Enhancing Soil Moisture Estimation through Integration of Fengyun MWRI and MERSI Data with a Piecewise Artificial Neural Network Model

Land surface soil moisture (SSM) is a critical parameter in agriculture, hydrology, and climate studies. Microwave remote sensing technology, which is extensively used for acquiring SSM information at the regional scale, faces accuracy limitations due to influencing factors such as vegetation coverage. This study first comprehensively assessed the accuracy of SSM products derived from the Microwave Radiation Imager (MWRI) carried by Chinese Fengyun-3 (FY-3) satellites. The results revealed a clear decline in correlation and an increase in unbiased root-mean-square error (ubRMSE) with increasing vegetation density, particularly when the normalized difference vegetation index (NDVI) exceeded 0.5. To address this, a piecewise artificial neural network (pANN) model was developed by integrating FY-3 MWRI multi-frequency brightness temperature (BT), FY-3 Medium Resolution Spectral Imager (MERSI) NDVI, and auxiliary data (land cover, soil texture, and date information). This approach significantly improved SSM estimation accuracy across diverse vegetation conditions by effectively capturing the complex interactions between vegetation, soil, and the atmosphere. The pANN model enhanced the correlation coefficient (R) with ground measurements from 0.240 to 0.517 in sparsely vegetated areas (NDVI 0.0–0.3) and from −0.050 to 0.346 in densely vegetated areas (NDVI > 0.5). Furthermore, the pANN-based product demonstrated competitive performance against the international soil moisture active passive (SMAP) level-3 (L3) soil moisture product and substantially increased spatial coverage compared to the original FY-3 data. The refined SSM dataset proves to be a reliable resource for applications in drought and flood monitoring, as well as weather and climate modeling.