itsdm: Isolation forest‐based presence‐only species distribution modelling and explanation in r

Abstract

Abstract Multiple statistical algorithms have been used for species distribution modelling (SDM). Due to shortcomings in species occurrence datasets, presence‐only methods (such as MaxEnt) have become increasingly widely used. However, sampling bias remains a challenging issue, particularly for density‐based approaches. The Isolation Forest (iForest) algorithm is a presence‐only method less sensitive to sampling patterns and over‐fitting because it fits the model by describing the unsuitable instead of suitable conditions. Here, we present the itsdm package for species distribution modelling with iForest, which provides a workflow wrapper for the algorithms in iForest family and convenient tools for model diagnostic and post‐modelling analysis. itsdm allows users to fit and evaluate an iForest SDM using presence‐only occurrence data. It also helps the users to understand relationships between species and the living environment using Shapley values, a suggested technique in explainable artificial intelligence (xAI). Additionally, itsdm can make spatial response maps that indicate how species respond to environmental variables across space and detect areas potentially affected by a changing environment. We demonstrated the usage of the itsdm package and compared iForest with other mainstream SDMs using virtual species. The results enlightened that iForest is an advantageous presence‐only SDM when the actual distribution range is unclear.

Keywords

• explainable artificial intelligence (xAI)
• isolation forest
• presence only
• shapley values
• species distribution modelling (SDM)