Neural Network Integration of SMAP and Sentinel-1 for Estimating Soil Moisture at High Spatial Resolution

The possibility of improving the spatial resolution of Soil Moisture (SM) mapping from microwave satellite radiometers is extremely interesting for hydrological studies in small catchments as well as applications to precision farming. In this study, an algorithm based on Artificial Neural Networks (ANN) is proposed, with the aim of improving significantly the spatial resolution of the Soil Moisture Active Passive (SMAP) Enhanced 9 km Soil Moisture (SMC) product, by integrating SMAP and Sentinel 1 (S1) data. The ANN is trained with data at 9 km resolution, obtained by combining the Sentinel-1 data downsampled to the SMAP resolution and the corresponding SMAP SMC product. After training the ANN is applied pixel by pixel to the Sentinel-1 images at full resolution for generating the enhanced SMC maps. The method has been tested in an agricultural area located in Central Italy, for which in-situ SMC measurements were available: the Active/Passive synergy resulted in an appreciable improvement of both retrieval accuracy and spatial resolution.