This paper examines the usefulness of the non‐stationary generalized extreme value (GEV) distribution in modelling extreme rainfall. We modelled the annual maxima of daily (AMP1) and 2‐day (AMP2) rainfall data observed during the summer rainy season, dating up to 2007 in 28 stations in South Korea. We fitted the GEV distribution to the data for each location. The location parameter of the GEV distribution was formulated as a function of time to explore the temporal trends in maximum precipitation over the course of climatic change and to predict future behaviours. We found evidence of non‐stationarity in the form of increasing trends for six stations from AMP1 and for five stations from AMP2. This trend is consistent with the results from a regional climate model derived by the A1B emission forcing of IPCC AR4. The stationary Gumbel distribution provided a good fit to the AMP1 data for 18 stations and to the AMP2 data for 15 stations. We quantified the changes in extreme rainfall for each station; the return levels and their 95% confidence intervals for various return periods are provided. Copyright © 2010 Royal Meteorological Society

A model weighting scheme is important in multi-model climate ensembles for projecting future changes. The climate model output typically needs to be bias corrected before it can be used. When a bias-correction (BC) is applied, equal model weights are usually derived because some BC methods cause the observations and historical simulation to match perfectly. This equal weighting is sometimes criticized because it does not take into account the model performance. Unequal weights reflecting model performance may be obtained from raw data before BC is applied. However, we have observed that certain models produce excessively high weights, while the weights generated in all other models are extremely low. This phenomenon may be partly due to the fact that some models are more fit or calibrated to the observations for a given applications. To address these problems, we consider, in this study, a hybrid weighting scheme including both equal and unequal weights. The proposed approach applies an “imperfect” correction to the historical data in computing their weights, while it applies ordinary BC to the future data in computing the ensemble prediction. We employ a quantile mapping method for the BC and a Bayesian model averaging for performance-based weighting. Furthermore, techniques for selecting the optimal correction rate based on the chi-square test statistic and the continuous ranked probability score are examined. Comparisons with ordinary ensembles are provided using a perfect model test. The usefulness of the proposed method is illustrated using the annual maximum daily precipitation as observed in the Korean peninsula and simulated by 21 models from the Coupled Model Intercomparison Project Phase 6.

Attempts to assess the changes between the observed (or historical) and future projected daily rainfall extremes for 59 stations throughout Korea have been made with descriptive statistics and extreme value analysis. For the comparison, three different periods and four different data sets are considered: observation and historical data from 1976 to 2005 (period 0), simulation from 2021 to 2050 (period 1) and from 2066 to 2095 (period 2). The historical and projected rainfalls are obtained fromRCP4.5 andRCP8.5 scenarios, which are based on a regional climate modelHadGEM3‐RA. For the comparison of extreme values, the 20‐ and 50‐year return levels and the return period estimates are obtained by using the best one between two extreme value distributions, the method of L‐moments and the regional frequency analysis. From the descriptive statistics, we find that the numbers of heavy rainfall events will increase in the future. The total precipitation is projected to remain unchanged or slightly increased, compared to the observation. From the extreme value analysis, we realize that a 1‐in‐20 year and a 1‐in‐50 year annual maximum daily precipitation will likely become a 1‐in‐10 year and a 1‐in‐16 year event, respectively, when compared to the observation (a 1‐in‐5 year and a 1‐in‐7 year event, compared to the historical data), by the end of the 21st century. But this finding is based on only one simulation model, which confines the confidence of the result and suggests an ensemble approach based on multiple models to get more reliable result.