A data assimilation-based method for optimizing parameterization schemes in a land surface process model

Abstract

Optimizingtheparameters of a land surface process model (LSPM) through data assimilation (DA) can not only improve and perfect theparameterization schemes in the LSPM through the physical mechanism, but also increase its regional adaptability and simulation capability. This has practical importance for improving simulation results and the climate-prediction capability of general circulation models (GCMs) and regional climate models (RCMs). This paper presents a DA-based method for optimizingtheparameterization schemes in LSPMs. We optimize the unsaturated-soil water flow (UnSWF) model as an example by developing a soil-moisture assimilation scheme based on the UnSWF model and the extended Kalman filter (EKF) algorithm, and then combining them with the Variable Infiltration Capacity (VIC) model. Using a month as the assimilation window, we used the Shuffled Complex Evolution-University of Arizona (SCE-UA) algorithm to minimize the objective function through simulated and assimilated soil moisture, achieved the best fit with the given objective function measurement, and optimized the parameters of the UnSWF model, including the saturated-soil hydraulic conductivity, moisture content, matrix potential, and the Clapp and Hornberger constant. The optimal values of the model parameters were obtained during the DA period (the year 1986), and then the optimized parameters were used to improve the UnSWF model. Finally, numerical simulation experiments were carried out from 1986 to 1993 to evaluate the simulation capability of the improved model and to explore and realize the DA-based method for optimizing the soil water parameterization scheme in LSPMs. The experimental results indicated that the optimized model parameters improved and perfected the model based on the physical mechanism, and increased its simulation capability; the optimized model parameters had good temporal portability and their adaptability was stronger, achieving the aim of improving the model. Therefore, this method is reasonable and feasible. This paper provides a good reference for DA-based optimization of the parameterization schemes in LSPMs.