Soil moisture estimation over flat lands in the Argentinian Pampas region using Sentinel-1A data and non-parametric methods

Abstract

A procedure for soil moisture (SM) estimation over flat lands in the Argentinian Pampas region, using the water balance equation that considers SM to be the result of water inflows and outflows to the soil system, is presented. In recent years, remotely sensed data with Synthetic Aperture Radar (SAR) and radiometer sensors have been used to develop different methodologies to obtain SM maps. Thus, a variety of methodologies with different levels of complexity are available nowadays. These models require soil information such as soil physical properties and mineral composition, not readily available in Argentina and many other remote areas of the world. The procedure presented in this paper takes into account water input and output processes of the soil system and represents them with different hydro-environmental variables and SAR data. The water balance equation was solved with Multiple Linear Regression (MLR), Multivariate Adaptive Regression Splines (MARS) and Artificial Neural Network (ANN) statistical models, fed with readily available data over Comisión Nacional de Actividades Espaciales (CONAE) core site located in Cordoba province, Argentina. The resulting models were obtained with precipitation (PP), air temperature (T a ) and relative humidity (RH) observations and with SAR data from the Sentinel-1A satellite mission. The accuracy of the model estimates represents 10% of the observed measured values of SM and is in line with state of the art algorithms. Results suggest that any model can be used with similar precision, since they show similar errors, although the MLR method allows analyzing and quantifying the errors introduced by the variables.