How do you see soil moisture at the scale of a single field when your data comes from space? Satellite and land-surface-model products map surface soil moisture across whole regions, but only at kilometre-scale resolution — far too coarse to capture the variability that matters for a small catchment or a precision-agriculture plot. Our latest study, just published in Remote Sensing Applications: Society and Environment, offers a way to bridge that gap.
We developed a two-step Random Forest downscaling framework that carries a satellite soil-moisture signal all the way down to the centimetre scale. A first model sharpens the 1 km satellite product to 30 m resolution, and a second model then draws on features derived from an uncrewed aerial system (UAS) to reach a 16 cm super-resolution map. Tested at the experimental catchment of Monteforte Cilento in southern Italy, both steps performed strongly, with coefficients of determination of 0.805 and 0.837, and the final 16 cm map matched in-field measurements with a Pearson correlation of 0.71 and an unbiased RMSE of 0.0473 cm³ cm⁻³, while keeping soil-moisture patterns spatially coherent across every scale from 1 km to 16 cm.
What we find most telling is how the drivers change with scale: regional temperature metrics dominate at the coarser step, but at sub-metre resolution the decisive predictors become diurnal thermal contrast and micro-topography — the small differences that make one corner of a field wetter than another. The result is a single, consistent workflow that links regional monitoring to field-scale decisions, opening the door to precision irrigation and fine-grained hydrological modelling.
Zhuang, R., Manfreda, S., Zeng, Y., Zhang, L., Szabó, B., Nasta, P., Romano, N., & Su, Z. (2026). Bridging Satellite and UAS Scales for Surface Soil Moisture Mapping: A Two-Step Random Forest Downscaling Framework. Remote Sensing Applications: Society and Environment, 102178.