Spatiotemporal Trends of Precipitation and Natural Streamflow in the Upper Yangtze River Basin from 1951 to 2020

Abstract

The Yangtze River Basin is vital to China’s water security and flood management yet lacks a basin-wide quantitative assessment of long-term hydroclimatic changes. This study uses the high-resolution CMFD 2.0 dataset and the VIC model to evaluate spatiotemporal trends in precipitation and natural streamflow from 1951 to 2020. The results show a significant increase in annual precipitation in the upper basin (1.10 mm yr⁻¹, p < 0.05), particularly during the wet season, with spatially concentrated increases along the eastern Tibetan Plateau. The VIC model performed reliably across major stations, with NSE > 0.9 and PBIAS within ±10% during calibration. Natural streamflow trends are spatially heterogeneous: upper mainstream stations (e.g., Shigu, Panzhihua, Zhutuo) exhibit significant increases (6.25–14.58 m³/s per year), while lower stations remain stable or decline. Seasonally, wet-season streamflow increased in the upper basin, whereas dry-season streamflow decreased in the middle and lower reaches. At Yichang, natural simulations reveal growing seasonal extremes, with rising wet-season and declining dry-season flows (−19.06 m³/s yr⁻¹). Human interventions have partially offset these extremes. Since 1990, observed peak discharge at Yichang during the wet season has decreased by 10.04% compared to natural streamflow, while the dry-season minimum discharge has increased by 27.63%. This shows that large reservoirs help reduce flood peaks and increase low flows. These findings highlight the intensifying impacts of climate variability and human regulation on hydrological processes and provide a scientific basis for adaptive water resource management in large river basins.

1. Introduction

The Yangtze River Basin is the largest and most water-abundant river system in China, covering an area of over 1.8 million km². It plays a critical role in sustaining water security for nearly one-third of the country’s population and economic activities, particularly in central and eastern regions. However, the basin has long experienced frequent floods and droughts, posing persistent challenges for water resources management, flood control, and drought mitigation. Under the influence of global warming, the climate system has undergone significant changes, and the global hydrological cycle has become more intense [1,2]. This is manifested in the increase in atmospheric water vapor and the growing frequency of extreme climatic events. Consequently, regional precipitation and temperature patterns in the Yangtze River Basin have changed notably. These shifts, together with intensified human activities, have had profound effects on the hydrological regime of the basin. As a result, changes in water availability and aquatic ecosystems in the Yangtze River Basin have become a major concern among researchers in China and worldwide [3,4,5,6,7].

Previous research has suggested that annual precipitation across the Yangtze River Basin showed no significant long-term trend during 1955–2011. Nevertheless, precipitation was responsible for approximately 80 percent of the variability in streamflow [8]. This highlights that, despite increasing human influence, precipitation remains a key driver of streamflow change and a critical factor in flood and drought management within the basin. Numerous studies have investigated precipitation and extreme climate trends at various spatial and temporal scales. For instance, Jiang et al. [9] analyzed data from 147 stations between 1961 and 2000, revealing a significant increase in summer precipitation and in the frequency of heavy rainfall events. Similarly, Buda et al. [10] found, based on records from 51 stations (1951–2002), that summer precipitation had increased, rainfall in adjacent months had declined, and extreme events had become more frequent across most regions of the basin. Sun et al. [11] observed a declining trend in annual precipitation over the upper Yangtze Basin from 1961 to 2005, mainly attributed to reductions in winter and spring rainfall. In addition, several studies have identified the El Niño–Southern Oscillation as a key factor influencing precipitation variability in the region [12]. However, most of these analyses rely on station data, which are spatially limited, particularly in the mountainous headwater regions of the upper basin. Due to sparse station coverage and complex topography, existing studies have not adequately captured the pronounced spatial heterogeneity of precipitation in these critical upstream areas.

A series of major reservoirs and hydropower projects—such as the Three Gorges, Xiluodu, Xiangjiaba, and Baihetan—have been successively constructed in the Yangtze River Basin, significantly altering its natural streamflow regime. These large-scale engineering interventions, combined with the intensifying effects of climate change, have brought profound changes to the basin’s hydrological processes [3,4,13]. Previous studies have primarily focused on observed streamflow trends or on quantifying the impact of reservoir regulation by comparing regulated and historical streamflow [9,14,15]. For instance, research has indicated that flood-season streamflow in the middle and lower Yangtze increased significantly between 1961 and 2000, potentially due to changes in atmospheric circulation such as a weakening summer monsoon [9]. Further modeling-based analyses have shown that the Three Gorges Reservoir operation has reduced flood peaks and enhanced low flows during the dry season, altering the seasonal streamflow distribution by 41–61% [3,16]. Modeling studies have attempted to isolate the role of reservoir operations from climatic variability by reconstructing natural streamflow conditions [17,18,19]. Chai et al. [17] suggested that, in extremely dry years, climate variability still plays the dominant role in controlling streamflow patterns at key mainstream stations. Similarly, Zhang et al. [19] reconstructed natural streamflow at Yichang Station, which is located downstream of the Three Gorges Dam and serves as a key hydrological control station in the upper Yangtze River Basin, playing a crucial role in monitoring and regulating streamflow in this region. They found that reservoir regulation increased streamflow by 400–1900 m³/s from December to June and reduced it by 300–3400 m³/s from July to November. However, long-term spatiotemporal trends in natural streamflow under unregulated conditions—particularly across different parts of the upper Yangtze—remain insufficiently understood. Therefore, systematically assessing natural streamflow dynamics and separating the effects of climate change from those of reservoir regulation are crucial for advancing hydrological understanding and improving basin-scale water resources management.

This study focuses on the Yangtze River Basin, utilizing 3-hourly precipitation data and multi-site streamflow observations from 1951 to 2020 to analyze long-term hydrological changes. Two trend detection methods are applied to investigate the spatiotemporal evolution of precipitation and natural streamflow. The objectives are: (1) to characterize the interannual and seasonal trends in precipitation and their spatial distribution across the basin; (2) to simulate natural streamflow at major control stations upstream of Yichang and evaluate its trends at both annual and seasonal scales; and (3) to assess the impacts of human activities by comparing simulated natural streamflow with observed discharge. This comparison is used to quantify the influence of human interventions and to evaluate the reliability of the hydrological model. The study aims to close existing knowledge gaps regarding long-term natural streamflow variability in the Yangtze River Basin, and to provide scientific evidence and technical support for integrated water resources management, reservoir operation optimization, and regional water security policymaking. The novelty of this study lies in the use of long-term high-resolution CMFD v2.0 data, the integration of VIC model simulations with explicit consideration of reservoir regulation, and a comprehensive spatiotemporal assessment of precipitation and natural streamflow trends across the upper Yangtze Basin, which together provide a more refined understanding compared with previous studies.

2. Data and Methods

2.1. Study Area

The Yangtze River originates from the Geladandong Snow Mountain in the Tanggula Range on the Qinghai–Tibet Plateau and flows eastward for approximately 6300 km before emptying into the East China Sea. Its basin covers around 1.8 million km², representing nearly one-fifth of China’s total land area [8,17]. The Yangtze River Basin spans 19 provinces, autonomous regions, and municipalities, making it one of the most densely populated, economically dynamic, and land use diverse basins in the country.

The Yangtze River Basin is conventionally divided into three sections: the upper reaches (upstream of Yichang Station), the middle reaches (from Yichang to Hukou), and the lower reaches (from Hukou to the river mouth) (Figure 1). The upper basin is dominated by high plateaus and mountainous terrain, featuring pronounced wet and dry seasons, with streamflow primarily generated by high-altitude snowmelt and summer precipitation. The main channel above Yichang extends approximately 4504 km [17]. The middle reaches are characterized by relatively flat topography and strong hydraulic connectivity, where streamflow is heavily influenced by major tributary inflows. In contrast, the lower reaches consist of low-lying alluvial plains with dense river networks and are highly susceptible to flood hazards due to concentrated rainfall and tidal backwater effects. Major tributaries such as the Jialing, Min, Tuo, Wu, Han, Xiang, and Gan Rivers play essential roles in regulating and supplementing the mainstem, forming a critical component of the basin’s hydrological system.

The Yangtze River Basin spans multiple climatic zones, ranging from the Qinghai–Tibet Plateau to the central subtropics and southeastern coastal regions, and exhibits distinct gradients in temperature and precipitation. Annual precipitation is relatively low in the western plateau but increases substantially toward the southeast. The basin is strongly influenced by the East Asian monsoon, with the majority of annual rainfall concentrated in June to August. In this study, the wet season is defined as May to October, and the dry season as November to April of the following year [17].

2.2. Data Collection

The meteorological forcing data used in this study are derived from the China Meteorological Forcing Dataset Version 2.0 (CMFD 2.0), which provides the primary atmospheric input for hydrological simulations in the Yangtze River Basin. This dataset was jointly developed by the National Tibetan Plateau Data Center, the Institute of Tibetan Plateau Research, Chinese Academy of Sciences, and the research group led by Professor Yang Kun at the Department of Earth System Science, Tsinghua University [20]. CMFD 2.0 integrates ERA5 reanalysis data from the European Centre for Medium-Range Weather Forecasts with dense ground-based observations from the national meteorological network in China. It also incorporates artificial intelligence techniques, such as ISCCP-ITP-CNN and TPHiPr, to improve the accuracy of key variables including radiation and precipitation. Compared to its predecessor CMFD 1.6, which was based mainly on interpolation, CMFD 2.0 offers significant improvements in bias correction and trend fidelity, particularly in regions with complex terrain such as the Tibetan Plateau and the southwestern mountains. The dataset provides a spatial resolution of 0.1° × 0.1° and a temporal resolution of 3 h, making it highly suitable for hydrological modeling in mountainous basins.

Snow processes play a secondary but non-negligible role in the upper Yangtze Basin, particularly in the headwater regions of the Tibetan Plateau and western Sichuan, where seasonal snowmelt contributes to streamflow regulation. However, ground-based snow observations are sparse and unevenly distributed in these mountainous areas, which limits their direct application. Therefore, this study mainly relies on the CMFD v2.0 reanalysis product, which integrates station and satellite data to provide continuous coverage. While this improves the spatial representation of hydroclimatic variables, uncertainties in snow estimation remain and are further discussed in the simulation results.

To represent the spatial variability of natural streamflow, eight major hydrological control stations located on the upper mainstream and first-order tributaries of the Yangtze River were selected: Shigu, Panzhihua, Gaochang, Zhutuo, Beibei, Cuntan, Wulong, and Yichang. Trends in simulated annual natural streamflow at these stations were analyzed over the period 1951–2020. Although water regulation, irrigation, and hydropower infrastructure exist in the upper Yangtze, natural streamflow in this region is generally less influenced by anthropogenic activities compared to the middle and lower reaches, where complex systems of lakes, river–lake connections, and inter-basin transfer projects—such as the South-to-North Water Diversion Project—are present. Therefore, the upper basin offers a clearer representation of the hydrological response to climate forcing. In this study, “natural streamflow” refers to VIC-simulated discharge under unregulated conditions, while “observed streamflow” refers to gauged discharge influenced by both climate variability and human activities. Observed streamflow were obtained from the Yangtze River Hydrological Bureau, and all time series were standardized to daily values from 1 January 1951 to 31 December 2020. Details of the hydrological and water level stations are listed in

Table 1. Basic information of the hydrological stations involved in this study.

ID	Station Name	Location	Time Scale	Period	Drainage Area (km2)
1	Shigu	Yangtze River	Daily	1951–2020	212,883.4
2	Panzhihua	Yangtze River	Daily	1951–2020	256,952.7
3	Gaochang	Min River	Daily	1951–2020	134,792.3
4	Zhutuo	Yangtze River	Daily	1951–2020	667,232.4
5	Beibei	Jialing River	Daily	1951–2020	157,401.2
6	Cuntan	Yangtze River	Daily	1951–2020	839,325.1
7	Wulong	Wu River	Daily	1951–2020	83,212.7
8	Yichang	Yangtze River	Daily	1951–2020	980,426.3

.

2.3. Hydrological Model

The Variable Infiltration Capacity (VIC) model is a physically based, semi-distributed hydrological model widely applied in basin-scale water cycle simulation and climate impact studies [21,22]. The VIC model operates on a gridded structure, dividing the study basin into individual computational cells, and explicitly represents key hydrological processes including precipitation, infiltration, surface runoff, groundwater recharge, and evapotranspiration. It accounts for the spatial heterogeneity of land surface characteristics, including soil properties, vegetation types, and topography. A core feature of the model is its runoff generation scheme based on variable infiltration capacity, which enables realistic simulation of streamflow responses under varying land use and climate conditions.

In this study, the VIC model was employed to perform long-term simulations of natural streamflow across the upper Yangtze River Basin. The model was driven by high-resolution meteorological forcing data to evaluate hydrological responses to changes in precipitation. Specifically, the CMFD v2.0 dataset with a spatial resolution of 0.1° × 0.1° (≈10 km) and a temporal resolution of 3 h was used as input. The VIC model was implemented on a 0.125° grid with a 3-hourly time step for runoff generation. For river routing, the Muskingum routing module embedded in the Skyshare system (System for Hydrodynamic Assembling and Realtime Evaluation), developed by the China Institute of Water Resources and Hydropower Research, was applied with a 1-hourly time step. This module adopts a node–river network–sub-basin topology and allows for flood wave propagation simulation in complex branching river systems. In this work, it was used to transform grid-based runoff outputs into streamflow hydrographs at the selected control stations throughout the basin.

2.4. Trend Analysis

The Mann–Kendall (M–K test is a widely used non-parametric method for detecting monotonic trends in time series data without requiring any specific distributional assumptions [23,24]. It is particularly well-suited for hydroclimatic variables and has been recommended by the World Meteorological Organization (WMO) for identifying statistically significant increasing or decreasing trends [25]. In this study, the M–K test was applied to both precipitation and natural streamflow time series, with statistical significance evaluated at the 95% confidence level. The main steps are as follows:

The M–K test assumes that the observations (x₁, x₂, …, xₙ) are independent and identically distributed. The test statistic $S$ is defined as:

$$S=\sum _{k=1}^{n-1}\sum _{j=k+1}^n \mathrm{s}\mathrm{g}\mathrm{n}\left(x_j-x_k\right)$$

(1)

where S follows a normal distribution with a mean of zero and a variance of $Var\left(S\right)={\displaystyle \frac{n(n-1)(2n+5)}{18}}$;

The standardized test statistic $Z$ is calculated as:

$$Z=\left\{\begin{array}{c}{\displaystyle \frac{S-1}{\sqrt{Var\left(S\right)}}},S>0\\ 0,S=0\\ {\displaystyle \frac{S+1}{\sqrt{Var\left(S\right)}}},S<0\end{array}\right.$$

(2)

A positive $Z$ value indicates an upward trend, while a negative $Z$ suggests a downward trend. If $\left|Z\right|$ exceeds the critical value for a given significance level $\alpha$, the null hypothesis is rejected, indicating a statistically significant trend in the time series. To estimate the magnitude of the trend, Sen’s slope estimator was applied. This method is robust to outliers and computes the median of all possible pairwise slopes, representing the median rate of change per year [26]. While the M–K test is powerful for detecting the existence and direction of monotonic trends, it does not quantify their magnitude with confidence bounds. Therefore, a complementary parametric approach was also applied.

In addition to the non-parametric M–K test, this study also employed the parametric Ordinary Least Squares (OLS) linear regression to quantify trends in annual mean precipitation and natural streamflow. To address common issues such as residual autocorrelation and heteroscedasticity in hydrometeorological time series, the Newey–West Heteroskedasticity and Autocorrelation Consistent (HAC) estimator was used to adjust the standard errors of the OLS estimates [27]. By combining these two complementary methods—M–K and OLS—this study ensures robust detection of both the direction and magnitude of trends, while also mitigating the influence of extreme years and providing a more reliable basis for assessing hydroclimatic changes in the Yangtze River Basin.

2.5. Model Calibration

To ensure accurate simulation of hydrological processes, the Variable Infiltration Capacity (VIC) model requires parameter calibration tailored to the specific characteristics of the target basin. However, due to its large number of parameters and computational intensity, conventional brute-force or global optimization methods are often inefficient—particularly when performing joint calibration at multiple sites, which substantially increases the computational burden.

To address this challenge, a surrogate modeling approach was adopted in this study to calibrate the VIC model at eight representative hydrological control stations in the upper Yangtze River Basin. Surrogate models are statistical or machine learning approximations of high-cost objective functions. Unlike the VIC model, which operates explicitly on a gridded spatial structure, surrogate models do not have a physical spatial scale. Instead, they act as mathematical emulators that learn the relationship between input parameters (e.g., infiltration parameter, baseflow coefficient) and performance metrics such as NSE and PBIAS at hydrological stations [28,29]. In this study, the surrogate model was trained on a limited set of VIC simulations evaluated at eight representative stations and then used as a computationally efficient proxy during optimization. By iteratively refining the response surface through comparisons with additional VIC runs, the surrogate gradually improved its approximation of the optimal parameter space, thereby substantially reducing the computational burden of calibration without introducing additional spatial complexity. The Python Surrogate Optimization Toolbox (pySOT, version 0.3.3) under Python 3.9.18 was used to implement this framework.

To ensure basin-wide consistency, a hierarchical calibration strategy was employed, progressing from upstream to downstream stations based on hydrological control areas and spatial distribution. This approach ensured spatial coherence in model performance while maintaining computational efficiency. Calibration focused on both natural streamflow accuracy and regional-scale representativeness.

In the VIC model calibration process, several sensitive parameters were adjusted to optimize the simulation performance. The main parameters include the infiltration parameter (B_inf), baseflow parameters (D_s, D_m), soil moisture capacity parameter (W_s), and soil layer thicknesses (d₁, d₂, d₃). Their typical value ranges, physical meaning, and relevance are summarized in

Table 2. Key calibrated parameters of the VIC model, their physical meaning, and typical value ranges.

Parameter	Physical Meaning	Typical Range
Binf	Infiltration parameter controlling the amount of water that can infiltrate into the soil	0–10
Ds	the fraction of maximum base flow	0–1.0
Dm	the maximum velocity of base flow (mm/day)	0–30
Ws	the fraction of maximum soil moisture content of the third layer	0–1.0
d1	thickness of the second soil layer (mm), influencing root-zone storage	50–2000
d2	thickness of the second soil layer (mm), influencing root-zone storage	50–2000
d3	thickness of the third soil layer (mm), influencing baseflow storage and slow runoff	50–2000

. These parameters, respectively, control infiltration, baseflow generation, and soil water storage, and have been identified as the most sensitive for basin-scale calibration [21,30].

The simulation period was divided into three segments: 1980–2000 for model calibration, 1951–1979 for validation, and 2001–2020 for testing. This setup enabled evaluation under both relatively natural and regulated hydrological conditions. Model performance was assessed using the Nash–Sutcliffe Efficiency (NSE) and Percent Bias (PBIAS), with calibration aimed at maximizing NSE and minimizing PBIAS across all stations.

$$NSE=1-{\displaystyle \frac{{\sum }_{i=1}^N{(x_i-y_i)}^2}{{\sum }_{i=1}^N{(x_i-\overline{x})}^2}}$$

(3)

$$PBIAS={\displaystyle \frac{{\sum }_{i=1}^N(y_i-x_i)}{{\sum }_{i=1}^N{x}_i}}\times 100\%$$

(4)

where $x_i$ represents the observed value at time i, $y_i$ represents the predicted value at time i, and N represents the number of data points. $\overline{x}$ is the average value of the observation dataset. All values and performance metrics (NSE, PBIAS) are based on daily aggregated streamflow series.

3. Results

3.1. Spatial and Temporal Trends of Precipitation

3.1.1. Basin-Wide Trends in Annual and Seasonal Precipitation

To evaluate long-term changes in precipitation across the Yangtze River Basin, this study first conducted trend analyses on basin-averaged annual precipitation from 1951 to 2020 (Figure 2). The M–K test revealed no significant trend in annual precipitation over the full period, while the OLS regression suggested a weak but non-significant declining trend. This discrepancy may be attributed to the influence of extreme wet years during the early 1950s, especially in 1954 when a historic flood was recorded. However, precipitation records before 1955 are subject to larger uncertainties, since the national meteorological observation network was still in its early stage of development and many stations had incomplete or missing data [31]. In addition, the extreme precipitation of 1954 exerts a disproportionate influence on statistical tests, potentially masking the underlying long-term trends. Therefore, to ensure consistency and reliability, subsequent analyses in this study focus on the period 1955–2020. Results for this period showed a significant increasing trend using the OLS method (slope = 1.095 mm/yr, p < 0.05), while the M–K test still did not detect a statistically significant monotonic trend. Therefore, subsequent analyses in this study focus on the period 1955–2020.

Several extreme precipitation years were identified during the post-1955 period, including 1983, 1998, 2016, and 2020, when annual precipitation far exceeded the long-term mean. These years correspond with historically recorded flood events across the basin, which also serves to validate the consistency and reliability of the precipitation data used.

To further explore seasonal hydrological changes, this section analyzes trends in monthly mean precipitation for both the wet season (May–October) and dry season (November–April) across the basin. As shown in Figure 3, OLS regression indicates a significant increasing trend in wet-season precipitation, with a slope of 0.14 mm/yr (p = 0.024), while the dry-season trend is negligible and statistically insignificant (0.01 mm/yr, p = 0.664). The M–K test, however, fails to detect a significant monotonic trend for either season, suggesting that seasonal precipitation variability does not consistently increase or decrease over time within a nonparametric framework.

Extreme values further highlight the contrasting characteristics of wet and dry seasons. The wettest season occurred in 2020, with monthly mean precipitation approaching 175 mm, while the driest occurred in 1978, with a basin average of about 115 mm. For the dry season, 2016 saw the maximum monthly mean precipitation of over 60 mm, and the minimum dropped below 35 mm in 2011. Although dry-season precipitation exhibits considerable interannual variability, it lacks a sustained upward or downward trend. In contrast, wet-season precipitation demonstrates not only a statistically significant linear increase but also a growing frequency of extreme events in recent decades.

These results indicate that while basin-wide annual and dry-season precipitation show no statistically significant long-term trends, wet-season precipitation has increased notably since 1955—both in magnitude and frequency of extreme events—highlighting rising flood risks under intensified summer rainfall conditions.

3.1.2. Regional Differences Across Upper, Middle, and Lower Reaches

Averaging precipitation across the entire Yangtze River Basin may obscure localized variations, especially given the basin’s spatial heterogeneity. To address this, annual precipitation series were extracted for the upper, middle, and lower reaches of the basin and analyzed for trends and statistical significance over the period 1955–2020 (Figure 4). Results indicate that the upper basin has experienced a relatively stable but significantly increasing trend in annual precipitation, with a Sen’s slope of 0.864 mm/yr (p = 0.018) and an OLS slope of 1.065 mm/yr (p = 0.014), both statistically significant at the 95% confidence level. Despite modest interannual variability, the precipitation trend in the upper basin is robust and upward.

In contrast, the middle reaches exhibit higher annual precipitation levels (ranging from approximately 1100 to 1800 mm), but no statistically significant trends are detected using either method. The middle basin also shows the widest interannual variability, with values below 800 mm in dry years and exceeding 2000 mm in wet years. Although both Sen’s slope (2.138 mm/yr, p = 0.111) and the OLS slope (3.074 mm/yr, p = 0.053) are positive, they fall short of the 95% significance threshold, indicating that no consistent monotonic trend can be confirmed. The lower reaches also show no statistically meaningful changes in annual precipitation.

Figure 5 further examines seasonal precipitation trends across the three subregions. The upper basin again displays a statistically significant increase in wet-season precipitation, with a Sen’s slope of 0.17 mm/yr (p = 0.003) and an OLS slope of 0.21 mm/yr (p = 0.001). In the dry season, the upper basin shows a slight but significant decrease, with a Sen’s slope of –0.04 mm/yr (p = 0.035). The middle reaches show no significant trend in either season, reinforcing the conclusion of long-term stability. In the lower basin, dry-season precipitation displays a marginally significant upward trend, with an OLS slope of 0.16 mm/yr (p = 0.047), though no significant change is detected for the wet season.

This spatial pattern—marked by intensified wet-season precipitation in the upper basin, overall stability in the middle basin, and modest recovery of dry-season precipitation in the lower basin—has important implications for both flood and drought management. In particular, the confirmed increase in wet-season rainfall in the upper basin highlights the growing need for enhanced flood preparedness and infrastructure resilience in the headwater regions.

In summary, while the middle and lower reaches of the Yangtze River Basin exhibit little to no systematic changes in annual or seasonal precipitation, the upper basin stands out with a consistent and statistically significant increase—particularly during the wet season—raising concern over heightened flood risks in mountainous headwaters.

3.1.3. Spatial Distribution of Trend Slopes

Figure 6 illustrates the spatial patterns of annual precipitation trends across the Yangtze River Basin from 1955 to 2020. The Qinghai–Tibet Plateau region in the upper basin exhibits a notable upward trend, with most grid cells showing increased precipitation. However, localized areas—including parts of the Three Gorges Reservoir region and the upper Poyang Lake Basin—display significant downward trends. Across most other regions, trend magnitudes are small and generally fall within ±8 mm/yr. The spatial distributions derived from the M–K test and OLS regression are broadly consistent, though the OLS method identifies a slightly larger area of significant decline in the Poyang Lake region.

Figure 7 further depicts the spatial distribution of seasonal precipitation trends during the wet (May–October) and dry (November–April) seasons. A clear spatial heterogeneity is observed: wet-season precipitation shows a pattern of “increase in the west and east, and decrease in the central basin,” while the dry season exhibits a widespread but moderate decline.

In the wet season, the M–K test (Figure 7a) indicates significant increasing trends primarily concentrated in the mountainous regions of the upper basin, particularly along the western margin of the Sichuan Basin, the western Sichuan Plateau, and the Jinsha River Basin. In these areas, trend slopes typically exceed 0.8 mm/yr and locally surpass 1.4 mm/yr. In contrast, most parts of the middle and lower basin exhibit non-significant changes or slight declines. The OLS results (Figure 7c) reaffirm these findings, with a similar spatial pattern of significant increases in the upper basin.

During the dry season, although the magnitude of precipitation trends is generally lower, their spatial extent is broader. The M–K test (Figure 7b) shows that significant decreasing trends are widespread across the middle and lower reaches, particularly in the Jianghan Plain and Poyang Lake Plain. Most of these areas exhibit slopes ranging from −0.4 to −0.1 mm/yr. The OLS method (Figure 7d) confirms these trends and even indicates a wider spatial coverage of statistically significant change during the dry season than during the wet season. This suggests that, despite relatively minor absolute changes, the dry-season trend signal is statistically more robust, especially across the low hills and plains in the eastern basin.

In contrast, the area experiencing significantly increased wet-season precipitation remains limited, concentrated mostly in the southwestern highlands. Meanwhile, the spatial extent of dry-season decline covers a broader portion of the basin, indicating a systematic and widespread reduction in precipitation during this period.

Overall, the spatial analysis reveals that while wet-season precipitation increases are concentrated in the mountainous upper reaches—posing heightened flood risks—the dry season has experienced widespread and statistically significant declines across the eastern basin, amplifying concerns over water scarcity and drought vulnerability during the low-flow period.

3.2. Trends in Natural Streamflow

3.2.1. Evaluation of Hydrological Model Calibration

To reliably reproduce the long-term dynamics of natural streamflow in the upper Yangtze River Basin, the VIC model was employed to simulate streamflow from 1951 to 2020. Model calibration and validation were conducted for a series of representative control stations on both the mainstem and primary tributaries, such as Yichang, Beibei, Wulong, and Zhutuo. Model parameters were optimized individually for each station to enhance local accuracy and preserve spatial consistency across the basin. It should be noted that precipitation trend analysis in this study starts from 1955 to avoid uncertainties in early meteorological records and the disproportionate influence of the 1954 extreme flood year, whereas streamflow simulations start from 1951 to utilize relatively complete discharge observations for model validation.

Table 3. Performance evaluation of daily streamflow simulations at hydrological stations.

Station	Calibration Period (1980–2000)		Validation Period (1951–1979)		Test Period (2001–2020)
	NSE	PBIAS	NSE	PBIAS	NSE	PBIAS
Shigu	0.91	6.7%	0.78	−8.8%	0.89	5.8%
Panzhihua	0.92	4.9%	0.82	−10.9%	0.89	−0.3%
Gaochang	0.73	2.8%	0.69	−13.7%	0.66	8.8%
Zhutuo	0.91	0.7%	0.82	−12.4%	0.71	−3.8%
Beibei	0.79	−0.8%	0.74	−0.7%	0.73	−8.8%
Cuntan	0.93	0.9%	0.87	−10.9%	0.88	−3.6%
Wulong	0.81	6.0%	0.77	7.6%	0.41	2.1%
Yichang	0.94	0.8%	0.89	−6.3%	0.79	−1.4%

presents the performance metrics of the VIC model at each site, evaluated using NSE and PBIAS across three periods: calibration (1980–2000), validation (1951–1979), and testing (2001–2020). Results show strong simulation performance at most stations. NSE values during the calibration period exceeded 0.80 at nearly all sites, while PBIAS remained within ±5% in most cases. Mainstream stations exhibited particularly high accuracy, with calibration NSE values above 0.90. For example, Yichang station achieved an NSE of 0.94 and a PBIAS of 0.8% during the calibration period, and an NSE of 0.89 and PBIAS of –6.3% during validation, indicating strong model stability and transferability across time periods.

At mountain headwater stations such as Shigu and Panzhihua, where snowmelt and orographic effects dominate, the model also performed well, with calibration NSE values of 0.91 and 0.92, respectively. In tributary regions, performance was slightly lower but still satisfactory. For instance, Beibei and Wulong stations achieved calibration NSE values of 0.79 and 0.81, respectively.

The test period (2001–2020) reflects conditions under intensified human interventions, including major reservoir operations and land use changes. Despite these disturbances, the VIC model maintained acceptable performance. Most stations showed NSE values above 0.70 and PBIAS within ±10%, demonstrating that the model retained predictive capability even under anthropogenic influence. Notably, Figure 8 shows a clear increase in dry-season observed streamflow at Yichang station in recent years, illustrating the compensatory effect of the Three Gorges Reservoir during low-flow periods—a deviation well captured by the model.

In summary, the VIC model—combined with distributed parameter calibration and multi-period validation—achieved high simulation accuracy and generalizability across a wide range of hydrological and human-influenced conditions. This provides a robust foundation for subsequent trend analysis and future scenario simulations. It should be noted that the calibration was performed on natural streamflow, without explicitly accounting for reservoir regulation, irrigation, or urbanization. The potential implications of these anthropogenic factors, particularly reservoir operations and urbanization, are discussed later in Section 4.2 and Section 4.3.

3.2.2. Annual Scale Trends in Natural Streamflow

To reveal the overall response of natural streamflow to changes in precipitation, this section analyzes annual natural streamflow series at major mainstream control stations in the upper Yangtze from 1951 to 2020, evaluating their long-term trends and statistical significance. In general, annual natural streamflow in the upper Yangtze exhibits pronounced regional differences: stations upstream of Zhutuo predominantly show increasing trends, while downstream stations exhibit either decreasing trends or no significant change. This reflects varying sensitivities of different hydrological control zones to climate change and basin response mechanisms.

Figure 9 displays the temporal trends in annual natural streamflow at each station. The results show clear regional differentiation. Stations located upstream of Zhutuo—such as Shigu, Panzhihua, Gaochang, and Zhutuo itself—exhibit significant increasing trends. For instance, Sen’s slope estimates for Shigu and Panzhihua are 6.25 m³/s·yr⁻¹ and 6.23 m³/s·yr⁻¹, respectively, while Zhutuo shows the strongest increase at 14.58 m³/s·yr⁻¹. These results are consistent with the OLS regression analysis and suggest a persistent upward trend in natural streamflow in the uppermost mainstream reaches, likely driven by increased highland precipitation and enhanced snowmelt.

In contrast, downstream mainstream stations such as Cuntan and Yichang show no statistically significant trends, indicating a more stable flow regime. On the other hand, tributary stations, particularly Beibei and Wulong, display significant decreasing trends. At Beibei, the Sen’s slope is −9.09 m³/s·yr⁻¹, and at Wulong −4.10 m³/s·yr⁻¹, suggesting that tributary basins may be more sensitive to local climatic variability, land use change, or vegetation shifts.

Over the full 70-year period, most mainstream stations experienced reduced streamflow during the 1970s–1980s, followed by recovery and a notable rise since 2016. In particular, the high flows observed in 2020 were comparable to those in 1954, a historically extreme flood year. This resurgence is likely linked to more frequent and intense wet-season rainfall in recent years, reinforcing the potential for elevated future flood risk in the upper Yangtze.

A comparative analysis of interannual streamflow and precipitation trends further supports the dominant role of climatic forcing. In wet years such as 1954, 1998, and 2020, both precipitation and streamflow peaked basin-wide. Conversely, drought years—including 1959, 1972, 2006, and 2011—consistently corresponded to streamflow minima across nearly all stations. The close synchronization between precipitation and streamflow confirms that, in the upper Yangtze, natural streamflow is predominantly governed by basin-wide climatic variability rather than anthropogenic alterations.

3.2.3. Seasonal Scale Trends in Natural Streamflow

To evaluate seasonal dynamics in natural streamflow, this section analyzes the long-term trends in wet-season (May–October) and dry-season (November–April) flow at eight control stations across the upper Yangtze River Basin from 1951 to 2020.

Table 4. Seasonal trends in natural streamflow at upper Yangtze River stations.

Station	Wet Season				Dry Season
	M–K		OLS		M–K		OLS
	Slope	p	Slope	p	Slope	p	Slope	p
Shigu	11.55 *	0.00	11.18 *	0.00	0.44	0.35	0.35	0.66
Panzhihua	11.93 *	0.00	12.13 *	0.00	−0.12	0.77	−0.25	0.78
Gaochang	7.64	0.10	8.25	0.14	0.32	0.78	−0.34	0.75
Zhutuo	34.74 *	0.00	32.83 *	0.00	−3.16	0.24	−3.65	0.17
Beibei	−12.42 *	0.03	−14.80 *	0.02	−3.69 *	0.00	−5.79 *	0.00
Cuntan	18.01	0.19	18.04	0.25	−8.73 *	0.01	−9.61 *	0.01
Wulong	−2.94	0.43	−3.92	0.29	−5.20 *	0.00	−6.26 *	0.00
Yichang	13.35	0.44	11.42	0.55	−17.81 *	0.00	−19.06 *	0.00

* indicates statistical significance at the 95% confidence level.

shows that most mainstream stations exhibit significant increases in wet-season streamflow. Zhutuo, Panzhihua, and Shigu stations all passed both the M–K and OLS significance tests (p < 0.05), with OLS trend slopes reaching 32.83, 12.13, and 11.18 m³/s·yr⁻¹, respectively. These results indicate a consistent upward trend in high-flow season discharge across the upper mainstream, which aligns with observed increases in wet-season precipitation and may aggravate future flood risks.

By contrast, several tributary stations exhibit declining trends in wet-season streamflow. Most notably, Beibei on the Jialing River shows a significant downward trend of −14.8 m³/s·yr⁻¹ (p = 0.02), while Wulong on the Wu River also shows a decreasing but non-significant trend. These results suggest a weakening wet-season contribution from specific tributaries, potentially driven by altered precipitation patterns, land use changes, or catchment responses. No significant trends are observed at downstream mainstream stations such as Gaochang, Cuntan, and Yichang. This may indicate the dampening effect of regulated streamflow and flow routing below Zhutuo, where the hydrological signal of climate variability becomes increasingly obscured by human influence.

In contrast to the wet season, most stations downstream of Zhutuo show significant and persistent decreases in dry-season flow. Beibei (−5.79 m³/s·yr⁻¹), Wulong (−6.26 m³/s·yr⁻¹), Cuntan (−9.61 m³/s·yr⁻¹), and Yichang (−19.06 m³/s·yr⁻¹) all passed the MK and OLS significance thresholds (p < 0.01), suggesting a systematic reduction in baseflow during winter and spring. The decline at Yichang is particularly notable, highlighting potential threats to ecological flows and water supply reliability even under natural (unregulated) conditions. In contrast, upstream stations such as Shigu and Panzhihua show no significant trends in dry-season flow, implying greater hydrological stability in headwater areas likely maintained by snowmelt and baseflow buffering. However, it should be noted that the scarcity of ground-based snow observations in the headwater regions may introduce uncertainties in reproducing spring snowmelt contributions, although the overall seasonal streamflow trends remain consistent.

Figure 10 shows that wet-season streamflow exhibits more pronounced interannual fluctuations compared to the dry season, with low-flow episodes occurring basin-wide during the 1970s–1980s. In recent decades (2010–2020), wet-season flow has increased steadily, coinciding with more frequent extreme precipitation events under climate warming. In comparison, dry-season streamflow shows stronger persistence and synchronicity across stations, likely reflecting its sensitivity to long-term climatic drying and catchment desiccation.

In summary, seasonal streamflow in the upper Yangtze River Basin demonstrates marked spatial and temporal contrasts. While wet-season flow is increasing in upstream mainstream areas—raising flood management challenges—dry-season flow is declining in many downstream regions, posing risks for water supply and ecological health. These divergent trends underscore the need for differentiated water resource strategies that account for both climatic drivers and basin-scale hydrological heterogeneity.

4. Discussion

4.1. Reliability Analysis of Precipitation Data from 1951–1955

Trend analysis of basin-wide precipitation revealed anomalously high annual averages during the early 1950s (1951–1955), particularly in 1954. As discussed in Section 3.1.1, this period coincides with the early stage of China’s national meteorological network, raising concerns about data completeness and accuracy. To further evaluate their reliability, the present study assessed the usability of the CMFD v2.0 precipitation dataset from 1951 to 1955 by testing its performance in hydrological modeling. Specifically, streamflow was simulated using the VIC model at four major mainstream control stations in the upper Yangtze River, and the results were compared with observed streamflow records.

As shown in Figure 11, while the VIC model notably underestimated some peak flows at Shigu and Zhutuo in certain years, it captured the interannual variability and flow dynamics at downstream stations (Cuntan and Yichang) with high accuracy. One notable case is Shigu Station, where the VIC simulation produced higher flows in 1955 than in 1954. In reality, observed discharge in 1954 was greater, consistent with the catastrophic basin-wide flood. This discrepancy reflects the sensitivity of Shigu, a high-altitude station, to localized precipitation in the upper Jinsha Basin, snowmelt contributions, and antecedent soil moisture conditions. By contrast, the extreme 1954 flood was primarily driven by intense regional precipitation downstream of Shigu, which explains why Zhutuo, Cuntan, and Yichang all recorded much higher flows in 1954 than in 1955.

With respect to the exceptionally high basin-mean precipitation in 1952 and 1954 shown in Figure 2, the 1954 extreme aligns with well-documented historical records of a catastrophic flood driven by persistent, widespread monsoon rainfall. For 1952, although no basin-wide catastrophic flood is recorded, the observed hydrograph at Cuntan shows a peak nearly as high as in 1954, suggesting that 1952 was likely a wet year in the mid–upper mainstream. This lends plausibility to the high CMFD v2.0 precipitation for 1952. Nevertheless, given limitations in early-station coverage, confirming its meteorological causes and rarity will require future work that cross-validates multiple precipitation datasets and historical archives.

Despite such local anomalies and documented early-period uncertainties, model–data comparisons indicate robust basin-scale behavior. At the downstream stations, the NSE exceeded 0.86 and PBIAS remained within ±10%, indicating strong model performance despite potential errors in precipitation input. These results suggest that, even under early-stage observational conditions, the CMFD v2 dataset can adequately reproduce basin-scale hydrological responses.

Moreover, given that observed streamflow records from 1951 to 1955 are relatively complete and that human disturbances (e.g., regulation, land use change) were minimal during this period, the VIC simulations can be interpreted as close approximations of natural streamflow processes.

In summary, although precipitation data from 1951 to 1955 may contain some uncertainties due to limitations in the early meteorological network, their performance in reproducing hydrological processes at the basin scale is robust. As such, these data are deemed sufficiently reliable for use in long-term hydrological trend analysis and historical scenario reconstruction within the Yangtze River Basin.

4.2. Influence of Human Activities on Natural Streamflow

As one of the most heavily human-influenced basins in China, the Yangtze River has undergone substantial modifications to its natural streamflow regime over the past several decades. These changes have been driven by large-scale water conservancy projects, intensified hydropower development, and widespread land use transformation [18,32,33]. Among these, the operation of the Three Gorges Dam since 2003 has emerged as the most significant anthropogenic intervention, exerting strong regulation effects on both wet- and dry-season streamflow [14,15,19].

To evaluate the influence of major regulation projects on seasonal natural streamflow, this section compares observed and simulated natural streamflow at two key mainstream stations—Cuntan and Yichang—over the period 1990 to 2020. The year 1990 was selected as the starting point because several large reservoirs (e.g., Gezhouba, Ertan, Wujiangdu) had already been constructed and operational in the 1990s, and their regulation effects became increasingly significant during this period, preceding the full operation of the Three Gorges Dam in 2003. These stations were selected due to their strategic location downstream of large hydropower reservoirs and their representativeness in capturing basin-wide regulation effects.

Figure 12 presents the changes in wet-season maximum and dry-season minimum streamflow at both Cuntan and Yichang stations. The results clearly show a divergence between natural and observed streamflow patterns since the early 2000s. At Yichang, observed flood-season peak discharges have been consistently lower than simulated natural values, while dry-season minimum discharges have increased significantly. This shift became particularly pronounced after 2003, coinciding with the full operation of the Three Gorges Dam. The resulting regulation pattern follows a typical reservoir management strategy—storing water during the wet season to attenuate peak discharges and releasing water during the dry season to supplement low flows. Quantitatively, between 1990 and 2020, the wet-season peak discharge at Yichang decreased by 10.04% relative to natural conditions, while the dry-season minimum increased by 27.63%, corroborating previous findings [16].

A similar pattern is observed at Cuntan, where dry-season streamflow has steadily increased since 1995, with a more rapid rise after 2015. This corresponds to the commissioning of major upstream reservoirs—including Wudongde, Baihetan, Xiluodu, and Xiangjiaba—which have collectively enhanced the inter-seasonal redistribution of flow. From 1990 to 2020, the dry-season minimum discharge at Cuntan rose by 26.09% relative to simulated natural streamflow. Interestingly, the observed wet-season peak discharge at Cuntan is slightly higher than the natural simulation (3.89%), which may reflect coordinated reservoir operations that either delay peak runoff or release water in response to downstream hydropower demand.

In summary, regulation projects—particularly the Three Gorges Dam and the cascade of upstream hydropower reservoirs—have substantially reshaped the seasonal dynamics of natural streamflow in the upper Yangtze mainstream. Wet-season streamflows have been effectively attenuated, while dry-season streamflows have been augmented. Although these anthropogenic modifications support water supply security and hydropower generation, they also pose emerging challenges for maintaining ecological flow regimes and managing flood risks under a changing climate. The findings of this study provide scientific support for adaptive reservoir operation and for enhancing flood and drought risk management strategies under climate change.

4.3. Limitations

This study employed long-term precipitation records and simulated natural streamflow data, in combination with trend detection and spatial analysis, to reconstruct the natural hydrological variability of the upper Yangtze River over the past 70 years as comprehensively as possible. Nonetheless, several limitations remain due to constraints in data availability, model structure, and methodological scope.

First, the natural streamflow series were generated using the VIC model under a specific calibration framework. Although rigorous calibration and validation were conducted at multiple representative stations, the issue of parameter equifinality (non-uniqueness) introduces uncertainty in the simulated results. This is particularly evident in the model’s reduced ability to accurately reproduce extreme peak flows in certain years. In addition, snow processes in the headwater regions of the Tibetan Plateau and western Sichuan are not well constrained due to sparse ground observations, and the reanalysis product may under- or overestimate snow accumulation and melt, which could introduce further uncertainties in seasonal runoff simulations.

Second, long-term projections of hydrological trends are essential for supporting basin-wide flood control, drought mitigation, and water resources planning. Existing studies combining hydrological models with climate projections suggest a likely increasing trend in precipitation in the Laixi River Basin (a key sub-basin in the upper Yangtze), especially during May–September [34]. Moreover, under projected warming and modest precipitation increases, the upper Yangtze is expected to experience rising mean annual streamflow, enhanced seasonal high flows, and more frequent daily peak discharges throughout the 21st century [35]. The frequency of extreme floods at key mainstream stations—such as Cuntan, Yichang, and Datong—is also projected to rise significantly [5]. Given the strong climatic sensitivity of streamflow in the middle and lower Yangtze, incorporating updated climate projection datasets and extending the modeling domain downstream should be prioritized in future research.

Finally, with regard to spatial representativeness, this study focused primarily on the upper Yangtze Basin upstream of Yichang. While the selected stations are representative of hydrological conditions in this region, they do not fully capture the flow evolution and anthropogenic impacts in the middle and lower reaches. Future work should incorporate additional control stations and assimilate multi-source datasets from across the basin to provide a more complete, integrated assessment.

In addition, this study did not explicitly consider the long-term effects of urbanization and land use change, which may alter runoff generation and hydrological responses in rapidly developing regions. This omission represents another source of uncertainty and should be addressed in future research. In summary, despite the systematic design of this study in terms of methodology and data, limitations remain in model generalization, scenario coverage, and spatial completeness. Future research should emphasize model inter-comparison, data assimilation, and more detailed process-based representations to further enhance the robustness, scalability, and scientific rigor of hydrological assessments in the Yangtze River Basin.

5. Conclusions

Based on high-resolution CMFD v2.0 meteorological data and the VIC distributed hydrological model, this study systematically assessed the spatiotemporal trends of precipitation and natural streamflow in the Yangtze River Basin from 1951 to 2020 and further explored the hydrological impacts of human activities. The main conclusions are as follows:

(1) Significant spatial and seasonal variability in precipitation trends. Between 1955 and 2020, the basin-wide mean annual precipitation showed a significant increasing trend (1.095 mm/yr, p < 0.05), with particularly pronounced increases in the eastern margin of the Tibetan Plateau and the mountainous areas of western Sichuan. The rate of increase in wet-season (May–October) precipitation was much higher than that in the dry season, especially in the upper basin, which led to increased flood risk during the wet season.

(2) Reliable simulation of long-term natural streamflow. The VIC model achieved strong performance at major mainstream stations such as Shigu, Panzhihua, Zhutuo, Cuntan, and Yichang. During the calibration period (1980–2000), NSE values were all above 0.9 and PBIAS within ±10%. In the validation (1951–1979) and test periods (2001–2020), NSE remained above 0.7, indicating that the model can reproduce both interannual and seasonal variability of natural streamflow.

(3) Marked regional heterogeneity in streamflow trends. Between 1951 and 2020, annual streamflow at upper mainstream stations—Shigu (6.25 m³/s·yr⁻¹), Panzhihua (6.23 m³/s·yr⁻¹), and Zhutuo (14.58 m³/s·yr⁻¹)—increased significantly (p < 0.05). In contrast, no significant trends were observed at downstream mainstream stations such as Yichang and Cuntan. Tributary stations (Beibei, Wulong) exhibited significant decreasing trends. At the seasonal scale, wet-season streamflow increased upstream of Zhutuo, while dry-season streamflow declined downstream, forming a distinct “wetter in the upper basin, drier in the lower basin” spatial pattern.

(4) Strong and increasing anthropogenic influence on streamflow regimes. After the operation of the Three Gorges Dam in 2003, observed streamflow at Yichang shows marked changes: since 1990, peak monthly discharge during the wet season decreased by 10.04%, while minimum monthly flow during the dry season increased by 27.63%, compared to simulated natural streamflow. Notably, trend analysis indicates that under natural conditions, Yichang would face a rising wet-season flow and sharply decreasing dry-season flow (−19.06 m³/s·yr⁻¹, p < 0.01), reflecting a growing risk of flood/drought extremes. These results underscore the regulatory benefits of large reservoirs in flattening seasonal contrasts and maintaining downstream water security.

In conclusion, the natural hydrological regime of the upper Yangtze River Basin over the past 70 years has undergone substantial spatial and temporal changes under the combined influences of climate variability and human interventions. This study is novel in combining high-resolution CMFD v2.0 data with VIC model simulations and reservoir regulation analysis to provide a comprehensive assessment of precipitation trends across the entire Yangtze River Basin and natural streamflow trends in the upper basin over the past seven decades. This study elucidates the response of natural streamflow to evolving precipitation patterns and provides a scientific foundation for integrated water resources planning, flood control, and drought mitigation in the basin. Future work should incorporate multi-model ensemble approaches and assimilate diverse data sources to enhance the detection of extreme events, improve process understanding, and strengthen predictive capabilities for long-term hydrological change under a changing climate.