A regional scale performance evaluation of SMOS and ESA-CCI soil moisture products over India with simulated soil moisture from MERRA-Land

Abstract

Abstract Three multi-decadal satellite soil moisture (SM) products, obtained by merging two active and six passive, all-active-merged (CCI-ACT), all-passive-merged (CCI-PAS) and all-active-passive-merged (CCI-COMBINED), and Level-3 SM retrieved from Soil Moisture Ocean Salinity (SMOS) mission were evaluated over India. The evaluation strategy employed is twofold: (a) time series and correlation analysis of SM datasets with respect to the Modern Era Retrospective-analysis for Research and Applications-Land (MERRA-L) SM simulation and the India Meteorological Department (IMD) gridded rainfall; (b) investigate the spatial distribution of random error of the satellite products using Triple Collocation (TC) approach. The Pearson's correlation analysis showed that the performance of CCI-ACT and CCI-COMBINED are comparable to each other and they agree well with the MERRA-L simulated SM time series. They also had a good rank correlation with rainfall. The random error from TC is represented in terms of fractional Root Mean Square Error ( fRMSE TC ). It also represents the sensitivity of satellite retrievals to changes in true state. The analysis of fRMSE TC showed that descending swath of SMOS SM has a lower error than ascending for 71% of the pixels over India. CCI-ACT was found to have the most number of pixels with the lowest errors, having a mean fRMSE TC of 0.7188, compared to 0.7705 for CCI-COMBINED, 0.7828 for CCI-PAS and 0.8308 for SMOS-D. However, the error in CCI-ACT was highest in arid desert regions of western India. The error in CCI-COMBINED, CCI-PAS and SMOS-D grew with an increase in vegetation density. The fRMSE TC maps were analysed against the maps of the probability of occurrence of Radio Frequency Interference (RFI), Normalized Difference Vegetation Index (NDVI), soil texture (percentage of clay, sand, and silt) and modified Koppen-Geiger climate classification. The climate classification map was used to classify fRMSE TC against the different homogeneous climate classes. The analysis of the maps revealed that the inconsistency in SMOS is because of the RFI events over India. However, a multiple linear regression based attribution study showed that SMOS-D is the least affected by vegetation (4%) and the spatial distribution of CCI-ACT and CCI-COMBINED error showed more affinity towards soil texture than vegetation density.