An effective parameter screening method based on sensitivity analysis

Abstract

Earth System Model(ESM) is a useful approach to predict the future climate changes. Physical parameterizations are its crucial components and contain a large number of parameters, which introduces tremendous uncertainties regarding the models′ performance. Traditional sensitivity analysis methods are widely used to identify the most influential parameters, however, they do not consider the impact of the dynamic changes of the parameter interactions on the screening process. In this study, a typical benchmark from the literature is used to analyze the bias of the screening results caused by this change. A Dynamic Sensitivity Analysis Method(DSAM) is proposed. The method was used on a numerical function, which is a typical screening test problem in computer experiment literature, as well as on a ESM(Single column Community Atmosphere Model, SCAM). DSAM and traditional methods have a quantitative difference up to 29%. In parameter optimization based on the screening results, the optimization results of the model increased by 67%.

Keywords

• parametric screening
• Earth System Models(ESMs)
• sensitivity analysis
• dynamic