Improve the performance of the Noah-MP model by hybridizing sequential data assimilation and machine learning
Abstract
Improve the performance of the Noah-MP model by hybridizing sequential data assimilation and machine learning
Xinlei He1, Shaomin Liu1, Tongren Xu1, Sayed M. Bateni2 1 State Key Laboratory of Earth Surface Processes and Resource Ecology, Faculty of Geographical Science, Beijing Normal University, Beijing, China 2 Department of Civil and Environmental Engineering and Water Resources Research Center, University of Hawaii at Manoa, Honolulu, Hawaii, USA In many eco-hydrological practices, the land data assimilation system (LDAS) has been increasingly used to dynamically improve various land surface variables such as soil moisture, leaf area index, and evapotranspiration (ET). However, accurately describing model errors in simplified physical models poses a major challenge to the successful implementation of data assimilation. In this paper, data-driven soil moisture bias correction models are integrated into the ensemble Kalman filter (EnKF) method to reduce the predicted bias of the Noah-MP model and to provide more robust soil moisture and ET estimates. Data-driven soil moisture bias correction models are established based on the machine learning (ML) method to account for the bias of the Noah-MP model. Then, the data assimilation system based on the unbiased Noah-MP model is used to estimate ET. The estimated ET from the developed EnKF (EnKF-ML) method is compared with the large-aperture scintillometer (LAS) ET measurements at the grassland (Arou), cropland (Daman), and shrubland (Sidaoqiao) sites in the Heihe River Basin (HRB). The findings indicate that the EnKF-ML method performed well across various hydrological and vegetative conditions. The root mean square deviation (RMSD) of daily ET estimates from EnKF-ML are 43.09%, 52.77%, and 65.22% lower than those of Noah-MP at the Arou, Daman, and Sidaoqiao sites, respectively. This study indicates that ML methods can exploit in-situ SM observations to improve the performance of land data assimilation systems and provide more accurate estimates of land surface variables.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.B11B..01H