Species distribution models (SDMs) can be important tools for proactive conservation management if they are realistic. Unfortunately, achieving and assessing SDM realism is challenging given the general limitations of scientific models and empirical species data. We addressed the issue of achieving realism with high model quality and reproducibility by reviewing 200 SDMs and cataloguing methods for data availability, response and predictor variables, model fitting, and model performance. We addressed the issue of assessing SDM realism by comparing known and predicted distributions with simulated data for various model fitting choices. Finally, we applied and compared subsequent lessons to empirical, ensemble SDMs for the exotic ball python (Python regius) and invasive Argentine black and white tegu (Salvator merianae) as case studies for Florida mitigation management practices. Fundamental SDM standards were addressed inconsistently in the literature and lacked transparency and replicability. This decreases SDM quality and increases method confusion. We provided a new checklist with well-supported methods to aid in greater method consistency (thus quality and reproducibility) and realism. Model realism varied based on algorithm choice but was consistent across sample sizes and species types. No algorithm was perfectly realistic, but eight consistently produced high rates of realism and performance (and reality and performance were not strongly correlated). Ensemble strategies were consistently more robust than individual algorithms, so we recommended a new ensemble based on those eight high-performing algorithms. We applied this ensemble strategy to our empirical SDMs along with other ensemble groupings (including the most popular individual algorithm) from the literature to inform novel SDMs. Ensemble SDMs consistently performed well with the empirical data and outperformed the individual algorithm. Results here help inform general SDM method guidance for a variety of native and nonnative species (with both simulated and empirical demonstrations) to improve SDM realism and applications in the future.
Hannah Bevan
Dr. David Jenkins
Read More