Comparison of Metrics for Probabilistic Climate Change Projections of Mediterranean Precipitation (COMEPRO)

气候减缓和适应战略需要对未来气候变化进行可靠的估计，而耦合的大气-海洋环流模式（aogcm）仍然是最先进的工具。然而，不同气候模式的气候预估结果在地球某些区域和某些气候变量（例如降水）方面差别很大，这意味着气候影响研究方面存在重大的不确定性。气候模式模拟之间的差异源于未知的初始条件、不同的分辨率和强迫因子、不同的物理参数化和假设的排放情景。由于无法预测哪种模式预测最真实，最全面的气候变化信息由所有可用的气候变化模拟组成。通过将这种多模式信息转换成概率密度函数（PDF），可以直接评估高于或低于给定气候变化水平的概率。这种概率预测确实与规划过程有关，因为它们提供了代表期望的气候变化的多模式平均估计，以及模式不确定性的定量测量。该项目旨在评估CMIP3、CMIP5、EMSEMBLES和CORDEX多模式集成以及统计降尺度方法对未来降水变化的概率预测。重点是地中海地区作为所谓的气候变化热点，以及季节平均值和极端值之间pdf的比较。本文的主要新颖之处在于，通过对参与气候变化预估的适当加权因子，可以在多大程度上减小多模式pdf的宽度。这意味着一定数量的模型不确定性必须被接受，因为它是由未知的初始条件或不确定的经验参数引起的内在问题。权重因子通过贝叶斯统计、回归分析、时空滤波、最优指纹和模型性能等各种比较指标进行评估，这些指标涉及动态模式、环流类型和统计降尺度技术的重要预测因子的表示。此外，还评估了多模式组合的pdf对降水产生过程链中的各种现象的影响，包括辐射、蒸发、平流和云量。这将有助于确定不同气候模式预估之间开始出现差异的水平。

Climate mitigation and adaptation strategies require reliable estimates of future climate change for which coupled atmosphere-ocean general circulation models (AOGCMs) still represent the state-of-the-art tool. However, climate projections from different climate models vary considerably in some regions of the Earth and in terms of some climatic variables like for instance precipitation, implying important uncertainties on the side of climate impact research. The differences between climate model simulations arise from the unknown initial conditions, different resolutions and forcing agents, different physical parameterizations and the assumed emission scenarios. As it cannot be anticipated which model projection is most realistic, the most comprehensive climate change information is composed of all available climate change simulations. By transforming this multi-model information into a probability density function (PDF) the probability above or below a given level of climate change can directly be assessed. Such probabilistic predictions are indeed relevant for planning processes because they provide a multi-model mean estimate of climate change which represents expectation, and a quantitative measure of model uncertainty. The proposed project aims at evaluating such probabilistic predictions of future precipitation changes from the CMIP3, CMIP5, EMSEMBLES and CORDEX multi-model ensembles and from statistical downscaling approaches. The focus is on the Mediterranean region as a so-called hot spot of climate change and on the comparison of PDFs between seasonal means and extremes. The major novel aspect is to investigate to what extent the width of the multi-model PDFs can be reduced by appropriate weighting factors of the participating climate change projections. This implies that some amount of model uncertainty has to be accepted because it is an intrinsic problem arising from the unknown initial conditions or from uncertain empirical parameters. The weighting factors are assessed by various comparative metrics like Bayesian statistics, regression analysis, spatio-temporal filtering, optimal fingerprinting, and model performance concerning the representation of dynamical modes, circulation types and important predictors for statistical downscaling techniques. In addition, the PDFs over the multi-model ensembles are also assessed for various phenomena in the process chain towards the generation of precipitation, including radiation, evaporation, advection and cloudiness. This will help to identify the level at which the dispersion between different climate model projections begins.