INTERNATIONAL JOURNAL OF CLIMATOLOGY Int. J. Climatol. 33: 1140–1152 (2013) Published online 27 April 2012 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/joc.3499 Copula-based spatio-temporal patterns of precipitation extremes in China Qiang Zhang,a,b,c * Jianfeng Li,a,b Vijay P. Singhd,e and Chong-Yu Xuf a b Department of Water Resources and Environment, Sun Yat-sen University, Guangzhou, China Key Laboratory of Water Cycle and Water Security in Southern China of Guangdong High Education Institute, Sun Yat-sen University, Guangzhou, China c School of Geography and Planning, and Guangdong Key Laboratory for Urbanization and Geo-simulation, Sun Yat-sen University, Guangzhou, China d Department of Biological and Agricultural Engineering, Texas A&M University, College Station, TX, USA e Department of Civil and Environmental Engineering, Texas A&M University, College Station, TX, USA f Department of Geosciences and Hydrology, University of Oslo, Blindern, Oslo, Norway ABSTRACT: Daily precipitation data from 590 stations in China covering a period of 1960–2005 are analysed using copulas, the modified Mann–Kendall (MK) trend test and linear regression. Changing characteristics of eight precipitation indices are investigated in both time and space. Results indicate that (1) the regions west of 100 ° E, particularly northwest China, exhibit a wetting tendency as reflected by increasing/decreasing number of consecutive rain/non-rain days; (2) the drying tendency is observed mainly in the regions covered by the Yellow River basin, the Huaihe River basin, and the Haihe River basin, and relatively moderate changes in precipitation indices are found in northeast China; (3) precipitation extremes are intensifying in the regions east of 100 ° E, particularly in case of south China, specifically the lower Yangtze River basin, the southeast rivers and the Pearl River basin. The intensification of precipitation extremes in south China is mirrored mainly by the decreasing number of rain days and increasing number of consecutive non-rain days. Besides an increasing percentage of P90 to the annual total precipitation, (4) the intensification of precipitation extremes has the potential to increase the probability of occurrence of natural hazards, particularly floods and droughts. The spatial distribution of floodand drought-affected crop areas is in agreement with that of precipitation extremes, showing considerable impacts of precipitation extremes on meteor-hydrological hazards. An increasing number of consecutive non-rain days in south China will cause a higher risk of droughts. The regions east of 100 ° E are heavily populated and are economically developed. Food security, water security, and sustainable socioeconomy in China urgently call for effective water resource management policy. Copyright 2012 Royal Meteorological Society KEY WORDS precipitation extremes; copula functions; climate changes; spatio-temporal distribution; natural hazards; agricultural responses Received 2 September 2011; Revised 15 March 2012; Accepted 26 March 2012 1. Introduction Climate change, characterized by global warming, is now believed to be intensifying the hydrological cycle (Allen and Ingram, 2002; World Meteorological Organization, 2003; Ziegler et al., 2003), altering the spatio-temporal patterns of precipitation and increasing occurrences of precipitation extremes, such as floods and droughts (Easterling et al., 2000; Vörösmarty et al., 2010). Arnell (1999) indicated that the hydrological cycle would intensify with more evaporation and more precipitation and the extra precipitation would be unequally distributed around the globe and hence would cause more frequent floods and droughts. Changing properties of precipitation and hydrological extremes have been investigated in China and other ∗ Correspondence to: Q. Zhang, PhD, Department of Water Resources and Environment, Sun Yat-sen University, Guangzhou 510275, China. E-mail: zhangq68@mail.sysu.edu.cn Copyright 2012 Royal Meteorological Society places of the world. Groisman et al. (1999) indicated that the probability of daily precipitation exceeding 50.8 mm in mid-latitude countries increased by ∼20% in the later twentieth century. Suppiah and Hennessy (1998) showed that heavy precipitation events in most parts of Australia have increased. Zhai et al. (1999) indicated that extreme precipitation events have increased in western China since 1950. Increasing extreme precipitation events can also be found in south-eastern China (e.g. Zhang et al., 2009a). Zolina et al. (2010) found the lengthening of wet periods in Europe, indicating that heavy precipitation events during the past two decades have become much more frequently associated with longer wet spells and intensified in comparison with the 1950s and the 1960s. Analysis of precipitation extremes in the Pearl River basin, south China, also indicated increased precipitation variability and high-intensity rainfall (Zhang et al., 2009a). Besides, Zhang et al. (2009a) also found that the COPULA-BASED SPATIO-TEMPORAL PATTERNS OF PRECIPITATION EXTREMES amount of rainfall has changed little, but its variability has increased. The probabilistic behaviour of hydro-meteorological variables has been receiving increasing attention in recent years, perhaps due to increasing risk of natural hazards under the influence of climate change and human activities. Leonard et al. (2008) considered maximum values that occur within a given season and the relationship between seasonal maxima and annual maxima. Hashino (1985) generalized the Freund bivariate exponential distribution (Freund, 1961) to represent the joint probability distribution of rainfall intensity and maximum storm surge in Osaka Bay, Japan. Using copulas, De Michele et al. (2005) and Zhang and Singh (2006, 2007a) derived bivariate distributions for modelling flood peak and volume; Singh and Zhang (2007) and Zhang and Singh (2007b) determined joint rainfall frequencies; and Kao and Govindaraju (2010) and Song and Singh (2010a, 2010b) characterized droughts. The flexibility of copulas for constructing joint distributions is evident from related studies on rainfall frequency analysis (Zhang and Singh, 2006, 2007a, 2007b; Kao and Govindaraju, 2008). Therefore, copulas are becoming important tools in the bivariate analysis of precipitation or streamflow extremes (e.g. Leonard et al., 2008). Copulas are a flexible representation of multivariate distributions, as each marginal distribution can have a different distributional form and there is a wide range of copulas to select from to be able to yield the correlated joint distribution (Leonard et al., 2008). China, the most populous country, is heavily dependent on agriculture for food production to support the growing population. However, agricultural development is prone to be affected by precipitation extremes and hence droughts and floods. Understanding of the probabilistic behaviour of precipitation extremes, particularly the joint probability of precipitation extremes, is the first step towards the scientific management of agricultural activities and the mitigation of the influence of floods and droughts on agriculture. Precipitation extremes have been analysed in both time and space (e.g. Zhang et al., 2011) However, no such reports are available so far concerning changing probabilistic characteristics of precipitation extremes in China. This constituted the motivation for this study. The objectives of this study therefore are (1) to analyse the joint probabilistic behaviour of precipitation extremes using copulas and (2) to discuss the implications of the probabilistic behaviour for occurrences of floods and droughts and related impacts on agriculture in China. 1141 two days were filled in by average precipitation values of neighbouring days. If consecutive days had missing data, the missing values were replaced with long-term averages of the same days. It was assumed that this gap filling method would have no influence on the longterm temporal trend. Furthermore, the data consistency was checked by the double-mass method and the result showed that all the data series used in the study were consistent (Zhang et al., 2009b, 2011). Eight extreme precipitation indices were defined and are displayed in Table I, which are similar to the precipitation indices used in other studies (e.g. Tebaldi et al., 2006; Fatichi and Caporali, 2009; Zhang et al., 2010). The NW and CDD are the number of rainy days, and the maximum number of consecutive dry days; the D90, P90, and I90 are the number of days, total precipitation, and precipitation intensity of the extreme heavy precipitation events; and the D10, P10, and I10 are the number of days, total precipitation, and precipitation intensity of the extreme weak precipitation events. 3. 3.1. Methodology Marginal probability distributions Extreme precipitation indices can be categorized into two types: discrete and continuous. NW, D90, CDD, and D90 are discrete indices, whereas P90, I90, P10, and I10 are continuous indices. There are several discrete probability distributions, e.g. uniform distribution, Poisson distribution, geometric distribution, binomial distribution, and negative binomial distribution (Spiegel et al., 2000). Wilks (1999) employed geometric, negative binomial, and mixed geometric distributions to represent the number of wet days in America. On the basis of the virtual attributes of precipitation indices, binomial, Poisson, geometric, and negative binomial distributions were used 2. Data Daily precipitation data from 590 rain gauge stations covering a period of 1960–2005 were obtained from the National Climate Center of China Meteorological Administration. The spatial distribution of rain stations is shown in Figure 1. Missing data of one day or Copyright 2012 Royal Meteorological Society Figure 1. Locations of the meteorological stations and river basins. The black triangles denote meteorological stations. Numbers denote river basins, i.e. 1, the Songhuajiang River; 2, the Liaohe River; 3, the Haihe River; 4, the Yellow River; 5, the Huaihe River; 6, the Yangtze River; 7, the southeast rivers; 8, the Pearl River; 9, the southwest rivers; and 10, the northwest rivers. Int. J. Climatol. 33: 1140–1152 (2013) 1142 Q. ZHANG et al. Table I. Definitions of precipitation extreme indices. Indices NW D90 P90 I90 CDD D10 P10 I10 Definitions Unit Number of wet days in a year. The wet day is defined as the day with precipitation ≥1 mm Annual number of days with daily precipitation exceeding the 90th percentile threshold. The 90th percentile threshold is calculated based on the precipitation series of all wet days at each station Total precipitation of D90 in a year Precipitation intensity of D90 in a year Maximum number of consecutive dry days of a year. The dry day is defined as the day with precipitation <1 mm Annual number of days with precipitation less than the 10th percentile threshold. The 10th percentile threshold is calculated based on the precipitation of all wet days at each station Total precipitation of D10 of a year Precipitation intensity of D10 of a year d d to analyse the marginal probability properties of NW, D90, CDD, and D10. Moreover, generalized extreme value (GEV), generalized Pareto (GP), Pearson III type, lognormal, Wakeby, and exponential distributions were analysed and the right distribution was selected for based on the goodness-of-fit test. The maximum likelihood estimate method was used to estimate parameters of the discrete probability distribution, while the L-moments method (Hosking, 1990) was applied to calculate parameters of the continuous probability distribution. The goodness-of-fit of the probability distributions was evaluated using the Kolmogorov–Smirnov (K-S) test (Frank and Masse, 1951) at >95% confidence level. After selecting the probability distribution with high goodness-of-fit, the specific precipitation indices corresponding to different return periods, denoted as z values, were computed. Let t denote return period, the z value related to t can be obtained as follows: z = F −1 (1 − 1/t) for NW, D90, P90, I90, CDD, and D10 and z = F −1 (1/t) for P10 and I10 (1) (2) where F −1 is the inverse cumulative distribution function. For NW, D90, P90, I90, CDD, and D10, values ≥z were considered; for P10 and I10, values ≤z were considered. 3.2. Copulas are a flexible representation of multivariate distributions. Assume X and Y are two random variables, with the marginal distributions as F (x) = P [X ≤ x] and G(y) = P [Y ≤ y]. The joint distribution, H (x, y) = P [X ≤ x, Y ≤ y], can be formulated using the copula function C as follows: H (x, y) = C[F (x), G(y)] (3) The Copula function, C, is the bivariate joint distribution of X and Y . Copulas can be used for multivariate joint distributions (Nelsen, 2006). Copyright 2012 Royal Meteorological Society d mm mm d−1 {(xk , yk )}nk=1 denotes the observed X and Y series the size of which is n. Then, the empirical Copula, Ce , can be expressed as follows: Ce i j , n n = ni,j n (4) where 1 ≤ i, j ≤ n, x(i) denotes the rank i of the x series arranged in ascending order; y(j ) denotes the rank j of the y series arranged in the ascending order, and ni,j denotes the number of observed values that xk ≤ x(i) , yk ≤ y(j ) . The Gumbel–Hougaard copula, Clayton copula, Frank copula, Gauss copula, and t copula are employed in this study. The Gauss copula and t copula are families of the elliptical copula family, and the coefficient of linear correlation ρ is the parameter. The other three copulas belong to the Archimedean copula family. Each Archimedean copula was constructed using a different generating unit ø with parameter θ. 3.3. Identification of joint probability distribution The identification of a joint probability distribution follows the procedure proposed by Genest and Rivest (1993): (1) The Kendall correlation coefficient τ is given by τ= Copula functions mm mm d−1 d sign[(xi − xj )(yi − yj )] i<j n(n − 1) 2 (5) where n denotes size of the sample, i, j = 1, 2, . . . , n; if xi ≤ xj and yi ≤ yj , then sign() = 1, else sign() = −1. (2) For the Gumbel–Hougaard, Clayton, and Frank copulas, the generating unit parameter θ was computed by the relation of θ and τ . For the Gauss and t copulas, parameter ρ was also calculated by τ . (3) Construct the copula function with θ or ρ. (4) Employ a statistical method, such as the Akaike information criterion (AIC; Akaike, 1974), to select the most appropriate copula family. Int. J. Climatol. 33: 1140–1152 (2013) COPULA-BASED SPATIO-TEMPORAL PATTERNS OF PRECIPITATION EXTREMES 3.4. Akaike information criterion The AIC (Akaike, 1974) was used to evaluate the goodness-of-fit of the copulas. Two segments were considered in AIC: bias of fit and the unreliability from the number of the model parameters. AIC is expressed as follows: AIC = n log(RSS/n) + 2m (6) where n is the sample size, m is the number of parameters, and RSS is the residual sum of squares. The probability distribution function with the minimum AIC value is the right choice. 3.5. Bivariate joint return period For precipitation variables X and Y , let Fx (x) be the marginal distribution of X, Fy (y) be the marginal distribution of Y , and F (x, y) be the joint distribution function of these two variables. Two types of the bivariate joint return periods (Salvadori and Michele, 2004) were used in this study: T{X>x,Y >y} = 1 P (X > x, Y > y) 1 (7) = 1 − Fx (x) − Fy (y) + F (x, y) T{X>x,Y ≤y} = 1 P (X > x, Y ≤ y) 1 = Fy (y) − F (x, y) (8) where T{X>x,Y >y} denotes the joint return period with both X and Y exceeding the specific threshold; T{X>x,Y ≤y} denotes the joint return period with X exceeding the specific thresholds and Y less than and equal to the specific threshold; T{NW, CDD:X>x,Y >y} denotes the joint return period of the event that NW and CDD exceed their specific thresholds simultaneously; T{P90, P10:X>x,Y ≤y} denotes the joint return period of the event that P90 exceeds its specific threshold, and P10 is less than and equal to its specific threshold. The bivariate joint return periods of extreme precipitation indices and related definitions are listed in Table II. Table II. Joint return periods of extreme precipitation and their meaning. Return periods T{NW, CDD:X>x,Y >y} T{D90, D10:X>x,Y >y} T{P90, I90:X>x,Y >y} T{P90, P10:X>x,Y ≤y} T{I90, I10:X>x,Y ≤y} T{D10, The trend was detected by the MK test (Mann, 1945; Kendall, 1955) and linear regression. The MK trend test is a rank-based nonparametric method that does not require any assumption about a distribution. It should be noted that the persistence within a hydro-meteorological series can heavily influence the MK test results (von Storch, 1995; Hamed and Rao, 1998). Hamed and Rao (1998) proposed a modified MK test based on the equivalent sample size to eliminate the effect of persistence. In their modification, the modified variance of the MK statistic was proposed to replace the original one if the lag-i autocorrelation coefficients were significantly different from zero. von Storch (1995) Copyright 2012 Royal Meteorological Society P10:X>x,Y ≤y} Definition Joint return period of the event that numbers of wet days and dry days in a year exceed their specific thresholds Joint return period that D90 and D10 events occur simultaneously Joint return period that P90 and I90 events occur simultaneously Joint return period that P90 and P10 events occur simultaneously Joint return period that I90 and I10 events occur simultaneously Joint return period that D10 and P10 events occur simultaneously suggested pre-whitening to eliminate the lag-1 autocorrelation before the use of the MK test. In pre-whitening, if the lag-1 autocorrelation coefficient, c, was larger than 0.1, then the analysed time series (x1 , x2 , . . . , xn ) should be replaced by (x2 − cx1 , x3 − cx2 , . . . , xn+1 − cxn ). However, pre-whitening has the potential to underestimate the trend in a time series (Yue et al., 2002). Moreover, significant lag-1 autocorrelation is still detected even after pre-whitening, because only lag-1 autocorrelation is considered in pre-whitening. The modified MK considers lag-i autocorrelation and its robustness has been demonstrated by Hamed and Rao (1998) and was used successfully in a meteor-hydrological study by Daufresne et al. (2009). The modified MK test was used in this study to analyse trends within the time series. Besides, a linear regression was also employed to estimate the linear trend in the series. The student t-test was then employed to check the significance of the estimated linear trend. The significance of trend was tested at the >95% confidence level. 4. 4.1. 3.6. Trend detection 1143 Results Case study To show the computation procedure, we used data from the Huiyang station in southeast China as a case study. Parameters of probability functions were estimated for eight extreme precipitation indices. Then, the K-S test was used to evaluate the reliability of the probability distribution function and the probability function with the highest goodness-of-fit was accepted as the best choice. The probability function with the largest p value by the K-S test was the most appropriate choice. The p values by the K-S test for the discrete precipitation indices are provided in Table III, and the p values for the consecutive precipitation indices are provided in Table IV. It should be noted that due to the statistical characteristics, it is possible that parameters of a certain probability Int. J. Climatol. 33: 1140–1152 (2013) 1144 Q. ZHANG et al. Table III. p values of the K-S test for discrete precipitation indices. Distributions Negative binomial Poisson Geometric Binomial NW D90 CDD D10 0.7753a N/A 0.8719a N/A 0.2461 0.0471 N/A 0.4949 N/A N/A N/A N/A N/A 0.0544a N/A 0.5315a N/A = the distribution parameters cannot be estimated for the index, or that distribution fails to pass the K-S test. a The distribution passes the K-S test. Table IV. p values of the K-S test for consecutive precipitation indices. GEV P90 I90 P10 I10 0.9820 0.9591 0.6950 0.7044 GP N/A N/A N/A N/A Pearson Lognormal Wakeby ExpIII onential 0.9819 0.9793 0.7174 0.7184 0.9842 0.9813 0.6889 0.7007 0.9995a 0.9871a 0.8818a 0.7398 N/A N/A N/A N/A N/A = the distribution parameters cannot be estimated for the index or the distribution fails to pass the K-S test. a The distribution passes the K-S test. distribution cannot be estimated. For example, parameters of the negative binomial distribution can be estimated for a time series, it is unnecessary that parameters of the binomial distribution are obtained for the same series. Besides, even though the parameters of a distribution are available, the distribution failing to pass the K-S test will still be ignored. The goodness-of-fit of the probability density functions of GEV, Pearson III, Wakeby, and exponential distributions for P90 were evaluated (Figure 2). It can be seen from Table IV that the exponential distribution failed to pass the K-S test. The p values of GEV and Pearson III were similar. However, it can be observed from Figure 2 that these two functions fitted very similarly. The p value for the Wakeby distribution was the largest, showing a good goodness-of-fit. The 10 year return period for P90 was estimated with parameters obtained as above based on Equations (1) and (2) (Table V). Different return periods for the extreme precipitation indices for the rainfall stations considered in this study were estimated following this computation procedure. To illustrate the computation of joint return periods of two extreme precipitation indices, an example computing the joint return period for P90 and I90 was considered. Figure 2. Probability density functions for P90 at the Huiyang station as a case study. Selection of the copula function was done following the procedure proposed by Genest and Rivest (1993). Parameters of copulas were calculated by τ . AIC was used to evaluate the goodness-of-fit of the copulas. Parameters and AIC values of the copulas are shown in Table VI. The t copula with the minimum AIC value was the most appropriate copula for P90 and I90 at the Huiyuan station. For the marginal distribution of P90 and I90, the joint cumulative distribution function is shown in Figure 3. The joint return period that P90 exceeding the P90 with 10 year return period and I90 exceeding the P90 with 10 year return period occurred simultaneously, i.e. T{P90, I90:X>x,Y >y} = 31.25 from Equation (7). The joint return periods of other precipitation extremes were estimated by the same computation procedure as introduced above. 4.2. Spatial distribution of precipitation indices with 10 year return period Increasing NW was identified in the direction from northwest to southeast China (Figure 4(a)). The maximum NW was observed in the Pearl River basin and southern parts of the Yangtze River basin. Thus, southeast China is characterized by abundant rainy days and northwest China by Table VI. Parameters and AIC values for the copulas of P90 and I90 at the Huiyang station.a . Gumbel Clayton Frank Gauss t Parameter 1.4744 0.9487 3.1681 0.4841 0.4841 AIC −15 954 −16 046 −16 036 −16 190 −16 199b a Parameters, θ, for the Gumbel, Clayton, and Frank copulas; parameters, ρ, for the Gauss and t copula functions. b The copula functions with best fits. Table V. Precipitation values corresponding to 10 year return periods at the Huiyang station. Precipitation values (mm) NW D90 P90 I90 CDD D10 P10 I10 123 15 1084.32 90.53 54 289 21.14 0.08 Copyright 2012 Royal Meteorological Society Int. J. Climatol. 33: 1140–1152 (2013) COPULA-BASED SPATIO-TEMPORAL PATTERNS OF PRECIPITATION EXTREMES Figure 3. Joint CDF of the P90 and I90 at the Huiyang station. scarce rainy days. Smaller changes can be observed for CDD in space (Figure 4(b)) when compared with that of NW (Figure 4(a)). CDD reaches its peak values in northwest China and the trough values in southeast China. The northwest rivers, the southwest rivers, the upper Yellow River, and the upper Yangtze River are dominated by CDD being longer than 100 d and other regions by CDD being below 50 d. It can be observed from Figure 4(c) that D90 is increasing from northwest to southeast China, indicating higher occurrence probability of heavy rains in southeast 1145 China in comparison to northwest China. The opposite spatial patterns of D10 (Figure 4(d)) can be found in comparison with D90 (Figure 4(c)). Therefore, larger occurrence probability of low precipitation is detected in northwest China when compared with that in southeast China (Figure 4(c) and (d)). In general, Figure 4(a)–(d) imply that southeast China suffers the most extreme strong precipitation and northwest China endures the longest extreme weak precipitation and consecutive dry days. This result illustrates the occurrence probabilities or natural risks of floods and droughts in both space and time across the entire country. Figure 4(e) shows the largest P90 in the Pearl River basin and the lower Yangtze River basin. Northwest China is still characterized by smaller P90. Relatively complicated spatial patterns of P10 can be found in Figure 4(f). Generally, two areas are dominated by larger P10 values, i.e. northeast China and the areas covered by the upper Yangtze River basin, the Pearl River basin, and the east parts of the southwest rivers. The lower P10 is found in the Yellow River basin, the northwest rivers, the Haihe River, the Huaihe River, and the lower Yangtze River. I90 represents the extreme strong precipitation intensity. Figure 4(g) indicates that I90 increases from northwest to southeast China. The maximum I90 value is observed in the Pearl River basin. Besides, I90 larger than Figure 4. Spatial distribution of precipitation indices corresponding to ten year return periods across China: (a) NW, (b) CDD, (c) D90, (d) D10, (e) P90, (f) P10, (g) I90, and (h) I10. Copyright 2012 Royal Meteorological Society Int. J. Climatol. 33: 1140–1152 (2013) 1146 Q. ZHANG et al. Figure 5. Spatial distribution of the modified Mann–Kendall trends for precipitation indices across China: (a) NW, (b) CDD, (c) D90, (d) D10, (e) P90/ATW, (f) P10/ATW, (g) I90, and (h) I10. ATW, annual total precipitation. 60 mm d−1 was observed mainly in the lower Yangtze River basin and the Pearl River basin, showing increasing precipitation extremes in these regions, which is in agreement with the result by Zhang et al. (2011). I10, the precipitation index showing extreme weak precipitation intensity, is in a relative complex spatial pattern. Higher I10 was found in the southern parts of the southwest rivers, the lower Pearl River, and northeast China and lower I10 was observed in other regions (Figure 4(h)). Figure 4(e)–(h) indicates higher probability of heavy precipitation events in southeast China and weak precipitation in northwest China. 4.3. Trends in precipitation indices Figure 5 displays the spatial patterns of trends of precipitation indices. It can be seen that northwest China and the southwest China are characterized by increasing NW, showing wetting tendency in these regions, and the increasing rate is 0.1–0.2 d year−1 (Figure 6(a)). Significant decreasing NW can be observed in the upper Pearl River basin, the upper Yangtze River basin, the Yellow River basin, the Huaihe River basin, and parts of northeast China, implying drying tendency in these areas. The changing rate of NW is −0.1 to −0.2 d year−1 Copyright 2012 Royal Meteorological Society (Figure 6(a)). NW changes in other regions are not statistically significant except some regions are distributed sporadically with significant NW changes. The changing magnitudes of NW in these regions are nearly 0 d year−1 . Significant decreasing CDD can be found in parts of the Tibet Plateau and the northwest China and the changing rate can reach −0.2 to −0.6 d year−1 (Figure 6(b)). Increasing CDD with the changing rate of 0–0.2 d year−1 (Figure 6(b)) is detected in the Haihe River, the Yellow River, the Huaihe River, and north parts of the Yangtze River (Figure 6(b)). Therefore, heavy drying tendency can be expected in these regions due to significant decreasing NW and increasing CDD. Northwest China and the Tibet Plateau tend to be subjected to a wetting tendency (Zhang et al., 2011). Figure 6(c) and (d) show the spatial distribution of D90 and D10. Similar but opposite spatial distribution properties can be found from Figure 6(c) and (d) for D90 and D10. Significant increasing (decreasing) D90 (D10) is detected in the regions west of 100 ° E; the upper Pearl River basin, the upper Yangtze River basin, the Yellow River basin, and the Huaihe River basin and parts of northeast China are characterized by significant decreasing (increasing) D90 (D10). The increasing rate of D90 in northwest China and Tibet Plateau is Int. J. Climatol. 33: 1140–1152 (2013) COPULA-BASED SPATIO-TEMPORAL PATTERNS OF PRECIPITATION EXTREMES 1147 Figure 6. Spatial distribution of linear coefficients of precipitation indices across China: (a) NW; (b) CDD; (c) D90; (d) D10, d year−1 ; (e) P90/ATW; (f) P10/ATW; (g) I90; and (h) I10. Unit for (a)–(d) is d year−1 ; unit for (e)–(f) is 10−2 % year−1 ; unit for (g) is mm year−1 ; and unit for (h) is 10−2 mm year−1 . 0.04–0.08 d year−1 and the decreasing rate of D90 is 0 to −0.04 d year−1 (Figure 6(c)). The increasing and decreasing rate of D10 is 0.1–0.2 and about −0.2 d year−1 , respectively. It can be seen from Figure 6(c) and (d) that the changing magnitude of D90 is smaller than that of D10. Most regions of China are dominated by nearly 0 d year−1 of D10 in the changing magnitude. A similar phenomenon was found in the spatial distribution properties of P90/ATW and P10/ATW. P90/ATW denotes the percentage of P90 to the annual total precipitation. Specifically, P90/ATW represents the percentage of the precipitation of extreme heavy precipitation events to the annual total precipitation. Similarly, P10/ATW denotes the percentage of P10 (the precipitation of extreme weak precipitation events) to the annual total precipitation. It can be seen from Figure 6(e) that the majority of regions in China are dominated by increasing P90/ATW. Significant P90/ATW can be detected in the lower Pearl River basin, the middle and the lower Yangtze River basin, and the southern parts of the Huaihe River basin. Other regions with significant increasing P90/ATW are distributed sporadically across China without obvious spatial patterns. Figure 6(f) shows that the regions characterized by significant increasing P90/ATW are roughly Copyright 2012 Royal Meteorological Society featured by significant decreasing P10/ATW, showing the precipitation changes in these regions shifting to heavy precipitation and to the extreme side. Figure 6(e) and (f) shows distinctly different changing properties of the changing magnitude of P90/ATW and P10/ATW when compared with those of MK trends (Figure 6(e) and (f)). Larger changing magnitude of P90/ATW and P10/ATW can be found in northwest China but not in the regions covered by significant MK trends. It should be due to the higher nonlinear precipitation changes and larger changing variability in the east and southeast China when compared with northwest China. The changing magnitude of P90/ATW and P10/ATW in east and southeast China is nearly zero. However, a large increase (decrease) in P90/ATW (P10/ATW) can be found in northwest China and parts of the Tibet Plateau regions. The absolute changing rate is about 0.2–0.8 × 10−2 % year−1 . Figure 5(g) shows that a majority of regions in east China, southeast China, southwest China, and northwest China are characterized by increasing I90. Significant I90 is observed mainly in the lower Yangtze River basin, the Pearl River basin, the southwest rivers, and the north parts of northwest China. As for changes in I10 (Figure 5(h)), significant I10 was detected mainly in Int. J. Climatol. 33: 1140–1152 (2013) 1148 Q. ZHANG et al. Figure 7. Spatial distribution of joint return period of different precipitation indices across China: (a) T{NW, CDD:X>x,Y >y} ; (b) T{D90, (c) T{P90, I90:X>x,Y >y} ; (d) T{P90, P10:X>x,Y >y} ; (e) T{I90, I10:X>x,Y ≤y} ; and (f) T{D10, P10:X>x,Y ≤y} . the regions east of 100 ° E. Parts of the region west of 100 ° E are characterized by significant increasing I10. Figure 6(g) also indicates that most of the regions in China are dominated by increasing trends of I90. The changing magnitude is about 0–0.02 mm year−1 . The largest increase is detected in the lower Pearl River basin, the lower Yangtze River, and the Huaihe River basin, which can reach 0.04 mm year−1 . The spatial distribution of linear trends of I10 (Figure 6(h)) is similar to that of the MK trends (Figure 5(h)). Decreasing trends in I10 can be found mainly in southeast, southern, and southwest China. The decrease is −0.02 to −0.08 × 10−2 mm year−1 . The largest decrease is detected in southeastern rivers, being −0.08 × 10−2 mm year−1 . 4.4. Copula-based joint return periods The spatial patterns of joint return periods of NW versus CDD (or NW-CDD thereafter) are the same for other couples of precipitation indices, D90 versus D10, P90 versus I90, P90 versus P10, I90 versus I10, and D10 versus P10, as illustrated in Figure 7. Figure 7(a) shows the distribution of joint return periods of NWCDD, T{NW, CDD:X>x,Y >y} . Larger joint return periods imply smaller probability that NW and CDD occur simultaneously and vice versa. Figure 7(a) shows relatively smaller probability that NW and CDD occurs simultaneously in one year in the Pearl River basin, the southeast rivers, the north parts of northeast China, the upper Yangtze River basin and the upper and the lower Yellow River basin, and also in the Huaihe River basin, implying larger risk of natural hazards, floods, or droughts in these regions. In other words, precipitation changes in these regions tend to shift to the extreme sides, i.e. extreme heavy or low precipitation. The spatial distribution of T{D90, D10:X>x,Y >y} is illustrated in Figure 7(b). Copyright 2012 Royal Meteorological Society D10:X>x,Y >y} ; Comparison between Figure 7(a) and (b) shows similar spatial distribution properties of T{NW, CDD:X>x,Y >y} and T{D90, D10:X>x,Y >y} . Relatively higher probability of D90 and D10 can be detected in the middle and the lower Pearl River basin, the upper and the lower Yangtze River basin, the southwest rivers, and parts of northeast China. Our previous investigations also found increasing precipitation extremes or precipitation maxima in the lower Yangtze River basin and the Pearl River basin (Zhang et al., 2008, 2009a, 2009b). Other regions are dominated by relatively longer joint return periods. The joint return periods of >1000 years imply that it is almost impossible that D90 and D10 occur simultaneously in one year. Thus, in this sense, the increasing precipitation extremes occur mainly in eastern, southeastern, southwestern, and northeast China. The spatial distribution of T{P90, I90:X>x,Y >y} is shown in Figure 7(c), showing joint probability that the frequency of extreme precipitation indices with both heavy precipitation and high precipitation intensity. It can be observed from Figure 7(c) that the joint probability of P90 and I90 is relatively even in space across China. The highest joint probability can be observed in the south parts of northwest China. Besides, relatively higher probability can be found in northeast China, the lower Yangtze River basin, and the upper Pearl River basin. Figure 7(d) shows the spatial distribution of T{P90, P10:X>x,Y ≤y} aiming to investigate the joint probability that extreme heavy and weak precipitation events occur simultaneously in the same year. Relatively higher joint probability (longer joint return periods) can be found in the Pearl River basin, the lower Yangtze River basin, the Huaihe River basin, and parts of northeast China and southwest river. On the basis of the meaning of the Int. J. Climatol. 33: 1140–1152 (2013) COPULA-BASED SPATIO-TEMPORAL PATTERNS OF PRECIPITATION EXTREMES T{P90, P10:X>x,Y ≤y} , the above-mentioned regions may be subjected to a higher risk of heavy precipitation events. Relatively lower joint probability is found in northwest China. Roughly similar spatial distribution properties can be found for T{I90, I10:X>x,Y ≤y} (Figure 7(e)) when compared with those for T{P90, P10:X>x,Y ≤y} (Figure 7(d)). Higher joint probability is found along the coastal regions of east China and also in the upper Yangtze River, the Yellow River basin, the Huaihe River basin, and the Haihe River basin, implying higher risk of heavy precipitation larger than the 10 year return period precipitation value. The spatial distribution of T{D10, P10:X>x,Y ≤y} is demonstrated in Figure 7(f), showing longer dry duration and at the same time the low precipitation volume. It can be seen from Figure 7(f) that relatively higher joint probability of D10 and P10 can be found in the regions west of 100 ° E, the Yellow River basin, the Huaihe, and the Haihe River basins, implying drying tendency of these regions reflected by longer no-rain days and correspondingly less total precipitation. 5. Discussion The spatial distribution of precipitation extremes exerts a tremendous influence on the occurrence of floods and droughts in both time and space. China is an 1149 agricultural country and food requirements are heavily dependent on agricultural development and production. In this sense, analysis of precipitation extremes in both time and space is of great scientific and practical value in the socioeconomic development of China. Many studies have addressed the changing characteristics of precipitation extremes (e.g. Zhai et al., 1999; Zhang et al., 2008, 2011). However, no such reports are available in China concerning probabilistic characteristics of precipitation extremes based on copula functions. Besides, we link the spatial distribution of precipitation extremes to the influence of floods and droughts on agriculture. Discussions on the implications of precipitation extremes for agriculture are also presented in this study. Figure 8 illustrates the province-based statistics of population, crop areas, and flood- and drought-affected crop area. The bars inside provinces are the MK trend results reflected by the province-based statistics. The significance of trend was detected at the >95% confidence level. Related time intervals of the data can be referred to the complimentary materials in Table VII. The first glance at Figure 8 shows that the population of China is increasing, particularly in Beijing, Guangdong, and Jiangsu. Increasing trends can also be observed in the crop area in most provinces of China, except Jiangsu, Shannxi, Guangdong, Shanxi, Tijin City, Beijing City, Figure 8. Province-based changes in population growth, crop area, and flood- and drought-affected crop areas over China. Copyright 2012 Royal Meteorological Society Int. J. Climatol. 33: 1140–1152 (2013) 1150 Q. ZHANG et al. Table VII. Length of time series of population growth, crop fields, flood-affected crop fields, and drought-affected fields in the provinces of China. Provinces Population Crop area Flood-affected crop area Drought-affected crop area Beijing Tianjin Hebei Shanxi Inner Mongolia Liaoning Jilin Heilongjiang Jiangsu Zhejiang Anhui Fujian Jiangxi Shandong Henan Hubei Hunan Guangdong Guangxi Hainan Chongqin Sichuan Guizhou Yunnan Tibet Shannxi Gansu Qinghai Nixia Xinjiang 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1960–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 1978–2000 / 1960–2000 1960–1998 / / / 1960–1985 1960–1998 1960–1987 1960–2000 / / 1960–2000 1960–2000 / / 1960–2000 1960–1994 1960–2000 / / / 1960–2000 / / 1960–2000 1960–2000 1960–1987 / / 1972–2000 1960–2000 / / 1960–1994 / / 1960–1985 1960–1987 1960–2000 / 1960–1990 / 1960–2000 / 1960–1999 1960–2000 1960–1997 1960–1993 / / 1960–2000 1960–2000 / / 1963–1990 1960–2000 1960–1987 / 1960–1985 / = no records are available. and Liaoning. However, the focus of this study is on the flood- and drought-affected crop areas. It can be shown in Figure 8 that crops are easily affected by droughts in the provinces of north China, such as Shandong, Tianjin City, Heilongjiang, Shannxi, and Hubei. However, floodaffected crop areas are observed mainly in the provinces of the south China, such as Guangdong, Guangxi, Hunan, Guizhou, and Zhejiang. The drought-affected crop areas in northwest China are increasing but not statistically significant. The spatial distribution of flood- and drought-affected crop areas is in agreement with that of precipitation extremes. Precipitation is relatively abundant in the south China, which is reflected by the spatial patterns of precipitation indices of the 10 year return periods in Figure 4 such as NW, P90, and P10. In general, the Yangtze River and Pearl River basins are characterized by relatively more precipitation than other river basins, particularly the basins in northwest China. Northwest China is having a wetting tendency, which is mirrored by increasing wet days and decreasing consecutive dry days. Copyright 2012 Royal Meteorological Society In southeast China, the number of consecutive dry days is increasing and precipitation intensity defined by the mean daily precipitation of rainy days with daily precipitation exceeding the 90th percentile is also increasing, showing a higher risk of heavy rains. Increasing CDD in southeast China shows a higher risk of the occurrence of droughts. In this sense, southeast China will be subjected to increasing hazard risks of both droughts and floods. The lower Pearl River basin is economically developed with many mega-cities. The East River, a tributary of the Pearl River, is the source of water supply for Shenzhen and Hong Kong, where about 80% of Hong Kong’s annual water demands are met by the East River. Therefore, the precipitation changes shifting to the extreme sides in the Pearl River basin mean much for the water resource management and sustainable development of the regional social-economy. The regions covered by the Yellow River basin, the Huaihe River basin, and the Haihe River basin are characterized by a drying tendency as reflected by the increasing number of consecutive dry days, i.e. CDD and decreasing wet days, i.e. NW. D10 is increasing but P10/ATW is decreasing. The increasing rate of D10 Int. J. Climatol. 33: 1140–1152 (2013) COPULA-BASED SPATIO-TEMPORAL PATTERNS OF PRECIPITATION EXTREMES in north China is 0.2–0.4 d year−1 , implying weak precipitation. Different climate systems control precipitation changes over China. Precipitation changes in the regions east of 100 ° E are controlled mainly by the east Asian summer monsoon (Wang et al., 2004). Our previous studies indicated the weakening of east Asian summer monsoon after about the mid-1970s and increasing geopotential height in north China, South China Sea, and west Pacific regions, all of which combine to negatively affect the northward propagation of vapour flux. Besides, weaker east Asian summer monsoon in recent decades does not benefit the northward propagation of water vapour flux and has potential to cause increasing (decreasing) moisture content and moisture budget in the regions south (north) to the Yangtze River basin (Zhang et al., 2008, 2011). Precipitation changes are heavily affected by the transport of moisture in the latitudinal direction (Jin et al., 2006). Increasing precipitation in northwest China should be attributed to the increasing geopotential height in Siberia and the decreasing geopotential height in the Iran Plateau. Besides, the increasing water vapour convergence after 1987 also contributes to the increasing precipitation in the western part of northwest China. However, the precipitation changes in the eastern part of the northwest China are still partly influenced by the strength variations in the east Asian summer monsoon (Chen and Dai, 2009). 6. Conclusions The probabilistic behaviours of precipitation extremes based on copulas are investigated in both space and time. Trends and changing magnitudes of precipitation extremes are also studied using the modified MK trend test and linear regression. The following conclusions are drawn from this study: (1) Precipitation is increasing from northwest to southeast China. Regions west of 100 ° E are dominated by increasing NW; the decreasing NW is found mainly in the regions west of 100 ° E. Significant decreasing NW is observed in the Yellow River basin, the Huaihe River basin, the upper Yangtze River, and the upper Pearl River basin, and these regions are also characterized by significant increasing CDD. These results imply a wetting tendency in the northwest China and a drying tendency in northern China. (2) In south China, NW is decreasing but is not significant and is increasing in southeast China with parts of the regions dominated by significant increasing CDD. The percentage of P90 to the annual total precipitation and I90 are both increasing in south China, especially in the lower Yangtze River basin and the Pearl River basin. Besides, larger occurrence probability of heavy precipitation events is also found in these regions, implying a higher risk of flood hazards in the lower Yangtze River basin and the Pearl Copyright 2012 Royal Meteorological Society 1151 River basin. Moreover, joint return periods of the events that P90 and I90 are both exceeding their 10 year return period values shows larger probability of heavy precipitation in the regions east of 100 ° E, implying that precipitation changes in east China are shifting to the extreme side. Extreme tendency of precipitation in northwest China is not evident. (3) The spatial distribution of flood- and drought-affected crop areas is in agreement with that of precipitation extremes, showing a considerable influence of precipitation changes, particularly the variations in precipitation extremes, on the occurrence of floods and droughts. Generally, agriculture in the south China is mostly affected by floods; however, agriculture in north China is mostly affected by droughts. The drought-affected crop area in the northwest China is increasing but is not significant at >95% confidence level, which should be due to increasing precipitation in recent decades. A drying tendency reflected by the increasing number of consecutive non-rain days should be paid enough attention. Generally, the precipitation extremes in the regions east of 100 ° E tend to be intensifying and enhancing. It should be noted here that the regions east of 100 ° E are usually heavily populated and bears the heavy responsibility for agricultural production and socioeconomy of China. Therefore, effective water resource management and scientific management of agricultural activities and related water-saving irrigation facilities are urgently required for the sustainable development of socioeconomy and enhancing social resilience and human mitigation of natural hazards. Acknowledgements This work was financially supported by Xinjiang Technology Innovative Program (Grant Nos. 201001066 and 200931105), the National Natural Science Foundation of China (Grant Nos. 41071020 and 50839005), the Project of the Guangdong Science and Technology Department (Grant Nos. 2010B050800001 and 2010B050300010), and by a grant from the Research Grants Council of the Hong Kong Special Administrative Region, China (Project No. CUHK405308). Thanks should be owed to the National Climate Centre of China for providing meteorological data. The last but not the least, our cordial gratitude should also be extended to the editor, Prof. Dr. Glenn McGregor, and reviewers for their constructive and pertinent comments and suggestions, which greatly helped improve the quality of this article. References Akaike H. 1974. A new look at the statistical model identification. IEEE Transactions on Automatic Control 19(6): 716–723. Allen M, Ingram WJ. 2002. Constraints on future changes in climate and the hydrologic cycle. Nature 419: 224–232. Arnell NW. 1999. Climate change and global water resources. Global and Environmental Change 9: S31–S49. Chen DD, Dai YJ. 2009. Characteristics and analysis of typical anomalous summer rainfall patterns in Northwest China over the Int. J. Climatol. 33: 1140–1152 (2013) 1152 Q. ZHANG et al. last 50 years. Chinese Journal of Atmospheric Sciences 33(6): 1247–1258 (in Chinese). Daufresne M, K Lengfellner, U Sommer. 2009. Global warming benefits the small in aquatic ecosystems. Proceedings of the National Academy of Sciences of the United States of America 106(31): 12788–12793. De Michele C, Salvadori G, Canossi M, Petaccia A, Rosso R. 2005. Bivariate statistical approach to check adequacy of dam spillway. Journal of Hydrologic Engineering 10(1): 50–57. Easterling RD, Meehl AG, Parmesan C, Changnon AS, Karl RT, Mearns OL. 2000. Climate extremes: observations, modeling, and impacts. Science 289: 2068–2074. Fatichi S, Caporali E. 2009. A comprehensive analysis of changes in precipitation regime in Tuscany. International Journal of Climatology 29(13): 1883–1893. Frank J, Masse J. 1951. The Kolmogorov–Smirnov test for goodness of fit. Journal of the American Statistical Association 46(253): 68–78. Freund JE. 1961. A bivariate extension of the exponential distribution. Journal of the American Statistical Association 56: 71–977. Genest C, Rivest LP. 1993. Statistical inference procedures for bivariate Archimedean copulas. Journal of the American Statistical Association 88(423): 1034–1043. Groisman PY, Karl TR, Easterling DR, Knight RW, Jameson PF, Hennessy KJ, Suppiah R, Page CM, Wibig J, Fortuniak K, Razuvaev V, Douglas A, Rorland E, Zhai PM. 1999. Changes in probability of heavy precipitation: important indicators of climatic change. Climatic Change 42: 243–283. Hamed KH, AR Rao. 1998. A modified Mann–Kendall trend test for autocorrelated data. Journal of Hydrology 204(1–4): 182–196. Hashino M. 1985. Formulation of the joint return period of two hydrologic variates associated with a Poisson process. Journal of Hydroscience and Hydrolic Engineering 3(2): 73–84. Hosking JRM. 1990. L-moments: analysis and estimation of distributions using linear combinations of order statistics. Journal of the Royal Statistical Society 52(1): 105–124. Jin LY, Fu JL, Chen FH. 2006. Spatial and temporal distribution of water vapor and its relationship with precipitation over the Northwest China. Journal of Lanzhou University 42(1): 1–5 (in Chinese). Kao SC, Govindaraju RS. 2008. Trivariate statistical analysis of extreme rainfall events via the Plackett family of copulas. Water Resources Research 44: W02415, DOI: 10.1029/2007WR006261. Kao SC, Govindaraju RS. 2010. A copula-based joint deficit index for droughts. Journal of Hydrology 380: 121–134. Kendall MG. 1955. Rank Correlation Methods. Griffin: London. Leonard M, Metcalfe A, Lambert M. 2008. Frequency analysis of rainfall and streamflow extremes accounting for seasonal and climatic partitions. Journal of Hydrology 348: 135–147. Mann HB. 1945. Nonparametric tests against trend. Econometrica 13: 245–259. Nelsen RB. 2006. An Introduction to Copulas. Springer Science: Portland, 7–269. Salvadori G, Michele C. 2004. Frequency analysis via copulas: theoretical aspects and applications to hydrological events. Water Resource Research. DOI: 10.1029/2004WR003133. Singh VP, Zhang L. 2007. IDF curves using the Frank Archimedean copula. Journal of Hydrologic Engineering 12(6): 651–662. Song S, Singh VP. 2010a. Meta-elliptic copulas for drought frequency analysis of periodic hydrologic data. Stochastic Environmental Research and Risk Assessment 24(3): 425–444. Song S, Singh VP. 2010b. Frequency analysis of droughts using the Plackett copula and parameter estimation by genetic algorithm. Stochastic Environmental Research and Risk Assessment 24(5): 783–805. Copyright 2012 Royal Meteorological Society Spiegel MR, Schiller J, Srinivasan RA. 2000. Outlines of Theory and Problems of Probability and Statistics, 2nd edn. McGraw-Hill: USA, 2–7. von Storch H. 1995. Misuses of statistical analysis in climate research. In Analysis of Climate Variability: Applications of Statistical Techniques. Springer-Verlag: Berlin, 11–26. Suppiah R, Hennessy K. 1998. Tends in seasonal rainfall, heavy raindays, and number of dry days in Australia 1910–1990. International. Journal of Climatology 18: 1141–1155. Tebaldi C, Hayhoe K, Arblaser JM, Meechl GA. 2006. Going to the extremes, an inter-comparison of model-simulated historical and future changes in extreme events. Climatic Change 79: 185–211. Vörösmarty CJ, Mclntyre PB, Gessner MO, Dudgeon D, Green P, Glidden S, Sullivan CA, Liermann CR, Davies PM. 2010. Global threats to human water security and river biodiversity. Nature 467: 555–561. Wang BJ, Huang YX, He JH. 2004. Relationship between moisture transportation during the East Asian Monsoon period and the droughts in the northwest China. Plateau Meteorology 23(6): 912–918. Wilks DS. 1999. Interannual variability and extreme-value characteristics of several stochastic daily precipitation models. Agricultural and Forest Meteorology 93: 153–169. World Meteorological Organization (WMO). 2003. Statement on the Status of Global Climate in 2003. WMO Publication No. 966. WMO: Geneva, Switzerland. Yue S, Pilon P, Phinney B, Cavadias G. 2002. The influence of autocorrelation on the ability to detect trend in hydrological series. Hydrological Processes 16(9): 1807–1829. Zhai PM, Ren FM, Zhang Q. 1999. Detection of trends in China’s precipitation extremes. Acta Meteorologica Sinica 57(2): 208–216. Zhang L, Singh VP. 2006. Bivariate flood frequency analysis using the copula method. Journal of Hydrologic Engineering 11(2): 150–164. Zhang L, Singh VP. 2007a. Bivariate rainfall frequency distributions using Archimedean copulas. Journal of Hydrology 332: 93–109. Zhang L, Singh VP. 2007b. Gumbel–Hougard copula for trivariate rainfall frequency analysis. Journal of Hydrologic Engineering 12(4): 431–439. Zhang Q, Xu C-Y, Becker S, Zhang ZX, Chen YD, Coulibaly M. 2009a. Trends and abrupt changes of precipitation extremes in the Pearl River basin, China. Atmospheric Science Letter 10: 132–144. Zhang Q, Xu C-Y, Gemmer M, Chen YD, Liu C-L. 2009b. Changing properties of precipitation concentration in the Pearl River basin, China. Stochastic Environmental Research and Risk Assessment 23: 377–385. Zhang Q, Xu C-Y, Chen XH, Zhang ZX. 2011. Statistical behaviors of precipitation regimes in China and their links with atmospheric circulation 1960–2005. International Journal of Climatology. DOI: 10.1002/joc.2193. Zhang Q, Xu C-Y, Zhang ZX, Chen YD, Liu C-L. 2008. Spatial and temporal variability of precipitation maxima during 1960–2005 in the Yangtze River basin and possible association with large-scale circulation. Journal of Hydrology 353: 215–227. Ziegler AD, Sheffield J, Maurer EP, Nijssen B, Wood EF, Lettenmaier DP. 2003. Detection of intensification in global- and continentalscale hydrological cycles: temporal scale of evaluation. Journal of Climate 16: 535–547. Zolina O, Simmer C, Gulev KS, Kollet S. 2010. Changing structure of European precipitation: longer wet periods leading to more abundant rainfalls. Geophysical Research Letters 37: L06704, DOI: 10.1029/2010GL042468. Int. J. Climatol. 33: 1140–1152 (2013)