Vine Copula在GCMs多模式降雨集成中的应用

Application of Vine Copula in GCMs multi-mode ensemble for precipitation

全球气候模式(global climate models,GCMs)数量众多,各有优劣,对降水量而言往往难以确定出一个最优模式,所以当将其输出用于水文长期预估时,通常需对各模式数据进行集成,以发挥不同模式的优势,提升水文预估的整体精度。采用Vine Copula构建GCMs与实测降雨的多维联合分布函数,并推求给定GCMs数据条件下实测降雨量的条件分布,再由该条件分布实现多维数据的综合。以淮河王家坝以上流域6个GCMs降雨数据的综合为例进行应用研究,并与贝叶斯模型平均和多元分位数回归2种多变量集成方法进行比较。结果表明,基于Vine Copula的多模式集成结果优于任意原始单模式,且具有整体最优的集成效果,为GCMs在水文中的应用提供了一种途径。

There are many global climate models(GCMs)with different advantages and disadvantages.Especially for precipitation prediction,it is difficult to determine an optimal model.Therefore,when the output of GCMs is used for hydrological long-term prediction,it is usually necessary to integrate the data of each model to take advantage of different models and improve the overall accuracy of hydrological prediction.In this paper,the multi-dimensional joint distribution function of GCMs and observed precipitation is constructed by using Vine Copula,and the conditional distribution of observed precipitation with given GCMs data is derived,then the multi-dimensional data synthesis is realized by the conditional distribution.In this paper,the synthesis of six GCMs precipitation data in the watershed above Wangjiaba in Huaihe River is taken as an example,and compared with the traditional Bayesian model averaging and multivariate quantile regression.The results show that the multi-model ensemble based on Vine Copula is better than any original single model,and has the overall optimal ensemble effect,which provides a means for the application of GCMs in hydrology.