Statistical Multisite Simulation of Summertime Precipitation over South Korea and Its Future Change Based on Observational Data

Abstract

In this study, a weather generator for summer (May 19 – September 15) precipitation over South Korea is developed. Precipitation data for 33 years (1979-2011) observed at 57 stations of Korea Meteorological Administration (KMA) are used to develop a new weather generator. Using the cyclostationary empirical orthogonal function (CSEOF) technique, the observed precipitation data is described as a linear combination of deterministic evolution patterns during summer and corresponding stochastic amplitude (principal component) time series. The deterministic evolution patterns repeat themselves every year while their amplitudes vary from one year to another. Then, an autoregressive moving average (ARMA) model was fit to each of the detrended stochastic amplitude time series. The resulting ARMA models are used to generate synthetic amplitude time series of arbitrary length such that they have statistical properties similar to the original amplitude time series. Then, one hundred sets of new amplitude time series are constructed for the period of 1979-2061 (83 years). Based on these synthetic time series and the annually repeating evolution patterns, one hundred sets of synthetic summer precipitation records were generated. Statistical characteristics of the synthetic datasets are examined in comparison with those of the KMA observational record for the period of the observational record. The synthetic datasets, particularly for a future period, are also compared with the precipitation from HadGEM3-RA regional climate model (RCM) under various climate change (representative concentration pathways: RCP) scenarios for further validation of the performance of the statistical approach developed in the present study. The seasonal cycle in the synthetic precipitation is reproduced faithfully with typical bimodal peaks of summer precipitation although its amplitude is slightly overestimated. The RCM underestimates significantly the second peak of precipitation. The mean and the standard deviation of summer precipitation in the seasonal cycle also exhibit similar characteristics

the synthetic data reasonably reproduce the mean and the standard deviation of the seasonal cycle whereas the historical run data significantly underestimate them. The frequency-intensity relationship of the synthetic precipitation also looks similar to that of the observational data. The frequency-intensity relationship of the model precipitation under the historical run significantly underestimates the occurrence of no-rain days and rainfall events with precipitation greater than 20 mm day-1 whereas the number of rainfall events with precipitation between 0 and 20 mm day-1 is seriously overestimated. The distribution of extreme events is delineated in terms of generalized Pareto distribution (GPD). The distribution of extreme events in the synthetic data is reasonably similar to that of the observational data whereas the model data under the historical run underestimates the occurrence of extreme events. The spatial correlation patterns of both the synthetic precipitation and the historical run precipitation are fairly similar to that of the observational data

both the statistical approach and the model reproduce the spatial characteristics of summer precipitation in Korea reasonably. In the future period, precipitation amounts increases except in the precipitation range of (0,10) mm day-1 with nearly no change in the frequency of no-rain days

frequency increase is particularly conspicuous in the range of [100,500) mm day-1. The model precipitation under the RCP 4.5 and RCP 8.5 scenarios both exhibit increased frequency of heavy precipitation greater than 100 mm day-1 compared with the historical run. The relative frequency of heavy precipitation clearly increases according to the GPD distributions of RCP 4.5 and RCP 8.5 precipitation. Nonetheless, the future frequency of heavy precipitation in the range of [100,500) mm day-1 in the model datasets is still generally below that of the observational record. On the other hand, heavy precipitation events exceeding 500 mm day-1 are clearly more abundant than in the observational record.