spatialMaxent: Adapting species distribution modeling to spatial data.

Area of Science:

• Ecological modeling
• Biodiversity informatics
• Computational biology

Background:

• Conventional species distribution models often lack predictive accuracy due to unaddressed spatial data structures.
• Overfitting during model training compromises predictive performance on independent, spatially separated data.
• Reliable species distribution models are crucial for effective biodiversity conservation strategies.

Purpose of the Study:

• To introduce spatialMaxent, a novel software integrating advanced spatial modeling with Maxent.
• To enhance species distribution modeling by addressing spatial dependency and overfitting.
• To provide a user-friendly tool for ecological research and conservation practice.

Main Methods:

• spatialMaxent incorporates spatial cross-validation for variable selection, feature selection, and regularization multiplier tuning.
• The software explicitly considers the impact of spatial dependency within training data to mitigate overfitting.
• Performance was evaluated using a large, diverse dataset (NCEAS) comprising over 200 species globally.

Main Results:

• spatialMaxent demonstrated superior performance compared to conventional Maxent and non-spatially tuned models in 80% of evaluated cases.
• The implemented spatial tuning strategies significantly improved model reliability and predictive power.
• The software proved effective across diverse species and geographic regions.

Conclusions:

• spatialMaxent offers a significant advancement in species distribution modeling by effectively handling spatial data structures.
• Its user-friendly nature makes it accessible to a broad range of users, including researchers and conservation practitioners.
• The tool has strong potential to aid in addressing critical biodiversity conservation challenges through improved ecological predictions.