Assessing optimal set of implemented physical parameterization schemes in a multi-physics land surface model using genetic algorithm

S Hong,S K Park,B Myoung,X Yu,Y.-S Choi

doi:10.5194/gmd-7-2517-2014

Abstract

Abstract. Optimization of land surface models has been challenging due to the model complexity and uncertainty. In this study, we performed scheme-based model optimizations by designing a framework for coupling "the micro-genetic algorithm" (micro-GA) and "the Noah land surface model with multiple physics options" (Noah-MP). Micro-GA controls the scheme selections among eight different land surface parameterization categories, each containing 2–4 schemes, in Noah-MP in order to extract the optimal scheme combination that achieves the best skill score. This coupling framework was successfully applied to the optimizations of evapotranspiration and runoff simulations in terms of surface water balance over the Han River basin in Korea, showing outstanding speeds in searching for the optimal scheme combination. Taking advantage of the natural selection mechanism in micro-GA, we explored the model sensitivity to scheme selections and the scheme interrelationship during the micro-GA evolution process. This information is helpful for better understanding physical parameterizations and hence it is expected to be effectively used for further optimizations with uncertain parameters in a specific set of schemes.

Highlights

Land surface models (LSMs) have significantly advanced in recent years, but their optimization has been challenging due to their increased complexities and number of uncertainties, which require tremendous computing resources
The micro-genetic algorithm (GA), which enables smart and fast searching for the optimum, was introduced and applied to a new version of the Noah Land Surface Model (LSM) with multiple physics options (NoahMP) in various physical parameterization categories
We implemented an interface between Noah-MP and micro-genetic algorithm” (GA) and developed a coupled system (MP–MGA)

Summary

Introduction

Land surface models (LSMs) have significantly advanced in recent years, but their optimization has been challenging due to their increased complexities and number of uncertainties, which require tremendous computing resources. Model optimization, which calibrates uncertain parameters based on observations, is one of the widely used methods that apply model runs to large-scale studies. Such methods often include parameter sensitivity analyses for effective optimizations (Gupta et al, 2000; Jackson et al, 2003; Mo et al, 2008; Nasonova et al, 2011; Rosero et al, 2010; Williams and Maxwell, 2011). To make model runs more reliable, previous studies have calibrated several uncertain parameters in only one or two schemes related to their targeted variables (Cretat and Phol, 2012; Essery et al, 2013; MiguezMacho and Fan, 2012), sometimes multiple parameters in Published by Copernicus Publications on behalf of the European Geosciences Union

Methods

Results

Conclusion