Multi-Sensor Data Assimilation for Global Soil Moisture Estimation Using the Local Ensemble Transform Kalman Filter

Abstract

This study investigates whether assimilating soil moisture (SM) observations from multiple sensors can increase spatiotemporal coverage and improve SM estimates. A global SM data assimilation (DA) system based on the Local Ensemble Transform Kalman Filter is used to merge SM retrievals from the Soil Moisture Active Passive (SMAP) mission, the Soil Moisture and Ocean Salinity (SMOS) mission, the Advanced Scatterometer (ASCAT), and the Advanced Microwave Scanning Radiometer 2 (AMSR2) into the Joint UK Land Environment Simulator (JULES) land model. The SM retrievals are assimilated both separately and jointly over the boreal warm seasons (May–September) of 2015–2021. The resulting SM estimates are validated using an Instrumental Variable approach globally, and against insitu measurements in North America, Europe, and East Asia. The four single-sensor DA experiments result in a global average improvement of 0.045 in the anomaly correlation coefficient (R) compared to the model-only (Openloop) simulation. The assimilation of SMAP retrievals yields the best R over 48 % of the global land area, followed by ASCAT (24%), AMSR2 (16%), and SMOS (12%). Furthermore, validation against in-situ measurements shows that all single-sensor DA experiments improve surface and root-zone SM estimates relative to the Openloop. Again, SMAP provides superior single-sensor benefit. The multi-sensor assimilation achieves additional gains by merging complementary observations. These benefits are also evident at the sub-daily timescale, with skill improvement both during daytime and nighttime. The skill improvement of the multi-sensor DA over single-sensor DA at each sub-daily time step is associated with the overpass times of the individual sensors and their respective performance. Overall, this study demonstrates that expanding the spatiotemporal coverage of satellite observations through multi-sensor assimilation provides better SM estimates.