Exploring the combination of remote sensing and machine learning to correct Cosmic Ray Neutron Sensing soil moisture signals for variable biomass effects

Abstract

The monitoring and modeling of the hydrological cycle are prerequisites for understanding and adapting to the consequences of climate change for natural and agricultural systems. The method of Cosmic Ray Neutron Sensing promises to close the critical measurement gap between localized point scale observations and wider-area remote sensing. However, the isolation of the soil moisture signal in the aggregated hydrogen signal requires highly interdisciplinary research. For transferring the method from hydrological research into an applied system, the complex, nonlinear influence of dynamic vegetation biomass must be accounted for. Because current manual measurement-based vegetation correction procedures are difficult to apply to the large-scale operationalization of Cosmic Ray Neutron Sensing, this study investigates the synergy of remote sensing and machine learning as a possible alternative. Utilizing local data, from diverse monitoring networks, across different land cover types, in combination with Sentinel-1 radar backscatter and Sentinel-2 optical indices, these data sources were integrated using regression-based random forest models to predict a dynamic calibration parameter, guided by the vegetation information. The results demonstrated that this data-driven approach significantly improved soil moisture estimations across agricultural, grassland, forest, and orchard sites. Specifically, the methodology highlights the operational superiority of continuous radar data over cloud limited optical imagery for consistent vegetation tracking. In contrast to existing assumptions, the spatial footprint experiments of this study challenge the established theoretical, distance based radial weighting, which is frequently overridden by site specific, non-symmetric environmental heterogeneities in the local vegetation. By successfully bridging the disciplines of particle physics, remote sensing, and soil hydrology, this research provides a relevant, scalable framework to account for biomass variations. The findings of this study include valuable insights for the refinement and application of biomass correction methods, advancing the harmonization of Cosmic Ray Neutron Sensing soil moisture measurements.