Testing regression equations to derive long-term global soil moisture datasets from passive microwave observations

Abstract

Within the framework of the efforts of the European Space Agency (ESA) to develop the most consistent and complete record of surface soil moisture (SSM), this study investigated a statistical approach to retrieve a global and long-term SSM dataset from space-borne observations. More specifically, this study investigated the ability of physically based statistical regressions to retrieve SSM from two passive microwave remote sensing observations: the Advanced Microwave Scanning Radiometer (AMSR-E; 2003–Sept. 2011) and the Soil Moisture and Ocean Salinity (SMOS) satellite. Regression coefficients were calibrated using AMSR-E horizontal and vertical brightness temperature (TB) observations and SMOS level 3 SSM (SMOSL3; as a training dataset). This calibration process was carried out over the June 2010–Sept. 2011 period, over which both SMOS and AMSR-E observations coincide. Based on these calibrated coefficients, a global SSM product (referred here to as AMSR-reg) was computed from the AMSR-E TB observations during the 2003–2011 period. The regression quality was assessed by evaluating the AMSR-reg SSM product against the SMOSL3 SSM product over the period of calibration, in terms of correlation (R) and Root Mean Square Error (RMSE). A good agreement (mean global R = 0.60 and mean global RMSE = 0.057 m3/m3), was obtained between the AMSR-reg and SMOSL3 SSM products particularly over Australia, central USA, central Asia, and the Sahel. In a second step, the AMSR-reg SSM retrievals and commonly used AMSR-E SSM retrievals derived from the Land Parameter Retrieval Model (AMSR-LPRM), were evaluated against two kinds of SSM references (i) the global MERRA-Land SSM simulations and (ii) in situ measurements over 2003–2009. The results demonstrated that both AMSR-reg and AMSR-LPRM (better when considering global simulations) successfully captured the temporal dynamics of the references used having comparable correlation values. AMSR-reg was more consistent with MERRA-land than AMSR-LPRM in terms of unbiased RMSE (ubRMSE) with a global average of ubRMSE of 0.055 m3/m3 for AMSR-reg and 0.084 m3/m3 for AMSR-LPRM. In conclusion, the statistical regression, which is tested here for the first time using long-term spaceborne TB datasets, appears to be a promising approach for retrieving SSM from passive microwave remote sensing TB observations.