Too many candidates: Embedded covariate selection procedure for species distribution modelling with the covsel R package

Selecting the best subset of covariates out a panel many candidates is key and highly influential stage species distribution modelling process. Yet, there currently no commonly accepted widely adopted standard approach by which to perform this selection. We introduce two-step “embedded” covariate selection procedure aimed at optimizing predictive ability parsimony models fitted in context high-dimensional candidate space. The combines collinearity-filtering algorithm (Step A) with three model-specific embedded regularization techniques B), including generalized linear model elastic net regularization, additive null-space penalization, guided regularized random forest. evaluated through an example application habitat suitability 50 Switzerland from suite 123 covariates. demonstrated provide significantly more accurate as compared obtained alternative procedures. Model performance was independent characteristics data, such number occurrence records or their spatial across study area. implemented streamlined our covsel R package, paving way for ready-to-use, automated, tool that missing field modelling. All information required installing running package openly available on GitHub repository https://github.com/N-SDM/covsel.