复杂陆面过程模型的参数不确定性定量化研究

项目名称： 复杂陆面过程模型的参数不确定性定量化研究

项目编号： No.41505092

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 天文学、地球科学

项目作者： 甘衍军

作者单位： 中国气象科学研究院

项目金额： 21万元

中文摘要： 陆面过程模型定量刻画土壤—植被—大气间能量、动量和物质的交换过程，是大气环流模型、区域气候模型和数值天气预报模型的重要组成部分。如何量化陆面过程模型的参数不确定性、正确定义模型参数取值，对提高陆面过程模型本身及其耦合模型的模拟效果至关重要。由于陆面过程模型结构复杂、参数众多、计算成本高，地学领域中常用的不确定性分析方法对这种大复杂模型的适用性有限。本研究拟研发一个系统的不确定性定量化框架，以较少的实验研究陆面过程模型参数不确定性问题。具体包括：综合利用定性和定量敏感性分析方法的优点，逐步识别出控制模型的敏感参数；构建统计仿真器，替代原始陆面过程模型进行统计仿真，在此基础上进行多目标参数优化。最终量化并降低模型不确定性、优化模型参数、提高模型模拟能力，为解决大复杂系统模型参数不确定性定量化问题提供思路。

中文关键词： 陆面过程模型；不确定性定量化；统计仿真器；敏感性分析；参数优化

英文摘要： Land surface models (LSMs) quantitatively characterize the exchanging processes of energy, momentum, and material in the soil-vegetation-atmosphere systems. They are indispensable in general circulation models (GCMs), regional climate models (RCMs), and numerical weather prediction (NWP) models. How to quantify parametric uncertainty of a LSM and set proper parameter values for it, are not only essential for improving the performance of the LSM, but also for its coupling models. Due to the complex structure, high dimensionality, and high computational cost of LSMs, commonly used uncertainty quantification methods are sometimes inapplicable for this kind of large complex models. We plan to develop a systematic uncertainty quantification framework, for exploring parametric uncertainties of LSMs with less experiments. These include: stepwise identification of parameter sensitivities by combining the advantages of qualitative and quantitative sensitivity analysis methods; multi-objective optimization of model parameters using statistical emulator other than running the original LSM. The ultimate objectives are to quantify and reduce model uncertainties, and improve model performance. This study would put forward ideas for solving parametric uncertainty quantification problems of large complex system models.

英文关键词： Land surface model;Uncertainty quantification;Statistical emulator;Sensitivity analysis;Parameter optimization