Future Streamflow and Hydrological Drought Under CMIP6 Climate Projections

Abstract

Droughts caused by runoff are an important environmental issue in the context of global climate change, with profound impacts on ecosystems, agriculture and water resource management. To assess the impact of future climate change on the hydrological response of watersheds, this study combines the SWAT (Soil and Water Assessment Tool) and MODFLOW (MODular groundwater FLOW model) models to predict future changes in runoff and hydrological drought in watersheds using data from two scenarios under 15 CMIP6 climate models. The results show that: (1) The R² and NSE values of monthly runoff at the Caizuzi station in the Naoli River basin are greater than 0.60 in different periods; (2) the ensemble of climate models after screening can effectively improve the accuracy of runoff simulation and reduce the prediction uncertainty of a single climate model; (3) under different scenarios, the temperature generally increases, the precipitation increases and evapotranspiration increased under the SSP2-4.5 scenario and decreased under the SSP5-8.5 scenario; (4) runoff showed an increasing trend under the SSP2-4.5 scenario and the opposite trend under the SSP5-8.5 scenario; (5) the frequency of winter runoff droughts decreased in the future period, while the frequency of spring and summer droughts increased, with the change trend being more pronounced under the SSP5-8.5 scenario; (6) compared with the baseline period (1965–2014), under the SSP2-4.5 and SSP5-8.5 scenarios, the average annual temperature in the watershed increased by 1.89 °C and 3.22 °C, respectively, and the annual precipitation increased by 32% and 36.19%, respectively, but the summer and autumn runoff decreased; and (7) The SRI-3 model analysis indicates that hydrological droughts will significantly intensify under both future emission scenarios. Under the SSP5-8.5 scenario, droughts will worsen earlier and the abrupt change will occur earlier, while under the SSP2-4.5 scenario, although the abrupt change will occur later, the drought intensity will be higher. The critical drought transition periods are 2030–2047 (SSP5-8.5) and 2045–2055 (SSP2-4.5). This study provides important scientific basis for adaptive water resources management and drought mitigation strategies in cold-region watersheds under future climate scenarios.

1. Introduction

With the general rise in temperature, the frequency of runoff drought events continues to increase, which has a significant impact on agriculture, the ecological environment and the socio-economy [1,2]. The occurrence of runoff drought events is affected by a variety of driving factors, which seriously threaten the water security, food security and ecological security of the region [3]. Therefore, it is particularly important to conduct in-depth research on the causes and development trends of hydrological drought and establish scientific prediction and response mechanisms for future drought scenarios.

Drought is a multi-dimensional hazard that can be classified into four categories: hydrological, meteorological, agricultural and socio-economic [4]. Hydrological drought assessment is the most important type in water resource management, and the simplest way to detect drought conditions is through drought indices [5]. Drought indices help decision-makers by identifying the characteristics of droughts (e.g., severity, duration and frequency of occurrence) [6]. In recent years, there has been a growing interest in various drought assessment methods and techniques. A variety of indices based on runoff, streamflow and groundwater variables have been developed, such as the Standardized Runoff Index (SRI), the Standardized Soil Moisture Index (SSMI), the Streamflow Drought Index (SDI) and the Baseflow Index (BFI) [7]. The SRI can effectively characterize hydrological drought processes at different timescales by normalizing runoff data [8]. In contrast, the Standardized Soil Moisture Index (SSMI) can reflect soil moisture changes, but data acquisition is difficult and spatial continuity is poor, and there is a lag in the drought response [9]; the Stream Drought Index (SDI) is limited by a fixed time scale, and its calculation depends on the integrity of the time series, making it less reliable than SRI [10]; the Baseflow Index (BFI) focuses on the contribution of groundwater to runoff and is difficult to reflect the dynamics of short-term hydrological drought [11]. In contrast, SRI data is readily available, highly applicable, can be reliably adapted to different time scales and is more comprehensive. It is an important tool for monitoring and predicting hydrological drought [12].

CMIP6 (the Coupled Model Intercomparison Program, CMIP) is the latest release of the international coupled model comparison program [13]. CMIP6 has higher resolution in atmospheric and ocean models and combines different combinations of Shared Socio-economic Pathways (SSPs) and Representative Concentration Pathways (RCPs), which can better reflect the changing characteristics of various socio-economic factors related to human activities than CMIP5 [14]. It was found in the CMIP5 scenario that global drought problems will continue to increase, so it is necessary to study the drought problem in the CMIP6 scenario [15]. There have been many studies that combine CMIP6 climate models with different drought indices [16,17,18]. For example, the characteristics of global droughts were analyzed by screening the CMIP6 multi-model dataset and combining it with different drought indices [19]; and socio-economic droughts were analyzed based on the CMIP6 framework using SPI (Standardized Precipitation Index) and SRI combined with SWAT [20].

At present, many studies use global hydrological models or watershed hydrological models to assess the impact of climate change on runoff. For example, Chen et al. (2022) [21] quantified the uncertainty in streamflow predictions under climate change using the SWAT model and CMIP6 climate projections. Loritz et al. (2024) [22] developed the CAMELS-DE dataset, providing comprehensive hydro-meteorological time series and catchment attributes for 1555 catchments in Germany. Wu et al. (2021) [23] utilized integrated hydrological models to evaluate the impacts of climate change on discharges and extreme flood events in the Upper Yangtze River Basin. Feng et al. (2024) [24] introduced ΔHBV-Globe1.0-HydroDL, a global-scale hydrological model integrating deep learning and physics-informed approaches.

However, most studies focus on surface water processes, while the response mechanism of groundwater is often ignored [25]. In fact, groundwater–surface water interactions are particularly important under drought conditions, and traditional hydrological models have difficulty fully characterizing this coupling [26]. The application of the SWAT-MODFLOW coupled model in hydrological drought research remains relatively limited, particularly in terms of comprehensive assessments of basin-scale drought under CMIP6 scenarios. Therefore, it is important to integrate coupled surface water and groundwater models, such as SWAT-MODFLOW, to assess future hydrological droughts.

This study adopts a semi-coupled approach in which SWAT simulates surface hydrology and provides spatially distributed recharge to MODFLOW, which models groundwater flow and returns baseflow to channels. The integrated framework enhances the representation of surface–subsurface interactions, thereby improving the reliability of drought projections under future climate scenarios. This study aims to use the SWAT-MODFLOW coupled model, based on the CMIP6 SSP2-4.5 and SSP5-8.5 emission scenarios, to assess the impact of future climate change on runoff and hydrological drought in the Naoli River basin, reveal the role of surface water–groundwater coupling mechanisms in drought evolution and provide scientific basis for regional water resource management and drought prevention and control.

The Raohe River basin is located in the Sanjiang Plain of northeastern China’s Heilongjiang Province, covering a total area of approximately 24,863 square kilometers. It is an important grain production base in China, with its agriculture, irrigation and ecological environment being highly dependent on water resources [27]. The basin is mainly composed of low-lying alluvial plains and marshes, with significant interaction between surface water and groundwater. It is highly susceptible to climate change, resulting in frequent runoff drought events [22]. With increasing agricultural demand and accelerating urbanization and industrialization, water resources are under increasing pressure, and worsening droughts are causing serious problems such as declining agricultural production and ecological degradation [28]. Therefore, this study employed the SWAT-MODFLOW coupled model to analyze data under the two commonly used emission scenarios—SSP2-4.5 and SSP5-8.5—from the latest CMIP6 climate scenarios. The study assessed the impacts of future climate change on runoff and hydrological drought in the Naoli River basin, providing scientific basis for the prevention and control of runoff drought disasters in the basin.

2. Materials and Methods

2.1. Overview of the Research Area

As shown in Figure 1, the Naoli River basin is located in the northeast of Heilongjiang Province and has a temperate continental monsoon climate, with low annual precipitation and uneven distribution. Insufficient precipitation in spring and autumn leads to frequent runoff droughts, especially in late spring and early summer, when runoff decreases, seriously affecting agricultural irrigation [29]. As an important agricultural production area, the watershed relies heavily on water resources for crops such as rice and corn. Runoff droughts not only affect crop growth but may also lead to production reductions, threatening the stability of agricultural production in the watershed [30].

The annual average precipitation is approximately 550–600 mm, concentrated in the summer (June–August, accounting for 70% of annual precipitation). Runoff is mainly surface runoff, with summer runoff accounting for approximately 65% of the annual total. In winter, runoff is extremely low (less than 10%) due to the freezing period and sparse precipitation [31]. Soil resources are mainly black soil, marsh soil and paddy soil. Black soil is highly fertile but prone to erosion, while marsh soil is widely distributed but has poor drainage, limiting agricultural development potential [32]. With increasing agricultural demand and accelerating urbanization and industrialization, pressure on water and soil resources continues to grow, and worsening droughts are leading to declining agricultural production and ecological degradation [33].

At the same time, the watershed’s weak soil water retention capacity further exacerbates the impact of drought [34].

2.2. Data Sources

Meteorological station site data selection: The data for the Naoli River basin and its surrounding four meteorological stations include daily meteorological data such as temperature, rainfall and sunshine from 1970 to 2014. A digital elevation model (DEM) with a spatial resolution of 30 m was obtained through a geospatial data cloud. Then, the weighting of weather stations in the catchment area was obtained using the Taisong polygon method.These weights were multiplied and then summed to obtain comprehensive daily precipitation and temperature data at the catchment area scale, which was used to assess future climate changes relative to the observation period. Specific data sources are shown in

Table 1. Sources and scales of basic data for the model.

Data Type	Scale	Source
Meteorological data	1970–2014	China Meteorological Data Network (http://data.cma.cn, accessed on 26 April 2025)
DEM	30 km	Geospatial data cloud (https://www.gscloud.cn/, accessed on 26 April 2025)
Soil type	30 km	HWSD Soil Database (https://www.fao.org/soils-portal/data-hub/soil-maps-and-databases, accessed on 26 April 2025)
Land use	30 km	Chinese Academy of Sciences Center for Resources and Environmental Science and Data (https://www.resdc.cn/, accessed on 26 April 2025)
Hydrological data	2005–2012	Earth Resources Data Cloud Platform (www.gis5g.com, accessed on 26 April 2025)

.

CMIP6 data sources: From the National Aeronautics and Space Administration (NASA Climate Simulation Center) (https://www.nccs.nasa.gov, accessed on 26 April 2025), 15 meteorological models were selected, and their data were processed using bias correction and spatial resolution methods to obtain daily time scales and 0.25° × 0.25° grids. Two shared socio-economic pathways were selected, namely SSP2-4.5 and SSP5-8.5, covering the baseline simulation period from 1970 to 2014 and the future simulation period from 2016 to 2100, as shown in

Table 2. Overview of the 14 global climate patterns for CMIP 6.

Pattern Name	Country	Spatial Resolution	Pattern Name	Country	Spatial Resolution
ACCESS-CM2 ACCESS-ESM1-5	Australia	0.25° × 0.25°	EC-Earth3 IPSL-CM6A-LR	Europe	0.25° × 0.25°
NorESM2-LM NorESM2-MM	Norway		MIROC6 MIROC-ES2L MRI-ESM2-0	Japan
MPI-ESM1-2-HR MPI-ESM1-2-LR	Germany
			GFDL-CM4 GFDL-ESM4	United States
INM-CM4-8	Russia		CanESM5	Canada

.

3. Research Methods

3.1. SWAT Model

SWAT is a physically based distributed hydrological model suitable for runoff simulation in medium- to large-scale watersheds, capable of long-term continuous simulation of complex hydrological processes in complex watersheds. It supports efficient coupling with groundwater models such as MODFLOW to achieve integrated surface–groundwater simulation and is widely applied in global hydrological research. By inputting digital elevation models (DEMs) for watershed delineation and terrain parameter extraction, land use data (LUCC) to reflect surface cover characteristics, soil data (Soil) to provide soil physical and chemical properties, and long-term meteorological data such as precipitation and temperature as driving forces, the model can operate effectively [35]. In sensitivity analysis, parameters significantly affecting runoff simulation (such as curve number (CN) and effective soil moisture content) were identified using the SWAT-CUP tool. During the calibration period, a multi-objective optimization strategy was adopted to comprehensively fit total runoff, peak flow, and low-flow conditions. In the validation period, the Nash efficiency coefficient (NSE) and coefficient of determination (R²) were used to evaluate model accuracy, ensuring the reliability and scientific validity of the simulation results [22].

The water balance equation is:

$${\mathrm{SW}}_t={\mathrm{SW}}_0+{\displaystyle \sum _{i=1}^t\left(R_{\mathrm{day}}-Q_{\mathrm{surf}}-E_{\mathrm{a}}-W_{\mathrm{seep}}-Q_{\mathrm{gw}}\right)}$$

(1)

$S{W}_t:$ The soil moisture content at time t (mm) indicates the storage state of soil moisture. $S{W}_0:$ The initial soil moisture content (mm) simulates the soil moisture content at the beginning of the time period. $R_{\mathrm{day}}:$ The daily precipitation (mm) refers to the daily precipitation input into the watershed system. $Q_{\mathrm{surf}}:$ The surface runoff (mm) refers to the amount of water that is converted into surface runoff by precipitation and discharged from the watershed. $E_a:$ The evapotranspiration (mm) includes the total amount of water lost by plant transpiration and soil evaporation. $W_{\mathrm{seep}}:$ The deep infiltration (mm) refers to the amount of water that infiltrates deeply into the soil. $Q_{\mathrm{gw}}:$ Groundwater discharge (mm) refers to the amount of groundwater lost to rivers or other discharge areas.

3.2. SWAT-MODFLOW Model

The QGIS-based SWAT-MODFLOW coupled model uses the spatial analysis capabilities and data management functions of the QGIS platform to achieve an efficient coupling of surface water and groundwater systems [36]. QGIS is responsible for integrating and managing the spatial data required by SWAT and MODFLOW in this model, such as DEM, land use and soil properties, through a spatial mapping mechanism to allocate the deep infiltration calculated by the SWAT model to the finite difference grid cells of MODFLOW as groundwater recharge [37]. At the same time, the groundwater flow and water level changes simulated by MODFLOW are fed back to SWAT to adjust the base flow and hydrological processes, thereby optimizing the simulation accuracy of surface water–groundwater interactions [38]. This coupled model provides a more accurate tool for water resources management and analysis of hydrological processes in watersheds. The MODFLOW formula is as follows:

$${\displaystyle \frac{\partial }{\partial x}}\left(K_x{\displaystyle \frac{\partial h}{\partial x}}\right)+{\displaystyle \frac{\partial }{\partial y}}\left(K_y{\displaystyle \frac{\partial h}{\partial y}}\right)+{\displaystyle \frac{\partial }{\partial z}}\left(K_z{\displaystyle \frac{\partial h}{\partial z}}\right)-q=S_s{\displaystyle \frac{\partial h}{\partial t}}$$

(2)

$h$: Hydraulic head (units: m), representing the groundwater level or potentiometric surface. $K_x,K_y,K_z$: Hydraulic conductivity in the x, y, and z directions (m/s), describing the ease with which water can move through the aquifer in each direction. $q$: Source/Sink term (1/s), representing the rate of water added to or removed from the aquifer per unit volume (e.g., recharge, pumping, or evapotranspiration). $S_s$: Specific storage (1/m), defined as the volume of water released from or stored in a unit volume of aquifer per unit change in hydraulic head. $t$: Time (s), representing the temporal dimension of the groundwater flow process.

3.3. Calculation of Evapotranspiration

The Penman–Monteith formula is used to calculate reference evapotranspiration (ET₀), which is a widely used method for calculating evapotranspiration in hydrology, agriculture and meteorology [39]. The formula was proposed by Penman in 1948 and improved by Monteith in 1965 to make it more suitable for calculating evapotranspiration under plant canopy conditions [40]. The formula is as follows:

$$E{T}_0={\displaystyle \frac{0.408\Delta \left(R_n-G\right)+\gamma \frac{900}{T+273}{u}_2\left(e_s-e_a\right)}{\Delta +\gamma \left(1+0.34u_2\right)}}$$

(3)

$E{T}_0$: Reference crop evapotranspiration (mm/day); $R_n$: net radiation (MJ/m²/day); $G$: soil heat flux (MJ/m²/day); $T$: mean temperature (°C); $u_2$: wind speed at 2 m height (m/s); $e_s$: saturated water vapor pressure (kPa); $e_a$: actual water vapor pressure (kPa); $e_s-e_a$: vapor pressure difference (kPa); $\Delta$: slope of saturated vapor pressure at air temperature (kPa/°C); $\gamma$: wet bulb constant (kPa/°C).

3.4. Model Calibration

The SUFI-2 (Sequential Uncertainty Fitting ver. 2) method is an uncertainty analysis and parameter optimization technique based on Bayesian inference, which is widely used in the runoff calibration of SWAT models (USDA-ARS, Temple, TX, USA) [41]. The main parameters considered cover the key variables of the hydrological process in the watershed to ensure the rationality and accuracy of runoff simulation [42]. SUFI-2 has the advantages of high computational efficiency, strong parameter optimization ability and applicability to large-scale watershed simulation and can effectively quantify model uncertainty [43]. Therefore, in the runoff calibration study of the SWAT and SWAT-MODFLOW coupled model, SUFI-2 is still one of the most widely used optimization methods, which can improve the applicability and simulation accuracy of the model while ensuring computational efficiency [44].

The calibration of the MODFLOW model usually uses the basic hydrogeological parameters as initial values, and the model performance is optimized by gradually adjusting the parameter values, provided that the qualitative understanding of the hydrogeological conditions is met [45]. During the calibration process, the adjustment of hydrogeological parameters needs to take into account a reasonable understanding of the hydrogeological characteristics of the watershed, while ensuring that the simulation results can best match the runoff observation data, so as to improve the reliability and applicability of the model [46].

In this paper, the coefficient of determination (R²) and the Nash efficiency coefficient (NSE) are used as evaluation indicators of the degree of fit between the simulated and measured flows [47]. The calculation formulas are as follows:

$$R^2={\displaystyle \frac{{\left({\displaystyle \sum _{i=1}^n\left(Q_{m,i}-Q_{m,\mathrm{avg}}\right)}\left(Q_{p,i}-Q_{p,\mathrm{avg}}\right)\right)}^2}{{\displaystyle \sum _{i=1}^n{\left(Q_{\mathrm{m},i}-Q_{m,\mathrm{avg}}\right)}^2}{\displaystyle \sum _{i=1}^n{\left(Q_{p,i}-Q_{p,\mathrm{avg}}\right)}^2}}},$$

(4)

$$\mathrm{NSE}=1-{\displaystyle \frac{{\displaystyle \sum _{i=1}^n{\left(Q_{m,i}-Q_{p,i}\right)}^2}}{{\displaystyle \sum _{i=1}^n{\left(Q_{m,i}-Q_{m,a\mathrm{av}}\right)}^2}}}$$

(5)

where: $Q_{m,i}$ is the measured flow rate, m³/s; $Q_{p,i}$ is the simulated flow rate, m³/s; $Q_{m,\mathrm{avg}}$ is the average measured flow rate over multiple years, m³/s; $Q_{p,\mathrm{avg}}$ is the average simulated flow rate over multiple years, m³/s; and $n$ is the length of the measured time series. It is generally considered that the model fit is satisfactory when $R^2>0.6$ and $NSE>0.5$.

3.5. SRI Index and the Theory of Travel

The Streamflow Response Index (SRI) is an index that quantifies the response of streamflow to rainfall, and it is often used to assess the sensitivity of a watershed to precipitation changes [48]. The SRI analyzes the relationship between streams and precipitation and uses stream runoff formulas to quantitatively describe the hydrological response characteristics of watersheds under different climate and precipitation conditions [7]. Its calculation is based on the comparison of the precipitation flow time series, which reflects the runoff fluctuations of the basin under specific meteorological conditions [49]. The larger the SRI, the more sensitive the basin is to precipitation changes; conversely, the smaller the SRI, the more stable the hydrological response of the basin [50]. This indicator is widely used in the evaluation of hydrological models, water resources management and the analysis of the impact of climate change. It can provide a quantitative analysis of the hydrological processes of the basin and help to optimize water resources scheduling and basin management [51].

The travel theory has been widely used in the identification and characterization of drought events. The identification process mainly includes the following steps: (1) Preliminary identification of drought months: by setting the threshold of the drought index to, for example, $R_0$ = 0, $R_1$ = −0.3 and $R_2$ = −0.5, when the SRI is less than $R_1$, it is initially determined to be a dry month; (2) Elimination of small drought events: some drought events only last for 1 month, and the SRI is greater than $R_2$, so they are considered to be small drought events and are not included in the statistics; and (3) Combination of drought events: When two drought events are separated by only one month and both have an SRI of less than $R_0$, the two events are combined into a single drought event [52,53]. In order to accurately express regional hydrological drought, according to the Chinese standard “Meteorological Drought Classification: GB/T20481-2017” [54], drought is classified into five levels based on the drought index, as shown in

Table 3. SRI drought grade classification table.

Grade	Type	SRI
1	no drought	SRI > −0.5
2	light drought	−0.5 ≥ SRI > −1.0
3	moderate drought	−1.0 ≥ SRI > −1.5
4	severe drought	−1.5 ≥ SRI > −2.0
5	exceptionally drought	SRI ≤ −2.0

.

4. Model Building

This paper selects land use data from 2005 as input data for runoff simulation under the current conditions of the model. QGIS is used to crop the land use type data of the study area, and the cropped data is reclassified from secondary to primary to provide suitable input data for the model [55].

Because the soil classification of the FAO-90 soil classification system adopts the classification standard of the United States Department of Agriculture (USDA), there is no need to convert the soil particle size. The soil type map of the study area can be cropped using the mask tool of GIS, and the soil type can be reclassified [55]. Finally, a corresponding soil database was created according to the format requirements of the SWAT soil property database. In the calculation of soil properties, the effective soil water holding capacity, wet density and saturated hydraulic conductivity were calculated using the SPAW software (Version 6.02.75) [56]; the soil erosion factor and minimum infiltration rate were estimated using empirical formulas [57].

4.1. Sub-Basin Division and Model Coupling

After debugging, the catchment area threshold was set to 7.1 square kilometers, the Caizuzi hydrological station was added as a control station, and the Dali River basin was finally divided into 135 sub-basins, as shown in Figure 2a. The thresholds for land use, soil type and slope were all set to 10%, and the Dali River basin was finally divided into 596 HRUs, as shown in Figure 2b. The underground aquifer in the study area was set to range from 40 m to 60 m. The MODFLOW model divides the study area into a uniform grid based on the geological characteristics of the area. The grid in the Naoli River basin is 1000 m × 1000 m, with 22,843 active cells, as shown in Figure 2c.

The coupling of the SWAT-MODFLOW model is achieved by establishing a mapping relationship between the HRUs (hydrological response units) in the SWAT model and the grid cells (cells) in the MODFLOW model [58]. After completing the construction of the four linked files, the system will automatically generate the swatmf_link.txt file, which is used to integrate the parameters and data of the two to achieve the linked operation of the SWAT-MODFLOW model [59].

4.2. Model Calibration and Validation

The setting rate is the monthly average runoff value for the period 2005–2009 and the verification period 2010–2012, and the simulated value is then compared with the measured value. The MODFLOW model is calibrated mainly by combining parameter inversion and manual parameter adjustment to adjust the hydraulic conductivity (K) and water supply degree (μ) of the hydrogeological region. The parameter calibration is shown in

Table 4. Parameter rate determination for the Naoli River basin.

Parameter Name	Physical Meaning	Optimal Value	Range
V__SOL_K(..).sol	Saturated hydraulic conductivity of soil	1839.660522	1636.908936 ~ 2183.40918
V__SLSOIL.hru	Slope length for lateral subsurface flow	113.416641	68.897896 ~ 128.977173
V__RAINHHMX(..).wgn	Maximum half-hour rainfall	22.638643	−3.12659 ~ 71.124802
V__TIMP.bsn	Snowpack temperature lag factor	0.633103	0.582176 ~ 1.12978
V__ALPHA_BNK.rte	Baseflow alpha factor for bank storage	1.127053	0.689446 ~ 1.333934
V__SOL_BD(..).sol	Bulk density of soil layer	1.39	1.22 ~ 1.95
R__CH_S1.sub	Channel slope in sub-basin 1	−2.07	−2.35 ~ 0.24
R__CH_L2.rte	Channel length in routing reach 2	0.88	0.05 ~ 1.70
V__CH_N1.sub	Manning’s n value for the main channel in sub-basin 1	10.07	5.56 ~ 21.61
V__SURLAG.bsn	Surface runoff lag coefficient	4.107	1.06 ~ 11.03
V__CN2.mgt	Curve number for moisture condition II	91.81	78.23 ~ 104.10
V__GW_SPYLD.gw	Specific yield of shallow aquifer	0.10	−0.04 ~ 0.19
V__LAT_TTIME.hru	Lateral flow travel time	−93.95	−107.40 ~ −7.76
V__SLSUBBSN.hru	Average slope length of sub-basin	118.01	75.30 ~ 142.57
V__GWQMN.gw	Threshold depth of water in shallow aquifer	−1326.12	−4392.32 ~ −839.36
V__EPCO.hru	Plant uptake compensation factor	0.57	0.55 ~ 0.93
V__PLAPS.sub	Precipitation lapse rate	−490.06	−793.15 ~ −385.23
K	Permeability coefficient k (cm/s)	0.001	/
μ	Degree of water hardness	0.03	/

Note: “/” indicates no data available.

.

For the SWAT-MODFLOW coupled model, the simulated monthly average flow in the basin is the sum of the SWAT model’s surface simulation results and the MODFLOW groundwater recharge river simulation results. The runoff simulation results at the site are shown in Figure 3. As can be seen from the figure, during the simulation period, R² and NSE were both greater than 0.67, and PBIAS was −12.96%, indicating that the model had good fitting ability during this period. In the validation period, R² was 0.84, NSE was 0.799, and PBIAS was −4.79%, further proving the good performance of the model. The model’s degree of fit and predictive ability improved significantly during the verification period, which indicates that the model can effectively capture the trend of flow changes during both the simulation and verification stages and has strong predictive ability, making it of good practical application value.

5. Results and Analysis

5.1. CMIP6 Model Simulation Capability Assessment

To select the most suitable CMIP6 model for simulating future climate scenarios in a catchment area and for runoff and hydrological drought prediction, it is typically necessary to rank the models based on their performance [60]. The Taylor plot is a widely used robust tool that effectively evaluates the relative performance of competing models and tracks overall performance as models evolve [61]. It integrates three statistical metrics: correlation coefficient (r), root mean square error (RMSE) and spatial standard deviation ratio (SD), thereby comprehensively assessing model consistency with observed data and the accuracy of climate simulation [62]. Therefore, this study compared the Taylor plot of 15 CMIP6 models and the ensemble mean (Multi-Model Mean, MMM) of the five best models selected from them with observed data from the Raohe River basin (1970–2014) and ranked the model performance, as shown in Figure 4.

In terms of temperature prediction, all 15 CMIP6 models exhibited high correlation (r ≥ 0.97) and low RMSE (≤2.98). Additionally, the standard deviation range for maximum temperature (Tmax) was 14.10 to 14.88, and for minimum temperature (Tmin), it was 13.90 to 14.84. These results indicate that all CMIP6 models can effectively predict future temperature changes in the basin.

In terms of precipitation prediction, all models performed well, but only five models demonstrated high correlation (r ≥ 0.75), low RMSE (≤30.99) and high similarity with the standard deviation of observed data (41.28). These five models are EC-Earth3, IPSL-CM6A-LR, MPI-ESM1-2-HR, MPI-ESM1-2-LR and NorESM2-MM. Therefore, these five models were selected to construct the ensemble mean, resulting in the optimal multi-model ensemble mean model (Multi-Model Mean-Best, MMM-Best). This model exhibits the highest correlation (r = 0.80), the smallest RMSE (26.15) and high consistency with the standard deviation of observed data (41.28). Ultimately, the MMM-Best model was selected as the optimal climate model for predicting future runoff and hydrological drought in the Raohe River basin from 2025 to 2100.

5.2. Future Trends in Hydro-Meteorological Changes

Based on the two scenarios of SSP2-4.5 and SSP5-8.5 under the meteorological mode of MMM-Best after ensemble averaging, a linear regression analysis was performed on the future (2016–2100) precipitation, temperature and evapotranspiration in the Naoli River basin and the runoff changes at the Caizui Station obtained by the SWAT model driven by future meteorological data (Figure 5) (

Table 5. Annual trends in temperature, rainfall and evapotranspiration in the MMM-Best climate model for the period of 2016–2100.

Scenario	Time Slot	Maximum Temperature	Lowest Temperature	Precipitation	ET	Runoff
SSP2-4.5	2016–2100	0.315 **	0.376 **	0.58 **	0.95 **	0.04
	2016–2060	0.362 **	0.048 **	0.98 *	0.96 **	0.39
	2061–2100	0.027 **	0.023 **	−0.23	1.03 **	−3.53
SSP5-8.5	2016–2100	0.082 **	0.094 **	1.16 **	2.42 *	−0.44
	2016–2060	0.057 **	0.073 **	1.67 *	1.71	2.81
	2061–2100	0.100 **	0.113 **	1.77 *	2.99	1.71

Notes: * indica significativo a 0.05, ** indica significativo a 0.001.

). The results show that the future maximum and minimum temperatures are both indicated to increase significantly. The rate of increase in maximum temperature for the whole year and the near-future scenarios is greater for SSP2-4.5 than for SSP5-8.5, while the opposite is true for the long-term scenario; for minimum temperature, the rate of increase for SSP2-4.5 is greater than that for SSP5-8.5 for the near-future scenario, while the opposite is true for the other periods. Rainfall shows a significant increasing trend in all periods except for the long-term scenario for SSP2-4.5, except for a non-significant decrease. The increasing trend of SSP5-8.5 is significantly greater than that of SSP2-4.5. Evapotranspiration shows a significant increasing trend at different times in the SSP2-4.5 scenario, with the long-term increasing trend being slightly greater than the rest of the period. Under the SSP5-8.5 scenario, there is an overall significant decreasing trend, with a slight increasing trend in evapotranspiration in the near future. Runoff shows a non-significant trend in both scenarios but a strong downward trend in the long term under SSP-2.45. There is also a downward trend in the overall time period under SSP5-8.5. In the near term, SSP5-8.5 shows a significant growth trend that is far more obvious than the growth trend of the same period under SSP2-4.5.

The total runoff volume can be initially studied through time series analysis, as shown in Figure 5. In the SSP2-4.5 and SSP5-8.5 scenarios, there are large fluctuations in runoff volume in the short term, indicating that in the early stages of climate change, runoff volume is affected by extreme weather and seasonal changes, resulting in large fluctuations. Over time, long-term runoff fluctuations have decreased significantly, especially in the SSP2-4.5 scenario, where the trend in runoff volume has become more stable. The analysis of the SSP scenarios shows that under the SSP2-4.5 scenario, the interannual fluctuations in runoff gradually tend to stabilize, with some fluctuations in the short term, but in the long term, runoff shows a slight increasing trend, especially in the multi-year moving average, which reflects the gradual accumulation of the long-term effects of climate change. Under the SSP5-8.5 scenario, runoff fluctuates more in the short term, reflecting the impact of higher greenhouse gas concentrations and more drastic climate change on runoff. In the long term, runoff changes are more complicated, but there is an overall increasing trend. The fluctuation trends of each scenario are obvious in the 10-year moving average curve, and a stable upward or downward trend is observed in the 40-year moving average. The overall trend of runoff under different moving averages is the same.

Based on a comprehensive analysis of the various components of the hydrological cycle, future scenarios were compared with data from the baseline period (1965–2014), as shown in Figure 6. Under future scenarios, the annual average temperature in the basin is higher than that in the baseline period, with increases of 1.89 °C and 3.22 °C under the SSP2-4.5 and SSP5-8.5 scenarios, respectively. In terms of annual average precipitation, the SSP2-4.5 and SSP5-8.5 scenarios show increases of 32% and 36.19%, respectively. In different seasons of precipitation and temperature, the increases in the SSP5-8.5 scenario are generally higher than those in the SSP2-4.5 scenario, except in autumn. In terms of runoff, runoff decreases in summer and autumn, while it increases in other seasons, with the most significant increase in winter; except for the larger increase in the SSP5-8.5 scenario in autumn, the increases in other seasons are smaller than those in the SSP2-4.5 scenario.

5.3. Trend and Sudden Change Tests

To analyze the seasonal variability of hydrological drought, this study employs the 3-month Standardized Precipitation Index (SRI-3) as a drought indicator. Figure 7 shows the drought trend and abrupt change test results based on SRI-3. The results indicate that the hydrological drought sequences under SSP2-4.5 and SSP5-8.5 scenarios both exhibit a significant intensification trend, but the trend under the SSP5-8.5 scenario becomes evident earlier (June 2016 compared with April 2016), has more frequent pre-tipping signs (September 2030–January 2031 compared with November–December 2045), and has earlier onset of abrupt changes (April 2060 compared with July 2074). The SSP2-4.5 scenario reached higher intensity (UF = 15.05 compared with 10.40) by July 2074, indicating that mid-emission scenarios may lead to more prolonged drought impacts, while high-emission scenarios accelerate the intensification of droughts. The periods from September 2030 to 2047 (SSP5-8.5) and November 2045 to 2055 (SSP2-4.5) are critical transition periods.

5.4. Future Changes in Hydrological Drought

The interannual variability characteristics of hydrological drought, characterized by the 12-month SRI (SRI-12), are shown in

Table 6. Drought characteristics of the Rehe River basin under different future scenarios.

Scenario	Time Slot	Frequency of Drought	Frequency of Drought	Average Duration	Average Drought Intensity	Average Drought Severity	Heavy, Very Dry, Frequent	SRImin
SSP2-4.5	2016–2060	13	36.7%	15.23	−1.1	−16.32	8	−2.33
	2061–2100	10	25.0%	12	−0.88	−12.1	3	−2.03
SSP5-8.5	2016–2060	9	27.7%	16.67	−1.22	−37.62	5	−4.04
	2061–2100	11	35.2%	15.36	−0.91	−12.54	0	−1.45

and Figure 8. Between 2016 and 2100, 23 and 20 drought events occurred under the SSP2-4.5 and SSP5-8.5 scenarios, respectively, with maximum drought peaks of 2.33 and 4.04, both occurring in January 2016. Overall, the trend change rates under the two scenarios were 0.0000037/month and −0.0000015/month, respectively. Under the SSP2-4.5 scenario, the average drought intensity was −1.1 and the average drought severity was −16.32 during the period from 2016 to 2060, with eight severe drought events occurring; during the period from 2061 to 2100, drought intensity weakened to −0.88, average drought severity decreased to −12.1, and the number of severe drought events reduced to three, indicating that drought conditions began to ease after approximately 2066. Under the SSP5-8.5 scenario, the overall duration of drought and the frequency of long-term drought increased, while the frequency of short-term drought decreased, with the trend change rate showing a declining trend, indicating that drought conditions are gradually worsening. Specifically, during the 2016–2060 period, drought intensity reached −1.22, with an average drought severity as high as −37.62, indicating frequent and severe early extreme drought events; during the 2061–2100 period, drought intensity weakened to −0.75, with average drought severity significantly decreasing to −12.54, no severe drought events occurring and the long-term drought frequency slightly increasing to 3 5.2%.

Based on the analysis of monthly SRI-1 changes under different future scenarios shown in Figure 9, annual runoff drought exhibits significant seasonal differences. Under the SSP2-4.5 scenario, winter runoff drought shows a continuous mitigation trend, while spring and summer droughts generally worsen, with drought severity continuing to intensify from May to July, manifested by the deepening and expansion of red areas. In contrast, under the SSP5-8.5 scenario, winter runoff drought also shows a continuous mitigation trend, with SRI values maintaining a relatively stable state around zero over the long term. Runoff drought experiences some relief in April and August, with an overall improvement characterized by a gradual reduction in the extent of red zones and a lightening of their color. However, the trend from May to July remains consistent with that under the SSP2-4.5 scenario, and runoff drought continues to intensify, indicating that regardless of the emission scenario, runoff drought risks during the late spring and early summer period will face severe challenges.

6. Discussion

The SWAT-MODFLOW model, driven by future meteorological data such as precipitation and temperature based on the CMIP6 climate model, is divided into future meteorological conditions. It can be observed that future temperature, precipitation and evapotranspiration all show a significant upward trend under different scenarios. Compared with the SSP2-4.5 scenario, the SSP5-8.5 scenario exhibits a more pronounced decrease in runoff and hydrological drought. These findings are consistent with the trends reported by Rashid, H et al. [63] and Wang, S et al. [64].

In addition, this study selected multiple meteorological model schemes for screening and used the optimal model for ensemble averaging. However, downscaling methods were not used to correct seasonal biases in future data based on historical periods. Relying solely on spatial disaggregation (SD) and bias correction (BC) methods applied to the data itself still has certain limitations and is unable to effectively address prediction biases caused by increased frequency and uncertainty of future climate change [65]. Although the ensemble averaging of different models can reduce time uncertainty, and some studies have shown that multi-model processing has high simulation accuracy for meteorological indicators in the northeast region, the impact of this method on the temporal homogeneity of extreme hydro-meteorological simulations cannot be ruled out [58]. Analysis of runoff using a SWAT-MODFLOW model driven by future meteorological data indicates that runoff under different future meteorological models exhibits significant fluctuations. As precipitation gradually decreases and evapotranspiration continues to increase, runoff will gradually decrease under constant soil water storage capacity. Among these, the impact of the high-emission scenario SSP5-8.5 is the most pronounced [66]. This paper simulates future runoff and hydrological drought changes in the entire basin but lacks specific descriptions of the spatial distribution of drought in the basin. At the same time, the relationship between meteorological drought and hydrological drought in future scenarios and changes compared with historical periods remains to be explored [17,67]. In addition, the SWAT model used default values when simulating atmospheric CO₂ concentrations, which introduced systematic errors into future scenario simulations and requires further improvement [62]. Future scenarios not only include changes in climate factors but also consider the impact of land use changes [68]. For runoff prediction, the SWAT-MODFLOW model can be combined with artificial intelligence methods (such as machine learning and deep learning) to correct future prediction biases [69]; at the same time, it is necessary to conduct a comprehensive analysis combining meteorological drought and runoff drought, which is of great significance for improving the accuracy of future basin drought simulations [70]. The multi-model ensemble averaging, bias correction (BC) and spatial decomposition (SD) methods used in this study are transferable to a certain extent and are applicable to other regions with similar hydrological characteristics or facing complex climate change scenarios. They provide valuable references for multi-climate model comparisons, integrated simulations and uncertainty control.

In future research, to address the current limitations, the following improvements will be made: first, statistical downscaling methods based on observational data (such as CDF-t and SDSM) will be used to correct seasonal biases in CMIP6 meteorological data to improve the reliability and temporal consistency of future meteorological driving data [71]; second, by combining remote sensing data with GIS spatial analysis methods, we will conduct a detailed description of the spatial distribution of future hydrological droughts to reveal the spatiotemporal characteristics of drought evolution in different regions [72]; third, by constructing quantitative relationship models between drought indices (such as SPI and SDI), we will systematically analyze the coupling mechanisms and time lag effects between meteorological drought and hydrological drought, thereby enhancing our understanding of future drought risks and improving early warning capabilities [73]; fourth, land use/land cover change (LUCC) simulation models (such as CLUE-S and PLUS models) will be introduced and combined with climate change scenarios to drive the SWAT-MODFLOW model, simulating hydrological responses under multi-factor coupling [68]; and finally, we will explore simulation bias correction and result optimization strategies based on machine learning (such as random forests and LSTM) and deep learning methods to improve the accuracy and adaptability of complex system simulations [74]. These improvements are expected to significantly enhance the model’s applicability and predictive capabilities across multiple scenarios, thereby advancing scientific assessments and management of future hydrological processes and drought risks in river basins.

7. Conclusions

In this paper, 15 CMIP6 model data were used. Based on the measured data, a Taylor analysis was performed on the historical period of temperature and rainfall under future scenarios in the Naoli River basin, and five optimal models were selected. The ensemble mean (MME-Best) of the optimal models was used to drive the SWAT-MODFLOW model to predict the future runoff changes in the basin. Based on the runoff data, the monthly scale (SRI-1, SRI-12, SRI-3) was calculated to analyze the future trend of hydrological drought. The conclusions are as follows:

(1)

The SWAT-MODFLOW hydrological model constructed based on measured meteorological data is of good applicability. The monthly runoff rate determination and verification R² for the Caizuzi hydrological station under the control of the watershed is 0.71 and 0.84, and the NSE is 0.673 and 0.799.

(2)

The multi-model after ensemble mean screening can better improve the accuracy of the simulation and reduce the instability of a single CMIP6 climate model.

(3)

The results of future climate and runoff changes in the Naoli River basin under different scenarios show that the maximum and minimum temperatures will increase significantly in the future, and precipitation will generally increase, with a greater increase in SSP5-8.5. Evapotranspiration will increase under SSP2-4.5 and generally decrease under SSP5-8.5. Runoff will show no significant trend under both scenarios but will show a downward trend in the long term.

(4)

Runoff changes frequently within 10 years in each scenario. The future runoff in the Naoli River basin in each scenario shows a trend of first increasing and then decreasing. The turning point in the SSP2-4.5 scenario is in the long term, and in the SSP5-8.5 scenario, it is in the short term.

(5)

The SRI under the SSP2-4.5 scenario shows an increasing trend, and the frequency of droughts and extremely severe droughts is highest in the short term. The frequency of extreme droughts is zero under the SSP5-8.5 scenario, and the SRI shows a decreasing trend, but the frequency of runoff droughts is relatively high in the long term.

(6)

In future scenario simulations, compared with the baseline period (1965–2014), both the annual average temperature and precipitation in the watershed increased significantly. Under the SSP2-4.5 and SSP5-8.5 scenarios, the temperature increased by 1.89 °C and 3.22 °C, respectively, and the annual precipitation increased by 32% and 36.19%, respectively. Although overall precipitation and temperature showed an increasing trend, runoff during summer and autumn decreases.

(7)

Analysis based on the SRI-3 model indicates that both future emission scenarios show a significant strengthening trend in hydrological drought. Under the SSP5-8.5 scenario, drought intensification occurs earlier (June 2016), with more frequent early warning signals and an earlier onset of abrupt changes (2060). In contrast, under the SSP2-4.5 scenario, although the onset of abrupt changes is later (2074), there is ultimately higher drought intensity (UF = 15.05). Additionally, the periods from 2030 to 2047 (SSP5-8.5) and from 2045 to 2055 (SSP2-4.5) are identified as critical drought transition periods, indicating that different emission pathways will exert distinct influences on the temporal evolution and intensity of future droughts.