Improving the fusion of global soil moisture datasets from SMAP, SMOS, ASCAT, and MERRA2 by considering the non-zero error covariance

https://doi.org/10.1016/j.jag.2022.103016Get rights and content
Under a Creative Commons license
open access

Highlights

  • Multi-source surface soil moisture datasets were merged.

  • Data error covariance was considered in calculating merging weights.

  • Merged dataset exhibited dramatically reduced uncertainties.

  • The impacts of cross error on data fusion quality were analyzed.

  • The relative strengths of individual parent datasets were identified.

Abstract

Surface soil moisture (SSM) estimates from different sources have distinct error characteristics. Multi-source data combination is an efficient way to obtain improved SSM data with reduced uncertainties. Previous data merging studies based on the linear weight averaging scheme rarely considered the impacts of data error covariance (EC) and usually needed a reference dataset, which can lead to suboptimal merging weights. This study applied the quadruple collocation (QC) to estimate EC and combine four SSM datasets simultaneously without the need for a reference. Specifically, two passive microwave satellite datasets (the L3 Soil Moisture Active Passive (SMAP)-V7 and the L3 Soil Moisture and Ocean Salinity -INRA-CESBIO (SMOS-IC)-V2), one active microwave dataset from the Advanced Scatterometer (ASCAT), and one model dataset from the Modern-Era Retrospective analysis for Research and Application, Version 2 (MERRA2) were combined. Generally, QC-based data combination reduced SSM data uncertainties with significantly reduced unbiased Root Mean Square Error (ubRMSE) scores against in situ observations and globally decreased fMSE scores. Moreover, in situ evaluation showed that the QC-based fusion products exhibited better skills than the Tripe Collocation (TC)-based products without considering EC. There were statistically significant differences in Pearson correlation coefficients and ubRMSE metric between the QC and TC -based products. Ignoring the EC between SMAPV7 and SMOS-ICV2 caused overestimations in their relative contributions to fusion data and degraded fusion accuracy. Specifically, the QC-based merging weight was reduced averagely by 0.27 (0.28) for SMAP (IC) when their error cross-correlation increased roughly from −0.42 to 0.9. This study can provide guidance for the generation of improved merged datasets at a global scale.

Keywords

Surface soil moisture
Multi-source data merging
Quadruple collocation
Error covariance

Data availability

Data will be made available on request.

Cited by (0)