International Journal of Applied Earth Observations and Geoinformation (Jun 2024)
A stepwise method for downscaling SMAP soil moisture dataset in the CONUS during 2015–2019
Abstract
The Soil Moisture Active Passive (SMAP) product, which delivers spatially continuous soil moisture (SM) data with reliable accuracy, has been widely used in ecologically and hydrologically related studies. However, the active radar of SMAP failed after operation for approximate three months, leading to the loss of the original 3 km (called A3) and 9 km (called AP9) SMAP SM dataset afterwards. Since merely the 36 km SMAP dataset (called P36 in this research) is available up to now, the relatively coarse spatial resolution is difficult to meet the requirement for monitoring at the regional scale. In this paper, the 36 km SMAP SM dataset in the conterminous United States (CONUS) was downscaled to 3 km after the failure of the active radar. To reduce the great uncertainty in downscaling (i.e., a process from 36 km to 3 km directly), a stepwise strategy was proposed. Specifically, our currently published 9 km SMAP SM dataset (called VIP9, from April 2015 to December 2019) was used as input for downscaling, which was downscaled to 3 km by fusion with auxiliary data (e.g., Land Surface Temperature (LST) and Normalized Difference Vegetation Index (NDVI)) at fine spatial resolution based on area-to-point regression kriging (ATPRK). Furthermore, a random forest (RF)-based correction process was developed to enhance the quality of the downscaled SM dataset by fusing with the ground-based SM data (i.e., in-situ data). As a result, a 16-day composited SM dataset at 3 km spatial resolution was produced from 2015 to 2019, with a mean correlation coefficient (CC) and unbiased root mean square error (ubRMSE) in temporal (spatial) validation of 0.888 (0.912) and 0.021 (0.023), respectively. Experimental results demonstrated that the AP9 dataset (although only available in the three months) is helpful to increase the accuracy of downscaling. Moreover, the difference between satellite- and ground-based SM data can be further reduced through the correction process. The produced 3 km SMAP SM dataset has great potential to extend the defunct A3 data and support related studies. The contributions of the paper are twofold, including the development of the stepwise method for downscaling and the generation of the 3 km SMAP SM dataset in the CONUS during 2015–2019.
