Species distribution models are receiving increasing attention for their ability to forecast the spatial distribution of environmentally suitable habitat under scenarios of climate change; however, researchers are confronted with numerous methodological decisions for which there is no widely accepted approach. The majority of studies that employ species distribution models rely upon observational data which record species occurrence, yet provide little information concerning  where the species may be absent. In order to use statistical algorithms requiring both presence and absence data, the generation of pseudo-absences has been widely adopted.  Several strategies for selecting pseudo-absences have been proposed, however, researchers have yet to reach a methodological consensus. The objectives of this study are to compare various methods of selecting pseudo-absences and to provide insight into the formulation of a pseudo-absence strategy for species distribution models. A case study will be presented for wetland plant species in northeastern North America, which have been largely understudied in climate change research. Models were constructed with climatic and edaphic data using the BIOMOD platform (Thuiller et al. 2009). Pseudo-absence strategies that were tested include random pseudo-absence generation, as well as targeted selection based on the geographic or environmental distance from observed presences. The predictive accuracy of the models was measured by calculating the area under the receiver operating characteristic curve.

Results/Conclusions

The results of these tests indicate that the model predictions vary depending on the pseudo-absence strategy that is used. Additionally, we recorded differences in model outcomes and performance according to different ratios of pseudo-absences to observed presences. While the environmental envelop technique was selected as the most appropriate method in our case, the choice of pseudo-absence strategy is dependent on the objectives of the researcher, the spatial bias in the occurrence data, and the scale at which environmental variables are measured. From the results of this study, recommendations have been produced to provide perspective and guidance for the development of pseudo-absence strategies in future studies. This study is part of a larger multi-taxa project examining the effects of climate change on biodiversity in Quebec, Canada.

References

Thuiller, W., B. Lafourcade, R. Engler, and M. B. Araújo. 2009. BIOMOD - a platform for ensemble forecasting of species distributions. Ecography 32:369-373.