资源环境科技发展态势分析平台

Global climate models (GCMs) provide the fundamental information used to assess potential impacts of future climate change. However, the mismatch in spatial resolution between GCMs and the requirements of regional applications has impeded the use of GCM projections for impact studies at a regional scale. This study applied statistical post-processing methods that preserve long-term temporal trends, bias-correction/spatial disaggregation with detrended quantile mapping (SDDQM) and BCSD with quantile delta mapping (SDQDM), to downscale 20 CMIP5 GCM climate projections for daily precipitation, minimum temperature, and maximum temperature over South Korea. Using the downscaled CMIP5 climate projections, we investigated absolute changes in extreme indices between the reference and three 30-year future periods. In addition, the biases in change signals from GCM projections for different statistical downscaling methods were compared to evaluate how well long-term trends in indices are preserved. The results showed that the statistical downscaling methods significantly improved the skill in reproducing extreme indices. For temperature-related extreme indices, we found strong significant trends while trends for precipitation-related indices varied depending on the index and climate projection horizon. Specifically, more frequent, longer duration, and more intense hot extremes may occur under the CMIP5 climate projections, while corresponding decreases may occur for extreme cold indices. Prominent upward trends are found in extreme precipitation events. Regarding analysis of the bias in change signals, SDQDM, which explicitly preserves changes in all quantiles of the underlying variables, better preserved long-term trends in extreme indices simulated by GCMs.