SDM profiling: A tool for assessing the information-content of sampled and unsampled locations for species distribution models

Species distribution models (SDMs) are key tools in biodiversity and conservation, but assessing their reliability in unsampled locations is difficult, especially where there are sampling biases. We present a spatially-explicit sensitivity analysis for SDMs - SDM profiling - which assesses the leverage that unsampled locations have on the overall model by exploring the interaction between the effect on the variable response curves and the prevalence of the affected environmental conditions. The method adds a 'pseudo-presence' and 'pseudo-absence' to unsampled locations, re-running the SDM for each, and measuring the difference between the probability surfaces of the original and new SDMs. When the standardised difference values are plotted against each other (a 'profile plot'), each point's location can be summarized by four leverage measures, calculated as the distances to each corner. We explore several applications: visualization of model certainty; identification of optimal new sampling locations and redundant existing locations; and flagging potentially erroneous occurrence records.

Automated summary

Species distribution models (SDMs) are important tools for biodiversity and conservation. Assessing their reliability in unsampled locations is challenging, especially with sampling biases. We propose a sensitivity analysis technique called SDM profiling, which examines the impact of unsampled locations on the overall model by analyzing the effect on variable response curves and the prevalence of affected environmental conditions. The method involves adding pseudo-presence and pseudo-absence data to unsampled locations and comparing the probability surfaces of the original and modified SDMs. This approach can be used for visualizing model certainty, identifying optimal sampling locations, pinpointing redundant locations, and identifying potentially erroneous occurrence records.