Response of discharge in the Upper Yellow River Basin to changing climate over the long time period (1850–2023)

Human influence has unequivocally caused global warming (IPCC 2021). Beyond natural climate variability, anthropogenic climate change has led to widespread adverse impacts, resulting in losses and damages to nature and people (IPCC 2022). A direct and significant impact of global warming is the acceleration of the water cycle through temperature-driven processes (Curry et al. 2003; Hua et al. 2013). The acceleration of the water cycle alters surface runoff characteristics by redistributing precipitation across both spatially and temporally (Field et al. 2012; Taylor et al. 2012). It cannot be ignored that anthropogenic climate change has altered local and regional runoff in various parts of the world (IPCC 2021). From the 1950s to the 2010s, river runoff exhibited decreasing trends in regions of Western and Central Africa, Eastern Asia, Southern Europe, Western North America, and Eastern Australia, while increasing trends were observed in Northern Asia, Northern Europe, and Northern and Eastern North America (Tang et al. 2016; Gudmundsson et al. 2017, 2019; Li et al. 2020; Masseroni et al. 2020). Runoff changes in different regions of China vary significantly. Since 1950, annual runoff in the Yangtze, Yellow, Haihe, Huaihe, and other river basins has shown decreasing trends, whereas runoff from the northwestern mountain pass has increased (Zhang et al. 2007; Huang et al. 2016; Liu & Du 2017). Research on regional river runoff changes under climate change provides a theoretical basis for basin water resource management and supports improvements in governance.

Integrating hydrological models (HMs) with climate models is a widely adopted approach to obtain discharge sequences for assessing climate change impacts on discharge. By incorporating multiple hydrological criteria, multiple HMs, and multiple hydrological stations in Hydrological modeling, several studies have produced notable results (Eghdamirad et al. 2019; Puertes et al. 2019; Wen et al. 2020). Incorporating multiple hydrological criteria allows for a comprehensive assessment of the reliability of simulation results (Höllermann & Evers 2017; Qin et al. 2019; Yang et al. 2022). HM ensembles are used in several studies to reduce the uncertainties associated with individual HM results (Vetter et al. 2015; Su et al. 2017; Gao et al. 2019). In most HM simulations, only the simulated runoff at the outlet is prioritized; however, evaluating HM performance across different river sections within the basin can enhance reliability by using multiple hydrological stations (Krysanova et al. 2018). Climate models have advanced considerably over the past three decades (Lambert & Boer 2001; Torres & Marengo 2014; Huang et al. 2018). The Coupled Model Intercomparison Project (CMIP) has developed the most comprehensive climate model database to date (Zhou et al. 2019). Now in its sixth phase (CMIP6), it is the preferred source for driving HMs to assess climate change's impact on discharge. Therefore, this study incorporates multiple hydrological criteria, HM ensembles, and multiple hydrological stations to construct comparable discharge sequences. CMIP6 climate model data are then applied as input for HMs to assess climate change impacts on discharge.

The Yellow River, China's second-longest river, originates in the northern foothills of the Bayan Har Mountain on the Tibetan Plateau and flows through nine provinces before entering the Bohai Sea. The Upper Yellow River Basin (UYR Basin, referring to the region above the Lanzhou station on the Yellow River), a critical area for water conservation and replenishment in the Yellow River Basin, constitutes 28.8% of the basin's total area (Meng et al. 2016; Wei et al. 2021). Furthermore, the UYR is highly sensitive to climate change. Since 1961, the warming in the UYR has exceeded the national average (Wang et al. 2018a, 2018b). Previous studies on historical climate change impact on the discharge in the UYR typically begin analysis from the earliest available observed data, which dates to 1960, covering a maximum period of 64 years (Jin et al. 2020; Wang et al. 2020; Ni et al. 2022). However, this limited temporal scope overlooks a broader historical context that could provide valuable insights into long-term climate change impacts on the discharge. Additionally, most prior studies rely on individual climates or HMs, which increase uncertainty due to model limitations. The absence of studies spanning the longer historical period of 1850–2023, especially those using ensemble modeling approaches, creates a notable gap in the literature. This extended period enables an unprecedented examination of climate change impacts on discharge that have evolved since the Industrial Revolution, providing insights unavailable from shorter-term studies. To address this gap, our study reconstructs discharge sequences for the UYR over an extended historical period (1850–2023) using both historical and pre-industrial climate simulations from CMIP6. By employing an ensemble of six Global Climate Models (GCMs) and three HMs (SWAT, HBV, and VIC), this study aims to reduce model uncertainties. Additionally, we utilize Bayesian model averaging (BMA) to integrate the results, ensuring a more reliable and comprehensive assessment of climate change impacts on discharge. BMA integrates outputs from multiple models, reducing uncertainty while providing probabilistic weighting for a nuanced understanding of discharge patterns under climate change. This study's approach fills a gap in long-term discharge research and offers a framework for assessing discharge in other climate-sensitive regions globally, enabling more effective water resource management.

The UYR has a typical continental climate and is situated in an arid to semi-arid region of northwest China. From 1961 to 2021, the average annual temperature and precipitation were −0.63 °C and 488.13 mm, respectively. The basin's average elevation exceeds 3,500 m, with the terrain gradually decreasing from southwest to northeast.

The observed meteorological data used in this study were collected from over 2,400 ground-based meteorological stations across China. These data were provided by the National Meteorological Information Centre of China Meteorological Administration (https://data.cma.cn/en). The dataset includes daily precipitation, mean temperature, maximum and minimum temperatures, relative humidity, near-surface wind speed, and sunshine hours from 1961 to 2021. To enhance data utility, daily meteorological data were interpolated to a spatial resolution of 0.5° × 0.5° using an anomaly approach (Wu & Gao 2013).

Daily discharge data from 1961 to 2014 at Jimai, Maqu, Tangnaihai, and Lanzhou in the UYR were obtained from the Hydrological Yearbook of the People's Republic of China –Yellow River. The Yearbook –Yellow River is an annual compilation of hydrological data organized according to standardized requirements (Hydrology Bureau Ministry of Water Resources, P. R. China 1961 − 2014). The UYR saw the commencement of operations of two major reservoirs: the Liujiaxia Reservoir in 1969 and the Longyangxia Reservoir in 1987. These reservoirs contribute 97% of the regulating capacity in the UYR (Peng et al. 2018; Bian et al. 2019). Hydrological data from 1962 to 1968 were used to calibrate and validate three HMs, given the minimal human impact on river discharge during that period.

The climate model data consist of outputs from GCMs provided by CMIP6, which contain both the ‘historical simulation’ and the ‘piControl’. The CMIP6 models were chosen as they represent the latest advancements in climate modelling. These data form the foundation for reliable hydrological simulations in the UYR. Daily data for precipitation, mean temperature, maximum, and minimum temperature, relative humidity, near-surface wind speed, and surface downward shortwave radiation from 1850 to 2023 were collected from the historical simulation and the piControl to drive HMs. The historical simulation and the piControl, respectively, represent the scenarios with or without climate change (Gao et al. 2019). The historical simulation spans 1850–2014, and the period from 2015 to 2023 is supplemented by shared socio-economic pathway (SSP)2–4.5 scenario data, medium radiative forcing scenarios in CMIP6, chosen for its alignment with China's current socio-economic development pathways and representation of a moderate emission trajectory (Institute of Climate Change and Sustainable Development of Tsinghua University et al. 2022). This makes it suitable for extending the historical simulation while maintaining time-scale consistency with the piControl experiment. The piControl is conducted under conditions representative of the pre-industrial, with 1850 as the reference year (Eyring et al. 2016). In this study, the selected period for the piControl is 1850–2023. Currently, CMIP6 includes fifty-three climate models; of these, six models with all required variables and suitable simulation periods were selected for this study. The selected models are CMCC-ESM2, INM-CM4-8, IPSL-CM6A-LR, MPI-ESM-1-2-HAM, MPI-ESM1-2-HR, and MPI-ESM1-2-LR (Table 1). The discharge response to climate change in the UYR is assessed across six time periods: 1850–1880, 1881–1910, 1911–1940, 1941–1970, 1971–2000, and 2001–2023.

Model outputs were statistically downscaled using spatial disaggregation (SD) (Epstein & Ramírez 1994) and bias-corrected using Equidistant Cumulative Distribution Functions (EDCDF) (Li et al. 2010) to a uniform 0.5° × 0.5 ° resolution. SD was applied to downscale coarse GCM outputs to the resolution needed for the UYR's complex terrain. SD preserves spatial covariance structures, which are crucial for capturing local climate variability. This approach ensures that the local-scale responses are accurately reflected in the HMs, offering a better representation of spatial heterogeneity across the basin. Studies have shown that SD effectively improves climate data accuracy in regions with significant topographical variability (Wilby & Wigley 1997; Fowler et al. 2007). EDCDF was applied for bias correction to ensure that simulated climate variables, particularly precipitation and temperature, align with observed values. EDCDF adjusts both the mean and distribution of climate data, capturing central tendencies and extremes. This method is computationally efficient and has proven effective in reducing biases across diverse climate scenarios, making it ideal for enhancing the reliability of hydrological simulations (Piani et al. 2010; Teutschbein & Seibert 2012).

The geospatial data used in this study include the digital elevation model (DEM), soil data, and land use data. The DEM, with a resolution of 90 m, is derived from the Shuttle Radar Topography Mission database (http://srtm.csi.cgiar.org/). The soil data, with a resolution of 1 km, is sourced from the Harmonized World Soil Database (Fischer et al. 2008), which uses the FAO-90 soil classification system. The land use data, also with a resolution of 1 km, is obtained from the Resources and Environmental Science Data Center, Chinese Academy of Sciences (http://www.resdc.cn). It is based on human-visually interpreted Landsat 8 remote sensing images. The temporal reference for the data used in this study is 1960. The land use classes follow a three-level classification system, with the broadest level including forest, grassland, barren land, cropland, urban land, and water. The verification accuracy of these data exceeded 95%, based on random selections conducted twice in the Sanjiangyuan region (Yellow River source) between 2004 and 2005 (Ning et al. 2018).

The SWAT, HBV, and VIC HMs were chosen for their ability to simulate discharge processes across diverse climatic conditions and their extensive validation in large basins. This ensemble approach reduces model-specific uncertainties, enhancing discharge simulations.

The Soil and Water Assessment Tool (SWAT) is a HM developed by the United States Department of Agriculture in the 1990s. It is based on physical principles and provides continuous simulations of runoff and other related processes. The SWAT model can compute all components of water balance, including assessing climate change impacts on runoff in a given watershed. The SWAT uses a combination of DEM, soil data, meteorological data, and land use data to simulate the water balance (Arnold et al. 1998).

The SWAT is a highly efficient HM for water resource management, environmental planning, and decision-making processes (Femeena et al. 2020; Chen et al. 2021; Ayalew et al. 2023; Wendell et al. 2024). Stakeholders can use it to evaluate climate change impacts on runoff and identify strategies to enhance water resource sustainability in watersheds (Tigabu et al. 2022).

The VIC model simulates key components of the water cycle, including precipitation, evapotranspiration, snow accumulation and melt, soil moisture, and runoff. Like the SWAT, VIC also uses DEM, soil data, meteorological data, and land use/cover data as inputs.

The VIC model has been widely used in flood forecasting, ecological modeling, water resource management, and climate change impact assessments. A key feature of the VIC model is its versatility, allowing for customization to suit different research fields and objectives.

The Hydrologiska Byråns Vattenbalansavdelning (HBV) model is a conceptual HM developed by the Swedish Meteorological and Hydrological Institute in the 1970s (Bergström & Forsman 1973; Krysanova et al. 1999). In this model, the river basin is divided into multiple sub-basins based on underlying surface conditions, such as elevation and land use type, with each sub-basin treated as a single, homogeneous unit for hydrological simulation. The model uses a simple time-lag method for routing. The input data of HBV are similar to those for SWAT and VIC.

The HBV model has been applied across a wide range of catchments from small headwater streams to large river basins and has been adapted to various hydrological and climatic conditions (Seibert & Bergström 2022). Its simplicity and user-friendliness make it an ideal choice for hydrological modeling.

Despite differences in complexity, mathematical process formulation, and spatial resolution, all three models have been widely applied to a broad range of hydrological issues and climate impact studies.

A comprehensive model evaluation method (Krysanova et al. 2018; Wen et al. 2020) is applied to assess the performance of HMs during calibration and validation. The method includes the following steps:

• (1) Verify the accuracy of the meteorological input data, focusing particularly on temperature and precipitation.

• (2) Ensure consistency in the basin's internal properties by calibrating and validating discharge at all four hydrological stations simultaneously. Evaluate the performance of the three HMs over historical periods by comparing observed and simulated discharges. The model parameterization is considered more reliable when observed and simulated discharges show high consistency.

• (3) Use Nash–Sutcliffe efficiency (NSE), Kling–Gupta efficiency (KGE), and the ratio of root mean square error to the standard deviation of the observations (RSR) to evaluate the performance of the three HMs. Criteria for ‘good’ simulation include NSE ≥ 0.7, KGE ≥ 0.7, and RSR ≤ 0.6 (Gao et al. 2019).

Using these methods, we obtained the following results on temperature, precipitation, and discharge patterns in the UYR.

From 1961 to 2014, the multi-year average temperature in the UYR was −0.8 °C, and the multi-year average precipitation was 505.5 mm. The multi-year average temperature exhibits a fluctuating upward trend, with an increase of 0.3 °C per decade. The annual precipitation shows no significant upward trend, with a tendency rate close to 0.

The annual Q10 discharge (representing the high flow, with only 10% of the discharge exceeding this value) simulated by all three HMs shows strong consistency with observations, with correlation coefficients ranging from 0.87 to 0.96. However, the annual Q90 discharge (representing low flow, with 90% of the discharge exceeding this value) was less accurately simulated, with correlation coefficients ranging from 0.52 to 0.66.

From 1850 to 2023, without the climate change impact, the multi-year average temperature in the UYR is −1.7 °C, the annual mean temperature shows no notable trend (Figure 5(a)), while the multi-year average precipitation is 543.6 mm and the annual precipitation shows a slight upward trend (Figure 5(b)). Comparing temperature and precipitation under scenarios with and without climate change, both annual mean temperature and annual precipitation show increased differences in the late 20th and early 21st centuries compared to previous periods.

Q10 and Q90 discharges are used to represent discharge extremes. Q90 represents the extremely low flow, indicating that 90% of the discharge in a given year is above this value. Conversely, Q10 represents the extremely high flow, indicating that 10% of the discharge in a given year is above this value.

From 1850 to 2023, the multi-year average extreme high flow in the UYR is 2,366.5 m³/s with climate change impact and 2,450.9 m³/s without it. The multi-year average extreme high flow with the climate change impact is 3.4% lower than without it (Figure 10(b)). Over the same period, the tendency rates of the extremely high flow in the UYR are −8.5 m³ per decade with climate change impact and −15.4 m³ per decade without it. The decrease in the extremely high flow is more pronounced with climate change impact.

For the multi-year average extreme high flow, climate change from 1850 to 1880 results in a 1.4% increase, whereas, in the five periods from 1881 to 2023, it causes a decrease. Overall, as time progresses, the climate change impact on the multi-year average extreme high flow intensifies, leading to greater decreases. The multi-year average extreme high flow shows similar decreases from 1971 to 2000 and 2001 to 2023, decreasing by 6.7 and 6.3%, respectively (Figure 11(b)).

The decreases in extreme high and low flows indicate greater water supply variability, potentially increasing the risk of both drought and flood events. These findings highlight the critical need to incorporate climate projections into regional flood and drought management strategies.

Previous studies indicate that in the UYR, the annual mean temperature has increased since 1961 when observed data became available. Additionally, annual precipitation has shown a slight upward over time (Liu et al. 2012; Wang et al. 2021; Xiao et al. 2021). These studies also show that the annual average discharge in the UYR has a downward trend due to climate change (Kong et al. 2016; Li et al. 2017; Liu et al. 2021). Unlike studies in Northern Europe and North America, where discharge has shown increasing trends due to regional climate patterns (Gudmundsson et al. 2017; Li et al. 2020), our findings show a consistent discharge decrease in the UYR, aligning with trends observed in other Asian basins. This underscores regional variability in climate change's impact on hydrology and highlights the unique sensitivity of the UYR to climate change compared to other basins globally. Our work is consistent with earlier findings based on available observed data but also extends the duration back to the Industrial Revolution, capturing long-term changes in temperature, precipitation, and discharge in the UYR. Insights from this study provide a valuable basis for adaptive water management strategies. Examining long-term climate change impacts on discharge offers critical information for water resource managers, aiding sustainable water allocation, especially in regions such as the UYR where water scarcity is a concern. This information can support drought planning, reservoir operation optimization, and effective water demand management policies.

This study primarily assesses climate change's impact on discharge in the UYR through the difference between the historical simulation and the piControl. The historical simulation includes anthropogenic climate forcing, natural climate forcing, and climate internal variability, whereas the piControl includes only climate internal variability (Menary et al. 2018; Wang et al. 2024). Therefore, differences in discharge between the historical simulation and piControl in the UYR are caused by anthropogenic and natural climate forcing. Comparison of discharge differences between the historical simulation and the piControl is a proven method for studying the impact of climate change on discharge and has been applied in various basins (Forbes et al. 2019; Kalugin 2022; Wang et al. 2024). While anthropogenic climate forcing plays a dominant role in global climate change, determining the primary contributor to discharge changes in this specific study area requires further research. To distinguish the impacts of anthropogenic climate forcing, natural climate forcing, and climate internal variability, an additional scenario with only natural climate forcing and climate internal variability could be introduced, alongside the historical simulation and the piControl. This approach enables more precise attribution of climate change impacts on discharge.

While this study primarily investigates climate change impacts on discharge by analyzing differences between historical simulations and piControl, we recognize that land use changes, such as urbanization, deforestation, and agricultural expansion, also play a role in shaping hydrological processes in the UYR. Future research could integrate land use data with climate scenarios to better distinguish between discharge changes driven by climate forcing and those related to land use changes. This would provide a more detailed attribution of discharge variability, especially as rapid development and land conversion continue to impact the region. Integrating land use impacts with climate change analysis could offer deeper insights into effective water resource management strategies in the UYR. This research primarily addresses climate-forcing impacts without incorporating potential socio-economic changes and land use patterns. Future research could expand on these aspects to improve discharge simulation accuracy and facilitate a more integrated approach to water management, which is essential for adapting to ongoing climate and land use changes in the region. This research also has direct implications for the local communities and agricultural sectors in the UYR, where water availability is crucial for livelihoods. The analysis of discharge trends, particularly extreme low and high flows, can aid farmers and community planners in anticipating periods of water scarcity or flood risks, enabling proactive measures to protect crops and manage resources. Understanding potential seasonal and annual variations can guide crop selection, irrigation planning, and disaster preparedness.

Integrating climate models and HMs is essential for studying the climate change impact on discharge. Although the downscaled and bias-corrected climate model data used in this study are suitable for the UYR, inherent structural differences among models mean that some uncertainty remains. Additionally, this study assumes that the optimal parameters calibrated and validated for the historical period remain valid over the entire simulation duration. The calibration and validation of HMs aim to meet the criteria of NSE, KGE, and RSR; however, other criteria are not addressed. Regardless of the HM used, these models simplify complex hydrological phenomena and cannot fully replicate the basin's hydrological processes. Although three HMs were used and the BMA method was applied to integrate results, the improvement in result reliability remains limited. On a larger scale, this study provides valuable data for policymakers developing climate adaptation and mitigation plans. The findings underscore the increasing climate change impact on discharge patterns, stressing the need for policy frameworks that incorporate long-term hydrological changes. This approach, combining ensemble modeling with BMA, could serve as a model for other basins facing similar climate challenges, supporting robust, evidence-based climate adaptation policies across regions.

This study reconstructed the UYR discharge sequences from 1850 to 2023, both with or without climate change impact, and quantitatively assessed climate change impacts on annual average discharge, monthly discharge, and discharge extremes. The research framework presented can also serve as a reference for studies in watersheds with similar climate and hydrological characteristics. Additionally, as the piControl in some climate models extends to 2100, the simulated discharge under piControl can be compared with that under different SSPs to assess climate change impacts on discharge across scenarios. The methodology and findings of this study offer valuable insights for future research in hydrology and climate science. This approach, focusing on long-term historical periods, provides a framework for evaluating climate adaptation strategies and monitoring mitigation efforts aimed at reducing climate impacts on hydrology. Overall, this study provides critical insights for sustainable water resource management, proactive climate adaptation strategies, and effective policy planning for climate-sensitive regions like the UYR.

In summary, this study underscores significant shifts in UYR hydrological patterns due to climate change, marked by decreasing discharge trends, altered seasonal discharge patterns, and intensified decreases in extreme flows over the past century and a half. These findings emphasize the increasing impact of climate variability on water resources and the urgent need for adaptive strategies to address potential water scarcity risks. Based on the detailed analysis, the following summary and conclusions consolidate the key outcomes of our research, offering a basis for future studies and water management initiatives in climate-sensitive regions such as the UYR.

This paper uses six GCMs from CMIP6 to drive optimal parameter-fixed HMs, studying climate change impacts on discharge in the UYR from 1850 to 2023. The three HMs were calibrated and validated using a comprehensive model evaluation method to obtain the optimal parameters that are suitable for the entire basin. After identifying optimal parameters, six downscaled and bias-corrected GCMs, including the historical simulation (under climate change) and the piControl (with GHGs, aerosols, ozone, and solar irradiance fixed at 1850 levels), were used as inputs for the three HMs. BMA was then used to average the output of the three HMs, providing the final daily simulated discharge. The evaluation was conducted over six time periods: 1850–1880, 1881–1910, 1911–1940, 1941–1970, 1971–2000, and 2001–2023. Finally, the climate change impact on annual discharge, monthly discharge, and discharge extremes was evaluated. The main conclusions are as follows:

From 1850 to 2023, global climate change significantly impacted the UYR, causing marked shifts in regional temperature, precipitation, and discharge patterns. The annual mean temperature rose by 0.1 °C per decade, while precipitation decreased by approximately 1.2 mm per decade, with the most pronounced changes occurring in the late 20th and early 21st centuries. With climate change impact, the annual average discharge shows a stronger decreasing trend of −8.6 m³/s per decade, compared to −4.1 m³/s per decade without climate change impact. Over the six time periods since 1850, discharge decreases have become progressively severe, with decreases of up to 6.8% from 1971 to 2000 and 6.6% from 2001 to 2023, as simulated in the historical simulation compared to the piControl. These findings underscore the escalating climate change impact on hydrological patterns in the UYR, particularly in the most recent periods.

From 1850 to 2023, climate change has impacted monthly discharge patterns in the UYR, with increases observed in March, April, and May and decreases in other months. October experienced the most significant decreases across all periods, with the change rate reaching 24.8% from 2001 to 2023, as simulated in the historical simulation compared to the piControl. This trend indicates a shifting seasonal discharge pattern, where early spring months see increased discharges, while late summer and autumn months face substantial decreases. These shifts are particularly pronounced in recent decades, underscoring the intensifying climate change impact on seasonal discharge variability in the UYR.

From 1850 to 2023, climate change has progressively decreased extreme low (Q90) and high flows (Q10) in the UYR. The multi-year average extreme low flow decreased by 2.8% with climate change, and extreme high flow fell by 3.4%. The tendency rates show that climate change accelerates the decrease of discharge extreme, with extreme low flows decreasing by −1.2 m³ per decade and extreme high flows by −8.5 m³ per decade. This trend is particularly evident in recent decades, with decreases of around 8% for low flows and 6.5% for high flows from 1971 to 2023, as simulated in the historical simulation compared to the piControl, underscoring a significant intensifying impact over time.

These findings highlight the pressing need for adaptive water management strategies in the UYR, as intensified climate change is expected to heighten water scarcity risks, particularly in the months of late summer and autumn. The observed trends suggest that, if current climate change patterns persist, the UYR may encounter increasing challenges in water management. Furthermore, although this study provides robust insights into long-term hydrological changes, inherent model uncertainties suggest that future studies should refine techniques and incorporate broader climate variables to enhance discharge simulations.

This study is financially supported by the International Cooperation Program between the National Science Foundation of China (NSFC) and the United Nations Environment Program (UNEP) (Grant no. 42261144002). The first author is thankful the PhD scholarship provided to study modeling hydrological process in the Kiel University for 2 years by the China Scholarship Commission. The authors are thankful for the support from the High-level Talent Recruitment Program of Nanjing University of Information Science and Technology (NUIST). The authors would like to thank the World Climate Research Program's working group on coupled modeling for producing and making available their model simulation and projection.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.