Conservation biology is a discipline with a deadline. Suitable habitats for nearly all species on nearly all continents are being altered or destroyed at rates unprecedented in human history (Wilson 1988, Ehrlich and Wilson 1991, Myers 1994). With limited time and resources, ecologists and conservation planners are faced with making difficult decisions and setting priorities to conserve what habitat remains. Conservationists have proposed numerous criteria to identify areas of highest conservation priority, including high species richness, high levels of endemism, high concentrations of rare species, unique habitats, unusual ecological or evolutionary phenomena, extraordinary richness or endemism in higher taxa, high ecosystem service value, or intense levels of threat (Myers 1988, Williams and Gastón 1994, Dinerstein et al. 1995, Long et al. 1996, Williams et al. 1996, Daily 1997, Olson and Dinerstein 1998, Ricketts et al. in press). Ideally, conservation priorities will be established based on as many of these factors as possible. Species richness, however, will likely remain a central component of most priority-setting studies because of the intuitive importance of biodiversity “hotspots”: if species extinction is the ultimate crisis to avoid, areas that contain many species are logical targets for biodiversity conservation.

The problem with using overall species richness as a criterion for setting conservation priorities is that for the vast majority of species, range data are not available. This situation is unlikely to improve; conducting exhaustive biotic surveys with the limited time and resources normally available for conservation is almost certainly impossible. Indeed, one effort to accomplish an exhaustive biotic survey in Costa Rica self-destructed almost immediately (Kaiser 1997).

Conservationists have responded to this lack of data with a widespread search for indicator taxa: relatively well known groups of organisms (i.e., those with relatively well understood ranges and taxonomies) whose distributions can be used as a surrogate measure of the distribution patterns of other taxa, or of biodiversity overall (e.g., Prendergast et al. 1993b, Scott et al. 1993, Sisk et al. 1994, Beccaloni and Gastón 1995, Williams et al. 1996, Carroll and Pearson 1998, Lawton et al. 1998). For example, Sisk et al. (1994) used a richness index based on mammals and butterflies to estimate overall species richness for 135 nations. They assumed that the overall species richness of a country could be estimated by counting the number of species present from these two groups. In other words, they assumed that mammals and butterflies are good indicator taxa of overall species richness at broad scales. The United States Gap Analysis Project also employs indicator taxa (vertebrates, butterflies, and occasionally plants) as proxies for overall biodiversity patterns (Scott et al. 1993).

Despite the common and increasing use of indicator taxa, few studies have tested explicitly the utility of different groups as indicators at broad scales. Prendergast et al. (1993b) tested the coincidence of “hotspots” (areas of extraordinarily high species richness) for birds, butterflies, dragonflies, liverworts, and aquatic angiosperms on a grid of 100 km2 squares covering Britain. They found only weak support for any pairwise concordance and no grid squares that were hotspots for all groups. Pearson and Carroll (1998) compared data on butterfly, bird, and tiger beetle richness for sets of large (approximately 90,000 km2) grid squares in North America, India, and Australia. Their findings were equivocal. For instance, they found that tiger beetles are useful indicators of butterfly species richness, but not bird species richness, in North America. Jaarsveld et al. (1998) studied the distributions of eight taxa (mammals, birds, plants, butterflies, termites, antlions, scarab beetles, and buprestid beetles) on a grid of 625 km2 squares in the Transvaal region of South Africa. For each taxon, a rarity-based selection algorithm identified the minimum set of grid squares required to capture all species in that taxon at least once, thus selecting the most efficient, or complementary, set of grid squares. The authors found that the mean overlap of the resulting complementary sets between pairs of taxa was less than 10%. In these and other studies (e.g., Currie 1991, Balmford and Long 1994, Kerr 1997), the authors made a variety of pairwise comparisons among taxa; however, few studies have examined any one taxon's ability to predict species richness in all other taxa, which is an issue of more direct interest to biologists setting priorities for conservation activities.

In this article, we test not only pairwise correlations among taxa but also the utility of different groups of organisms as indicators of overall species richness. We use a database originally compiled for a World Wildlife Fund (WWF) conservation assessment of North America north of Mexico (Ricketts et al. in press). The data set includes distribution or richness data for almost all species in nine major groups—birds, butterflies, mammals, reptiles, amphibians, land snails, trees, nontree vascular plants, and tiger beetles—totaling over 20,000 species. To our knowledge, this database is the most taxonomically diverse in existence for species distributions at a continental scale. The taxonomic groups included in the database represent plant and animal, large and small, vagile and sessile, homeothermic and poikilothermic, and vertebrate and invertebrate organisms. Indeed, North America is one of the few regions for which such comprehensive data are available.

From the data on these nine taxa, we calculate an index of overall richness, which we use as a surrogate for species richness in all taxa. Because the index includes so many and such diverse taxonomic groups, it may be a more accurate measure of total richness than indexes based on fewer or more closely related taxa (Scott et al. 1993, Sisk et al. 1994, Long et al. 1996). The availability of this index therefore provides a unique opportunity to test not only pairwise correlations among groups, but also the utility of distribution patterns in certain taxa for predicting overall species distribution patterns at broad scales.

We use this data set to address three initial questions. First, to what extent are patterns of species richness in the nine taxa correlated with each other? Second, how well is the pattern of species richness of each taxon correlated with total species richness (as approximated by our overall richness index)? Third, how reliable are birds, butterflies, and mammals, when used together, as proxies for overall richness patterns? The taxonomy and distribution of these three taxa are known well enough to allow their use as indicator groups in most parts of the world; thus, they are used most often as indicators of species distributions in other taxa (Scott et al. 1993, Sisk et al. 1994, Long et al. 1996).

We then examine two additional questions that illuminate two potential pitfalls associated with the use of indicator taxa at broad scales. First, to what degree are any correlations influenced by the effects of latitude on species richness? In other words, would holding latitude constant severely weaken any correlations we observe? If a correlation in richness patterns among taxa were attributable primarily to the strong North American latitudinal gradient, then applying these results to other regions would be misleading. Second, is the residual error of each index-by-taxon relationship distributed at random across North America? Or is there a geographic pattern to each relationship? Even if the species richness pattern in a taxon were correlated with overall richness, a strong geographic pattern of residual error could still result in substantial and nonrandom over- or underprediction in certain areas. Awareness of any patterns underlying the simple correlation coefficient statistic may deepen the understanding of the predictive ability of indicator taxa and improve the ability of conservation biologists to use them wisely.

Ecoregion map and species distributions

The geographic units we used are the 110 terrestrial ecoregions composing the continental United States and Canada (Figure 1). This ecoregion map was first introduced as the framework for a WWF conservation assessment of North America (Ricketts et al. in press).

Figure 1.

Map of the terrestrial ecoregions of the continental United States and Canada. Ecoregion numbers relate to the box on pages 372–373. Ecoregion names and full descriptions are given in Ricketts et al. (in press).

Figure 1.

Map of the terrestrial ecoregions of the continental United States and Canada. Ecoregion numbers relate to the box on pages 372–373. Ecoregion names and full descriptions are given in Ricketts et al. (in press).

Ecoregions are relatively coarse biogeographic divisions of a landscape. They delineate geographically distinct areas that share broadly similar environmental conditions and, often, natural communities. Because of the complexity with which environmental and ecological factors vary across a landscape, most efforts to map ecoregion boundaries combine equal parts quantitative data and gestalt (Bailey 1996). The boundaries are necessarily approximate and usually represent areas of transition rather than sharp divisions. Nevertheless, ecoregions are extremely useful because they allow broad-scale conservation planning, reporting, and monitoring for biologically meaningful geographic units as opposed to arbitrary political boundaries or grid cells. The WWF ecoregions are based on three established ecoregion mapping projects: Omernik (1995), for the conterminous United States; Ecological Stratification Working Group (ESWG 1995), for Canada; and Gallant et al. (1995), for Alaska. During two expert workshops, the WWF project joined the maps into a single international framework and modified certain areas. Details of the mapping methodology can be found in Ricketts et al. (in press).

We collected published and unpublished data on the distribution of species in nine taxonomic groups (Table 1). For vascular plants, John Kartesz and Amy Farstad (North Carolina Botanical Garden) provided a richness estimate for each ecoregion, assembled from their mostly county level database. For land snails, Barry Roth (Museum of Paleontology, University of California-Berkeley) provided richness estimates for the ecoregions west of the Rocky Mountains from his database of museum and research collection localities. For all other taxa (and for eastern land snails), we compared the range data for each species, in the form of published range maps, to the ecoregion map; we then recorded the species as present or absent in each ecoregion. The range maps in many field guides represent only coarse outlines of species occurrence; significant portions of a species' mapped range often do not actually support populations of the species. We therefore used the habitat preference information found in the text of most species accounts to modify the species ranges shown on the range maps. Although this method introduces a certain amount of recording error, any error in our methodology is likely to be of a lower magnitude than the error inherent in the maps themselves (see Ricketts et al. in press for a discussion). Finally, we queried the resulting database to produce a species richness estimate for each taxon in all ecoregions (see box above; distribution maps of species richness for all taxa individually are given in Ricketts et al. in press).

Table 1.

Taxonomic groups used in this analysis, with number of total species considered in each group and sources of distribution data.

Table 1.

Taxonomic groups used in this analysis, with number of total species considered in each group and sources of distribution data.

Ecoregion richness values for each taxon and for the three richness indexes used in the analysis

Ecoregion numbers correspond to the numbers in Figure 1 and follow those in Ricketts et al. (in press). Missing ecoregion numbers (1, 3, 4, 5, 73, and 74) correspond to the six ecoregions in Hawaii and Puerto Rico, which were not considered in the analysis. The “Overall index” uses all nine taxa, whereas Imbb uses mammals, birds, and butterflies only and Ibtr uses butterflies, trees, and reptiles only (see text for details).

Image: bisi.1999.49.5.369-un01.gif

It is worth noting that some of the groups listed in Table 1 are not mono-phyletic, and others do not represent the majority of species in certain taxa. For example, there is controversy over the monophyly of reptiles (BentonandClarkl988,Evansl988), tiger beetles are only a small fraction of beetle diversity in North America (Lawrence and Newton 1982), and trees are a subset of vascular plants. Nevertheless, we grouped the species as we did because distribution data are often collected according to these groups. For instance, tree distributions are often known, whereas those of the rest of the vascular flora are not. To be useful as a guide to the application of indicator taxa, the groups we test must be the same groups for which data are likely to be available in other regions of the world. The richness values we report for vascular plants exclude trees because we grouped trees separately. For convenience, however, we nevertheless refer to this nontree vascular flora as “vascular plants.” Throughout this article, we use the word “taxon” interchangeably with the word “group” to refer to the species groups listed in Table 1.

Correlations in species richness among taxa

We tested pairwise correlations among all taxonomic groups as well as between each taxon and the overall richness index. Because in some taxa the richness values of ecoregions are not distributed normally, we used nonparametric methods to test for correlations.

Pairwise correlations. All pairwise correlations among the nine taxa are significant (Kendall's rank correlation coefficients, n = 110, P < 0.01), and most are quite strong (Table 2). In general, then, major taxa appear to have broadly correlated patterns of species richness across North America. However, at least two additional points in Table 2 are worth noting. First, the four smallest correlation coefficients in the table all involve mammals, a well-known group worldwide. Therefore, one of the most practical choices for an indicator taxon appears to be a relatively poor predictor of species richness in other taxa. Second, the correlation between trees and the rest of the vascular flora, although significant, is relatively weak, even though the two groups are commonly assumed to covary (e.g., Currie 1991).

Table 2.

Kendall's rank correlation coefficients (τ) for all pairwise comparisons among taxa and between each taxon and the overall richness index.a

Table 2.

Kendall's rank correlation coefficients (τ) for all pairwise comparisons among taxa and between each taxon and the overall richness index.a

Correlations between taxa and the overall richness index. As already noted, a more central question than “how effectively do birds predict richness patterns in butterflies?” is “how effectively do birds indicate patterns of species richness overall?” To address this question and to test the ability of different taxa to serve as indicators of overall diversity, we used our database to construct an overall richness index, modified from Sisk et al. (1994). The richness index for each ecoregion, e, is:

where n is the number of taxonomic groups used in the index (maximum 9), Gi(e) is the number of species of group i in the ecoregion, and Gm is the total number of species of group i in the database.

This index computes, for each taxon, the fraction of North American species that is found in an ecoregion and then averages the fractions across all taxa. It therefore gives equal weight to the taxonomic groups rather than to individual species, removing the dominating effect of speciose taxa. (A simple sum or average of all species from the nine groups, by contrast, would result in an index that is swamped by vascular plants, because in most ecoregions the species richness of vascular plants is two orders of magnitude greater than that of any other group.)1 The overall species richness index is included in the box on pages 372–373 and is mapped onto the ecoregions in Figure 2.

Figure 2.

Map of overall richness index. This map incorporates richness values for all nine taxa: mammals, birds, reptiles, amphibians, butterflies, tiger beetles, land snails, trees, and nontree vascular plants.

Figure 2.

Map of overall richness index. This map incorporates richness values for all nine taxa: mammals, birds, reptiles, amphibians, butterflies, tiger beetles, land snails, trees, and nontree vascular plants.

We used this overall index to test the correlation between each taxon and overall species richness. To preserve independence between the correlates (Sokal and Rohlf 1995), we recalculated the overall richness index, excluding the taxon in question before testing each correlation. For example, we removed birds from the richness index before testing it for correlation with birds. We also created a subindex composed of just three taxa—mammals, butterflies, and birds—and tested the correlation between this index and overall species richness. It is these three taxa whose taxonomic and distribution data are probably best known worldwide (Scott et al. 1993, Sisk et al. 1994, Ceballos and Brown 1995, Long et al. 1996), which makes them among the best practical choices for indicator taxa at broad scales. This subindex, Imbb, is calculated in the same way as the overall index, but it includes only mammals, birds, and butterflies. As in the single-taxon tests, the taxa composing Imbb were excluded from the overall index before we tested correlation.

The lowest line of Table 2 shows that each taxonomic group, in addition to being correlated with all other groups, is strongly correlated with the index of overall species richness. Imbb is also strongly correlated with overall species richness. These results suggest that any taxon or combination of well-known taxa could be used as an indicator of total species richness at these coarse scales.

However, species occurrence data such as these are likely to include some degree of spatial autocorrelation, which makes interpreting the significance of correlations problematic (Jongman et al. 1995). A species' range typically overlaps with several ecoregions; therefore, ecoregions are not independent data points in that they do not accrue their richness values independently. With standard statistical techniques (e.g., rank correlations) that assume independence of sample points, spatial autocorrelation inflates the number of degrees of freedom used in significance testing. Carroll and Pearson (1998) have addressed this problem in a clever way by using a geostatistical modeling technique instead of correlations to test for significance. In this article, we provide a simpler, conservative test, which should give a lower bound to the adjusted degrees of freedom due to nonindependence. In addition to testing the significance of the correlations for the observed degrees of freedom (n = 110 ecoregions), we also test it for n = 15, under the arbitrary assumption that spatial autocorrelation reduced the degrees of freedom over sevenfold. Even under this extreme assumption, the correlations between each taxon and the overall richness index remain significant (n = 15; P < 0.01 in all cases; Table 2).

Potentially problematic patterns

The significant correlations between the richness of each taxon and overall species richness suggest a reason for optimism in the search for useful indicator taxa. It appears initially that almost any group could serve well as a surrogate for species richness patterns in other taxa or for overall species richness. Two patterns lying within these correlation coefficients, however, suggest that this conclusion requires careful qualification.

The effect of latitude. To determine how latitude affects species richness in the nine groups, we regressed the species richness values of each taxon by the weighted mean latitude of the ecoregions. Table 3 shows that latitude alone explains between 33% and 74% of the variation in species richness in the nine taxa. The slopes of the richness-by-latitude regression lines vary widely among taxa, but they are all significantly negative (Figure 3). This result indicates that all taxa decrease in richness with increasing latitude, albeit at differing rates.

Table 3.

Results of linear least-squares regressions of taxon richness against area and latitude.a

Table 3.

Results of linear least-squares regressions of taxon richness against area and latitude.a

Figure 3.

Linear regression lines for each taxon regressed on latitude, measured in kilometers north from an arbitrary point off the southern tip of Florida. The slopes differ widely among taxa, but all slopes are negative. The line for vascular plants is not plotted because its intercept is one or two orders of magnitude greater than that of the other taxa, but it too has a negative slope (Table 3).

Figure 3.

Linear regression lines for each taxon regressed on latitude, measured in kilometers north from an arbitrary point off the southern tip of Florida. The slopes differ widely among taxa, but all slopes are negative. The line for vascular plants is not plotted because its intercept is one or two orders of magnitude greater than that of the other taxa, but it too has a negative slope (Table 3).

We used partial correlation analyses to test the correlations in species richness that remain once latitude is removed from the data (Table 4). The correlations between each taxon and the overall richness index are substantially weakened when they are adjusted for latitude (this result is similar to that of Flather et al. [1997] for North American birds, butterflies, and tiger beetles). All correlations are still significant for n = 110, but none remains significant under the reduced degrees of freedom (n = 15). Some pairwise correlations among individual taxa actually become negative. Moreover, among the taxa whose correlations are most reduced are the groups often used as indicators: butterflies, birds, and mammals. The correlation coefficient between the overall richness index and the subindex constructed of only these three taxa (Imbb) weakens from 0.67 (in Table 2) to 0.15 (in Table 4) when latitude effects are removed.

Table 4.

Partial correlation coefficients, with the cirects of latitude removed. Pairwise comparisons among all taxonomic groups are listed, as are correlations between each taxon and the overall richness index (Kendall's rank correlation coefficients x).

Table 4.

Partial correlation coefficients, with the cirects of latitude removed. Pairwise comparisons among all taxonomic groups are listed, as are correlations between each taxon and the overall richness index (Kendall's rank correlation coefficients x).

Therefore, when latitude is held constant, the utility of these nine taxa for indicating overall patterns of richness is severely diminished, especially in the three best-known groups (mammals, birds, butterflies). Because the strong latitudinal gradient in North America is responsible for a large portion of the correlation among taxa, the strength of the correlations in Table 2 may be misleading if they are assumed in other regions of the world.

Geographic patterns of prediction error. When one or a few taxa are used to indicate overall species richness patterns, one pattern is being predicted based on another. In this way, the process can be seen as one of regression, with the taxon in question being the predictor variable and the overall richness index the response variable. We regressed the overall richness index (again excluding the taxon under investigation) against each taxon and assigned each ecoregion a shade of gray based on the sign and magnitude of its residual error. (For these analyses, we used the species richness data without latitude effects removed.) We then shaded the ecoregion map accordingly to reveal any geographic patterns of residuals. Figure 4 provides an illustration of our approach, using butterflies as an example.

Figure 4.

Regression of overall richness index by butterfly richness. Each circle represents an ecoregion, and the heavy line plotted through the middle of the points is the linear regression line. The shading on either side of this line indicates whether the ecoregions have highly positive residuals (black), positive residuals (dark gray), slightly positive or slightly negative residuals (medium gray), negative residuals (light gray), or strongly negative residuals (white). The thresholds between shades of gray were selected arbitrarily and were kept constant among all taxa to clarify comparisons among the taxa.

Figure 4.

Regression of overall richness index by butterfly richness. Each circle represents an ecoregion, and the heavy line plotted through the middle of the points is the linear regression line. The shading on either side of this line indicates whether the ecoregions have highly positive residuals (black), positive residuals (dark gray), slightly positive or slightly negative residuals (medium gray), negative residuals (light gray), or strongly negative residuals (white). The thresholds between shades of gray were selected arbitrarily and were kept constant among all taxa to clarify comparisons among the taxa.

When the shaded ecoregions are mapped, the geographic patterns of residual error become apparent (Figure 5). Based on these patterns, the nine taxa fall into three categories. Taxa of the first category (butterflies, birds, mammals, and vascular plants) have strongly positive residuals (darkest shades) concentrated in the southeastern part of the continent (Figures 5a-5d). In this area, the overall richness index is higher than predicted on the basis of richness in any of these taxa alone. Areas where the overall richness index is lower than predicted (lighter shades) or relatively close to the predicted value (medium gray) are generally found to the north and west of the continent.

Figure 5.

Maps of the residuals from regressions of overall richness index on individual taxa. Shading is assigned as in Figure 4. Dark areas do not indicate where the taxon is high in species richness, but rather where the overall richness index is higher than predicted by the taxon alone. Hatched ecoregions contain no species of the taxon in question. Butterflies (a), birds (b), mammals (c), and nontree vascular plants (d) all fall in the first broad pattern category, in which overall richness is higher than predicted in the southeastern United States. Trees (e), amphibians (f), and land snails (g) all fall in the second broad pattern category, in which overall richness is higher than predicted in the southwestern United States. Reptiles (h) and tiger beetles (i) do not fall in either category.

Figure 5.

Maps of the residuals from regressions of overall richness index on individual taxa. Shading is assigned as in Figure 4. Dark areas do not indicate where the taxon is high in species richness, but rather where the overall richness index is higher than predicted by the taxon alone. Hatched ecoregions contain no species of the taxon in question. Butterflies (a), birds (b), mammals (c), and nontree vascular plants (d) all fall in the first broad pattern category, in which overall richness is higher than predicted in the southeastern United States. Trees (e), amphibians (f), and land snails (g) all fall in the second broad pattern category, in which overall richness is higher than predicted in the southwestern United States. Reptiles (h) and tiger beetles (i) do not fall in either category.

For taxa of the second category (trees, amphibians, and land snails), ecoregions with positive residuals (i.e., overall richness higher than predicted) are generally concentrated in the central and southwestern region of the United States (Figures 5e-5g). The third category consists of two taxa—reptiles and tiger beetles—each of which shows a pattern that prevents inclusion in either the first or second general category. Both of these taxa tend to under-predict overall richness in the southeastern United States and in the Rocky Mountains and the West Coast of the United States (Figures 5h, 5i).

Even without detailed knowledge of the mechanisms responsible for these differences in pattern, a conservation lesson applicable to the use of indicator taxa can be gleaned from the patterns themselves. The three best-known taxa worldwide—mammals, birds, and butterflies—all show the same prediction bias pattern (Figures 5a-5c). Thus, the richness index constructed of these three groups (Imbb) only reinforces the geographic bias already present in each of the three individual taxa (Figure 6).

Figure 6.

Map of the residuals from regression of overall richness index by Imbb (a subindex composed of mammals, birds, and butterflies). Shading is assigned as in Figure 4.

Figure 6.

Map of the residuals from regression of overall richness index by Imbb (a subindex composed of mammals, birds, and butterflies). Shading is assigned as in Figure 4.

However, these categories suggest another way to create a subindex: by choosing a taxon from each of the three geographic bias categories. Accordingly, we construted an analogous subindex using butterflies, trees, and reptiles (Ibtr). The residuals generated by regressing overall richness on Ibtr show a much weaker geographical pattern of error (Figure 7) than the residuals generated by regressing overall richness on Imbb (Figure 6) or on any single taxon (Figure 5). Furthermore, Ibtr improves on Imbb in two other respects: It is correlated more strongly with the overall richness index (τ = 0.81; Table 2), and it remains strongly correlated with overall richness (τ = 0.54) even when latitude is held constant (Table 4).

Figure 7.

Map of the residuals from regression of overall richness index by Ibtr (a subindex com posed of butterflies, trees, and reptiles). The pattern of residuals is weaker than in Figure 6. Shading is assigned as in Figure 4.

Figure 7.

Map of the residuals from regression of overall richness index by Ibtr (a subindex com posed of butterflies, trees, and reptiles). The pattern of residuals is weaker than in Figure 6. Shading is assigned as in Figure 4.

In selecting butterflies, trees, and reptiles to compose a more balanced index, we have attempted to maintain practicality and improve accuracy by choosing the best-known taxon in each of the bias pattern categories. The point of this article, however, is not to advocate the immediate and worldwide use of butterflies, trees, and reptiles as indicators of species richness. Rather, our example serves to illustrate a more general point: there are probably geographic patterns to the error in prediction of any indicator taxon or index. Recognizing these patterns will help conservation biologists choose taxa that will combine to form a more accurate surrogate measure for overall richness.

Applicability to other regions

This analysis has allowed us to test some general ideas about indicator taxa because distribution data on a large number of species are available for the United States and Canada. But a key question remains: to what extent can these results be applied to other regions, where range data for the majority of species (including those used in this article) are poorly known and where indicator taxa would be especially useful?

At the continental scale, many factors may influence the distribution patterns of species. Every continent has a unique history of climate, isolation, topography, and evolution, each of which may affect different taxa in different ways. For example, the Pleistocene glaciations in North America and resulting shifts in vegetation communities have left a distinct mark on current biogeographic patterns (Braun 1950, Graham 1988, Pielou 1991). Taxa found to be useful indicators on one continent may, therefore, be difficult to apply with confidence in other parts of the world. Indeed, Pearson and Carroll (1998) found that the best indicators for predicting richness in tiger beetles, birds, and butterflies differed among the United States, India, and Australia. The differing tectonic histories and current climates of these three regions make Pearson and Carroll's (1998) work an extreme (but appropriate) test of the applicability of results from one part of the world to another. It is likely that for regions with climates more similar to that of North America (such as Europe, temperate Asia, or temperate South America), the relationships described in this article will be more applicable. Nevertheless, Pearson and Carroll's (1998) results suggest that significant relationships found in one region do not guarantee the general utility of particular taxa as indicators of overall species richness.

Factors that vary within a continent, however, may be useful in illuminating some general determinants of species richness. For example, we have shown that latitude is an important factor in North America. Other studies have shown it to be almost universally important (Brown and Gibson 1983, Currie 1991, Lawton et al. 1993, Begon et al. 1996, Flather et al. 1997). However, after holding latitude constant, several pairs of taxa in our dataset remain significantly correlated. These taxa presumably respond in a similar manner to other determinants of species richness. Many physical and climatic variables have been proposed to explain broad-scale patterns of species richness, including longitude, precipitation, mean annual insolation, elevation, topographic heterogeneity, potential evapotranspiration, and peninsular location (Currie 1991, Lawton et al. 1993, Begon et al. 1996, Bóhning-Gaese 1997). Understanding the causal factors of regional species richness in different groups may make it possible to choose appropriate indicator taxa for the region of interest (Flather et al. 1997).

Another consideration when applying our results to other regions is the issue of spatial scale. Several authors have shown that the correlative relationships among taxa are sensitive to the scale of observation (e.g., Weaver 1995, Bóhning-Gaese 1997). At the broad scale of our analysis, physical and climatic factors such as mean annual precipitation and potential evapotranspiration are likely to be important. At finer scales, factors such as resource specificity and habitat diversity probably dominate. Taxa will respond differently to the most influential factors at these two scales, depending on habitat specificity, vagility, and other natural-history characteristics. Ideally, it would be helpful to understand how these differing responses affect the strength of correlations among taxa as the scale varies. Generally, though, our results are probably best applied in other temperate regions at a similar spatial scale. In tropical areas and at substantially finer scales, the relationships we report are less likely to be useful.

Biases in species richness data

The data we have used for these analyses (like all species richness data derived from compiled species occurrence records) may suffer from at least two biases that could inflate our correlation results. First, the ecoregions we used in this study span three orders of magnitude in size, from 103 km2 to 106 km2. Hundreds of studies have shown that, as the sample area increases, the species richness of the sample generally increases (e.g., MacArthur and Wilson 1967, Rosenzweig 1995). Therefore, it is possible that a substantial portion of the correlation we observed among taxa is due to the wide range of ecoregion sizes in our study, with large ecoregions relatively species rich in all groups and small ecoregions relatively depauperate in all groups. However, we found no significant relationship between ecoregion area and species richness in any taxon (Table 3). This counterintuitive result may be due in part to the trend toward larger ecoregions as latitude increases (Figure 1). An imposed positive correlation between latitude and ecoregion area may counteract the negative correlation between latitude and species richness, obscuring the often-observed relationship between species richness and area. However, in our study latitude explains only 4% of the variation in ecoregion area, as compared to 32–74% of the variation in species richness (Table 3).

More likely, the delineation of the ecoregions themselves has made species-area laws less applicable. Ecoregions by definition delineate distinct groups of related ecosystems, some small and highly diverse, others large and relatively uniform. Thus, the delineation process may have partially normalized the species richness of ecoregions by isolating small, diverse areas as separate ecoregions (e.g., ecoregion 72, California coastal sage and chaparral) and by lumping large areas of relatively uniform habitat into single ecoregions (e.g., ecoregion 58, Northern short grasslands). Whatever the reason, statistically removing area from the richness data before testing for correlation would not alter our results significantly.

A second bias could result from a common problem with the use of species richness data: collection effort is not equal among sampling units, because biologists are likely to have spent more effort compiling species occurrence data in some areas than in others. This bias can, like an area effect, inflate intertaxon correlations in species richness. Several studies have sought ways to correct for this disparity in sampling effort (e.g., Prendergast et al. 1993a, Fagan and Kareiva 1997). Although correcting for this bias would be ideal, it would be difficult to measure total collecting effort for each ecoregion. In addition, it is unlikely that the data in this analysis suffer substantially from this bias, for two main reasons. First, we relied mostly on range map data, instead of compiled species lists for each ecoregion. Because many range maps tend to “fill in” gaps to make ranges more continuous, species are often assumed to be present in poorly sampled areas if they are surrounded by known occurrences. Second, several of the taxa we used (birds, mammals, trees, and butterflies) have been studied thoroughly in North America and are therefore unlikely to be absent from ecoregions because of insufficient sampling effort. Correlations among these taxa are not systematically lower than those among the less well known groups, suggesting that correlations among the less well known groups are unlikely to be substantially inflated by sampling bias.

Conclusions

Selecting indicator taxa involves balancing the dual requirements of practicality and accuracy. An indicator must, by definition, be relatively well known or at least easy to study, but it also must be informative, accurate, and reliable. The results presented in this article suggest difficulties in the search for indicators of broad-scale patterns of species diversity. The three most practical (i.e., best-known) choices for indicator taxa worldwide are not the most (indeed, are often the least) informative and accurate indicators, when employed either individually or in combination. Interestingly, the problems with using these taxa as indicators are not immediately apparent from simple analyses. Results of correlation analyses initially suggested that any of the taxa or composite indexes we tested are effective indidators of overall species richness at coarse (i.e., ecoregion) scales. However, two subtler features of these relationships—that is, latitude effects and geographic patterns of prediction error—indicated that this initial conclusion requires careful qualification.

One common application of documented patterns of species richness is to help establish conservation priorities (Scott et al. 1993, Sisk et al. 1994). Clearly, however, conservation biologists should take into account more than richness “hotspots” when establishing priorities for conservation. Data on species richness should be combined with information on centers of species endemism, complementarity with existing reserves, richness and endemism of higher taxa, levels of threat, and ecosystem service value. Overall species richness, however, will continue to be important in selecting conservation priorities. Understanding the patterns illuminated in this article may help conservation biologists make better choices of taxa for use as proxies for overall species richness and to make more informed decisions, with more confidence and efficiency, in the face of ever-increasing conservation challenges.

Acknowledgments

We thank Karen Carney, Prashant Hedao, Patrick Hurley, and John Fay for help with the species data and GIS analysis. John Kartesz and Amy Farstad graciously provided unpublished data on vascular plants, as did David Pearson at Arizona State University for tiger beetles and Barry Roth for land snails. Gerardo Ceballos, Andy Weiss, and Stu Weiss lent many valuable ideas early in the project, and Carol Boggs, Karen Carney, Gretchen Daily, Paul Ehr-lich, Jennifer Hughes, Alan Launer, Reed Noss, Tore Schweder, Peter Vitousek, and two reviewers improved the manuscript immensely with their comments. This work was supported at World Wildlife Fund by the Commission for Environmental Cooperation, the Environmental Protection Agency, Melvin Lane, and an Armand G. Erpf Conservation Fellowship to E. D. Support at Stanford University was provided by the Koret Foundation, Peter and Helen Bing, and a National Science Foundation Graduate Fellowship to T. H. R.

References cited

Bailey
RG
.
1996
.
Ecosystem Geography
.
New York
:
Springer-Verlag
.
Balmford
A
Long
A
.
1994
.
Across-country analyses of biodiversity congruence and current conservation effort in the tropics
.
Conservation Biology
 
9
:
1539
1547
.
Banfield
AWF
.
1974
.
The Mammals of Canada
.
Toronto (Canada)
:
University of Toronto Press
.
Beccaloni
GW
Gastón
KJ
.
1995
.
Predicting the species richness of neotropical forest butterflies: Ithomiinae (Lepidoptera: Nymphalidae) as indicators
.
Biological Conservation
 
71
:
77
86
.
Begon
M
Harper
JL
Townsend
CR
.
1996
.
Ecology: Individuals, Populations, and Communities
.
Boston
:
Blackwell Science
.
Benton
MJ
Clark
J
.
1988
.
Archosaur phylogeny and the relationships of the Crocodylia
. Pages
295
338
in
Benton
MJ
, ed.
The Phylogeny and Classification of the Tetrapods
 .
New York
:
Oxford University Press
.
Bóhning-Gaese
K
.
1997
.
Determinants of avian species richness at different spatial scales
.
Journal of Biogeography
 
24
:
49
60
.
Braun
LE
.
1950
.
Deciduous forests of eastern North America
.
Philadelphia
:
Blakiston
.
Brown
JH
Gibson
AC
.
1983
.
Biogeography
.
St. Louis (MO)
:
C. V. Mosby
.
Burt
HB
.
1980
.
A Field Guide to the Mammals: North America North of Mexico
.
New York
:
Houghton Mifflin
.
Carroll
SS
Pearson
DL
.
1998
.
Spatial modelling of butterfly species richness using tiger beetles (Cicindelidae) as a bio-indicator taxon
.
Ecological Applications
 
8
:
531
543
.
Ceballos
G
Brown
JH
.
1995
.
Global patterns of mammalian diversity, endemism, and endangerment
.
Conservation Biology
 
9
:
559
568
.
Conant
R
Collins
JT
.
1991
.
A Field Guide to Reptiles and Amphibians: Eastern/Central North America
.
Boston
:
Houghton Mifflin
.
Cook
FR
.
1984
.
Introduction to Canadian Amphibians and Reptiles
.
Ottawa (Canada)
:
National Museum of Natural Sciences, National Museums of Canada
.
Currie
DJ
.
1991
.
Energy and large-scale patterns of animal- and plant-species richness
.
The American Naturalist
 
137
:
27
49
.
Daily
GC
, ed.
1997
.
Nature's Services: Societal Dependence on Natural Ecosystems
.
Washington (DC)
:
Island Press
.
Dinerstein
E
Olson
DM
Graham
DJ
Webster
A
Primm
SA
Bookbinder
MP
Ledec
G
.
1995
.
A conservation assessment of the terrestrial ecoregions of Latin America and the Caribbean
.
Washington (DC)
:
The World Bank
.
Dobbyn
J
.
1994
.
Atlas of the Mammals of Ontario
.
Toronto (Canada)
:
Federation of Ontario Naturalists
.
[ESWG] Ecological Stratification Working Group
.
1995
.
A National Ecological Framework for Canada
.
Ottawa/Hull (Canada)
:
Agriculture and Agri-food Canada, Research Branch, Centre for Land and Biological Resources Research; and Environment Canada, State of the Environment Directorate, Ecozone Analysis Branch
.
Ehrlich
PR
Wilson
EO
.
1991
.
Biodiversity studies: Science and policy
.
Science
 
253
:
750
752
.
Evans
SE
.
1988
.
The early history and relationships of the Diapsida
. Pages
221
260
in
Benton
MJ
, ed.
The Phylogeny and Classification of the Tetrapods
 .
New York
:
Oxford University Press
.
Fagan
WF
Kareiva
PM
.
1997
.
Using compiled species lists to make biodiversity comparisons among regions: A test case using Oregon butterflies
.
Biological Conservation
 
80
:
249
259
.
Farrar
JL
.
1995
.
Trees in Canada
.
Ottawa (Canada)
:
Fitzhenry and Whiteside Ltd. and the Canadian Forest Service
.
Flather
CH
Wilson
KR
Dean
DJ
McComfc
WC
.
1997
.
Identifying gaps in conservation networks: Of indicators and uncertainty in geographic-based analyses
.
Ecological Applications
 
7
:
531
542
.
Forsyth
A
.
1985
.
Mammals of the Canadian Wild
.
Camden East (Canada)
:
Camden House Publishing
.
Gallant
AL
Binnian
EF
Omernik
JM
Shasby
MB
.
1995
.
Ecoregions of Alaska
.
Washington (DC)
:
US Geological Survey
.
Graham
RW
.
1988
.
The role of climate change in the design of biological reserves: The paleoecology perspective for conservation biology
.
Conservation Biology
 
2
:
391
394
.
Hubricht
L
.
1985
.
The distributions of the native land mollusks of the eastern United States
.
Fieldiana Zoology
 
24
:
1
191
.
Jaarsveld
AS
, et al.
1998
.
Biodiversity assessments and conservation strategies
.
Science
 
279
:
2106
2108
.
Jongman
RHG
Ter Braak
CJF
Van Tongeren
OFR
, eds.
1995
.
Data Analysis in Community and Landscape Ecology
.
New York
:
Cambridge University Press
.
Kaiser
J
.
1997
.
Unique, all-taxa survey in Costa Rica ‘self-destructs’
.
Science
 
276
:
893
.
Kerr
JT
.
1997
.
Species richness, endemism, and the choice of areas for conservation
.
Conservation Biology
 
11
:
1094
1100
.
Lawrence
JF
Newton
AFJ
.
1982
.
Evolution and classification of beetles
.
Annual Review of Ecology and Systematics
 
13
:
261
290
.
Lawton
JH
Nee
S
Letcher
AJ
Harvey
PH
.
1993
.
Animal distributions: Patterns and processes
. Pages
41
58
in
Edwards
PJ
May
RM
Webb
NR
, eds.
Large-Scale Ecology and Conservation Biology
 .
Oxford
:
Blackwell Science
.
Lawton
JH
Bignell
DE
Bolton
B
Bloemers
GF
Eggleston
P
Hammond
PM
Hodda
M
Holt
RD
Laresen
TB
.
1998
.
Biodiversity inventories, indicator taxa and effects of habitat modification on tropical forest
.
Nature
 
391
:
72
76
.
Little
EL
Jr
.
1976a
.
Conifers and Important Hardwoods
.
Washington (DC)
:
US Department of Agriculture, Forest Service
.
Little
EL
Jr
.
1976b
.
Minor Western Hardwoods
.
Washington (DC)
:
US Department of Agriculture, Forest Service
.
Little
EL
Jr
.
1977
.
Minor Eastern Hardwoods
.
Washington (DC)
:
US Department of Agriculture, Forest Service
.
Long
AJ
Crosby
MJ
Stattersfield
AJ
Wege
DC
.
1996
.
Towards a global map of biodiversity: Patterns in the distribution of restricted-range birds
.
Global Ecology and Biogeography Letters
 
5
:
281
304
.
MacArthur
RH
Wilson
EO
.
1967
.
The Theory of Island Biogeography
.
Princeton (NJ)
:
Princeton University Press
.
Myers
N
.
1988
.
Threatened biotas: ‘Hot spots’ in tropical forests
.
The Environmentalist
 
8
:
187
208
.
Myers
N
.
1994
.
Biodiversity II: Losses
. Pages
380
BioScience
Vol.
49
No. S110–142 in
Meffe
GK
Carroll
CR
, eds.
Principles of Conservation Biology
 .
Sunderland (MA)
:
Sinauer Associates
.
Olson
DM
Dinerstein
E
.
1998
.
The Global 200: A representation approach to conserving the Earth's most biologically valuable ecoregions
.
Conservation Biology
 
12
:
502
515
.
Omernik
JM
.
1995
.
Level III Ecoregions of the Continental United States
.
Washington (DC)
:
US Environmental Protection Agency
.
Pearson
DL
Carroll
SS
.
1998
.
Global patterns of species richness: Spatial models for conservation planning using bio-indicators and precipitation
.
Conservation Biology
 
12
:
809
821
.
Pearson
DL
Barraclough
T
Vogler
AP
.
1997
.
Distributional maps for North American species of tiger beetles (Coleóptera: Cicindelidae)
.
Cicindela
 
29
:
33
84
.
Petrides
AP
.
1988
.
A Field Guide to Eastern Trees: Eastern United States and Canada
.
New York
:
Houghton Mifflin
.
Petrides
AP
.
1992
.
A Field Guide to Western Trees: Western United States and Canada
.
New York
:
Houghton Mifflin
.
Pielou
EC
.
1991
.
After the Ice Age: The Return of Life to Glaciated North America
.
Chicago
:
University of Chicago Press
.
Prendergast
JR
Wood
SN
Lawton
JH
Eversham
BC
.
1993a
.
Correcting for variation in recording effort in analyses of diversity hotspots
.
Biodiversity Letters
 
1
:
39
53
.
Prendergast
JR
Quinn
RM
Lawton
JH
Eversham
BC
Gibbons
DW
.
1993b
.
Rare species, the coincidence of diversity hot-spots and conservation strategies
.
Nature
 
365
:
335
337
.
Ricketts
TH
, et al. In press.
Terrestrial Ecoregions of North America: A Conservation Assessment
.
Washington (DC)
:
Island Press
.
Rosenzweig
ML
.
1995
.
Species diversity in space and time
.
Cambridge (UK)
:
Cambridge University Press
.
Scott
JA
.
1986
.
The Butterflies of North America: A Natural History and Field Guide
.
Stanford (CA)
:
Stanford University Press
.
Scott
MJ
, et al.
1993
.
Gap analysis: Protecting biodiversity using geographic information systems
.
Wildlife Monographs
 
123
:
1
41
.
Scott
SL
.
1995
.
Field Guide to the Birds of North America
.
Washington (DC)
:
National Geographic Society
.
Sisk
TD
Launer
AE
Switky
KR
Ehrlich
PR
.
1994
.
Identifying extinction threats: Global analyses of the distribution of biodiversity and the expansion of the human enterprise
.
BioScience
 
44
:
592
604
.
Sokal
RR
Rohlf
FJ
.
1995
.
Biometry
.
New York
:
W.H. Freeman
.
Stebbins
RC
.
1985
.
A Field Guide to Western Reptiles and Amphibians
.
New York
:
Houghton Mifflin
.
Weaver
JC
.
1995
.
Indicator species and the scale of observation
.
Conservation Biology
 
9
:
939
942
.
Williams
PH
Gastón
KJ
.
1994
.
Measuring more of biodiversity: Can higher-taxon richness predict wholesale species richness?. Biological Conservation
 
67
:
211
217
.
Williams
PH
Gibbons
D
Margules
CR
Rebelo
A
Humphries
CJ
Pressey
RL
.
1996
.
A comparison of richness hotspots, rarity hotspots, and complementarity areas for conserving diversity of British birds
.
Conservation Biology
 
10
:
155
174
.
Wilson
EO
, ed.
1988
.
Biodiversity
.
Washington (DC)
:
National Academy Press
.
1
An alternative index that also weights taxa equally is where n is the number of taxonomic groups used in the index, Gi(e) is the number of species of group i in ecoregion e, and G, and SGi are the mean species richness and standard deviation, respectively, of group i across all ecoregions. In this index, the frequency distribution of species richness values for each taxon is standardized to be of mean 0, variance 1, and then the taxa are averaged. We repeated all analyses, substituting this index for the original, and found that all results are almost exactly the same (results not presented here). This result is encouraging; it suggests that our findings are robust to other measures of overall richness and are not due simply to our choice of index.

Author notes

1
(ricketts@leland.stanford.edu) Doctoral Student and Research Scientist at the Center for Conservation Biology, Department of Biological Sciences, Stanford University, Stanford, CA 94305-5020. He is interested in patterns of species diversity at a range of spatial scales in both natural and human-dominated landscapes, the mechanisms generating these patterns, and how they may better inform conservation efforts.
2
Director and Chief Scientist, Conservation Science Program, World Wildlife Fund-US, Washington, DC 20037-1175. His work focuses on innovative methods of setting continental and global conservation priorities in Latin America, North America, southern Asia, Africa, and the world's oceans.
3
Senior Scientist, Conservation Science Program, World Wildlife Fund-US, Washington, DC 20037-1175. His work focuses on innovative methods of setting continental and global conservation priorities in Latin America, North America, southern Asia, Africa, and the world's oceans.
4
Conservation/Gis Analyst, Conservation Science Program, World Wildlife Fund-US, Washington, DC 20037-1175. His work focuses on innovative methods of setting continental and global conservation priorities in Latin America, North America, southern Asia, Africa, and the world's oceans.