Finnish researchers use Sentinel-1 and ML to achieve 90% soil moisture accuracy

A team led by Alireza Hamedianfar at the Geological Survey of Finland (GTK) and collaborators has developed a high-resolution method for sediment-specific surface soil moisture retrieval using time series from Copernicus Sentinel-1 C-band SAR, auxiliary Sentinel-2 optical data, and IoT-enabled capacitance sensors deployed at a limestone quarry in southeastern Finland. The study, published on arXiv (2606.24364), compared machine learning models including XGBoost, LightGBM, Random Forests, linear regression, and k-nearest neighbors. The best results came from tree-based ensemble methods using a comprehensive feature set that combined Sentinel-1 backscatter, time-series soil moisture indices, optical, topographic, and temperature predictors. The models achieved RMSE as low as 0.037 m³/m³ (3.7 volumetric percent) and R² values reaching 0.90 across multiple sediment types (clay, organic soil, flotation sand, gravel). The approach demonstrates that sediment-specific calibration significantly improves accuracy over baseline SAR-only methods by more than 2 vol% when using single-sensor inputs, but the benefit diminishes when richer multi-source features are available.

Notably, the accuracy varied by sediment texture — lowest errors were observed for clay and organic soils, while coarser materials like flotation sand and gravel showed higher errors. The study highlights the potential of combining operational satellite data (Sentinel-1 and Sentinel-2) with ground-based IoT sensors and machine learning to enable continuous, high-resolution moisture monitoring over mining sites. This has direct implications for environmental management, mine safety, and water resource optimization in extractive industries. The research also demonstrates a practical workflow for integrating multi-sensor SAR and optical data with in-situ measurements, offering a scalable template for precision agriculture, landslide monitoring, and other land-surface applications where soil moisture dynamics are critical.