Using Matlab´s Supervised Machine-Learning Tools to Retrieve Surface Soil Moisture from Sentinel-1 SAR Data Over the Valencia Anchor Station (Spain)

The urgency on detailing surface soil moisture content worldwide, especially in agricultural soils, is well established. The efforts of the European Space Agency (ESA), regarding the Sentinel-1 mission, facilitated a synthetic aperture radar (SAR) sensor that, in conjunction with machine-learning-based methods, can be useful  and fruitful responding to this technological demand. This paper aims at  exploring the possibility of the Valencia Anchor Station, near the city of Valencia, Eastern Spain, to provide 1 km x 1 km soil moisture products using its ground-based reference meassurements. The results suggest that, among several options, an artificial neural network using the Levenberg-Maquardt learning algorithm, based on soil moisture recovery from Sentinel-1 SAR radar data should be preferred for this site. Among other options, the so called fine-tree regression also presented relevant results. All of this allows us to gain insights into the complexity of the relation SAR´s backscatter – surface soil moisture relation for this site, also aiming at the potential extension of this knowledge to other sites where Sentinel-1 data is available, for example, in framework of the Joint Research Center "Ground-Based Observations for Validation (GBOV) of Copernicus Global Land Products" Project.