While climate models are an invaluable tool for increasing our understanding and therefore, the predictability of the Earth’s system for decades, their increase in complexity and resolution has put a considerable, growing strain on the computational resources of research centers and institutions worldwide. The statistics community has a long history of developing stochastic models as a means to save computational time, but the emergence of storage as an additional cost for climate investigations has prompted a reformulation of the aim of statistical models in model-based environmental science. Can stochastic approximations be useful as a mechanism for saving both computational time and storage? We focus on a collection of simulations from a climate model and propose several statistical models of increasing complexity. By analyzing and discussing the associated costs for each model, we demonstrate how computation and storage are closely intertwined, and how a statistical model of increasing complexity is justified only to the extent that information at a fine spatial and/or temporal scale is sought to be preserved.Supplementary materials accompanying this paper appear online.

虽然气候模型是增加我们的理解的宝贵工具，因此，几十年来地球系统的可预测性，它们的复杂性和分辨率的增加已经给全球研究中心和机构的计算资源带来了相当大的、越来越大的压力。统计界在开发随机模型作为节省计算时间的手段方面有着悠久的历史，但存储作为气候调查的额外成本的出现促使人们重新制定基于模型的环境科学中统计模型的目标。随机逼近是否可以用作节省计算时间和存储空间的机制？我们专注于气候模型的一系列模拟，并提出了几个越来越复杂的统计模型。