一种基于敏感性分析的改进参数筛选方法

An effective parameter screening method based on sensitivity analysis

地球系统模式是预测未来气候变化的重要工具。物理参数化方案是其关键部件,其包含的大量不确定参数会严重影响模型模拟性能。而参数不确定性分析面临参数个数过多的挑战。传统敏感性分析方法被广泛用于识别重要参数,然而它并没有考虑参数交互作用的动态变化对筛选过程的影响。对此,利用一个典型非线性数学函数分析该变化导致的筛选结果偏差,进而提出了动态敏感性分析方法(Dynamic Sensitivity Analysis Method,DSAM)。在一个数值函数和单柱大气模式的实验中,DSAM和传统方法的定量筛选结果偏差高达29%。将筛选结果应用于参数优化,模式的优化性能提升67%。

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 to29%.In parameter optimization based on the screening results,the optimization results of the model increased by67%.