Finding constraints on the abundance of species: A path forward in large-scale distribution modeling

Background/Question/Methods

Depending on situation, purpose and evaluation technique, large-scale distribution models only account for a small percentage of the variation in observed species occurrences or abundances. Various issues haunt the field of distribution modeling such as scale mismatch, dynamics in environment and distributions, and lack of ecological mechanisms underlying the models. While sometimes these problems have been ascribed to a failure of distribution modelers, we believe that most of the time inadequate data, not ignorance, prevents distribution modelers from building better models. Since collecting adequate data over large-extents is nearly impossible we suggest that other ways forward should be explored. We want to encourage distribution modelers to use the biotic and abiotic constraints on presence and abundance to more narrowly predict the distribution of an organism compared to using simple climatic envelopes.
Results/Conclusions

Beginning with the coarsest, presence/absence view of a species range, we suggest several ways in which to come to a more differentiated view. First, environmental envelopes may identify unsuitable areas within a range. Second, energetic constraints and other measures of fundamental niche could refine this envelope. Third, using an output of probability of occurrence instead of presence/absence may give a correlation to relative abundance. Fourth, using abundance as input rather than presence/absence allows for a much more differentiated view of range structure that may be transferable across space and onto new conditions. Fifth, spatial autocorrelation due to processes such as dispersal could be used to improve predictions into new areas. And finally, general community constraints, for example in the form of species abundance distributions, could give clues to constraints on individual species due to biotic interactions. Again these constraints may be transferable in space and time, at least in a coarse and general form. Taken together we will provide visions for the path forward in distribution modeling that take realistic data constraints and transferability in space and new conditions into account.