Fusing satellite-based surface soil moisture products over a typical region with complex land surface characteristics

Abstract

• A high accurate SM product was generated by fusing the ESA CCI and SMAP products. • Accuracy of the fused SM product was evaluated. • The proposed approach provided a new insight for fusing SM products. Accurate surface soil moisture (SSM) information is essential for various investigations and implications in agronomy, hydrology, meteorology and ecology. In this study, an improved SSM product from 2015 to 2018 was generated by combining the triple collocation (TC) method and entropy value (EV) method (TC&EV) to fuse the European Space Agency Climate Change Initiative (version 04.7, ESA CCI v04.7) and Soil Moisture Active Passive (SMAP, version 5) products in the Yangtze River Delta (YRD) region. The ESA CCI v04.7 and SMAP products were corrected against a reference model-based SSM product using the cumulative distribution function matching method. These two corrected SSM products were fused for pixels in which they were significantly correlated ( p < 0.05) with the model-based SSM product. Their fusing weights were determined by three methods including the equal-weight (EW) method, TC method, and TC&EV. In TC&EV, for pixels in which both corrected SSM products were significantly correlated ( p < 0.05) with the five auxiliary factors, the EV method was applied to determine their fusing weights. Otherwise, the TC method was used. The SSM fused by TC&EV was compared with those fused by EW and by TC. The SSM fused by TC&EV outperformed the other two with the highest correlation coefficient (R) of 0.766 and lowest unbiased root mean square error (ubRMSE) of 0.022 m 3 m −3 . It also had higher R, lower ubRMSE, and lower absolute mean bias compared to the ECA CCI and SMAP products. Moreover, for pixels in which the EV method was used, the relative importance of the five auxiliary factors was ranked as land surface temperature > normalized difference vegetation index > precipitation > water area percentage > Gini-Simpson index. This study proposed an approach for accquiring accurate SSM information for the YRD region with complex land surface characteristics, which would be useful for various modeling and implication purposes.