Environmental determinants and spatial patterns of soil antibiotic resistance gene diversity across Europe using interpretable geographically weighted machine learning

The dissemination of antibiotic resistance genes (ARGs) in soils poses significant threats to ecosystems and public health. Their diversity is associated with a complex interplay of environmental factors and exhibits pronounced spatial heterogeneity. However, the spatial patterns and key associated factors remain insufficiently characterized. In this study, ARGs data from the topsoil of 649 sites across Europe, obtained from the 2018 Land Use/Cover Area frame statistical Survey (LUCAS, 2018), were used to develop an interpretable geographically weighted machine learning model. The model was used to quantify the relative importance of environmental factors associated with soil ARG diversity and to characterize spatial distribution patterns. The results indicate that by incorporating spatial heterogeneity in ARG diversity is associated with improved model performance, with geographically weighted machine learning models outperform conventional machine learning models. In particular, the geographically weighted extreme gradient boosting (GW-XGBoost) model achieved a higher R 2 and lower RMSE than the other models, with the test set R 2 improving by 22.3% compared to XGBoost. Soil pH showed the highest relative importance, accounting for 41.3%, with lower ARG diversity observed under neutral to alkaline conditions (pH > 6.7). Both pH and precipitation showed strong direct associations with ARG diversity, whereas land cover, temperature, and elevation were associated with ARG diversity through indirect pathways related to soil properties. Spatial mapping revealed higher ARG diversity in northern Europe, particularly Sweden and Finland, and lower ARG diversity in southern regions, including Spain and Italy. This study characterizes the environmental factors associated with soil ARG diversity and their spatial patterns, providing a scientific basis for soil management and public health protection, and offering a useful reference for modeling in environmental studies.