Abstract

Under changing environment, the feasibility and potential impact of an inter-basin water transfer project can be evaluated by employing the coincidence probability of runoff in water sources area (WSA), water receiving area (WRA), and the downstream impacted area (DIA). Using the Han River to Wei River Water Transfer Project (HWWTP) in China as an example, this paper computed the coincidence probability and conditional probability of runoff in WSA, WRA and DIA with the copula-based multivariate joint distribution and quantified their acceptable and unfavorable encounter probabilities for evaluating the water supply risk of the water transfer project and exploring its potential impact on DIA. Results demonstrated that the most adverse encounter probability (dry–dry–dry) was 26.09%, illustrating that this adverse situation could appear about every 4 years. The acceptable and unfavorable probabilities in all encounters were 44.83 and 55.17%, respectively, that is the unfavorable situation would be dominant, implying flood and drought risk management should be paid greater attention in project operation. The conditional coincidence probability (dry WRA & dry DIA if dry WSA) was close to 70%, indicating a requirement for an emergency plan and management to deal with potential drought risk.

HIGHLIGHTS

  • Coincidence probability of annual runoff in all main impacted areas is investigated.

  • Copula-based method to capture the encounter situations of water transfer project.

  • Acceptable and unfavorable probabilities in all encounters imply water transfer risk.

  • Conditional coincidence probability with dry water resources area was computed.

  • An emergency water scheduling plan is required to deal with potential drought risk.

Graphical Abstract

Graphical Abstract
Graphical Abstract

INTRODUCTION

Inter-basin water transfer (IBT) means building water transfer projects that span two or more basins for transferring water from a basin with abundant water resources to those in shortage and for redistributing water resources among the basins to meet water demand in the water-deficient area (Zhuang 2016). IBTs project, as an important safeguard to join different water systems, has been widely applied in many countries and regions, e.g. Australia, Canada, China, India and the United States, with the purpose of supporting economic and societal development (Manshadi et al. 2013; Yan & Chen 2013; Du et al. 2019). However, it is being debated that the hydrological cycle has altered in some basins or regions under the combined influence of climate change and human activities (Zou et al. 2018), with the result that the feasibility of IBT projects planned or under construction may be questionable. The questions include whether there will be enough water to be diverted and the new potential influence on water use, ecological protection and disaster control (e.g. drought and pollution) in the downstream areas of the water resources area under the changing environment. Thus, the coincidence probability analysis of annual runoff in these areas, including synchronous and asynchronous probabilities between or among the water sources area (WSA), water receiving area (WRA) and the downstream impacted area (DIA), is appealing for evaluating the feasibility and potential influence of IBT projects.

In previous studies, the coincidence probability analysis of IBT projects played an important role in determining project feasibility. Some multivariate approaches as main analysis tools were applied to calculate synchronous–asynchronous probabilities of streamflow, precipitation and drought between water source area and WRA. For instance, He et al. (2012) employed a Bayesian network model to investigate the rich–poor rainfall encounter risk between WSA and WRA in the middle route of the South-to-North Water Transfer Project (SNWTP) in China and implemented real-time scenario simulations, reflecting the change in risk in the operation of water transfer projects. Liu et al. (2015) investigated the concurrent drought probability between the water source and destination regions of the central route of the SNWTP using Clayton copula and general circulation models (GCMs) and found that the probability of concurrent drought events was highly likely to increase during 2020–2050, representing the WSA probably having insufficient amounts of water for diversion while the WRA urgently needing the diversion of water in concurrent drought years. Du et al. (2019) investigated the synchronous and asynchronous exceedance probabilities of wet versus dry conditions of precipitation in WSA and WRA of the Shuhe to Futuan Water Transfer Project (SFWTP) using a bivariate copula joint distribution function, and results indicated that climate change had positive impacts on the exceedance probabilities, demonstrating that the project risk was decreasing. In addition, the coincidence probability analysis of precipitation in WSA, WRA and DIA had been reported by Yan & Chen (2013). This analysis quantified the synchrony and asynchrony of precipitation for the middle route of SNWTP and verified the effectiveness of trivariate Clayton copula in the study area, and obtained the combination frequencies for the middle route of SNWTP, representing that the amount of transferable water was generally assured, but the possibility for water transfer was very small if extreme deficit rainfall events occurred in the WRA.

It is found from the literature that bivariate and trivariate copulas have been widely used in the coincidence probability analysis of hydrological variables in the relevant areas of IBT projects (Yan & Chen 2013; Du et al. 2019). Therein, bivariate copulas are generally used to investigate the hydrological combination frequencies between WSA and WRA for determining the feasibility of a water transfer project, and that between WSA and DIA for analyzing the potential influence of the project on the downstream areas (Yan & Chen 2013). Trivariate copulas are employed to calculate the hydrological combination frequencies and conditional frequencies among WSA, WRA and DIA by capturing the spatial dependence structure of hydrological variables influencing the coincidence probabilities of asynchrony and synchrony. Recent years have seen a growing interest in applying copulas to hydrological frequency analysis, which can be attributed to manifold advantages of copulas in modeling joint distributions, representing flexibility in selecting arbitrary margins and dependence structure, the ability to deal with three variables or more, and the separability in analyzing marginal and dependence structure (Salvadori & De Michele 2007; Serinaldi et al. 2009; Zhang & Singh 2019).

In this study, trivariate combination frequencies and conditional frequencies of annual runoff series in WSA, WRA and DIA influenced by the Han River to Wei River Water Transfer Project (HWWTP) in China, were investigated through copula-based multivariate probability distribution, with the purpose of estimating water supply risk of water transfer project and exploring its potential impact on the downstream area of the water source areas. Different from the previous relevant studies on the water transfer project, it uses annual runoff data from three gauges located in WSA, WRA and DIA, respectively and provides a reference for the scheduling of inter-basin transfer projects. The bivariate and trivariate copulas were employed to build the joint probability distribution between the annual runoff of three regions (WSA, WRA and DIA) and trivariate coincidence probability and conditional coincidence probability were analyzed to obtain information for flood and drought risk management for the project.

STUDY AREA AND DATA

Han River to Wei River Water Transfer Project

As an important project to resolve the water conflict in northwest China due to increasing industrial and domestic water consumption, the HWWTP connecting the upper Han River basin (WSA) and Wei River basin (WRA) in China (as shown in Figure 1) is designed to annually transfer an average of 1,000 million m3 of water in 2025 and 1,500 million m3 in 2030 from Huangjinxia reservoir and Sanhekou reservoir in the upper Han River basin to large-scale industrial zones and urban areas in the Wei River basin, with the main purpose of alleviating watershed water conflict (Wu et al. 2019). Because 70% of water is for industrial development, it requires stable water transfer, implying high requirement in controlling water supply risk, especially in the changing environment.

Figure 1

Study area and locations of three gauges potentially impacted by the Han River to Wei River Water Transfer Project (HWWTP) in China.

Figure 1

Study area and locations of three gauges potentially impacted by the Han River to Wei River Water Transfer Project (HWWTP) in China.

The WSA and the DIA of HWWTP are located in the upper Han River basin, with a drainage area of 95,200 km2 and have a mainstream length with approximately 925 km (Figure 1). The basin is influenced by the north subtropic monsoon, representing the annual mean temperature ranging between 12 and 16 °C (Li et al. 2009) and average annual precipitation varying from 700 to 1,800 mm, and 80% of annual precipitation falls in the wet period from May to October in which the corresponding runoff is about 4.11 × 1010 m3, accounting for 70% of the total annual runoff. Also, the runoff of the entire basin demonstrates a large annual and interannual variability (Li et al. 2008). Therein, the WSA mainly refers to the watershed above Huangjinxia and Sanhekou reservoirs and the DIA is downstream of them, primarily covering the watershed below Shiquan gauge.

The WRA of HWWTP mainly refers to Guanzhong plain region located in the central part of Shaanxi Province and belongs to the Wei River basin. The Guanzhong plain region covers a total area of 55,500 km2, as the most economically developed area in Shaanxi Province, including five prefecture-level cities, i.e. Xi'an city, Baoji city, Xianyang city, Weinan city, Tongchuan city and one agricultural high-tech industries demonstration zone as Yangling (Tian et al. 2020). The Guanzhong plain region has an average elevation of approximately 500 m and belongs to the continental monsoon climate, with the annual mean temperature of 6–13 °C and precipitation of 500–800 mm (Deng et al. 2020).

Data

According to the water transfer route of the HWWTP, the data used in this paper included annual runoff data from Shiquan (SQ) gauge on the WSA, Danjiangkou gauge (the watershed above SQ is not covered or the data deducted that from at SQ, DSDJK) on the DIA, and Xianyang (XY) gauge on the WRA (Figure 1) (see Supplementary Material). Therein, the observed runoff data series from SQ and DSDJK gauges were selected during 1956–2000. Considering that measured runoff is greatly affected by human activities in WRA, representing nonstationary feature (Lan et al. 2020), the natural runoff data, covering 45 years from 1956 to 2000, were selected from the XY gauge. The runoff data were obtained from hydrological manuals published by the Hydrological Bureaus of the Yellow River Conservancy Commission and Yangtze River Conservancy Commission. Statistical features of the annual data series are shown in Table 1.

Table 1

Statistical features of annual runoff during 1956–2000

GaugesMean (108m3)Standard deviation (108m3)Coefficient of variationCoefficient of skewness
Shiquan (SQ) 327.5 143.6 0.4385 0.9964 
Xianyang (XY) 49.8 20.6 0.4125 0.7392 
Danjiangkou (DSDJK) 368.1 136.4 0.3703 1.0753 
GaugesMean (108m3)Standard deviation (108m3)Coefficient of variationCoefficient of skewness
Shiquan (SQ) 327.5 143.6 0.4385 0.9964 
Xianyang (XY) 49.8 20.6 0.4125 0.7392 
Danjiangkou (DSDJK) 368.1 136.4 0.3703 1.0753 

Note: Runoff at Danjiangkou does not include the runoff from the area above Shiquan gauge.

METHODOLOGIES

Copula functions

Sklar (1959) introduced the copula theorem to model the stochastic nature of multi-dimensional processes by using univariate marginal distributions to derive multivariate distribution functions. Nowadays, copula models play an active role in the hydrological field (Yan & Chen 2013; Zhang & Singh 2019), such as drought assessment (Chanda et al. 2014), flood risk analysis (Favre et al. 2004; Serinaldi et al. 2009; Chebana et al. 2012) and streamflow simulation (Chen et al. 2015). According to Sklar's theorem, the copula function C can be used to describe a joint multivariate probability distribution of n correlated variables (X1, X2, X3,…, Xn). If variables (X1, X2, X3,…, Xn) have arbitrary marginal distribution functions , respectively, there will exist a copula function C, which can combine these marginal distribution functions to give the joint distribution function, , as follows: 
formula
(1)
where are uniformly distributed random realizations of the variables, defined as , and the copula parameter is θ. A unique copula function C(⋅) exists, when these marginal distributions are continuous (Sraj et al. 2015).

There are many types of copula functions, such as Plackett copula (Plackett 1965), Archimedean copulas and elliptical copulas (Fang et al. 2002). Because the Archimedean copulas can be easily constructed and are capable of capturing a variety of dependence structures with several desirable properties, such as symmetry and associativity, they have become very popular (Nelsen 2006; Hofert 2008).

The widely used Archimedean family copulas include the Frank copula, Clayton copula, Ali–Mikhail–Haq copula (AMH) and Gumbel–Hougaard (G–H) copula, with a parameter θ. For bivariate copula function, Kendall's correlation coefficient τ has a certain relation with parameter θ, as shown in Table 2. The appropriate copulas can be determined by different values of Kendall's τ from observations (Hofert 2011; Xie & Wang 2013).

Table 2

Types of Clayton, Frank and GH copula (Nelsen 2006)

Archimedean copulaCθθ′ rangeRelation between θ and τ
Frank    
Clayton    
AMH    
G–H    
Archimedean copulaCθθ′ rangeRelation between θ and τ
Frank    
Clayton    
AMH    
G–H    

Moreover, parameters of multi-dimensional copulas can be estimated by the maximum-likelihood estimator (MLE) or inference of functions for margins (IFM) method (Joe 1998, 2005). In view of the well-known optimality properties of the MLE, it would be the preferred option for estimating θ. Nevertheless, it is found from application that the more flexible IFM method is preferable to the MLE. Although the conceptual bases of the two methods are very similar, and the efficiency of these two methods is almost the same in many cases, the IFM method is easier to calculate than the MLE method.

For example, the trivariate copula parameter θ is estimated in two steps in the IFM method. In the first step, the parameters αk (k = 1, 2, 3) of each marginal distribution are estimated separately via Xki, i= 1,2,…,n, and this estimator is expressed as . The second step is that θ is estimated by replacing for αk in the log-likelihood (as shown in Equation (2)) and the IFM estimate of θ is as follows: 
formula
(2)

Marginal distribution

The Pearson type III (P-III) distribution, Gumbel distribution and Lognormal distribution were employed in this study for describing the marginal distribution of single annual runoff.

The cumulative distribution function (CDF) of P-III distribution can be defined as 
formula
(3)
where , and are the shape, scale and location parameters of the P-III distribution, respectively.
The CDF of Gumbel distribution can be defined as 
formula
(4)
where and k are the scale and the location parameters of the Gumbel distribution, respectively.
The CDF of Lognormal distribution can be defined as 
formula
(5)
where and are the mean and standard deviation formed by the natural logarithm of the variable x, respectively.

The linear moment method (Hosking 1990) was employed to estimate the parameters of the above three distributions, and a fitting test was used to choose the suitable marginal distribution of a single variable.

Fitting test

In this paper, Kolmogorov–Smirnov (K–S) test (Dn, Massey 1951), root mean square error (RMSE, Zhang & Singh 2006, 2019), Akaike's information criterion (AIC, Zhang & Singh 2007, 2019) were employed to measure the goodness of fit of the marginal and joint distributions which can be expressed as follows: 
formula
(6)
 
formula
(7)
 
formula
(8)
where N is the sample size, k is the number of parameters of different distributions, PEi and PTi are the empirical frequency and theoretical frequency, respectively. Therein, the empirical joint probability always plays a significant role in the selection of a copula function, which is used as a criterion for judging and selecting the best theoretical distribution from the Archimedean copula functions. Assuming the variables X and Y with the same length, the empirical frequency of bivariate variables (X and Y) was estimated by using the Gringorten formula (Gringorten 1963), as follows. 
formula
(9)
where is the combination of the ith values in the X and Y series arranged in increasing order, i is 1: N, and N is the length of the series. For the trivariate copula function also, the above equation was used. The theoretical frequencies can be obtained by the models of marginal and copula distributions.

It is noted that according to RMSE and AIC criteria, the smaller AIC and RMSE are, the better the fitness of the distribution.

Coincidence probability

The coincidence probability is usually defined as the probability of two or more events that happen at the same time. For the IBT project, it is considered that the runoff coincidence is generally defined as the simultaneous occurrence of runoff in two or more basins, which represent the frequency P of wetness, dryness or normal of one basin when the other basin is in a condition of wetness, dryness or normal. It is reported that the 37.5 and 62.5% quantiles are usually used as thresholds to define the condition of wetness or dryness in precipitation coincidence (Liu & Zheng 2002; Yan & Chen 2013), and this paper used them due to high rainfall–runoff relation and simultaneous frequency.

For the annual runoff variable X, Xw and Xd were assigned the values of Pw = 62.5% and Pd = 37.5%, respectively, as the threshold quantiles of wetness and dryness, respectively. The degree of wet–dry runoff can be described as wetness (XXw = X62.5%), dryness (X < Xd=X37.5%) and normal (Xw > XXd). In terms of bivariate copula, there were nine encounter situations (X and Y), and the trivariate copula (X, Y and Z) had 27 combinations. Some of all 36 combination functions were as follows.

Wet–wet periods coincidence probability, Pww: 
formula
(10)
Wet–dry periods coincidence probability, Pwd: 
formula
(11)
Normal–wet periods coincidence probability, Pnw: 
formula
(12)
Wet–wet–wet periods coincidence probability, Pwww: 
formula
(13)
Wet–normal–dry periods coincidence probability, Pwnd: 
formula
(14)
Dry–dry–dry periods coincidence probability, Pddd: 
formula
(15)
where u, v and are the marginal distributions of X, Y and Z, and subscripts w, n and d mean wet, normal and dry conditions. The composite letters note the coincidence condition. For example, ddd presents the dry–dry–dry condition (period), and Pddd is the coincidence probability of this condition. Here, the product notation is used to express the simultaneous occurrence of events. Other combination probabilities listed in Tables 3 and 4 were calculated in a similar way.
Table 3

Nine coincidence combinations of bivariate copula

XY
WetNormalDry
Wet    
Normal    
Dry    
XY
WetNormalDry
Wet    
Normal    
Dry    
Table 4

Twenty-seven coincidence combinations of trivariate copula

XYZ
WetNormalDry
Wet Wet    
Normal    
Dry    
Normal Wet    
Normal    
Dry    
Dry Wet    
Normal    
Dry    
XYZ
WetNormalDry
Wet Wet    
Normal    
Dry    
Normal Wet    
Normal    
Dry    
Dry Wet    
Normal    
Dry    

RESULTS AND DISCUSSION

Parameter estimation and fitting test of marginal distributions

The parameters of the P-III, Gumbel, and Lognormal distributions mentioned above were estimated using the linear moment method, and results are shown in Table 5.

Table 5

Estimated parameters of P-III, Gumbel and Lognormal distributions

GaugesP-III
Gumbel
Lognormal
SQ 4.0371 71.4811 38.9582 110.7327 263.6216 327.5376 142.0196 
XY 8.1332 7.2092 −8.7950 15.8512 40.6888 49.8382 20.3298 
DSDJK 3.4689 73.1856 114.2435 105.0921 307.4579 368.1180 134.7853 
GaugesP-III
Gumbel
Lognormal
SQ 4.0371 71.4811 38.9582 110.7327 263.6216 327.5376 142.0196 
XY 8.1332 7.2092 −8.7950 15.8512 40.6888 49.8382 20.3298 
DSDJK 3.4689 73.1856 114.2435 105.0921 307.4579 368.1180 134.7853 

Also, the K–S test (Dn), RMSE and AIC were employed to test the feasibility of fitting these three distributions to runoff data, and results are shown in Figure 2.

Figure 2

Marginal distributions fitted with empirical frequency and models.

Figure 2

Marginal distributions fitted with empirical frequency and models.

The statistical results in Figure 2 indicate that the P-III, Gumbel and Lognormal distributions all passed the K–S test with the critical value . At SQ gauge, it was seen that the AIC of the Gumbel distribution was the smallest of the three models, while the P-III distribution had the smallest RMSE. After consideration, the P-III distribution was finally selected to fit the annual runoff data from the SQ gauge. For XY and DSDJK gauges, the Gumbel distribution had both the smallest AIC and RMSE, thus the Gumbel distribution was employed to fit at the two gauges. The curves in Figure 2 represent the performances of three models in fitting annual runoff data from the three gauges visually.

Parameter estimation and fitting test of joint distribution

Various forms of the copula function have different requirements for the correlation of variables. For example, the G–H copula and Clayton copula functions are suitable for random variables with positive correlation, the Frank copula function applicable to random variables with positive and negative correlations, and the AMH function is often applied to random variables with weak correlations. For this, Pearson's correlation coefficient (R) and Kendall's tau (τ) are first calculated to describe the dependence (concordance) of annual runoff for any two of the three gauges, and the values of R and τ are shown in Table 6.

Table 6

Pearson's correlation coefficient (R) and Kendall’ tau (τ) between runoff series from three gauges

GaugesRτ
SQ–XY 0.7983 0.5993 
SQ–DSDJK 0.8542 0.6721 
XY–DSDJK 0.7355 0.4611 
GaugesRτ
SQ–XY 0.7983 0.5993 
SQ–DSDJK 0.8542 0.6721 
XY–DSDJK 0.7355 0.4611 

In Table 6, it is seen that runoff series from any two of the three gauges had positive correlation, thus the G–H copula and Clyton cupola functions were considered suitable for the data series analysis.

Then, Kendall's correlation coefficient τ was counted between annual runoff data from any two of the three gauges, and the corresponding copula parameters θ can be computed in terms of the equations of the G–H and Clyton copulas listed in Table 2. The values of τ and θ are shown in Table 7.

Table 7

Parameter values of bivariate copula functions

GaugesτΘ
G––HClayton
SQ–XY 0.5993 2.494 2.988 
SQ–DSDJK 0.6721 3.049 4.098 
XY–DSDJK 0.4611 1.855 1.711 
GaugesτΘ
G––HClayton
SQ–XY 0.5993 2.494 2.988 
SQ–DSDJK 0.6721 3.049 4.098 
XY–DSDJK 0.4611 1.855 1.711 

Furthermore, the bivariable joint distributions from the Clayton copula and G–H copula were constructed by using the estimated parameters and the equations in Table 2, respectively, while the parameters of trivariate Clayton and G–H copula functions need to be estimated through Equation (2) for the construction of trivariate distributions, which is different from the bivariable joint distribution. So, the fitness results of all joint distributions, representing Dn, RMSE and AIC, are shown in Tables 8 and 9.

Table 8

K–S test (Dn), RMSE and AIC of bivariate copulas

GaugesDn
RMSE
AIC
G–HClaytonG–HClaytonG–HClayton
SQ–XY 0.1600 0.1280 0.0854 0.0811 −219.47 −224.12 
SQ–DSDJK 0.1279 0.1086 0.0700 0.0701 −237.33 −237.16 
XY–DSDJK 0.1840 0.1472 0.1016 0.0995 −203.82 −205.71 
GaugesDn
RMSE
AIC
G–HClaytonG–HClaytonG–HClayton
SQ–XY 0.1600 0.1280 0.0854 0.0811 −219.47 −224.12 
SQ–DSDJK 0.1279 0.1086 0.0700 0.0701 −237.33 −237.16 
XY–DSDJK 0.1840 0.1472 0.1016 0.0995 −203.82 −205.71 
Table 9

K–S test (Dn), RMSE and AIC of trivariate copulas

Trivariate copula functionsΘDnRMSEAIC
G–H copula 3.49 0.0921 0.1646 −212.66 
Clayton copula 3.27 0.1160 0.1648 −191.85 
Trivariate copula functionsΘDnRMSEAIC
G–H copula 3.49 0.0921 0.1646 −212.66 
Clayton copula 3.27 0.1160 0.1648 −191.85 

Tables 8 and 9 show that bivariate and trivariate copulas, based on the G–H and Clayton copula, both passed the K–S test (Dn less than 0.207). The smallest values of RMSE and AIC (0.0811, −224.12) in Table 8 demonstrate that Clayton copula was the best-fitted bivariable joint distribution between runoff series from SQ and XY gauges and was also suitable for that between XY and DSDJK gauges, with RMSE and AIC of 0.0995 and −205.71. Differently, the best-fitted bivariable joint distribution between SQ and DSDJK gauges was the G–H copula, and the smallest values of RMSE and AIC were 0.0700 and −237.33. Additionally, it can be seen from Table 9 that the smallest values of RMSE and AIC are 0.1646 and −212.66, respectively, implying the G–H copula was the best suitable function among trivariate copula functions and can describe the joint distribution among three gauges.

Coincidence probability of bivariate copula

According to the chosen and established bivariate copula functions, the bivariate joint runoff distribution between any two gauges was obtained, and the joint coincidence probability was calculated under certain conditions. The joint cumulative probability distribution and contours at any two of the three coupled gauges are shown in Figure 3. From the contours in Figure 3, one can obtain the probabilities that annual runoff of any two gauges was less than a certain value at the same time, and that possible combinations of annual runoff from different gauges under a certain probability, such as the combination of annual runoff from the SQ gauge with a 30% probability and corresponding annual runoff from the XY gauge, or the combination of annual runoff from the XY gauge with a 30% probability and annual runoff from the SQ gauge. At this time, the joint cumulative probability in Figure 3 was able to analyze the worst situation of simultaneous dry periods occurring from any two gauges. For instance, it can be calculated that the joint probability was 10%, that the annual runoff was less than 2 × 1010 m3 at SQ gauge and less than 0.27 × 1010 m3 at XY gauge; that the joint probability was 90%, that the annual runoff was less than 0.86 × 1010 m3 at XY gauge and less than 7 × 1010 m3 at DSDJK gauge. For other examples, the possible combination of annual runoff at SQ and DSDJK gauges with a joint probability less than 50% was that the annual runoff was less than 5 × 1010 m3 at SQ gauge and less than 3.5 × 1010 m3 at DSDJK gauge, or that the annual runoff was less than 3.05 × 1010 m3 at SQ gauge and less than 6 × 1010 m3 at DSDJK gauge.

Figure 3

Contours and joint cumulative probability distribution: (a) joint cumulative probability distribution of annual runoff from the SQ and XY gauges; (b) contours of the joint cumulative probability distribution of annual runoff from the SQ and XY gauges; (c) joint cumulative probability distribution of annual runoff from the SQ and DSDJK gauges; (d) contours of the joint cumulative probability distribution of annual runoff from the SQ and DSDJK gauges; (e) joint cumulative probability distribution of annual runoff from the XY and DSDJK gauges; (f) contours of the joint cumulative probability distribution of annual runoff from the XY and DSDJK gauges.

Figure 3

Contours and joint cumulative probability distribution: (a) joint cumulative probability distribution of annual runoff from the SQ and XY gauges; (b) contours of the joint cumulative probability distribution of annual runoff from the SQ and XY gauges; (c) joint cumulative probability distribution of annual runoff from the SQ and DSDJK gauges; (d) contours of the joint cumulative probability distribution of annual runoff from the SQ and DSDJK gauges; (e) joint cumulative probability distribution of annual runoff from the XY and DSDJK gauges; (f) contours of the joint cumulative probability distribution of annual runoff from the XY and DSDJK gauges.

Through the constructed bivariate joint distributions, the nine coincidence probabilities of annual runoff data mentioned in Table 3 were computed at any two gauges. In the IBT project, the most adverse encounter situation was considered as dry–dry periods, which means that a shortage of water resources appeared both in the WSA and the WRA. The statistical results showed that Pdd was not too high, 30, 29.19 and 26.49% in SQ–XY, XY–DSDJK and SQ–DSDJK, respectively. The synchronous coincidence probabilities generally pointed to wet–wet periods coincidence probability, normal–normal periods coincidence probability and dry–dry coincidence probability, while other probabilities were regarded as asynchronous coincidence probabilities. The synchronous and asynchronous coincidence probabilities at any two gauges were computed, as shown in Table 10. Obviously, it is seen from the table that the synchronous coincidence probability was generally higher than asynchronous coincidence probability. It can be explained well due to the close geographical position among three gauges and similar climatic conditions. The larger synchronous coincidence probabilities indicated when there was enough water to be diverted in WSA, WRA or DIA was also likely to be wet, implying water demand in WRA for transferring water was small or the potential influence caused by the water transfer in DIA was low. Contrarily, when WSA was dry and cannot provide abundant water for diversion, water demand in WRA for transferring water was very high or the potential influence in DIA was greatly strong due to WRA or DIA being probably dry at that time. This was not expected in the IBT project, because it took on the project with a big risk, threatening the water security in WSA, WRA and DIA. Of course, when there was enough water to be diverted in WSA, WRA just happened to be dry and DIA was wet, and transferring water in the inter-basin provided the benefit at this time and caused the low impact in the downstream river basin. Thus, it was clear that coincidence probability of bivariate copula failed to provide a valuable reference when considering WSA, WRA and DIA of the IBT project

Table 10

Coincidence probability of bivariate copula (%)

ProbabilityPeriodsSQ–XYXY–DSDJKSQ–DSDJK
Synchronous probability Wet–wet 26.78 30.43 23.33 
Normal–normal 10.50 12.06 8.38 
Dry–dry 30.00 29.19 26.49 
Total 67.29 71.68 58.19 
Asynchronous Probability Wet–normal 8.86 5.85 9.89 
Wet–dry 1.86 1.22 4.28 
Normal–wet 8.86 5.85 9.89 
Normal–dry 5.64 7.09 6.73 
Dry–wet 1.86 1.22 4.28 
Dry–normal 5.64 7.09 6.73 
Total 32.72 28.32 41.8 
ProbabilityPeriodsSQ–XYXY–DSDJKSQ–DSDJK
Synchronous probability Wet–wet 26.78 30.43 23.33 
Normal–normal 10.50 12.06 8.38 
Dry–dry 30.00 29.19 26.49 
Total 67.29 71.68 58.19 
Asynchronous Probability Wet–normal 8.86 5.85 9.89 
Wet–dry 1.86 1.22 4.28 
Normal–wet 8.86 5.85 9.89 
Normal–dry 5.64 7.09 6.73 
Dry–wet 1.86 1.22 4.28 
Dry–normal 5.64 7.09 6.73 
Total 32.72 28.32 41.8 

Coincidence probability of trivariate copula

The coincidence probability of trivariate copula certainly provided an effective tool to comprehensively analyze the influence of the IBT project. Based on the G–H copula, this study constructed trivariate joint distributions, by which 27 coincidence probabilities from the trivariate copula, corresponding to the combinations shown in Table 4, were computed. Similarly, the wet–wet–wet coincidence probability Pwww, normal–normal–normal coincidence probability Pnnn, and dry–dry–dry coincidence probability Pddd were considered as synchronous coincidence probabilities of trivariate copula, while other probabilities were asynchronous coincidence probabilities. Furthermore, synchronous coincidence probabilities and asynchronous coincidence probabilities at three gauges (SQ–XY–DSDJK) can be obtained, as shown in Table 11.

Table 11

Coincidence probability of trivariate copula (%)

PeriodsWet (SQ)
Normal (SQ)
Dry (SQ)
Wet (XY)Normal (XY)Dry (XY)Wet (XY)Normal (XY)Dry (XY)Wet (XY)Normal (XY)Dry (XY)
Wet (DSDJK) 29.08 2.23 0.05 2.23 2.78 0.34 0.05 0.34 0.39 
Normal (DSDJK) 2.23 2.78 0.34 2.78 7.99 2.40 0.34 2.40 3.75 
Dry (DSDJK) 0.05 0.34 0.39 0.34 2.40 3.75 0.39 3.75 26.09 
Sum of synchronous probability (Pwww + Pnnn + Pddd63.16 
Sum of asynchronous probability (Pother36.84 
Unfavorable probability (Pwww + Pddd55.17 
Acceptable probability (Pother44.83 
PeriodsWet (SQ)
Normal (SQ)
Dry (SQ)
Wet (XY)Normal (XY)Dry (XY)Wet (XY)Normal (XY)Dry (XY)Wet (XY)Normal (XY)Dry (XY)
Wet (DSDJK) 29.08 2.23 0.05 2.23 2.78 0.34 0.05 0.34 0.39 
Normal (DSDJK) 2.23 2.78 0.34 2.78 7.99 2.40 0.34 2.40 3.75 
Dry (DSDJK) 0.05 0.34 0.39 0.34 2.40 3.75 0.39 3.75 26.09 
Sum of synchronous probability (Pwww + Pnnn + Pddd63.16 
Sum of asynchronous probability (Pother36.84 
Unfavorable probability (Pwww + Pddd55.17 
Acceptable probability (Pother44.83 

As is known, the purpose of HWWTP was to balance the non-uniform temporal and spatial distributions of water resources in WSA and WRA and increase water supply for guaranteeing industrial and domestic water safety in Guanzhong plain, China. Considering this, the wet–wet–wet and dry–dry–dry encounter situations were determined as unfavorable circumstances, in which the wet–wet–wet situation implying flood risk and the dry–dry–dry situation implying drought risk in this study. Other encounter situations were also regarded as acceptable conditions. These results are shown in Table 11 where the values of synchronous coincidence probabilities Pwww, Pnnn and Pddd were 29.08, 7.99 and 26.09%, respectively. Therein, the most adverse encounter situation was in dry–dry–dry periods, with the coincidence probability (Pddd) of 26.09%. Through the probability, the recurrence interval was obtained in the worst encounter situation, and the value was 3.8 years, which means the most adverse situation (dry–dry–dry) could appear once every 3.8 years on average. Also, it can be seen from the table that unfavorable probability (Pwww + Pddd) reached 55.17%, greater than the acceptable probability, implying the hydrological condition was very unfavorable to water transfer in more than half of the whole operation period. It undoubtedly posed a big challenge to the scheduling of inter-basin projects, threatening the safety of water delivery.

Furthermore, in order to implement the optimal scheduling of the transfer project, it was necessary to identify the possible runoff in WRA and DIA when different runoff rhythms happened in WSA. Fortunately, the joint distribution model established by the copula function provided an effective tool for solving this problem. So three equations were constructed to calculate dry–dry coincidence probabilities of annual runoff from the XY and DSDJK gauges with the condition that SQ gauge was in wet, normal and dry periods. The conditional dry–dry coincidence probabilities were obtained as follows, where X notes runoff from the SQ gauge, Y is that from the XY gauge and Z refers to the runoff from the DSDJK gauge.

Wet SQ, dry XY and dry DSDJK, 
formula
(16)
Normal SQ, dry XY and dry DSDJK, 
formula
(17)
Dry SQ, dry XY and dry DSDJK, 
formula
(18)
where u, v and are the marginal distributions of X, Y and Z, and subscript w, n and d mean the wet, normal and dry conditions, respectively.

The above calculation showed that the coincidence probability of wet SQ, dry XY and dry DSDJK was 1.04%; that of normal SQ, dry XY and dry DSDJK was 44.83%; that of dry SQ, dry XY and dry DSDJK was 69.57%. Therein, the probability that dry XY and dry DSDJK happened at the same time reached 69.57%, in the case that dry SQ was occurring, and that of dry XY and dry DSDJK arrived at 44.83% even though normal runoff occurred at SQ gauge. They implied that water supply deficit in WRA due to the insufficient transfer of water caused by low watershed yield in WSA, and water supply deficit in DIA due to low flow from the upstream watershed caused by normal water transfer, could disturb the operation of the HWWTP, when there was not abundant water yield in WSA.

To visualize the conditional coincidence probability among three gauges mentioned above, the distribution and contours of different magnitudes of runoff from XY and DSDJK gauges in the case that SQ gauge was in dry period were drawn, as shown in Figure 4. From the figure, the conditional coincidence probability of a certain magnitude was obtained directly from the contours. For example, when SQ gauge was dry, the conditional coincidence probability of dry XY and dry DSDJK was 20%, in which dry XY illustrated the annual runoff at XY gauge was less than 0.5 × 1010 m3 and dry DSDJK demonstrated that it was less than 2.2 × 1010 m3.

Figure 4

Conditional coincidence probability of XY–DSDJK given the runoff of SQ gauge.

Figure 4

Conditional coincidence probability of XY–DSDJK given the runoff of SQ gauge.

Also, the conditional coincidence probability of the most adverse encounter situation (dry A and dry B if dry C) for the IBT project was 69.57%, representing the encounter probability of dry XY and dry DSDJK in the case that SQ was in the dry period. Compared with the coincidence probability (dry SQ and dry XY and dry DSDJK) of trivariate copula in Table 11, the conditional coincidence probability (dry XY and dry DSDJK if dry SQ) better showed the importance of drought risk, indicating the dry–dry–dry coincidence probability in the study area was close to 70% when the drought event occurred in WSA. Thus, it should be emphasized again that the implementation effect of designed water transfer could not be guaranteed to a large extent, once WSA was in a dry period during the operation of HWWTP.

CONCLUSIONS

The Han River to Wei River Water Transfer Project (HWWTP) is a strategic project serving the water resources allocation in northwest China, aiming to mitigate water shortage in the Guanzhong Plain. It is important for HWWTP to estimate the water supply risk of the water transfer project and explore its potential impact on the downstream of the water sources areas. To discuss this issue, this study analyzed the coincidence probability of annual runoff series between or among the water source area (WSA), WRA and the DIA. In this analysis, the Archimedean copula-based method was used to capture the encounter situations of various water yields in different areas by constructing bivariate and trivariate joint distributions and calculated various combination frequencies, implying the water supply risk of the HWWTP in China. The main conclusions of the study can be summarized as follows.

The coincidence probability of annual runoff series from WSA (SQ gauge), WRA (XY gauge) and DIA (DSDJK gauge) was obtained for different encounter situations by using multivariate copula functions, which provided a valuable reference for water resources scheduling of the IBT project because they covered all primary areas impacted by the project.

The most adverse encounter in all situations was the case that three hydrological gauges (WSA, WRA and DIA) were in the dry period at the same time, i.e. dry–dry–dry, and the coincidence probability was 26.09% obtained by the trivariate copula. Accordingly, the recurrence interval was calculated with a value of 3.8 years, illustrating that this adverse situation would appear about every 4 years. According to the coincidence probability of annual runoff series from the three gauges, it was found that the acceptable and unfavorable probabilities were 44.83 and 55.17%, respectively. Obviously, the unfavorable probability was greater than the acceptable probability and should be paid greater attention to flood and drought risk management during the operation of the project.

Considering the most adverse encounter situation in WRA (XY gauge) and DIA (DSDJK gauge) under the condition that WSA (SQ gauge) was in the dry period, i.e. dry XY and dry DSDJK if dry SQ, the conditional coincidence probability was computed with a value of 69.57%. Different from the coincidence probability (dry–dry–dry) of trivariate copula, it better emphasized the risk of drought for the water transfer project, representing the probability of drought event simultaneously occurring in WRA and DIA was close to 70% when a drought event occurred in WSA. Therefore, it was inferred that the volume of designed transferable water could not be guaranteed to a large extent once WSA was in a dry period during the operation of HWWTP, which could lead to water shortages caused by drought in WRA and DIA could not be alleviated effectively and could further threaten regional water security. Thus, it is suggested that an emergency water resources scheduling plan and management programs should be drawn up, with the purpose of dealing with potential flood and drought risk of the IBT project.

ACKNOWLEDGEMENTS

The authors gratefully acknowledge the financial support from the National Natural Science Foundation of China (No. 51979005 and 51809005), the Natural Science Basic Research Program of Shaanxi (No. 2020JM-250) and State Key Laboratory of Eco-hydraulics in Northwest Arid Region, Xi'an University of Technology (No. 2018KFKT-4).

AUTHOR CONTRIBUTIONS

All authors contributed to this study. H.Z. provided the writing idea; X.W. carried out data analyses and wrote the first manuscript draft; C. D. and S. S. contributed to the analysis of results; Y.W. drew the figures; H.Z. and V.S. revised the manuscript. All authors read and approved the final manuscript.

DATA AVAILABILITY STATEMENT

All relevant data are included in the paper or its Supplementary Information.

REFERENCES

REFERENCES
Chanda
K.
Maity
R.
Sharma
A.
Mehrotra
R.
2014
Spatiotemporal variation of long-term drought propensity through reliability-resilience-vulnerability based drought management index
.
Water Resources Research
50
(
10
),
7662
7676
.
doi:10.1002/2014WR015703
.
Chebana
F.
Dabo-Niang
S.
Ouarda
T. B. M. J.
2012
Exploratory functional flood frequency analysis and outlier detection
.
Water Resources Research
48
(
4
),
4514
.
doi:10.1029/2011WR011040
.
Chen
L.
Singh
V. P.
Guo
S.
Zhou
J.
Zhang
J.
2015
Copula-based method for multisite monthly and daily streamflow simulation
.
Journal of Hydrology
528
(
528
),
369
384
.
doi:10.1016/j.jhydrol.2015.05.018
.
Du
H.
Wang
Y.
Liu
K.
Cheng
L.
2019
Exceedance probability of precipitation for the Shuhe to Futuan Water Transfer Project in China
.
Environmental Earth Sciences
78
(
7
),
1
12
.
doi:10.1007/s12665-019-8207-2
.
Fang
H. B.
Fang
K. T.
Kotz
S.
2002
The meta-elliptical distributions with given marginals
.
Journal of Multivariate Analysis
82
(
1
),
1
16
.
doi:10.1006/jmva.2001.2017
.
Favre
A. C.
Adlouni
S. E.
Perreault
L.
Thiemonge
N.
Bobee
B.
2004
Multivariate hydrological frequency analysis using copulas
.
Water Resources Research
40
(
1
),
1
12
.
doi:10.1029/2003WR002456
.
Gringorten
I. I.
1963
A plotting rule for extreme probability paper
.
Journal of Geophysical Research
68
(
3
),
813
814
.
doi:10.1029/JZ068i003p00813
.
He
X. C.
Kang
L.
Cheng
X. J.
Ding
Y.
2012
Flood risk analysis in the middle route of South-to-North Water Diversion Project of China based on Bayesian network
.
South-to-North Water Diversion and Water Science & Technology
10
(
4
),
10
13
.
doi:10.3724/sp.j.1201.2012.04010
.
Hofert
M.
2008
Sampling Archimedean copulas
.
Computational Statistics & Data Analysis
52
(
12
),
5163
5174
.
doi:10.1016/j.csda.2008.05.019
.
Hofert
M.
2011
Efficiently sampling nested Archimedean copulas
.
Computational Statistics & Data Analysis
55
(
1
),
57
70
.
doi:10.1016/j.csda.2010.04.025
.
Hosking
J. R.
1990
L-moments: analysis and estimation of distributions using linear combinations of order statistics
.
Journal of the Royal Statistical Society Series B – Methodological
52
(
1
),
105
124
.
doi:10.1111/j.2517-6161.1990.tb01775.x
.
Joe
H.
1998
Multivariate models and dependence concepts
.
Journal of the American Statistical Association
93
(
443
).
doi:10.2307/2669872
.
Joe
H.
2005
Asymptotic efficiency of the two-stage estimation method for copula-based models
.
Journal of Multivariate Analysis
94
(
2
),
401
419
.
doi:10.1016/j.jmva.2004.06.003
.
Lan
T.
Zhang
H.
Xu
C. Y.
Singh
V. P.
Lin
K.
2020
Detection and attribution of abrupt shift in minor periods in human-impacted streamflow
.
Journal of Hydrology
584
,
124637
.
doi:10.1016/j.jhydrol.2020.124637
.
Li
S.
Gu
S.
Liu
W.
Han
H.
Zhang
Q.
2008
Water quality in relation to the land use and land cover in the Upper Han River basin, China
.
Catena
75
(
2
),
216
222
.
doi:10.1016/j.catena.2008.06.005
.
Li
S.
Liu
W.
Gu
S.
Cheng
X.
Xu
Z.
Zhang
Q.
2009
Spatio-temporal dynamics of nutrients in the upper Han River basin, China
.
Journal of Hazardous Materials
162
(
2
),
1340
1346
.
doi:10.1016/j.jhazmat.2008.06.059
.
Liu
C.
Zheng
H.
2002
South-to-north water transfer schemes for China
.
International Journal of Water Resources Development
18
(
3
),
453
471
.
doi:10.1080/0790062022000006934
.
Liu
X.
Luo
Y.
Yang
T.
Liang
K.
Zhang
M.
Liu
C.
2015
Investigation of the probability of concurrent drought events between the water source and destination regions of China's water diversion project
.
Geophysical Research Letters
42
(
20
),
8424
8431
.
doi:10.1002/2015GL065904
.
Manshadi
H. R. D.
Niksokhan
M. H.
Ardestani
M.
2013
Water allocation in inter-basin water transfer with the virtual water approach
. In:
World Environmental & Water Resources Congress: Showcasing the future
,
American Society of Civil Engineers
,
Reston, VA
, pp.
2510
2521
.
doi:10.1061/ 9780784412947.247
.
Massey
F. J.
1951
The Kolmogorov-Smirnov test for goodness of fit
.
Journal of the American Statistical Association
46
(
253
),
68
78
.
doi:10.2307/2280095
.
Nelsen
R. B.
2006
An Introduction to Copulas
.
Springer
,
New York
. .
Plackett
R. L.
1965
A class of bivariate distributions
.
Journal of the American Statistical Association
60
(
310
),
516
.
doi:10.1080/01621459.1965.10480807
.
Salvadori
G.
De Michele
C.
2007
On the use of copulas in hydrology: theory and practice
.
Journal of Hydrologic Engineering
12
(
4
),
369
380
.
doi:10.1061/(ASCE) 1084-0699(2007)12:4(369)
.
Serinaldi
F.
Bonaccorso
B.
Cancelliere
A.
Grimaldi
S.
2009
Probabilistic characterization of drought properties through copulas
.
Physics & Chemistry of the Earth Parts A/B/C
34
(
10–12
),
596
605
.
doi:10.1016/j.pce.2008.09.004
.
Sklar
C. A.
1959
Fonctions de répartition à n dimensions et leurs marges
.
Publication de l'Institut de Statistique de l'Université de Pair
8
,
229
231
.
Sraj
M.
Bezak
N.
Brilly
M.
2015
Bivariate flood frequency analysis using the copula function: a case study of the Litija gauge on the Sava River
.
Hydrological Process
29
(
2
),
225
238
.
doi:10.1002/hyp.10145
.
Tian
H.
Wang
P.
Tansey
K.
Zhang
S.
Li
H.
2020
An IPSO-BP neural network for estimating wheat yield using two remotely sensed variables in the Guanzhong Plain, PR China
.
Computers and Electronics in Agriculture
169
,
105180
.
doi:10.1016/j.compag.2019.105180
.
Xie
H.
Wang
K.
2013
Joint-probability methods for precipitation and flood frequencies analysis
.
Third International Conference on Intelligent System Design & Engineering Applications
,
Los Alamitos, CA
.
doi:10.1109/ISDEA.2012.217
.
Zhang
L.
Singh
V. P.
2006
Bivariate flood frequency analysis using the copula method
.
Journal of Hydrologic Engineering
11
(
2
),
150
164
.
doi:10.1061/(ASCE)1084-0699(2006)11:2(150)
.
Zhang
L.
Singh
V. P.
2007
Gumbel-Hougaard copula for trivariate rainfall frequency analysis
.
Journal of Hydrology
12
(
4
),
409
419
.
doi:10.1061/(ASCE)1084-0699(2007)12:4(409)
.
Zhang
L.
Singh
V. P.
2019
Copulas and Their Applications in Water Resources Engineering
.
Cambridge University Press
,
Cambridge
,
UK
.
Zhuang
W.
2016
Eco-environmental impact of inter-basin water transfer projects: a review
.
Environmental Science & Pollution Research.
23
(
13
),
12867
12879
.
doi:10.1007/s11356-016-6854-3
.
Zou
L.
Xia
J.
She
D.
2018
Analysis of impacts of climate change and human activities on hydrological drought: a case study in the Wei River Basin, China
.
Water Resources Management
32
(
4
),
1421
1438
.
doi:10.1007/s11269-017-1877-1
.

Supplementary data