High spatial resolution soil moisture information is important for hydrological, climatic and agricultural applications. The lack of high resolution soil moisture data over large areas at the required accuracy is a major impediment for such applications. This study investigates the feasibility of downscaling satellite soil moisture products to 1 km resolution. This study was undertaken in the semi-arid Goulburn River Catchment, located in south-eastern Australia. The Soil Moisture Active Passive (SMAP)-Enhanced 9 km (L3SMP-E) and Soil Moisture and Ocean Salinity (SMOS) 25 km gridded (SMOS CATDS L3 SM 3-DAY) radiometric products were compared with in-situ soil moisture observations and a regression tree model was developed for downscaling based on thermal inertia theory. Observations from a long-term soil moisture monitoring network were employed to develop a regression tree model between the diurnal temperature difference and the daily mean soil moisture for soils with different clay content and vegetation greenness. Moderate-resolution Imaging Spectroradiometer (MODIS) land surface temperatures were used to estimate the soil moisture at high spatial resolution by disaggregating the satellite soil moisture products through the regression model. The downscaled SMAP-Enhanced 9 km and SMOS 25 km gridded soil moisture products showed unbiased root mean square errors (ubRMSE) of 0.07 and 0.05 cm3/cm3, respectively, against the in-situ data. These ubRMSEs include errors caused by measuring instrument and the scale mismatch between downscaled products and in-situ data. An RMSE of 0.07 cm3 /cm3 was observed when comparing the downscaled soil moisture against the passive airborne L-band retrievals. The findings here auger well for the use of satellite remote sensing for the assessment of high resolution soil moisture.
- Soil moisture