190–198.

Hill, D., Fasham, M., Tucker, G., Shewry, M. & Shaw, P. (2005).

Handbook of Biodiversity Methods: Survey, Evaluation and Monitoring.

Cambridge University Press, Cambridge.

Knight, A.T., Cowling, R.M. & Campbell, B.M. (2006). An oper-

ational model for implementing conservation action. Conserv.

Biol., 20, 408–419.

Lambin, E.F., Turner, B.L., Geist, H.J., Agbola, S.B., Angelsen, A.,

Bruce, J.W. et al. (2001). The causes of land-use and land-cover

change: moving beyond the myths. Glob. Environ. Change, 11,

261–269.

McDonald, D.J., Cowling, R.M. & Boucher, C. (1996). Vegetation-

environment relationships on a species-rich coastal mountain

range in the fynbos biome (South Africa). Vegetatio, 123, 165–

182.

Meir, E., Andelman, S. & Possingham, H.P. (2004). Does con-

servation planning matter in a dynamic and uncertain world?

Ecol. Lett., 7, 615–622.

Moilanen, A. & Cabeza, M. (2007). Accounting for habitat loss

rates in sequential reserve selection: Simple methods for large

problems. Biol. Conserv., 136, 470–482.

Mucina, L. & Rutherford, M. (2006). The Vegetation of South Africa,

Lesotho and Swaziland. SANBI, Pretoria.

Murdoch, W., Polasky, S., Wilson, K.A., Possingham, H.P., Kare-

iva, P. & Shaw, R. (2007). Maximizing return on investment in

conservation. Biol. Conserv., 139, 375–388.

Myers, N., Mittermeier, R.A., Mittermeier, C.G., da Fonseca,

G.A.B. & Kent, J. (2000). Biodiversity hotspots for conservation

priorities. Nature, 403, 853–858.

Pearson, R.G., Raxworthy, C.J., Nakamura, M. & Townsend Pet-

erson, A. (2007). Predicting species distributions from small

numbers of occurrence records: a test case using cryptic geckos

in Madagascar. J. Biogeogr., 34, 102–117.

Perhans, K., Kindstrand, C., Boman, M., DjupstrOM, L.B., Gu-

stafsson, L., Mattsson, L. et al. (2008). Conservation goals and

the relative importance of costs and benefits in reserve selection.

Conserv. Biol., 22, 1331–1339.

Phillips, S.J., Anderson, R.P. & Schapire, R.E. (2006). Maximum

entropy modeling of species geographic distributions. Ecol.

Model., 190, 231–259.

Possingham, H.P., Grantham, H. & Rondinini, C. (2007). How can

you conserve species that haven�t been found? J. Biogeogr., 34,

758–759.

Pressey, R.L., Watts, M.E. & Barret, T.W. (2004). Is maximising

protection the same as minimizing loss? Efficiency and retention

as alternative measures of the effectiveness of proposed reserves

Ecol. Lett., 7, 1035–1046.

Reddy, S. & Davalos, L.M. (2003). Geographical sampling bias and

its implications for conservation priorities in Africa. J. Biogeogr.,

30, 1719–1727.

Reyers, B., Rouget, M., Jonas, Z., Cowling, R.M., Driver, A., Maze,

K. et al. (2007). Developing products for conservation decision-

making: lessons from a spatial biodiversity assessment for South

Africa. Divers. Distrib., 13, 608–619.

Rondinini, C., Wilson, K.A., Boitani, L., Grantham, H. & Poss-

ingham, H.P. (2006). Tradeoffs of different types of species

occurrence data for use in systematic conservation planning.

Ecol. Lett., 9, 1136–1145.

Underwood, E.C., Shaw, M.R., Wilson, K.A., Kareiva, P., Klaus-

meyer, K.R., McBride, M.F. et al. (2008). Protecting biodiversity

when money matters: maximizing return on investment. PLoS

ONE, 3, e1515.

Wilson, K.A., Underwood, E.C., Morrison, S.A., Klausmeyer,

K.R., Murdoch, W.W., Reyers, B. et al. (2007). Conserving

biodiversity efficiently: what to do, where, and when. PLoS

Biol., 5, e223.

Editor, Tim Benton

Manuscript received 14 December 2008

First decision made 14 January 2009

Manuscript accepted 28 January 2009



Delaying conservation actions for improved knowledge: how long should we wait

Documents

Transcript of Delaying conservation actions for improved knowledge: how long should we wait