Multi-objective adaptive surrogate modeling-based optimization for distributed environmental models based on grid sampling

Abstract:

Parameter optimization is needed for reliable simulations and predictions of environmental models. Efficient parameter optimization for complex models requires surrogate modeling to reduce the number of model evaluations. However, building a surrogate of a distributed environmental model with many output variables is computationally intensive because it involves a large number of expensive model simulations. In this study, a novel calibration method called the multi-objective adaptive surrogate modeling-based optimization using grid sampling is introduced. This method first samples representative grids from the spatial domain based on the parameter sensitivity information of each grid, then conducts the surrogate modeling-based multi-objective optimization on the representative grid points. Through running the model at a representative set of grids as well as reducing the number of model runs, a considerable amount of computational cost can be saved. We apply this method to calibrating the Noah‐MP land surface model focusing on two output variables - gross primary production (GPP) and latent heat flux (LH), for two vegetation types across the continental United States. Results demonstrate that conducting multi-objective optimization on only 10% of total grids sampled using the representative grid screening strategy can significantly improve simulations of GPP and LH simultaneously. Our calibration strategy shows similar effectiveness but is much more efficient than the traditional calibration strategy based on evaluating cell-to-cell performance on all grid cells. This strategy will greatly enhance the computational efficiency of model calibration to improve model performance and advance our understanding of the environmental systems.