Soil Moisture Retrieval Using Quad-Polarized SAR Data from Radar Imaging Satellite 1 (RISAT1) Through Artificial Intelligence-Based Soft Computing Techniques

Abstract

The present study aims to explore and compare the potential of different Artificial Intelligence-based Soft Computing (AISC) techniques to prepare surface Soil Moisture Content (SMC) map using fine-resolution (~ 5 m), quad-polarized Synthetic Aperture Radar (SAR) data obtained from Radar Imaging Satellite 1 (RISAT1). Potential of three different AISC techniques, i.e. Support Vector Machine (SVM), Random Forest (RF) and Genetic Programming (GP), is explored. The estimated surface SMC is validated with the field soil moisture values in both bare and vegetated lands (< 30 cm height). Different techniques have their own merits and demerits; however, we recommend GP to be most useful due to its other features. For example, GP provides the mathematical relationship, importance and sensitivity of each individual input to the surface SMC. This helps us to quantify the contribution of quad-polarized backscattering coefficients and soil texture information. It is noticed that the use of only SAR data without soil texture information may be acceptable with reasonable accuracy with an enormous benefit of its applicability to the locations without soil texture information. Using this, an exemplary fine-resolution (~ 5 m) SMC map is developed. Such high-resolution maps for large spatial extent are expected to be highly useful in many applications.