Estimating top-soil moisture at high spatiotemporal resolution in a highly complex landscape

Soil moisture is a critical variable in precision agriculture, hydrological modeling, and environmental monitoring, influencing crop productivity, irrigation planning, hydrological processes and water resource management. Advances in Earth Observation (EO) technologies enable high-resolution soil moisture estimation by integrating synthetic aperture radar (SAR), multispectral imagery, and ground-based measurements. This study describes a comprehensive methodology which is currently under development for near surface soil moisture estimation tailored to the diverse agricultural landscapes of Greece.

The primary objective is to develop and implement a national-scale soil moisture estimation methodology utilizing data from Sentinel-1 and Sentinel-2 satellites, supplemented by an in-situ soil moisture sensors network. The study region encompasses agricultural areas with heterogeneous soil types, land cover, and topographic variations, addressing the complexity of soil moisture dynamics in Mediterranean climates.

Ground truth data for model calibration and validation is provided by a network of IoT-based soil moisture sensors strategically placed to capture diverse soil textures and land cover classes. The network builds on existing stations and introduces additional sensors to enhance spatial coverage and data representativeness for top-soil moisture dynamics. The monitoring network was designed using geospatial analysis techniques considering all the biophysical features influencing soil moisture dynamics.

The methodology includes preprocessing dual-polarization backscatter data (VV and VH) from Sentinel-1 SAR imagery. Vegetation effects on the backscatter signal are corrected using the Water Cloud Model (WCM), parameterized with NDVI from Sentinel-2 and empirical coefficients derived from field measurements. Corrected soil backscatter is combined with ancillary data and fed into machine learning models, including Random Forest and Artificial Neural Networks, trained on in-situ soil moisture observations. Model performance is evaluated using metrics such as RMSE and R² to ensure predictive accuracy. The resulting high-resolution soil moisture maps reflect dynamic spatial and temporal variations with enhanced precision.

Preliminary results highlight the feasibility of integrating satellite and in-situ data for national-scale soil moisture mapping. WCM-based corrections significantly enhance SAR-derived backscatter accuracy, while machine learning models demonstrate strong predictive performance. The scalable methodology offers valuable insights for optimizing agricultural practices and water resource management.