TESTING THE ACCURACY OF ECOLOGICAL NICHE MODELS USING THE LAST GLACIAL MAXIMUM FOSSIL RECORD OF MAMMALS

The geographic distributions of many species are expected to change dramatically over the next century as they accommodate global climate change. Conservation biologists model these expected changes using a suite of techniques called Ecological Niche Models (ENMs) or Species Distribution Models (SDMs). ENMs sample physical environmental parameters (temperature, precipitation, seasonality) and vegetation parameters from known modern distributions of species, creating a multidimensional niche space that describes the current realized niche of the species. This realized niche space can then be mapped on future climate surfaces, producing hypotheses of future species distributions. These hypotheses have been used to plan conservation efforts, and some researchers have suggested programs of assisted migration to move individual organisms into their expected new optimal environments. We test the accuracy of the most commonly used ENMs by projecting a large number of them onto paleoclimate surfaces generated from Last Glacial Maximum (LGM, ~18-21 Ka) boundary conditions and comparing the hindcast distributions to the known LGM fossil record of the species. We have created ENMs for 50 species of North American Mammals with LGM fossil records known from at least 4 US states. We chose these species to balance taxonomic, ecological, and geographic diversity, but choices were purposefully biased toward the western USA, to favor stronger LGM fossil coverage. We drew all fossil data from the geographically explicit FAUNMAP II database of Pliocene to Holocene North American fossil mammal published records, limiting ourselves to sites with dates from 30-10 Ka that clearly cross the LGM, ~20Ka. We compared hindcast distributions to the distribution of all fossil sites from our LGM window, tallying the number of sites within the hypothesized species distribution that contain and do not contain the species and making a similar tally of sites outside the distribution. In this way, we were able to quantify the mismatch between hindcast and fossil distributions.