Do Random Forest-Driven Climate Envelope Models Require Variable Selection? A Case Study on Crustulina guttata...

Area of Science:

• Ecology
• Biogeography
• Computational Biology

Background:

• Climate Envelope Models (CEMs) use bioclimatic variables for species distribution, but variable selection is challenging.
• Ecological relevance is often assumed, yet species' biological responses are frequently unknown.
• Random Forest (RF) is a robust method for CEMs, handling complex variable relationships.

Purpose of the Study:

• To test the full model hypothesis using all 19 bioclimatic variables in an RF model.
• To compare the predictive performance of full models against reduced and randomly selected variable sets.
• To assess the impact of variable selection on species distribution modeling accuracy.

Main Methods:

• Employed Random Forest (RF) for species distribution modeling.
• Compared four model variants: 2, 7, 10, and 19 bioclimatic variables.
• Validated models against 1000 randomly assembled models with equivalent variable counts.
• Used *Crustulina guttata* as a case study for distribution prediction.

Main Results:

• All tested models demonstrated high predictive performance.
• The full model (19 variables) consistently outperformed models with fewer variables.
• Random variable selections showed comparable performance to ecologically or statistically selected sets of the same size.
• Omitting variables risked losing crucial predictive information.

Conclusions:

• The full model hypothesis, utilizing all available bioclimatic variables, enhances predictive accuracy in RF-based CEMs.
• Variable selection may not offer significant advantages over random selection when using RF.
• In data-limited scenarios, employing all variables preserves potentially important predictors.
• Further research is needed to confirm these findings across diverse taxa and environments.