Abstract

Understanding the connections between climate, anthropogenic impacts, and hydrology is fundamental for assessing future climate change. However, a comprehensive methodology is lacking to understand significant changes in the discharge regime and their causes. We propose an approach that links change point tests with hydrologic metrics applied to two Vietnamese catchments where both climatic and anthropogenic changes are observed. The change points in discharge series are revealed by six widely used change point tests. Then, 171 hydrologic metrics are investigated to evaluate all possible hydrological changes that occurred between the pre- and post-change point period. The tests showed sufficient capabilities to detect hydrological changes caused by precipitation alterations and damming. Linking the change point tests to the hydrological metrics had three benefits: (1) the significance of each detected change point was evaluated, (2) we found which test responds to which hydrologic metric, and (3) we were able to disentangle the hydrological impacts of the climatic and anthropogenic changes. Due to its objectivity, the presented method can improve the interpretation of anthropogenic changes and climate change impacts on the hydrological system.

INTRODUCTION

Long-term series of daily discharge measurements are the most frequently used data type in many aspects of hydrology, such as hydrological modeling and flood frequency analysis. A long-term data set enables both the precise description of past and current hydrological conditions, as well as predictions for future developments. A core assumption of this approach is that there are constant hydrological conditions within the whole data period, i.e., stationarity (Raff et al. 2009). However, this assumption is often not valid in reality, where some hydro-meteorological records may be exposed to abrupt changes caused by human intervention (e.g., land-use, water diversion, damming) or by a changing climate (Huang et al. 2016). Thus, for accurate estimations, it is required to check whether the assumption of stationarity is valid.

One way to investigate stationarity is an analysis of the change points that identify a sudden change in a data set. Several change point tests have been developed, using different methods for detection. Kundzewicz & Robson (2004) summarized that change point tests are based on different assumptions, including: (i) distribution of the data (normal or standard); (ii) number of abrupt changes (one or multiple); and (iii) location of the change point (known or unknown). A so-called ‘jump change’ can indicate a change in the mean, median, variance, or a combination of those. Different techniques may be required for different kinds of investigations. Thus, the detected change points can vary in time, and yield conflicting results when different tests are applied to the same series. This raises the question of which change point tests are appropriate and how many tests need to be compared to obtain reliable results. Therefore, a need has arisen for a careful discussion and comparison of these methods, along with recommendations concerning which procedure(s) are best to use in commonly encountered situations (Reeves et al. 2007).

Since there is no single, well-accepted statistical test for reliably identifying change points, the combination of multiple tests has become more common to check whether similar or different change points are detected by different tests. Examples can be found in Matoušková et al. (2011) and Sharma et al. (2016). These studies investigated changes in the mean and median in hydrologic time series, and consistently found that a single test only is not able to detect weak changes or changes that have not lasted long. Vezzoli et al. (2012) considered four different tests to detect breaks on annual minimum and maximum discharge, and concluded that the existence of shifts in the series can be rejected by a single test, but may not be rejected by the others. The results of these studies support the statement that a comparison is needed among multiple tests. Thus, the detection of change points remains challenging due to possibly contrasting change point detections by different tests, and a robust procedure to identify the most appropriate change point(s) based on the results of several change point tests is still unclear.

Change points may be presented within the observed data series due to different reasons. For example, Stolte & Herrington (1984) explained changes in a river regime by precipitation and evapotranspiration changes, as well as shifts in land use and agricultural practices in a Canadian catchment. Furthermore, the impact on the river regime through hydropower (Xuefei et al. 2010), land-use change, and climate variability (Mwangi et al. 2016) was deduced from detected change points in discharge and water level records. To understand changes in catchments, knowledge about the reason for the change is required. However, many situations exist where the reasons for the existence of a change point is unclear (Lund & Reeves 2002), and little attention has been given to comparing methods to understand changes in hydrological time series.

To overcome this issue, hydrologic metrics are often used to characterize the hydrological regime which is based upon five fundamental properties (Richter et al. 1996) and, more recently, in hydrological modeling (Kiesel et al. 2017). Hydrologic metrics (also termed indices or statistics) are numbers that characterize statistical properties of the long-term hydrologic regime of rivers (Kennard et al. 2010). The first consistent set of hydrologic indices, the Indicators of Hydrologic Alteration (IHA), were defined by Richter et al. (1996) and include 32 hydrologic indices which are categorized into five groups: magnitude, duration, timing, frequency, and rate of changes. The IHA variables are considered ecologically relevant and can be calculated from historical flow data (Richter et al. 1996; Poff et al. 1997). Thus, they can be used to depict hydrological change by contrasting flow conditions of the unimpaired and impaired records. Costigan & Daniels (2012) studied the changes in the median values of the IHA parameters between pre-dam and post-dam periods in nine large rivers, and found that both the magnitude and duration decreased similarly through the system, while the number of annual hydrograph reversals (change from increasing to decreasing flow and vice versa) increased significantly, and rise and fall rates became faster. Zhang et al. (2016) assessed changes in IHA using the range of variability approach (RVA) to evaluate the hydrologic alteration caused by impoundment. Huang et al. (2016) are the first to couple the change point method and IHA to analyze multiple changes in hydrologic regimes of large rivers. The study of Huang et al. (2016) focused on one single test that detected multiple change points and quantified the change of individual indicators in general, but did not attempt to investigate whether different tests respond to different IHA. Therefore, the hypothesis that differences in the identified year of change as detected by several change point tests leads to differences in the most-changed indicators or indicator groups has not been addressed yet.

In this study, we undertake a comprehensive method to detect and understand hydrological changes, in consideration of both change point tests and hydrologic metrics. The main objective of this paper is to unbiasedly and reliably detect and understand the most significant hydrologic shifts occurring in the discharge regime of an anthropogenically influenced river basin. First, different change point tests are applied to evaluate if they provide different results. Second, the whole set of 171 hydrologic metrics is used to assess changes in pre- and post-change point time series. This makes it possible to examine the link between particular tests and the hydrologic metrics, and to find which metrics are detectable by the particular test. Third, observed changes caused by both climatic and anthropogenic impacts are interpreted using the results of the applied methodology. By combining change point tests with hydrologic metrics, we analyze how hydrological change can be interpreted and assessed in a more understandable manner.

STUDY AREA AND DATA

Study area

The Vu Gia-Thu Bon catchment (Figure 1) is the largest inland river basin in central Vietnam, with an area of 10.350 km2. The river system is formed by two main rivers named Vu Gia and Thu Bon (VGTB), which originate in the high mountains on the eastern side of Truong Son range and drain to the North Pacific Ocean in the Han and Dai River mouths. These two rivers are connected by a natural channel, Quang Hue, in the downstream region, and through the reservoir cascade Dakmi4 in the upstream region (see Figure 1). As located in south of the Bach Ma hill range and east of the West Truong Son mountain range, the climate in this catchment has similar characteristics to the South-Central Vietnamese climate, with wet-dry tropical climate with a strong seasonality. The very high rainfall during the rainy season (September–December) comprises approximately 65% to 83% of the total annual rainfall (2,500–4,200 mm), and causes regular river flooding in the downstream, delayed by about one month. The biggest floods are recorded in October and November, which are due to the most frequent and heaviest rainfalls or typhoons, comprising 65–85% of the total annual flow. The dry season lasts for the remaining months, with the lowest flows occurring in August (Mid-regional Center for Hydro-Meteorology in Vietnam 2012).

Figure 1

Vu Gia – Thu Bon catchment (left) and sub-basins up to Thanh My and Nong Son (right). The hydropower plants in this catchment are labeled by the year of their operation. The dashed line shows water diversion from Vu Gia to Thu Bon through the step-reservoir Dakmi4.

Figure 1

Vu Gia – Thu Bon catchment (left) and sub-basins up to Thanh My and Nong Son (right). The hydropower plants in this catchment are labeled by the year of their operation. The dashed line shows water diversion from Vu Gia to Thu Bon through the step-reservoir Dakmi4.

The main topographic characteristic of the catchment is the distinct differentiation into mountains and plains from up- to downstream. The rivers flow through many complex topographies, including the relatively narrow mountainous area with a maximum elevation of 2.598 m at Ngoc Linh mountain that features a large number of steep tributaries. The mountainous and hilly area accounts for over 60% of the total area with an average slope over 30%. The regions of interest of this study are the upper part of Vu Gia and Thu Bon (Figure 1), up to the gauges Thanh My and Nong Son, with a drainage area of 1.850 km2 and 3.155 km² respectively.

Observed changes in Vu Gia – Thu Bon catchment

Observed changes that have occurred in the catchment characteristics and management are important information for assessing how well the tests can detect these changes in the hydrological records and how and which metrics changed.

According to Souvignet et al. (2014), river discharge and rainfall in VGTB show an increasing trend since the 1980s, and trends in precipitation explain up to 74% of change in discharge. Therefore, to understand significant changes in precipitation, data from 1976 to 2015 at six rainfall gauges (Figure 1) were used to reveal any break points by applying the Pettitt test (Pettitt 1979) with a significance level of 5% (p-value). Change points in precipitation in both Thanh My and Nong Son climate stations occurred in 1995 and 1998 (Table 1).

Table 1

Change points in rainfall time series at climate stations in Vu Gia sub-basin (up to Thanh My) and Thu Bon (up to Nong Son)

Sub-basinsStationsp-valueChange point (date)
Vu Gia Kham Duc 9.33 × 10−47 30-07-1998 
Thanh My 3.49 × 10−19 07-05-1995 
Thu Bon Tien Phuoc 9.90 × 10−21 27-08-1998 
Tra My 3.86 × 10−19 23-08-1998 
Hiep Duc 4.35 × 10−08 04-09-1998 
Nong Son 1.06 × 10−16 08-05-1995 
Sub-basinsStationsp-valueChange point (date)
Vu Gia Kham Duc 9.33 × 10−47 30-07-1998 
Thanh My 3.49 × 10−19 07-05-1995 
Thu Bon Tien Phuoc 9.90 × 10−21 27-08-1998 
Tra My 3.86 × 10−19 23-08-1998 
Hiep Duc 4.35 × 10−08 04-09-1998 
Nong Son 1.06 × 10−16 08-05-1995 

Due to the topography and high discharges, the area is highly suitable for hydropower production. Within ten years, from 2005 to 2015, 24 hydropower plants were newly built in the whole catchment, and are able to store approximately 2.16 billion m3 of water. This is a typical example of the current situation in Vietnam and in South East Asia, where 61 gigawatts (GW) of new hydroelectric is planned to be constructed through 2020, in which Vietnam contributes up to 20 GW (Mayes 2015). Alterations in the flow regime are the major consequences and need to be well studied. In Vu Gia, the Dakmi4A plant is located about 50 km upstream of Thanh My and started operation in 2012. It is the most significant power plant in the area since the water released from its turbines is diverted to Thu Bon (through Dakmi4B, C, see Figure 1), upstream of Nong Son. Due to their location, the gauges are appropriate to investigate changes in the river regime of Vu Gia, both by damming and diversion. Therefore, daily discharge time series at both stations were gathered from the Mid-regional Center for Hydro-Meteorology in Vietnam for the period 1976–2015, in which gaps were filled and common errors (typing mistakes or errors due to technical issues) were corrected. The annual mean flow in Thanh My is 120.3 m3/s, while in Nong Son it is 283.5 m3/s. The highest flow event occurs in October and November with the historical peak in 10/2009 in Thanh My (4,540 m3/s) and 11/1998 Nong Son (8,920 m3/s), respectively.

The Dakmi4A dam impounds approximately 310 million m3, which accounts for 8.1% of the annual discharge volume at the downstream discharge gauge Thanh My. In Thu Bon, the first plant started operation in 2007, but by 2015, eight plants had been built, with a maximum retention volume of about 815 million m3, which amounts to 9.1% of the annual discharge volume at Nong Son. It is hypothesized that this causes a hydrologic regime differing significantly from the pre- and post-impoundment natural flow regime, especially downstream of Dakmi4A. Hence, the most significant impact through damming occurred in 2007 at Nong Son and in 2012 at Thanh My.

Over 80% of the study region is covered by natural mountainous areas and forests. Deforestation is negligible in the upstream part and occurred only for construction purposes, such as areas for reservoir storage, or new roads for hydropower plans, providing access to remote locations. Laux et al. (2016) pointed out that deforestation in the area caused no observable impacts to surface air temperature, and negligible impacts to the precipitation as well as the discharge regime. Therefore, we consider impacts of land use and land cover change on the hydrologic regime as negligible and focus on the two observed changes of precipitation and damming.

METHODOLOGY

Detecting change points using common change point tests

In this study, six change point tests are compared to investigate the stationarity of discharge series. Table 2 shows details about the employed tests used to detect changes in hydro-meteorological time series. Hereafter, we will refer to their abbreviations.

Table 2

List of applied change-point tests, test characteristics, and abbreviations

No.TestsAbbreviationParametricNon-parametricBased on
Change point in mean CPTM  Likelihood-based 
Break point BRPT  CE method 
Student t-test STUT  t-distribution 
Cumulative deviations CUMD  Rescaled 
Mann–Whitney test MWNT  Rank-based 
Pettitt test PETT  Rank-based 
No.TestsAbbreviationParametricNon-parametricBased on
Change point in mean CPTM  Likelihood-based 
Break point BRPT  CE method 
Student t-test STUT  t-distribution 
Cumulative deviations CUMD  Rescaled 
Mann–Whitney test MWNT  Rank-based 
Pettitt test PETT  Rank-based 

The present section establishes several definitions that appear in the following methodology. The sequence of data is defined as Xt = {x1, x2, … xn} (where X represents daily average discharge and n the number of time steps) and may have a change point at location T. The whole data set is then separated into two sub-series: a pre-change point segment Xi= {x1, x2, …, xT} and a post-change point segment Xj= {xT+1, xT+2, …, xn}, where i and j designates the pre- and post-change point period. The segments have distribution functions Fi(x) and Fj(x), respectively, and may differ in mean and/or variance.

To affirm the existence of location T, the null hypothesis of ‘no-change’ H0 and the alternative hypothesis of ‘change’ H1 are considered with a pre-assigned significance level (p-value) at 5%. Despite the fact that some of the tests summarized in Table 2 can detect more than one change point, here we only consider the most significant one.

There are two general approaches for checking whether an abrupt change occurs in a time series: parametric and non-parametric. Both of these are frequently used for hydrological data.

Parametric tests

Parametric tests have certain data requirements, such as data being normally distributed, independent, or homogeneous in variance. Therefore, data are tested with the Shapiro–Wilk normality test for normality, and chi-squared test for independence with a p-value at 5% significance level.

Change point in mean (CPTM): The tests use the ‘likelihood-ratio’ based approach to detect changes in mean, which was first introduced by Hinkley (1970) for normally distributed observations, then extended to detect changes in variances by Tang & Gupta (1987) under the same distributional assumption. The method considers the maximum log-likelihood in calculating both the null- and alternative hypotheses, over all possible change point locations. When the difference number between those is greater than a threshold, c, then the existence of a change point is true and the null hypothesis H0 is rejected. Information about appropriate values for c, as well as further information, can be found in Killick & Eckley (2013).

Break-point (BRPT): This is a multiple break-point detection test, which detects locations of change points using the cross-entropy (CE) algorithm combined with a modified Bayesian information criterion (mBIC). The observations are assumed to be normally distributed and continuous. The CE method turns the process of finding the abrupt change into a combinatorial optimization problem, which starts with defining a number of break-points and then subsequently generates a random sample from a four-parameter beta distribution. In each sample, the performance function mBIC is used to score this combinatorial arrangement. The process terminates when a stopping criterion (SC) is met, and the location T becomes a break-point if mBIC is optimal at that point. More detailed information can be found in Priyadarshana & Sofronov (2015).

Student t-test (STUT): Originated from t-test and followed by Student's t-distribution (Student 1908), a pseudonym of the statistician William Sealy Gosset, this is a standard parametric test for testing whether two samples have a difference in mean. The assumption is made that (a) the data follow a normal (Gaussian) distribution for the populations of the random errors and (b) there is no difference between the standard deviations of both population samples. At each location t, a computed value (Dn) is compared with a critical value (hn) and the null hypothesis H0 is rejected when Dn > hn.

Cumulative deviations (CUMD): The CUMD test (Buishand 1982) is first applied to examine whether records are homogeneous. The test calculates the difference between individual value and the mean value of the record. The CUMD from the mean or partial sums are calculated and then rescaled to make sure that the partial sums are not influenced by a linear transformation of the data. The test statistic Q is used as an indication for a change in level of significance. The change is affirmed if Q/ exceeds the critical value shown in Table 1 (Buishand 1982).

Non-parametric tests

Non-parametric tests have no particular data requirement, which only needs to be continuous. The two most widely used tests are the Mann–Whitney test (MWNT) and Pettitt test:

Mann–Whitney test (MWNT): The so-called Mann–Whitney-U or Wilcoxon–Mann–Whitney test was first proposed by Wilcoxon (1945), and further developed by Mann & Whitney (1947). The test evaluates differences of two independent sample groups, assuming that those sequences of data must be continuous. The test statistic Dij is evaluated at every time-step t by assigning a value of 1, 0 or −1 when Xi is greater, equal, or smaller than the subsequent value Xj, respectively. The significant level (p-value) is calculated at the time t at which Dij is largest, and then compared with the pre-assigned significant level (). The null hypothesis will be rejected when the p-value is less than , and the shift occurs at T.

Pettitt test (PETT): The non-parametric Pettitt test (Pettitt 1979) is based on the MWNT and can detect a single shift at a location T. The difference is that the MWNT assumes that the time of the change is known, while the Pettitt test detects changes at an unknown T.

The daily discharge series did not meet the requirements of normality and independence and were, hence, only used for the non-parametric tests (MWNT and PETT). The annual time series passed the Shapiro–Wilk test with a p-value of 0.56 and 0.2, respectively, and thus met the normality requirement. The chi-squared test evaluated for both stations' annual time series returned a p-value of 0 and confirmed their independence. Therefore, the annual discharge was used with the parametric tests (CPTM, BRPT, STUT, and CUMD). For all tests, a detected change point is considered if the pre- and post-periods differ at or below a significance level of 5%.

The tests were implemented using the following packages in the open source programing language R: ‘changepoint’ (Killick & Eckley 2013); ‘breakpoint’ (Priyadarshana & Sofronov 2015); ‘cpm’ (Ross 2015); ‘trend’ (Pohlert 2016); ‘climtrends’ (Gama 2016).

Linking hydrologic metrics to change point tests

The up-to-date set of 171 hydrologic metrics as listed by Olden & Poff (2003) are divided into five groups following the original set: Magnitude – Duration – Frequency – Timing – Rate of change in flow events as first suggested by Richter et al. (1996). We took the full set rather than previously published subsets for being able to evaluate all possible changes, and to examine the sensitivity of particular metrics to the tests. Each group is furthermore subdivided into Average-, High- and Low-flow conditions. We will henceforth refer to the combination of groups and flow conditions as ‘subgroups’. The abbreviation and number of metrics in each subgroup are shown in Table 3.

Table 3

List of subgroups, abbreviations, and number of metrics within each subgroup of hydrologic metrics

Groups
MagnitudeDurationFrequencyTimingRate
(MAG)(DUR)(FRE)(TIM)(RATE)
Flow conditions      
Average flow (A) MAGA, 45   TIMA, 3 RATE, 9 
High flow (H) MAGH, 27 DURH, 24 FREH, 11 TIMH, 3 
Low flow (L) MAGL, 22 DURL, 20 FREL, 3 TIML, 4 
Total 171 metrics 94 44 14 10 9 
Groups
MagnitudeDurationFrequencyTimingRate
(MAG)(DUR)(FRE)(TIM)(RATE)
Flow conditions      
Average flow (A) MAGA, 45   TIMA, 3 RATE, 9 
High flow (H) MAGH, 27 DURH, 24 FREH, 11 TIMH, 3 
Low flow (L) MAGL, 22 DURL, 20 FREL, 3 TIML, 4 
Total 171 metrics 94 44 14 10 9 
To evaluate changes of those metrics between the pre- and post-change point period, the daily discharge series at Thanh My and Nong Son were split at the change points detected by each test. The ‘EflowStats’ R-package (Henriksen et al. 2006) was used to calculate the hydrologic metrics for both periods. Percent-change for each metric was compared between both periods. For instance, naming a specific metric I and using the subscripts ‘pre’ and ‘post’ to indicate the periods before and after the change point, the change C, expressed in %, is written as Equation (1). The higher the percentage value presents the higher difference of the metrics due to the change point.  
formula
(1)
 
formula
(2)
To enable a better interpretation of the changes, we considered changes in the above-defined subgroups, intensity and variation of single metrics within the corresponding subgroup. At first, the subgroups were investigated by taking the median in C’ of all included single metrics. This is to identify the dominant subgroup at the change points regardless of the direction of change. Second, the link of the test to the hydrologic metrics was examined based on the average amount of change at each change point detected. The higher amount of change quantified, the more significant is the change point, and the stronger the test is linked to the metrics. Third, to evaluate whether the tests can detect changes in certain metrics, the ten most-changed metrics in each catchment over each change point were selected and it was checked whether they are simultaneously presented in both catchments. The ones which appear in the top ten list in both catchments and are linked to the same test will be considered as detectable by the corresponding test. Finally, the variability in single metrics within each group is considered by a simplified classification. We consider five classes of change: no- (0%), low- (0–25%), medium- (25–75%), high- (75–100%), and very high change (>100%) with its direction of change. The number of metrics in each change-class enables a better understanding of the variability in change.

Interpreting the observed changes in basins

To investigate the alteration in hydrologic metrics by the actual changes in the catchment, we related the major changed metrics to the dominant groups as explained in the section ‘Linking hydrologic metrics to change point tests’. We mainly focused on the alteration of the subset of 33 ecologically relevant metrics which describe the primary characteristic of the natural flow regime (Table 4 and Archfield et al. 2013), along with the most sensitive metrics that were identified in the section ‘Linking hydrologic metrics to change point tests’. The change C (%) was preferred in use to figure out the intensity and direction of changes. The strongly changed metrics, which occurred at change point in precipitation or dam operation, were attributed to that change. Having estimated the whole set of metrics, we further took the variability of change of the metrics into consideration. This enables a better understanding of the alteration of the observed changes in the discharge regime.

Table 4

Definition of 33 hydrologic metrics presented in Figure 4, obtained from Archfield et al. (2013) 

GroupMetricsDescriptionUnit
Magnitude of flow (MAG) ma5 Skewness in daily flows (mean divided by median daily flows) –** 
ma16 Mean monthly flow in May m³/s 
ma18 Mean monthly flow in July m³/s 
ma21 Mean monthly flow in October m³/s 
ma22 Mean monthly flow in November m³/s 
ma26 Variability in monthly flow in March – 
ma29 Variability in monthly flow in June – 
ma34 Variability in monthly flow in November – 
ma37 Variability across monthly flow – 
ma39 Coefficient of variation in mean monthly flows – 
mh14 Median of annual maximum flows – 
mh16 High flow discharge index – 
mh26 Mean of the high peak flow during the high flow event above a threshold (7 times median annual flow) divided by median annual daily flow – 
ml13 Variability across minimum monthly flows – 
ml14 Mean of the ratios of minimum annual flows to the median flow for each year – 
ml17 Base flow, defined as 7-day minimum flow divided by mean annual daily flows averaged across all years – 
ml18 Variability in base flow – 
Duration of low (DUR) dh5 Annual maximum 90-day average flow m³/s 
dh10 Variability in annual maximum 90-day average flow – 
dl1 Annual minimum 1-day average flow m³/s 
dl2 Annual minimum 3-day average flow m³/s 
dl5 Annual minimum 90-day average flow m³/s 
dl6 Variability in annual minima of 1-day mean of daily discharge – 
dl9 Variability in annual minima of 30-day mean of daily discharge – 
dl10 Variability in annual minima of 90-day mean of daily discharge – 
dl18 Mean annual number of days having zero daily flow 1/year 
Frequency of flow (FRE) fh2 Variability in average number of high flow* (high pulse count) – 
fh3 Average number of high flood events that above 3 times median annual flow 1/year 
fh4 Average number of high flood events that above 7 times median annual flow 1/year 
fl1 Average number of low flow* (low pulse count) 1/year 
fl2 Variability in average number of low flow* (low pulse count) – 
Rate of change (RATE) ra1 Mean of rise rate (positive changes in flow from one day to the next) m³/s 
ra4 Variability in mean of fall rate (negative changes in flow from one day to the next) – 
GroupMetricsDescriptionUnit
Magnitude of flow (MAG) ma5 Skewness in daily flows (mean divided by median daily flows) –** 
ma16 Mean monthly flow in May m³/s 
ma18 Mean monthly flow in July m³/s 
ma21 Mean monthly flow in October m³/s 
ma22 Mean monthly flow in November m³/s 
ma26 Variability in monthly flow in March – 
ma29 Variability in monthly flow in June – 
ma34 Variability in monthly flow in November – 
ma37 Variability across monthly flow – 
ma39 Coefficient of variation in mean monthly flows – 
mh14 Median of annual maximum flows – 
mh16 High flow discharge index – 
mh26 Mean of the high peak flow during the high flow event above a threshold (7 times median annual flow) divided by median annual daily flow – 
ml13 Variability across minimum monthly flows – 
ml14 Mean of the ratios of minimum annual flows to the median flow for each year – 
ml17 Base flow, defined as 7-day minimum flow divided by mean annual daily flows averaged across all years – 
ml18 Variability in base flow – 
Duration of low (DUR) dh5 Annual maximum 90-day average flow m³/s 
dh10 Variability in annual maximum 90-day average flow – 
dl1 Annual minimum 1-day average flow m³/s 
dl2 Annual minimum 3-day average flow m³/s 
dl5 Annual minimum 90-day average flow m³/s 
dl6 Variability in annual minima of 1-day mean of daily discharge – 
dl9 Variability in annual minima of 30-day mean of daily discharge – 
dl10 Variability in annual minima of 90-day mean of daily discharge – 
dl18 Mean annual number of days having zero daily flow 1/year 
Frequency of flow (FRE) fh2 Variability in average number of high flow* (high pulse count) – 
fh3 Average number of high flood events that above 3 times median annual flow 1/year 
fh4 Average number of high flood events that above 7 times median annual flow 1/year 
fl1 Average number of low flow* (low pulse count) 1/year 
fl2 Variability in average number of low flow* (low pulse count) – 
Rate of change (RATE) ra1 Mean of rise rate (positive changes in flow from one day to the next) m³/s 
ra4 Variability in mean of fall rate (negative changes in flow from one day to the next) – 

*Low-flow and high flows are defined as flow events below the 25th percentile and above the 75th percentile of the entire flow record.

** These metrics are dimensionless.

RESULTS

Different change points provided by the selected tests

Figure 2 illustrates the occurrences of the significant change points returned by the tests at Thanh My and Nong Son gauges. The values of the test variables mentioned in the section ‘Detecting change points using common change point tests’ are reported in Table S1 (in supplementary material).

Figure 2

Mean annual discharge time series in Thanh My and Nong Son and the occurrence of the detected change points. Dashed lines refer to change points detected by parametric tests on the annual time step, while the dotted lines refer to the non-parametric tests on daily time step.

Figure 2

Mean annual discharge time series in Thanh My and Nong Son and the occurrence of the detected change points. Dashed lines refer to change points detected by parametric tests on the annual time step, while the dotted lines refer to the non-parametric tests on daily time step.

In Thanh My, change in precipitation in 1995 was detected only by CUMD on annual basis. Another change point close to that is detected in 1996 by BRPT. The observed change caused by damming was detected by MWNT and PETT in the same year, but not at the same date. PETT returned a change point on 08/02/2012, closer to the time of initial dam operation than 28/02/2012 by MWNT. Further change points that are close to the time of damming in 2011 were detected by CPTM and STUT based on the annual time series.

By contrast, in Nong Son, the change in precipitation in 1995 is detected by a higher number of tests (CPTM, STUT, and CUMD) on annual basis. The abrupt change in rainfall series upstream of Nong Son in 1998 was found by PETT applied on the daily series. Similar to Thanh My, damming in 2007 was identified only by MWNT using daily discharge, but also not at the same date. To summarize, some change point tests applied on discharge succeeded in finding the year of change in both precipitation and damming.

Linking hydrologic metrics to the change point tests

Values of hydrologic metrics of the same pre- and post-change point periods are equal for the tests that returned the same change point. To better relate the change of hydrologic metrics over the change points to the observed changes, we add damming (DAM) and precipitation (PREC) as additional change points, presented alongside the other points given by the tests. Results from the 11 subgroups (Table 3) are presented first, followed by the analysis of individual metrics.

Quantifying changes in subgroups of hydrologic metrics in pre- and post-period

Figure 3 illustrates the changes in C’ (%) calculated by Equation (2) for each subgroup and each test at both stations. The last column and last row show the average change over the rows and columns. The alteration in hydrologic metrics over the change points in Thanh My is more significant than in Nong Son.

Figure 3

Changes in percentage of hydrologic metrics in subgroups (x-axis) by the change point tests and years of change (y-axis). The y-axis also includes the observed changes caused by precipitation (PREC) and damming (DAM). AVE shows the average change by each test (by row) and by each sub-group of metrics (by column).

Figure 3

Changes in percentage of hydrologic metrics in subgroups (x-axis) by the change point tests and years of change (y-axis). The y-axis also includes the observed changes caused by precipitation (PREC) and damming (DAM). AVE shows the average change by each test (by row) and by each sub-group of metrics (by column).

In Thanh My, the pattern of change in subgroups is split into two major sets. For the precipitation change points (PREC 1995 and BRPT 1996), the frequency of low flow (FREL) is the dominant subgroup. For the change points related to DAM in 2012, the magnitude of low flow (MAGL) is dominant with over 60% of change. Overall, the changes in MAGL due to damming are up to three times higher than those changes related to precipitation. The duration- and timing-related groups show less significant changes.

Similarly, the two identified clusters according to precipitation and damming are also shown in Nong Son. However, the hydrologic metrics respond differently to the tests as compared to Thanh My. The duration-related metrics are subject to the highest change which is mainly attributed to the precipitation-related change points detected in 1995, 1996, and 1998. There, 43% of change is observed in duration of low flow (DURL), followed by magnitude of low flow (MAGL) with an average of 23% change. For the change points in 2007, FREL is the dominant subgroup with over 30% of change. Regarding DURL, the impact by DAM is half as high as by PREC. Timing of flow shows no significant change. Hence, the major change of hydrologic metrics in Nong Son occurs in low flow conditions. The change points that can be attributed to precipitation mainly show changes in duration, while the change points caused by damming show changes in frequency and magnitude. Comparing both catchments, the change points in Thanh My are better explained by magnitude-related metrics for low flow, while the change points in Nong Son are strongly related to flow duration.

Significance of the change points confirmed by the link to the metrics

Figure 3 additionally shows the average amount of change at each detected point, thus, shows how strongly each test is linked to hydrologic metrics. Having quantified the largest amount of change, MWNT and PETT had the strongest link to the metrics. The test CUMD and BRPT had the weakest link since the detected change point is subject to the lowest change in both catchments.

Sensibility of the change point test to the most changed individual metrics

The intensity and directions of change of each single metric in both catchments and the identified most sensitive metrics are shown in Figure S1 (in supplementary material). The subset of 33 hydrological metrics and the sensitive metrics is presented in Figure 4. The tests and observed changes are shown alongside the corresponding year of change.

Figure 4

Variability in metrics in each group (x-axis) by year of change point, change point test, and observed change (y-axis). Definitions of metrics are listed in Tables 4 and 5. The levels of changes are: no- (0%), low- (0–25%), medium- (25–75%), high- (75–100%), and very high change (>100%). MAG, DUR, FRE, TIM, and RATE refer to the five categories magnitude, duration, frequency, timing, and rate of change, respectively. The y-axes are the percent changes. Each individual row represents year of change detected by the tests or observed changes.

Figure 4

Variability in metrics in each group (x-axis) by year of change point, change point test, and observed change (y-axis). Definitions of metrics are listed in Tables 4 and 5. The levels of changes are: no- (0%), low- (0–25%), medium- (25–75%), high- (75–100%), and very high change (>100%). MAG, DUR, FRE, TIM, and RATE refer to the five categories magnitude, duration, frequency, timing, and rate of change, respectively. The y-axes are the percent changes. Each individual row represents year of change detected by the tests or observed changes.

Table 5

Link between change point tests, most significantly changed metrics, and the related cause of the change

MetricsDescriptionsTestsObserved changes
mh3 Mean maximum flow in March MWNT and PETT DAM 
mh8 Mean maximum flow in August BRPT  
mh9 Mean maximum flow in September BRPT and CUMD PREC 
dh24 Flood-free days MWNT DAM 
dl5 Annual minimum 90-day average flow CPTM, STUT  
dl9 The variability in annual minima of 30-day mean of daily discharge CPTM, STUT, BRPT, CUMD PREC 
dl10 The variability in annual minima of 90-day mean of daily discharge CPTM, STUT, BRPT, CUMD PREC 
th2 Variability in Julian date of annual maximum MWNT DAM 
MetricsDescriptionsTestsObserved changes
mh3 Mean maximum flow in March MWNT and PETT DAM 
mh8 Mean maximum flow in August BRPT  
mh9 Mean maximum flow in September BRPT and CUMD PREC 
dh24 Flood-free days MWNT DAM 
dl5 Annual minimum 90-day average flow CPTM, STUT  
dl9 The variability in annual minima of 30-day mean of daily discharge CPTM, STUT, BRPT, CUMD PREC 
dl10 The variability in annual minima of 90-day mean of daily discharge CPTM, STUT, BRPT, CUMD PREC 
th2 Variability in Julian date of annual maximum MWNT DAM 

In Thanh My, in the magnitude-related subgroup (MAG), two groups can be roughly distinguished. For the change points in 1995 and 1996, few metrics are in the very high change class. Different metrics show very high changes for the remaining change points. This clearly shows that the alteration in these metrics is linked to the change points, and that the major changed magnitude metrics are more related to damming. Variability in monthly flow in March (ma26) is subject to the highest amount of change in 2012 and is the most impacted index by damming. In the duration group (DUR), more metrics are in the very high change level for the first three change points, which are mainly related to low flows. For the last three points, this is observed only for the metric variability in annual maximum 90-day average flow (dh10). CPTM and STUT detected a change point in 2011, which is close to damming (2012). However, the pattern of most strongly changed single metrics is more related to precipitation (1995, 1996). This shows that, the change point in 2011 is different from 2012 due to the change in the duration metrics. The variability in annual minima of 90-day mean of daily discharge (dl10) is the most changed index for these years. In Nong Son, the highest changed metrics are relatively homogeneous across the change point years. The highest changes are apparent in DUR, similar to Thanh My, especially for the dl10 metric.

Although variability in average number of low flow events (fl2) is subject to very high change only in year of change in precipitation (1995, 1996) in Thanh My, this is not observed in Nong Son. Thus, the relevance to this climate factor is uncertain. For timing-related metrics, variability in Julian date of annual maximum (th2) is significantly impacted by damming in both catchments. Hence, this metric can be linked to damming. Variability in reversals (ra9) appears consistently in the ‘Rate of change’ category (RATE) throughout the years and in both catchments. That means, ra9 is the most changeable metrics in general, and cannot be attributed to a particular cause of the change.

The strongest linkages between the change point test and the individual hydrological metrics are listed in Table 5. The links are selected by comparing the simultaneous presence of the ten most changed metrics in each catchment for each change point shown in Figure S1. There, the most important metrics impacted by DAM are detected by the non-parametric tests, while the metrics impacted by PREC are detected by parametric ones. Since variability in reversals (ra9) cannot be linked to a certain cause and test, it is not listed in Table 5.

Understanding the actual changes based on intensity and direction of changes in hydrologic metrics

Figure 4 additionally shows directions of change resulting from change in precipitation and dam operation in hydrologic metrics. The change in precipitation leads to an increase in the majority of metrics, while damming causes different impacts in both catchments.

The change in precipitation in 1995 strongly impacts the DURL through increases in the variability of annual minima 1-/3-/7-/30-/90-day means (dl6–dl10). Annual minima 90-day mean of daily discharge (dl5) and its variability (dl10) is significantly increased. In addition, maximum monthly flows in August and September (mh8, mh9) rise. Noticeably, variability in annual low pulse (fl2) significantly increases in Thanh My and explains the dominant group FRE in this basin. This metric is however absent in Nong Son.

Dams cause most significant changes in March and April in the hydrological regime. In Thanh My, discharges decrease by about 50% during these months, while their variability (ma26, ma27) increases. This causes the flows during the dry season to become less predictable. This is contrary to Nong Son, where an increase in flow in March and April is observed (ma14, ma15 in Figure S1). The variability in Julian date of annual maximum flow (th2) increases, which means the occurrence of extreme high flow in the impaired period is spread over a longer time period.

DISCUSSION

This study linked hydrologic metrics to change point tests in order to understand the reasons for the occurrence of hydrological change, and to assess if change point tests can be related to changes in hydrologic metrics. Based on the objectives and our results, we will here discuss the following points: difference in change points detected by the tests, benefit of linking change point tests and hydrologic metrics, and understanding of hydrological change due to the actual changes in the catchment.

Difference in change points detected by the tests

The long-term discharge series in Vu Gia – Thu Bon exhibits multiple abrupt changes as detected by the set of change point tests. Precipitation in this area was subject to significant changes in the 1990s, which was detected by the majority of tests applied on discharge series. The changes in discharge regimes are significantly impacted by the diversion dam in Thanh My, which explains why more tests detected changes close to the year of damming there. The parametric tests CUMD and BRPT provided consistent results in 1995 and 1996 in both stations, in agreement with Kundzewicz & Robson (2004), who stated that ‘the test is relatively powerful in comparison with other tests for detecting a change point that occurs towards the center of time series’. CPTM and STUT detected the change close to the year of damming in Thanh My, and a change point in coincidence with the precipitation change in Nong Son. Thus, both tests are able to detect changes caused by damming and precipitation. The non-parametric tests MWNT and PETT perform well in terms of detecting changes caused by damming. This is in disagreement with Machiwal & Jha (2006), who stated that non-parametric tests are less powerful than parametric ones.

Linking change point test and hydrologic metrics

The benefit of linking change point tests to the whole set of hydrologic metrics is, that (1) the significance of the detected change point can be evaluated and (2) it can be assessed how well the test is linked to the metrics. Linking tests and metrics revealed that the change point detected by one test (PETT in 1998, MWNT in 2007) shows changes in hydrologic metrics that are more important than changes detected by multiple tests. This seems to be counterintuitive, since it is generally assumed that change points detected by one test are less significant than changes detected by multiple tests (Kundzewicz & Robson 2004). In that regard, it is an important finding that the identified change points by multiple tests coinciding at the same time are statistically significant but can be less relevant in describing actual hydrological change than through a single test. This result highlights the importance of investigating detected change points regarding their actual impact on the hydrological regime, rather than relying on the statistical significance of the change point tests alone. The link between change points and hydrologic parameters further shows the relation of individual metrics to the certain change point tests, i.e., the metrics related to DURL are better linked to the parametric tests, while the magnitude of high monthly flow is linked to non-parametric ones. One reason for this result might be the different requirements of the tests to their input data, which led to annual and daily input time series, although further investigation is needed to draw a final conclusion.

Interpreting hydrological changes caused by climatic and anthropogenic impacts

Our study addresses the question whether the combination of both change point tests and hydrologic metrics can help to explain the actual change that occurred in the basins. Changes in hydrologic metrics by both climatic- and anthropogenic-related factors are clearly shown in this study.

The change in precipitation in 1995 results in similar discharge regime changes in the two catchments. Several studies have pointed out that precipitation pattern changes resulting from climate change are expected to increase drought intensity (see Wanders & Wada 2015). This is in line with our findings, since both the duration and variability of low flows increase in our study. The effects of precipitation changes on the hydrologic regime depend strongly on the location of the climate station. Tra My is located in the region of the highest rainfall in Vietnam, with an annual precipitation over 4,000 mm and a strong correlation to discharge in Nong Son (r = 0.91). This explains why the precipitation-related change in 1998 and 1995 results in similar hydrologic metrics.

Since the geographical and meteorological conditions in both catchments are similar, the reasons for the different change points in Thanh My and Nong Son are related to the anthropogenic impacts of dam constructions. The major change in Thanh My occurs in magnitude, and in Nong Son in the frequency of low flow. March and April are the periods most affected by damming, as agreed by Yan et al. (2010). We further found contradicting results in early dry season flow in the two below-dam gauges, a decrease in Thanh My and an increase in Nong Son. These differences are related to the design of reservoirs. The Dakmi4A in Thanh My has more than ten-fold the capacity of the first operated hydropower plant in Nong Son, Khedien (in 2007), 310 million m3 and 26 million m3, respectively. That is why the discharge in Thanh My is subject to more magnitude-related impacts, while the change in Nong Son is more frequency-related. The comparable plant in Nong Son is SongTranh 2 (733 million m3 storage), which started operation in 2010. However, there is no change point detected in that year, and the records of hydrologic metrics after 2010 show no significant alteration. We therefore agree with Gao et al. (2009) that storage ratio does not necessarily play a significant role in determining which metrics are most representative of flow regimes, but rather how the reservoirs are operated (Yang et al. 2008). This emphasizes the crucial role of the diversion hydropower plant, Dakmi4A, which is designed to divert water to downstream Dakmi4B and C, before releasing water to Thu Bon. With over 8% of the annual runoff diverted, this transfer apparently results in changes of magnitude and timing of extreme high flow and rate of changes in the discharge regime. Ordinary hydropower dams alleviate droughts, e.g., as observed in Nong Son, but due to the water transfer of Dakmi4A, drought events for this dam are aggravated downstream. Evidently, monthly flow in the early dry season in Thanh My significantly decreases due to the presence of the hydropower plant. Change of metrics in the post-dam change point period shows that, the occurrence of maximum flows is spread over a longer time period, which potentially decreases the predictability of floods. This link was also observed by Le et al. (2014). Additionally, regular dams cause an increase in the number of flow changes from fall to rise or vice versa (hydrograph reversals) (Costigan & Daniels 2012). However, the diversion dam results in less reversals, which means that flow is divided into distinct phases, strongly depending on the regulation of reservoirs.

The alteration in hydrological metrics is further impacted by the discharge record length, especially for those describing timing of flow events and variability in reversals (Kennard et al. 2010). In this study, the post-period after dam constructions, three and eight years in Thanh My and Nong Son, respectively, is much shorter than of changes in precipitation (20 years). Although the impacts of damming on hydrologic regime in this case is more visible and significant than of precipitation, the less accurate estimation of alteration by dams in some metrics is unavoidable and can only be mitigated by the availability of longer discharge time series after dam construction.

In this study, the impact of anthropogenic factors on the river regime was found to exceed climate-related factors. This is affirmed by the higher changes in hydrologic metrics resulting from dams than by precipitation in both catchments. This connection enables a clear attribution of changes to climate or dams, and hence represents a method for further study that could be extended to a larger scale, to investigate spatial and temporal patterns of change caused by both factors.

CONCLUSION

We presented a comprehensive analysis to understand the actual change of the hydrologic regime due to climatic and anthropogenic impacts. By considering six different change point tests, we found that the detected change points vary among the tests and catchments. However, the comparison of a high number of change points leads overall to an agreement of the major change in each catchment, which is dam operation in Thanh My and change in precipitation in Nong Son.

Linking change point detection tests to hydrologic metrics clearly showed the difference in intensity of change among the change points and the catchments, which helped to assess the significance of detected change points. The sensitivity of the tests to certain individual metrics showed that, variability in reversals is the most impacted metric in general. In addition, we were able to understand what detailed impact the climatic and anthropogenic changes had on the hydrological regime: The change driven by precipitation results mainly in increased variability of annual minimum 30-/90-day means of daily discharge. The appearance of dams in the two catchments causes different impacts on the hydrologic regime, and depends on the type of operation of the dams, e.g., the diversion of water from one catchment to the other causes different hydrological impacts than ordinary hydropower dams. In Vu Gia, the magnitude of monthly flows in the dry season decreases and less reversals are observed, in contrast to Thu Bon. However, results in both basins affirm that occurrence of extremely high flows become less predictable in post-impoundment, and that damming causes a more significant impact than the change in precipitation.

We can conclude that change point tests alone are not sufficiently reliable to detect hydrological changes, and they cannot supply information to understand the causes of change points. The combined analysis with hydrologic metrics showed the impacts on the hydrologic regime more clearly and improved the interpretability of the causes of change. Furthermore, we emphasize the applicability of this method to other regions, to larger scales, as well as in future climate change impact studies to examine the systematic change by anthropogenic impacts and climate change.

ACKNOWLEDGEMENTS

This work was carried out under the special program ‘Sustainable Water Management’ (NaWaM) from the German Academic Exchange Service (DAAD) with resources from the Federal Ministry of Education and Research (BMBF). The second author (JK) was funded through the ‘GLANCE’ project (Global change effects in river ecosystems; 01LN1320A) supported by the German Federal Ministry of Education and Research (BMBF). The third author (BG) was funded by DFG (Deutsche Forschungsgemeinschaft) via project GU 1466/1-1 (Hydrological Consistency in Modelling). For the use of the discharge data in the river basins, we thank the Mid-regional Center for Hydro-Meteorology in Vietnam. For these analyses, we used several R-packages and we would like to thank the writers and all the maintainers who are responsible for those. We further thank Dr. Georg Hörmann for his initial help with data analysis and Dr. Lisa Neef for her kind proofreading of the manuscript. We greatly appreciate the constructive comments of the anonymous reviewers who improved the quality of the manuscript.

SUPPLEMENTARY DATA

The Supplementary Data for this paper is available online at http://dx.doi.org/10.2166/wcc.2018.068.

REFERENCES

REFERENCES
Archfield
S. A.
,
Kennen
J. G.
,
Carlisle
D. M.
,
Wolock
D. M.
2013
An objective and parsimonious approach for classifying natural flow regimes at a continental scale
.
River Research and Applications
30
(
2
),
307
328
.
http://doi.org/10.1002/rra.2710
.
Costigan
K. H.
,
Daniels
M. D.
2012
Damming the prairie: human alteration of Great Plains river regimes
.
Journal of Hydrology
444–445
,
90
99
.
http://doi.org/10.1016/j.jhydrol.2012.04.008
.
Gama
J.
2016
climtrends: Statistical Methods for Climate Sciences
.
R package version 1.0.6. Retrieved from https://cran.r-project.org/package=climtrends
Gao
Y.
,
Vogel
R. M.
,
Kroll
C. N.
,
Poff
N. L.
,
Olden
J. D.
2009
Development of representative indicators of hydrologic alteration
.
Journal of Hydrology
374
(
1–2
),
136
147
.
http://doi.org/10.1016/j.jhydrol.2009.06.009
.
Henriksen
J. A.
,
Kennen
J. G.
,
Nieswand
S
.
2006
Users’ Manual for the Hydroecological Integrity Assessment Process Software (Including the New Jersey Assessment Tools): U.S. Geological Survey
.
Biological Resources Discipline, Open File Report 2006-1093
, p.
71
.
Hinkley
D. V.
1970
Inference about the change-point in a sequence of random variables
.
Biometrika
57
(
1
),
1
17
.
http://doi.org/10.1093/biomet/57.1.1
.
Huang
F.
,
Zhang
N.
,
Ma
X.
,
Zhao
D.
,
Guo
L.
,
Ren
L.
,
Wu
Y.
,
Xia
Z.
2016
Multiple changes in the hydrologic regime of the Yangtze river and the possible impact of reservoirs
.
Water (Switzerland)
8
(
9
),
1
20
.
http://doi.org/10.3390/w8090408
.
Kennard
M. J.
,
Mackay
S. J.
,
Pusey
B. J.
,
Olden
J. D.
,
Marsh
N.
2010
Quantifying uncertainty in estimation of hydrologic metrics for ecohydrological studies
.
River Research and Applications
26
,
137
156
.
http://doi.org/10.1002/rra.1249
.
Kiesel
J.
,
Guse
B.
,
Pfannerstill
M.
,
Kakouei
K.
,
Jähnig
S. C.
,
Fohrer
N.
2017
Improving hydrological model optimization for riverine species
.
Ecological Indicators
80
(
April
),
376
385
.
http://doi.org/10.1016/j.ecolind.2017.04.032
.
Killick
R.
,
Eckley
I.
2013
Changepoint: an R Package for changepoint analysis
.
Lancaster University
58
(
3
),
1
15
.
http://doi.org/10.1359/JBMR.0301229
.
Kundzewicz
Z. W.
,
Robson
A. J.
2004
Change detection in hydrological records – a review of the methodology
.
Hydrological Sciences Journal-Journal Des Sciences Hydrologiques
49
(
1
),
7
19
.
http://doi.org/10.1623/hysj.49.1.7.53993
.
Laux
P.
,
Nguyen
P. N. B.
,
Cullmann
J.
,
Van
T. P.
,
Kunstmann
H.
2016
How many RCM ensemble members provide confidence in the impact of land-use land cover change?
International Journal of Climatology
37
(
4
),
2080
2100
.
http://doi.org/10.1002/joc.4836
.
Le
A. T.
,
Dao
T. T.
,
Dang
N. V.
,
Pham
T. D. M.
,
Lam
T. T. S.
2014
Impacts of Operation in Flood Season and Storage Water Into Reservoir in Downstream of Vu Gia – Thu Bon River System: From Simulation to Reflection of the Local Residents
.
Technical report
.
http://doi.org/10.13140/2.1.1856.4805
Lund
R.
,
Reeves
J.
2002
Detection of undocumented changepoints: a revision of the two-phase regression model
.
Journal of Climate
15
(
17
),
2547
2554
.
http://doi.org/10.1175/1520-0442(2003)016<3383:CODOUC > 2.0.CO;2
.
Machiwal
D.
,
Jha
M. K.
2006
Time series analysis of hydrologic data for water resources planning and management: a review
.
Journal of Hydrology and Hydromechanics
54
(
3
),
237
257
.
Mann
H. B.
,
Whitney
D.
1947
On a test of whether one or two random variables is stochastically larger than the other
.
Annals of Mathematical Statistics
18
(
1
),
50
60
.
Matoušková
M.
,
Kliment
Z.
,
Ledvinka
O.
,
Královec
V.
2011
Application of selected statistical tests to detect changes in the rainfall and runoff regime
.
Die Bodenkultur
62
(
1–4
),
95
100
.
Mayes
F.
2015
Beyond China, Other Southeast Asian Countries Plan for Significant Hydroelectric Additions
.
EIA
.
Available online, 1–2. Retrieved from https://www.eia.gov/todayinenergy/detail.php?id=23552#
Mid-regional Center for Hydro-Meteorology in Vietnam
2012
Hydro-meteorological Characteristics in Quang Nam
.
Mwangi
H. M.
,
Julich
S.
,
Patil
S. D.
,
McDonald
M. A.
,
Feger
K. H.
2016
Relative contribution of land use change and climate variability on discharge of upper Mara River, Kenya
.
Journal of Hydrology: Regional Studies
5
,
244
260
.
http://doi.org/10.1016/j.ejrh.2015.12.059
.
Olden
J. D.
,
Poff
N. L.
2003
Redundancy and the choice of hydrologic indices for characterizing streamflow regimes
.
River Research and Applications
19
(
2
),
101
121
.
http://doi.org/10.1002/rra.700
.
Pettitt
A. N.
1979
A non-parametric approach to the change-point problem
.
Journal of the Royal Statistical Society
28
(
2
),
126
135
.
Poff
N. L.
,
Allan
J. D.
,
Bain
M. B.
,
Karr
J. R.
,
Prestegaard
K. L.
,
Richter
B. D.
,
Sparkes
R. E.
,
Stromberg
J. C.
1997
The natural flow regime
.
BioScience
47
(
11
),
769
784
.
http://doi.org/10.2307/1313099
.
Pohlert
T.
2016
Package ‘Trend’: Non-Parametric Trend Tests and Change-Point Detection
, p.
26
.
http://doi.org/10.13140/RG.2.1.2633.4243
.
Priyadarshana
W. J. R. M.
,
Sofronov
G.
2015
Multiple break-points detection in array CGH data via the cross-entropy method
.
IEEE/ACM Transactions on Computational Biology and Bioinformatics
12
(
2
),
487
498
.
http://doi.org/10.1109/TCBB.2014.2361639
.
Raff
D. A.
,
Pruitt
T.
,
Brekke
L. D.
2009
A framework for assessing flood frequency based on climate projection information
.
Hydrology and Earth System Sciences
13
(
11
),
2119
2136
.
http://doi.org/10.5194/hess-13-2119-2009
.
Reeves
J.
,
Chen
J.
,
Wang
X. L.
,
Lund
R.
,
Lu
Q.
2007
A review and comparison of changepoint detection techniques for climate data
.
Journal of Applied Meteorology and Climatology
46
,
900
915
.
http://doi.org/10.1175/JAM2493.1
.
Richter
B. D.
,
Baumgartner
J. V.
,
Powell
J.
,
Braun
D. P.
1996
A method for assessing hydrologic alteration within ecosystems
.
Conservation Biology
10
(
4
),
1163
1174
.
http://doi.org/10.2307/2387152
.
Ross
G. J.
2015
Parametric and nonparametric sequential change detection in R: the cpm package
.
Journal of Statistical Software
66
(
3
),
1
20
.
Sharma
S.
,
Swayne
D. A.
,
Obimbo
C.
2016
Trend analysis and change point techniques: a survey
.
Energy, Ecology and Environment
1
(
3
),
123
130
.
http://doi.org/10.1007/s40974-016-0011-1
Souvignet
M.
,
Laux
P.
,
Freer
J.
,
Cloke
H.
,
Thinh
D. Q.
,
Thuc
T.
,
Cullman
J.
,
Nauditt
A.
,
Flugel
W.-A.
,
Kunstmann
H.
,
Ribbe
L.
2014
Recent climatic trends and linkages to river discharge in Central Vietnam
.
Hydrological Processes
28
(
4
),
1587
1601
.
http://doi.org/10.1002/hyp.9693
.
Stolte
W. J.
,
Herrington
R.
1984
Changes in the hydrologic regime of the Battle River basin, Alberta, Canada
.
Journal of Hydrology
71
(
3–4
),
285
301
.
http://doi.org/10.1016/0022-1694(84)90102-1
.
Student
1908
The probable error of a mean
.
Biometrika
6
(
1
),
1
25
.
Tang
J.
,
Gupta
A. K.
1987
On testing homogeneity of variances for Gaussian models
.
Journal of Statistical Computation and Simulation
27
(
2
),
155
173
.
http://doi.org/10.1080/00949658708810988
.
Vezzoli
R.
,
Pecora
S.
,
Zenoni
E.
,
Tonelli
F.
2012
Data analysis to detect inhomogeneity, change points, trends in observations: an application to Po River discharge extremes
.
CMCC Research Papers
RP0138
,
1
15
.
Wanders
N.
,
Wada
Y.
2015
Human and climate impacts on the 21st century hydrological drought
.
Journal of Hydrology
526
,
208
220
.
http://doi.org/10.1016/j.jhydrol.2014.10.047
.
Wilcoxon
F.
1945
Individual comparisons by ranking methods
.
Biometrics Bulletin
1
(
6
),
80
83
.
Xuefei
M.
,
Van Gelder
P. H. A. J. M.
,
Vrijling
J. K.
2010
Hydrologic data analysis for pre-and post-dam construction in China
.
Water
2
.
Yan
Y.
,
Yang
Z.
,
Liu
Q.
,
Sun
T.
2010
Assessing effects of dam operation on flow regimes in the lower Yellow River
.
Procedia Environmental Sciences
2
(
5
),
507
516
.
http://doi.org/10.1016/j.proenv.2010.10.055
.
Yang
T.
,
Zhang
Q.
,
Chen
Y. D.
,
Tao
X.
,
Xu
C.-Y.
,
Chen
X.
2008
A spatial assessment of hydrologic alteration caused by dam construction in the middle and lower Yellow River, China
.
Hydrological Processes
22
,
3829
3843
.
http://doi.org/10.1002/hyp
.
Zhang
H.
,
Wang
B.
,
Lan
T.
,
Shi
J.
,
Lu
S.
2016
Change-point detection and variation assessment of the hydrologic regime of the Wenyu River
.
Toxicological & Environmental Chemistry
98
(
3–4
),
358
375
.
http://doi.org/10.1080/02772248.2015.1123480
.

Supplementary data