Time of unprecedented climate for extreme temperature in winter averaged over Korea

For extreme temperature, we used climate extreme indices provided by CLIVAR (Climate and Ocean-Variability, Predictability, and Change) ETCCDI (Expert Team on Climate Change Detection and Indices). ETCCDI has provided 27 climate extreme indices not only with global reanalysis datasets but with CMIP5 simulations. The indices data are available on-line and the results with CMIP5 simulations were summarized by Sillmann et al. [2013]. For our analysis, we downloaded a monthly minimum of daily minimum surface air temperature (TNn) and a monthly maximum of daily maximum temperature (TXx). Among the CMIP5, 27 model results available on their website, we used 23 model results containing both of the TNn and TXx for all of the historical, RCP 4.5 and 8.5 experiments. Since our focus is on boreal-winter extreme temperature, we selected the lowest TNn and highest TXx among the three months of December-January-February every year from 1861 to 2005 for the historical simulation and from 2006 to 2099 for the RCP 4.5 and RCP 8.5 scenario. Before the spatial averaging over the analysis domain (34°N-43°N in latitude and 124°E-131°E in longitude including the Korean Peninsula), we had remapped all of the simulation data onto a 1.5° x 1.5° horizontal resolution. The time of unprecedented climate (TUC) for extreme temperature is defined in this study as the beginning year when the extreme temperature projected for the future climate scenarios exceed a critical value in all subsequent years during the RCP scenario runs. In this study, the critical value for extreme temperatures is specified as a 50-year return level which is rather arbitrary but refers to a rough estimate for the social lifetime of a man. One may find the return level empirically from historical data, but this study estimates it using a Generalized Extreme Value distribution function as suggested by Kharin et al. [2007]. Based on the CMIP5 historical simulation data using R, we obtained three parameters determining a GEV distribution for each model, respectively for TNn and TXx. The GEV distribution for each model and variable has been verified using a Q-Q (quantile-quantile) plot if it adequately describes the CMIP5 historical data. All of the models showed the Q-Q plot within the 95% confidence range (Figure 1a for GFDL-ESM2G TXx for an instance). Then, we estimated the return level from the distribution and TUC from the RCP scenario runs for the wintertime TNn and TXx averaged over Korea.

针对极端气温研究，我们采用了气候与海洋变率、可预报性及变化计划（CLIVAR）下属气候变化检测与指数专家小组（ETCCDI）提供的气候极端指数数据集。ETCCDI共提供27项气候极端指数，其数据不仅基于全球再分析数据集，还涵盖耦合模式比较计划第五阶段（CMIP5）的模拟结果。该指数数据可在线获取，其基于CMIP5模拟的相关结果由Sillmann等人[2013]汇总整理。本研究选取了两项指数：日最低地表气温月最小值（TNn）与日最高气温月最大值（TXx）。在该网站公开的27个CMIP5模式结果中，我们选取了同时包含TNn与TXx数据的23个模式结果，用于历史试验、典型浓度路径4.5（RCP4.5）及典型浓度路径8.5（RCP8.5）三类模拟试验。鉴于本研究聚焦北半球冬季极端气温，我们针对历史模拟试验（1861-2005年）及RCP4.5、RCP8.5情景（2006-2099年），分别从每年12月、1月、2月中选取最低TNn与最高TXx值。在对包括朝鲜半岛在内的纬度34°N-43°N、经度124°E-131°E的研究区域进行空间平均前，我们将所有模拟数据重映射至1.5°×1.5°的水平分辨率网格。本研究将极端气温的空前气候年份（TUC）定义为：在RCP情景模拟中，未来气候情景预估的极端气温在后续所有年份均超过临界值的起始年份。本研究将极端气温的临界值设定为50年重现期水平，该设定虽带有一定主观性，但参考了人类社会平均寿命的粗略估算值。重现期水平可通过历史数据以经验方法求得，本研究则采用Kharin等人[2007]提出的广义极值分布（GEV）函数进行估算。基于CMIP5历史模拟数据并借助R语言，我们为每个模式分别针对TNn与TXx拟合得到了广义极值分布的三个参数。我们通过分位数-分位数图（Q-Q图）对各模式、各变量对应的GEV分布进行了验证，以判断其是否能够合理拟合CMIP5历史模拟数据。所有模式的Q-Q图均落在95%置信区间内（例如GFDL-ESM2G模式TXx的结果见图1a）。随后，我们基于拟合得到的分布估算了重现期水平，并通过RCP情景模拟结果，计算得到朝鲜半岛冬季平均TNn与TXx对应的空前气候年份TUC。