Evaluation of the value of spatially improved SMAP soil moisture products in enhancing streamflow forecast skills

Soil moisture (SM) measurements over large areas are vital for many operational applications such as flood forecasting, irrigation scheduling, and drought monitoring. Although obtaining SM over extensive areas was difficult until recently, the advent of satellite remote sensing technologies such as passive microwave satellites (e.g., SMOS and SMAP) opened a new way. Nevertheless, the utilization of SM products of these satellites is often impeded because of their coarse spatial resolution (i.e., about 40 km). A number of studies have been attempted to improve the coarse resolution satellite SM products via downscaling. However, despite of many downscaling efforts, subsequent use of downscaled satellite SM products for operational applications has not yet been fully explored. Thus, the objective of this study is to evaluate the value of SMAP SM in enhancing short-term streamflow forecast skills. The random forest machine learning technique was used to downscaled SMAP SM from 36 km to a range of resolutions from 1 to 9 km (i.e., 9, 3, and 1 km). Thereafter, a host of experiments were carried out to update a physically-based distributed hydrological model through direct ingestion of the original SMAP SM (e.g., 36 km), SMAP enhanced SM (i.e., 9 km), and downscaled SMAP SM at different spatial resolutions (e.g., 9, 3 and 1 km). A non-updated model was used as a benchmark for comparison. The result shows that the downscaled SMAP SM has presented better spatial detail than its corresponding native resolution and updating the model state with SMAP SM products (i.e., with the native and downscaled products) shows promising potential for improving short term flood forecasting. Finally, this will in turn helps in better water resources management.