Study on the Method of Multi-Objective Parameters Optimization for Land Surface Models Based on Data Assimilation

Abstract

Data assimilation is one effective method of integrating the advantages of multiple observation data from ground stations and remote sensing, optimizing the land surface process model parameters and promoting the regional simulation capability of land surface models, which is of great significance for improving the simulation effect of the process of hydrological cycle and obtaining effective temporal-spatial continuous hydrological factors so as to provide effective data supporting for efficient agricultural water management. We develop a multi-objective parameters optimization method for land surface models based on multi-element data assimilation. Taking two interrelated factors, soil moisture and land surface brightness temperature as examples, we build multi-element data assimilation scheme of soil moisture and land surface brightness temperature based on Ensemble Kalman Filter (EnKF) and Noah land surface model, for the purpose of eventually improving temporal-spatial continuous soil moisture simulation. Then we build multi-objective optimization function which is set by using observed values, simulated values and assimilation values of soil moisture and land surface brightness temperature based on multi-objective optimization concept. Meanwhile, the Shuffled Complex Evolution (SCE-VA) algorithm is used to minimize the multi-objective function value of the soil moisture and land surface brightness temperature, so as to achieve best fit between the observed values, simulation values, and assimilation values of the soil moisture and land surface brightness temperature and optimize the parameters related to soil moisture (saturated hydraulic conductivity, saturated soil water potential, Clapp and Hornberger constant, vegetation roughness, etc.). Finally, we evaluate the validation of optimization results. Taking Hetao irrigation district as an example, we obtain the soil moisture data retrieved by multi-remote sensing data through the TVDI index method and land surface brightness temperature derived from single-window algorithm. We set up a data assimilation model and carry out method tests. The result shows that the method of multi-objective model parameters optimization based on data assimilation of multi-hydrological factors is better than the original model simulation, which has indeed improved the simulation capability of the land surface model. We provide the valuable experience for study on multi-objective parameters optimization method and multi-element data assimilation method based on land surface model.