Delineation of the climatic rainfall regions of South Korea based on a multivariate analysis and regional rainfall frequency analyses

Regional frequency analysis (RFA) may provide more accurate estimates of rainfall quantiles than at-site frequency analysis (ASFA), especially in regions with short records. In this study, RFA is applied to 1-, 2-, 3-, 6-, 9-, 12-, 15-, 18-, 24-, and 48-h annual maximum rainfall series at 67 sites in South Korea. Procrustes analysis is used to select 33 rainfall-related and geographical variables that represent most of the statistical information from among 42 candidate variables. Both factor analysis and cluster analysis, such as fuzzy c-means (FCM) and Ward's method, are used to identify the homogeneous regions, and five regions are identified through heterogeneity measures. It is found that FCM-based regions are more appropriate to the precipitation in South Korea in terms of the homogeneity of the identified regions. To investigate the effectiveness of FCM-based regions, region-of-influence approach was applied. It is shown that the spatial pattern of rainfall is affected by main mountain ranges, the prevailing Westerlies, and the proximity of the coast. Six distributions are applied, and the generalized extreme value distribution is selected as the best-fit distribution from goodness-of-fit measures. RFAs, such as the index flood method (IFM) and regional shape estimation based on the regional L-moments algorithm, are applied to determine the growth curves of regions. Using Monte-Carlo simulations, it is concluded that RFA is more accurate than ASFA for the annual maximum rainfall data of South Korea. The IFM provides more accurate estimates than regional shape estimation in homogeneous regions, while regional shape estimation is more appropriate for use in heterogeneous regions and in homogeneous regions with a lower L-CV (coefficient of L-variation) and for the estimation of quantiles in higher tails over a 100-year return period for the applied data.