Community-level model performance across large temporal scales and periods of climatic novelty
Multiresponse (i.e. multivariate) species distribution models (SDMs) offer a promising new approach to predict species- and community-level responses to global climate change. These models, here referred to as community-level models (CLMs), incorporate species associations through co-occurrence/co-exclusion matrices with environmental data to estimate a species potential distribution and community assembly. Similar to traditional SDMs, CLMs are subject to transferability issues across space and time. Here we compare a suite of CLMs to their traditional SDM counterparts through the last 21,000 years in eastern North America. We test model predictability against the observed fossil pollen and mammal records and compare model performance to climatic novelty in each time period examined. We then combine these two datasets and test model performance with the incorporation of two trophic levels, an underutilized extension of CLMs (especially in the fossil record).
Overall, model performance drops significantly for both SDMs and CLMs as climatic novelty increases; however, CLMs predict better than SDMs in times with high climatic novelty. In addition, CLMs improve predictions of rare or low abundance taxa. These conclusions demonstrate the caution in using models to predict species and community response to global climate change but suggest that CLMs may better predict ecological responses during novel climates, especially for rare taxa. Fossil evidence suggests trophic cascades shape species distributions and species responses to climate, demonstrating the need to incorporate multiple trophic levels into models.