Résumé
Earthworms, as crucial ecosystem engineers, contribute significantly to various soil functions and ecosystem services such as water regulation, nutrient dynamics, and biomass production. Recognized as sensitive indicators of soil health, their conservation is pivotal for maintaining biodiversity and ecosystem stability. However, the factors influencing their biogeography and community composition are still not well understood. Our study aimed to assess and compare the effectiveness of five modeling algorithms in predicting three key earthworm community parameters: total abundance, total biomass, and taxonomic richness in France. We utilized the LandWorm database (FRB 2023-2025), which primarily contains data on earthworm communities across mainland France. To enhance our analysis, we integrated environmental variables from additional databases related to climate (temperature, rainfall…) and soil properties (texture, carbon, pH…). The algorithms we evaluated included Generalized Linear Models (GLM), Generalized Additive Models (GAM), Random Forest (RF), Generalized Boosted Models (GBM), and Artificial Neural Networks (ANN). This comparative approach allows us to determine the most effective model for predicting each parameter. Our results highlight that the Random Forest model performed best for predicting earthworm abundance, achieving the highest R² of 0.48 with a RMSE of 18 individuals/m². Similarly, for taxonomic richness, the Random Forest model yielded the best R² of 0.55 with a RMSE of 1.8 taxa. This research not only advances our understanding of earthworm community distribution but also supports the design of targeted conservation strategies, ensuring the protection and sustainability of vital soil functions. By highlighting the effectiveness of diverse modeling algorithms, this study contributes to the broader field of soil biodiversity assessments and the strategic planning of land use and environmental conservation initiatives.