Surface soil moisture (SM), as a crucial ecological element, is significant to monitor in semiarid mining areas characterized by aridity and little rainfall. The passive microwave remote sensing, which is not affected by weather, provides more accurate SM information, but the resolution is too coarse for mining areas. The existing downscaling method is usually pointed to natural scenarios like agricultural fields rather than mining areas with high-intensity mining. In this paper, combined with geoinformation related to SM, we designed a convolutional neural network (SM-Residual Dense Net, SM-RDNet) to downscale SMAP/Sentinel-1 Level-2 radiometer/radar soil moisture data (SPL2SMAP_S SM) into 10 m spatial resolution. Based on the in-site measured data, the root mean square error (RMSE) was utilized to verify the downscaling accuracy of SM-RDNet. In addition, we analyzed its performance for different data combinations, vegetation cover types and the advantages compared with random forest (RF). Experimental results show that: (1) The downscaling from the 3 km product with the combination of auxiliary data NDVI + DEM + slope performs best (RMSE 0.0366 m3
); (2) Effective data combinations can improve the downscaling accuracy at the range of 0.0477–0.1176 m3
(RMSE); (3) The SM-RDNet shows better spatial completeness, details and accuracy than RF (RMSE improved by 0.0905 m3
). The proposed SM-RDNet can effectively obtain the fine-grained SM in semiarid mining areas. Our method bridges the gap between coarse-resolution microwave SM products and ecological applications of small-scale mining areas, and provides data and technical support for future research to explore how the mining effect SM in semiarid mining areas.
This is an open access article distributed under the Creative Commons Attribution License
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.