Niche models relating current species distribution to environmental variables are commonly used to predict future shifts in species’ ranges in response to climate change. When constructing these models, the choice of scale is generally large (>1 km²), making it difficult to include soil and topographical variables as predictors, which generally are too spatially variable to make up-scaling meaningful at these large grid sizes. The question we ask in this experiment is what is the importance of these variables when scale-appropriate, high quality data are available? Our data come from a large ecological inventory of southern Quebec, Canada. At 4874 forested locations within this area, 400 m² quadrats were sampled and the presence/absence of  plant species were recorded, as well as numerous site and soil characteristics. Climate variables were taken from a smoothed surface of current climate. A pre-selection of climate variables was made to reduce the multicollinearity of the data using the VARCLUS technique in SAS. The variables of growing season growing degree days (base of 5° C), total annual precipitation, and annual temperature range where retained.

Results/Conclusions We fit models to the 98 species of plants that met our requirements of having at least 100 occurrences. Generalized linear models with logit link function were fit using a stepwise procedure in R, and the deviance squared statistic (D²) was calculated. The average D² of the final model was 26 (standard deviation of 11). In general, the edaphic variables were more important than the climatic variables in explaining species presence or absence (average of 69% and 31% of total D², respectively with a standard deviation of 18% for both). This work highlights the importance of considering edaphic conditions when projecting future range shifts under changed climates.