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