By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
*Correspondence: David J. Currie, Ottawa-Carleton Institute of Biology, University of Ottawa, Ottawa, Ontario, Canada K1N 6N5. E-mail: firstname.lastname@example.org
Aim Anthropogenic habitat loss is usually cited as the most important cause of recent species’ extinctions. We ask whether species losses are in fact more closely related to habitat loss than to any other aspect of human activity such as use of agricultural pesticides, or human population density (which reflects urbanization).
Methods We statistically compared areas in Canada where imperiled species currently occur, versus areas where they have been lost. Using multiple regressions, we relate the numbers of species that had suffered range reductions in an ecoregion to variables that represent present habitat loss, pesticide use and human population density.
Results We find high losses of imperiled species in regions with high proportions of agricultural land cover. However, losses of imperiled species are significantly more strongly related to the proportion of the region treated with agricultural pesticides. The relationship between species losses and area treated with pesticides remains significant after controlling for area in agriculture.
Main conclusions Our results are consistent with the hypothesis that agricultural pesticide use, or something strongly collinear with it (perhaps intensive agriculture more generally), has contributed significantly to the decline of imperiled species in Canada. Habitat conversion per se may be a less important cause of species declines than how that converted habitat is used.
Much of the evidence supporting this generalization has been based on studies of individual endangered species. Such studies typically identify multiple contributing factors to a species’ decline. These virtually always include some aspect of anthropogenic habitat loss among those threats (Primack, 1995; Wilcove etal., 1998). For example, Czech etal. (2000) reviewed the accounts of the threats to 877 US endangered species, and they concluded ‘Collectively, the studies have shown that habitat loss is the most prevalent cause of species endangerment’. A similar study in Canada concluded that habitat loss affected 84% of endangered species and was the greatest overall cause of endangerment based on the threats given when the species were listed as being endangered (Venter etal., 2006). Kerr & DeGuise (2004) found that the numbers of endangered species that occurred in 15 Canadian ecozones were related to the extent of broad-scale habitat conversion. Certainly, it is a truism that elimination of a species’ habitat leads to extirpation of the species in the wild. Based on this logic, both governmental and private efforts to preserve imperiled species typically target habitat preservation (e.g. the Endangered Species Act in the United States and the recently enacted Species at Risk Act in Canada; The Nature Conservancy, Wildlife Habitat Canada).
Curiously, although agriculture is often cited as one of the main threats to endangered species, there is rarely specific mention of pesticides. Czech etal. (2000) list agriculture as the third most frequent contributor to species declines in the USA (after non-native species and urbanization), but they do not mention pesticides. Venter etal. (2006) list habitat lost as the most prevalent threat to endangered species, with agriculture and urbanization as the most common human activities contributing to habitat loss. Again, they do not mention pesticides.
Hypotheses regarding which human modifications of the environment pose the greatest threat to imperiled species have been examined by relating broad-scale patterns of endangerment to habitat characteristics. Broad-scale multispecies studies to date have focused on identifying hot spots of endangered species (e.g. Dobson etal., 1997; Flather etal., 1998; Kerr & Deguise, 2004). However, hot spots of endangered species could result from several processes. They may be places where some factor(s) is (are) causing species to become endangered. Alternatively, they may be places where endangered species have avoided extinction, having been lost elsewhere (e.g. Channell & Lomolino, 2000). Knowing where the most endangered species remain may be relevant to the establishment of reserves. Knowing what distinguishes places where species persist from places where they do not persist is relevant to establishing what causes species losses in the first place.
The objectives of this study are to determine how areas where imperiled species persist differ from areas where imperiled species have suffered serious range reductions. We examine this question by comparing imperiled species losses in ecoregions across southern Canada. More specifically, are there hot spots of species losses? Are species losses most closely related to habitat loss (to urban and agricultural development), use of agricultural pesticides, or human population density (which reflects urbanization)? It is unquestionable that many factors contribute to the loss of particular species (e.g. Czech etal., 2000); in this study we are looking for the strongest effects that are most consistent among imperiled species in general. This study is the first, to our knowledge, that quantifies the spatial variation in numbers of species losses (versus the number of extant imperiled species) and that statistically examines correlates of those losses.
Species distribution data
Species distribution data were obtained from reports prepared for the Committee on the Status of Endangered Wildlife in Canada (COSEWIC), the body that officially designates the conservation status (e.g. threatened, endangered) of species in Canada. To determine where species have been lost, both historic and current distribution data are required. Unfortunately, the historic, and sometimes current, distributions of most COSEWIC-listed plants and aquatic taxa are unknown. This study was therefore limited to terrestrial mammals, birds (breeding distribution only), amphibians, and reptiles whose historic and current range distributions have been confidently described or mapped. The most recently reported distribution of a species in the COSEWIC reports was defined as the current distribution (generally between the 1980s and the late 1990s). The historic distribution of a species was acquired from the earliest known or recorded range distribution and dated anywhere from the early 1800s to the early 1900s.
Within the four major taxa examined, those species listed as extinct, extirpated, endangered, threatened or vulnerable/rare were included in this study. Subspecies or specific populations of species were not included in our analysis unless only one subspecies or population of a species had ever existed in Canada, or if Canadian distribution data for the remainder of the species were included in the subspecies or population report. Our study only considered the Canadian portions of species’ distributions. Consequently, in any ecoregion that straddles the US–Canadian border, we consider imperiled species declines, agriculture and pesticide use only in the Canadian area (mainly since US and Canadian ecoregions are not coordinated). See Appendix S1 in Supporting Information for a list of all species included in our analyses.
The spatial resolution of our study was the terrestrial ecoregion, a subunit of the coarser ecozones that delineate areas of reasonably homogeneous physical and biotic characteristics (Ecological Stratification Working Group, 1995; Fig. 1). Although fine-scale information on species distributions is often lacking, presence/absence and range reduction data at the ecoregion scale are reliable. For each ecoregion, we tallied the number of imperiled species currently extant, the number historically present, and the number that have suffered significant range losses.
Ecoregion attributes and land-cover data
Ecoregion attributes and land-cover data were obtained from several sources. Land cover data for Canada were compiled and classified by Marshall etal. (1999) in pixels of 1.1 km2. They distinguished natural cover versus land dominated by built-up areas (e.g. towns, roads, industrial), croplands, or domestic livestock rangelands. Habitat loss was measured as the area of human-dominated land cover, which includes all land cover classifications other than natural cover. Human-dominated land cover is an imperfect surrogate for habitat loss but at broad scales and for broad taxa, it is the best approximation available. We also obtained estimates from E-Stat 1999 (Statistics Canada, 1999) of the areas of croplands, the area treated with insecticide, and the area treated with herbicide in each ecoregion. These estimates were based on the 1999 release of the 1996 Census of Agriculture. The Census of Agriculture was carried out simultaneously with the national Census of Population. Any household that responded positively to the question, ‘Is anyone in this household a farm operator?’ was asked to complete the Census of Agriculture form. Details can be found at http://www.statcan.ca/english/freepub/95F0301XIE/about.htm.
Following Mineau & Whiteside (2006), we use area treated with pesticide as the primary independent variable in this study. In principle, an estimate of the total toxicity of pesticides applied per ecoregion would have been preferable to the area treated with pesticide. Toxicity may vary according to the quantity of pesticide applied, the formulation of the pesticide (aqueous, dry powder, granular, etc.), or carriers applied with the pesticide. Toxicity will also vary among the imperiled species in question. To combine amounts of pesticide used, beyond area treated with pesticides, would have required much more information than was available for a very broad-scale study.
The landscape data used here are recent (1990s), and they therefore do not directly measure historic landscape characteristics. However, historic land use data are rare. Because colonial settlement and wide-scale human land use are relatively recent in Canada (population in 1760 was less than 100,000; Wynn, 1991), we assume that the differences in current landscape characteristics among ecoregions are proportional to those that existed while species’ declines occurred. This assumption argues that, to a first approximation, the ranking of agricultural intensity among ecoregions has remained fairly constant: only a small proportion of natural areas and agricultural areas have traded places. Where this assumption is not true, it will tend to add noise to the data, obscuring statistical relationships.
Using multiple regressions, we tested the hypothesis that the numbers of species that have suffered range reductions in an ecoregion are related to variables that represent present habitat loss, pesticide use and human population density. We carried out these regressions using the number of bird, mammal, amphibian and reptile species combined, as well as for birds and mammals separately. The numbers of species of amphibians and reptiles were too small for independent statistical analysis.
As our main hypothesis dealt with effects of pesticides versus habitat loss, we related numbers of species lost to one variable describing land cover conversion, one pesticide variable, and total regional species richness. Land cover was described by either land in agriculture, or by total human-dominated cover (agriculture plus urban). Pesticides variables were area treated with herbicides, area treated with insecticides, or total area treated with pesticides (herbicide + insecticides + fungicides). We did not include other combinations of variables because these variables are strongly collinear (Appendix S2). Tolerance values of the independent variables in these regressions (i.e. the proportion of their variance that is uncorrelated with other independent variables) were reasonably high (> 0.6). We excluded the 131 ecoregions in northern Canada in which neither habitat loss nor pesticide use was detectable, because habitat loss and pesticide use are perfectly collinear in those samples. This left 86 southern ecoregions with a total area of 3.6 × 106 km2. Note that collinearity between pesticide use and habitat loss in the remaining data reduces the probability of detecting effects of pesticides after controlling for area in agriculture. Thus, our tests of the pesticide hypothesis are conservative.
Total regional species richness was included in the regression models since the number of species lost from an ecoregion seemed likely to depend on the number of species originally present in that region. Bird species richness per ecoregion was compiled from feral distributions and was provided by Parks Canada. Mammal, amphibian and reptile species richness per mainland ecoregion was estimated from feral regional distribution data compiled by Currie (1991). Richness in island ecoregions was estimated from feral distributions in Banfield (1974, mammals) and Cook (1984, amphibians and reptiles). Using species richness as a covariate in regression models serves the same function as expressing the number of species losses as a proportion of total richness, but it avoids using a ratio as the dependent variable.
Many of the variables in this study were strongly positively skewed. We therefore used power transformations to make the distributions of both dependent and independent variables as close to Gaussian as possible: X′ = (X + 0.5Xmin)a, where Xmin is the limit of detection of X (e.g. for species counts, Xmin = 1 species). The coefficient a was iteratively varied from 0 to 1, and agreement with the Gaussian distribution was assessed with a Komolgorov–Smirnov test.
A further statistical complication is that ecoregions do not necessarily accrue or lose species independently. This may produce spatial autocorrelation in the data (i.e. ecoregions may not be independent data points). The predictor variables in this study are also spatially structured. Since we hypothesize that environment is driving species losses, we first carried out ordinary least squares regression, and we then tested for residual autocorrelation in the residuals (Legendre, 1993). We carried out conditional autoregressive regressions using SAM (Spatial Analysis in Macroecology; Rangel etal., 2006) on the final models to test whether any contributing variables become non-significant when the spatial structure of the data is taken into account.
Statistical analyses were performed using systat version 10 and SAM.
A total of 62 COSEWIC-listed species of birds (n = 37), mammals (n = 12), amphibians (n = 6) and reptiles (n = 7) whose historic and current range distributions are known were identified for this study. Of these, approximately 9% were listed as extinct or extirpated, 31% listed as endangered, 24% listed as threatened and 36% listed as vulnerable/rare (COSEWIC, 2004). These numbers represent approximately 60% of the species in these groups listed by COSEWIC. Birds are the best represented with over 80% of COSEWIC-listed species included in this data set, and amphibians are the least with approximately 40% of the listed species represented here. Excluded species were those for which historic distributions could not be accurately determined.
COSEWIC status reports do not explicitly mention pesticides as a major threat. Among the COSEWIC-listed terrestrial vertebrate species, land conversion to agriculture was listed as a threat for 64% of species. ‘Agricultural pollutants’ were cited as a threat for only 21% of species.
Distributions and losses of COSEWIC-listed species
The geographical locations of hot and cold spots of species’ losses, with several of their respective and agriculture attributes are presented in Table 1. Hot spots are areas where many imperiled species have suffered historic range losses. Cold spots of losses are ecoregions in which many imperiled species currently persist and few have experienced historic range losses.
Table 1. Location, landcover and agricultural pesticide data for hot and cold spots of losses of COSEWIC-listed species. Hot spots are areas where many imperiled species have suffered historic range losses. Cold spots of losses are areas where many imperiled species currently persist and have not experienced historic range losses. Landcover, pesticide and population data are from 1991 Statistics Canada census (Statistics Canada, 1999).
Hot or cold spot
Area (103 km2)
% human landuse
% treated with herbicides
% treated with insecticides
Population density (persons km−2)
Mixedwood Plains (2 southern ecoregions)
Mixedwood Plains (southernmost ecoregion)
Prairies/southern Boreal Plains
Pacific Maritime (southwestern tip)
Pacific Maritime (southern coast and ranges of mainland)
High concentrations of imperiled species were apparent in several southern ecoregions of Canada (Fig. 2a). The most prominent hot spot of listed species is in the Mixedwood Plains ecoregion of southern Ontario and an adjacent ecoregion of the Boreal Shield ecozone. High numbers of imperiled species were also found in the Prairies and Boreal Plains ecozones, in the southern portion of the Montane Cordillera, and in the southwest corner of the Pacific Maritime ecozone.
The greatest number of losses of imperiled species occurred in the two southernmost ecoregions of the Mixedwood Plains ecozone; 30 of the 62 species included in this study suffered significant range losses in the southernmost ecoregion (Fig. 2b). A much larger but less intense (up to 20 losses in a single ecoregion) hot spot of species’ losses is located in the Prairies ecozone and bordering ecoregions of the Boreal Plains ecozone. Smaller hot spots were found in the southwest corner of the Pacific Maritime ecozone (which contains the city of Vancouver, 12 species losses), and ecoregions of the Atlantic Maritime ecozone (generally seven to nine losses per ecoregion). These regions of Canada include the most heavily agricultural parts of the country, but not the most urbanized areas (with the exception of Vancouver).
Equally interesting are the ecoregions containing many imperiled species that have suffered no known or detectable range losses in those ecoregions (i.e. ‘cold spots’ of species’ losses) (Fig. 2c). The most notable cold spot is found in the southern Montane Cordillera ecozone.
Patterns of imperiled bird species richness and species’ losses resemble those for all species combined (Fig. 3a,b): species richness and species’ losses are greatest in the Mixedwood Plains and Prairies ecozones where land is dominated by extensive agriculture and human settlement. The southern Montane Cordillera is also the most noteworthy cold spot for losses of imperiled bird species (Fig. 3c). In contrast, richness of imperiled mammal species is greatest in a large area of mountainous and taiga habitat encompassing much of northwestern mainland Canada (Fig. 4a); this region is also a cold spot of losses for imperiled mammals (Fig. 4c). Listed mammal species have experienced their greatest losses in the Prairies ecozone and bordering ecoregions of the Boreal Plains ecozone (Fig. 4b). As with imperiled birds, hot spots of imperiled amphibians and reptiles are found in the southernmost part of Canada. Low sample sizes obscure patterns of declines for amphibians and reptiles, but the southernmost Mixedwood Plains experienced the most extirpations.
Correlates of losses of COSEWIC-listed species
The number of imperiled species lost per ecoregion was strongly correlated with habitat loss and, with rates of use of pesticides, but only weakly correlated with human population density (Table 2). The number of species lost is also strongly correlated with the total regional species richness. These correlations reflect, in part, the strong climatic gradients in Canada, which have concentrated both agriculture and total species richness in southern Canada (Rivard etal., 2000).
Table 2. Simple Pearson correlations between the numbers per ecoregion of imperiled species lost during approximately the last century, extant imperiled species richness (SR), regional species richness (SR), measures of, agricultural pesticide use, landscape fragmentation and human population. ‘Combined’ refers to the combination of birds, mammals, amphibians and reptiles. Correlation coefficients ≥ 0.133, ≥ 0.174, ≥ 0.190 and ≥ 0.222 are significant at P < 0.05, P < 0.01, P < 0.005 and P < 0.001, respectively.
Imperiled combined SR0.33
Measured as the area of human-dominated land cover.
After controlling for the geographical gradient in total richness, the relationship between species losses and herbicide use (transformed to be bivariate normal), is approximately linear (Fig. 5). There is no evidence of a threshold effect. The relationship between species losses and area in agriculture, after controlling for regional species richness, is similar but noisier (Table 3). Models that related imperiled species losses to area treated with herbicides account for 56–70% of the variance, compared to models relating losses to area in agriculture (46–57%). The statistical effect of herbicides was significant above and beyond the amount of agriculture. Herbicides account for additional variability after accounting for agriculture, whereas the inverse is not true (Table 3).
Table 3. Multiple regressions relating the number of losses of imperiled species (birds, mammals, amphibians and reptiles combined; birds; and mammals) to habitat characteristics in the 86 ecozones across southern Canada that have experienced some habitat loss. Variables were transformed as necessary to improve normality (superscripts indicate power transformations). ΔAIC is the difference in Akaike's Information Criterion between the best model for a given dependent variable and competing models. ΔAIC > 10 indicates a significantly inferior model. In all cases, the best model includes total species richness and the area treated with herbicides. Area in crops and human-dominated land cover were significantly poorer predictors of the numbers of species lost per ecozone. All the individual terms in the models reported here were significant at P < 10−5, except log(area in crops) when log(area treated with herbicides) was already in the model. In those cases, log(area in crops) was not significant (P > 0.05). n = 86.
Measured as the area of human-dominated land cover.
Our results indicate that the most prominent driver of species losses in Canada is more than simply habitat conversion to agriculture and urbanization. Collectively, the hot spots of species losses in Canada contain 12% of Canada's area, 61% of the human population, 84% of farmland, 87% of lost habitat and 90% of herbicide-treated croplands. The southernmost ecoregions of the Mixedwood Plains ecozone, where bird, amphibian and reptile losses are greatest, are densely populated (112 persons km−2) and 71% of the landscape has human-dominated land cover (mostly farmland). With more horticultural crops (e.g. fruits and vegetables) than any other ecozone in Canada, agriculture is very chemically intensive in the Mixedwood Plains (Agriculture & Agri-Food Canada, 1998). The Prairies and southern Boreal Plains constitute a large hot spot of losses that, in comparison to the Mixedwood Plains, is less intense for birds but more so for mammals. This area has low human population density (5.2 persons km−2), but high agricultural intensity. A total of 63% of agricultural pesticide expenditures in Canada are from this area due to vast field crops (e.g. wheat, canola, barley), but the areal rate of use is less intense than in the Mixedwood Plains hot spots.
Although Venter etal. (2006) list urbanization and agriculture as the main human activities contributing to habitat loss, and habitat loss as the main cause of species loss, we found that the relationship between species losses and urbanization is weak. Many ecoregions with large population centres (e.g. Montreal, Halifax) or with the longest history of human alteration of the environment (e.g. the St. Lawrence River lowlands and areas surrounding the Bay of Fundy, settled beginning in the 17th century) have had relatively few species losses. Some losses may have occurred in these areas before systematic species inventories occurred. If this is the case, then species’ losses will have been underestimated in these regions. While accounts of historic bird distributions appear to be relatively thorough, those of mammals and, particularly, both amphibians and reptiles, appear less so.
Cold spots of losses of listed species were characterized by high amounts of natural area and little use of agricultural pesticides. For mammals, these were typically found in mountainous and northern taiga habitats. However, cold spots of losses of birds and amphibians are located in southern regions with greater human population density and habitat loss, but where neither agriculture land cover (0.7–14.5% of is cropland) nor agricultural pesticide coverage is extensive.
Our observation that species losses are significantly more strongly related to pesticide use than to agricultural area indicates that something related to agriculture beyond habitat conversion affects imperiled species persistence. However, our study cannot exclude the hypothesis that species declines result from some other characteristic of agriculture that is correlated with pesticide use, rather than from pesticide use per se. Areas with high pesticide use are also likely to have very large farms (and consequent habitat homogenization), frequent habitat disturbance (e.g. through use of heavy machinery), low plant diversity, etc. It is possible therefore that pesticide use is a surrogate for agricultural intensity in general. Moreover, the marginal effect of herbicides in our study, after controlling for agricultural area, was only moderate. A data set that disentangles this collinearity would be necessary to distinguish between these two competing hypotheses. In practice, such data would be difficult to obtain over large spatial scales.
Our results are consistent with the hypothesis that agriculture, irrespective of pesticide use, is a significant contributor to species losses (Krebs etal., 1999; Kerr & Cihlar, 2004). There is growing evidence that many aspects of the increasing intensification of agriculture, e.g. monocultures, changes in crop type and harvest methods, etc. (Krebs etal., 1999), can all have negative effects on species.
Two main ideas regarding how to lessen the negative impacts of agriculture on wildlife have been put forth by Green etal. (2005). The ‘wildlife-friendly’ option involves reducing the amount of pesticides and fertilizers applied to crops and incorporating more seminatural land within agricultural areas. The other approach is built around the idea of ‘land sparing’ and involves increasing the yield of current agricultural land in order to spare new land from being converted for agriculture. Current literature suggests that conservationists look at the land sparing approach as a viable option (Balmford etal., 2005; Mattison & Norris, 2005). The increase in yield that is recommended by this approach would need to be met through the increased use of pesticides, fertilizers and possibly genetically modified crops (Cassman etal., 2003; Balmford etal., 2005; Green etal., 2005). Our results suggest that further research needs to be done into the effects of pesticides on wildlife populations before choosing the land sparing approach as a solution.
The range of variability of factors that potentially influence the persistence of imperiled species may determine the extent to which our results can be generalized beyond the situation in southern Canada. In West Africa, for example, mammal declines have been linked to bushmeat hunting (Brashares etal., 2004). In Canada, the human population is relatively small, agricultural intensity varies greatly, there is relatively little subsistence hunting, and there are only limited areas of severe industrial pollution. In parts of the world where these other factors are more variable, or agricultural use of pesticides is less variable, spatial variation in the persistence of imperiled species may correlate more strongly with other variables. A further test of the hypotheses that declines are specifically related to pesticides would be to examine this relationship in such regions.
This study indicates that hot spots of imperiled species richness can comprise two classes of areas: areas where many imperiled species have suffered range losses (hot spots of species’ losses, e.g. birds) and areas where many imperiled species persist with their historic ranges intact (cold spots of species’ losses, e.g. mammals). If one examined only the current distribution of imperiled mammal species richness in Canada (Fig. 4a), it may appear that, because relatively few imperiled mammals exist in the Prairies ecozone, the survival of mammals is generally not at risk in this region. And, it may appear that mammals are at great risk in the mountainous and taiga regions of western Canada because many imperiled mammals are found in these areas. In contrast, it is clear that imperiled mammals have undergone extensive losses throughout the prairies, but have suffered little loss throughout the western mountain ranges and taiga (Fig. 4b,c). Identifying and examining hot and cold spots of losses of species, compared with studying only hot spots of imperiled species richness, provides a clearer picture of where species are most threatened and what factors pose a risk to the survival of imperiled species.
In conclusion, our results are consistent with the hypothesis that conversion of natural habitat to human-dominated land cover is a contributor to species losses. However, losses have occurred not simply in areas where native vegetation has been converted to agriculture or human settlement; rather, losses are concentrated in areas where agriculture is chemically intensive (i.e. widespread areal coverage of pesticides). Either pesticides per se, or something correlated with their use (other characteristics of intensive agriculture) apparently contributes to species losses. Conservation strategies to protect endangered species that focus mainly on habitat losses, rather than on patterns of surrounding habitat use, may be inadequate to prevent species losses. In particular, more research is needed on the role that pesticide use plays in species losses and on the possibility of reducing pesticide use as a means to help conserve species.
This project was funded by a Natural Sciences and Engineering Research Council of Canada research grant to DJC. We would like to thank Don Rivard (Parks Canada) for his suggestions and for access to ecoregion, land cover and COSEWIC data. Lenore Fahrig, John Arnason, Francois Chapleau, David Green, Frances Pick and Anthony Francis provided helpful suggestions and comments.