A spatial downscaling method for SMAP soil moisture through visible and shortwave-infrared remote sensing data

Abstract

Soil moisture (SM) plays an indispensable role in many practical applications, such as drought monitoring, hydrologic applications and agricultural management. Passive microwave remote sensing has proven capable of capturing changes in SM. However, the coarse spatial resolution (approximately 25–40 km) may greatly limit many regional hydrological and agricultural applications. In this study, we present a SM downscaling method based on the visible and shortwave-infrared (SWIR) remote sensing data toward improved spatial resolution of Soil Moisture Active Passive (SMAP) SM. In the proposed method, the Land Surface Temperature (LST), Normalized Difference Vegetation Index (NDVI), horizontally polarized Brightness Temperature (TBh), vertically polarized Brightness Temperature (TBv) and SWIR reflectance collected in Inner Mongolia from April to September 2018 and April to September 2019 are used as input data and combined with topographic information to downscale the SMAP SM L3 product from the original 36 km to 1 km. Random forest (RF) was used to link the input data and SMAP SM. Finally, 30 in situ station SM measurements distributed in Inner Mongolia and precipitation data were used to verify the downscaled SM. The results show that the correlation range of the downscaled SM and the in situ SM is 0.246 to 0.739, with an average of 0.535. The root mean square error (RMSE) range is 0.02 to 0.152 m3/m3, and the mean RMSE is 0.059 m3/m3. The ubRMSE has a range of 0.02 to 0.06 m3/m3, and the mean performance is 0.04 m3/m3. The bias ranges from −0.145 to 0.087 m3/m3, and the mean bias is 0.008 m3/m3. The downscaled SMAP SM can capture the spatial heterogeneity of the SMAP SM and the dynamic changes in measured soil moisture. The downscaled SMAP SM also has a high spatial correspondence with the original SMAP SM, providing more detailed soil water information than the original 36 km resolution.