Spatio-temporal analysis of extreme precipitation regimes across South Korea and its application to regionalization

Abstract

Assessing spatio-temporal variability of extreme rainfall is required to establish future plans and policies for water resource management. One of the main objectives of this study is to introduce an effective approach based on circular statistics for assessing the seasonality of the extreme precipitation. Circular statistics explicitly reflect the seasonal pattern of precipitation with maximum frequency of the timing of daily and monthly maximums. In southern Korea, a dominant frequency was identified in early July. The timing of the monthly maximum has been delayed in northern Korea. In the case of the daily maximum, the end of June is the period of most intense rainfall, with the exception of the east coast near Gangrung. A long-term temporal variation of timed monthly and daily maximums was investigated by a 30-year moving average for main stations. Monthly peak times of Seoul and Gangrung continuously moved backward while monthly peak times of Mokpo and Busan has moved forward since the 1960’s. These features could be influenced by inherent variations in the East Asian monsoon system. Given the identified spatio-temporal pattern, this study was extended to characterize regional patterns of extreme rainfall over Korea. A new concept in regionalization procedures was developed on the basis of existing approaches that mainly utilize simple moments of data. In this study, the K-means method was incorporated with the temporal pattern of the extreme rainfalls in order to better characterize hydrologic patterns for regional frequency analysis. The results showed that the proposed approach is promising for the region in term improving the physical understanding of extreme rainfall.

Introduction

We used hydrologic frequency analysis to produce guidance about the expected behavior of a specified event. The underlying assumption is that hydrologic events are random variables and are stationary, meaning that the statistical properties of hydrologic events will not change in the future. Estimation of the frequency of extreme events is often of particular interest, and point frequency analysis using a single set of data is well established. However, uncertainties associated with limited sample size in an “on site” analysis are a well documented problem in existing literature (Andres et al., 2007, Faustini and Kaufmann, 2007).

Statistically homogeneous samples of data are of particular interest in hydrologic frequency analysis, and these may be observations of the same variables scrutinized by multiple sources in terms of measuring sites and instruments. If extremes from multiple sources are statistically similar, more robust estimates can be expected by pooling all data, and this is a regionalization approach. Frequency analysis in conjunction with regionalization is known as regional frequency analysis in hydrological applications.

Regionalization is a term used to seek a homogeneous area in hydrologic frequency analysis. The aim is to maximize information available at gauging stations due to limitations imposed by the finite length and sampling intervals and the augmented sample size from a well-established homogeneous region to enhance the accuracy of frequency analysis. Various methods of regionalization for frequency analysis have been proposed over the decades. Cunnane (1988) and Hosking and Wallis (1997) provided a rigorous review of regionalization. A homogeneous region can usually be defined by hydrologically similar sites, while not necessarily constrained by geographical adjacency. Previously, statistical moments such as mean and variance have been widely used as hydrological properties in clustering approaches. An alternative approach, considering different hydrologic properties such as seasonality (Gingras et al., 1994) and likelihood-ratio statistics (Wiltshire, 1985) has been applied and considered in the partitioning of sites. Many clustering techniques such as K-means, factor and principal component analysis with hydrological properties are commonly used to label meteorological sites as regional identification (White, 1975, Acreman and Sinclair, 1986, Burn, 1988, Burn, 1989, Guttman, 1993, Fragoso and Gomes, 2008, Hsu and Li, 2010, Rao and Srinivas, 2006).

One of the difficulties in various regionalization approaches is the need to determine homogeneity of sites having similar hydrological characteristics. These issues are more challenging in the case of stream-flow data due to the influence of watershed size, channel network and topography. Consequently, scaling schemes of hydrologic variables for regionalization have been proposed and investigated by, among others (Eaton et al., 2002, Moussa, 2008, Richards-Pecou, 2002). Hosking and Wallis (1997) suggested several statistics that can be applied to facilitate the process of defining regions of reasonably homogeneous sites.

As mentioned, existing regionalization approaches have been based on statistical and watershed characteristics in conjunction with clustering techniques. Many, if not most previous studies have focused on properties of magnitude as well as clustering techniques, and temporal aspects (e.g. timing of extreme and seasonality). Hydrologic variables have not been given much attention. However, spatio-temporal characteristics of hydrologic variables are a crucial component in defining hydrologically homogeneous areas. This study aims to develop a systematic method to determine regionalization using temporal characteristics driven by the circular statistics which are suitable for estimating cyclical time. This work will benefit the possible improvement of a pragmatic modeling approach for determining hydrologically homogeneous regions in regional frequency analysis. Therefore, our objective is to extend a previously developed regionalized statistical procedure to define homogeneous regions for rainfall using multiple pieces of hydrologic characteristic information, and to use these to yield “design rainfall” in regional frequency analysis. We first evaluated spatio-temporal patterns of extremes and monthly rainfall series through circular statistics in order to gain insight into how the spatio-temporal pattern can be incorporated into the regionalization procedure. As part of this investigation, we considered the appropriateness of employing temporal characteristics. The study area and the data used are first described briefly. The models and their applications are next discussed in the order of the key research questions indicated above. Conclusions and recommendations are then presented.