Data from: A Spatial Dependent Model for Climate Emulation

For studying impacts and policy issues related to climate change, it is often critical to be able to forecast the future climate for a range of forcing scenarios. Complex climate models can be used to study climate change, but they are expensive to run, and thus, can only be used to investigate a limited number of scenarios. For some climate summaries, it is possible to develop a statistical emulator of the climate model that accurately and quickly reproduces the climate model output. Training such an emulator based on a small number of model runs can be challenging, especially when emulating at a fine spatial resolution. This work considers developing such an emulator for a specific climate model, CCSM3, as a function of the past trajectory of atmospheric CO₂ concentrations. We propose a new approach to fitting an emulator for annual temperature at the pixel level of the climate model by combining a spatially varying coefficient model and an infinite distributed lag model. The approach can capture the annual mean temperature at grid-cell level of climate model output in transient climates based on model runs from just a single CO₂ trajectory. We apply the approach to annual temperature emulation over North America and Africa, and show that the resulting emulator predicts annual temperature quite well and that the emulator can be fit in a computationally efficient manner. We show that the emulator outperforms procedures that do not take account of the spatial structure.

为研究气候变化相关影响与政策议题，针对一系列气候强迫情景开展未来气候预测往往是至关重要的核心工作。复杂气候模型可用于气候变化研究，但运行成本高昂，因此仅能对有限数量的情景展开探究。针对部分气候统计分析场景，可构建气候模型统计模拟器（statistical emulator），以精准且快速地复现气候模型的输出结果。基于少量模型运行结果训练此类模拟器颇具挑战，尤其是在精细空间分辨率下开展模拟时。本研究针对特定气候模型CCSM3，以大气二氧化碳（CO₂）浓度的历史变化轨迹为自变量，尝试构建此类模拟器。本研究结合空间变系数模型与无限分布滞后模型，提出了一种在气候模型像素尺度下构建年气温统计模拟器的全新方法。该方法仅需基于单条CO₂浓度变化轨迹的模型运行结果，即可在瞬变气候场景下精准还原气候模型输出的格点尺度年平均气温。本研究将该方法应用于北美与非洲地区的年气温模拟任务，结果表明所构建的模拟器能够精准预测年气温，且其训练过程具备较高的计算效率。同时证实，该模拟器的性能优于未考虑空间结构的同类建模方法。