Theor Appl Climatol DOI 10.1007/s00704-011-0497-6
ORIGINAL PAPER
Design rainfall depth estimation through two regional frequency analysis methods in Hanjiang River Basin, China
Yue-Ping Xu & Chaofeng Yu & Xujie Zhang & Qingqing Zhang & Xiao Xu
Received: 19 March 2011 /Accepted: 4 August 2011 # Springer-Verlag 2011
Abstract Hydrological predictions in ungauged basins are 1 Introduction of significant importance for water resources management. In hydrological frequency analysis, regional methods are Under enhanced greenhouse conditions, the increases in regarded as useful tools in estimating design rainfall/flood the frequency and magnitude of hydro-meteorological for areas with only little data available. The purpose of this events like storm, flood, typhoon, and droughts are very paper is to investigate the performance of two regional likely (Fowler and Ekström 2009). This consequently methods, namely the Hosking’s approach and the cokriging increased public awareness of extreme events, particularly approach, in hydrological frequency analysis. These two for regions which are very sensitive to climate change. methods are employed to estimate 24-h design rainfall Determination of design storm/flood with a specified depths in Hanjiang River Basin, one of the largest return period therefore becomes important for flood risk tributaries of Yangtze River, China. Validation is made assessment, flood management and infrastructure design through comparing the results to those calculated from the under anticipated climate change. Climate change greatly provincial handbook approach which uses hundreds of increases the uncertainty already existed in design storm/ rainfall gauge stations. Also for validation purpose, five flood obtained by hydrological frequency analysis. In hypothetically ungauged sites from the middle basin are order to reduce the uncertainty in design storm/flood, chosen. The final results show that compared to the several possible solutions are available, including consid- provincial handbook approach, the Hosking’s approach eration of climate change (Fowler et al. 2007)in often overestimated the 24-h design rainfall depths while hydrological frequency analysis, appropriate choice of the cokriging approach most of the time underestimated. probability distributions and parameter estimation meth- Overall, the Hosking’ approach produced more accurate ods (Xu and Booij 2007), incorporating physical mecha- results than the cokriging approach. nism in frequency analysis (Pandey et al. 1998;Xuand Tung 2009), and use of “regional” or “prehistoric” informa- tion (Frances and Salas 1994; Benito and Thorndycraft 2005; Neppel et al. 2010). : : : Under climate change, it is increasingly important to Y.-P. Xu (*) X. Zhang Q. Zhang X. Xu investigate the impact of climate change on the frequency Institute of Hydrology and Water Resources, Civil Engineering, and magnitude of extreme hydro-meteorological events and Zhejiang University, Hangzhou, China associated uncertainties. For example, Jones and Reid e-mail: [email protected] (2001) studied possible changes in extreme precipitation intensities through frequency analysis using regional C. Yu climate model (with greenhouse gas changes following Shanghai Water Conservancy – Engineering Design & Research Institute, the IS92a scenario for 2080 2100) over Britain. Results Shanghai, China indicated dramatic increases in the heaviest precipitation Y.-P. Xu et al. events over Britain. Prudhomme et al. (2003) proposed the approach to analyze the 24- and 2-h design precipitation for implementation of appropriate methods that incorporate Washington State. Yang et al. (2010) investigated the spatial climate change uncertainty in flood risk assessment. Fowler patterns of the flood risks in the Pearl River Delta region and Ekström (2009)used13regionalclimatemodel using Hosking’sapproach. integrations from the Prediction of Regional Scenarios and The use of geostatistical methods is another prevailing Uncertainties for Defining European Climate change risks alternative to estimate design storm/flood for ungauged areas and Effects (PRUDENCE) ensemble together with extreme in recent years. For example, Daviau et al. (2000)used value analysis to assess changes to seasonal precipitation geostatistical methods to obtain map sets of L-mean and L- extremes in nine UK rainfall regions by 2070–2100 under CV for flood timing and magnitude. Nezhad et al. (2010) the SRES A2 emissions scenario. They found out that the employed residual kriging in physiographical space to handle multimodel ensembles project increases across the UK in effectively any possible spatial trends within the hydrological winter, spring, and autumn extreme precipitation; although variables for regional frequency analysis. Castiglioni et al. there is uncertainty in the absolute magnitude of increases, (2010) applied Physiographic-Space Based Interpolation and these range from 5% to 30% depending upon region and Top-Kriging for the regionalization of Q355 (i.e., the stream- season. In general, these researches show that the impact of flow that is equaled or exceeded 355 days in a year, on climate change can be significant in hydrological frequency average) over a broad geographical region in central Italy. analysis. These researches demonstrated the usefulness of geostatis- The second topic of interest in hydrological frequency tical methods in regional frequency analysis. analysis is the use of “regional” or “prehistoric/historic” The purpose of this paper is twofold: (1) to data. The data from a single site for frequency analysis are investigate the spatial patterns of design storm/rainfall usually insufficient to obtain accurate estimates of quan- depth for different return periods for Hanjiang River tiles. Therefore, regional and historical information can be Basin, one of the largest tributaries of Yangtze River, often incorporated into hydrological frequency analyses to China; (2) to compare the performance of two regional reduce the uncertainty in the design storm/flood estimation. frequency analysis methods, i.e., geostatistical method The use of prehistoric or historic data can be found in, e.g., and Hosking’s approach in design rainfall depth estima- Fenske (2003) and Reis and Stedinger (2005). Fenske tion in ungauged areas. The structure of this paper is as (2003) investigated the use of paleoflood data for risk follows. First, study area and data used in this study analysis of US Army Corps of Engineers projects and will be described. Second, L-moment approach used in found out that the usefulness of paleoflood data in flood regional frequency analysis is briefly described. Third, frequency analysis depends on their accuracy and uncer- the two regional frequency analysis methods will be tainty. Reis and Stedinger (2005) have employed Bayesian introduced. Afterwards an exploratory data analysis is methods to incorporate historic flood information in flood made followed by results. Finally discussion and frequency analysis. Similarly, the use of regional informa- conclusion are given. tion is also very useful for hydrological frequency analysis, particularly in ungauged basins (Hosking and Wallis 1997). It may reduce the sampling uncertainty and result in more 2 Study area and data robust estimates by introducing more data as long as homogeneous regions can be established. Frequency anal- The data used in this study is from Hanjiang River, one of ysis using regional information is widely found in the the largest tributaries of Yangtze River, China. This river is literature. Yu and Chen (1998) developed regional rainfall the source of the Middle Route of well-known South-to- intensity-duration-frequency (IDF) relationship by general- North Water Transfer Project. It stretches 1′577 km from izing the parameters of dimensionless frequency curves Shannxi to Hubei Provinces, returning Yangtze River in using regression equations. Yu et al. (2004) developed IDF Wuhan city. The river passes through Gansu, Sichuan, scaling formulas based on simple scaling concept combined Henan, and Chongqin Provinces. The Hanjiang River is with Gumbel distribution and parameters of IDF scaling located at 106°15′∼114°20′ east longitude, 30°10′∼34°20′ formulas were then regionalized based on the relationship north latitude, with its basin area of 159,000 km2.Itis between the scaling exponent and the average of annual 1- 820 km long from northwest to southeast, 320 km at the day maximum rainfall. Norbiato et al. (2007) characterized widest, and 180 km at the narrowest. The basin has a sub- the severity of a flash flood generating storm occurred on tropical monsoon climate and the average annual precipi- August 29, 2003 on the upper Tagliamento river basin using tation is about 700∼1,200 mm. More precipitation can be Hosking’s approach (Hosking and Wallis 1997). Wallis et al. observed in the south than the north, in the west than the (2007) combined the PRISM mapping system with Hosking’s east in the upper basin. In the middle and lower basin, the Design rainfall depth estimation in Hanjiang River Basin, China precipitation increases. It is noted that precipitation at combination of order statistics and known to be robust to mountainous region is much more than that at hillock areas outliers and virtually unbiased for small samples. Details and the plain. about the L-moment approach can be found in Hosking and Daily rainfall data of 20 gauge stations are obtained from Wallis (1997). The L-moment is given by: China Meteorological Administration (CMA) and Bureau of Hydrology (BOH), Yangtze Commission. New time Z1 Xr r k series of annual maximum daily data are then abstracted » ðÞ 1 ðÞr þ k ! l þ ¼ xðFÞP ðFÞdF ¼ b ; ð1Þ from these daily data. Figure 1 shows the locations of r 1 r ðÞ2ðÞ k k¼0 k! r k ! Hanjiang River Basin and 20 rainfall gauge stations, with 0 basic information shown in Table 1. r ¼ 0; 1; 2;
3 L-moment approach Pr r k R1 P»ðFÞ¼ ð 1Þ ðrþkÞ!Fk b ¼ xðFÞFrdF With r 2 and r ðk!Þ ðr kÞ! 3.1 L-moment theory k¼0 0 Where F(x) is a cumulative distribution function and Both regional frequency analysis approaches above are x(F) is the quantile function. based on the L-moment theory. L-moments are linear The first four L-moments are therefore: