Print PageSpecies distribution models are increasingly being used in ecology. Such uses include predicting suitable habitat for species, identifying areas of conservation importance, predicting the advances of invasive species and response of species to global climate change. However, there is limited research in validating species distribution models with field surveys to identify if the predictions accurately represent observed species distributions and subsequent species richness. Urban green space is an area of concern because they often are habitat islands in a mosaic of development and could harbor biodiversity. These areas often have similar climate to adjacent green spaces in less urbanized areas (on the scale most species distribution models use), but vary in habitat and land use. These differences could lead to heterogeneous projections of suitable habitat and species richness between urban and natural green spaces. This study uses species distribution models and field surveys (published and conducted) to test the validity of these models in urban and non-urban green spaces within and adjacent to the greater Cincinnati, OH metropolitan area. We used accumulated amphibian species richness maps derived from species distribution models to identify if predicted amphibian species richness represented known species richness. We also identified variables associated with predicted and observed amphibian richness in an attempt to understand if habitat or environmental variables explain the variation in predicted species richness.
Results/Conclusions
Within the urban matrix, predicted amphibian species richness was highest in urban green spaces. However, the predicted species distribution models did not accurately represent observed species richness patterns within urban or non-urban green spaces and tended to over-predict species richness. For example, mean species richness derived from our most liberal threshold (minimum training presence) over-predicted species richness by 127% compared to observed richness; whereas our most conservative threshold (maximum training sensitivity plus specificity) over-predicted by 31% on average. These differences were statistically significant (liberal, t = -13.605, P ≤ 0.001, conservative; t = -2.641, P = 0.022). Our results suggest species distribution models built with the landscape variables we selected did not represent amphibian richness accurately, or that amphibian surveys did not detect all species that were present. Furthermore, the differences between predicted species richness among thresholds represents a significant challenge to land managers and conservation biologist in need of a tool for modeling biodiversity. Lastly, the results of this study highlight potential issues when using species distribution models in heterogeneous environments, such as an urban landscape.
