Changes in Streamflow Pattern and Complexity in the Whole Yangtze River Basin

Abstract

The assessment of streamflow patterns and their complexity variations across multiple timescales within river basins is a crucial aspect of water resource management and policy formulation. In this study, the Hurst coefficient, Mann–Kendall nonparametric test method, streamflow pattern indices, and sample entropy (SampEn) analyses were used to investigate the streamflow pattern in the whole Yangtze River basin at annual, monthly, and daily scales. The results show that with the increase in the time resolution, the streamflow shows more complex changing characteristics and streamflow changes more obviously on the monthly timescale than on the annual one. The annual mean streamflow decreases only in some of the tributaries, while the monthly streamflow shows significant increasing trends in the dry season and significant decreasing trends in the late wet season in almost the whole basin. Results also show that the minimum extreme streamflow indices increase in almost the whole basin. The maximum indices show decreasing trends in most of the tributaries and the Yichang gauge in the main reach. The streamflow complexity in the tributaries is higher and the complexity increases from upstream to downstream along the main reach in the basin. Along the main reach, the average SampEn increases downstream of the reach with values of 0.05, 0.07, 0.10, 0.12, 0.14, and 0.14 at Shigu, Pingshan, Cuntan, Yichang, Hankou, and Datong, respectively. These findings are helpful for understanding the hydrological characteristics and water resource management in the Yangtze River basin.

1. Introduction

Streamflow is a major path linking land and ocean for water, energy, and nutrients, which plays an important role in the natural water cycle and ecosystem health [1,2,3]. Streamflow changes could affect water resource availability and the biogeochemical processes in river systems [4]. The maintenance of environmental streamflow for natural water bodies is essential for sustainable ecosystem services [5]. Meanwhile, rivers are also important habitats for human survival and development. Hydrological regime changes could have significant effects on water supply, irrigation, flood control and electric power production [6,7]. Investigating streamflow index changes is especially important for understanding hydrological mechanisms and water resource management.

Many studies have shown that streamflow around the world has been affected both by climate change and human activities in the past and will be affected by them in the future, which has received much attention [8,9,10]. For example, Zhang et al. found that streamflow decreases mainly in northern China and the upper reaches of the Yangtze and the Pearl River basins and significant decreasing streamflow was found in the Yellow River, the Liaohe River, and the Haihe River basins in China [11]. The characteristics of streamflow, including changing trends, abrupt points, complexity, etc., are often diversified over different time scales [12,13,14], and are also different between upstream and downstream areas in a certain basin, especially for the large basins [15,16]. As far as we know, policy makers are continuously seeking to know more about the characteristics of streamflow index changes under changing environments for different management purposes. For example, water supply management departments focus on the spatiotemporal variations in water resources on monthly or annual scales, while emergency management departments urgently need to obtain the runoff changes and the regional flood composition over a short duration for flood control. Hence, it is necessary to find the potential hydrological characteristics and their spatiotemporal variations over different scales in a river basin.

The Yangtze River basin (YRB), the world’s third largest river basin, spans 19 provinces and includes the eastern, central, and western economic zones of China. Fluctuations in the streamflow of the Yangtze River have a profound impact on both the local socio-economic landscape and the ecological environment. For example, the severe drought that hit the middle and lower reaches of Yangtze in 2011 caused water levels to drop to historic lows, severely affecting agricultural irrigation, urban water supply, and navigation safety [17]. Understanding streamflow patterns at multiple timescales is essential for monitoring water resources and making adaptive management strategies to mitigate impacts on agriculture, socioeconomics, and nature. Previous studies showed that the streamflow along the main reach of the Yangtze River has experienced significant changes in recent decades due to environmental factors and human activities, indicating that the discharge series are non-stationary [18,19]. Jiang et al. [20] showed that streamflow increased in the headwater of the YRB, while it had little influence downstream. Gao et al. [21] indicated that flow regime changes vary in different seasons in the upper reaches of the Yangtze River; annual streamflow decreased in the period from 1961 to 2008, while autumn streamflow evidently decreased after the 1980s. Zhang et al. [22] showed a significant increase in streamflow in the middle of the Yangtze River by analyzing the streamflow indices at three stations, namely Yichang, Hankou, and Datong. Previous studies have also analyzed hydrological characteristics and their changes caused by the operation of the Three Gorges Dam in the lower reaches of the Yangtze River, showing that the water level and discharge significantly increase during most months from January to March and significantly decrease from August to November, and the date of the start of the dry season has been advanced in the lower reach [23,24]. Recently, some studies have also investigated the spatial distribution and dynamic changes in streamflow complexity in the middle and lower reaches of the Yangtze River using six gauges along the main reach showing an obvious spatial difference and an increasing trend [24]. However, according to our literature research, previous studies are mainly focused on the streamflow changes in certain subregions or the mainstream of the Yangtze River and analyses over multiple temporal scales are also scarce. The Yangtze River, China’s longest river, exhibits pronounced differences in terrain and climatic conditions across its upper, middle, and lower reaches. With an enormous annual flow volume of approximately 9600 × 108 m³, it contributes a significant 36% to China’s total river flow [25], underscoring its pivotal role in water resource management, agriculture, and flood control. Analyzing streamflow at daily, monthly, and annual scales provides a comprehensive view of temporal variations, encompassing extremes, intra-annual distribution patterns, and total water resource availability, respectively. This multi-scale investigation deepens our understanding of hydrological dynamics, facilitates adaptation to climate change, and informs precise, adaptive, and science-based strategies for water management. So, the comprehensive analysis of streamflow changes in the whole Yangtze River basin at different spatiotemporal scales is still required.

In order to better understand the characteristics of streamflow index changes in the YRB, a trend analysis, an abrupt analysis, and a complexity analysis are conducted for streamflow at different timescales from upstream to downstream in the YRB using the most recent dataset during 1960–2018. The main objectives of this study are as follows: (1) to investigate the annual and monthly water yield changes, (2) to quantify extreme changes in streamflow over the past year, and (3) to analyze the streamflow complexity across the whole Yangtze River basin.

2. Study Area and Data

2.1. Description of the Study Area

The Yangtze River basin (YRB), depicted in Figure 1, spans latitudes 25° N to 35° N and longitudes 91° E to 122° E, encompassing a vast drainage area of 1.80 million square kilometers, representing 18.8% of China’s total landmass. As China’s longest river at approximately 6300 km, the YRB stretches from the Qinghai–Tibet Plateau to the East China Sea, traversing a remarkable elevation drop of 6600 m. The basin’s geography is diverse, with the upper reaches extending from its source to Yichang, covering 4504 km and a drainage area of 1 million km², dominated by mountainous terrain. Conversely, the middle reaches, stretching from Yichang to Hukou, span 955 km and feature primarily fluvial plains, with a drainage area of 0.68 million km². The lower reaches, extending from Hukou to the river’s estuary, encompass 938 km and a drainage area of 0.12 million km². Climate-wise, the majority of the YRB experiences a subtropical monsoon climate, contributing to its status as a region of abundant precipitation. Specifically, the average annual precipitation is approximately 1126.7 mm [26], though there are notable variations. The source area receives roughly 400 mm of the annual precipitation, while most other regions within the basin experience between 800 and 1600 mm. This uneven spatial distribution of precipitation, decreasing from southeast to northwest, is attributed to the interplay of local circulation patterns and topography. The YRB’s complex terrain and climate and significant river flows require the study of river flows at daily, monthly, and annual scales. This research deepens the understanding of hydrological dynamics, supports adaptation to climate change, and guides precise, adaptive, science-based water management strategies.

2.2. Streamflow Data

Daily streamflow data from 1960 to 2018 from 12 hydrological stations—Gaochang, Beibei, Wulong, Chenglingji, Hukou, Huangjiagang, Shigu, Pingshan, Cuntan, Yichang, Hankou, and Datong (Figure 1)—covering the YRB were used in this study, and were obtained from the Hydrological Bureau of the Ministry of Water Resources and the Changjiang Water Resources Commission, China. The 12 gauges represent the streamflow characteristics of the mainstream and tributaries in the whole river basin. The details of each gauge, including the station name, river basin, and the catchment area, are shown in

Table 1. The hydrological stations used in this study.

Station ID	Name	River Basin	Catchment Area (104 km2)	Time Series
1	Gaochang	Minjiang River basin	13.54	1960~2018
2	Beibei	Jialingjiang River basin	15.61	1960~2018
3	Wulong	Wujiang River basin	8.3	1960~2018
4	Chenglingji	Dongting Lake basin	26.28	1960~2018
5	Hukou	Poyang Lake basin	16.22	1960~2018
6	Huangjiagang	Hanjiang River basin	9.52	1960~2018
7	Shigu	Upper Jinshajiang River basin	21.42	1960~2018
8	Pingshan	Jinshajiang River basin	48.51	1960~2018
9	Cuntan	Yangtze mainstream	86.66	1960~2018
10	Yichang	Yangtze mainstream	100.55	1960~2018
11	Hankou	Yangtze mainstream	148.80	1960~2018
12	Datong	Yangtze mainstream	170.54	1960~2018

.

3. Methods

3.1. Alteration Change Method

In this study, the Hurst coefficient method was firstly used to judge whether a series alteration existed and the degree of the alteration. If there was an alteration, the Mann–Kendall test, moving T-test method, and the Mann–Kendall–Sneyers test were used to perform a detailed analysis on trends and abrupt points in the sequence.

The Hurst coefficient method [27,28] can be used to determine if there is an alteration and the alteration degree. According to the value of H, we can observe three cases. When H equals 0.5, it means that this is a random series. When H is greater or less than 0.5, it indicates a future trend that is the same or opposite to the present trend, showing whether there is a positive continuous effect or reverse continuous effect. Moreover, the more H deviates from 0.5 indicates a larger alteration degree. In this study, the rescaled range analysis (R/S analysis) method, which is commonly used, was selected to calculate the Hurst coefficient H by following the steps below.

Considering a time series {X(t)}, t = 1, 2, …, for any positive integer t ≥ 1, the mean series is defined as

$${\overline{X}}_l={\displaystyle \frac{1}{\tau }}{\displaystyle \sum _{i=1}^{\tau }X\left(t\right)},\tau =1,2,\cdots \cdots ,n$$

(1)

The cumulative deviation ς(t) is given as

$$\varsigma \left(t,\tau \right)={\displaystyle \sum _{u=1}^t\left[X\left(u\right)-{\overline{X}}_{\tau }\right]},\hspace{1em}1\le t\le \tau$$

(2)

The range R is calculated via

$$R\left(\tau \right)=\underset{1\le t\le \tau }{\max }\varsigma \left(t,\tau \right)-\underset{1\le t\le \tau }{\min }\varsigma \left(t,\tau \right),\hspace{1em}\tau =1,2,\cdots ,n$$

(3)

The standard deviation S is

$$S\left(\tau \right)={\left\{{\displaystyle \frac{1}{\tau }}{\displaystyle \sum _{t=1}^{\tau }{\left[X\left(t\right)-{\overline{X}}_{\tau }\right]}^2}\right\}}^{{\displaystyle \frac{1}{2}}},\hspace{1em}\tau =1,2,\cdots ,n$$

(4)

For a given sequence,

$$R\left(\tau \right)/S\left(\tau \right)={\left(c\tau \right)}^H,\hspace{1em}\ln \left[R\left(\tau \right)/S\left(\tau \right)\right]=H\left(\ln c+\ln \tau \right)$$

(5)

According to the observation results, the least squares method was used to calculate the parameter c and the Hurst coefficient H. The relationship between the fractional Brownian motion correlation function and the Hurst coefficient H is shown as follows.

$$C\left(t\right)=-{\displaystyle \frac{E\left[B_H\left(-t\right)B_H\left(t\right)\right]}{E{\left[B_H\left(t\right)\right]}^2}}=2^{2H-1}-1$$

(6)

The correlation function value C(t) is tested by using the statistical test method. Under the given significance level α, where C(t) is less than the critical value r_α, the long-term correlation is not evident. Where C(t) is larger than the critical value r_α, the sequence alteration is evident. The alteration and its degree are determined as shown in

Table 2. Classification of alteration degree.

Correlation Function C(t)	Hurst Coefficient H	Alteration Degree
0 ≤ C(t) < rα	0.5 ≤ H < hα	No alteration
rα ≤ C(t) < rβ	hα ≤ H < hβ	Weak alteration
rβ ≤ C(t) < 0.6	hβ ≤ H < 0.839	Medium alteration
0.6 ≤ C(t) < 0.8	0.839 ≤ H < 0.924	Strong alteration
0.8 ≤ C(t) ≤ 1.0	0.924 ≤ H ≤ 1.0	Huge alteration

, based on the Hurst coefficient of the hydrological sequence and its grade interval of alteration.

Due to the advantages without particular underlying distributions [29], the Mann–Kendall (M–K) nonparametric test method [30,31] has been widely used to detect hydrological changing trends. In our study, the autocorrelation of the selected streamflow sequences was eliminated via the pre-whitening method [32], as a higher autocorrelation may lead to a larger error in the M–K trend analysis [33]. A positive M–K statistics Z indicates an increasing trend, while a negative Z indicates a decreasing trend. More details of the M–K trend test can be found in Mann [30] and Kendall’s [31] work.

The abrupt changes in streamflow sequences were detected using the moving t-test method and the Mann–Kendall–Sneyers test [34] in this study. Within the t-test, the original series was divided into two subsequences; if there is an obvious difference detected between the mean value of the two subsequences exceeding a certain significant level, it indicates that the time sequence has a point of alteration.

The non-parametric Mann–Kendall–Sneyers test is widely employed to identify abrupt changes in climatic factors and streamflow time series [35,36]. Given a time series of data represented by x₁, …, x_n, the methodology involved calculating t_k, the cumulative sum of m_i, where m_i is the cumulative count of instances where a later value in the series is greater than its preceding value.

$$t_k={\displaystyle {\sum }_{i=1}^k{m}_i}\hspace{1em}(2\le k\le n)$$

(7)

Under the assumption of random independence among the data points in the time series, the mean and variance of the statistic can be derived as specified in the relevant equations.

$$\overline{t_k}=E(t_k)=k(k-1)/4$$

(8)

$$Var(t_k)={\displaystyle \frac{k(k-1)(2k+5)}{72}}$$

(9)

To obtain the statistic U_BK, the streamflow data time series was rearranged in reverse order, and the calculations analogous to those in Equations (7)–(9) were repeated, with the initial condition UB1 set to 0. At a predetermined significance level, if the U_FK and U_BK curves intersected within the confidence interval, the point of intersection was designated as the theoretical abrupt change point.

$$U_{FK}={\displaystyle \frac{(t_k-\overline{t_k})}{\sqrt{Var(t_k)}}}$$

(10)

3.2. Streamflow Pattern Indices

Using the daily streamflow data spanning from 1960 to 2018, collected by 12 hydrological gauges distributed across the YRB, we calculated annual and monthly runoff series for each site to capture water yield variations across different time scales. Given the YRB’s propensity for flooding, as evidenced by the extreme floods recorded in 1954, 1998, 2010, and 2020, extreme floods remain a primary concern in our study. To comprehensively analyze these extreme flood events, we have selected several indicators that capture both the magnitude and duration of peak flows and low flows. Specifically, for extremely high streamflow, we have chosen the annual maximum daily streamflow (MAX1), annual maximum 3-day streamflow (MAX3), annual maximum 7-day streamflow (MAX7), and annual maximum 15-day streamflow (MAX15). These indicators have been chosen for two primary reasons: Firstly, they effectively reflect the changing characteristics of flood peaks and flood volumes, providing crucial insights into the severity and duration of extreme flooding events. Secondly, they align with the data series commonly used in China for designing flood control measures and managing flood risks, thereby ensuring that our findings can inform practical strategies for flood prevention and mitigation. Additionally, given the significance of dry season streamflow for maintaining riverine ecosystems, we have also included indicators of extremely low flow: the annual minimum daily streamflow (MIN1), annual minimum 3-day streamflow (MIN3), annual minimum 7-day streamflow (MIN7), and annual minimum 15-day streamflow (MIN15).

3.3. Streamflow Complexity Indices

Streamflow complexity is defined as the variability and uncertainty of streamflow and could reflect the dynamic structure of streamflow. In this study, the sample entropy (SampEn) analysis, initially proposed by Richman and Moorman [37], was utilized to quantify the complexity of streamflow. SampEn, which is a modification of the approximate entropy, has been employed to measure the complexity of time series data, where larger entropy values indicate a higher degree of complexity. The method is advantageous in that it is less dependent on the length of the data and demonstrates relative consistency across a broader range of parameter values [37,38]. SampEn has been used to characterize the complexity of hydrological time series in several recent studies [39,40,41].

For a certain time series $\left\{x\left(1\right),x\left(2\right),\cdots ,x\left(N\right)\right\}$, SampEn is calculated as follows:

Create an m-dimension vector:

$$X_m\left(i\right)=\left\{x\left(i\right),x\left(i+1\right),\cdots ,x\left(i+m-1\right)\right\}, i=1,2,\cdots ,N-m+1$$

(11)

Define the distance between $X_m\left(i\right)$ and $X_m\left(j\right)$:

$$d\left[X_m\left(i\right),X_m\left(j\right)\right]=\max \left\{\left|x\left(i+k\right)-x\left(j+k\right)\right|\right\}, k=0,1,\cdots ,m-1$$

(12)

Given a tolerance criterion r, count the $d\left[X_m\left(i\right),X_m\left(j\right)\right]$ that are smaller than r. The function $C_i^m\left(r\right)$ and the average similarity $C^m\left(r\right)$ are defined as

$$C_i^m\left(r\right)=\left\{\mathrm{the}numberofd\left[X_m\left(i\right),X_m\left(j\right)\right]<r\right\}/\left(N-m\right)$$

(13)

$$C^m\left(r\right)={\displaystyle \sum C_i^m\left(r\right)}/\left(N-m+1\right)$$

(14)

where i = 1, 2, …, N − m + 1.

Change the dimension to m + 1 and repeat the above steps to calculate $C^{m+1}\left(r\right)$. The $SampEn\left(m,r,N\right)$ of a time series can be defined as follows:

$$SampEn\left(m,r,N\right){=-\ln [\mathrm{C}}^{m+1}\left(r\right){/\mathrm{C}}^m\left(r\right)]$$

(15)

SampEn, an index of complexity in time series analysis, employs the variables m (embedding dimension) and r (tolerance) [39] to quantify the conditional probability that two sequences, similar within r for at least m consecutive points, will remain similar at their (m + 1)th point. The choice of m determines the minimum sequence length for a similarity assessment, impacting sensitivity to pattern complexity. Meanwhile, r sets the maximum allowed difference for sequence similarity, balancing between capturing fine details and ensuring sufficient matches. SampEn excludes self-matches and considers only the first N − m vectors, facilitating the evaluation of short-term pattern predictability in time series dynamics.

4. Results

4.1. Changes in Inter-Annual and Intra-Annual Streamflow

Figure 2 shows the annual mean streamflow changes at the 12 gauges in the Yangtze River basin. The results indicate that the annual mean flow shows decreasing trends at most of the gauges including Gaochang, Beibei, Wulong, Pingshan, Cuntan, and Yichang in the upper stream and Chenglingji, Huangjiagang, and Hankou in the lower basin. The streamflow shows increasing trends only in Shigu in the upper stream and Hukou and Datong in the middle and lower reaches (

Table 3. Changing characteristics of annual mean streamflow.

Station ID	Name	Changing Trend (108 m3/y)	Zc	Significance Level 1	Hurst	Alteration Degree 2	Abrupt Point
1	Gaochang	−1.86	−2.18	**	0.75	++	1969
2	Beibei	−2.54	−1.66	*	0.71	+	1994
3	Wulong	−0.99	−1.20	-	0.68	+	2005
4	Chenglingji	−12.19	−2.71	***	0.73	++	1971
5	Hukou	4.75	1.31	-	0.64	-	-
6	Huangjiagang	−2.28	−1.70	*	0.73	++	1991
7	Shigu	0.28	0.58	-	0.70	+	1967
8	Pingshan	−0.31	0.16	-	0.66	-	-
9	Cuntan	−5.11	−1.43	-	0.68	+	1969
10	Yichang	−5.93	−1.62	-	0.65	-	-
11	Hankou	−3.10	−0.51	-	0.60	-	-
12	Datong	3.00	0.29	-	0.63	-	-

Notes: ¹ in this study, the significance level α is set to be 0.01, 0.05, and 0.1 and the corresponding values of Z_1−α/2 are 2.32, 1.96, and 1.64. ***, **, and * indicate significance levels of 0.01, 0.05, and 0.1, respectively; - means significance exceeds 0.1. ² ++ and + indicate medium alteration and weak alteration, respectively. - means no alteration according to the Hurst coefficient method.

).

Table 3 shows the results of the trend analysis based on the Mann–Kendall test. The results show that the changes in trends are significant at Gaochang (α = 0.05) in the Minjiang River, Beibei (α = 0.1) in the Jialingjiang River in the upper stream and Chenglingji (α = 0.01) in Dongting Lake and Huangjiagang (α = 0.1) in the Hanjiang River in the middle reach. However, there are no obvious changes in trends at Wulong in the Wujiang River, Hukou in Poyang Lake and the gauges along the main reach.

The abrupt point analysis is also shown in Table 3. From the results, we can see there are changes in Gaochang (abrupt point at 1969), Beibei (abrupt point at 1994), Wulong (abrupt point at 2005), Shigu (abrupt point at 1967), and Cuntan (abrupt point at 1969) in the upper stream and Chenglingji (abrupt point at 1971) and Huangjiagang (abrupt point at 1991) in the middle reach. The other gauges, including Hukou in Poyang Lake and Pingshan, Yichang, Hankou, and Datong along the main reach, indicate no obvious alterations according to the Hurst coefficient method.

Figure 3 and

Table 4. Abrupt detection of monthly mean streamflow in the Yangtze River basin.

Station ID	Name	Abrupt Detection	January	February	March	April	May	June	July	August	September	October	November	December
1	Gaochang	Alteration degree	+++	+++	++	-	-	+	++	-	++	++	++	++
		Abrupt point	2011	2011	2011	-	-	2006	1986	-	2005	1995	1970	2011
2	Beibei	Alteration degree	++	++	++	++	+	-	-	-	++	++	+	++
		Abrupt point	2004	2004	2004	1970	1969	-	-	-	1986	1976	1970	2011
3	Wulong	Alteration degree	++	++	++	-	++	+	+	+	++	+	-	+
		Abrupt point	2001	2001	2001	-	1978	2005	2004	1979	1990	1990	-	2011
4	Chenglingji	Alteration degree	++	++	+	-	++	-	+	-	++	++	++	+
		Abrupt point	1990	1982	1980	-	1979	-	1971	-	1970	1976	1971	1971
5	Hukou	Alteration degree	+	-	+	+	-	-	+	++	++	+	-	++
		Abrupt point	1983	-	1973	1973	-	-	1989	1992	1991	1980	-	1993
6	Huangjiagang	Alteration degree	++	++	+	+	++	+	+	+	-	++	++	++
		Abrupt point	1972	1971	1974	1970	1971	1995	1995	2014	-	1987	1986	1971
7	Shigu	Alteration degree	++	++	++	+	++	-	-	-	++	+	++	+
		Abrupt point	1980	1981	1981	1978	1967	-	-	-	1967	1985	1987	1980
8	Pingshan	Alteration degree	+++	+++	+++	++	-	-	-	-	-	-	-	-
		Abrupt point	2013	2013	2013	2013	-	-	-	-	-	-	-	-
9	Cuntan	Alteration degree	+++	+++	+++	-	-	-	-	-	++	++	-	++
		Abrupt point	2013	2013	2013	-	-	-	-	-	1991	1994	-	2014
10	Yichang	Alteration degree	+++	+++	+++	+	++	-	++	-	++	+++	-	++
		Abrupt point	2010	2010	2011	2013	2012	-	2001	-	2002	2002	-	2014
11	Hankou	Alteration degree	+++	+++	++	-	++	-	+	-	-	++	-	+
		Abrupt point	1995	1989	1989	-	1978	-	2001	-	-	1996	-	2013
12	Datong	Alteration degree	++	++	++	-	+	-	+	-	++	++	+	-
		Abrupt point	1995	1989	1989	-	1978	-	2004	-	2011	1997	1991	-

Notes: +++, ++, and + indicate strong, medium, and weak alterations, respectively. - means no alteration according to the Hurst coefficient method.

show the trends and abrupt point detection results of the monthly streamflow in the 12 gauges based on the Mann–Kendall test. In the upper reach, the Gaochang gauge in the Minjiang River shows significant increasing trends from January to April and significant decreasing trends from July to November. Table 4 shows that the changes are also moderate to significant from January to March, in June and July, and from September to December. The abrupt point analysis shows similar points in the dry season, while the changes are different in the flood season. The Beibei gauge representing the runoff changes in the Jialingjiang River shows increasing trends from January to March and decreasing trends in September and October (Figure 3). The abrupt point analysis shows that there was no change from June to August, while the changes are small to moderate in other months. The abrupt point analysis also shows similar points after 2000 in the dry season, while the changes occurred around 1970 and 1980 in the flood season. The Wulong gauge at the outlet of the Wujiang River shows significant increasing trends from January to March and decreasing trends in May, October, and November. The results of change detection show that there are no changes in April and November. The abrupt points are in 2001 from January to March and around 1980 (May and August), 1990 (September and October) and after 2004 (June, July, and December).

In the middle stream of the YRB, the data at the Chenglingji gauge in Dongting Lake indicate increasing trends from January to March and decreasing trends in April, May, and June until November. The results show that changes occur in most months except April, June, and August. The abrupt points in January, February, and March occur in 1990, 1982, and 1980, and the abrupt points in the remaining months are all in the 1970s. The Hukou gauge in Poyang Lake shows increasing trends in January to March, September, and October, while in the other months, there are no obvious trends. The abrupt point in 1973 in March and April and in the 1980s and 1990s during the other months exhibit changes. The data from the Huangjiagang gauge representing the streamflow of the Hanjiang river indicate significant decreasing trends in autumn exceeding the significance level of 0.1, while in the other months, there are no obvious trends. The changes are obvious throughout the year except for September, with an abrupt point in the 1970s in winter and spring and in the 1980s, 1990s and 2010s, in summer and autumn.

In the upper stream of the main reach, there are no obvious trends in most months except for in April at the Shigu gauge. The results show that the changes are clear in spring, autumn, and winter, and the abrupt points occur around the 1980s and 1970s. At the Pingshan gauge, the results show that streamflow increases significantly from January to May, while the trends are not obvious from June to December. The changes are significant from January to March and moderate in April, while in the other months, there are no alterations. The abrupt points in the four months are the same in 2013. At the Cuntan gauge, the streamflow increases significantly from January to April and decreases significantly from August to November. From the results of the abrupt point detection, the alterations occur mostly in the dry season, and the abrupt point is around 2013 from December to March and in early 1990 in autumn. At the Yichang gauge, the streamflow shows increasing trends from January to April and significant decreasing trends from September to November. The changes are clear in most months, and the abrupt points are after 2010 in spring and winter and around 2002 in autumn. The results show that streamflow increases significantly from January to March in the middle and lower reaches of the mainstream at the Hankou and Datong gauges. The streamflow decreases significantly from September to November at Hankou, while at the Datong gauge, the streamflow decreases only in October and November. The changes in the streamflow at the Hankou and Datong gauges are similar from January to August with the same abrupt point, while the changes are clear in October and December at the Hankou gauge and in September, October, and November at the Datong gauge.

4.2. Changes in Extreme Streamflow Indices

The minimum and maximum streamflow at 1 day, 3 days, 7 days, and 15 days were selected, and then, the changing characteristics were analyzed in the range of 1960–2018, as shown in Figure 4 and

Table 5. Abrupt detection of extreme streamflow in the Yangtze River basin.

Station ID	Name	Abrupt Detection	MIN1	MIN3	MIN7	MIN15	MAX1	MAX3	MAX7	MAX15
1	Gaochang	Alteration degree	++	++	+++	+++	++	++	++	++
		Abrupt point	2012	2012	2012	2011	2004	1998	1998	1996
2	Beibei	Alteration degree	+++	++	++	+	+	+	++	++
		Abrupt point	1991	1991	2008	2008	1990	1990	1990	1990
3	Wulong	Alteration degree	-	-	-	-	++	++	++	++
		Abrupt point					2004	2004	2004	2004
4	Chenglingji	Alteration degree	++	++	++	++	+	+	++	++
		Abrupt point	2007	2012	2012	2012	2000	2000	2000	2000
5	Hukou	Alteration degree	++	++	++	++	-	-	+	-
		Abrupt point	1992	1992	1992	1992			2000
6	Huangjiagang	Alteration degree	++	++	++	++	++	++	++	++
		Abrupt point	1974	1971	1971	1971	1985	1985	1985	1985
7	Shigu	Alteration degree	+	+	++	++	-	-	-	-
		Abrupt point	1988	1988	1988	1988
8	Pingshan	Alteration degree	+++	+++	+++	+++	-	-	-	-
		Abrupt point	2014	2014	2014	2014
9	Cuntan	Alteration degree	+++	+++	+++	+++	-	-	-	-
		Abrupt point	2000	2000	2000	2000
10	Yichang	Alteration degree	+++	+++	+++	+++	++	++	+	+
		Abrupt point	2009	2009	2009	2009	2006	2006	2006	2006
11	Hankou	Alteration degree	+++	+++	+++	+++	++	++	++	++
		Abrupt point	2000	2000	2000	2000	2006	2006	2006	2006
12	Datong	Alteration degree	+++	+++	+++	+++	++	++	++	++
		Abrupt point	2000	2000	2000	2000	2000	2000	2000	2000

Notes: +++, ++, and + indicate strong, medium, and weak alterations, respectively. - means no alteration according to the Hurst coefficient method.

. In the upper stream of the Yangtze River basin, the maximum streamflow indices at the Gaochang gauge show significant decreasing trends, while the minimum streamflow indices show no significant increasing trends. The abrupt point detection shows the minimum and maximum streamflow indices exhibit medium to strong changes, and the abrupt points are around 2010 and 2000, respectively. At the Beibei gauge, only the minimum 1-day and 3-day flows show significant decreasing trends. From the results of the abrupt point detection, the changes are clear for all the streamflow indices. The abrupt point is in 1991 for the 1-day and 3-day minimum streamflow and in 2008 for the 7-day and 15-day minimum flows. The abrupt point for the 1-day, 3-day, 7-day, and 15-day maximum streamflow is in 1990. For the Wulong station, the minimum streamflow indices all show significant increasing trends, while the maximum streamflow indices all show decreasing trends. The changes are clear for the maximum streamflow, and the abrupt point is in 2004, while the minimum streamflow shows no change.

In the middle stream of the Yangtze River basin, significant increasing and decreasing trends were detected for the minimum and maximum streamflow indices, respectively. The abrupt point detection shows that the extreme flow indices all change, and the abrupt points are in 2007 for the 1-day minimum flow, 2012 for the 3-day to 15-day minimum flows, and 2000 for the maximum flow indices. At the Hukou gauge, the minimum streamflow indices increase significantly, while the maximum flow indices show no obvious decreasing trends. The results also show that the minimum streamflow indices exhibit an abrupt change in 1992, while the change is only obvious for the 7-day maximum streamflow in 2000. For the Huangjiagang gauge, the annual minimum 1-day streamflow increases, and the maximum streamflow indices all show significantly decreasing trends. The abrupt point detection results show that the minimum streamflow indices clearly change around early 1970, and the maximum streamflow indices change in 1985.

In the mainstream of the Yangtze river, the Shigu and Pingshan gauges in the upper reach show no change in trends for extreme flows except for the 15-day minimum flow at Pingshan, while the changes are clear for the minimum streamflow indices, with the abrupt point in 1988 and 2014 for the two gauges. At the Cuntan gauge, the minimum streamflow for 1 day, 3 days, 7 days, and 15 days increases significantly, and changes are also significant, with the abrupt point in 2000. However, the maximum streamflow indices show no obvious changes at the Cuntan gauge. In the middle and lower reaches of the mainstream, the results show that the minimum streamflow indices are all increasing at the Yichang, Hankou, and Datong gauges, and the changes are all very significant, with abrupt points in 2009, 2000, and 2000, respectively. For the maximum streamflow, the indices decrease significantly only at the Yichang gauge, while there are no obvious trends for the Hankou and Datong gauges. The results of the abrupt point detection show that all maximum streamflow indices at the Yichang, Hankou, and Datong gauges change, and the abrupt points are in 2006, 2006, and 2000, respectively.

To further investigate the changes in the extreme flows, the annual minimum and maximum streamflow before and after the abrupt points were analyzed, as shown in

Table 6. Changing characteristics of annual minimum and maximum streamflow.

Station ID	Name	Annual Minimum Flow					Annual Maximum Flow
		Before Abrupt		After Abrupt		Change Percentage	Before Abrupt		After Abrupt		Change Percentage
		Streamflow	Most Frequently Occurs	Streamflow	Most Frequently Occurs		Streamflow	Most Frequently Occurs	Streamflow	Most Frequently Occurs
1	Gaochang	574	2	744	5	29.6%	16,519	8	11,876	7	−28.1%
2	Beibei	327	2	256	2	−21.7%	24,677	7	19,458	7	−21.1%
3	Wulong	-	-	-	-	-	12,370	6	8137	6	−34.2%
4	Chenglingji	1408	1	1751	1	24.4%	28,298	7	24,126	7	−14.7%
5	Hukou	−3994	9	−1601	7	59.9%	15,995	6	13,806	6	−13.7%
6	Huangjiagang	203	2	399	12	96.3%	10,337	7	3996	7	−61.3%
7	Shigu	368	2	387	2	5.2%	-	-	-	-	-
8	Pingshan	1253	3	1694	1	35.2%	-	-	-	-	-
9	Cuntan	2653	3	3210	2	21.0%	-	-	-	-	-
10	Yichang	3436	2	5781	12	68.2%	49,161	7	37,662	7	−23.4%
11	Hankou	6525	2	9078	12	39.1%	54,767	7	48,300	7	−11.8%
12	Datong	8383	2	10,974	1	30.9%	60,518	7	55,042	7	−9.0%

. The results show that annual minimum streamflow increases in all the gauges except for Beibei in the Jialingjiang River. The biggest increase in the annual minimum streamflow occurs at the Huangjiagang gauge at about 96.3%, followed by the Yichang gauge at about 68.2%, and the Hukou gauge at about 59.5%. In the upper reach of the Yangtze River, the annual minimum streamflow increases downstream of the main reach from 5.2% to 35.2% but decreases at Cuntan, as indicated by the reduction at the Beibei gauge. In the middle and lower reaches, the annual minimum streamflow decreases downstream of the main reach from 68.2% to 30.9%. Meanwhile, the results also show that the monthly minimum streamflow occurs one or two months after it occurs in the main reach.

The results show the annual maximum streamflow decreases after the changing point. In the upper stream, it decreases by about 28.1%, 21.1%, and 34.2% at the Gaochang, Beibei, and Wulong gauges, respectively, while in the middle stream, it decreases by about 14.7%, 13.7%, and 61.3% at Chenglingji, Hukou, and Huangjiagang, respectively. In the middle and lower main reaches of the Yangtze River, the annual maximum streamflow decreases downstream of the main reach from 23.4% to 9.0%. The monthly maximum streamflow shows no obvious changes.

4.3. Streamflow Complexity Change Analysis

The changing characteristics of the sample entropy (SampEn) in the YRB are shown in

Table 7. Changing characteristics of the streamflow complexity.

Station ID	Name	SampEn	Zc	Significance Level 1	Hurst	Alteration Degree 2	Abrupt Point
1	Gaochang	0.17	2.38	***	0.84	+++	2010
2	Beibei	0.11	1.36	-	0.81	++	2013
3	Wulong	0.19	1.65	*	0.83	++	2010
4	Chenglingji	0.21	−0.50	-	0.59	-	-
5	Hukou	0.25	−1.56	-	0.60	-	-
6	Huangjiagang	0.17	−1.44	-	0.81	++	1998
7	Shigu	0.05	2.18	**	0.68	+	1974
8	Pingshan	0.07	1.73	*	0.82	++	2012
9	Cuntan	0.10	−1.11	-	0.69	+	1978
10	Yichang	0.12	0.13	-	0.58	-	-
11	Hankou	0.14	−2.83	***	0.74	++	1977
12	Datong	0.14	−0.63	-	0.63	-	-

Notes: ^1,2 the meanings are same as in Table 1. ***, **, and * indicate significance levels of 0.01, 0.05, and 0.1, respectively; - means significance exceeds 0.1. +++, ++ and + indicate strong alteration, medium alteration and weak alteration, respectively. - means no alteration according to the Hurst coefficient method.

. In the upper stream, the average SampEn is 0.17, 0.11, and 0.19 at Gaochang, Beibei, and Wulong, respectively. In the middle stream of the basin, the SampEn is higher, with values of 0.21, 0.25, and 0.17 at Chenglingji, Hukou, and Huangjiagang, respectively. Along the main reach, the average SampEn increases downstream of the reach, with values of 0.05, 0.07, 0.10, 0.12, 0.14, and 0.14 at Shigu, Pingshan, Cuntan, Yichang, Hankou, and Datong, respectively.

Figure 5 and Table 7 illustrate that the SampEn shows increasing trends at Gaochang, Beibei, and Wulong in the upper stream of the basin, and the trends are obvious at Gaochang and Wulong according to the Mann–Kendall test. However, the SampEn decreases at Chenglingji, Hukou, and Huangjiagang in the middle stream of the basin. Along the main reach of the Yangtze River, the SampEn shows obvious increasing trends at Shigu and Pinshan in the upper stream, while there are no obvious decreasing or increasing trends at the Cuntan and Yichang gauges. In the middle and lower reaches, the SampEn decreases significantly at Hankou, and the decreasing trend is not obvious at Datong.

The abrupt point detection results in Table 7 show that changes are moderate to significant at most of the gauges, and they are not clear in Chenglingji, Hukou, Yichang, and Datong. In the upper stream, the changes are moderate at Beibei and Wulong and significant at Gaochang, with an abrupt point around 2010. In the middle stream of the basin, the changes are moderate at Huangjiagang, and the abrupt point is in 1998. Along the main reach of the Yangtze River, the changes are clear at Shigu (abrupt in 1974), Pingshan (abrupt in 2012), Cuntan (abrupt in 1978), and Hankou (abrupt in 1977).

5. Discussion

The spatiotemporal characteristics of the streamflow indices vary over different time scales from upstream to downstream in the Yangtze River basin from 1960 to 2018, as seen from the results above. On the annual time scale, the mean streamflow shows significant decreasing trends in only some of the tributaries of the Yangtze River such as the Minjiang River, Jialingjiang River, Dongting Lake, and Hanjiang River, and the change is also moderate only in the Minjiang River, Dongting Lake, and Hanjiang River. However, the annual mean streamflow shows no significant trends along the main reach from upstream to downstream. The annual streamflow changes are directly related to the precipitation and evapotranspiration changes. Previous studies have shown that precipitation increases mainly in the eastern Tibet Plateau and the middle and lower Yangtze River basin, like Poyang Lake and decreases in the region of the Minjiang River, Jialingjiang River, Wujiang River and Hanjiang River. For the whole basin, precipitation increased and then decreased from 1960 to 2015. Meanwhile, the evapotranspiration decreased slightly in the Yangtze River basin according to the previous research. A warming–wetting trend is detected in the southeastern and northwestern regions, while there is a warming–drying trend in the middle regions of the basin [42].

However, on the monthly time scale, the streamflow shows significant increasing trends in the dry season, especially from January to March, at most of the gauges in the whole basin, and the change detection results show similar characteristics at each gauge. In the late wet season, the streamflow shows significant decreasing trends at most of the gauges. Particularly, streamflow decreases significantly in the flood season from July to November. The streamflow is noted to have passively increased from January to March, a phenomenon that is likely attributed to the melting of snowpacks and the enhanced precipitation in upstream regions [43,44] driven by seasonal shifts in weather patterns. The passive increase in minimum streamflow may be a consequence of altered precipitation patterns, enhanced evaporation rates due to climate change, or the cumulative effect of water management strategies. Similarly, the decrease in maximum streamflow from July to September could be attributed to changes in runoff generation mechanisms, dam operations, and water diversions [45]. A possible reason for this is the large number of reservoirs in the basin used to store water in the wet season and subsequently supply water in the dry season, which reduces the streamflow in the wet season and increases it in the dry season. Another reason may be the decreasing precipitation in the wet season and increasing precipitation in the dry season. Jiang et al. [46] found significant changes in many monthly precipitation datasets collected between 1961 and 2000. Gemmer et al. [47] reported that a statistically significant negative trend can be found in September over the Yangtze River basin, especially in the middle region. Along the main reach, the streamflow decreases in the flood season from August to November at the Cuntan gauge, indicating a decreasing inflow at the Three Gorges Dam; however, the streamflow decreases only in September, October, and November in the Yichang gauge due to regulation by the Three Gorges Dam. Additionally, the reduction degree reduces at Hankou and Datong in the middle and lower reaches. The abrupt change points of the monthly streamflow are similar in the dry season, while they are more complex in the wet season, which indicates the more complex effects of climate change and human activities on the streamflow in the wet season.

Over the daily time scale, the minimum extreme streamflow indices show increasing trends in most of the tributaries, while the maximum extreme streamflow indices show decreasing trends in the Minjiang River, Wujiang River, Dongting Lake and Hanjiang River. The results also show that the minimum streamflow occurs one or two months in advance along the main reach, while the maximum streamflow mainly occurs in June and July. As we know, the changes in extreme streamflow can be related to reservoir regulation and extreme precipitation. Guan et al. [48] showed that maximum 1-day and 5-day precipitation and maximum consecutive dry days have all increased significantly. Li et al. [49] showed that the extreme precipitation intensities and the frequency of extreme precipitation have significantly increased. Lü et al. [50] reported that 73% of the stations exhibited rising trends in annual maximum 1-day precipitation and the June–July–August seasonal maximum 1-day precipitation, with the most pronounced increases in the Yangtze River basin. From these previous studies, we can infer that the extreme streamflow changes are directly caused by reservoir regulation, which increases the ecological environmental flow in the dry season. It should be noted that the minimum 1-day and 3-day streamflow indices show decreasing trends at the Beibei gauge in the Jialingjiang River. Along the main reach of the Yangtze River, the minimum extreme streamflow indices show increasing trends from upstream to downstream, while the maximum extreme streamflow indices show significant decreasing trends only in the Yichang gauge due to the Three Gorges Dam.

The sample entropy (SampEn) values, indicating the streamflow complexity in the river basin, show an increasing trend from upstream to downstream in the main reach, and the SampEn values in the tributaries are generally higher than those in the mainstream [51]. As we know, a higher SampEn value means a more complex river system. In the upper stream of the main reach, the streamflow complexity is lower due to the lower impact of human activities, climate, and underlying surface conditions. From upstream to downstream, as the runoff of the tributaries flows into the main reach, the complexity increases due to the combined effects of human activities like reservoir regulation, human consumption, land use change, and climate change [52]. The higher SampEn values in the tributaries indicate that the streamflow is more unpredictable and complex on the regional scale in the Yangtze River basin. The results also show that the sample entropy in the Dongting and Poyang lakes is higher than that at all the other sites, which may be due to the complex water exchange between the Yangtze River and the Dongting and Poyang lakes, determined by both the flow and water level of the Yangtze main reach and the two lakes [53]. The data show that the SampEn values increase in the upper stream and decrease in the lower stream of the basin both in the main reach and tributaries. A possible reason for this is the increasing anthropic effects caused by the construction of large-scale hydraulic projects [54].

Although the spatiotemporal characteristics of the streamflow indices across the Yangtze River basin are detected in this study, a more reasonable quantitative analysis of the contributions of climate change and human activities to changes in different streamflow indices over various time scales should be conducted in our following studies. For instance, an analysis of the effects of climate change and human activities on streamflow on a monthly scale could help improve the efficiency of water resource regulation, and a quantitative analysis of factors impacting the streamflow on a daily scale could be useful for flood control. Consequently, this could lead to a better understanding of the main influencing factors and hydrological process mechanisms, providing more scientific support for water resource management in the river basin.

6. Conclusions

In this study, changes in the streamflow indices have been investigated from upstream to downstream in the whole Yangtze River basin over different time scales. The changing characteristics of the annual and monthly extreme minimum and maximum streamflow and streamflow complexity have been studied. The annual mean streamflow shows significant decreasing trends in only some of the tributaries, while it shows no significant trends along the main reach from upstream to downstream. The monthly streamflow shows significant increasing trends in the dry season, especially from January to March, and significant decreasing trends in the late wet season at most of the gauges. The abrupt change points of the monthly streamflow are similar in the dry season, while they are more complex in the wet season in the basin. The minimum extreme streamflow indices show increasing trends almost over the whole basin, while the maximum extreme streamflow indices show decreasing trends in most of the tributaries and only at the Yichang gauge in the main reach. The streamflow changes are more complex in the tributaries than those in the main reach. In addition, the complexity increases from upstream to downstream along the main reach in the basin. These findings can help us to understand the hydrological characteristics in the Yangtze River basin.