Comparing habitat configuration strategies for retaining biodiversity under climate change

Authors


Correspondence author. E-mail: karel.mokany@csiro.au

Summary

  1. Establishing new conservation reserves is a key management response to promote the persistence of biodiversity under climate change. Although there are many approaches to designing reserves, quantitatively assessing the performance of alternative habitat configuration strategies in retaining biodiversity has been limited by the lack of suitable modelling frameworks.
  2. Here, we apply a new dynamic macroecological modelling approach to compare the outcomes under climate change for plant biodiversity in Tasmania (all 2051 species) when new conservation reserves are established according to four contrasting reserve design strategies: connectivity; aggregation; representativeness; and a balanced approach.
  3. The most effective reserve design strategy under climate change depended on the specific conservation goal. New reserves focussed on improving representativeness most effectively promoted regional gamma diversity; however, the aggregation and balanced strategies best promoted the mean area of occurrence across all species.
  4. As the modelled level of dispersal increased, the connectivity strategy became relatively less effective, and the aggregation strategy relatively more effective in retaining biodiversity.
  5. Synthesis and applications. Our results demonstrate that adherence to a single habitat configuration strategy, such as connectivity, is unlikely to result in the best outcomes for biodiversity under climate change. The best reserve design strategy under climate change will vary between regions due to unique combinations of attributes and between taxa due to contrasting dispersal abilities. Quantitative assessments, such as ours, are required to identify configurations that will best retain the biodiversity of each region under climate change.

Introduction

Climate change is predicted to have substantial negative impacts on biodiversity for a wide variety of taxa across many regions of the world (Botkin et al. 2007; Bellard et al. 2012). Observed and predicted impacts include the contraction of suitable habitat for species (Huntley et al. 2008), poleward and elevational shifts in occurrence (Randin et al. 2009; Chen et al. 2011), and the disassembly of communities, with potential loss of key interspecific interactions (Harley 2011; Urban, Tewksbury & Sheldon 2012). Importantly, bioclimatic niche modelling predicts that existing conservation reserves are likely to become less suitable for many of the species they currently harbour (Araújo et al. 2011).

One of the primary adaptive management responses we can make to reduce the predicted negative impacts of climate change on biodiversity is to increase the amount of habitat protected in conservation reserves (Peters & Darling 1985; Heller & Zavaleta 2009). However, the creation of new conservation reserves is typically limited by the resources available to purchase and manage them (Naidoo et al. 2006). It is therefore imperative that new reserves are designed, so that they best contribute to retaining biodiversity in a rapidly changing climate (Hannah et al. 2005; Mumby et al. 2011).

Traditionally, assessment of priorities for new conservation reserves focussed on trade-offs between the number and size of new reserves (aggregation; Hodgson et al. 2009), and their representativeness (Margules, Pressey & Williams 2002). While new reserves that better represent the diversity of a region may increase the initial diversity of taxa and environments conserved, Island Biogeography Theory suggests that large contiguous reserves are more likely to retain viable populations over long time periods (Diamond 1975).

With increasing concern over the impacts of climate change, the focus for new conservation reserves has dramatically shifted to increasing the connectivity between reserves through habitat corridors, to enable species to migrate with their climatic niche (Williams et al. 2005; Heller & Zavaleta 2009). Although increased connectivity would intuitively appear to promote the persistence of biodiversity under climate change, there are innate risks with this approach, including the potential spread of exotic species, fire and disease, as well as higher maintenance costs per unit reserve area (Simberloff & Cox 1987). Some conservation scientists have also questioned the biological effectiveness of corridors, arguing that established strategies focussing on reserve aggregation and representativeness are more robust in the face of climate change (Hodgson et al. 2009).

Substantial effort has gone into developing approaches to prioritise areas for new conservation reserves (Pressey et al. 2007; Carroll, Dunk & Moilanen 2010); however, quantitative comparisons of the effectiveness of different reserve design strategies in retaining biodiversity over time are surprisingly rare. One exception is a study applying correlative species distribution modelling for 1200 plant species across Europe under climate change, to compare the effectiveness of six reserve design criteria (Araújo et al. 2004), with selection based on ‘habitat suitability’ outperforming reserve clustering (i.e. aggregation) in retaining habitat for species. Other studies have applied dynamic landscape and metapopulation models to assess the relative benefits of corridors compared with larger reserves (Falcy & Estades 2007) and spatially dynamic versus static protected areas (Rayfield et al. 2008) in maintaining populations for a focal species over time.

One of the primary limitations in quantitatively assessing the effectiveness of alternative habitat configuration strategies has been the lack of suitable ecological modelling frameworks. Purely correlative modelling approaches ignore key ecological processes in predicting change in biodiversity over time (Kearney & Porter 2009), while more mechanistic approaches have focussed on individual species (Keith et al. 2008), in which interspecific interactions are commonly ignored, and the amount of information required for each species limits analysis to only a small number of species (Mokany & Ferrier 2011).

Here, we apply a new dynamic macroecological modelling approach (Mokany et al. 2012) to compare the outcomes for biodiversity under climate change when new conservation reserves are created based on four contrasting reserve design strategies, aimed at maximising: (1) connectivity; (2) aggregation; (3) representativeness; and (4) a balanced approach (Fig. 1). Our macroecological modelling framework incorporates key ecological processes (i.e. dispersal, community assembly) in predicting biodiversity change for large numbers of species over space and time under alternative habitat configurations. We applied our approach to assess the potential benefits of the four broad reserve design strategies in retaining plant biodiversity in Tasmania under climate change.

Figure 1.

Three distinct reserve design strategies may be to maximise connectivity, aggregation or representativeness of the reserve estate. Alternatively, each of these objectives may be achieved in part, through a balanced strategy. In this study, we compared outcomes for biodiversity when new reserves were created under these four contrasting reserve design strategies (Fig. S1, Supporting Information).

Material and methods

Designing new reserves

Our analysis focused on projecting changes in plant biodiversity (2051 species) across Tasmania, Australia (42° 01′ S; 146° 36′ E), at a 250 m grid cell resolution. To compare the capacity of alternative reserve design strategies to retain plant biodiversity under climate change, we randomly selected ten regions of dimensions 40 × 40 km within Tasmania (Fig. 2a). We applied two criteria in selecting regions for analysis: (1) the area of existing reserves must be >1% (16 km2) and <60% (960 km2) of the total area of the region (from CAPAD2008); (2) the region selected must have an area of unreserved habitat >5% (80 km2) of the total area of the region (from TASVEG 2.0). These basic criteria were applied to ensure adequate areas of both reserved and unreserved habitat with which to generate and contrast alternative reserve design strategies.

Figure 2.

(a) The ten 40 × 40 km regions used to analyse alternative reserve configurations in Tasmania, Australia. (b) An example of the 16 km2 of new reserves selected for one region (region 3) based on the four alternative reserve design strategies compared in this study.

We allocated a modest amount of new reserves totalling 1% (16 km2 or 256 grid cells) of the land area of each region based on the four reserve design strategies (Fig. 1: connectivity, aggregation, representativeness, balanced). New reserves were allocated only to areas that are currently unmodified natural habitat, as determined by the TASVEG 2.0 (2009) vegetation classification system. Ecological outcomes were then assessed assuming that the reserve estate provided the only viable habitat. Note that here we consider only the ecological impacts of different reserve design strategies, ignoring differences in the cost of purchasing and managing alternative locations.

For the ‘connectivity’ reserve design strategy, new reserves were allocated as 250-m-(one grid cell) wide corridors connecting existing reserve areas by the shortest distance, with priority given to connecting larger reserves over smaller reserves (Fig. 2b, Fig. S1a,b, Supporting Information). For the ‘aggregation’ reserve design strategy, new reserves were randomly allocated to areas of available habitat abutting the edge of existing reserves (Fig. 2b, Fig. S1a,b, Supporting Information).

For the ‘representativeness’ reserve design strategy, new reserves were allocated to areas that would best improve the current representation of biodiversity in the reserve estate. To quantify the value of each grid cell for improving representativeness, we applied the model of current (2010) compositional dissimilarity for Tasmanian plants (described below) to calculate the predicted summed similarity to the existing reserves (SSR) of each grid cell i (math formula), where sij is the complement of the predicted Sørensen's dissimilarity between cells i and j (i.e. sij = 1 − dij) and hj is the current reserve status of cell j (i.e. 1 = reserved, 0 = not reserved) (Allnutt et al. 2008). Grid cells with the lowest SSR values are predicted to be most poorly represented in the reserve estate, so for each region, we selected those grid cells to be the new reserves under the representativeness strategy (Fig. 2b, Fig. S1a,b, Supporting Information).

For the ‘balanced’ reserve design strategy (Fig. 1), new reserves were allocated to contiguous areas of poorly represented habitat, using the cost–benefit approach of Drielsma, Ferrier & Manion (2007), which assigns a value to each cell based on the habitat value of surrounding cells and their connectivity to the focal cell. Specifically, the current habitat value for each cell j (Hj) was determined by its representativeness using the SSR values above {i.e. Hj = 1 – [SSRj/max(SSR)]}, and the link permeability values between cells i and j (wij) calculated as suggested by Drielsma, Ferrier & Manion (2007) (i.e. wij = ea·Dij, where 1/a is the average movement ability for a given spatial process (here = 0·75), and Dij is the linear distance between grid cells i and j). The ‘connected-SSR’ value for each cell i (conSSRi) was then calculated using all cells j within a 1·25-km radius of i (math formula) (Drielsma, Ferrier & Manion 2007). Grid cells with the highest connected-SSR values are the most contiguous poorly represented cells in the reserve estate, so for each region, we selected those grid cells to be the new reserves under the balanced reserve design strategy (Fig. 2b, Fig. S1a,b, Supporting Information). This results in selected grid cells being more contiguous (greater aggregation and connectivity), but less effective in improving the representativeness of the reserve estate than the pure representativeness strategy (Fig. S4, Supporting Information).

For each reserve design strategy in each of the 10 regions, we quantified changes in the connectivity, aggregation and representativeness following the addition of the new reserves. (Fig. S4, Supporting Information). Connectivity in each region was assessed using the number of isolated reserve fragments (where lower values indicate higher connectivity). Reserve system aggregation in each region was assessed using the aggregation index of He, DeZonia & Mladenoff (2000), which ranges from 0 (completely disaggregated) to 1 (perfectly aggregated). Current (2010) reserve system representativeness in each region was assessed using the summed similarity to the reserves (SSR) of each grid cell, as quantified above. The change in representativeness (ΔSSR) was then calculated using the average SSR of all available habitat from which the new reserves could be selected (math formula), and the average SSR of all cells selected to be new reserves (math formula), such that math formula. Randomly selected reserves would on average result in no change in representativeness, negative values indicate that new reserves are worse than random at improving representativeness, while positive values indicate new reserves are better than random at improving representativeness.

Assessing reserve design strategies with dynamic macroecological modelling

To assess the effectiveness of the alternative reserve design strategies in retaining biodiversity, we applied a dynamic macroecological modelling approach (Fig. S5, Supporting Information) (Mokany et al. 2012). In this approach, the initial composition of each community in a metacommunity (i.e. each cell in a regular spatial grid) is predicted using the DynamicFOAM algorithm (Mokany et al. 2011). This optimisation algorithm constructs species lists for each community that best meet the constraints of modelled estimates of the number of species present (α-diversity), the dissimilarity in species composition between each pair of communities (β-diversity) and any available data on the occurrences of specific species at specific sites (Mokany et al. 2011). Although communities predicted by DynamicFOAM can be partly or entirely comprised of hypothetical species (e.g. undescribed species), for the current implementation, we applied only the 2051 described Tasmanian native plant species.

The DynamicFOAM predictions of the current composition of every community are then applied as the starting conditions for the M-SET metacommunity model (Metacommunity–Space Environment Time) (Fig. S5, Supporting Information) (Mokany et al. 2012). The M-SET model is a dynamic, spatially explicit metacommunity model that predicts changes in the occurrence (presence/absence) of each species in each community over time in response to both endogenous and exogenous drivers. M-SET is a neutral metacommunity model that can be applied to any taxonomic group in any region, with individual species not possessing unique attributes (apart from their initial occurrences), and all species sharing the same dispersal kernel. The key inputs for M-SET are models of α-diversity, pair-wise β-diversity and dispersal. The model projects compositional change for each community over time under specified climate and habitat scenarios, incorporating stochastic dispersal processes and community assembly consistent with the predicted changes in α- and β-diversity (Mokany et al. 2012). The input models of α- and β-diversity implicitly incorporate interspecific interactions plus community constraints and are assumed to reliably predict the compositional dissimilarity and potential species richness of communities over time, under changing environmental conditions.

We used this dynamic macroecological modelling approach to project climate change impacts on the flora of Tasmania. As described in Mokany et al. (2012), correlative models of α-diversity (< 0·001, proportion deviance explained = 0·259) and β-diversity (< 0·001, proportion deviance explained = 0·532) were derived using all available species occurrence data and complete spatial environmental data at 250-m resolution across Tasmania (72 350 km2). The α- and β-diversity models were then combined with the occurrence records to predict the current (2010) composition of all communities across Tasmania (n = 1 157 587) using DynamicFOAM (Mokany et al. 2011, 2012), with five replicate predictions of metacommunity composition generated (mean absolute error in predicted dissimilarity = 0·078 ± 0·002). These predictions of current community composition across Tasmania were then used as the starting point for the M-SET model. The other inputs for the M-SET model were the α- and β-diversity models derived previously, and a negative square power law model of dispersal probability {Pdij = [1/(2πrij)] [2K/(πλ (1 + (rij/λ)2))]}; where rij is the radial distance between locations i and j, the assumed median dispersal distance (λ) was 250 m, and the scaling factor (K) was 4500) (Mokany et al. 2012). Although this dynamic macroecological modelling approach and the input models of α- and β-diversity were previously applied for plant biodiversity in Tasmania (Mokany et al. 2012), here we applied entirely new simulations for the four specific reserve design strategies assessed.

For reserves allocated under each reserve design strategy in each region, we applied the M-SET model 25 times (five times for each five replicate DynamicFOAM initial composition prediction) for the entire reserve estate across Tasmania (32 277 km2) plus the new reserves for the focal region (16 km2), with a yearly time step from 2010 to 2100. We also applied M-SET using only the current reserve estate, to quantify the benefits of adding the new reserves to a region. To incorporate the impacts of likely climate change, we applied bias-adjusted fine scale (0·1°) dynamically downscaled climate projections from the Climate Futures for Tasmania project (Corney et al. 2010; Grose et al. 2010), using the A2 climate scenario and CSIRO-Mk3.5 general circulation model. The climate projections were statistically downscaled in the present analysis to 250-m grid resolution using ANUCLIM v6.1 (Xu & Hutchinson 2010). The models of α- and β-diversity were applied to the mean decadal climate projections and static environmental variables to project species richness and compositional dissimilarity for each decade from 2010 to 2100 under climate change, with projections for non-decadal years (i.e. 2xx1 to 2xx9) linearly interpolated between the two closest decadal projections (Mokany et al. 2012).

The parameters we applied in the dispersal model for our analysis were chosen to approximate the likely dispersal properties of Tasmanian plant species (Mokany et al. 2012). To assess the sensitivity of our results to alternative dispersal capacities, we compared the outcomes under 12 different levels of dispersal probability, derived by altering the scaling factor (K) in the dispersal model (Fig. S2, Supporting Information). This resulted in probabilities of successful dispersal at a given distance ranging over four orders of magnitude. For each level of dispersal examined (= 25; 45; 225; 450; 1000; 1500; 2250; 4500; 22 500; 45 000; 225 000; 450 000), we projected outcomes for Tasmanian plant biodiversity when new reserves were allocated under the four reserve design strategies, plus the existing reserves only, in the same manner as described previously. Due to the computational requirements of this analysis, we selected a single region over which to compare alternative reserve design strategies under different levels of dispersal, being region 10, in which the relative performance of the four reserve design strategies was most similar to the average across all regions (Fig. S3, Supporting Information).

Analysis of reserve design performance

The M-SET model projected the composition of every community in the reserve estate in every year from 2010 to 2100. To apply this data in comparing the capacity of alternative reserve design strategies to retain biodiversity, we examined two measures of conservation benefit: (1) the number of additional species retained to 2100 in each region following addition of the new reserves; and (2) the mean area of occurrence gained per species by 2100 following the addition of the new reserves. The first measure relates to the benefits of new reserves in retaining regional γ-diversity, while the second measure relates to the benefits of the new reserves for promoting the area occupied by all species. For both measures of conservation benefit, we partitioned the benefits into the portion derived purely from the new (added) reserves (direct benefits), and the benefits that the new reserves indirectly provided to the existing reserves (indirect benefits). We used anova to assess the significance of differences in the benefits for biodiversity achieved between reserve design strategies, using R (R Development Core Team 2012). Changes in reserve system connectivity, aggregation and representativeness when new reserves were allocated under each reserve system design strategy in each region were related to the number of additional species retained and the mean area of occurrence gained per species through simple linear regression.

Results

When considering the modelling results from all ten regions combined, the strategy based on representativeness was significantly more successful at increasing the total number of species retained in a region to 2100 under climate change than the other three reserve design strategies assessed (Fig. 3a: = 14·61, < 0·001). These benefits of the representativeness strategy for regional gamma diversity (Fig. 3a) were realised directly through significantly more species being supported solely in the new reserve areas (= 131·05, < 0·001) rather than indirectly through flow on benefits to the existing reserve areas (= 0·73, = 0·53).

Figure 3.

The predicted biodiversity benefits of allocating new conservation reserves under the four reserve design strategies, with respect to: (a) the mean number of additional species retained in each region and (b) the mean area of occurrence gained per species in each region, in 2100 under climate change. The total benefit under each reserve design strategy for both γ-diversity (b), and mean area of occurrence (b), are partitioned into the direct benefit (grey: attributable to the new reserve areas themselves) and the indirect benefit (black: flow on benefits to existing reserve areas). Different symbols represent significant differences (< 0·05) between reserve design strategies (‘a, b’ = total benefits; ‘A, B, C’ = direct benefits; ‘#, *’ = indirect benefits). Error bars are for the partitioned data.

In contrast, when the conservation benefit was quantified as the mean area of occurrence gained per species to 2100 (Fig. 3b), the aggregation and balanced reserve design strategies performed significantly better than the connectivity and representativeness strategies (= 13·84, < 0·001). The mean area of occurrence per species gained directly from the new reserve areas was significantly greater under the balanced strategy compared with the other reserve design strategies (Fig. 3b: = 24·55, < 0·001). Despite this, there were significantly greater indirect benefits for increasing the mean area of occurrence per species within existing reserves for the connectivity and aggregation strategies compared with the representativeness and balanced strategies (Fig. 3b: = 224·08, < 0·001).

When the results for each reserve design strategy in each region were assessed, there were few clear relations between changes in actual reserve system connectivity, aggregation or representativeness, and the two biodiversity benefits quantified: number of additional species retained in the region (Fig. 4a–c), and the mean area of occurrence gained per species (Fig. 4d–f). The strongest relationship observed was for increased reserve representativeness to result in an increased number of additional species retained in each region (Fig. 4c; < 0·001, R2 = 0·253). The positive relationship between change in number of reserve fragments and the number of additional species retained in each region (Fig. 4a; = 0·032, R2 = 0·116) is likely to be an artefact of the positive relationship between change in reserve representativeness and change in number of reserve fragments (< 0·001, R2 = 0·357). The greatest increase in reserve representativeness was often achieved by many small fragmented new reserves (Fig. S1a,b, Supporting Information).

Figure 4.

The relationship between changes in the level of reserve connectivity, aggregation or representativeness and: (a, b, c) the number of additional species retained in each region or (d, e, f) the mean area of occurrence gained per species, from 2010 to 2100 under climate change. Data are means for new reserves allocated under each of the four reserve design strategies in each of the ten regions examined. Linear relations are significant for (a) (= 0·032, R2 = 0·116) and (c) (< 0·001, R2 = 0·253).

Our analysis of 12 different levels of dispersal in region 10 suggested that dispersal ability had no significant effect (> 0·05) on the ability of any of the four reserve design strategies to retain additional species in the region (Fig. 5a). Across all levels of dispersal tested, the representativeness strategy retained significantly more species in region 10 compared with the three other reserve design strategies (Fig. 5a: = 35·99, < 0·001). In contrast, higher levels of dispersal resulted in significantly greater area of occurrence gained per species for all four reserve design strategies (Fig. 5b: = 2793·97, < 0·001). The relative performance of each reserve design strategy in terms of the area of occurrence gained per species also changed with increasing levels of dispersal. For example, the aggregation approach performed worst at low levels of dispersal, but the equal best at high levels of dispersal, while the performance of the connectivity approach relative to the other strategies decreased as the level of dispersal increased (Fig. 5b).

Figure 5.

(a) The number of additional species retained and (b) the area of occurrence gained per species, from 2010 to 2100 under climate change, for the four reserve design strategies in region 10, under 12 different levels of dispersal probability (Fig. S2, Supporting Information). Linear trends in (a) are not significant (> 0·05) and are shown for interpretative purposes only. For (b), alternative reserve design strategies are significantly different (< 0·05) within each level of dispersal, except where indicated by an enclosing grey circle. Note that dispersal scalar values (x-axis) are shown on a log-scale.

Discussion

Comparing broad reserve design strategies

There are different possible conservation goals by which to assess the effectiveness of alternative habitat configurations in retaining biodiversity under climate change (Pressey et al. 2007). Here, we compared reserve design strategies using two measures of conservation benefit: the benefit of new reserves in increasing the total number of species retained in the region over time, and the benefit in increasing the average area of occurrence for all species (Fig. 3). Our analysis suggests that the best approach to creating new reserves to retain biodiversity under climate change depends on the specific conservation goals.

For all ten regions combined, allocating new reserves solely on the current (2010) representativeness of habitat was most effective in retaining regional gamma diversity (Fig. 3a). In contrast, when the conservation goal was to promote the mean area of occurrence for all species, the most effective reserve design strategies were to increase the aggregation of reserves, and balance connectivity with representativeness and aggregation (Fig. 3b). Our results indicate that the most effective reserve design strategy will strongly depend on the specific conservation goals, reiterating the importance for ecologists and conservation managers to explore and identify clear biological objectives for conservation actions (Margules & Pressey 2000).

For both of the conservation goals assessed here, increasing connectivity by allocating new reserves to corridors was an inferior reserve design strategy for retaining Tasmanian plant biodiversity under climate change (Fig. 3a,b). These results highlight the need for much closer scrutiny of pure connectivity as the dominant habitat configuration strategy in retaining biodiversity under rapid climate change (Williams et al. 2005; Falcy & Estades 2007; Hodgson et al. 2009). In our analysis, using the same amount of land to increase aggregation and representativeness of existing reserves was more effective in retaining regional gamma diversity and increasing the average area of occurrence across all species (Fig. 3a,b). Although it has been empirically shown that corridors can be effective in promoting the movement of organisms (Gilbert-Norton et al. 2010), the conservation benefits of corridors are less clear, especially when compared to more established reserve design strategies (Hodgson et al. 2009).

An additional consideration in our comparison of reserve design strategies is the locations in which the benefits of the new reserves are achieved. In our analysis, we partitioned benefits for the two measures assessed into components that were realised in the new reserves (direct benefits) or in the existing reserves (indirect benefits) (Fig. 3). This partitioning analysis revealed that reserve design strategies based on connectivity and aggregation resulted in greater benefits for biodiversity in the existing reserve areas, compared with the representativeness-based strategies (representativeness and balanced). These results highlight an important trade-off in reserve system objectives (Fig. 1): approaches aiming to increase representativeness will often require allocating new reserves away from existing reserves, limiting the capacity for the new reserves to contribute to biodiversity benefits in existing reserves.

In a combined assessment of the two measures of conservation benefit examined here, the reserve design strategy balancing representativeness with aggregation and connectivity could be considered to have performed best overall in retaining biodiversity (Fig. 3). Indeed, most quantitative approaches to reserve selection attempt to balance representativeness, aggregation and connectivity in identifying priorities for new reserves (Moilanen 2007; Ball, Possingham & Watts 2009). Future studies comparing projected biodiversity benefits for large numbers of species under alternative ‘balanced’ reserve selection strategies would be highly valuable and could easily be achieved through modelling approaches such as ours.

The importance of context for reserve design

Summarising the benefits of different reserve design strategies across all ten regions examined here (Fig. 3) is useful in demonstrating their overall relative benefits for retaining biodiversity under climate change. However, there was substantial variation between individual regions as to which reserve design strategy best retained biodiversity, and the magnitude of the benefit achieved (Fig. 4). The variable performance of different reserve design strategies in different regions was not related in any simple way to the attributes of the regions considered (Fig. 4), such as the extent of reserved areas, or their level of fragmentation (aggregation). In addition, the only consistent evidence of reserve design effects at the regional scale was that the number of additional species retained in a region was positively related to the change in reserve representativeness (Fig. 4c). Note that the relatively modest gain in the gamma diversity for a region (0·6–41·8 species) is a direct result of the relatively modest area added as new reserves in our analyses (16 km2).

These results highlight the importance of a complex combination of regional attributes in determining which reserve design strategy will produce the greatest benefits for biodiversity under climate change, and the likely magnitude of those benefits. For any given region, the amount and spatial configuration of existing reserves, the representativeness of existing reserves, the amount and spatial configuration of unreserved natural habitat, and the spatiotemporal nature of projected climate change for that region will all combine in complex ways to influence the effectiveness of different reserve design strategies in retaining biodiversity over time (Hodgson et al. 2009). This complexity of interacting factors means that assessing priorities for new reserves in a region requires thorough quantitative assessment, based on the unique attributes of each region. Following a single broad reserve design strategy (e.g. connectivity) in every region is unlikely to provide the best outcomes in retaining biodiversity under climate change.

Alternative climate futures for a region should also be considered when assessing reserve design strategies that best retain biodiversity over time. Due to the number of analyses undertaken here, we applied a single climate scenario (CSIRO-Mk3.5 model, A2 emissions scenario). This climate future for Tasmania represents more severe warming and drying than projected by most other general circulation models and is therefore likely to approximate a ‘worst case’ scenario for plant biodiversity. Assessing outcomes for biodiversity over much longer time periods than examined here (90 years) would also help in identifying the most robust habitat configurations for a region. Although our dynamic modelling analyses directly quantify the ‘extinction debt’ (Tilman et al. 1994; Thomas et al. 2004) incurred by a given habitat configuration to 2100, longer simulations would more thoroughly account for the full implications of alternative habitat configuration on biodiversity.

Reserve design and dispersal

The capacity for species to disperse to new habitats is likely to be an important factor in determining the effectiveness of alternative habitat configuration strategies in retaining biodiversity under climate change. In this study, we assessed how the performance of the four reserve design strategies changed with dispersal ability for one of the regions examined (region 10). In terms of the number of additional species retained in the region, the level of dispersal had no significant effects, with the representativeness strategy consistently retaining significantly more species under climate change (Fig. 5a). The high variability in the number of additional species retained in the region clouded any effect of dispersal. This variability in regional gamma diversity results from the stochastic factors in our model influencing whether very rare species persist or become regionally extinct.

In contrast, higher levels of dispersal led to significantly greater benefits in terms of the gain in area of occurrence per species for all reserve design strategies (Fig. 5b). Importantly, the relative performance of the four reserve design strategies varied as the level of dispersal increased, with the connectivity strategy becoming relatively less effective and the aggregation strategy becoming relatively more effective (Fig. 5b). Our results demonstrate that dispersal abilities can be important in influencing which configuration of habitat provides the greatest benefits for biodiversity. The obvious implication of this finding is that for taxonomic groups with different dispersal abilities, retaining biodiversity under climate change would be best achieved through different reserve configurations. Assessments of how alternative reserve configurations retain overall biodiversity for multiple taxonomic groups with different dispersal abilities would be highly valuable in identifying strategies that best retain biodiversity as a whole under climate change.

Looking forward

Here, we have demonstrated that our dynamic macroecological modelling approach can be very useful in comparing the effectiveness of alternative configurations of habitat in retaining biodiversity under climate change. However, we also suggest that our modelling approach could itself be used as a strategy for designing conservation reserves. By applying our modelling approach to assess a much larger number of possible reserve configurations, including many randomly generated alternatives, we could identify those configurations likely to best retain biodiversity over time. The advantage over existing methods for designing conservation reserves would be that our approach explicitly incorporates the key ecological processes of dispersal and community assembly in projecting outcomes collectively for thousands of species.

Here, we applied our dynamic macroecological modelling approach in a neutral manner, with species possessing no unique attributes. Future applications could extend the realism and utility of these analyses by incorporating some key differences between species. For example, where good information exists on the dispersal attributes of different species or functional groups, this information could be applied to avoid our assumption that all species possess identical dispersal kernels. Other functional attributes could also be included, in either driving the model of community assembly and turnover or assessing the likely outcomes of compositional changes for the functional properties of communities. Including these attributes, or other information (e.g. phylogenetic diversity), could change the outcomes of the model and the relative conservation benefits of different habitat configurations.

In conclusion, we have demonstrated the capacity of our dynamic macroecological approach to assess alternative reserve design strategies and highlighted its potential to identify reserve configurations that may best retain biodiversity under climate change. Our analyses for plant biodiversity in Tasmania show that the best reserve design strategy will strongly depend on the primary conservation goals, reiterating the importance of having clearly identified ecological goals for conservation actions. Finally, we emphasise the role of regional context in influencing which reserve design strategy will best retain biodiversity over time, under climate change.

Acknowledgements

We thank D.M. Summers and K.J. Williams for comments on an earlier version of this manuscript.

Ancillary