An improved inversion algorithm for spatio-temporal retrieval of soil moisture through modified water cloud model using C- band Sentinel-1A SAR data

Abstract An approach to incorporate vegetation fraction into Modified Water Cloud Model (MWCM) and the evaluation of potential of multi-target random forest regression (MTFER) were done for the retrieval of spatio-temporal variability of soil moisture (SM) in Varanasi district of Uttar Pradesh, India. The Sentinel-1A SAR images were acquired for three different dates (19/12/2016, 05/02/2017 and 25/03/2017) for two types of spatial regions covered with vegetated and sparse vegetated soil field for SM retrieval. The vegetation fraction (fveg), computed from Landsat – 8 satellite data, was inserted into the modified water cloud model (MWCM) as a modification factor. Leaf area index (LAI) was used as a vegetation descriptor parameter (V) in the MWCM. Subsequently, a machine learning based MTRFR algorithm with a regularization routine was used for stable and optimum solution for complex problems related to the inversion of the MWCM for the accurate estimation of SM. The coefficient of determination (R2), root mean square error (RMSE) and nash sutcliffe efficiency (NSE) indicated significantly better results in the region-2 for all the temporal changes occurred than those of region-1. The results showed that incorporation of fveg to the MWCM provided high potential to retrieve spatio-temporal SM in the region-2 where soil fields were mostly covered with wheat and barley crops rather than in region-1 having sparse vegetated soil field. The overall accuracy of spatio-temporal retrieval of SM after incorporating vegetation factor to MWCM showed significantly better R2 = 0.82, RMSE = 3.18 (%) and NSE = 0.85. The inversion results proximate that the MTRFR techniques applied to the MWCM, including vegetation factor, has great a capability for an accurate SM retrieval in the vegetated soil field.