Predicting soil biological indicators of soil health and identifying environmental constraints on soil biodiversity across European landscapes using earth observation data and machine learning

Under global change and land-use intensification, changes in soil physical and chemical properties propagate to soil biota, with cascading effects on ecosystem functioning and ecosystem service provision. In turn, soil organisms regulate key soil properties such as aggregation, nutrient cycling, and organic matter stabilization, creating a tightly coupled biophysical feedback loop that underpins soil health and ecosystem resilience. However, across Europe, strong environmental heterogeneity and fragmented datasets have made it difficult to identify biological soil health indicators that are robust across land-use systems and pedoclimatic regions.

We used datasets on soil biotic and abiotic properties generated within the SOB4ES project, land-use information and Earth observation-derived climatic (ERA-5), vegetation (Sentinel-2 NDVI) and topographic (NASA SRTM DEM 30m) variables across European sites. Our aim was to investigate scalable, data-driven approaches for soil health assessment under global change and human pressures. State-of-the-art machine-learning models were used to identify the relative importance of natural environmental drivers, soil state variables and human-induced pressures, shaping soil organism abundance and diversity across spatial scales.

Diversity metrics across multiple  taxa consistently showed stronger relationships with environmental gradients than population densities, highlighting diversity as a more sensitive indicator of environmental change than density. Among the most dominant cross-taxa drivers of species richness was soil pH and organic carbon, with highest biodiversity associated with alkaline, carbon-rich soils under moderate moisture conditions. In contrast, high soil moisture and high relative humidity, reflecting both climatic forcing and land-use effects, reduced abundance and diversity across multiple groups, indicating broad sensitivity of soil biota to excess moisture stress under global change. Microbial biomass and nematode density showed particularly strong and accurately captured responses to soil carbon availability, soil texture and elevation, highlighting their value as integrative indicators of soil resource status and ecosystem functioning. Overall, our results demonstrate that biological indicators respond consistently to large-scale gradients in climate, soil chemistry and land-use, supporting their application in spatially explicit soil health assessments and in evaluating the impacts of environmental change and land management across Europe. By integrating microbial, soil faunal indicators across multiple European countries and contrasting pedoclimatic regions, our analysis shows that soil communities are governed by broadly shared environmental controls under global change and land-use pressures, rather than by idiosyncratic, site-specific effects.

The strong contribution of specific soil properties and  Earth-observation-derived variables, combined with the ability of machine-learning models to integrate heterogeneous datasets, demonstrates a powerful and scalable approach for identifying robust biological soil health indicators across regions and land-use systems.

Acknowledgments: The work and all the authors were supported by the Horizon Europe project SOB4ES (“Integrating Soil Biodiversity to Ecosystem Services”) under Grant Agreement No. 101112831. We acknowledge all participating investigators from the SOB4ES consortium who contributed to the existing sample collection and the field sampling for the generation of the spatial database used in the current analysis. Partners from KNAW, UVIGO, NUID UCD, UNICT, KU Leuven, CU, ARO, IBB, UL, UoC, SLU, EFWSL, Airfield, MFO, and INRAe provided these contributions.