Linear Meta-Model optimization for regional climate models (LiMMo version 1.0)

Abstract

A new tool for objective parameter tuning of regional climate models is presented. The climate model output was emulated using a linear regression approach for each grid point on a monthly mean basis. This linear approximation showed decent accuracy over a 6-year period. The root-mean-square error norm between the Meta-Model and the observational data sets was minimized using the gradient-based, limited-memory Broyden-Fletcher-Goldfarb-Shanno method with box constraints. We refer to this framework as LiMMo (Linear Meta-Model optimization). The LiMMo framework was applied to the state-of-the-art regional climate model ICON-CLM, tuned to the E-OBS and HOAPS observational data sets. Different optimization objectives were explored by assigning varying weights to model variables in the error norm definition. The combination of a linear emulator with fast gradient-based optimization allows the proposed method to scale linearly with the number of model variables and parameters, facilitating the tuning of dozens of parameters simultaneously.