The potential damage caused by coastal flooding is increasing due to sea-level rise and coastal development. Flood control planning is of greater importance in coastal areas. In this paper, annual extreme precipitation in the Taihu Basin of China was used to investigate the difference between two commonly used parameter estimation methods, the method of Conventional Moments (CM) and the method of L-moments (LM). Generalized Extreme Value (GEV) distribution was used to describe the extreme rainfall events. The robustness and unbiasedness of the two parameter estimation methods were evaluated. Monte Carlo simulation was applied to compute the confidence bounds quantifying the uncertainty resulting from the use of these two methods.
Rainstorm induced flood is one of the most severe natural disasters in coastal area. The interaction between global climate change and climate variability has increased the occurrence of natural disasters (Bowles et al., 1996; Vrijling, 2001; Apel et al., 2004). The Taihu Basin is located in the Yangtze River Delta plains, near the East China Sea and is often attacked by storms and floods. As one of the richest area of China, economic loss inflicted by flood and heavy rainfall is considerable, which makes flood control planning of great importance in this area. Estimation of precipitations of specific return periods is an important task for offshore structure planning and design.
Statistical frequency analysis is widely used in hydrology for estimating the relationship between the magnitude and the occurrence frequency of various hydrological events (Ashkar and Nwentsa, 2007). Method of Conventional Method (CM) is one of the traditional parameter estimation methods used in statistical frequency analysis. In China, the CM-method together with Person III (P-III) distribution, have been written into the basic handbook of hydrology and applied for decades (MWR, 2006). In practical applications, the goodness-of-fit of theoretical curves are sometimes examined by eyes and adjustments are made by considering more detailed information such as the goodness-of- fit of the upper and lower tails (Chen et al., 2002). Results obtained by different people using this method may vary from 15% to 20% due to its full empirical nature (Jin, 1999; Liu et al., 2010). Recently, more and more researches and applications indicate that the traditional methods are not good enough to effectively guarantee consistent results and safety (Lin et al., 2006; Wang et al., 2008). Other parameter estimation methods like the method of maximum likelihood (Benjamin & Cornel, 1970), the method of probability weighted moments (Greenwood, 1979) and the method of L-moments (Hosking, 1990) have also been widely used. In addition, some other theoretical distribution models, such as the Generalized Extreme Value (GEV) distribution, the generalized logistic (GLO) distribution and Weibull distribution are also widely used. In the UK, the L-moment (LM) fitting of the GLO and GEV distributions is recommended for hydrologic frequency analysis (Hosking and Wallis, 1997; Robson and Reed, 1999; NERC, 1975). In the United States, the Log-Pearson Type III (LP-III) distribution is recommended as the basic distribution for defining the annual flood series, and the CM-method is used to determine the distribution parameters (IACWD, 1982). In the Taihu Basin, Xu and Huang (2011) used long-term data at the Wusong station, fitted GEV distributions using the maximum likelihood method (ML) for estimating the extreme water level near the City of Shanghai in the Yangtze River estuary. Zhou et al. (2014) applied P-III and GEV distributions to estimate the annual maximum precipitation at the Changxi Station in the Taihu Basin. After using ML, LM, and CM as parameter estimation methods, GEV distributions with parameters estimated using the LM-method were found to fit well with the observed data at the Changxi Station.