Soil moisture mapping at high resolution by merging SMAP, Sentinel1 and COSMO SkyMed with the support of machine learning

In recent years, the possibility of estimating Soil Moisture (SM) at different resolution scales improved greatly with the launch of the latest satellite microwave sensors and in particular of the Soil Moisture Active and Passive (SMAP) radar + radiometer and Sentinel-1 (S-1) Synthetic Aperture Radar (SAR). However, the tradeoff between temporal and spatial resolution offered by each of these two sensors is still unable to meet the requirements of many users. The SM SMAP products are available through the NSIDC data portal at resolution varying between 1 and 36 Km [1, 2]. This study exploited the possibility of merging SMAP, S-1 and COSMO-SkyMed X-band SAR (CSK) data through Artificial Neural Networks (ANN) for obtaining SM products at high spatial resolution, with the aim of evaluating the benefits of assimilating higher resolution SM into hydrological models. The algorithm has been implemented and validated on a test area in Tuscany (Val d’Elsa, center coordinates of Ponte a Elsa: 43°41′20.37″N 10°53′42.38″E), in central Italy. The area is characterized by a partially hilly landscape, including agricultural and urban areas areas and forests, with heterogeneities that set important constraints to the potential of SMAP observations for SM monitoring. The SMAP, S-1 and CSK acquisitions available between 2019 and 2020 on the area have been considered for the algorithm development. The reference SM values for validation purposes have been derived from in-situ observations carried out in the framework of the ASI ‘Algoritmi’ project [3], which also provided the CSK images considered in this study. The improvement of spatial resolution of the output SMC product is still under investigation; however, the preliminary results seem showing that the method is able to map SM from the SMAP, S-1 and CSK synergy at a resolution better than 100m, with correlation coefficient R≃0.89 and RMSE≃0.025 m3/m3.