Remote sensing of soil moisture at high spatiotemporal resolution using thermal inertia

Abstract

Reliable knowledge of soil moisture at daily, field-scale (∼30 m) resolution is essential for flood forecasting, agricultural water management, and other applications that depend on accurate characterization of near-surface hydrologic conditions.  This work presents a thermal-inertia-based framework that integrates multi-sensor remote sensing, in situ observations, and physically based modeling to estimate soil moisture at high spatial and temporal resolution while maintaining physical interpretability.  Hourly land surface temperatures (LST) are obtained at 30 m resolution by downscaling geostationary GOES observations using Landsat-derived spatial covariates, enabling construction of spatially-detailed diurnal surface temperature time series.  Hourly surface energy-balance components are retrieved from ERA5, providing net radiative forcing information, from which ground heat flux is calculated as a residual.  ERA5 fluxes represent spatially aggregated thermal forcing across widely heterogeneous landscapes.   Local (30 m scale) subsurface heat dynamics vary greatly due to differences in soil properties, vegetation, and moisture state.  In situ observations from the International Soil Moisture Network (ISMN) are used to model how local subsurface heat dynamics depart from those implied by coarse-scale energy forcing. This step employs a support vector regression (SVR) modeling strategy that is well suited for multiple, non-linear, and non-independent predictors. The SVR model derives a physically interpretable quantity representing heat dynamics within the soil profile as a function of surface thermal response (LST magnitude, diurnal amplitude, and phase lag) and geographic context.  Soil moisture is subsequently estimated using an SVR approach. In this step, soil moisture is inferred from observed LST amplitude and LST lag relative to radiative forcing, together with the modeled representation of subsurface heat dynamics. To address periods when acceptable thermal remote sensing is unavailable, the framework is coupled with a physically based water-balance model. This model propagates soil moisture states through intervals with limited thermal data.  The resulting hybrid framework enables daily, field-scale soil moisture estimation that is physically grounded and well suited for hydrologic forecasting, agricultural decision support, and environmental monitoring.