Mapping uncertainty of habitat suitability models of four north American tree species

Habitat suitability models (HSMs) are powerful tools that encourage informed decision making in conservation planning, research, and policy making. They are typically derived from the correlation of environmental variables with the location of known species occurrences. Most publications within the last fifty years on HSMs or related habitat modeling techniques do not mention error, noise, or uncertainty. When uncertainty is not reported, these models may elicit a misleading sense of confidence or accuracy in model output. Sources of uncertainty are understood, but the methodology for addressing these sources are not well established. I investigated multiple sources of uncertainty within my data and simulated them using the novel modeling package, Hyper-envelope modeling interface (HEMI 2) for four North American tree species whose geographic ranges are known (redwood, red alder, paper birch, and Douglas-fir). I also investigated the impact of incomplete occurrence data and modeling extent on HEMI 2 using the same species. My study demonstrated that simulating multiple sources of uncertainty within HSMs can increase the coverage of predicted habitat suitability within a species' geographic range. However, HSMs in general tended to predict a large proportion of potential habitat beyond the documented geographic range of each species. Additionally, HSMs had variable results when this predicted range was compared to each species' geographic range, regardless of a uniform method for comparison. When manual model adjustment was appropriate, my study also demonstrated that geographic range coverage can increase in accuracy when the species response to its environment is known. With respect to modeling extent, I also found that as it increased beyond the geographic range of each species, the value of habitat suitability tended to decrease. My study provides a comprehensive methodology aimed at quantifying the uncertainty inherit within a few commonly used data sources (FIA, WorldClim, and BioClim), where my results support the overall importance of modeling uncertainty in HSMs.