POC plots: calibrating species distribution models with presence-only data

Area of Science:

• Ecology
• Biodiversity research
• Computational biology

Background:

• Statistical models are crucial for predicting species distributions and responses to environmental variables.
• Model performance is assessed by discrimination (distinguishing presence/absence) and calibration (predicting presence probability).
• Common metrics like AUC primarily measure discrimination, neglecting calibration.

Purpose of the Study:

• Introduce a novel tool, the presence-only calibration plot (POC plot), for evaluating species distribution model calibration.
• Develop a method to assess model accuracy using readily available presence-only data.
• Provide a tool for visual exploration and recalibration of statistical models.

Main Methods:

• Generalized predicted/expected curves to create a presence-only analogue of traditional calibration curves.
• Utilized presence-only evaluation data, which is more abundant than presence-absence data.
• Applied POC plots to recalibrate models generated by the DOMAIN method.

Main Results:

• The POC plot effectively measures model calibration using presence-only data.
• Recalibration using POC plots significantly improved the predictive performance of DOMAIN models.
• Demonstrated the utility of POC plots across 226 species in six global regions.

Conclusions:

• POC plots offer a valuable method for assessing and improving the calibration of species distribution models.
• The approach enhances model reliability by utilizing widespread presence-only occurrence data.
• This tool has broad applicability for ecological modeling and biodiversity assessments.