Ken Norris, Centre for Wildlife Assessment and Conservation, School of Animal and Microbial Sciences, University of Reading, Whiteknights, PO Box 228, Reading RG6 6AJ, UK (fax +44 01189310180; e-mail email@example.com).
1The management of threatened species is an important practical way in which conservationists can intervene in the extinction process and reduce the loss of biodiversity. Understanding the causes of population declines (past, present and future) is pivotal to designing effective practical management. This is the declining-population paradigm identified by Caughley.
2There are three broad classes of ecological tool used by conservationists to guide management decisions for threatened species: statistical models of habitat use, demographic models and behaviour-based models. Each of these is described here, illustrated with a case study and evaluated critically in terms of its practical application.
3These tools are fundamentally different. Statistical models of habitat use and demographic models both use descriptions of patterns in abundance and demography, in relation to a range of factors, to inform management decisions. In contrast, behaviour-based models describe the evolutionary processes underlying these patterns, and derive such patterns from the strategies employed by individuals when competing for resources under a specific set of environmental conditions.
4Statistical models of habitat use and demographic models have been used successfully to make management recommendations for declining populations. To do this, assumptions are made about population growth or vital rates that will apply when environmental conditions are restored, based on either past data collected under favourable environmental conditions or estimates of these parameters when the agent of decline is removed. As a result, they can only be used to make reliable quantitative predictions about future environments when a comparable environment has been experienced by the population of interest in the past.
5Many future changes in the environment driven by management will not have been experienced by a population in the past. Under these circumstances, vital rates and their relationship with population density will change in the future in a way that is not predictable from past patterns. Reliable quantitative predictions about population-level responses then need to be based on an explicit consideration of the evolutionary processes operating at the individual level.
6Synthesis and applications. It is argued that evolutionary theory underpins Caughley's declining-population paradigm, and that it needs to become much more widely used within mainstream conservation biology. This will help conservationists examine critically the reliability of the tools they have traditionally used to aid management decision-making. It will also give them access to alternative tools, particularly when predictions are required for changes in the environment that have not been experienced by a population in the past.
In its most recent assessment of the status of the world's bird populations, Birdlife International estimated that 12% of 9946 species should be considered threatened with extinction (Stattersfield & Capper 2000). While of immense concern to many people, the situation for birds is much less severe than for other vertebrate taxa (e.g. fish), certain invertebrate groups and plants (http://www.redlist.org). For many other taxa, it is doubtful whether comprehensive data will ever be available that permits a quantitative assessment of endangerment.
Against this background, decisions taken by conservationists concerning how to manage threatened species are clearly pivotal to reducing extinction risk and the loss of biodiversity. But, how should we decide what is appropriate management? Should each threatened species be considered on a case-by-case basis, or are there general ecological principles that underpin all management issues? In his seminal review, Caughley (1994) argued that conservation biology has two main threads: a small-population paradigm that attempts to understand how low levels of abundance influence population persistence, and a declining-population paradigm that attempts to understand how and why abundance reaches critically low levels, and to use such insights to design remedial management. He went on to argue that the former is characterized by elegant theory that has had little practical impact, whereas the latter lacks a theoretical framework but has provided practical benefits. This distinction has been criticized because both paradigms are often important in saving threatened species from extinction (Hedrick et al. 1996; Asquith 2001). Furthermore, recent work has begun to reveal the importance of the small-population paradigm in understanding the population biology of wild populations (Keller & Waller 2002) and population restoration (Madsen et al. 1999) (both examples deal with inbreeding effects). Despite these more recent insights, however, the declining-population paradigm has played and continues to play a pivotal role in understanding anthropogenic effects on wild populations, and in the design of management for threatened species (Caughley 1994; Green 1995, 2002).
According to Caughley (1994) the declining-population paradigm encompasses two main areas of theory: the causes of extinction and the means by which the causes of decline are diagnosed. While our understanding of variation in extinction rates between taxa and the ecological factors that influence this variation is still relatively poor (May 1999), recent work has begun to address this issue (Bennett & Owens 1997; Owens & Bennett 2000; Purvis et al. 2000; Reed & Shine 2002). Furthermore, the exposure of vulnerable taxa to anthropogenic agents of decline has been considered as a means of identifying conservation priorities (Balmford 1996). Nevertheless, the practical benefits of understanding the ecology of extinction in relation to anthropogenic effects have still to be realized.
In contrast, diagnosing the causes of population decline is the principle component of virtually all projects aimed at saving threatened species from extinction. In very general terms, a population decline can be characterized into three phases within which conservationists can intervene (Fig. 1). It is becoming increasingly important that conservationists are able to quantify the likelihood of a population declining in the future in the face of specific changes in the environment (phase 1; Fig. 1). Historically, the role of conservationists has been concerned with populations in the process of declining (phase 2; Fig. 1) or that have stabilized at low abundance levels following a decline (phase 3; Fig. 1). In these cases, halting the decline, increasing abundance and thereby improving population persistence are the primary aims of management. For populations that continue to decline to critically low levels of abundance, the urgency for action and a lack of data often preclude a detailed analysis of the causes of endangerment. In such cases, maintaining the population through the bottleneck becomes the primary aim of management in the short term. However, in the longer term even a rudimentary diagnosis of putative causes of the decline might be necessary in order to restore populations in the wild (e.g. Mauritius kestrels Falco punctatus L., Jones et al. 1995; Lord Howe's woodhen Tricholimnas sylvestris Sclater, Miller & Mullette 1985).
Caughley (1994) argued that the declining-population paradigm is relevant to most problems of conservation, but is urgently in need of more theory. Population decline (a negative, deterministic trend in population size over time) is a pattern not a process. The process driving all population declines is the way the life-history and behavioural strategies of individuals in the population are affected by and respond to environmental change caused by man. Life-history and behavioural strategies in a population are the outcome of evolutionary processes that depend on the fitness of particular strategies under prevailing environmental conditions (Stearns 1992; Krebs & Davies 1997). When humans alter the environment, the fitness of existing strategies changes. If fitness is drastically reduced, and the strategies are either not sufficiently ‘plastic’ to respond to the environmental change or an evolutionary response to the altered environment is slow relative to the rate of environmental change, the population will decline. It is for this reason, for example, that bird species with ‘slow’ life histories (high survival rates, low reproductive rates) are vulnerable to the introduction of alien predators that dramatically raise mortality rates (Owens & Bennett 2000). Evolutionary theory therefore underpins the declining-population paradigm. Under what circumstances (if any) is it necessary to bring evolutionary theory explicitly into conservation planning?
To develop this debate my review has two primary components. First, I critically review the ecological tools used by conservationists to design management for threatened species. All such tools are being used to make qualitative and quantitative predictions about future levels of abundance, population growth and persistence in order to identify appropriate management options. For the purposes of this review, I recognize three broad classes of predictive tool used in this way: statistical models of habitat use, demographic models and behaviour-based models. For each of these, I describe each technique, illustrate its use in detail using a case study then examine critically its application.
Secondly, I examine critically the reliability of each ecological tool in making predictions about population behaviour (abundance, growth and persistence) under future environmental conditions. To do this, I distinguish two different types of future environment. The first concerns the recreation of a future environment that is comparable with that experienced by the population in the recent past prior to its decline in abundance. The second concerns the creation of an entirely novel future environment. Novelty in this sense is defined as a unique combination of intrinsic and extrinsic factors not experienced by the population, or at least not in the recent past when the population was the subject of an ecological study.
In structuring the review around ecological models, I do not wish to imply that models are the only means of addressing population decline and its management (for a discussion see Peery et al. 2004). For example, an experimental approach is sometimes appropriate, and practical constraints can limit the use of models (time, money, data). However, models have an important role to play in testing hypotheses concerning (i) the cause(s) of decline and (ii) the response of populations to future changes in the environment (some of which may be driven by management), particularly when dealing with large (for humans) spatial scales. Furthermore, this role has increased dramatically over the last decade (Beissinger & McCulloch 2002). For these reasons, a consideration of the ‘ecological toolbox’ seems both timely and important.
Tools for managing threatened species
statistical models of habitat use
Statistical models of habitat use describe patterns in the distribution of individuals across habitat types. In this sense, habitat might include vegetation characteristics, environmental characteristics (e.g. altitude, temperature), predation, disease, competitors and human activities. There are essentially two types of statistical model of habitat use that guide management decisions. The first is widely applied and involves modelling a population statistic, such as presence/absence or population density, as the response variable in relation to a range of predictor variables describing habitat characteristics (Fig. 2a), or comparing habitat actually used with that available in the environment (Cowley et al. 2000; Burrow et al. 2001; Lane, Alonso & Martin 2001; Gil-Sanchez & Alba-Tercedor 2002; Muchai, Lens & Bennun 2002; Russo, Jones & Migliozzi 2002). The model provides a temporal snapshot of how the population varies spatially in relation to variables that are considered potentially important. Some inference is then made using the model about how management decisions might affect the population. For example, if an animal is more likely to be found in habitat A than habitat B, then conservationists might argue that management should be targeted to maintain or increase the available area of habitat A. Over recent years, a number of papers have been published that examine the construction and testing of this type of statistical model (Augustin, Mugglestone & Buckland 1996; Fielding & Bell 1997; Manel et al. 1999; see also above empirical examples).
The second type of statistical model is similar to these habitat-use models, but involves using a measure of population change as the response variable. For example, sites that had retained (1) or lost (0) individuals might be used to quantify population decline (Stowe et al. 1993). Temporal changes in population density can be used in a comparable way (Norris et al. 1998). Such response variables can then be modelled in relation to a range of candidate predictor variables based on current data or, if comparable historical data are available, population change can be modelled directly as a function of temporal changes in potentially important correlates (Fig. 2b,c). This method, usually termed the comparative method (Green 1995, 1999, 2002), relies on the fact that when a population declines there is often spatial variation in the rate of the decline that can be used to identify putative causes of the decline.
Case study: corncrakes in the UK
The corncrake Crex crex L. is a summer visitor to northern Eurasia wintering in south-eastern Africa. In the UK, the population has declined for more than a century and is now confined largely to the northern and western isles of Scotland and the west of Ireland (Stowe et al. 1993). During the breeding season corncrakes are strongly associated with agricultural grassland managed for the production of hay or silage. Initial studies suggested that the decline in corncrake numbers between 1850 and 1940 varied in different parts of the UK and was associated with the introduction of mechanized hay production (Green 1995).
While the initial work suggested that changes in farming practices might be responsible for the declines in different areas, the data could not exclude other contributory factors such as habitat loss. Neither could it reveal the specific mechanism via which mechanization might have precipitated population declines, because a number of changes occurred contemporaneously, for example mechanization meant that hay meadows were cut earlier and more frequently as well as mechanically. For these reasons, a number of statistical models was constructed to explore the putative agents of decline in more detail. Stowe et al. (1993) identified sites that had been occupied by corncrakes during censuses conducted in 1978–79 and that had retained (1) or lost (0) singing males by 1988. Next, they used logistic regression to model the relationship between this response variable and a range of habitat variables that described the vegetation cover at each site collected in 1988. The statistical modelling showed that sites that continued to be occupied were characterized by the presence of tall vegetation and hay meadows. Green & Stowe (1993) developed this work further to show that sites that lost singing male corncrakes experienced a decrease in habitat suitability [as determined by Stowe et al.'s (1993) logistic regression model] between the 1978–79 and 1988 censuses, whereas sites that retained singing males maintained suitable habitat over this time. Finally, Green (1996) examined factors correlated with population density and showed that density was highest in areas with tall vegetation in spring and summer and that were mown in late July or later.
The statistical models strongly suggested that agricultural practices were the most likely cause of the decline, resulting in the loss of suitable habitat and inappropriate management of the suitable habitat that remained. Further work established the mechanism and demographic consequences of grassland management: mowing fields early in the season reduced breeding success due mainly to increased mortality among nests and chicks (Green et al. 1997; Tyler, Green & Casey 1998). A simulation model was used to estimate the impact of mowing practices on the birth rate, and a simple deterministic model of population growth showed that the expected demographic response to changes in farming practices were sufficient to explain the observed timing and rate of regional population declines in corncrakes (Green et al. 1997). As a result of these studies, management recommendations were made that aimed to reduce the impact of mowing operations on breeding success, and populations in some areas have subsequently started to recover (Green & Gibbons 2000).
The corncrake case study illustrates the important role statistical models of habitat use can play in diagnosing the cause of population decline. Within this framework, the role of statistical models is to provide a means of identifying putative agents of decline that then form the basis of further hypothesis testing (Caughley 1994; Green 1995, 2002). When taken in isolation, the results of the statistical modelling are a poor guide to management decisions for two main reasons.
First, because the models are correlative their interpretation strongly depends on the candidate predictor variables originally considered for inclusion in the modelling process. While this problem is widely recognized, it is nevertheless critical and sometimes very difficult to tackle. For example, consider the endangered remnant population of a formerly widespread species. A habitat-use model is likely to reveal disproportionate use of particular habitats, but as Caughley (1994) points out this process could reveal habitats least favourable to the agent of the decline, rather than habitats favourable to the species in question. It is important, therefore, that candidate predictor variables cover a plausible but broad range of explanations for the decline.
Secondly, spatial variation in the timing and rate of population declines can be generated by a buffer effect (Green 1995). The buffer effect describes patterns of density-dependent habitat use by individuals: poor-quality sites (in terms of individual fitness) become increasingly used as total population size increases due to competition for space in the good-quality sites. Consider a population that breeds in a series of sites that vary in quality (in terms of breeding success), and individuals are distributed between these sites according to the ideal despotic distribution (i.e. individuals occupy exclusive breeding territories; Fretwell & Lucas 1970). The population experiences a significant increase in mortality during the non-breeding period as a result of anthropogenic factors that act in a density-independent way, and abundance starts to decline. As the total population declines, the decline starts first in the poor-quality breeding sites because these are only occupied at relatively high population sizes. Constructing a habitat-use model based on breeding season habitat data would reveal significant habitat correlates of the variation in population decline between sites, but this spatial variation would be unrelated to the actual cause of the decline.
As a result of these problems, statistical models are used to identify plausible causes of population decline that can then be tested critically as a working hypothesis (Caughley 1994; Green 1995, 2002). Testing would ideally take the form of replicated experiments, but such replicated population-level experiments have only rarely been attempted in ecology (Hudson, Dobson & Newborn 1998) and constraints are often even more severe in conservation due to the lack of replicate populations, urgency for action and a shortage of resources. An alternative approach is to understand the demographic mechanism driving the population decline (Green 1995, 1999, 2002; Caswell 2001), which itself may be more amenable to experiment, or at least spatial variation in the putative agent of decline might be used as pseudo-experimental treatments. The case study on corncrakes illustrates how this process works.
While these approaches to hypothesis testing have produced practical conservation benefits, both are vulnerable to misdiagnosis due to buffer effects. Consider the above buffer effect example. Detailed autecological studies into demographic mechanisms would reveal poorer breeding success in those sites that started declining in abundance first, and initially declined most rapidly. In this case, if there were little data available on survival rates then conservation management would almost inevitably be designed to manage habitat in a way that improved breeding success, even though this was not the demographic mechanism causing the decline. This management decision may, depending on the relative effects of management and the anthropogenic factor on population growth, actually arrest the population decline, but it may not.
There are two ways to check for spurious habitat relationships driven by the buffer effect in a declining population. First, determine whether the demographic differences between habitat areas are sufficient to cause the observed decline in abundance (Green 2002, box 7·1). Secondly, model statistically spatial variation in the rate of population decline as a function of spatial variation in habitat change (Green 1996; Norris et al. 1998). This provides a more powerful test of causation between habitat and spatial variation in the rate of the decline.
Such models can be used in the management of threatened species in three main ways. First, models can be used directly in threat assessment by estimating the probability that a population will decline to a particular abundance level over a specified future time period. The World Conservation Union (IUCN) has incorporated such quantitative predictions from demographic models into criteria designed to identify threatened species (Mace & Lande 1991). In management terms, this might be considered as an estimate of the impact on population persistence of maintaining the status quo. Secondly, models can be used to identify the life-history stage(s) that has (have) the strongest relative effect on population growth in order to target (at least in principle) management measures efficiently to improve future population growth using sensitivity or elasticity analysis (reviewed by Benton & Grant 1999). Thirdly, models can be used to assess the effectiveness (in terms of reducing extinction risk, increasing population growth or population size) of a range of management options in order to guide management decision making (reviewed by Beissinger & Westphal 1998; Beissinger & McCulloch 2002; Reed et al. 2002). The latter two applications are frequently termed population viability analysis (PVA), although a PVA does not always involve the use of a demographic model (reviewed by Reed et al. 2002).
Case study: loggerhead turtles in the USA
The conservation of marine turtles world-wide is urgent: of seven species, five are listed as endangered or critically endangered (http://www.mcsuk.org). Traditionally, management efforts have focused on ‘headstarting’, which involves various measures to protect nests in situ or the captive rearing of hatchlings from eggs harvested from the wild. The objective of management is to maximize the birth rate and/or first-year survival rate. Although headstarting produces demonstrable ecological benefits, Crouse, Crowder & Caswell (1987) used elasticity analysis to examine whether management was being focused on the life-history stage that had the greatest impact on population growth in a population of loggerhead turtles Caretta caretta L. that nest on beaches in south-eastern USA. To do this, they constructed a demographic model that recognized seven life-history stages, and used the best available demographic data to parameterize the model. The elasticity analysis showed that population growth was most sensitive to small changes in the survival rate of juveniles and subadults, and that the birth rate had a relatively low elasticity compared with the free-living marine life-history stages. It had been recognized for some time that turtles experience mortality from entanglement in fishing nets, so Crouse, Crowder & Caswell (1987) argued that management measures to reduce this source of mortality should be implemented because of the potential impact on population growth. Turtle excluder devices (TED) allow turtles to escape from fishing nets without getting entangled, and Crowder et al. (1994) showed, using a similar demographic model, that the use of TED on fishing nets could promote significant population growth. There is evidence that the introduction of TED in the late-1980s was associated with a significant decline in the stranding of dead turtles on beaches (Crowder et al. 1995). However, the demographic model was also used to compare the relative effectiveness of headstarting and TED on population growth (Grand & Beissinger 1997). This work showed that the use of TED in isolation could only generate positive population growth when nest survival rates exceeded 32–43%, otherwise both nest protection and TED were necessary to maintain positive population growth rates. In areas where nest survival rates are typically low, the demographic modelling suggests a suite of management measures is likely to be necessary to ensure population persistence (Grand & Beissinger 1997).
The use of demographic models in conservation has promoted considerable recent debate (Ludwig 1999; Mills, Doak & Wisdom 1999, 2001; Caswell 2000; Wisdom, Mills & Doak 2000; Coulson et al. 2001; Ehrlén, van Groenendael & de Kroon 2001; Brook et al. 2002; Ellner et al. 2002; Reed et al. 2002). Demographic modelling is attractive because it provides an explicit description of population dynamics that can be examined critically; it provides precise predictions about extinction, population growth and population size in relation to management; and the number of software packages available for constructing and running demographic models makes the technique widely available. These potential benefits have led a number of authors to advocate the widespread use of demographic models in the management of threatened species (Carroll et al. 1996; Morris et al. 2002). In contrast, other authors have argued that demographic models should be applied with considerable caution because of uncertainties in their predictions (Coulson et al. 2001; Reed et al. 2002). While the details of this debate are beyond the scope of this review, it centres on the availability of data for constructing a demographic model and the extent to which the data are sufficient to capture the dynamics of the population under future management scenarios.
Although the complexity of demographic models varies considerably between study systems, they have two fundamental data requirements if they are to be applied to management problems. First, demographic data are required to model the current dynamics of the population. Secondly, estimates of the expected demographic response to various management measures are required to predict the future dynamics of the population. How are techniques such as elasticity analysis and PVA used in this respect, and do they provide a reliable guide to appropriate management?
Elasticity analysis involves identifying the life-history stage whose demography has the greatest relative impact on population growth (see above for details; referred to as the ‘key’ vital rate here). The practical limitations of this approach are widely recognized: the analysis does not include an assessment of the ability of management to alter the key vital rate, so it might be impossible to practically realize this population growth potential (for a practical example see Hiraldo et al. 1996). The corollary of this is that management may have the potential to improve population growth significantly by altering a particular vital rate even if this rate has a low elasticity (Ehrlén, van Groenendael & de Kroon 2001). This strongly suggests that a prudent course of action would be to estimate the possible impact on population growth of all plausible management options irrespective of the elasticity of particular vital rates. This realization seriously questions whether elasticity analysis has any practically valuable role to play in the management of threatened species.
PVA is routinely used to quantify the population dynamic consequences of a range of management alternatives. The widespread application of PVA is severely hampered by a chronic lack of data. Green & Hirons (1991) argued that even the most basic demographic data are not available for the majority of endangered birds world-wide. In a recent review, Morris et al. (2002) showed that less than 25% of recovery plans for endangered species in the USA proposed to collect ‘all the necessary data’ for undertaking a PVA based on a demographic model, although no formal definition of what constitutes a complete data set is given. These constraints raise two important questions. First, if high-quality data are available are predictions based on a PVA reliable? Secondly, how can uncertainty be incorporated in management decision making using PVA?
In a recent paper, Brook et al. (2000) used 21 long-term data sets on vertebrate populations to investigate the reliability of PVA, both in terms of specific predictions generated by models and the consistency in predictions generated using different software. A demographic model was constructed using the first half of the data set for each population and then used to predict the expected dynamics for the second half, which could be compared with observed dynamics. Observed and predicted patterns were similar, leading Brook et al. (2000) to conclude, ‘PVA is a valid and sufficiently accurate tool for categorizing and managing endangered species’. This conclusion may be overly optimistic because observed and predicted dynamics would be similar if the fundamental dynamics of each population changed little between the two time periods (i.e. that used to construct a model and that used to test it). This sheds little light on the reliability of PVA when used to make predictions in relation to changes in demography or abundance caused by management. This is because the objectives of management are to alter population dynamics fundamentally in order to make them more favourable to abundance, population growth or population persistence in the future. As Coulson et al. (2001) point out, a PVA can only be reliable if the distributions of vital rates and population growth remain stationary in the future, or if any changes can be accurately predicted. As the objective of management is to promote beneficial environmental change, the key issue is the extent to which we understand how population processes are likely to respond to management. This is an important general issue and is discussed in detail below, but it is important to note that we often have a very poor idea of the quantitative impact of particular management proposals. As a result, certain authors have argued that when data are poor PVA should be ‘treated with extreme caution and possibly even ignored entirely’ (Coulson et al. 2001) or ‘used to focus research on gathering needed data’ (Reed et al. 2002), although this is not a consensus view (for details of this debate see Beissinger & Westphal 1998; Ludwig 1999; Fieberg & Ellner 2000; Brook et al. 2002; Ellner et al. 2002).
One way to deal with uncertainties in the data used in PVA is to perform a sensitivity analysis on model predictions to see whether particular management options consistently produce the desired population response across a range of parameter values used in the model. Despite a range of methods for sensitivity analysis (reviewed by Reed et al. 2002; examples in Drechsler, Burgman & Menkhorst 1998; Drechsler 2000; Cross & Beissinger 2001), its outcome depends crucially on the range of plausible ecological responses to management incorporated into the analysis. If there are no data from the past on the population's response to comparable environmental conditions, then assessing ‘plausible’ responses is subjective and uncertain. Furthermore, as management aims to alter population dynamics in a way that improves population persistence, then a lack of adequate past data will be the rule rather than the exception. If the range of plausible responses is not specified correctly and the actual population response lies outside the range included in the analysis, then the analysis is relatively meaningless. This is particularly worrying given the paucity of information available on processes that have a strong influence on population dynamics and persistence, such as density dependence (Pascual, Kareiva & Hillborn 1997; Runge & Johnson 2002) and catastrophes (Ludwig 1999), let alone how these processes might respond to management in the future (see Future environments below).
Behaviour-based models are fundamentally different to statistical models of habitat use and demographic models because they deal with processes acting at the level of individuals, and attempt to derive population parameters, for example vital rates and abundance, from an understanding of individual decision making (hence the term behaviour-based). In contrast, statistical models of habitat use and demographic models describe the population-level patterns that arise as a consequence of these individual-level processes. Note also that behaviour-based and individual-based models (DeAngelis & Gross 1992) are fundamentally different. The latter describes demographic patterns across individuals in a population and uses these to model population dynamics (Walters, Crowder & Priddy 2002), whereas the former explains the processes that cause these patterns.
Behaviour-based models are derived from the more general approach taken by behavioural ecologists over the past 20 years, developed to understand animal decision making using evolutionary theory (Krebs & Davies 1997). This approach assumes that each animal in a population makes decisions in order to maximize its fitness. For example, it consumes prey types or feeds in particular patches of prey that minimizes its risks of mortality, or breed in locations that maximize its reproductive success. Optimality and game theory can then be used to estimate the evolutionarily stable strategy (ESS) for a population competing to exploit a specified resource (Maynard-Smith 1982; Houston et al. 1988). The ESS describes the behavioural strategies that each individual adopts in the population to maximize its fitness, given that every other individual is attempting to achieve the same goal. Although this approach has been applied to a wide range of behavioural problems, each model of behavioural decision making contains the same standard elements. It describes the (i) resources that individual animals in the population are trying to exploit, for example food and breeding sites; (ii) behavioural strategies that individuals are permitted to adopt, for example territory owner or floater, and the extent to which these are limited by an individual's phenotype; (iii) mechanism of competition, for example territoriality, scramble or contest competition; and (iv) fitness consequences (usually measured as fitness components such as mortality or breeding success) associated with acquiring a certain level of resources, or associated with adopting a particular strategy. Population-level parameters such as mortality and birth rates are then derived from the fitness of individual animals in the population (Fig. 4).
Case study: habitat change and waterbirds in north-west Europe
The conservation of coastal habitats and their wildlife is of global concern. In north-west Europe each year vast numbers of waterbirds (waders, ducks, geese and swans) spend the winter, or migratory stopovers, feeding on intertidal invertebrates in estuaries, making these sites internationally important under various international conservation conventions. Conservation issues affecting these sites are varied, but most revolve around habitat loss and habitat change. When habitat is lost, existing food resources have to support a higher population density than they did prior to habitat loss. This intensifies competition for food resources and could have a detrimental effect on bird populations if the remaining food resources are insufficient to support all or part of the population for the season in which it was normally occupied. Because the impact depends on resource availability, competition for resources and the demographic consequences of this competition, making predictions has been undertaken using behaviour-based models. Models have been developed for a number of different systems (Sutherland & Allport 1994; Percival, Sutherland & Evans 1998; Clark & Butler 1999; Stillman et al. 2000; Gill, Sutherland & Norris 2001). The models vary in detail but each contains a description of (i) spatial and temporal variation in food availability; (ii) the distribution of birds in space and time in response to food availability and the process of competition for resources; and (iii) a measure of the population consequences of competition. This latter effect can be measured in various ways, for example expected life-time reproductive success (Clark & Butler 1999), mortality rates (Stillman et al. 2000) and sustainable population sizes (Sutherland & Allport 1994; Percival, Sutherland & Evans 1998; Gill, Sutherland & Norris 2001). The main differences in detail between models concern the process of competition for food resources and how these impact on different individuals in the population.
These models explicitly link the process of resource competition to population-level consequences. As a result, if changes in resources after habitat loss or change can be predicted accurately, the model can be used to predict how the population statistic of interest is affected by habitat loss or change. For example, oystercatchers Haematopus ostralegus L. are specialist feeders on bivalve molluscs that are also fished commercially. There is considerable concern that fishing could adversely affect oystercatcher populations, so a behaviour-based model was constructed to estimate potential impacts (Stillman et al. 2001). The model describes food availability and how this is affected by fishing; how birds distribute themselves between food patches in relation to food availability and their susceptibility to interference competition (which reduces food intake rates for vulnerable individuals); and how their consumption of food resources affects starvation risk. The model's reliability was checked by comparing predicted and observed mortality rates and the correspondence was good (Stillman et al. 2000). Note the model was built with data from years that did not contribute to mortality rate comparisons during which population density was relatively low, and was used to estimate mortality rates under novel (an increasing range of population density) conditions. The model was then used to examine the impact of a range of future fishery management scenarios (Stillman et al. 2001).
To date, models have been constructed for a very narrow range of systems (mainly waterbird populations) and applied to issues of future habitat loss or change (Sutherland & Allport 1994; Percival, Sutherland & Evans 1998; Stillman et al. 2001). However, there is a growing body of theory that allows models to be constructed for a range of different systems that differ in the resources being exploited and the mechanisms of competition evident in resource exploitation (Sutherland 1996; Kokko & Sutherland 1998). Furthermore, the wider role of these models is being considered (Lima & Zollner 1996; Bradbury et al. 2001) and behaviour-based models involving taxa other than birds are being developed (Smith, Reynolds & Sutherland 2000). This interest is partly because behaviour-based models are regarded as being able to produce accurate predictions outside the range of conditions for which they are parameterized (Norris & Stillman 2002). In this respect, behaviour-based models are, at least in principle, extremely powerful tools for informing management decisions because environmental conditions not experienced by the population in the recent past can be investigated in terms of their impact on demography, abundance, population growth and population persistence.
Nevertheless, as pointed out by Lima & Zollner (1996), a role for behaviour-based models has to be based on something more than a ‘standard of plausibility’. The potential will only be realized provided changes in a model's parameters under a specific future management scenario or environmental change are accurately known. Sometimes it may be sufficient simply to be able to quantify how resource availability might change in the future, and derive population predictions from the consequences such changes have for behavioural decisions (Stillman et al. 2001). However, to provide accurate predictions about environmental change such models need to be able to incorporate the full range of behavioural strategies individuals might adopt in order to determine the ESS for a particular set of environmental conditions. This involves being able to define alternative strategies and quantify the fitness of individuals adopting a particular strategy. It is likely that the range of behavioural strategies actually adopted by individuals in a population will depend on the prevailing environmental conditions. For example, individuals competing for access to breeding territories only adopt strategies that involve delayed reproduction if the availability of high-quality territories becomes limited by habitat quality and/or increasing population density (Komdeur 1992; Kokko & Sutherland 1998). Furthermore, individuals often only use poor-quality habitat at relatively high population sizes (Vickery et al. 1995; Gill et al. 2001). It would be unwise to assume, therefore, that the observed strategy set in a particular population necessarily applies to any future environmental conditions being considered. Furthermore, stochastic events are very important in understanding population persistence (Ludwig 1999) but have not been incorporated into behaviour-based models of populations as yet (Norris & Stillman 2002), even though the role of environmental variance in the evolution of behavioural and life-history strategies has been widely studied (Stearns 1992; Krebs & Davies 1997). These issues mean that there is a clear need for the development of a wider range of theoretical and empirical behaviour-based models, and a critical appraisal of their predictions across a range of environmental conditions.
In broad terms, the management of threatened species encompasses the production of two different types of future environment. The first concerns the recreation of a environment that is comparable to that experienced by the population in the recent past for which some population-level data on demography and abundance are available. The second concerns the creation of an entirely novel environment. Novelty in this sense is defined as a unique combination of intrinsic (to the population) and extrinsic factors not experienced by the population previously, or at least not in the recent past when ecological data on the population of interest were collected. This distinction is useful because it allows us to examine the phases in which conservationists are able to intervene within a declining population (Fig. 1) and ask exactly which type of future environment is being considered and how the different ecological tools perform in making predictions.
Phase 1 intervention asks whether or not a particular population is likely to experience a decline in abundance following a change in the environment. As the environment deteriorates it will alter the mean and variance in important vital rates, and their relationship with population density, in ways that are not necessarily consistent with past patterns. Therefore, it is inevitable that such environmental change tends to produce novel future environments. An intervention during phase 2 identifies the environmental change that is causing the population to decline in abundance, and management is then designed to restore more favourable environmental conditions. That is, management aims to produce environmental conditions that the population experienced in the recent past prior to its decline, and uses population data collected during the decline to identify favourable environmental conditions. Intervention during phase 3 asks how population growth, abundance or persistence in the future will respond to a range of management options that are selected because they are considered to provide more favourable environmental conditions. The objective of management in this case is to alter vital rates (their mean and variance) and their relationship with population density, and so is by definition aimed at producing novel future environmental conditions.
Statistical models of habitat use in declining populations and demographic models have been used in isolation and in combination to make management recommendations for declining populations that are aimed at arresting the decline, restoring more favourable conditions for population growth, and thereby increasing abundance and population persistence in the longer term (phase 2; Fig. 1). To do this, assumptions are made about population growth or vital rates that will apply when environmental conditions are restored, based on either past data collected under favourable environmental conditions (e.g. the corncrake case study) or estimates of these parameters when the agent of decline is removed (e.g. the loggerhead turtle case study). This application has a basis in evolutionary theory. This is because if we assume that the life-history and behavioural strategies of individuals in the population prior to the decline were in a state of evolutionary equilibrium with the prevailing environmental conditions, then conservation management is implicitly attempting to restore the environment associated with that equilibrium in order to ensure population persistence. As a result, spatial and temporal patterns of abundance and demography are implicitly linked to the evolutionary processes underlying them.
When based on adequate data from a declining population, both statistical models of habitat use and demographic models have an important role to play in designing management for threatened species. However, neither tool can produce reliable quantitative predictions for novel future environments. When the environment changes the population response to the change is determined by the strategies employed by individuals in the population competing for resources and the fitness consequences of that competition. This means that the vital rates and their relationship with population density in the future are specific to the nature of the environmental change, and so are unlikely to resemble patterns seen in the population in the past. For example, Sutherland & Dolman (1994) developed a theoretical model based on the ideal free distribution with unequal competitors, and used it to explore the consequences of habitat loss on population decline. The rate of population decline depended on whether high-quality or low-quality habitat was being lost. While intuitive, this occurred because density-dependent mortality was much more severe when good-quality habitat was being lost. That is, the form of the density-dependent function describing how mortality changed with increasing population density was specific to the future environmental change. Furthermore, these alternative future functions were different from that which would have been evident in the population prior to the decline had it been measured. Comparable conclusions can be drawn from theoretical models of other processes of resource competition between individuals and the strategies they adopt (Kokko & Sutherland 1998). Therefore, the degree to which future patterns differ from the past is only quantifiable by explicitly considering the evolutionary processes underlying demography. Without doing this, vital rates included in demographic models of novel future environments should be viewed as a guess of how the population might respond with no firm basis in evolutionary theory.
This is a concern given the increasing role demographic models in particular are playing in making predictions about population behaviour in novel future environments. Empirical assessments of how unreliable demographic models might be in such circumstances are few, but recent work suggests that discrepancies between these approaches in their ability to describe population dynamics can be significant. Stephens et al. (2002) developed a suite of demographic models of an alpine marmot Marmota marmota L. population in southern Germany and compared predictions derived from these models with a behaviour-based model that considered explicitly the behavioural strategies employed by individual animals with their social system. This study highlighted a number of important general issues. First, a spatially explicit individual-based model that ignored the evolutionary dynamics of behavioural strategies predicted different dynamics to all other models because it forced individuals to make maladaptive decisions as it was based solely on patterns of vital rates between individuals. This shows that assumptions about how past patterns in demography might relate to the future can be seriously flawed if they have no basis in evolutionary theory. Secondly, predictions requiring an understanding of transient dynamics would only have been adequate if based on the behaviour-based model. Thirdly, the behaviour-based model allowed the characterization of an important Allee effect. Such effects can have a profound impact on population dynamics, but are notoriously difficult to study in natural populations (Stephens & Sutherland 1999). This highlights the wider issue of the degree to which density-dependent processes are adequately represented in demographic models used for conservation management (Stephens et al. 2002).
Behaviour-based models provide an alternative ecological tool that can, at least in principle, be used for predicting population-level responses to novel environmental change because, by definition, past patterns in vital rates and their relationship with population density will not apply in the future. Furthermore, how they will change in the future can only be quantified by explicitly understanding the evolutionary processes underlying these rates. However, there are clearly outstanding issues in the development and application of behaviour-based models that need to be addressed (see above), which includes a need for a wider range of models (both theoretical and empirical) and a more rigorous critical appraisal of their application to novel environmental conditions. Nevertheless, in my view, the development and application of evolutionary theory in this way to the management of threatened species is a vastly under-utilized resource in mainstream conservation.
Evolutionary theory underpins the declining-population paradigm. Effective conservation needs to be based on a reliable diagnosis of the cause(s) of decline and the use of ecological tools that aid an assessment of how management might assist in population restoration. In evolutionary terms, conservation management implicitly aims to restore the fitness environment associated with the life-history and behavioural strategies employed by individuals in the population prior to the decline. This is often achieved using data on spatial and temporal patterns in abundance (statistical models of habitat use) and demography (demographic models). These tools are adequate when data are available on the population characteristics (population growth, vital rates) associated with that environment because then management decisions are firmly rooted in evolutionary theory, albeit implicitly. These tools are not adequate when environmental change promotes changes in vital rates (means, variances, covariances) and their relationship with population density that are not reflected in past patterns. Predictions using these tools are then, at best, qualitative guesses with no firm basis in evolutionary theory. While qualitative predictions can have practical value, quantitative predictions must be regarded as unreliable under these circumstances (for an example see Brook et al. 1997). In contrast, behaviour-based models, because they deal explicitly with evolutionary processes acting at the individual level and derive population-level parameters from these, can be used to make predictions about population behaviour when a population faces deteriorating (phase 1) or improving (phase 3) environmental conditions that affect demography in ways that are not predictable from past population behaviour. However, to do this reliably, the availability of resources in the future must be predicted with accuracy, and the life-history and behavioural strategies that could be employed in the new environment, together with their fitness consequences, must be known. In my view, therefore, there is a clear dichotomy between the roles of the different ecological tools within the declining-population paradigm.
There is a pressing need for conservation biologists to bring evolutionary theory into their work. This would provide a greater understanding of the potential pitfalls in the ecological tools they routinely use (statistical models of habitat use and demographic models) and would provide opportunities for developing and using alternative tools (behaviour-based models) that have a limited role at present in mainstream conservation. We need to recognize that conservation management concerns managing an appropriate fitness environment for the life-history and behavioural strategies of individuals in a particular population. The tools we use must be appropriate to this goal. That is, we can only make assumptions about how past patterns of abundance and demography relate to future environments provided we consider such assumptions in terms of evolutionary theory. At the very least, we need to begin to examine the potential reliability of existing tools in the face of novel future environmental change by comparing their predictions with models that explicitly incorporate evolutionary theory in making population-level predictions (Stephens et al. 2002).
There is also a challenge here for evolutionary ecologists to become more involved in the application of their subject to conservation management (Lima & Zollner 1996). This is happening in certain areas, for example the development of theoretical tools to understand linkages between evolutionary processes at the individual level with their population dynamic consequences (Sutherland 1996; Kokko & Sutherland 1998), and the application of ecological theory to understanding the vulnerability of species to agents of population decline (Owens & Bennett 2000; Purvis et al. 2000; Reed & Shine 2002). Furthermore, there is a growing recognition that certain management initiatives may have failed in the past because they failed to consider life-history and behavioural strategies adequately (Knight 2001). However, much more widespread involvement is needed. Applied field projects can provide unique opportunities for testing evolutionary ideas (Komdeur 1992), so such involvement is also potentially beneficial to evolutionary ecologists.
In his seminal review, Graeme Caughley (1994) argued that the declining-population paradigm was in need of more theory. In a sense he was wrong. The theory is well established. It just needs to be put to work in conservation biology.
I would like to thank Rhys Green, Bill Sutherland, Barry Brook, Steven Beissinger and two anonymous referees for their very insightful and constructive comments on an earlier draft of the paper.