Abstract
Soil moisture (SM) with high precision and spatiotemporal resolution is crucial for crop yield estimation and water resource management, yet the spatial resolution of widely used passive microwave-based SM products remains low (~tens of kilometers), making them inadequate for regional-scale applications. Spatial downscaling technique provides a viable solution to enhance the spatial resolution of passive microwave SM products. Despite extensive efforts made so far, surface heterogeneity which is an essential factor that causes differences in SM across coarse and fine scales by affecting processes such as water infiltration, evaporation, and storage, has often been overlooked in previous algorithms, limiting the effectiveness of SM downscaling in heterogeneous regions. To address this knowledge gap, this study proposed a new SM downscaling method, termed heterogeneity-aware downscaling algorithm (HADA), that integrates surface heterogeneity including heterogeneity in land cover, soil texture, terrain, and vegetation coverage by using the data-driven machine learning (i.e., Random Forest) approach. Moreover, a weighted scheme based on the importance ranking of heterogeneity parameters was developed to perform a more physically reasonable spatial correction of downscaling residuals. The proposed HADA was adopted to downscale the SMAP SM products generated by the newly developed microwave SM index (SMI) from 0.25° to 0.05°. Finally, the downscaling results were assessed using ground SM observations from 1260 sites across various regions worldwide and compared with existing methods and datasets. The results indicate the global distribution of the downscaled SM aligns well with that of the global aridity index, indicating a reasonable spatial behavior. Incorporating surface heterogeneity noticeably improves the fitting ability and estimation accuracy of the downscaling model. The downscaled products maintain accuracy comparable to the original data when validated byin situSM but exhibits enhanced spatial details. Compared with the traditional DisPATCH approach, SMAP active-passive, ERA5-Land, and SiTHv2 SM datasets, HADA is superior in terms of both absolute accuracy and the capability to capture SM dynamics. This study not only provides an effective and feasible method to downscale passive microwave based SM data, but also introduces a potential approach to mitigate uncertainties caused by surface heterogeneity when downscaling other satellite derived products, thereby offering high-quality data support for diverse applications.
Showing the abstract — retrieve the full paper via the Exa API.