A Two-Step Fusion Framework for Quality Improvement of Remotely Sensed Soil Moisture Product: A Case Study for ECV product over the Tibetan Plateau

Quality of remotely sensed soil moisture (SM) product, in terms of coverage and accuracy, is very important in hydrological and environmental applications. Essential climate variable (ECV) SM product is the first try to resolve this problem using multi-satellite observations, but the key issue is still not addressed. Given so, this study proposed a two-step fusion framework to enhance the temporal coverage and accuracy of the remotely sensed SM product. The first step aims to improve the temporal coverage by filling the gaps of remotely sensed SM with a machine learning method. The second step aims to improve the accuracy by rescaling the reconstructed product against another based on the cumulative distribution function (CDF) matching, and then both the reconstructed and rescaled SM products are fused by simple average. We chose the ECV and Fengyun SM products as the algorithm test to improve the quality of ECV over the Tibetan Plateau. Validation in eight fields shows that the R²-value between the fused SM and in-situ measurements ranges between 0.494 to 0.706., which is much higher than the R²-value between the reconstructed ECV SM and in-situ measurements that ranges between 0.368 to 0.647. Moreover, the temporal coverage is improved from 30.84% to 78.12% for the entire Tibetan Plateau (TP). The fused SM provides a better trade-off between temporal coverage and accuracy than the reconstructed ECV SM. This fused SM product has high temporal coverage, high accuracy and high consistency, and is expected to help us better understand the role of soil moisture in the water and energy cycles under global change. It has to be noted that this method could also be used to improve the temporal coverage and accuracy for other remotely sensed SM products.

Keywords: soil moisture; fusion; remote sensing; cumulative distribution function; ECV; Tibetan Plateau