Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Would environmental diversity be a good surrogate for species diversity?

Would environmental diversity be a good surrogate for species diversity? Representative conservation area-networks are needed to ensure persistence of species diversity within regions. Frequently, however, there are neither resources nor time to carry out detailed inventories before areas are selected. Consequently, areas may be chosen using information other than species. One promising approach is to represent as much environmental variation as possible (environmental diversity, ED) as a surrogate for species diversity (e.g. Anon. 1974, DeVellice et al. 1988, Belbin 1993, Faith and Walker 1996a). This would achieve great economies in all sectors, if true. To our knowledge no formal empirical tests have been made to assess the performance of environmental diversity as a surrogate for species diversity. Indeed, a positive relationship between these two measures has often been assumed rather than estimated. For example Pressey et al. (1996) and Woinarski et al. (1996) asked whether reserve networks sampled representative portions of environmental variation, but did not question whether this would represent biodiversity at a rate higher than expected by chance. We test this idea using species and environmental data for Europe. The p-median location-allocation model was applied to select representative portions of environmental-space (Faith and Walker 1996a, b). The consequences of this selection are compared to those of choosing areas at random and to solutions using an optimising area-selection algorithm (hotspots of complementarity). We show that ED does not always represent species at a rate consistently higher than that expected by chance, let alone approximate to that of the optimising solution. This is because particular distributions among restricted-range ECOGRAPHY 24:1 (2001) size species do not fit the underlying assumptions of the ED model. With these data, ED is a poor predictor of species diversity. European data Species diversity is measured as richness in European terrestrial vertebrates and higher plants. Data include 783604 records of occurrence for 186 mammal (Mitchell-Jones et al. 1999), 440 breeding bird (Hagemeijer and Blair 1997), 143 amphibian and reptile (Gasc et al. 1997), and 2294 plant species (Jalas and Suominen 1972– 96). These data vary with regard to taxonomic coverage; terrestrial vertebrates include all known species, whereas plants comprise : 20% of the European flora. The mapped area (2089 grid cells) excludes most of the eastern European countries (except for the Baltic States) because of low recording efforts in these areas (Williams et al. 2000). Nevertheless, this is still one of the world’s most extensive and representative data sets for species distribution on a consistent grid. Six environmental variables (Table 1) were selected from those used previously to model expected species distributions (e.g. Huntley et al. 1995); these variables were summarised into two axes of variation using PCA (Principal Components Analysis), after standardisation of data to zero means and unit variances (Table 1). The first component summarises a temperature gradient running from north-east to south-west, parallel to the 103 Table 1. Results of PCA on six environmental variables distributed across Europe. Data on potential evapotranspiration and altitude were obtained from UNEP (United Nations Environmental Program). Precipitation and temperature were obtained from NOAA (National Oceanic and Atmospheric Administration). Data were converted from 0.5° latitude-longitude maps to UTM (Universal Transverse Mercator) 50×50 km grid cells. PCA 1 Eigenvalue % variation per axis Cumulative % variation Altitude Potential evapotranspiration in January Potential evapotranspiration in July Mean annual precipitation Temperature in January Temperature in July 2.86683 47.8 47.8 −0.05659 0.73315 0.92857 0.00237 0.82344 0.88646 PCA 2 1.40779 23.5 71.2 0.79979 −0.00654 0.09515 0.84357 0.13499 −0.1708 The ED model Environmental diversity (ED) is measured in relation to the degree to which sampling variation is maximised within environmental-space; the greater the variation sampled, the greater the diversity of environments expected to be represented. As a first step towards selecting ED areas from our environmental-space, pairwise distances between pairs of principal-components scores were calculated using a shortest-path algorithm (Dijkstra 1959). These distances were used to solve p-median location-allocation models (Hikimi 1965), where p conservation areas are selected from n possible areas, so that the sum of the distances is minimised (example in Fig. 2b). This is conceptually equivalent to maximising ED when sampling environmental-space (Faith and Walker 1996a). Formally, minimise: Z = % % aidijIij i=1 j=1 Atlantic coast; the second summarises moisture conditions often associated with high altitudes (Fig. 1). As expected, the two PCA axes are unrelated (Spearman rank correlation, rs =0.11, p B0.001). Regional deviations from this relationship reveal patterns of covariation between the two PCA axes (Fig. 1, overlaying technique described by Williams and Gaston 1998). For example, the Mediterranean region has high scores for the first PCA axis (high temperature) and generally low scores for second axis (low moisture). Score variation in the second axis show regional differences in moisture availability from dry (e.g. eastern coast of Spain) to wet (e.g. north-west of the Iberian Peninsula). Subject to: % Iij = 1 j=1 n j=1 i = 1, … , n % Ijj = p i "j Iij 5 Ijj where: Z is the value of the objective function, n is the number of areas in the environmental-space, p is the number of ED areas to be selected, ai is the number of candidate areas for selection at location i, dij is the distance from ai to p at location j, Iij it takes the value of 1 if ai is allocated to p at j, 0 otherwise. Fig. 1. Geographical variation in the two PCA axes scores of environmental variables among UTM 50 ×50 km grid cells in Europe: a) scores for PCA axis 1; b) scores for PCA axis 2; c) overlay of the two PCA axes scores. Axes score maps (a and b) were divided into thirty three equal-frequency colour classes, so that maximum scores are shown in red and minimum scores are shown in blue. Scores for the overlay map (c) show the regional differences in the overall variation between the PCA axes. We use a 10 colour-scale class, where increasing intensities of blue represent increasing scores of PCA axis 1 and increasing intensities of green represent increasing scores of PCA axis 2. Black grid cells show low scores for both, white shows high scores for both, and shades of grey show linearly covarying scores for both. ECOGRAPHY 24:1 (2001) Fig. 2. Location of grid cells in relation to the environmental space of PCA axis 1 and 2. The example shows ED and optimising area-set solutions (hotspots of complementarity) for 211 areas. P-medians can be solved using optimal linear-programming or heuristic techniques. Most optimal techniques require, at some point, the application of a branch and bound algorithm. Because of this, it is possible that large problems take an inordinate amount of computation time to solve. In practice, the most robust of these techniques (Narula et al. 1977) still needs refining to solve p-medians for \ 900 areas (Church and Sorensen 1996). We used the heuristic vertex-substitution algorithm GRIA (Global-Regional Interchange Algorithm) (Densham and Rushton 1992), because this is one of the most robust and efficient heuristic procedures available to address large p-median problems (Church and Sorensen 1996). GRIA selects p (or ED) areas from among m candidate locations to represent an environmental space dispersed over n locations (here m =n). ED areas are located to minimise the value of an objective function (z): the sum of all n areas weighted by the distance separating them from their closest ED area. GRIA has two phases. The first phase (global exchange) itself consists of two parts: first, identify the ED area to drop from the current solution that least increases the value of z and, second, find the candidate to add to the solution which most reduces the value of z. In its second phase (regional exchange), GRIA ensures that all areas in the environmental space are represented by their closest ED area and that each ED area is located at the local median of the areas it represents. The two phases are applied ECOGRAPHY 24:1 (2001) iteratively until three conditions are met: 1) each and every ED area is the local median of the areas it represents; 2) each area in the space is allocated to its closest ED area; and 3) removing an ED area from the solution and replacing it with a candidate area not in the solution yields an increase in the value of z. These properties are necessary but not sufficient for a globally optimal solution. Optimising and random models The efficiency of ED to predict the location of important areas for species diversity was compared to that of an optimising solution and to that expected by chance. The models were solved for 52 (2.5% of the total areas), 105 (5%) and 211 (10%) areas respectively. A complementarity-based area selection procedure was used to identify optimising solutions (hotspots of complementarity) that maximise species representation in a given area (Church et al. 1996). The algorithm selects first all areas with taxa that are irreplaceable for a given representation goal. Then it follows a simple set of rules to select areas with the greatest complementary richness in just the rarest taxa. If there are ties it proceeds by selecting areas among ties richest in the next rarest taxa and so on. If there are persistent ties, it then selects areas among persistent ties with the lowest grid-cell number. This is an arbitrary rule rather than a random 105 choice among ties in order to ensure repeatability in tests. It then performs a test to reject any areas that in hindsight are redundant. It repeats all previous steps until the representation goal is achieved. Finally it re-orders areas by complementary richness and chooses the first n areas from the re-ordered area list. A random solution is obtained by simulating selection of a given number of areas with records at random; the selection is repeated 1000 times to calculate the 5% upper tail of the random distribution. This is used as simple test to assess differences from observed p-median and optimising solutions with that of expected by chance (p \0.05). The WORLDMAP (Williams 1999) software was used to implement both the optimising and random solutions. A test Only plants exhibit consistent, non-random positive patterns of representation (p B0.05) when areas are selected to maximise ED (Table 2). The best results were obtained when 211 ED areas were selected, representing : 3% more species than expected by chance (pB 0.05) and :21% less than expected from the optimising solution. In contrast, ED areas consistently failed to represent more amphibians and reptiles than expected by chance (p B 0.05). The worst results were Table 2. Percentages of species represented in selected areas: a) p-median solutions seeking to maximise environmental diversity; b) areas selected at random, with 1000 trials performed to calculate the 5% upper tail of the random distribution; c) optimising solutions seeking to maximise species representation, i.e. hotspots of complementarity. Where appropriate, multiple representations (rep) of all species are shown. a) ED (%) Plants 52 areas 105 areas 211 areas Birds 52 areas 105 areas 211 areas Mammals 52 areas 105 areas 211 areas Amphibians and reptiles 52 areas 105 areas 211 areas All 52 areas 105 areas 211 areas 57.80 69.88 79.25 83.41 92.73 93.64 79.57 90.32 94.09 72.73 83.22 87.41 63.50 75.02 82.60 b) Random (%) 57.45 67.31 76.16 86.82 91.14 94.32 82.26 88.71 93.55 78.32 85.31 91.66 63.63 72.05 80.28 c) Hotspots (%) 84.92 93.50 99.83 100×2 rep 100×4 rep 100×9 rep 100×2 rep 100×4 rep 100×9 rep 100×3 rep 100×6 rep 100×13 rep 86.55 94.16 99.38 obtained with 211 ED areas representing : 4% less than expected by chance (p B0.05), and nearly 12 – 13% less than expected from a near-optimum solution. Additionally, ED areas do not show consistent patterns of representation for breeding birds and mammals when different p-median solutions are considered. For example, ED areas perform better than random when 105 areas are selected (p B0.05), but perform badly when only 52 areas are selected (p B 0.05). Testing the performance of ED against the combined data set of plants and terrestrial vertebrates improves slightly on the results, but this is due to the overriding importance of plants (75% of the total) in relation to the other taxa. The idea that ED provides a useful framework for area selection in the context of species conservation, comes from the assumption that species’ distributions are at equilibrium with governing environmental factors (Hutchinson 1957, Whittaker 1975, Brown 1995). A unimodal bell-shaped response curve is often used to describe this relationship (Faith and Walker 1996a, b). However, non-equilibrium historical events such as extinction, speciation, barriers to dispersal, and biotic interactions also play major roles in determining current distributions. Consequently, the spatial arrangement of areas needed to maximise species representation may be clustered and/or biased towards some particular section of environmental-space (Fig. 2c– f). This contrasts with the idea that areas adequately spanning environmental-space would maximise species representation (Fig. 2b). Our data do not support this idea. Only optimising solutions for plants span the space well. However, they cluster showing that important areas for conservation may be environmentally autocorrelated rather than evenly dispersed in environmental space. The same pattern of clustering applies to optimising solutions for all other groups, except that they do not sample environmental-space as adequately as for plants. For example, areas needed to maximise representation of amphibians and reptiles do not require areas to be selected beyond a score of −1 on the first PCA axis, whilst the original samplingspace extends almost to a score of − 3 (Fig. 2f). Selected areas for breeding birds and mammals do not span the entire range of the second PCA axis (Fig. 2d, e). As expected, patterns of environmental autocorrelation among optimising solutions have the effect that the areas selected are also spatially autocorrelated (Fig. 3). ED areas have a mean number of 4.52% of nearest neighbours that are ED areas themselves, whilst optimising solutions are more aggregated: 16.47% nearest neighbours for plants; 23.70% for mammals; 19.50% birds; and 38.15% for herptiles. The relatively high performance of ED for plants follows logically from a well-established relationship between plant distributions and environmental limiting factors, such as temperature and precipitation (Whittaker 1975). A similar pattern of representation would ECOGRAPHY 24:1 (2001) be expected for amphibians and reptiles given that their distribution is known to be strongly limited by precipitation (especially for amphibians) and temperature (especially for reptiles) (Gasc et al. 1997). Why are ED areas generally poor surrogates for terrestrial vertebrates with this data? One hypothesis is that the impact of temperature and moisture in determining current species distributions is contingent to Fig. 3. Geographical location of the 211 selected UTM 50 × 50 km grid cells in Europe. Richness scores in the species that remain unrepresented were divided into a thirty-three equal-frequency colour-scale, so that maximum scores are shown in red and minimum (non zero) scores are shown in blue. ECOGRAPHY 24:1 (2001) many other unmeasured factors. To explore this idea further it is useful to investigate whether unrepresented species are a non-random sample within each group. If they are not then one can ask what underlying mechanisms might cause ED to fail as a surrogate for these species. An inspection of the residuals (resulting from selecting 211 ED areas) shows that unrepresented species are a non-random set with restricted-range sizes below the lower quartile value of the species-range-size distribution (Fig. 4). Furthermore, their distributions Fig. 3. ECOGRAPHY 24:1 (2001) Fig. 4. Species-range-size distributions for log10-transformed European geographic range sizes of plant, bird, mammal, and combined amphibian and reptile species. Range sizes are measured as the number of UTM 50×50 km grid cells occupied. Thick arrows indicate the lower quartile threshold-value for the log10-transformed speciesrange-sizes; thin arrows indicate the Medians for log10-transformed unrepresented species-rangessizes in the 211 ED areas model. do not fit equilibrium assumptions of the ED model. For example, unrepresented amphibian and reptile species (n =18) are either narrow endemics (72%), limited to a few areas in the Mediterranean (mountain-tops, islands, or peninsulas), or are at the edges of the their ranges (28%), having their core distribution eastwards or southwards. In the former case, environmentally suitable areas may be unoccupied due to species’ inability to colonise them. In the latter case, species may occur in unusual environmental conditions. Indeed, their peripheral position may be better described by a monotonic curve than by an unimodal bell-shaped response curve (TerBraak and Prentice 1988). Breeding birds (n=27) show a similar pattern of unrepresentation. They are relict populations (7%), species with distributions limited by lower-level variations in habitat (36%), species at the edges of ranges (36%), or introductions (21%). Similarly, unrepresented mammals (n = 11) are relict species or narrow endemics (36%), species at their distributional edge (27%), or introductions (36%). We suggest that the degree to which ED is suitable for predicting species diversity depends on the extent to which non-equilibrium events affect current distributions. It could be argued that marginal populations and introduced species are dubious targets for analyses, as they may be of no conservation concern or typical of environmental conditions that are atypical in the study area; in either case they would probably be better represented somewhere else. To explore for the extent, to which these two types of distributions affect the outcome of the analyses, we tested the performance of 211 ED areas selected to represent 88 native endemic vertebrate species to Europe (this subset of the data is fully described by Williams et al. 2000). Again, ED areas represented species at a rate lower (79.55%) than expected by chance (85.23%, p B0.05) and much lower ECOGRAPHY 24:1 (2001) than the optimising solution ( : fourteen representations of all species). Possible caveats and developments Our results provide only weak support for the idea that representative samples of environmental-space are also representative of species diversity. The most restrictedrange size species are under-represented and these tend to be more susceptible to extinction (e.g. Johnson 1998). Therefore, with these data, selecting areas with the ED criterion would be inadequate to achieve the goal of species persistence within regions. It may be that European data are not representative of other areas, specially because current distributions are highly affected by human history; it may also be that other taxonomic groups with different ecologies and life histories may generate distinct patterns of representation. Nevertheless, given that the sizes of geographic range within taxonomic assemblages are generally rightskewed (Gaston 1996; i.e. most species have relatively small range sizes while a few have relatively large ones) and that, at least in the northern hemisphere, speciesrange-sizes are expected to decline with latitude (Gaston et al. 1998), it is likely that ED might perform inadequately in many areas and for many taxa. This is especially true in the tropics and in mediterranean-climate areas, where there is a large number of endemic species with distributions strongly determined history, area and isolation (e.g. Major 1988). It is possible that adding further dimensions to environmental-space would improve the surrogacy value of ED. Furthermore it may also be a possibility that narrowing the spatial scale of analyses would provide further insights on ED as a surrogate for species diversity. In the absence of similar tests from elsewhere, however, it 109 would be unwise to ignore these provisional results. We suggest that ED should be used only when its surrogacy value has been empirically demonstrated. Naturally, this requires tests using high-quality data for species and these are rarely available. The ED framework could also be tested for Gap Analysis (Scott et al. 1993) where, in a first step, areas would be chosen given knowledge on the distribution of few, well-known, restricted-range size and endangered species (typically with non-equilibrium distributions); followed by the selection of additional areas using the ED criterion. This is likely to overcome some of the weaknesses of ED in representing some of the most valued species for conservation, whilst ensuring that a greater variety of environments would be represented. Acknowledgements – We thank the national atlas representatives and the many volunteer fieldworkers who contributed the atlas records; Dominique Richard (European Topic Centre for Nature Conservation) for bringing some of the authors together at the European Chorological Grid Reference System meeting in Paris during May 1998; Mike Sadka, Rosemarie Rees, and Sonia Gervas (The Natural History Museum) for ´ help in combining the species data; and Lera Miles (Univ. of Leeds) for assistance in converting the environmental data. Dan Faith, Dick Vane-Wright, Paul Williams and Stuart Pimm provided valuable comments on the manuscript. MBA is supported by the Portuguese PRAXIS XXI Studentship, under the sub-programme Science and Technology of the second European Union support framework. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Ecography Wiley

Would environmental diversity be a good surrogate for species diversity?

Loading next page...
 
/lp/wiley/would-environmental-diversity-be-a-good-surrogate-for-species-62Y0ItsyCV

References (21)

Publisher
Wiley
Copyright
Copyright © 2001 Wiley Subscription Services, Inc., A Wiley Company
ISSN
0906-7590
eISSN
1600-0587
DOI
10.1034/j.1600-0587.2001.240112.x
Publisher site
See Article on Publisher Site

Abstract

Representative conservation area-networks are needed to ensure persistence of species diversity within regions. Frequently, however, there are neither resources nor time to carry out detailed inventories before areas are selected. Consequently, areas may be chosen using information other than species. One promising approach is to represent as much environmental variation as possible (environmental diversity, ED) as a surrogate for species diversity (e.g. Anon. 1974, DeVellice et al. 1988, Belbin 1993, Faith and Walker 1996a). This would achieve great economies in all sectors, if true. To our knowledge no formal empirical tests have been made to assess the performance of environmental diversity as a surrogate for species diversity. Indeed, a positive relationship between these two measures has often been assumed rather than estimated. For example Pressey et al. (1996) and Woinarski et al. (1996) asked whether reserve networks sampled representative portions of environmental variation, but did not question whether this would represent biodiversity at a rate higher than expected by chance. We test this idea using species and environmental data for Europe. The p-median location-allocation model was applied to select representative portions of environmental-space (Faith and Walker 1996a, b). The consequences of this selection are compared to those of choosing areas at random and to solutions using an optimising area-selection algorithm (hotspots of complementarity). We show that ED does not always represent species at a rate consistently higher than that expected by chance, let alone approximate to that of the optimising solution. This is because particular distributions among restricted-range ECOGRAPHY 24:1 (2001) size species do not fit the underlying assumptions of the ED model. With these data, ED is a poor predictor of species diversity. European data Species diversity is measured as richness in European terrestrial vertebrates and higher plants. Data include 783604 records of occurrence for 186 mammal (Mitchell-Jones et al. 1999), 440 breeding bird (Hagemeijer and Blair 1997), 143 amphibian and reptile (Gasc et al. 1997), and 2294 plant species (Jalas and Suominen 1972– 96). These data vary with regard to taxonomic coverage; terrestrial vertebrates include all known species, whereas plants comprise : 20% of the European flora. The mapped area (2089 grid cells) excludes most of the eastern European countries (except for the Baltic States) because of low recording efforts in these areas (Williams et al. 2000). Nevertheless, this is still one of the world’s most extensive and representative data sets for species distribution on a consistent grid. Six environmental variables (Table 1) were selected from those used previously to model expected species distributions (e.g. Huntley et al. 1995); these variables were summarised into two axes of variation using PCA (Principal Components Analysis), after standardisation of data to zero means and unit variances (Table 1). The first component summarises a temperature gradient running from north-east to south-west, parallel to the 103 Table 1. Results of PCA on six environmental variables distributed across Europe. Data on potential evapotranspiration and altitude were obtained from UNEP (United Nations Environmental Program). Precipitation and temperature were obtained from NOAA (National Oceanic and Atmospheric Administration). Data were converted from 0.5° latitude-longitude maps to UTM (Universal Transverse Mercator) 50×50 km grid cells. PCA 1 Eigenvalue % variation per axis Cumulative % variation Altitude Potential evapotranspiration in January Potential evapotranspiration in July Mean annual precipitation Temperature in January Temperature in July 2.86683 47.8 47.8 −0.05659 0.73315 0.92857 0.00237 0.82344 0.88646 PCA 2 1.40779 23.5 71.2 0.79979 −0.00654 0.09515 0.84357 0.13499 −0.1708 The ED model Environmental diversity (ED) is measured in relation to the degree to which sampling variation is maximised within environmental-space; the greater the variation sampled, the greater the diversity of environments expected to be represented. As a first step towards selecting ED areas from our environmental-space, pairwise distances between pairs of principal-components scores were calculated using a shortest-path algorithm (Dijkstra 1959). These distances were used to solve p-median location-allocation models (Hikimi 1965), where p conservation areas are selected from n possible areas, so that the sum of the distances is minimised (example in Fig. 2b). This is conceptually equivalent to maximising ED when sampling environmental-space (Faith and Walker 1996a). Formally, minimise: Z = % % aidijIij i=1 j=1 Atlantic coast; the second summarises moisture conditions often associated with high altitudes (Fig. 1). As expected, the two PCA axes are unrelated (Spearman rank correlation, rs =0.11, p B0.001). Regional deviations from this relationship reveal patterns of covariation between the two PCA axes (Fig. 1, overlaying technique described by Williams and Gaston 1998). For example, the Mediterranean region has high scores for the first PCA axis (high temperature) and generally low scores for second axis (low moisture). Score variation in the second axis show regional differences in moisture availability from dry (e.g. eastern coast of Spain) to wet (e.g. north-west of the Iberian Peninsula). Subject to: % Iij = 1 j=1 n j=1 i = 1, … , n % Ijj = p i "j Iij 5 Ijj where: Z is the value of the objective function, n is the number of areas in the environmental-space, p is the number of ED areas to be selected, ai is the number of candidate areas for selection at location i, dij is the distance from ai to p at location j, Iij it takes the value of 1 if ai is allocated to p at j, 0 otherwise. Fig. 1. Geographical variation in the two PCA axes scores of environmental variables among UTM 50 ×50 km grid cells in Europe: a) scores for PCA axis 1; b) scores for PCA axis 2; c) overlay of the two PCA axes scores. Axes score maps (a and b) were divided into thirty three equal-frequency colour classes, so that maximum scores are shown in red and minimum scores are shown in blue. Scores for the overlay map (c) show the regional differences in the overall variation between the PCA axes. We use a 10 colour-scale class, where increasing intensities of blue represent increasing scores of PCA axis 1 and increasing intensities of green represent increasing scores of PCA axis 2. Black grid cells show low scores for both, white shows high scores for both, and shades of grey show linearly covarying scores for both. ECOGRAPHY 24:1 (2001) Fig. 2. Location of grid cells in relation to the environmental space of PCA axis 1 and 2. The example shows ED and optimising area-set solutions (hotspots of complementarity) for 211 areas. P-medians can be solved using optimal linear-programming or heuristic techniques. Most optimal techniques require, at some point, the application of a branch and bound algorithm. Because of this, it is possible that large problems take an inordinate amount of computation time to solve. In practice, the most robust of these techniques (Narula et al. 1977) still needs refining to solve p-medians for \ 900 areas (Church and Sorensen 1996). We used the heuristic vertex-substitution algorithm GRIA (Global-Regional Interchange Algorithm) (Densham and Rushton 1992), because this is one of the most robust and efficient heuristic procedures available to address large p-median problems (Church and Sorensen 1996). GRIA selects p (or ED) areas from among m candidate locations to represent an environmental space dispersed over n locations (here m =n). ED areas are located to minimise the value of an objective function (z): the sum of all n areas weighted by the distance separating them from their closest ED area. GRIA has two phases. The first phase (global exchange) itself consists of two parts: first, identify the ED area to drop from the current solution that least increases the value of z and, second, find the candidate to add to the solution which most reduces the value of z. In its second phase (regional exchange), GRIA ensures that all areas in the environmental space are represented by their closest ED area and that each ED area is located at the local median of the areas it represents. The two phases are applied ECOGRAPHY 24:1 (2001) iteratively until three conditions are met: 1) each and every ED area is the local median of the areas it represents; 2) each area in the space is allocated to its closest ED area; and 3) removing an ED area from the solution and replacing it with a candidate area not in the solution yields an increase in the value of z. These properties are necessary but not sufficient for a globally optimal solution. Optimising and random models The efficiency of ED to predict the location of important areas for species diversity was compared to that of an optimising solution and to that expected by chance. The models were solved for 52 (2.5% of the total areas), 105 (5%) and 211 (10%) areas respectively. A complementarity-based area selection procedure was used to identify optimising solutions (hotspots of complementarity) that maximise species representation in a given area (Church et al. 1996). The algorithm selects first all areas with taxa that are irreplaceable for a given representation goal. Then it follows a simple set of rules to select areas with the greatest complementary richness in just the rarest taxa. If there are ties it proceeds by selecting areas among ties richest in the next rarest taxa and so on. If there are persistent ties, it then selects areas among persistent ties with the lowest grid-cell number. This is an arbitrary rule rather than a random 105 choice among ties in order to ensure repeatability in tests. It then performs a test to reject any areas that in hindsight are redundant. It repeats all previous steps until the representation goal is achieved. Finally it re-orders areas by complementary richness and chooses the first n areas from the re-ordered area list. A random solution is obtained by simulating selection of a given number of areas with records at random; the selection is repeated 1000 times to calculate the 5% upper tail of the random distribution. This is used as simple test to assess differences from observed p-median and optimising solutions with that of expected by chance (p \0.05). The WORLDMAP (Williams 1999) software was used to implement both the optimising and random solutions. A test Only plants exhibit consistent, non-random positive patterns of representation (p B0.05) when areas are selected to maximise ED (Table 2). The best results were obtained when 211 ED areas were selected, representing : 3% more species than expected by chance (pB 0.05) and :21% less than expected from the optimising solution. In contrast, ED areas consistently failed to represent more amphibians and reptiles than expected by chance (p B 0.05). The worst results were Table 2. Percentages of species represented in selected areas: a) p-median solutions seeking to maximise environmental diversity; b) areas selected at random, with 1000 trials performed to calculate the 5% upper tail of the random distribution; c) optimising solutions seeking to maximise species representation, i.e. hotspots of complementarity. Where appropriate, multiple representations (rep) of all species are shown. a) ED (%) Plants 52 areas 105 areas 211 areas Birds 52 areas 105 areas 211 areas Mammals 52 areas 105 areas 211 areas Amphibians and reptiles 52 areas 105 areas 211 areas All 52 areas 105 areas 211 areas 57.80 69.88 79.25 83.41 92.73 93.64 79.57 90.32 94.09 72.73 83.22 87.41 63.50 75.02 82.60 b) Random (%) 57.45 67.31 76.16 86.82 91.14 94.32 82.26 88.71 93.55 78.32 85.31 91.66 63.63 72.05 80.28 c) Hotspots (%) 84.92 93.50 99.83 100×2 rep 100×4 rep 100×9 rep 100×2 rep 100×4 rep 100×9 rep 100×3 rep 100×6 rep 100×13 rep 86.55 94.16 99.38 obtained with 211 ED areas representing : 4% less than expected by chance (p B0.05), and nearly 12 – 13% less than expected from a near-optimum solution. Additionally, ED areas do not show consistent patterns of representation for breeding birds and mammals when different p-median solutions are considered. For example, ED areas perform better than random when 105 areas are selected (p B0.05), but perform badly when only 52 areas are selected (p B 0.05). Testing the performance of ED against the combined data set of plants and terrestrial vertebrates improves slightly on the results, but this is due to the overriding importance of plants (75% of the total) in relation to the other taxa. The idea that ED provides a useful framework for area selection in the context of species conservation, comes from the assumption that species’ distributions are at equilibrium with governing environmental factors (Hutchinson 1957, Whittaker 1975, Brown 1995). A unimodal bell-shaped response curve is often used to describe this relationship (Faith and Walker 1996a, b). However, non-equilibrium historical events such as extinction, speciation, barriers to dispersal, and biotic interactions also play major roles in determining current distributions. Consequently, the spatial arrangement of areas needed to maximise species representation may be clustered and/or biased towards some particular section of environmental-space (Fig. 2c– f). This contrasts with the idea that areas adequately spanning environmental-space would maximise species representation (Fig. 2b). Our data do not support this idea. Only optimising solutions for plants span the space well. However, they cluster showing that important areas for conservation may be environmentally autocorrelated rather than evenly dispersed in environmental space. The same pattern of clustering applies to optimising solutions for all other groups, except that they do not sample environmental-space as adequately as for plants. For example, areas needed to maximise representation of amphibians and reptiles do not require areas to be selected beyond a score of −1 on the first PCA axis, whilst the original samplingspace extends almost to a score of − 3 (Fig. 2f). Selected areas for breeding birds and mammals do not span the entire range of the second PCA axis (Fig. 2d, e). As expected, patterns of environmental autocorrelation among optimising solutions have the effect that the areas selected are also spatially autocorrelated (Fig. 3). ED areas have a mean number of 4.52% of nearest neighbours that are ED areas themselves, whilst optimising solutions are more aggregated: 16.47% nearest neighbours for plants; 23.70% for mammals; 19.50% birds; and 38.15% for herptiles. The relatively high performance of ED for plants follows logically from a well-established relationship between plant distributions and environmental limiting factors, such as temperature and precipitation (Whittaker 1975). A similar pattern of representation would ECOGRAPHY 24:1 (2001) be expected for amphibians and reptiles given that their distribution is known to be strongly limited by precipitation (especially for amphibians) and temperature (especially for reptiles) (Gasc et al. 1997). Why are ED areas generally poor surrogates for terrestrial vertebrates with this data? One hypothesis is that the impact of temperature and moisture in determining current species distributions is contingent to Fig. 3. Geographical location of the 211 selected UTM 50 × 50 km grid cells in Europe. Richness scores in the species that remain unrepresented were divided into a thirty-three equal-frequency colour-scale, so that maximum scores are shown in red and minimum (non zero) scores are shown in blue. ECOGRAPHY 24:1 (2001) many other unmeasured factors. To explore this idea further it is useful to investigate whether unrepresented species are a non-random sample within each group. If they are not then one can ask what underlying mechanisms might cause ED to fail as a surrogate for these species. An inspection of the residuals (resulting from selecting 211 ED areas) shows that unrepresented species are a non-random set with restricted-range sizes below the lower quartile value of the species-range-size distribution (Fig. 4). Furthermore, their distributions Fig. 3. ECOGRAPHY 24:1 (2001) Fig. 4. Species-range-size distributions for log10-transformed European geographic range sizes of plant, bird, mammal, and combined amphibian and reptile species. Range sizes are measured as the number of UTM 50×50 km grid cells occupied. Thick arrows indicate the lower quartile threshold-value for the log10-transformed speciesrange-sizes; thin arrows indicate the Medians for log10-transformed unrepresented species-rangessizes in the 211 ED areas model. do not fit equilibrium assumptions of the ED model. For example, unrepresented amphibian and reptile species (n =18) are either narrow endemics (72%), limited to a few areas in the Mediterranean (mountain-tops, islands, or peninsulas), or are at the edges of the their ranges (28%), having their core distribution eastwards or southwards. In the former case, environmentally suitable areas may be unoccupied due to species’ inability to colonise them. In the latter case, species may occur in unusual environmental conditions. Indeed, their peripheral position may be better described by a monotonic curve than by an unimodal bell-shaped response curve (TerBraak and Prentice 1988). Breeding birds (n=27) show a similar pattern of unrepresentation. They are relict populations (7%), species with distributions limited by lower-level variations in habitat (36%), species at the edges of ranges (36%), or introductions (21%). Similarly, unrepresented mammals (n = 11) are relict species or narrow endemics (36%), species at their distributional edge (27%), or introductions (36%). We suggest that the degree to which ED is suitable for predicting species diversity depends on the extent to which non-equilibrium events affect current distributions. It could be argued that marginal populations and introduced species are dubious targets for analyses, as they may be of no conservation concern or typical of environmental conditions that are atypical in the study area; in either case they would probably be better represented somewhere else. To explore for the extent, to which these two types of distributions affect the outcome of the analyses, we tested the performance of 211 ED areas selected to represent 88 native endemic vertebrate species to Europe (this subset of the data is fully described by Williams et al. 2000). Again, ED areas represented species at a rate lower (79.55%) than expected by chance (85.23%, p B0.05) and much lower ECOGRAPHY 24:1 (2001) than the optimising solution ( : fourteen representations of all species). Possible caveats and developments Our results provide only weak support for the idea that representative samples of environmental-space are also representative of species diversity. The most restrictedrange size species are under-represented and these tend to be more susceptible to extinction (e.g. Johnson 1998). Therefore, with these data, selecting areas with the ED criterion would be inadequate to achieve the goal of species persistence within regions. It may be that European data are not representative of other areas, specially because current distributions are highly affected by human history; it may also be that other taxonomic groups with different ecologies and life histories may generate distinct patterns of representation. Nevertheless, given that the sizes of geographic range within taxonomic assemblages are generally rightskewed (Gaston 1996; i.e. most species have relatively small range sizes while a few have relatively large ones) and that, at least in the northern hemisphere, speciesrange-sizes are expected to decline with latitude (Gaston et al. 1998), it is likely that ED might perform inadequately in many areas and for many taxa. This is especially true in the tropics and in mediterranean-climate areas, where there is a large number of endemic species with distributions strongly determined history, area and isolation (e.g. Major 1988). It is possible that adding further dimensions to environmental-space would improve the surrogacy value of ED. Furthermore it may also be a possibility that narrowing the spatial scale of analyses would provide further insights on ED as a surrogate for species diversity. In the absence of similar tests from elsewhere, however, it 109 would be unwise to ignore these provisional results. We suggest that ED should be used only when its surrogacy value has been empirically demonstrated. Naturally, this requires tests using high-quality data for species and these are rarely available. The ED framework could also be tested for Gap Analysis (Scott et al. 1993) where, in a first step, areas would be chosen given knowledge on the distribution of few, well-known, restricted-range size and endangered species (typically with non-equilibrium distributions); followed by the selection of additional areas using the ED criterion. This is likely to overcome some of the weaknesses of ED in representing some of the most valued species for conservation, whilst ensuring that a greater variety of environments would be represented. Acknowledgements – We thank the national atlas representatives and the many volunteer fieldworkers who contributed the atlas records; Dominique Richard (European Topic Centre for Nature Conservation) for bringing some of the authors together at the European Chorological Grid Reference System meeting in Paris during May 1998; Mike Sadka, Rosemarie Rees, and Sonia Gervas (The Natural History Museum) for ´ help in combining the species data; and Lera Miles (Univ. of Leeds) for assistance in converting the environmental data. Dan Faith, Dick Vane-Wright, Paul Williams and Stuart Pimm provided valuable comments on the manuscript. MBA is supported by the Portuguese PRAXIS XXI Studentship, under the sub-programme Science and Technology of the second European Union support framework.

Journal

EcographyWiley

Published: Feb 1, 2001

There are no references for this article.