Drivers and Future Regimes of Runoff and Hydrological Drought in a Critical Tributary of the Yellow River Under Climate Change

Abstract

China’s Yellow River basin encounters widespread risks of reduced runoff and intensified hydrological drought. This study focuses on the middle and upper reaches of the Dahei River, the Yellow River’s primary tributary. In this region, the Soil & Water Assessment Tool (SWAT) hydrological model was employed to simulate hydrological processes, identify runoff changes and hydrological drought characteristics, and conduct attribution analysis from 1983 to 2022, as well as to project trends over the next 40 years. The results indicate that total runoff during the impact period (1999–2022) decreased by 55.26% compared to the baseline period (1983–1998). Climate change accounted for a contribution rate of 38.6% to this decline, while human activities accounted for 61.4%. Additionally, climate primarily altered surface runoff (SURQ) and lateral groundwater flow (LATQ) through precipitation changes, while land use had a predominant influence on total runoff volume by modifying SURQ. Both factors exhibited relatively minor effects on LATQ. Moreover, human activities contribute to hydrological drought at a rate of 36.11% to 94.25%. Drought probability is significantly influenced by climate through precipitation and temperature changes, while land use primarily mitigates hydrological drought by impacting the three runoff components. It is predicted that over the next 40 years, total runoff will decrease by 2.08% to 60.16%, along with hydrological droughts that are more frequent, longer in average duration, and more intense; however, the Maximum Drought Duration is anticipated to shorten. In the east and northeast, hydrological drought presents a trend of intensification, with central and western regions exhibiting weaker or declining changes.

1. Introduction

The Yellow River Basin in China connects the Qinghai–Tibet Plateau, Loess Plateau, and North China Plain. As a core economic corridor for China’s Belt and Road Initiative, the basin serves as a vital link, promoting socioeconomic development across eastern, central, and western regions. Therefore, it is imperative to maintain the basin’s environmental stability for national socioeconomic progress and ecological security. Situated at the upper reaches of the Yellow River, the Dahei River Basin is one of the primary grain-producing regions in Inner Mongolia and constitutes an important economic belt within the Yellow River Basin. Consequently, it is necessary to gather scientific evidence and quantitative support in the middle and upper reaches of the Dahei River Basin. This will help to optimize water resources, formulate drought mitigation and disaster reduction decisions, and implement land use planning in the specified region.

The complex interplay between climate, land use, and hydrological processes makes it challenging to quantify the runoff changes they induce [1,2,3]. Numerous methodologies have been employed to evaluate the impacts of climate and land use change on runoff, chiefly using statistical analysis and hydrological modeling [4,5,6]. For instance, in the Jinghe River basin, distributed hydrological modeling identified the primary drivers of runoff variation, revealing that land use change contributed 71% to runoff variability [7]. Similar findings have been reported in the middle Yellow River in China [8], the Qinhuai River basin [9], the Tarbela catchment in Pakistan [10], and the Taral River basin in Iran [11]. Conversely, other studies, such as those on the upper Indus River near Besham in Pakistan, indicate that land use change and climate change contribute 2.53% and 97.47% to river runoff variation, respectively [12]. Comparable results have been obtained in China’s Changyang River basin [13], the Yellow River headwaters [14], the Jingjiang River basin [15], and India’s Dharoi catchment within the Sabarmati River basin [16]. Collectively, these studies demonstrate that climate change generally exerts a stronger influence on runoff variability than land use change.

The two key drivers, climate change and land use change, affect global and regional hydrological cycles through complex mechanisms. Climate change restructures the spatiotemporal distribution of water and energy, leading to alterations in precipitation patterns, evaporation rates, and river hydrological conditions [17]. It also intensifies issues such as uneven water distribution and alternating periods of drought and flooding, particularly in arid and semi-arid regions where hydrological processes are highly sensitive to climate variability [18,19]. The impacts of future climate variability on basin runoff remain uncertain, posing potential risks to agriculture, ecosystems, and human settlements, and calling for further research to inform policy and planning aimed at mitigating adverse effects. The Second Assessment Report of Working Group II (WG II) has highlighted the disturbing consequences of climate change, including increased frequency and intensity of compound risks and extreme events [20]. Climate change renders disaster risks increasingly transregional and trans-systemic [21], creating new challenges for disaster prevention and mitigation.

Land use change modifies watershed surface characteristics and thereby alters runoff generation processes during precipitation events, including infiltration, water storage, evaporation, and channel routing [22]. These effects are particularly pronounced in developing countries, where land use patterns are rapidly shifting. Runoff responses differ across temporal scales, depending on factors such as urbanization, cropland expansion, and forest loss [23]. However, most existing studies rely primarily on historical meteorological and land use datasets and pay limited attention to potential future changes. Therefore, it is essential to project future climate and land use change to better understand their impacts on hydrological processes.

Total runoff, a key component of the water cycle, comprises three primary elements: surface runoff (SURQ) entering rivers, groundwater flow (GWQ) entering rivers, and lateral flow (LATQ) entering rivers. Each component, playing a distinct role in hydrological processes, is influenced by climatic factors and basin bedrock characteristics. This can alter the spatiotemporal distribution patterns of water resources and increase the frequency of extreme hydrological events, thereby presenting severe challenges to the security and sustainable management of water resources. The independent and combined impacts of these factors on runoff and hydrological drought must be quantified to develop effective adaptation strategies. In northern China, where water scarcity and frequent droughts are common, such quantification is crucial [24,25,26,27]. Current studies on the evolution of water resource patterns in response to changing conditions and the evaluation of drought and flood disaster risks have predominantly focused on large river basins. However, research into small- and medium-sized basins remains relatively scarce. Alterations in basin water cycles and drought/flood disasters directly impact regional agricultural production; this, in turn, affects local livelihoods and obstructs the region’s secure and stable socioeconomic development [28,29,30].

This study focuses on the Dahei River Basin’s middle and upper reaches, located in an arid–semi-arid region. It aims to quantitatively analyze the impacts of climate and land use change on runoff and hydrological drought by employing the influencing factor separation method. Attribution analysis will be conducted for changes in runoff and various hydrological drought characteristics. Moreover, the study will integrate global climate models (GCMs) with the patch-level land use simulation (PLUS) model to predict runoff and hydrological drought trends under various scenarios over the next 40 years.

2. Materials and Methods

2.1. Study Area Overview

The Dahei River, located between 40°27′ N and 41°26′ N latitudes and 110°20′ E and 112°50′ E longitudes, is the upper Yellow River’s terminal first-order tributary. The region exhibits a continental semi-arid climate and has a basin area of 18,441 km² and a main stem length of 238 km. This study selected the basin’s middle and upper reaches, controlled by the Meidai and Sanliang hydrological stations, as the study area (Figure 1). The study area covers approximately 8173 km² of the basin area. The terrain moves from high in the north to low in the south, with elevations of 970–2240 m. The northern part is primarily hilly and mountainous, whereas the majority of the southern region comprises alluvial plains. The study area has a typical arid and semi-arid continental climate with distinct seasons. Additionally, the area experiences a short frost-free period, abundant sunshine, strong evaporation, frequent droughts and sandstorms, and annual precipitation of 330–460 mm. The long-term values for average evaporation, average air temperature, and average wind speed are 1826 mm (φ20 cm evaporation pan), 7.7 °C, and 1.8–4.0 m/s, respectively [31]. All data sources are provided in Section 2.2.

2.2. Data Sources and Preprocessing

Meteorological data were sourced from the China Resource and Environment Science Data Center (https://www.resdc.cn/); this consisted of records from seven stations in the study area from 1980 to 2022. Hydrological data were taken from monthly runoff records from the Meidai and Sanliang hydrological stations in the study area for 1980 to 2022. The China Geospatial Data Cloud (http://www.gscloud.cn) provided digital elevation model (DEM) data at a resolution of 12.5 m, while the Chinese Academy of Sciences Resource and Environment Science and Data Center presented spatial distribution data for land use, population density, and gross domestic product (GDP). Furthermore, land use data included 30 m grid data taken from 1990, 2000, and 2020. ArcGIS 10.2 (Esri, Redlands, CA, USA) reclassification tools were employed to categorize land use types into six primary land classes: arable land, forest land, grassland, water bodies, construction land, and unutilized land (Figure S2). The Harmonized World Soil Database (HWSD) published by the Food and Agriculture Organization (FAO) of the United Nations (UN) (http://www.fao.org) provided soil type maps at a resolution of 1000 m. The necessary soil parameters were calculated to construct the Soil & Water Assessment Tool (SWAT) model soil database [32]. Figure 1 presents the spatial distribution of soil types. The road data for all categories were provided by the OpenStreetMap website (https://www.openstreetmap.org/), and county government seat coordinates were sourced from the coordinate extraction system of Baidu Maps (https://api.map.baidu.com/lbsapi/getpoint/). The Euclidean distance tool in ArcGIS 10.2 was used to calculate the distances from each study site to primary roads, secondary roads, tertiary roads, and county government seats. The projections of GCMs were sourced from the Earth System Grid Federation (https://esgf.llnl.gov). This study selected three Sixth Coupled Model Intercomparison Project (CMIP6) GCMs with a strong track record of simulating China’s climate: ESM2-0, GFDL-ESM4, and EC-Earth3-veg-LR [33,34]. Three Shared Socioeconomic Pathways (SSPs), assessing the impacts of different climate change scenarios on basin hydrological processes, were considered: SSP1-2.6, SSP2-4.5, and SSP5-8.5. The Delta downscaling method was employed to downscale and correct the results for daily precipitation amount and temperature from 2025 to 2064, followed by a multi-model ensemble (MME) approach for subsequent SWAT model projections of future climate change scenarios. The PLUS model simulated and projected future land use change in the Dahei River Basin’s middle and upper reaches. Drawing upon land use data from 2010 and 2015, the model simulated 2020 land use patterns, validating accuracy through field measurements. Subsequently, it projected the spatiotemporal distribution of land use for 2045 under three scenarios: Business-As-Usual (BAU), Ecological Protection (EP), and Economic Development (ED).

2.3. Methods

2.3.1. SWAT Model Hydrological Processes’ Simulation, Model Calibration, and Validation

Developed by the Agricultural Research Service of the United States Department of Agriculture, the SWAT model is a distributed hydrological model, with its water balance equation expressed as:

$$S{W}_t=S{W}_0+\sum _{i=1}^t\left(R_{day}-Q_{surf}-E_a-W_{seep}-Q_{gw}\right)$$

Here, $S{W}_t$ is the soil water content at time $t$ in mm; $S{W}_0$ is the initial soil water content in mm; $t$ is the time step; $R_{day}$ is the precipitation amount on day $i$ in mm; $Q_{surf}$ is the surface runoff on day $i$ in mm; $E_a$ is the evaporation on day $i$ in mm; $W_{seep}$ is the infiltration and lateral flow within the soil profile on day $i$ in mm; and $Q_{gw}$ is the groundwater content on day $i$ in mm. Certain steps must be taken to establish the SWAT model: a DEM must be imported, and threshold values must be set to generate the watershed river network and sub-basins (Figure 1). The hydrologic response unit (HRU) is the model’s smallest computational unit; to generate it, land use types and soil types were used. Based on the natural river network and topography, the study area was divided into 42 sub-basins comprising a total of 679 HRUs to achieve reasonable water and sediment flow. Using land use data from 2000 and climate date from 1980–2022, a SWAT project was established as the local model. During the process of model calibration, the first three years of simulated outputs were treated as a warm-up period in order to minimize the impact of user-estimated parameter values [35]. After completing the calibration and validation of the SWAT model, this study directly used the standard model outputs of runoff components for analysis. Surface runoff (SURQ), lateral flow (LATQ), and groundwater runoff (GWQ) were all automatically computed by SWAT based on the soil water balance and flow routing processes. Specifically, under the unified set of calibrated parameters, monthly SURQ, LATQ, and GWQ outputs for each HRU were extracted and spatially aggregated at the subbasin scale and at the basin outlet for composition analysis and attribution assessment. The sum of these components is mass-conservative with the simulated total runoff, thereby ensuring consistency between the runoff components and the main runoff simulation results.

The observed discharge series (1983–2022) was split into a calibration period (1983–2014) and a validation period (2015–2022). The longer calibration period (80% of the record) was selected to ensure robust parameter estimation across a wide range of hydrological conditions, while the most recent years were reserved as an independent validation period to test the transferability of the calibrated parameters under recent climate and human influences. Model performance for both sub-periods was evaluated using NSE and R². This study employed the Sequential Uncertainty Fitting Program (SUFI-2) within the SWAT-CUP 2012 (Abbaspour, Zurich, Switzerland) software for calibration and uncertainty analysis to choose the 20 most sensitive parameters for model calibration (

Table 1. List of sensitive parameters calibrated based on global sensitivity analysis.

Order	Parameter	Meaning (with Units)	Best Parameters	t-Stat	p-Value
1	ESCO	Soil evaporation compensation factor: adjusts the depth in soil from which evaporation demand is met (dimensionless).	0.05	3.73	0.0008
2	SMFMN	Snowmelt factor for winter: the minimum melt rate of snow per degree-day (mm/°C-day).	16.44	−2.69	0.01
3	CN2	SCS runoff curve number for average moisture condition (II): an index of runoff potential for soil and land cover (dimensionless).	−0.26	−2.16	0.04
4	CH_N2	Manning’s roughness coefficient for the main channel: influencing flow resistance in the channel (dimensionless).	0.02	−2.07	0.05
5	SOL_K	Saturated hydraulic conductivity of the soil layer: measuring the ease of water movement through saturated soil (mm/h).	−0.87	−1.77	0.09
6	ALPHA_BF	Baseflow alpha factor: the groundwater baseflow recession constant that controls the rate of baseflow decline (1/day).	−0.15	0.95	0.34
7	HRU_SLP	Average slope steepness of the HRU (m/m).	−0.41	−0.94	0.35
8	SOL_BD	Soil bulk density: mass of soil per unit volume (g/cm3).	1.27	0.93	0.36
9	GWQMN	Threshold water depth in the shallow aquifer required for return flow to occur (m/m).	0.56	0.77	0.45
10	CANMX	Maximum canopy storage: the maximum water that can be held on the vegetation canopy (m/m).	0.78	−0.73	0.47
11	SOL_AWC	Available water capacity of the soil layer: fraction of water that can be stored in soil for plants (mm/mm).	0.82	−0.63	0.54
12	OV_N	Manning’s roughness coefficient for overland flow (dimensionless).	0.27	0.50	0.62
13	SLSUBBSN	Average slope length for overland flow; the distance of sheet flow before runoff concentrates into channels (m).	−0.24	0.42	0.67
14	EPCO	Plant uptake compensation factor: adjusts how deeply plant roots can draw water (dimensionless).	0.51	0.34	0.74
15	GW_REVAP	Groundwater re-evaporation coefficient: controls the fraction of water moving from the shallow aquifer up to the root zone (dimensionless).	0.77	−0.14	0.88
16	TIMP	Snowpack temperature lag factor: controls the influence of the previous day’s snowpack temperature on today’s (dimensionless).	0.57	0.14	0.89
17	GW_DELAY	Groundwater delay time: the lag between water percolation from the soil and its recharge to the shallow aquifer (days).	606.16	0.13	0.90
18	USLE_P	USLE support practice factor: ratio of soil loss with a given conservation practice to the loss with conventional farming (dimensionless).	1.65	−0.12	0.92
19	SFTMP	Snowfall temperature threshold: the mean air temperature at which precipitation is equally likely to be rain or snow (°C).	−0.87	0.09	0.93
20	REVAPMN	Threshold water depth in the shallow aquifer required for upward flow to soil or percolation to the deep aquifer (mm).	806.70	0.01	0.99

). Among these, the top eight parameters with high sensitivity were identified for simulating runoff: ESCO, SMFMN, CN2, CH_N2, SOL_K, ALPHA_BF, HRU_SLP, and SOL_BD.

To compare and evaluate the fit between observed and simulated data, the Nash–Sutcliffe (NS) efficiency and correlation coefficient (r²) were employed. The equations are presented below [36]:

$$NS=1-\sum _{i=1}^n{\left(Q_{obs}-Q_{sim}\right)}^2/\sum _{i=1}^n{\left(Q_{obs}-mean\left(Q_{obs}\right)\right)}^2$$

Here, $Q_{obs}$ indicate the observed data, $Q_{sim}$ refers to the simulated data, and n denotes the total number of data records. If the NS value is closer to 1, the simulation will be more accurate. $NS$ being negative indicates that the model’s prediction performance is inferior to the mean of the observed values.

$$r^2={\left[\sum _{i=1}^n\left(Q_{obs,i}-{\overline{Q}}_{obs}\right)\left(Q_{sim,i}-{\overline{Q}}_{sim}\right)\right]}^2/\sum _{i=1}^n{\left(Q_{obs,i}-{\overline{Q}}_{obs}\right)}^2\sum _{i=1}^n{\left(Q_{sim,i}-{\overline{Q}}_{sim}\right)}^2$$

Here, $Q_{obs}$ and $Q_{sim}$ represent the observed and simulated data, respectively, and $i$ refers to the $i$-th observed or simulated data point. Similar to $NS$ mentioned above, the closer $r^2$ is to 1, the more accurately the model will simulate the observed data. During the calibration period, once the objective function reaches a satisfactory value, the selected parameter values will be applied to test the simulation performance during the validation period.

2.3.2. Pettitt Change-Point Test

The Pettitt test does not rely on distributional assumptions, is suitable for data that are non-normally distributed or exhibit irregular variations, and can accurately identify a single change point, making it appropriate for detecting abrupt shifts in time series. In this study, the test is used to objectively determine the year of a significant change point in the 1983–2022 annual mean runoff series at the basin outlet (Sanliang hydrological station), thereby delineating the baseline and impact periods. This provides the essential temporal segmentation for subsequently applying the factor separation method to quantitatively analyze the effects of climate and land use changes on runoff and hydrological drought [37,38].

Assuming the occurrence of a break at time $\mathrm{t}$ in a hydrometeorological time series X, the sequence is divided into two parts: $\left(x_1,x_2,\dots ,x_t\right) \mathrm{and} \left(x_{t+1},x_{t+2},\dots ,n\right)$. The statistical measure $U_{t,n}$ is defined as follows:

$$U_{t,n}=\sum _{i=1}^t\sum _{j=1}^n\mathrm{s}\mathrm{g}\mathrm{n}\left(x_j-x_t\right)\left(\mathrm{t}=\mathrm{2,3},4,\dots ,n\right)$$

Here, $sgn\left(.\right)$ is the designated sign function. Consequently, the statistic $\mathrm{t}$ and the minimum point of $U_{t,n}$ is considered the breakpoint of the sequence.

The change point $P$’s significance level is defined as:

$$P=\mathrm{e}\mathrm{x}\mathrm{p}\left({\displaystyle \frac{-6U_{t,n}^2}{n^3+n^2}}\right)$$

If $P$ ≤ 0.05, a significant change has occurred.

2.3.3. Scenario Design

Using the Pettitt change-point test, the 1983–1998 period serves as the baseline period, thus making the 1999–2022 the impact period. Data on land use from 1990 and 2020 represent the land use patterns for these two periods. Based on projected climate for 2025 to 2064 and three land use datasets for 2045, the future projection period was established as 2025 to 2064, ensuring that both periods spanned 40 years for comparative analysis. S1–S4 are used for historical attribution, where S1 serves as the baseline scenario, S2 is used to extract the pure climate effect, S3 is used to identify the single-factor land use effect, and S4 is used to evaluate the combined and interaction effects of climate and land use. S5–S13 combine three climate pathways (SSP1-2.6, SSP2-4.5, SSP5-8.5) with three land use scenarios (BAU, EP, ED) to assess the risks of runoff and hydrological drought under different future climate–land use combinations. The specific scenarios for these cases are detailed in

Table 2. Scenario Design.

Scenario	Dataset
Baseline period scenario (S1)	1983–1998 climate and 1990 Land use
Climate change scenario (S2)	Climate from 1999–2022 and Land use in 1990
Land use change scenario (S3)	Climate from 1983–1998 and Land use in 2020
Integrated Change Scenario (S4)	Climate 1999–2022 and Land Use 2020
BAU1-2.6 Scenario (S5)	SSP1-2.6 Climate for 2025–2064 and BAU Land Use for 2045
BAU2-4.5 Scenario (S6)	SSP2-4.5 Climate Scenario for 2025–2064 and BAU Land Use for 2045
BAU5-8.5 Scenario (S7)	SSP5-8.5 Climate for 2025–2064 and BAU Land Use for 2045
EP1-2.6 Scenario (S8)	SSP1-2.6 climate for 2025–2064 and EP Land use for 2045
EP2-4.5 Scenario (S9)	2025–2064 SSP2-4.5 Climate and 2045 EP Land Use
EP5-8.5 Scenario (S10)	2025–2064 SSP5-8.5 Climate and 2045 EP Land Use
ED1-2.6 Scenario (S11)	2025–2064 SSP1-2.6 Climate and 2045 ED Land Use
ED2-4.5 Scenario (S12)	2025–2064 SSP2-4.5 Climate and 2045 ED Land Use
ED5-8.5 Scenario (S13)	SSP5-8.5 climate for 2025–2064 and ED Land use for 2045

.

2.3.4. Hydrological Drought Index

Standardized runoff index (SRI) is commonly used to analyze hydrological drought conditions in the study area. For a given period, assuming runoff volume x, the probability density function of x is expressed below:

$$g\left(x\right)={\displaystyle \frac{1}{{\beta }^{\alpha }\Gamma \left(\alpha \right)}}{x}^{\alpha -1}{e}^{{\displaystyle \frac{-x}{\beta }}}$$

Here, $\alpha$ and $\beta$ indicate the shape parameter and scale parameter ($\alpha >0, \beta >0, x >0$), respectively. Therefore, the calculation formula for $\Gamma \left(\alpha \right)$ will be:

$$\Gamma \left(\alpha \right)={\int }_0^{\mathrm{\infty }}{x}^{\alpha -1}{e}^{-x}dx$$

When the variable is less than the net flow $x$0, the probability $F$ can be calculated as given:

$$F={\int }_0^{x}g\left(x\right)\mathrm{d}x$$

$$\mathrm{S}\mathrm{R}\mathrm{I}=\mathrm{S}\left({\displaystyle \frac{t-c_0+c_{1}t+c_2{t}^2}{1+d_{1}t+d_2{t}^2+d_3{t}^3}}\right)$$

Here, $c_0=2.515,{ c}_1=0.803,{ c}_2=0.010,{ d}_1=1.434,{ d}_2=0.189,{ d}_3=0.001,t=\sqrt{ln{F}^{-2}}.$

2.3.5. Hydrological Drought Characteristics

For hydrological drought characteristic analysis, run theory is used to calculate the number of drought events, the start time and end time of each drought event, Drought Duration, Drought Severity, and Drought Peak Intensity. In this study, a drought event is defined as a period during which SRI is not higher than −0.5 and the Drought Duration is not shorter than 2 months. Drought Severity is defined as the sum of the absolute SRI values within a single drought event; Drought Peak Intensity is defined as the maximum absolute SRI value within the drought event; and Drought Duration is defined as the time length from the beginning to the end of the drought event. Based on the obtained drought characteristics, Mean Drought Duration and Maximum Drought Duration are derived from the durations of multiple drought events; Mean Drought Severity and Maximum Drought Peak Intensity are derived from the Drought Severity and Drought Peak Intensity of multiple drought events; and Drought Frequency is calculated as the ratio of the number of months with drought events to the total number of months in the study period.

2.3.6. Statistical and Regression Analysis Methods

To systematically evaluate the impacts of climate change and land use change on runoff and hydrological drought, the following statistical and regression analysis methods are employed:

(1)

Under fixed land use conditions, Newey–West robust regression is used to assess the effects of annual climatic variables on different runoff components;

(2)

Using 42 sub-basins as samples, ordinary least squares (OLS) regression is applied to analyze the effects of land use and slope on changes in the three runoff components;

(3)

Scenario-based decomposition and interaction-difference methods are used to quantitatively separate the contributions of climate change, land use change, and their interaction to runoff changes;

(4)

An SRI series is constructed from the baseline-period monthly runoff at Sanliang Station, drought characteristics are extracted for the two periods, and scenario-based attribution is performed accordingly;

(5)

A linear probability model (LPM) is employed to characterize the influence of climatic factors on the probability of hydrological drought occurrence at the monthly scale;

(6)

Under fixed climatic conditions, Spearman rank correlation analysis is used to identify through which runoff components land use change affects drought characteristics;

(7)

A mediation analysis based on nonparametric bootstrap is used to estimate the indirect effects of land use change on hydrological drought characteristics via SURQ, LATQ, and GWQ.

The Mann–Kendall nonparametric test was applied to the baseline and impact periods identified by the Pettitt test to examine the presence of a monotonic trend in annual runoff at the 5% significance level.

The mathematical formulations, parameter settings, and testing procedures for all methods are provided in the Supplementary Material (S1).

3. Results

3.1. Model Calibration and Validation of the SWAT Model for the Middle and Upper Dahei River Basin

The monthly runoff (m³/s) data measured from the hydrological stations of Meidai and Sanliang and the SWAT model simulation results are depicted in Figure 2. For the calibration period, the Nash–Sutcliffe efficiency (NSE) values were 0.97 and 0.79, with corresponding coefficients of determination (R²) of 0.98 and 0.88, p-factors of 0.78 and 0.75, and r-factors of 0.53 and 0.62, respectively. The NS values of the validation period were 0.86 and 0.92, respectively, whereas R² values were 0.86 and 0.92. These outcomes confirm that the SWAT model simulations exhibit high agreement with the observed values; thus, they can be used to simulate monthly runoff variations in the study area.

3.2. Pettitt Breakpoint Test

According to the Pettitt change-point test results (Figure 3a), 1998 was a significant abrupt change year. By applying the Mann–Kendall trend test to the two sub-periods before and after the change point identified by the Pettitt test, we confirm that no significant monotonic trends (|Z| ≤ 1.96) exist within either sub-period. This indicates that the runoff variation is primarily characterized by a single stepwise shift rather than multiple gradual changes, thereby providing a rational basis for adopting a two-stage attribution framework. Consequently, 1983–1998 was defined as the baseline period, with the impact period being 1999–2022. A linear regression analysis was conducted on the cumulative runoff of the outlet hydrological station (Sanliang Hydrological Station) during both the baseline and impact periods based on 1988 (Figure 3b). With p < 0.001, the slope went down from 90.24 to 41.815; this suggests extremely significant differences in annual runoff variation between the two periods and a decreasing cumulative runoff growth rate.

3.3. Attribution Analysis of Runoff Variations

3.3.1. Attribution Analysis of Total Runoff Variation

The outlet hydrological station (Sanliang Hydrological Station) recorded a decrease in measured annual runoff from 8779.6 × 10⁴ m³/year during the baseline period to 3934.8 × 10⁴ m³/year during the impact period, representing a total reduction of 55.26%. To quantify the drivers of this decline, we adopted a scenario-decomposition framework with a single calibrated parameter set. Under the climate-only scenario (S1–S2, fixed land use), climate change led to a 21.34% reduction in runoff (m³/year), corresponding to 38.6% of the total observed decrease. The land use-only and combined climate + land use scenarios indicate that land use change and its interaction with climate increased runoff by 12.97% and 4.95%, respectively, thereby partially offsetting the climate-induced decrease. The remaining unexplained reduction of approximately 51.84% is attributed to the combined impacts of human water withdrawals, reservoir regulation, irrigation and groundwater abstraction that are not explicitly represented as individual engineering units in the SWAT model but inferred as a residual term (Figure 4). In this integrated framework, human activities as a whole contribute 61.4% of the net runoff reduction, while climate change contributes 38.6%.

3.3.2. Impact of Climate Change on Runoff Components

Under fixed land use conditions, the effects of climatic variables on runoff components in the study area during 1983–2022 are investigated. The results indicate that precipitation exhibits significant positive correlations with SURQ and LATQ (standardized regression coefficients of 0.84 and 0.99, p < 0.001), which is the primary driver of interannual variability in these components. Conversely, the influence of precipitation on GWQ is comparatively weaker (standardized regression coefficient of 0.44, p < 0.05). Increased minimum temperatures led to significant reductions in SURQ and LATQ (standardized regression coefficients of −0.35 and −0.19, p < 0.05), while maximum temperatures did not present a considerable effect. Climatic factors strongly explained SURQ and LATQ (R² = 62%, 94%, respectively) but poorly explained GWQ (R² = 19%), which revealed that groundwater recharge processes are predominantly under the influence of non-climatic factors.

3.3.3. Effects of Land Use and Slope Variation on Three Runoff Components

Under fixed climatic conditions, the relative importance of changes in different land use types and slope for the three runoff components is investigated. As the sum of percentages for each land use type is equal to 100%, bare land (BARR), exhibiting the smallest standard deviation, was excluded as the reference category. This was done to avoid perfect multicollinearity. Based on the results, land use change serves as the best explanation for SURQ (R² = 0.98). Cultivated land and grassland expansion led to significant reductions in SURQ (standardized regression coefficients −1.57 and −0.95, respectively; p < 0.001), while the expansion of construction land, forest, and water bodies exhibited considerably negative effects (p < 0.05). On the other hand, GWQ and LATQ’s regression fits were extremely low (R² = 0.20 and 0.14, respectively), confirming that land use change is the primary influencer driving watershed hydrological processes by regulating SURQ, with limited direct effects on GWQ and LATQ.

3.3.4. Effects of Climate-Land Use Change Interaction Effect on Runoff Components

To quantitatively assess the impact of the interaction between climate change and land use change on runoff components. The results, illustrated in Figure 5, show that climate and land use change’s interaction effect accounted for 50% of the total change in SURQ. However, this interaction effect had a negligible effect on LATQ (r% = 6%) and virtually no impact on GWQ.

3.4. Attribution Analysis of Changes in Hydrological Drought Characteristics

Using observed monthly runoff (m³/s) data from the outlet hydrological station (Sanliang Hydrological Station), a hydrological drought sequence was constructed. Additionally, for the baseline period, a standardized SRI scale was established. For both periods, drought characteristics were calculated to obtain percentage changes. Contribution rates were quantified using attributive analysis based on scenario simulations (

Table 3. Attribution Analysis and Quantified Contribution Rates for Hydrological drought Characteristics.

Hydrological Drought Feature	Measured Relative Change (%)	Climate Effect %	Land Use Effect %	Interaction Effect %	Water Withdrawal and Regulation Effect %	Climate Contribution Rate	Human Activity Contribution Rate
Annual Mean Number of Drought Events	100	5.75	−21.05	3.51	111.79	5.75	94.25
Mean Drought Duration	42.71	8.37	−11.14	6.75	38.74	19.60	80.40
Maximum Drought Duration	133.33	0.00	−40.00	−20.00	193.33	0.00	100
Mean Drought Severity	110.32	20.89	−9.33	2.50	96.26	18.94	81.06
Maximum Drought Peak Intensity	19.55	12.49	0.00	−9.61	16.67	63.89	36.11
Drought Frequency	106.94	2.63	−22.37	3.95	122.73	2.46	97.54

). The outcomes indicate that human activities serve as the primary driver for the intensification of most hydrological drought characteristics (contribution rates ranging from 36.11% to 94.25%), while land use change was identified as the predominant influence mitigating hydrological drought. The combined effects of water abstraction, water regulation, irrigation, and groundwater activities were found to reduce runoff while exacerbating hydrological drought. Moreover, climate and land use change’s interaction effect is generally considered weak.

3.4.1. Impact of Climate Change on the Occurrence of Hydrological Drought

To quantify the impact of monthly scale climate change on Hydrological drought occurrence in the study area during 1983–2022. The results indicate the following: the average marginal effect value for precipitation is −0.10 (p < 0.001), while that for maximum and minimum temperature are 0.113 and −0.09, respectively (p < 0.001). These values demonstrate that climate change can cause considerable reductions in the probability of drought occurrence by increasing precipitation and raising minimum temperature, whereas rising maximum temperatures, exhibiting the opposite effect, can substantially increase the probability of hydrological drought.

3.4.2. Mechanism Analysis of Land Use Change and Interaction Effects on Hydrological Drought Characteristics

To identify the impacts of changes in different land use types and their interactions with climate on Hydrological drought characteristics in the study area. The results indicate the following: during the impact period, the Annual Mean Number of Drought Events, Mean Drought Severity, and Drought Frequency exhibited weak to moderate negative correlations with SURQ (with Drought Frequency displaying the strongest correlation, ρ = −0.35); Maximum Drought Peak Intensity presented moderate negative correlations with LATQ in both periods (ρ = −0.34–−0.41). Additionally, during the baseline period, maximum drought duration exhibited a moderate positive correlation with GWQ (p = 0.38–0.43). Climate and land use change’s interaction effect had a weak influence on all hydrological drought characteristics.

An OLS regression analysis was conducted using 42 sub-basins as samples to further compare the relative importance of land use changes on hydrological drought characteristics. The results indicated an extremely low model fit (R² = 5–6%) with no significant single factor, thereby suggesting that land use changes alone cannot explain variations in hydrological drought; their impact is primarily transmitted through three runoff components.

We conducted analyses under fixed climatic conditions to investigate how land use change indirectly influences hydrological drought characteristics by altering these three runoff components. Results indicate that grassland expansion exhibits in total effect of slightly reducing Drought Frequency, with a positive indicate effect through SURQ counterbalanced by a negative direct effect. Forest expansion shows an total effect of increasing Drought Frequency, featuring a positive indirect effect through SURQ and a negative indirect effect through LATQ. However, the mediation effects do not sufficiently counteract the positive direct effect of forest expansion on increasing drought frequency; other pathway effects were close to zero.

3.5. Simulation of Total Runoff in Future Periods and Analysis of Spatial Characteristics of Hydrological Drought

3.5.1. Multi-Scenario Total Runoff Simulation for Future Periods

As illustrated in Figure 6, the Sanliang Station’s average annual runoff (m³) from 2025 to 2064 under nine scenarios. The runoff decreased by 2.08 to 60.16% compared to the historical period (1983–2022). However, under identical land use scenarios, the annual average runoff under the SSP-1-2.6 climate scenario exceeds that of the SSP5-8.5 climate scenario, whereas the SSP2-4.5 climate scenario yields the lowest annual runoff. Under the same climate scenarios, the runoff is highest under the ED land use scenario but lowest under the EP scenario.

3.5.2. Spatial Characteristics of Hydrological Drought Analysis

The drought characteristics of the Dahei River Basin’s middle and upper reaches during the historical period are shown in Figure 7. Under the local model, the drought events’ annual mean number was 0.7 times/year, the mean drought duration was 3.82 months, the maximum drought duration was 13 months, the mean drought severity was 4.92, the maximum drought peak intensity was 3.19, and the drought frequency was 30.00%. In the eastern part of the basin, the Annual Mean Number of Drought Events (Figure 7a) was higher, with Sub-basin 16 experiencing the highest frequency at 1.20 times. Sub-basin 3 had the longest duration at 5.00 months for Mean Drought Duration (Figure 7b). In Maximum Drought Duration (Figure 7c), spatial variation was significant, ranging from 22 months in Sub-basin 3 to 6 months in Sub-basin 41. For Mean Drought Severity (Figure 7d), the southwest exhibited a higher intensity, while the east displayed a lower intensity, presenting a maximum-to-minimum difference of 1.73 for Maximum Drought Peak Intensity. In Maximum Drought Peak Intensity (Figure 7e), spatial variation was substantial, with the maximum value being twice the minimum. In terms of Drought Frequency (Figure 7f), Sub-basin 3 recorded the highest frequency at 42.92%, whereas Sub-basin 41 had the lowest frequency at 26.67%, with the maximum exceeding the minimum by 16.25%.

This study uses the local model to establish a unified SRI benchmark to calculate changes in hydrological drought characteristics across nine scenarios (S5–S13) relative to the local model, as depicted in Figure S7. The nine future scenarios show increases ranging from 83–114% and 43–117%, respectively, for annual mean number of drought events and Drought Frequency. In terms of spatial aspects, most areas in the eastern and northeastern parts of the basin exhibit relative increases, while the northwestern to central regions show weaker changes or declines. Relative changes compared to the historical period range from −14% to 56% and −11% to 58%, respectively, for Mean Drought Duration and Mean Drought Severity. It is important to note that reductions occur only under three economic development land use scenarios. Spatially, increases are generally observed in the east, while decreases usually occur in the west. Maximum Drought Duration decreases by 15% to 46% across scenarios, with significant reductions generally observed from the central to western regions. Across all scenarios, peak drought intensity remains largely unchanged. Collectively, the nine future scenarios exhibit the characteristic of “more frequent droughts, longer Mean Drought Duration, higher intensity, but shorter Maximum Drought Duration.” The three land use scenarios are consistently ranked under identical climatic conditions. For Mean Drought Duration, Maximum Drought Duration, Mean Drought Severity, and Drought Frequency, the EP scenario ranked highest, the ED scenario lowest, and the BAU scenario intermediate. However, for the Annual Mean Number of Drought Events, the EP scenario ranked the lowest.

4. Discussion

4.1. Model Simulation Performance

This study presents the foundational data support for the attribution analysis of changes in runoff and hydrological drought characteristics within the Dahei River Basin’s middle and upper reaches, utilizing simulations from the SWAT hydrological model. Previous studies that employed this model to simulate hydrological processes in the basin achieved accuracy metrics (R² and NS) nearing 0.8 [39,40]. The SWAT model reproduced the monthly runoff at the Meidai and Sanliang hydrological stations with Nash–Sutcliffe efficiency (NS) values of 0.97 and 0.79 and coefficients of determination (R²) of 0.98 and 0.88, respectively, during the validation period. This high simulation accuracy is primarily attributed to the use of station-based meteorological data, which represents actual precipitation more reliably than remote sensing products, and to the 40-year continuous simulation period, which enhances the stability and robustness of the hydrological model.

4.2. Attribution Analysis of Changes in Runoff Volume and Hydrological Drought Characteristics

Regarding climate (Figure S1), the monthly precipitation amount experienced a decline during the impact period (from 31.41 mm to 30.56 mm) in comparison to the baseline period. Specifically, precipitation during the critical runoff period (July–August) presented a decrease of 10% to 23%, diminishing the ability to generate flood season runoff. Concurrently, during the critical runoff period temperatures rose, with monthly maximum and minimum temperatures increasing by 0.7 °C and 1.07 °C, respectively. This led to an escalation in potential evapotranspiration (PET) [41]. Concerning land use changes (Figures S1–S3), urbanization resulted in a 47.9% expansion of construction land area, consequently elevating runoff. The mechanisms employed include reduced infiltration due to the proliferation of impervious surfaces and accelerated runoff processes [42]; simultaneously, ecological restoration projects (0.65% increase in forested land, 0.1% decrease in unutilized land) were found to mitigate runoff through vegetation transpiration and canopy interception [43]. The attribution analysis conducted in this study indicates that climate change reduced runoff by 21.34%, while land use change has contributed to a 12.97% increase in runoff. This result is consistent with findings reported by Hossainzadeh et al. [44,45,46]. Conversely, Yongxin Ni et al. [47] found that both climate and land use change reduced runoff in the middle Yellow River basin from 1961 to 2018. Such discrepancies may stem from accelerated urban expansion in this basin during the study period. Yet another reason could be the ecological restoration efforts focused on upstream areas, which have weaker hydrological impacts. Both of these allow the increases in runoff from construction land expansion to surpass the reductions from forest restoration. Under the aforementioned nine future scenarios, although the annual precipitation amount is projected to rise by 28.7 mm to 51.7 mm, the increase in annual mean maximum temperature (1.9 °C–2.5 °C) and minimum temperature (2.0 °C–2.8 °C) will substantially enhance evapotranspiration in the study area (Figures S4 and S5). Under three scenarios from 2000 to 2045, land use change expands construction land and forest land from 147.19% to 226.49% and 12% to 15%, respectively (Figure S6). Despite the trend of ongoing urbanization, runoff exhibited a reduction by 2.08 to 60.16% across all scenarios, indicating that climate change exerts a more pronounced impact on watershed runoff than land use change. This finding aligns with those of Iqbal et al. Furthermore, extreme precipitation events will become more frequent and the occurrence of extreme heavy rainfall may increase basin runoff under persistent climate warming scenarios. Simultaneously, the impact of land use change on runoff may be mitigated by increased emission concentrations [48].

The present study suggests that climate change has the potential to substantially lower drought probability by increasing precipitation and elevating minimum temperatures. This phenomenon can be due to the direct replenishment of water sources by increased precipitation and the acceleration of snowmelt due to rising minimum temperatures. Conversely, rising maximum temperatures may result in heightened evaporation and increased water loss, thereby intensifying drought probability. Land use change plays a primary role in influencing hydrological drought by regulating SURQ. Urbanization accelerates the runoff from rainfall into rivers, thereby enabling faster recovery from low-flow conditions after events of rainfall. This is in alignment with our findings: increased SURQ exhibits an inverse relationship with Drought Frequency, Annual Mean Number of Drought Events, and drought intensity. During precipitation-free periods, LATQ and GWQ enable soil and groundwater to sustain tributary flows to rivers, preventing prolonged low river flows and extreme drawdowns [49]. This is consistent with our findings: LATQ inversely correlates with Maximum Drought Peak Intensity, while GWQ inversely correlates with Maximum Drought Duration. Land use change alters the proportional relationships among SURQ, LATQ, and GWQ by impacting infiltration and runoff processes, thus influencing alterations in drought characteristics. This study demonstrates that climate and land use’s interaction effect affects the three runoff components in different ways across sub-basins rather than through simple linear addition. This implies that land use changes can significantly impact the process of converting precipitation into runoff.

4.3. Recommendations for Future Water Resources Management and Drought Mitigation Strategies

Changes in land surface characteristics alter the influence of climatic factors on hydrological processes. Therefore, water resources management strategies should adopt a comprehensive perspective that prioritizes the optimization of the eastern and northeastern parts of the study area. (1) Land and soil management to suppress SURQ and enhance slow recharge. In areas with steep slopes or compacted soils, prioritize measures that lower CN2 and increase SOL_K and SOL_AWC, such as residue retention/conservation tillage, organic amendments, contouring and terracing, grassed waterways and riparian buffers, and slope-scale forest–grass configurations. The mechanism is to attenuate storm-runoff peaks and rapid surface responses, increase the contributions of LATQ/GWQ, and thereby provide more stable supply in low-flow periods and mitigate drought intensity and frequency. (2) Groundwater management coordinated with seasonal recharge. Implement zonal pumping limits and warning thresholds in the dry season; in areas with suitable hydrogeological conditions, conduct managed aquifer recharge (MAR) pilots during the wet season, and establish a joint groundwater-level–baseflow monitoring and assessment framework to avoid baseflow decline and increased dry-season supply vulnerability caused by long-term extraction. (3) Water-saving and drought-resilient upgrading of irrigation systems. In view of the rising evaporative demand, advance quota-based/precision irrigation, promote drought-tolerant cultivars and a rational cropping structure, and pair these with on-farm water conservation and soil improvement measures to lower water-demand thresholds and increase the effective utilization coefficient of irrigation water; prioritize implementation in subbasins where drought frequency has increased markedly. The findings are directly relevant for the China Yellow River Water Conservancy Commission, China Hohhot Water Conservancy Bureau, China Inner Mongolia Autonomous Region Planning Institute, supporting integrated regulation of surface and groundwater, ecological restoration zoning, and drought contingency planning under combined climate–human pressures.

4.4. Uncertainty Analysis

This simulation and prediction methods in this study involve certain uncertainties. When using the SWAT model for runoff simulation, spatial data uncertainty is a critical issue that cannot be overlooked [50]. Although the long-term hydrological processes cannot be entirely described by the spatial information from this period, addressing the lack of precise spatial information can help facilitate the comparison of variations in hydrological processes over time [51]. Furthermore, variations in input data resolution and watershed thresholds may impact runoff process simulations. Similar findings have been brought up by various researchers [52,53]. This study employed SUFI-2 for model calibration and uncertainty analysis, resulting in high R² and NS values for predicted parameters. Moreover, appropriate parameter ranges and optimal values minimized uncertainty in this component.

It should be emphasized that the inferred “water abstractions and regulation” is obtained as a residual term under the adopted model structure, parameter set and scenario design, rather than from fully explicit simulation of each water use sector or project. This residual inevitably includes the effects of model structural and parameter uncertainty, input data errors, and unmodelled processes; hence, the reported percentage (51.84%) should be interpreted as an integrated and plausible magnitude of overall human influence on runoff, not as a precise quantification of individual regulation or abstraction measures.

The comparison between the baseflow indices derived from digital filtering and those estimated from SWAT-simulated runoff components provides a consistency check for our attribution results (detailed methodology and equations are provided in the Supplementary Material). Applying the Eckhardt filter to observed discharge yields a long-term baseflow index of approximately 0.27, indicating that runoff in the Dahei River is predominantly governed by fast responses, with a relatively small but non-negligible contribution from slow flow. When the lateral flow (LATQ) and groundwater flow (GWQ) simulated by SWAT are jointly treated as slow runoff, the corresponding model-based baseflow index is about 0.34. This discrepancy is hydrologically reasonable. First, the filter-based approach performs a two-component separation (quickflow vs. baseflow) at the monthly scale, whereas SWAT explicitly distinguishes surface runoff, lateral flow, and groundwater flow; LATQ in SWAT inherently aggregates both relatively rapid and slower subsurface processes. Second, digital filters rely solely on the statistical characteristics of hydrograph recession, while SWAT routes water through soil, topographic, and aquifer storages using physically based parameters. Therefore, exact numerical agreement between the two approaches is neither expected nor required. More importantly, both methods consistently indicate that (1) event-driven surface and near-surface runoff dominates total streamflow, and (2) slow runoff plays a crucial role in sustaining low flows and buffering hydrological droughts. Accordingly, in this study the SWAT-derived components (SURQ, LATQ, GWQ) are used as diagnostic indicators of the relative importance of fast and slow runoff generation, rather than as a unique “true” partitioning of flow paths. The baseflow separation results confirm that the simulated runoff composition is hydrologically plausible and thereby reinforce the robustness of our process-based attribution.

GCMs are indispensable in the study of climate change mechanisms and prediction of future climate shifts [54]. Nevertheless, considerable uncertainty persists in future climate models and emission scenarios, which can potentially introduce errors in regional precipitation and air temperature projections [55]. No single accurate prediction model exists currently, and the use of multi-model ensemble assessments is a subject of controversial debate [56]. It should be noted that the scenario simulations of future runoff and hydrological drought are affected by uncertainties in climate models. In this study, SWAT is driven by a multi-GCM ensemble, and the scenario results represent change signals in the sense of multi-model means. Differences among GCMs in the magnitude of precipitation changes and the degree of warming, especially under the high-emission SSP5–8.5 scenario, introduce a certain range of uncertainty into the projected values of runoff and drought indices. Therefore, the results presented in this paper should be interpreted as indications of robust directional trends and relative changes, rather than precise predictions of future absolute runoff values, and more detailed quantification of model-specific uncertainties remains a subject for future research. The Delta downscaling method this study employed for downscaling CMIP6 data still requires further refinement compared to dynamic downscaling methods in simulating extreme events and spatial variability.

5. Conclusions

(1)

The SWAT model performed well in simulating monthly runoff at the Meidai and Sanliang hydrological stations in the study area. Parameters such as ESCO, SMFMN, CN2, CH_N2, and SOL_K were discovered to be relatively sensitive parameters affecting runoff simulation.

(2)

Total runoff in the study area decreased by 55.26% during the impact period (1999–2022) as compared to the baseline period (1983–1998). In this decline, climate change accounted for a contribution rate of 38.6%, while human activities accounted for 61.4%. Particularly, climate change reduced runoff by 21.34%, while land use change increased runoff by 12.97%. The climate and land use’s interaction effect increased runoff by 4.95%. Climate primarily altered SURQ and LATQ through precipitation changes, while land use mainly influenced total runoff by modifying SURQ. Both factors had minor effects on subsurface runoff. Half the variation in SURQ is due to the interaction between climate and land use accounts.

(3)

Human activities significantly influence the intensification of hydrological drought (contribution rate: 36.11–94.25%). While climate change alone tends to worsen drought characteristics, land use change has the opposite effect. The impact of the interaction between climate and land use change has on all hydrological drought characteristics is comparatively weak. Climate change can notably reduce the probability of drought by increasing precipitation and raising minimum temperatures; in contrast, increasing maximum temperatures leads to an increase in drought probability. Land use impacts drought primarily through three runoff components.

(4)

It is predicted that over the next 40 years, total runoff will decrease by 2.08% to 60.16%. Under identical climatic conditions, ED > BAU > EP, and under identical land use conditions, SSP1-2.6 > SSP5-8.5 > SSP2-4.5. Hydrological drought characteristics exhibit the pattern of “more frequent, longer in average duration, and more intense hydrological droughts; however, Maximum Drought Duration is anticipated to shorten.” In the eastern and northeastern parts of the study area, hydrological droughts show strengthening trends; meanwhile, in the central and western regions, these droughts exhibit weaker or declining trends.