The Evolution of Runoff Processes in the Source Region of the Yangtze River Under Future Climate Change

Abstract

Climate change has intensified the melting of glaciers and permafrost in high-altitude cold regions, leading to more frequent extreme hydrological events. This has caused significant variations in the spatiotemporal distribution of meltwater runoff from the headwater cryosphere, posing a major challenge to regional water security. In this study, the HBV hydrological model was set up and driven by CMIP6 global climate model outputs to investigate the multi-scale temporal variations of runoff under different climate change scenarios in the Tuotuo River Basin (TRB) within the source region of the Yangtze River (SRYR). The results suggest that the TRB will undergo significant warming and wetting in the future, with increasing precipitation primarily occurring from May to October and a notable rise in annual temperature. Both temperature and precipitation trends intensify under more extreme climate scenarios. Under all climate scenarios, annual runoff generally exhibits an upward trend, except under the SSP1-2.6 scenario, where a slight decline in total runoff is projected for the late 21st century (2061–2090). The increase in total runoff is primarily concentrated between May and October, driven by enhanced rainfall and meltwater contributions, while snowmelt runoff also shows an increase, but accounts for a smaller percentage of the total runoff and has a smaller impact on the total runoff. Precipitation is the primary driver of annual runoff depth changes, with temperature effects varying by scenario and period. Under high emissions, intensified warming and glacier melt amplify runoff, while low emissions show stable warming with precipitation dominating runoff changes.

1. Introduction

Climate change is currently one of the most important global environmental issues [1,2,3]. The IPCC’s Sixth Assessment Report (IPCC6) indicates that the substantial rise in global temperature [4] and SST has caused widespread melting of snow and glaciers, altering the temporal and spatial distribution of water resources and increasing the frequency of extreme climatic phenomena [5]. The substantial hydroclimatic changes directly affect the ecological environment on a regional and even global scale [6]. This is especially obvious in alpine areas such as the Tibetan Plateau. The Tibetan Plateau is the source of numerous rivers [7], including the Yangtze, Yellow, Lancang, and Yarlung Tsangpo Rivers, which play crucial roles in regulating the climate, conserving water and soil, controlling wind and sand, releasing oxygen, sequestering carbon, and regulating the climate [8]. Due to its special geographic location and fragile and sensitive ecological environment, the Tibetan Plateau has responded more strongly to climate change compared to the global average. It experiences a warming trend that exceeds twice the global average temperature rise in the last 50 years [9,10]. Under the influence of warming and humidification, the change in runoff in the river source area has become a research hotspot [11,12]. According to recent studies, climate change and catastrophic events affecting water resources as well as climatic and hydrological extremes in the Tibetan Plateau would become more likely due to global warming [13]. As a result, concerns about disaster risk management and the security of water resources will receive greater attention. As the hinterland of the Tibetan Plateau and the main watershed of the Yangtze River headwaters, climate change in the Tuotuo River Basin (TRB) has received more attention [14,15]. Recent studies indicate that since the 1960s, the Yangtze River headwater basin has experienced significant upward trends in both temperature and runoff depth [16,17]. In contrast, precipitation trends have been less pronounced [18]. Notably, precipitation in the TRB has been identified as the primary climatic factor influencing runoff magnitude [19]. Specifically, increased summer precipitation has shown a strong correlation with elevated summer runoff. Glaciers and snowmelt, driven by rising mean annual minimum temperatures, are the secondary contributing factors to runoff variations [20,21], while changes in evapotranspiration have demonstrated less impact [22,23]. These findings highlight the significant spatial and temporal variability of runoff in the headwater region, which is influenced by complex interactions between climate factors (e.g., temperature and precipitation) and cryospheric processes (e.g., glacier and snowmelt) [24,25]. Such variability not only underscores the sensitivity of the headwater region to climate change but also emphasizes the need for further research to understand its implications for water resource management and ecosystem sustainability [26].

The Yangtze River headwater region, characterized by high elevation and climatic conditions, represents a data-scarce area where hydrological process simulations are significantly constrained by limited observational data [27]. In such data-scarce alpine river headwater regions, the HBV (Hydrologiska Byråns Vattenbalansavdelning) model has gained widespread application owing to its relatively simple structure and minimal data requirements [28,29]. Global climate models (GCMs) combined with regional hydrological models for the simulation of historical periods and prediction of future scenarios have become an important research tool for researchers in the fields of meteorology and hydrology [30]. Compared to its predecessor, the Fifth International Coupled Model Intercomparison Program (CMIP5), CMIP6 demonstrates significant improvements in simulating the spatial distribution of climate elements and reducing systematic biases [31,32]. Notably, CMIP6 exhibits enhanced performance in regional temperature and precipitation simulations [30,33]. Currently, CMIP6 data are widely utilized in various fields [34,35]. Numerous studies have systematically evaluated CMIP6 models’ capability in simulating China’s climate distribution and precipitation–temperature evolution patterns [36,37,38]. These evaluations reveal substantial improvements in model performance, characterized by continuous resolution enhancement and increased output stability. The reliable and consistent simulation results obtained using CMIP6 models provide valuable insights for future climate change projections in the Tibetan Plateau region [17,39]. Currently, most of the studies focus on applying CMIP6 data to study the future water resource changes in the Yellow River source area [40,41,42], while studies in the Yangtze River source region have primarily concentrated on historical meteorological element variations and their corresponding runoff characteristics [19,23]. Notably, research examining future climate change impacts on runoff evolution trends in the Yangtze River source area, particularly in the TRB, remains limited, highlighting a critical research gap that this study addresses.

For this purpose, this study investigates the runoff evolution of the TRB under different shared socioeconomic pathway (SSP) scenarios by driving the HBV model with the CMIP6 outputs. The research aims to (1) investigate future variations in temperature, precipitation, and runoff patterns under different climate scenarios, and (2) analyze the impacts of temperature and precipitation changes on runoff dynamics. These findings will offer valuable theoretical foundations for developing effective water resource management and disaster mitigation policies in response to climate change impacts.

2. Study Area

The Tuotuo River Basin (TRB) is situated in the southwestern Qinghai Province (Figure 1), within the core area of the Tibetan Plateau (89°48′–92°54′ E, 33°22′–35°12′ N). As the western source among the three headwaters of the Yangtze River, the basin originates from the Jianggendiru Glacier on the southwestern flank of the Geladaindong massif, the main peak of the Tanggula Mountains (elevation approximately 5500 m). The basin exhibits prominent high-altitude characteristics, with elevations ranging from 4489 m at its lowest point to 6468 m at its highest peak. It is geographically enclosed by two major mountain ranges: the Hoh Xil Mountains to the south and the Kunlun Mountains to the north. The basin covers a total area of 15,924 km² (with the Tuotuoheyan Hydrological Station serving as the control section), including a glacial coverage of 381 km² concentrated primarily in the southern regions of the basin. The terrain is significantly affected by glacial erosion and freeze–thaw action, forming a complex surface structure dominated by moraines and exposed bedrock. The vegetation type of the watershed is relatively simple, with alpine meadows, alpine grasslands, alpine swamp meadows, alpine deserts, and other ecological vegetation types, of which meadows dominate the plateau meadows and alpine grasslands. Hydrological stations are sparse. The basin has a meteorological station (Tuotuo River meteorological station) and a national hydrological station (Tuotuo River Hydrological Station), both geographically adjacent. However, due to the harsh climate and living conditions, the observation work of the hydrological station and the meteorological station of the water is only carried out from May to October every year.

The Tuotuo River Basin is located in the hinterland of the Tibetan Plateau, belonging to the typical high-altitude westerly wind belt control area, and its climatic characteristics are characterized as the transition zone between alpine semi-arid and semi-humid climates. Two major factors mainly control the climate of the region: one is the Bay of Bengal water vapor transport path from the Indian Ocean, and the other is the uplift effect of the plateau topography. Temperature data from the Tuotuo River Hydrological Station (2006–2020) show that the basin maintains a low temperature throughout the year, with a multi-year average temperature of −4.2 °C (Figure 2). July is the hottest month, with an average temperature of 7.5 °C; January is the coldest month, with an average temperature of −24.8 °C. Under the influence of low temperature, the perennial freezing period in the basin is as long as 7 months, and there are no obvious seasons throughout the year, which can only be divided into two seasons: the dry season (from October to April) and the wet season (from May to September).

Precipitation in the basin is characterized by noticeable spatial and temporal variations, with a multi-year average precipitation of 283.1 mm based on ground observations provided by the National Cryosphere Desert Data Center [2009–2020; http://www.ncdc.ac.cn (accessed on 17 August 2024)]. The seasonal distribution of precipitation is extremely uneven, with more than 70% of the precipitation concentrated from July to September. The hydrological process is mainly controlled by the recharge of snow and ice meltwater, with an average annual runoff depth of 51.9 mm derived from daily runoff records of the Tuotuo River Hydrological Station (2006–2020). It is worth noting that, due to the significant temperature difference between daytime and nighttime in the summer, the daily variation of the river level is more obvious, and this feature has an important impact on the hydrological process of the basin.

3. Data and Methods

3.1. CMIP6 Data

This study utilizes statistically downscaled and bias-corrected CMIP6 climate model data from the National Tibetan Plateau Science Data Center [43] as hydrological model input data trails for streamflow simulation. The dataset includes daily precipitation and temperature outputs from eight global climate models (

Table 1. Information on CMIP6 climate models.

Climate Models	Country	Resolution
ACCESS-ESM1-5	Australia	1.88° × 1.24°
BMA	Brazil	1.875° × 1.25°
CanESM5	Canada	2.81° × 2.81°
CNRM-CM6	Australia	1.88° × 1.24°
INMCM4-8	Russia	2.00° × 1.50°
IPSL	France	2.50° × 1.26°
MRI-ESM2-0	Japan	1.125° × 1.125°
UKESM1-0-LL	The United Kingdom	1.875° × 1.25°

):

The CMIP6 data spans historical simulations (1981–2014) and future projections under four shared socioeconomic pathways (SSPs) for 2015–2100: SSP1-2.6 (low emissions), SSP2-4.5 (medium), SSP3-7.0 (medium to high), and SSP5-8.5 (high emissions) [44]. The dataset was preprocessed using ANU-Spline statistical downscaling and regridded to the same spatial resolution as that of CMFD (0.1 × 0.1°). This reduced mean precipitation bias to <0.2 mm/day and temperature bias to <0.97 °C compared to raw model outputs [45]. In this study, the hydrologic model is driven by daily historical scenario precipitation and temperature data from eight climate models in the CMIP6 dataset.

3.2. HBV Model

For hydrological simulations, the HBV model was applied to simulate the future runoff in the TRB, which is a semi-distributed conceptual hydrological forecasting model developed and improved by the Swedish National Hydrometeorological Institute (SMHI). It has been widely used in many countries and regions around the world to simulate the hydrological simulation of river basins, due to the characteristics of the simple model structure, low data requirements, and flexibility [46,47]. The HBV model has been used in runoff simulations in most of the watersheds on the Tibetan Plateau, especially in the Yangtze and Yellow River source area watersheds [48,49]. The HBV model can be simplified into four modules. The structure of the HBV model is shown in Figure 3.

(1)

Snowmelt module

Rainfall and snowfall are two main forms of precipitation, and their specific forms mainly depend on whether the temperature is higher or lower than the critical temperature $\mathrm{T}\mathrm{T}\left(\mathrm{℃}\right)$. When the temperature is lower than the critical temperature $\mathrm{T}\mathrm{T}\left(\mathrm{℃}\right)$, all precipitation within the time step is considered to be in the form of snowfall. In order to compensate for the systematic errors in snowfall measurement and the evapotranspiration errors of snow that are not explicitly simulated in the model, the dimensionless precipitation correction coefficient $\mathrm{S}\mathrm{C}\mathrm{F}$ is introduced into the model. Daily snowmelt $\mathrm{M}(\mathrm{m}\mathrm{m}/\mathrm{d})$ is calculated by the degree-day factor model:

$$\mathrm{M}=\mathrm{C}\mathrm{F}\mathrm{M}\mathrm{A}\mathrm{X}·\left(\mathrm{T}\left(\mathrm{t}\right)-\mathrm{T}\mathrm{T}\right)·{\mathrm{C}}_{\mathrm{g}}·{\mathrm{C}\mathrm{F}}_{\mathrm{s}\mathrm{l}\mathrm{o}\mathrm{p}\mathrm{e}}$$

(1)

In the formula, $\mathrm{C}\mathrm{F}\mathrm{M}\mathrm{A}\mathrm{X}$ is the degree-day factor, T(t) is the daily average temperature (°C), and $\mathrm{T}\left(\mathrm{t}\right)-\mathrm{T}\mathrm{T}$ is the daily positive accumulated temperature (that is, when $\mathrm{T}\left(\mathrm{t}\right)-\mathrm{T}\mathrm{T}<0$, $\mathrm{T}\left(\mathrm{t}\right)-\mathrm{T}\mathrm{T}=0$). ${\mathrm{C}}_{\mathrm{g}}$ is the melting factor when the icemelt is more than the snowmelt. The parameter ${\mathrm{C}\mathrm{F}}_{\mathrm{s}\mathrm{l}\mathrm{o}\mathrm{p}\mathrm{e}}(\mathrm{m}\mathrm{m}/\mathrm{℃}/\mathrm{d}/°)$ is introduced in the model because of the difference in snow cover and glacier ablation in different slopes. Meltwater and rainfall are retained in the snow until they exceed a certain percentage of snow water equivalent ($\mathrm{C}\mathrm{W}\mathrm{H}$). When the temperature drops below the critical temperature, the refrozen liquid water $\mathrm{R}(\mathrm{m}\mathrm{m}/\mathrm{d})$ in the snow is calculated using the freezing coefficient $\mathrm{C}\mathrm{F}\mathrm{R}$:

$$\mathrm{R}=\mathrm{C}\mathrm{F}\mathrm{R}·\mathrm{M}$$

(2)

where M (mm/d) is the daily snowmelt calculated from Equation (1).

(2)

Soil module

According to the sum of rainfall and snowmelt input to the soil in a certain period of time, the soil flow production is calculated according to the following formula:

$${\displaystyle \frac{\mathrm{R}}{\mathrm{P}}}=({{\displaystyle \frac{\mathrm{S}\mathrm{M}}{\mathrm{F}\mathrm{C}}})}^{\mathrm{B}\mathrm{E}\mathrm{T}\mathrm{A}}$$

(3)

where $\mathrm{R}(\mathrm{m}\mathrm{m}/\mathrm{d})$ represents the daily runoff yield generated from the soil layer, $\mathrm{P}(\mathrm{m}\mathrm{m}/\mathrm{d})$ is the total water input to the soil (e.g., precipitation or snowmelt), $\mathrm{S}\mathrm{M}\left(\mathrm{m}\mathrm{m}\right)$ is the current soil water content, and $\mathrm{F}\mathrm{C}\left(\mathrm{m}\mathrm{m}\right)$ is the field capacity of the soil. BETA is an empirical exponent controlling the nonlinear relationship between soil moisture and runoff generation. Higher BETA values indicate stronger nonlinearity. The meaning of this formula is that the ratio of yield and discharge to input soil water and the ratio of soil water content to field water capacity are power function relations, which is one of the core ideas of the HBV model.

The calculation of soil evapotranspiration is shown in Equation (4), where ${\mathrm{E}}_{\mathrm{a}\mathrm{c}\mathrm{t}}\left(\mathrm{m}\mathrm{m}\right)$ is the actual evapotranspiration, ${\mathrm{E}}_{\mathrm{p}\mathrm{o}\mathrm{t}}\left(\mathrm{m}\mathrm{m}\right)$ is the potential evapotranspiration calculated as ${\mathrm{E}}_{\mathrm{p}\mathrm{o}\mathrm{t}}=0.72\times {\mathrm{E}}_{\mathrm{p}\mathrm{a}\mathrm{n}}$, ${\mathrm{E}}_{\mathrm{p}\mathrm{a}\mathrm{n}}\left(\mathrm{m}\mathrm{m}\right)$ is the pan evaporation data derived from national hydrological yearbooks (2006–2020), and $\mathrm{L}\mathrm{P}$ is the threshold of soil water content and field water capacity. Evapotranspiration data were integrated with observed temperature and precipitation records to drive the HBV model’s soil moisture module. Specifically, daily ${\mathrm{E}}_{\mathrm{p}\mathrm{o}\mathrm{t}}$ values served as the upper limit for evapotranspiration demand, while the ratio of soil moisture (SM) to FC dynamically regulated actual evapotranspiration through the LP threshold. If ${\displaystyle \frac{\mathrm{S}\mathrm{M}}{\mathrm{F}\mathrm{C}}}>\mathrm{L}\mathrm{P}$, the actual evapotranspiration is equal to the potential evapotranspiration. If ${\displaystyle \frac{\mathrm{S}\mathrm{M}}{\mathrm{F}\mathrm{C}}}<\mathrm{L}\mathrm{P}$, the actual evapotranspiration is less than the potential evapotranspiration and increases linearly with an increase in soil water content.

$${\mathrm{E}}_{\mathrm{a}\mathrm{c}\mathrm{t}}={\mathrm{E}}_{\mathrm{p}\mathrm{o}\mathrm{t}}·\mathrm{m}\mathrm{i}\mathrm{n}\left({\displaystyle \frac{\mathrm{S}\mathrm{M}}{\mathrm{F}\mathrm{C}·\mathrm{L}\mathrm{P}}},1\right)$$

(4)

(3)

Response Module

The HBV model uses three linear reservoirs to describe the underground runoff calculation, and the calculation formula is as follows:

$${\mathrm{Q}}_{\mathrm{G}\mathrm{W}}\left(\mathrm{t}\right)={\mathrm{K}}_2·\mathrm{S}\mathrm{L}\mathrm{Z}+{\mathrm{K}}_1·\mathrm{S}\mathrm{U}\mathrm{Z}+{\mathrm{K}}_0·\mathrm{m}\mathrm{a}\mathrm{x}(\mathrm{S}\mathrm{U}\mathrm{Z}-\mathrm{U}\mathrm{Z}\mathrm{L},0)$$

(5)

where ${\mathrm{K}}_2$, ${\mathrm{K}}_1$ and ${\mathrm{K}}_0\left({\mathrm{d}}^{-1}\right)$ are the outflow coefficients of flood peak, midsoil flow and base flow, respectively. $\mathrm{S}\mathrm{U}\mathrm{Z}\left(\mathrm{m}\mathrm{m}\right)$ is the upper soil water content, $\mathrm{S}\mathrm{L}\mathrm{Z}\left(\mathrm{m}\mathrm{m}\right)$ is the lower soil water content, and $\mathrm{U}\mathrm{Z}\mathrm{L}\left(\mathrm{m}\mathrm{m}\right)$ is the threshold of upper soil water content. In addition, an additional maximum permeability $\mathrm{P}\mathrm{E}\mathrm{R}\mathrm{C}(\mathrm{m}\mathrm{m}/\mathrm{d})$ is defined from top to bottom in the underground tank. PERC represents the maximum percolation rate from the upper to lower soil reservoir, which limits the vertical water transfer between the SUZ and SLZ. The upper soil reservoir (SUZ) receives water inputs and transfers it to the lower reservoir (SLZ) at a rate controlled by PERC.

(4)

Converging module

Eventually, the runoff transport process in the river channel is simulated in the river catchment module, and a linear or nonlinear reservoir model is used to describe the delay and attenuation effects of the runoff, and ultimately the flow process line at the outlet of the watershed is generated.

Model performance is quantitatively assessed by the Nash–Sutcliffe efficiency coefficient (NSE) and the coefficient of determination (R²), calculated below:

$$\mathrm{N}\mathrm{S}\mathrm{E}=1-{\displaystyle \frac{\sum {\left({\mathrm{R}}_{\mathrm{o}\mathrm{b}\mathrm{s}}-{\mathrm{R}}_{\mathrm{s}\mathrm{i}\mathrm{m}}\right)}^2}{\sum {\left({\mathrm{R}}_{\mathrm{o}\mathrm{b}\mathrm{s}}-\overline{{\mathrm{R}}_{\mathrm{o}\mathrm{b}\mathrm{s}}}\right)}^2}}$$

(6)

$${\mathrm{R}}^2={\displaystyle \frac{{\left[\sum \left({\mathrm{R}}_{\mathrm{o}\mathrm{b}\mathrm{s}}-\overline{{\mathrm{R}}_{\mathrm{o}\mathrm{b}\mathrm{s}}}\right)\left({\mathrm{R}}_{\mathrm{s}\mathrm{i}\mathrm{m}}-\overline{{\mathrm{R}}_{\mathrm{s}\mathrm{i}\mathrm{m}}}\right)\right]}^2}{\sum {\left({\mathrm{R}}_{\mathrm{o}\mathrm{b}\mathrm{s}}-\overline{{\mathrm{R}}_{\mathrm{o}\mathrm{b}\mathrm{s}}}\right)}^2\sum {\left({\mathrm{R}}_{\mathrm{s}\mathrm{i}\mathrm{m}}-\overline{{\mathrm{R}}_{\mathrm{s}\mathrm{i}\mathrm{m}}}\right)}^2}}$$

(7)

The observed daily runoff data (2006–2020) from the Tuotuo River Hydrological Station, retrieved from the national hydrological yearbooks, were split into a calibration period (2006–2011) for parameter optimization and a validation period (2012–2014) for performance evaluation.

3.3. Multi-Temporal-Scale Analysis

In this study, hydroclimatic variables were analyzed using CMIP6 datasets spanning the historical period 1981–2014 and future projections (2015–2090). To align with standard climate cycle definitions, the data were partitioned into three 30-year intervals: historical baseline (1981–2010; hist), near future (2031–2060; future1), and distant future (2061–2090; future2). Monthly and annual values of precipitation, temperature, and runoff were systematically examined across these periods.

To assess the statistical significance of intra-annual monthly variations in precipitation, temperature, and runoff depth within the TRB, a permutation-based resampling approach [51] is employed. The methodology involves randomly selecting sets of years without replacement from the combined historical and future periods, with each sample matching the duration of the future phase (30 years). This randomization process is iterated 1000 times, and for each iteration, monthly metrics (such as precipitation, temperature, and runoff depth) derived from the ensemble model mean are computed. Changes in these monthly metrics are deemed statistically significant at the 0.1 level if their values surpass the 95th or 5th percentiles of the distribution generated from the permutation results.

The Mann–Kendall (MK) test was applied to assess trends in basin-averaged annual precipitation, air temperature (spatially averaged over the TRB), and simulated runoff at the hydrological control section from 1981 to 2100. The standardized test statistic $\mathrm{Z}$ is calculated as

$$\mathrm{Z}={\displaystyle \frac{\mathrm{S}-\mathrm{s}\mathrm{i}\mathrm{g}\mathrm{n}\left(\mathrm{S}\right)}{\sqrt{\mathrm{V}\mathrm{a}\mathrm{r}\left(\mathrm{S}\right)}}}$$

(8)

The Mann–Kendall score statistic S is computed by comparing all possible pairs of observations in the time series. For each pair (${\mathrm{x}}_{\mathrm{i}}$,${\mathrm{x}}_{\mathrm{j}}$)(${\mathrm{x}}_{\mathrm{i}}$,${\mathrm{x}}_{\mathrm{j}}$) where j > i, a score of +1, 0, or −1 is assigned if ${\mathrm{x}}_{\mathrm{j}}$ is greater than, equal to, or less than ${\mathrm{x}}_{\mathrm{i}}$, respectively. The sum of these scores across all pairs defines S. Positive/negative S values indicate increasing/decreasing trends, and their magnitude reflects trend strength. Z > 1.96 or Z < −1.96 indicates statistically significant trends (p < 0.05). Positive/negative Z-values denote increasing/decreasing trends, respectively.

To disentangle the independent contributions of precipitation (P) and temperature (T) to future runoff (Q) variability under different climate scenarios, partial correlation analysis was employed. This method quantifies the direct relationship between two variables while statistically controlling for the confounding influence of a third variable, thereby isolating their independent effects. The partial correlation coefficient ${\mathrm{r}}_{\mathrm{x}\mathrm{y}|\mathrm{z}}$, representing the association between variables $\mathrm{x}$ and $\mathrm{y}$ after removing the effect of $\mathrm{z}$, was calculated using the following formula:

$${\mathrm{r}}_{\mathrm{x}\mathrm{y}|\mathrm{z}}={\displaystyle \frac{{\mathrm{r}}_{\mathrm{x}\mathrm{y}}-{\mathrm{r}}_{\mathrm{x}\mathrm{z}}{\mathrm{r}}_{\mathrm{y}\mathrm{z}}}{\sqrt{\left(1-{\mathrm{r}}_{\mathrm{x}\mathrm{z}}^2\right)\left(1-{\mathrm{r}}_{\mathrm{y}\mathrm{z}}^2\right)}}}$$

(9)

where ${\mathrm{r}}_{\mathrm{x}\mathrm{y}}$, ${\mathrm{r}}_{\mathrm{x}\mathrm{z}}$, and ${\mathrm{r}}_{\mathrm{y}\mathrm{z}}$ denote pairwise Pearson correlation coefficients. For each climate scenario, three sequential steps were implemented. First, Pearson correlation coefficients were computed between annual runoff, precipitation, and temperature to establish their pairwise linear relationships. Subsequently, partial correlations were derived to isolate the independent effects: ${\mathrm{r}}_{\mathrm{Q}\mathrm{P}|\mathrm{T}}$ (runoff–precipitation correlation controlling for temperature) and ${\mathrm{r}}_{\mathrm{Q}\mathrm{T}|\mathrm{P}}$ (runoff–temperature correlation controlling for precipitation). Finally, the statistical significance of these partial correlations was evaluated via a two-tailed t-test (α = 0.05), with p-values indicating the likelihood of observing such associations by random chance. This approach provides a robust statistical framework to attribute runoff changes to precipitation or temperature drivers under varying climate pathways, distinguishing their roles even when these variables are intrinsically linked.

The analysis was segmented based on the near future (2031–2060; future1) and far future (2061–2090; future2) as previously defined.

4. Results and Discussion

4.1. Calibration and Validation of HBV Model

The CMIP6 climate outputs are compared with observed station data during the overlapping period (2009–2014) to evaluate input the uncertainties and inter-model variability. The interannual distribution characteristics, illustrated by box plot analysis (Figure 4), reveal fundamental differences in precipitation simulation frameworks: while most models capture the observed median (280–560 mm), their physical parameterizations produce systematically narrower ranges (280–520 mm), particularly underestimating upper extremes. ACCESS-ESM1-5 and MRI-ESM2-0 demonstrate superior skill in replicating precipitation distribution patterns, likely attributable to refined snow process representations. Temperature simulations exhibit systematic cold biases across models (−1.76 °C to −1.74 °C range), with IPSL and MRI-ESM2-0 showing tighter observational alignment. These results reflect systematic cold biases prevalent in most simulations. These inter-model variations highlight uncertainties in climate projections, particularly for high-altitude regions where local feedback mechanisms (e.g., snow albedo effects) are sensitive to temperature biases.

The performance and uncertainties of eight climate models for precipitation and temperature are evaluated based on Taylor diagram analysis (Figure 5), showing substantial divergence in precipitation simulations with standardized standard deviations spanning 0.4583 (UKESM1-0-LL) to 1.0666 (CanESM5). CNRM-CM6 achieves optimal performance in precipitation simulations, evidenced by high spatiotemporal consistency (R = 0.7629) and minimal RMSE (0.631). Temperature simulations exhibit greater variability (STD = 0.6076–1.3093), with ACCESS-ESM1-5 demonstrating moderate correlation (R = 0.7285) and lowest RMSE (0.6281), contrasting sharply with MRI-ESM2-0’s high RMSE (1.7229). Most models exhibit poor phase-matching capability (R < 0.1) for temperature. These disparities underscore pronounced systematic uncertainties, particularly in high-altitude regions where temperature-sensitive feedback mechanisms amplify biases. The findings emphasize the need for cautious interpretation of multi-model ensemble reliability in climate projections.

In this study, the HBV hydrologic model is driven by daily historical scenario precipitation and temperature data from eight climate models (1981–2014) in the bias-corrected CMIP6 dataset, and the model calibration and validation are systematically carried out. The model calibration period is set for 2006–2011, the validation period for 2012–2014, and the model performance is assessed by comparative analysis with the measured May-October daily runoff data (2006–2020). This seasonal data limitation (May–October) introduces challenges in calibrating winter processes (e.g., snow accumulation), potentially propagating uncertainties to future simulations, especially for non-observational periods. This seasonal limitation inevitably affects the calibration and validation of the hydrological model, potentially introducing uncertainties in simulating hydrological processes during the unobserved cold season, especially under future climate scenarios. However, it is important to emphasize that our analysis specifically targets the dominant hydrological phase, during which approximately 90% of the annual runoff is generated. This period encompasses the primary meltwater- and rainfall-driven runoff events, making it the most hydrologically active and relevant season for basin-scale water resource assessment.

The HBV model was calibrated through a semi-automated parameter optimization procedure. Initially, a Monte Carlo simulation was implemented to extensively explore the parameter space and identify candidate parameter sets that maximize the Nash–Sutcliffe efficiency (NSE) and the coefficient of determination (R²). Subsequently, manual refinements were applied to ensure the physical realism of the selected parameters, guided by established hydrological principles and the specific physiographic and climatic characteristics of the study basin. The HBV-light model includes 14 parameters (Supplementary Table S1: Climate model simulation performance assessment (* = sensitive, ** = highly sensitive, – = fixed)), of which eight were identified as sensitive parameters. To reduce computational complexity, non-sensitive parameters were fixed within predefined plausible ranges, while the sensitive parameters were subjected to rigorous optimization. The final calibrated parameter sets corresponding to each climate model are presented in Supplementary Table S1.

To eliminate the influence of interannual fluctuations on the assessment results, this study uses multi-year daily average runoff data for comparison, focusing on capturing the overall trend and seasonal characteristics of the runoff simulation. As shown in

Table 2. Climate model simulation performance assessment.

Climate Models	Calibration Period		Validation Period
	NSE	R2	NSE	R2
ACCESS-ESM1-5	0.76	0.76	0.75	0.71
BMA	0.68	0.69	0.74	0.68
CanESM5	0.65	0.68	0.63	0.64
CNRM-CM6	0.68	0.63	0.64	0.71
INMCM4-8	0.70	0.71	0.71	0.72
IPSL	0.69	0.70	0.69	0.64
MRI-ESM2-0	0.73	0.75	0.69	0.72
UKESM1-0-LL	0.68	0.69	0.63	0.63

, the climate models perform well in the calibration period, with NSE and R² exceeding 0.65; the validation period indicators are also stable above 0.6, among which the ACCESS-ESM1-5 model has the best performance, with NSEs of 0.76 and 0.75, and R² s of 0.76 and 0.71 in the calibration and validation periods, respectively. It is worth noting that the average of the eight models’ ensemble results shows systematic underestimation in the August peak runoff simulation (Figure 6); this is probably due to the smoothing effect of precipitation grid data [52,53] and the failure to capture the precipitation center of summer storms caused by local convective heat transport. Additional factors may contribute to this underestimation: (1) simplified representation of snowmelt processes in HBV, particularly under extreme warming scenarios [54], and (2) limited observational constraints on winter snowpack dynamics (due to data availability only from May to October), which indirectly affect summer meltwater generation [55]. The results show that the HBV model can effectively characterize the hydrological response mechanism of alpine watersheds, and its stable simulation performance (NSE > 0.6, R² > 0.63) provides reliable technical support for runoff prediction in the cold region of the Tibetan Plateau.

4.2. Characteristics of Intra-Annual Variability in Precipitation, Temperature, and Runoff

In this study, based on the results of multi-scenario climate model simulations of eight climate models, the temporal differentiation characteristics of the future climate evolution in the TRB are systematically revealed. In terms of temperature evolution, all SSP scenarios show significant warming trends, and exhibit obvious phase characteristics and scenario dependence (Figure 7). The warming process in the near-future period (2031–2060) showed significant seasonal heterogeneity: the warming in summer (May–September) amounted to 3.8–6.1 °C (SSP5-8.5), which was 1.8–2.3 times higher than that in winter (December–February), which may be closely related to the seasonal differences in the feedback of snow albedo. It is noteworthy that the SSP5-8.5 scenario is generally warmer than the other scenarios by 1.2–1.5 °C due to strong radiative forcing effects, and in particular, the warming in summer (June–August) reaches up to 2.3 times that of the historical base period. Up to the distant period (2061–2090), the climate response shows a significant path divergence: the high forcing scenarios (SSP2-4.5, SSP3-7.0, and SSP5-8.5) show a 45–92% increase in annual mean temperature increase compared to the recent period, with the SSP5-8.5 scenario showing the largest warming. In contrast, under the carbon-neutral development pathway, the SSP1-2.6 scenario only increases long-term warming (ΔT = 2.1 °C) by 11.4% compared with the recent period, which confirms the effectiveness of deep emission reduction in regulating the climate system. The TRB exhibits a distinct warming trend under future climate change. This aligns with the amplified warming observed in the Tibetan Plateau (TP), where temperature rises exceed global averages due to synergistic effects of greenhouse gases, cryospheric feedbacks, and land-use changes [56].

The precipitation response shows the typical characteristics of “regionally increasing and seasonally differentiated”. In the near future (2031–2060), significant wetting occurs from May to September, peaking in August with an average increase of 10.0–15.5 mm (+23–34%). Under the SSP2-4.5 scenario, the largest June precipitation rise (13.45 mm, +29.7%) likely reflects an earlier onset of the Asian summer monsoon, driven by intensified thermodynamic forcing over the Qinghai–Tibetan Plateau. The pronounced August amplification aligns with enhanced convective activity during the plateau’s peak heating phase, which strengthens moisture convergence and elevates precipitation efficiency in late summer [57,58]. The precipitation response in the distant period (2061–2090) shows significant scenario divergence: the high-emission scenarios (SSP3-7.0 and SSP5-8.5) have mean precipitation increases of 176.75% and 260.39% during the monsoon period (May–October), respectively, with a 2.8-fold increase in the frequency of extreme precipitation events in August for the SSP5-8.5 scenario (p < 0.05). In contrast, the increase in the medium–low-emission scenarios (SSP1-2.6 and SSP2-4.5) is relatively moderate (35.00–79.51%), indicating the variability of the hydrologic cycle response mechanism under different emission reduction pathways. Combining the results of multi-model simulations, the TRB will experience a significant climate warming and humidification transition: by the end of the century, the mean annual temperature increase under the SSP5-8.5 scenario will reach 6.8 °C (vs. 1981–2010), and the annual precipitation will increase by 37.5%, with a particularly prominent feature of the water–heat coefficient in the months of June–September (ΔT = 7.2 °C, ΔP = 41.2 mm). This new hydrothermal configuration of “high temperature and heavy precipitation” may reshape the hydrological processes and ecosystem patterns in the alpine region by changing the hydrothermal state of permafrost and accelerating glacier retreat. This is consistent with studies linking high emissions to intensified glacial ablation in the TP [56].

In this study, we systematically analyze the spatial and temporal evolution characteristics of monthly runoff depth and its components in the TRB. We use a multi-model ensemble averaging method based on the runoff simulation results of eight global climate models (GCMs) under future scenarios from the HBV model. A permutation-based resampling approach was applied to assess the significance of trends and variability in the results. The findings reveal that the total runoff depth in the basin under different SSP scenarios exhibits a significant increasing trend, with its growth process showing similar seasonal characteristics to precipitation changes. Specifically, approximately 91% of the runoff increment is concentrated in the June–September water-abundant period (Figure 8). Particularly noteworthy is the amplified scenario dependence of runoff increase in the distant period (2061–2090), with SSP5-8.5 projecting a 2.42-fold rise (121% higher than SSP1-2.6’s 1.09). This divergence aligns with the preceding temperature and precipitation analysis: SSP5-8.5 exhibits substantially stronger warming and wetting trends in the distant period compared to near-term projections (2031–2060), whereas SSP1-2.6 shows stable hydroclimatic conditions between periods. Consequently, runoff under SSP1-2.6 remains resilient to inter-period variability, while SSP5-8.5’s pronounced warming–wetting synergy drives accelerated ice/snowmelt contributions.

Comparing the intra-annual distributions of precipitation and runoff, it is found that the inter-monthly variations of the total runoff depth in the distant period are highly consistent with the precipitation process, revealing a precipitation-dominated hydrological response mechanism. The in-depth analysis of the runoff components shows that the icemelt runoff and snowmelt runoff show differentiated evolution characteristics. Constrained by the high-altitude and low-temperature environment of the basin (mean annual temperature of −4.2 °C), the icemelt runoff is mainly distributed in May–October, and its increment in the near-term period is concentrated in June–September (accounting for 87% of the total increment), of which the increment in July accounts for 33% (Figure 8c). In the long term, the SSP5-8.5 scenario shows a 1.6-fold increase in July melt runoff compared to the recent period, with a significant delayed increase in November, which is directly related to the prolongation of the glacier ablation period due to the increased winter warming amplitude. In contrast, snowmelt runoff increment was concentrated in May–June (91% of the annual increment), but showed a weak decreasing trend in July–October, reflecting the pre-melting period. The forward snowmelt runoff increments under the high-emission scenarios (SSP3-7.0 and SSP5-8.5) increased by 3.1 times compared with the recent period.

Figure 9 illustrates the proportional contributions of different runoff components over distinct periods, highlighting changes in the hydrological regime under varying climatic conditions. The current baseline period (1981–2014) has about a 33% share of annual icemelt runoff, which is significantly higher than snowmelt runoff (2%). The future scenarios generally increase the share of icemelt runoff by 2.5 ± 0.3% in May–October (SSP2-4.5 to SSP5-8.5), with the SSP5-8.5 scenario reaching a 37.8% share in the far future, a 25% increase from the base period. Although the snowmelt runoff share is small in absolute terms (<3%), its relative increase is significant (+0.6 ± 0.2%) and shows a continuously increasing trend (+0.4% in the forward period compared to the recent period). It is worth noting that the change in the fraction of components in the SSP1-2.6 scenario was flat, confirming the maintenance of the stability of hydrological processes by the abatement measures. Taken together, the increase in runoff in the TRB is attributed to two main drivers: first, the incremental snowmelt runoff (contribution 58–67%), which is directly related to the accelerated glacier ablation triggered by summer warming; and second, the increase in precipitation (contribution 33–42%), especially the surface runoff response due to the enhanced intensity of precipitation during the monsoon period. Although snowmelt runoff shows an increasing trend, its limited share (<3%) makes its contribution to the overall runoff change less than 5%. This reconfiguration of hydrological components may trigger profound changes in the spatial and temporal distribution of water resources in the basin, and the compounding effects on the ecosystem of the forward shift of the peak of snowmelt runoff and the increase in winter baseflow need to be emphasized.

4.3. Characterization of Changes in Precipitation, Temperature, and Runoff Trends

In this study, the spatial and temporal evolution of hydrological elements in the TRB under multi-climatic scenarios is systematically analyzed based on the MK trend test method (

Table 3. Mann–Kendall trend test Z-values for precipitation, temperature, and simulated runoff under all scenarios (* indicates significance at the α = 0.01 level).

Variables	Scenarios	HIS	Near Future (2031–2060)	Far Future (2061–2090)
Precipitation	SSP1-2.6	2.28 *	1.18	−1.50
	SSP2-4.5		1.57	−0.21
	SSP3-7.0		2.07 *	3.64 *
	SSP5-8.5		2.85 *	4.10 *
Temperature	SSP1-2.6	5.39 *	4.10 *	−1.11
	SSP2-4.5		5.82 *	4.17 *
	SSP3-7.0		6.49 *	6.67 *
	SSP5-8.5		6.60 *	6.82 *
Total runoff	SSP1-2.6	2.11 *	1.32	−1.93
	SSP2-4.5		2.00 *	0.36
	SSP3-7.0		3.89 *	4.14 *
	SSP5-8.5		4.03 *	4.85 *
Snowmelt runoff	SSP1-2.6	−0.82	−0.11	0.11
	SSP2-4.5		0.89	0.57
	SSP3-7.0		0.68	1.18
	SSP5-8.5		2.25 *	2.93 *
Glacier-melt runoff	SSP1-2.6	4.57 *	2.57 *	−1.96 *
	SSP2-4.5		4.89 *	1.89
	SSP3-7.0		5.60 *	5.85 *
	SSP5-8.5		5.60 *	5.92 *

, Figure 10). The temporal divergence of the confidence intervals in Figure 10 reveals the nonlinear modulation of the synergistic effects of the hydrological elements by the intensity of climatic forcing. The confidence intervals for both precipitation and temperature in the historical period (1981–2014) are narrower (±8% for precipitation and ±0.2 °C/decade for temperature), with high inter-model agreement, confirming the deterministic response of hydrological processes under the constraints of observational data. The significant increase in precipitation (Z = 2.28*) coupled with persistent temperature rise (Z = 5.39*) during this period synergistically drives the steady runoff growth. However, future projections reveal escalating divergence in hydrological responses across emission scenarios: under the high-emission scenario (SSP5-8.5), the precipitation increase reaches 41.8% (Z = 4.10*), yet its confidence interval widens nearly three-fold (±22%) compared to historical baselines, likely due to heightened model disagreements on water vapor flux intensification under strong radiative forcing. Concurrently, temperature projections exhibit emission-dependent uncertainty amplification, with confidence intervals expanding from ±0.3 °C/decade (SSP1-2.6) to ±0.7 °C/decade (SSP5-8.5), despite maintaining statistically robust warming trends (Z = 6.82*). This pattern underscores the pathway-sensitive nature of both warming rates and hydroclimatic uncertainties.

The synergistic effects of precipitation and temperature on runoff demonstrate marked temporal and scenario dependencies. Under high-emission scenarios (SSP5-8.5) in the near-future period (2031–2060), the maintained coupling between precipitation increase (Z = 2.28*) and temperature rise (Z = 5.39*) drives runoff growth through dual pathways: direct surface replenishment by rainfall and indirect snowmelt acceleration (snowmelt runoff Z = 2.25*), collectively elevating the runoff confidence interval’s central value by 28% despite widening uncertainty (±15% to ±20%). This linear synergy transitions to nonlinear amplification in the far-future period (2061–2090) when temperatures exceed glacial ablation thresholds (>2 °C). Extreme precipitation trends (Z = 4.10*) combine with accelerated icemelt (snowmelt Z = 2.93*), producing a highly dispersed runoff confidence interval (central value +180%, width ±45%), indicative of threshold-driven systemic instability. Conversely, the low-emission scenario (SSP1-2.6) exhibits long-term reversal dynamics: precipitation and temperature shift to negative trends, confirming the effectiveness of high emission reductions in regulating the climate system, which together reduce glacier-melt (Z = −1.96*) and total runoff (Z = −1.93*) with narrowed uncertainty (±12%). Intermediate scenarios reveal transitional mechanisms: SSP3-7.0 maintains runoff growth (Z = 4.14*) through sustained precipitation increases (Z = 3.64*) and enhanced snowmelt (Z = 2.93*), whereas SSP2-4.5 experiences precipitation trend reversal (Z = −0.21) alongside decelerated warming (Z = 5.82 to 4.17), resulting in stagnant runoff (Z = 0.36).

The transition from precipitation-dominated runoff (historical) to temperature-controlled glacier-melt amplification (SSP5-8.5) mirrors threshold-dependent shifts in cryosphere-fed basins [59,60]. The 42% summer meltwater contribution under SSP5-8.5 versus 32% historically is consistent with accelerated glacial retreat rates [61]. Non-summer periods retain precipitation-dominated regulation, explaining SSP2-4.5’s stagnation when precipitation trends weaken. Although SSP5-8.5/SSP3-7.0 exhibit sustained runoff increases through snowmelt–precipitation synergies, their glacier-dependent growth faces inherent limits: persistent ice loss portends eventual runoff declines post-glacier extinction [62]. The above temporal heterogeneity suggests that the driving weights of precipitation and temperature on runoff adjust dynamically with climate forcing intensity and period. Runoff growth in the historical period was mainly dominated by precipitation, whereas the forward runoff variability of the high-emission scenario was more controlled by temperature-triggered snow- and icemelt. To quantitatively resolve this temporal specificity, precipitation–temperature–runoff correlation analyses are carried out in the three phases, historical, near future, and far future, to reveal the dynamic reconstruction process of the hydrological coupling mechanism and its adaptive thresholds under the evolution of climate forcing intensity.

4.4. The Contribution Analysis to Future Changes in Runoff

For a comprehensive analysis of precipitation and temperature impacts on annual runoff depth variations in the TRB, partial correlation analysis was applied to isolate their independent effects after controlling for confounding variables. The partial correlation coefficients between runoff and precipitation (controlling for temperature, ${\mathrm{r}}_{\mathrm{Q}\mathrm{P}|\mathrm{T}}$) and runoff and temperature (controlling for precipitation, ${\mathrm{r}}_{\mathrm{Q}\mathrm{T}|\mathrm{P}}$) across different scenarios and periods are summarized in

Table 4. Partial correlation coefficients between runoff and precipitation/temperature under future climate scenarios (* indicates significance at the p < 0.05 level).

Scenarios	Near Future (2031–2060)		Far Future (2061–2090)
	${\mathbf{r}}_{\mathbf{Q}\mathbf{P}|\mathbf{T}}$	${\mathbf{r}}_{\mathbf{Q}\mathbf{T}|\mathbf{P}}$	${\mathbf{r}}_{\mathbf{Q}\mathbf{P}|\mathbf{T}}$	${\mathbf{r}}_{\mathbf{Q}\mathbf{T}|\mathbf{P}}$
SSP1-2.6	0.941 *	0.675 *	0.966 *	0.461 *
SSP2-4.5	0.948 *	0.488 *	0.923 *	0.171
SSP3-7.0	0.931 *	0.604 *	0.910 *	0.759 *
SSP5-8.5	0.932 *	0.488 *	0.891 *	0.777 *

, highlighting emission-dependent divergence in dominant drivers. Under low-emission pathways (SSP1-2.6 and SSP2-4.5), partial correlation results confirm that precipitation remains the dominant regulator of runoff variability. Specifically, the partial correlation between runoff and precipitation (controlling for temperature, ${\mathrm{r}}_{\mathrm{Q}\mathrm{P}|\mathrm{T}}$) is exceptionally strong (0.941–0.966 for SSP1-2.6 and 0.948–0.923 for SSP2-4.5; all p < 0.05). In contrast, the independent temperature–runoff correlation (controlling for precipitation, ${\mathrm{r}}_{\mathrm{Q}\mathrm{T}|\mathrm{P}}$) diminishes in distant periods (e.g., ${\mathrm{r}}_{\mathrm{Q}\mathrm{T}|\mathrm{P}}=0.461$ for SSP1-2.6 and 0.171 for SSP2-4.5 in 2061–2090), aligning with findings in Section 4.2 and Section 4.3 that icemelt runoff contributions do not increase substantially in the long term. These results reinforce that under stabilized climate systems, runoff changes are primarily precipitation-driven, while temperature effects saturate over time.

For high-emission scenarios (SSP3-7.0 and SSP5-8.5), partial correlations reveal distinct mechanisms. While precipitation–runoff correlations remain robust (${\mathrm{r}}_{\mathrm{Q}\mathrm{P}|\mathrm{T}}=0.931-0.910$ for SSP3-7.0 and 0.932–0.891 for SSP5-8.5; p < 0.05), the temperature–runoff relationship strengthens markedly in later periods (${\mathrm{r}}_{\mathrm{Q}\mathrm{T}|\mathrm{P}}=0.759$ for SSP3-7.0 and 0.777 for SSP5-8.5 in 2061–2090; p < 0.05). This suggests a synergistic interplay where rising temperatures amplify runoff through accelerated glacial ablation, even as precipitation retains its primary role. Notably, the SSP2-4.5 scenario shows a nonsignificant temperature–runoff correlation (${\mathrm{r}}_{\mathrm{Q}\mathrm{T}|\mathrm{P}}=0.171$, p > 0.05) in the far future, contrasting with high-emission pathways and underscoring emission-dependent divergence in drivers.

These findings are consistent with historical studies identifying precipitation as the dominant runoff source in the TRB, while temperature exerts secondary control via glacial ablation [19,23]. Under high emissions, rising temperatures intensify meltwater contributions (e.g., summer meltwater reaching 42% under SSP5-8.5 versus 32% historically), though partial correlations indicate this process operates alongside precipitation-driven changes rather than displacing them. While glacial extinction is not projected by 2100, sustained meltwater increases signal a critical threshold: prolonged ice loss—even partial—threatens eventual runoff decline once reserves deplete.

5. Conclusions

This study employed historical simulation data (1980–2014) from eight CMIP6 global climate models, along with projections under four future climate scenarios (2015–2090) based on shared socioeconomic pathways (SSPs): SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5. The HBV-light hydrological model was applied to the TRB, using the outputs from each climate model as meteorological forcing to simulate and assess future runoff dynamics under varying climate change scenarios. The analysis yielded the following key findings:

• The TRB exhibits a distinct warming and increased precipitation trend under future climate change, characterized by increased precipitation concentrated from May to October and significant year-round temperature increases, particularly pronounced under high-emission scenarios.
• The overall trend of future yearly runoff is increasing with the forcing scenario. The increase in total runoff is concentrated in May–October, mainly due to the increase in icemelt runoff and increasing rainfall, while snowmelt runoff also shows an increase, but accounts for a smaller percentage of the total runoff and has a smaller impact on the total runoff. The average annual runoff in the watershed under the four future scenarios for 2031–2060 and 2061–2090, in descending order, was SSP3-7.0 < SSP1-2.6 < SSP2-4.5 < SSP5-8.5, with the largest increase in runoff in the SSP5-8.5 scenario.
• Precipitation becomes the main factor influencing changes in annual runoff depth, while temperature effects show scenario- and period-related complexity. In the low-emission scenario, far-reaching warming is stable and runoff changes are dominated by precipitation; in the high-emission scenario, warming is intense, melting is greater, and precipitation and temperature jointly drive runoff changes. In the long-term high-emission scenario, the amplification of runoff depth by temperature becomes significant, mainly through accelerated glacier ablation and altered precipitation patterns.

These findings highlight the substantial impact of future climate and socioeconomic policy changes on runoff dynamics in the TRB, particularly emphasizing the need for focused attention on summer runoff variations. While this study systematically analyzes runoff change characteristics in the TRB under future climate scenarios, it recognizes the complex interplay between water resources and climate change. Future research should enhance quantitative analyses of glacier and permafrost change impacts on runoff, complementing the current focus on precipitation and temperature effects. Such efforts would address critical knowledge gaps identified in foundational studies [56], particularly the need for integrated assessments of cryosphere–climate feedbacks in high-altitude regions with sparse observational data.