Downscaling SMAP Radiometer Soil Moisture Over the CONUS Using an Ensemble Learning Method

Abstract

Soil moisture plays a critical role in improving the weather and climate forecast and understanding terrestrial ecosystem processes. It is a key hydrologic variable in agricultural drought monitoring, flood forecasting, and irrigation management as well. Satellite retrievals can provide unprecedented soil moisture information at the global scale; however, the products are generally provided at coarse resolutions (25–50 km 2 ). This often hampers their use in regional or local studies. The National Aeronautics and Space Administration Soil Moisture Active Passive (SMAP) satellite mission was launched in January 2015 aiming to acquire soil moisture and freeze-thaw states over the globe with 2 to 3 days revisit frequency. This work presents a new framework based on an ensemble learning method while using atmospheric and geophysical information derived from remote-sensing and ground-based observations to downscale the level 3 daily composite version (L3_SM_P) of SMAP radiometer soil moisture over the Continental United States at 1-km spatial resolution. In the proposed method, a suite of remotely sensed and in situ data sets are used, including soil texture and topography data among other information. The downscaled product was validated against in situ soil moisture measurements collected from two high density validation sites and 300 sparse soil moisture networks throughout the Continental United States. On average, the unbiased Root Mean Square Error between the downscaled SMAP soil moisture data and in-situ soil moisture observations adequately met the SMAP soil moisture retrieval accuracy requirement of 0.04 m 3 /m 3 . In addition, other statistical measures, that is, Pearson correlation coefficient and bias, showed satisfactory results.