Spatio-Temporal Comparisons Between Microclimate Species Distribution Models and Mechanistic Models of Potential...

With projected decreases in biodiversity looming due to changing environmental conditions, it is important for conservation managers to have accurate predictions of species' distributions. Species distribution models (SDM) and mechanistic models that account for biophysical factors are important tools for predicting potential distributions for many species. Incorporating microclimate data into SDMs and mechanistic models has become an important step for developing biologically relevant models for organisms reliant on microclimatic regimes. However, there remains a need to compare and integrate the predictions from microclimate-derived SDMs and mechanistic models at fine spatio-temporal scales to improve predictive accuracy and quantify model uncertainty. We developed correlative SDMs and mechanistic models of potential resistance to surface activity for two salamanders using fine-resolution (3 m) microclimate data. Models were produced for the Great Smoky Mountains National Park, USA during the 2010, 2030, and 2050 time periods. We determined the spatio-temporal agreement between SDMs and mechanistic models to assess model uncertainty at varying spatial resolutions. We also modeled and assessed spatio-temporal variability within potential activity corridors. We found that agreement between fine-resolution microclimate SDMs and mechanistic models was generally poor and varied temporally, but model agreement increased and converged at varying coarser spatial resolutions. Furthermore, potential activity corridors spatio-temporally varied and demonstrated increased habitat fragmentation under future projections. The findings from this study highlight a contradiction in which we may need to model species distributions with microclimate data at finer, more biologically meaningful resolutions, but model agreement between correlative and mechanistic approaches may be weakened at these fine scales. Researchers may therefore need to strike a balance between increasing spatial resolution and study extent when integrating model approaches. Further quantifying model uncertainty and identifying alternative methods for integrating SDMs and mechanistic models will be an important step towards accurately predicting species distributions under changing environmental conditions.