Impact of Urbanization-Driven Land Use Changes on Runoff in the Upstream Mountainous Basin of Baiyangdian, China: A Multi-Scenario Simulation Study

Abstract

Urbanization in the Haihe River Basin in northern China, particularly the upstream mountainous basin of Baiyangdian, has significantly altered land use and runoff processes. The runoff is a key water source for downstream areas like Baiyangdian and the Xiong’an New Area, making it essential to understand these changes’ implications for water security. However, the exact implications of these processes remain unclear. To address this gap, a simulation framework combining SWAT+ and CLUE-S was used to analyze runoff responses under different land use scenarios: natural development (ND), farmland protection (FP), and ecological protection (EP). The model simulation results were good, with NSE above 0.7 for SWAT+. The Kappa coefficient for CLUE-S model validation was 0.83. The further study found that from 2005 to 2015, urban construction land increased by 11.50 km² per year, leading to a 0.5–1.3 mm rise in annual runoff. Although urban expansion continued, the other scenarios, which emphasized farmland and forest preservation, slowed this growth. Monthly runoff changes were most significant during the rainy season, with annual runoff in ND, FP, and EP varying by 8.9%, 10.9%, and 7.7%, respectively. While the differences in annual runoff between scenarios were not dramatic, these findings provide a theoretical foundation for future water resource planning and management in the upstream mountainous area of Baiyangdian and offer valuable insights for the sustainable development of Xiong’an New Area. Additionally, these results contribute to the broader field of hydrology by highlighting the importance of considering multiple land use scenarios in runoff change analysis.

1. Introduction

In the past few decades, global land use has undergone significant shifts. These changes primarily include the expansion of agriculture and tropical deforestation, reforestation or afforestation in temperate regions, intensification of cropland, and the growth of urban areas. Forested areas have increased by 2.24 × 10⁶ km², while barren land has decreased by 1.16 × 10⁶ km² [1]. Furthermore, urban growth has resulted in a global increase in built-up areas by 5.3 × 10⁴ km² between 2001 and 2018, with China accounting for 47.5% of this increase [2]. In the global context, rapid alterations in land use have profoundly influenced both human and natural environments. These modifications have affected interactions between land and atmosphere, hydrological cycles, and ecosystem functions at different scales [3,4,5].

Runoff, a crucial component of the basin water cycle, is notably influenced by shifts in land use patterns driven by human activities like urbanization and agricultural development [6,7,8]. With rapid socio-economic development and population growth, urban construction land is also expected to expand [9,10], therefore further altering the spatial distribution of land use in the future. Rapid urbanization has led to the conversion of large areas of agricultural into impervious surfaces. This transformation has enhanced surface runoff production, reduced runoff response time, altered watershed hydrology, and elevated flood risk [11,12]. Concurrently, societal demand for water resources will increase, intensifying the pressure on water resources [13,14,15,16]. Therefore, understanding how land use changes during urban expansion impact the watershed hydrological cycle is essential for efficient future water resource management.

Studies and analyses on the impact of land use on hydrology primarily utilize three methods: the statistical method based on time series analysis, “paired-catchment” experimental, and hydrological models [17,18,19]. A statistical method based on time series analysis [20] is a simple technique that utilizes climate data series from the target area for statistical analysis without accounting for the physical processes within the area. The method based on “paired-catchment” experimental is most appropriate for small basins that are under 100 km² in area [21]. The method based on hydrological models comprehensively considers the interactions between land use and hydrological processes [22,23,24]. Additionally, to evaluate the hydrological effects in different land use development directions, hydrological models often couple land use prediction models. Ali et al. [25] analyzed the future runoff and peak changes in Islamabad using the HEC-HMS model with land use scenarios constructed on a GIS-based platform. Zhou et al. [26] investigated streamflow changes in the Yangtze River Delta region amid future urbanization using SWAT and CLUE-S model. In the Yellow River Basin, Ji et al. [27] forecasted land use in 2050 with the CA-Markov model and simulated future runoff changes using the SWAT model. Lyu et al. [28] simulated the landscape pattern in 2035 using PLUS and evaluated the spatial and temporal dynamics of streamflow under the future scenario using SWAT. Land use pattern simulation and prediction in the aforementioned studies have progressed from subjective modifications of local land use distribution to objective and comprehensive models that integrate quantitative land use forecasts with spatial allocation methods. The selection of hydrological simulation models is more diverse, among which the distributed hydrological model SWAT model is widely applied in hydrological process simulation research due to its ease of operation, data availability and applicability in multiple basins [29,30,31,32,33]. Recently, the SWAT+ model, which improves the representation of spatio-temporal dynamics in hydrological processes, has been developed to overcome the limitations of earlier SWAT versions and to address potential challenges in water resource modeling [34]. Throughout the ongoing development of models, a persistent challenge has been to balance ease of use with the increasing demand for greater spatial complexity [35]. Focusing on a lowland catchment in Northern Germany, Wagner et al. [36] compared SWAT+ setups to an older SWAT model. The results highlighted SWAT+’s enhanced ability to model water resources by better capturing tile drainage flow, surface runoff, and the spatial heterogeneity of landscape characteristics. These improvements suggested the potential for more accurate and effective water resource management. Concurrently, it is also employed in multiple studies, demonstrating strong performance in simulating runoff [37,38,39,40,41,42].

Since 1980, the Haihe River Basin in China has encountered significant transformations and rapid economic development. By 2015, urban and rural land areas had steadily increased, with the largest expansion being 4164 km² [43]. The annual runoff in the Haihe River Basin has shown a significant declining trend, influenced by urbanization and other human activities [44]. A critical sub-basin of the Haihe River Basin, the upstream mountainous basin of Baiyangdian, plays a pivotal role in producing runoff for Baiyangdian Lake—the largest freshwater lake wetland in the North China Plain. This basin alone contributes approximately 80% of the total water inflow to Baiyangdian Lake. Consequently, it is crucial for supplying water to major reservoirs within the Baiyangdian Lake Basin, including Wangkuai, Xidayang and Angezhuang. These reservoirs are vital for maintaining the ecological security of the Xiong’an New Area, a nationally significant development region, by ensuring a stable water supply to the downstream Baiyangdian Lake area [45,46,47]. However, the implications of land use changes driven by socio-economic development in this basin on runoff production remain unclear, necessitating further investigation. Therefore, this study aimed to (1) construct a framework of coupled SWAT+ and CLUE-S; (2) establish various land use development scenarios reflecting urbanization trends; (3) assess and analyze the comprehensive effects of these scenarios on runoff in the upstream mountainous basin of Baiyangdian.

2. Materials and Methods

2.1. Study Area

The upstream mountainous basin of Baiyangdian, situated in the northern part of the North China Plain, belonging to the Haihe Basin, encompasses approximately 11,221 km² (113°39′~116°00′ E, 38°46′~40°03′ N). The exact geographical location is depicted in Figure 1a. Characterized by a warm temperate, semi-humid monsoon climate, the basin experiences an average annual temperature of 11 °C and annual precipitation of 680 mm. The upstream mountainous basin of Baiyangdian is connected to 15 counties, distributed across Hebei Province, Shanxi Province, and Beijing, with Hebei Province encompassing the largest portion (60%). The total population of the basin was approximately 1.68 × 10^6, and the GDP was around 6.34 × 10⁶ yuan in 2020. A detailed overview of the socio-economic conditions of each administrative region is given in Table S1.

From 1960 to 2019, the annual precipitation in the upstream mountainous basin of the Baiyangdian declined by 4.59 mm/10 year, while the annual temperature rose by 0.19 °C/10 year (Figure 1b,c). Within the basin, urban lands and farmlands are situated in the flat central areas. Forests are primarily found in the northeast of the basin. Grasslands are mostly located in the southwest and central regions. The upstream mountainous region of Baiyangdian acts as the primary water source for Baiyangdian Lake, essential for maintaining the stability of the Baiyangdian ecosystem. The sustainability of water resources in Baiyangdian Lake is closely connected to the future development of Xiong’an New Area. The upstream mountainous area of Baiyangdian is categorized into four major sub-basins based on the water sources that flow into downstream reservoirs (Figure 1a). They are the Shahe sub-basin (SRS), managed by the Fuping station; the Tanghe sub-basin (TRS), overseen by the Zhongtangmei station; the Jumahe sub-basin (JRS), controlled by the Luobaotan station; and the Dashihe sub-basin (DRS), supervised by the Manshuihe station. Details of all hydrological stations are provided in Table S2.

2.2. Research Framework for Runoff Response to Land Use Change

A coupling analysis framework (Figure 2) was developed in this study to analyze the changes in runoff under various land use development scenarios in the upstream mountainous area of Baiyangdian using the SWAT+ hydrological model [34] and CLUE-S land use simulation model [48]. First, a database for model simulation and verification was established. The data and their sources were thoroughly described in Section 2.3. Second, the SWAT+ hydrological model and CLUE-S land use model of the basin were established, calibrated, and verified. The CLUE-S model involved predicting the spatial distribution probabilities of various land use types by analyzing the driving factors influencing their distribution. The spatial allocation of land use demands under different scenarios was then executed through iterative processes. For the SWAT+ model, steps were followed to construct the model, the most important of which was the establishment of the Hydrological Response Units (HRUs). The HRUs were the basic computational unit of the SWAT+ model and were formed from combinations of land use types, soil types, and slope classes within different sub-basins. Detailed procedures for the construction, calibration, and validation of the SWAT+ and CLUE-S models were provided in Section 2.4 and Section 2.5. Finally, the runoff changes generated by various land use scenarios simulated by CLUE-S were quantified by the calibrated SWAT+ model. Section 2.6 provided a detailed explanation of the methods for land use scenarios setting up.

2.3. Data Source

The data used to drive both the SWAT+ and CLUE-S models primarily consist of basin spatial data, climate, and streamflow observation data. In addition, the socio-economic data are also required to drive the CLUE-S model. The nearest neighbor method was used to resample all raster data to 90 m, ensuring consistent resolution.

The Digital Elevation Model (DEM) was SRTMDEM from Geospatial Data Cloud (http://www.gscloud.cn/, accessed on 27 September 2022). The basin slope data were generated by the DEM data. The soil data were extracted from the World Soil Database (HWSD; http://fao.org/soils-portal/soil-survey, accessed on 27 September 2022). The soil characteristic parameters (soil saturated water conductivity, effective water holding capacity, and soil density) were computed by the software Soil–Plant–Atmosphere–Water [49] (SPAW). The historical land use data were acquired from the Resource and Environmental Science and Data Center (RESDC; https://www.resdc.cn, accessed on 19 October 2023). The average soil organic carbon density data of 0–5 cm, which was one of the driving factors of CLUE-S model modeling, was derived from the soil information data set SoilGrids 2.0 [50].

Historical climate data were daily observation data from three climate stations in the basin and four stations around the basin from 1959 to 2019, acquired from RESDC. In this study, a weather generator, WXGEN [51], integrated into the SWAT+ model was constructed to simulate climate data from 2020 to 2030 using the statistical characteristics of historical weather station data.

The observed data of monthly streamflow from six hydrological stations (Fuping, Zhongtangmei, Zijingguan, Zhangfang, Luobaotan, and Manshuihe) were collected from the China Hydrological Yearbook for 1969–1986.

The socio-economic data included river basin water use, river basin water withdrawal distance, population density, and GDP data, where population density and GDP were obtained from RESDC. The water withdrawal distance of the basin is the Euclidean distance between each grid and the river.

2.4. SWAT+ Model

Simulating the hydrological process in the upstream mountainous basin of Baiyangdian was achieved using SWAT+, an improved hydrological model based on SWAT. The SWAT+ uses a modular structure, allowing for greater flexibility in sub-basin discretization and configuration [34]. It also provides two options for calculating runoff, namely the Soil Conservation Service curve number (SCS-CN) procedure [52] and the Green and Ampt infiltration method [53].

Runoff simulation in this study utilized the SCS-CN method, noted for its efficiency and popularity [54,55,56]. The SCS-CN relies on an empirical formula created from extensive observations of rainfall and runoff data, incorporating various land use and other factors combinations. The specific calculation equations are as follows:

$$Q_{surf}={\displaystyle \frac{{\left(R_{day}-0.2S\right)}^2}{\left(R_{day}+0.8S\right)}}$$

(1)

where Q_surf is runoff (mm); R_day is the daily rainfall (mm); S represents the retention parameter (mm), which can be calculated by the following equation:

$$S=25.4\left({\displaystyle \frac{1000}{CN}}\right)-10$$

(2)

where CN is the curve number, typically obtained from the model user manual or determined through model parameter calibration. The CN in this study was determined by calibrating the parameter cn2.

Using the QSWAT+ plugin in QGIS, we constructed a SWAT+ model for the upstream mountainous basin of Baiyangdian, ultimately dividing it into 17 sub-basins with 3462 hydrological response units. The year range of 1966–1977 was set as the model calibration period (1966–1968 was the model warm-up period), and 1978–1980 was the model validation period in this study. Referring to existing studies [57] and sensitivity analysis using the Sobol method, 19 parameters associated with streamflow change were selected for calibration (Table S3). The detailed results of the parameter sensitivity analysis are provided in Figure S1. We utilized the multi-site calibration method [58] in this study, which employed streamflow data from multiple available hydrological stations across the entire basin for calibration. It is more effective in improving model simulation performance compared to using only the basin outlet site calibration [59,60]. Therefore, the monthly measured streamflow data from six hydrological stations were used for calibration, following the principle of upstream first and downstream later. First, the upstream hydrological stations of Fuping, Zhongtangmei, Zijingguan, and Manshuihe were calibrated. After the calibration results of the upstream stations met the accuracy requirements, the upstream parameter values were kept unchanged, and the downstream Zhangfang station was calibrated. Finally, we calibrated the Luobaotan station.

The accuracy of streamflow simulation was evaluated using three model evaluation performance indicators [41]: coefficient of determination (R²), Nash-Sutcliffe efficiency [61] (NSE), and percent bias [62] (PBIAS). The R² can be used to assess the linear correlation between the simulated and observed values, and results with R² exceeding 0.5 are generally deemed acceptable. The NSE measures the degree to which the hydrological model simulation matches the overall trend of the observed data. The PBIAS indicates the extent to which the model simulation results either overestimate or underestimate compared to the observed values. An NSE exceeded 0.7, and a PBIAS absolute value less than 25 indicates that the model performs well [63]. Following are the methods for calculating the three indicators:

$$R^2={\displaystyle \frac{{\left[\sum _{i=1}^n\left(O_i-O_{avg}\right)\left(P_i-P_{avg}\right)\right]}^2}{\sum _{i=1}^n{\left(O_i-O_{avg}\right)}^2\sum _{i=1}^n{\left(P_i-P_{avg}\right)}^2}}$$

(3)

$$\begin{array}{c}NSE=1-{\displaystyle \frac{\sum _{i=1}^n{\left(O_i-P_i\right)}^2}{\sum _{i=1}^n{\left(O_i-O_{avg}\right)}^2}}\end{array}$$

(4)

$$\begin{array}{c}PBIAS={\displaystyle \frac{\sum _{i=1}^n\left(P_i-O_i\right)}{\sum _{i=1}^n\left(O_i\right)}}\times 100\end{array}$$

(5)

where n is the number of observed data, O_i is the ith observed data, O_avg refers to the average value of the observations, P_i is the ith simulated data, and P_avg refers to the average value of the simulations.

Furthermore, model uncertainty was quantified using the 95% prediction uncertainty (95PPU) method [64]. A detailed explanation of the methodology and results can be found in the Supplementary text and Table S4.

2.5. CLUE-S Model

The land use simulation model CLUE-S [48] was employed in this study to predict future land use transformations. Assuming that in the short term, the connections among land use processes and natural and socio-economic drivers remain unchanged. The CLUE-S model simulates the spatial distributions of short-term land use change with the regression relationships between historical land use patterns and driving forces [65]. The model consists primarily of two modules: the land use demand module and the spatially explicit allocation module [66]. The spatial allocation process is accomplished using a stepwise logistic regression method, which effectively filters out driving factors that do not significantly contribute to changes in land use patterns:

$$L\mathit{og}\left({\displaystyle \frac{P_i}{1-P_i}}\right)={\beta }_0+{\beta }_1{X}_{1,i}+{\beta }_2{X}_{2,i}+\cdots +{\beta }_n{X}_{n,i}$$

(6)

where P_i is the probability that the land use type i is assigned to a certain grid cell, X_n,i is the value of driving factor n in land use type i, β₀ represents the intercept, and β_n represents the coefficient of the driving factor n in the logistic regression equation, used to indicate the correlation between the distribution probability of land use types and the driving factors.

Actual land use data from 2015 was used in this study as a simulation reference. We constructed a logistic regression model using natural climate factors (DEM, slope, soil organic carbon density, precipitation, temperature) and socio-economic factors (river water intake distance, population density, GDP) to generate the spatial suitability distribution for each land use type based on the 2015 land use pattern. The Receiver Operating Characteristic (ROC) curve was employed to assess the goodness of fit of the logistic regression model [67]. The area under the ROC curve (AUC) measures the trade-off between true and false positives across an extensive set of cut-off thresholds [68,69]. The closer the AUC result is to 1, the stronger the predictive ability and the higher the accuracy of the model.

A logistics regression model for land use, developed using the CLUE-S model, was presented in Table S5 in this study. For all land use types, AUC values exceeded 0.7, indicating a high spatial similarity between the modeled and actual land use in the established model.

To further assess CLUE-S model accuracy in predicting land use spatial distribution, we used the Kappa coefficient [70] to test the simulated land use results (Equation (7)). The Kappa coefficient spans from −1 to 1 and is deemed acceptable when above 0.7. The higher the coefficient, the more precise the simulation.

$$Kappa=\left(P_o-P_c\right)/\left(P_p-P_c\right)$$

(7)

where P_o is the correct simulation proportion; P_c is the correct simulation proportion for a random condition; P_p is the correct simulation proportion under perfect classification.

2.6. Land Use Development Scenarios Establishment

The land use demand module of the CLUE-S model is generally used in conjunction with external methods. Accordingly, this study first employed a Markov chain to predict the required quantities of different land use types across various scenarios against the backdrop of urban expansion. Subsequently, using the spatially explicit allocation module of CLUE-S and actual land use data from 2015, we simulated the spatial distribution of diverse land use development scenarios. Considering the relevant planning for future social and economic development, along with agricultural and ecological protection needs in the basin, we established three land use development scenarios:

The natural development scenario (ND): This scenario projected land use evolution to follow the 1980–2015 trend, with decreases in farmland, forest, and grassland areas and an increase in urban land areas.

The farmland protection scenario (FP): This scenario aimed to safeguard future agricultural development in the basin and mitigate potential urban land expansion in ND by reducing the probability of farmland conversion to urban areas by 50%, using conversion probabilities from 1980 to 2015.

The ecological protection scenario (EP): This scenario aimed to promote ecological restoration and curb the anticipated increase in urban land. The farmland, forest, grassland, and water were set to reduce the probability of converting into urban areas by 50%. Additionally, the forest, grassland, and water also had a 50% probability of being reduced into farmland.

3. Results

3.1. Model Calibration and Validation

Table 1. SWAT+ model parameter calibrated results (Refer to Table S3 for parameter definitions).

Parameter	Types of Change 1	Calibrated Value
		Zijingguan	Fuping	Manshuihe	Zhongtangmei	Zhangfang	Luobaotan
cn2	pctchg	5.47	−7.26	−6.12	13.60	−0.63	4.80
lat_len	abschg	−4.54	−27.79	−23.29	4.70	14.46	−8.19
lat_ttime	abschg	−0.21	5.61	25.07	−25.73	2.87	−16.56
esco	absval	0.25	0.02	0.85	0.07	0.89	0.18
epco	absval	0.34	0.35	0.69	0.57	0.06	0.75
slope	abschg	5.83	2.40	5.79	1.98	−24.37	−20.04
canmx	absval	1.94	6.36	23.10	11.68	87.06	84.20
ovn	abschg	4.87	−16.12	−24.95	−16.39	−10.26	−4.63
snofall_tmp	abschg	18.13	−8.10	17.57	26.11	−8.72	−3.42
snomelt_tmp	abschg	−3.38	3.04	7.56	−35.16	−3.50	−16.00
snomelt_max	absval	1.72	3.14	8.49	2.22	2.42	4.56
snomelt_min	absval	1.02	2.94	1.58	5.36	5.77	9.59
snomelt_lag	absval	0.13	0.17	0.97	0.64	0.34	0.43
alpha	absval	0.01	0.85	0.37	0.49	0.93	0.01
flo_min	abschg	13.37	22.53	−32.55	16.58	−21.90	−2.17
revap_co	abschg	14.26	−0.65	−31.23	25.86	25.76	8.80
revap_min	abschg	−11.60	−2.01	36.27	10.02	−11.88	19.92
k	abschg	8.19	34.85	−3.38	0.34	34.26	15.63
awc	abschg	−17.54	27.37	−7.39	−35.41	21.92	−16.74

¹ Note: pctchg refers to change the value by the specified percent; abschg refers to change the value by the specified amount; absval refers to change the value of the parameter.

displays the parameter calibration results from 1966 to 1977. The calibrated model performed well during the validation period, with simulated and observed streamflow changes showing consistent trends (Figure 3). Throughout the entire simulation period, the simulated streamflow during the non-flood season indicated lower simulation accuracy for low streamflow compared to high streamflow. This discrepancy might be attributed to inadequate modeling of groundwater changes in the region by the groundwater simulation module of the SWAT+ model [71]. Figure 3 also presents the simulation accuracy for each hydrological station. All these stations achieved NSE values exceeding 0.70, with R² values above 0.75, indicating satisfactory simulation accuracy and validating the reliability of the model results.

The constructed CLUE-S model was used to simulate the land use distribution for 2015. A Kappa coefficient of 0.83 (

Table 2. Comparison of simulation and observation results of land use patterns in 2015.

Kappa = 0.83		Number of Simulated Grids
		Farmland	Forest	Grassland	Water	Urban	Bare Land	Producer Accuracy%
Number of actual grids	Farmland	136,140	5170	14,068	2514	4739	16	83.70
	Forest	5428	635,004	33,398	1030	2216	13	93.78
	Grassland	14,276	32,878	442,689	1556	7335	32	88.76
	Water	1904	1236	1819	11,523	558	1	67.62
	Urban	4880	2787	6755	373	14,849	55	50.00
	Bare land	0	14	36	66	0	0	0.00
	User accuracy%	83.71	93.78	88.76	67.54	50.00	0.00	-

) was obtained, demonstrating a high level of consistency comparing the simulated land use with the actual land use in 2015. The overall accuracy of the land use simulation was considered acceptable, supporting the model application in future land use scenario simulations.

3.2. Historical Land Use Changes and Land Use Development Scenarios

The land use changes in the upstream mountainous basin of Baiyangdian displayed various characteristics in different periods (Figure 4). During the decade from 1980 to 1990, the forest in the watershed decreased rapidly (−5.98 km²/year), while the grassland increased significantly (5.52 km²/year). From 1990 to 2000, grassland areas began to decline, with other land use types increasing. Between 2000 and 2005, farmland decreased. From 2005 to 2015, farmland and grassland areas decreased sharply at rates of 7.20 km²/year and 5.91 km²/year, respectively, alongside an accelerated urban expansion rate of 11.50 km²/year. In summary, socio-economic development in the basin from 1980 to 2015 had led to an increasingly dominant annual rate of urban expansion. According to the land use transition matrix for this period (

Table 3. Land use transfer matrix in the upstream mountainous area of Baiyangdian from 1980 to 2015 (km²).

1980	2015
	Farmland	Forest	Grassland	Water	Urban	Bare Land	Transfer-Out Area
Farmland	1138.21	52.21	124.89	16.94	66.52	0.00	260.55
Forest	42.62	5175.68	271.85	10.59	27.97	0.11	353.14
Grassland	111.29	245.61	3628.92	14.45	59.69	0.23	431.28
Water	15.23	5.04	10.61	95.09	3.22	0.00	34.10
Urban	10.07	4.96	3.62	0.97	83.16	0.00	19.62
Bare land	0.02	0.92	0.11	0.00	0.00	0.60	1.05
Transfer-in area	179.23	308.74	411.08	42.95	157.40	0.34	1099.74
Net increase	−81.32	−44.40	−20.19	8.85	137.78	−0.71	-

), urban land saw the most significant increase (137.78 km²), while farmland experienced the largest decrease (81.32 km²). The expanded urban construction land primarily originated from farmland and grassland.

Land use changes from 1980 to 2015 differed among the various sub-watersheds, but all showed an increase in urban construction land (Figure 5). In SRS, the primary changes included the reduction of grassland and the increase of both forest and urban land. The predominant change in TRS was the growth of urban land. The JRS experienced the most extensive land use changes, marked by a 69.00 km² increase in urban land and reductions of 64.21 km² in farmland and 43.69 km² in forest. Despite the less dramatic land use changes in DRS, urban land continued to expand.

Figure 6 illustrates the projected land use areas in 2030 for three different development scenarios derived from the default land use (DLU) simulation of 2015. In scenario ND, a significant urban expansion at a rate of 33.65 km²/year dominated the land use change compared to DLU. By 2030, the urban area was anticipated to increase by 209.85%. In addition, farmland (−4.38%), forest (−5.45%), and grassland (−4.43%) all experienced reductions. In the scenario FP, the urban area expansion rate decreased to 24.65 km²/year. The final urban area increased by 153.69%, and the farmland increased by 3.66%. For the scenario EP, urban area expansion was the slowest at 8.28 km²/year, leading to a final increase of 51.64%. The grassland and water areas rose by 4.80% and 37.23%, respectively. Although the forest area was slightly lower than in DLU, it increased significantly compared to FP and EP scenarios. Generally, urban expansion remained the main trend for future development despite varying degrees of restrictions on the urban growth rate in the FP and EP scenarios.

Under varying land use scenarios, the dominant land use types in the four sub-basins remained unchanged, but the area ratios differed (Figure 7). The SRS sub-basin was predominantly covered by grassland (54.92–57.56%) and forest (37.41–38.36%). In the TRS and JRS, extensive areas of grassland (TRS: 49.38–53.88%; JRS: 19.19–21.81%), forest (TRS: 23.23–25.30%; JRS: 62.85–65.03%) and farmland (TRS: 16.18–20.63%; JRS: 7.99–11.20%) were noted. In the DRS, forest (67.56–76.58%) and urban areas (13.23–27.10%) were the primary land use types, with their proportions changing significantly.

3.3. The Impact of Historical Land Use on Runoff

Examining annual variation trends in simulated runoff differences under historical land use conditions in 1980 and 2015 provided insights into the hydrological responses of the upstream mountainous basin of Baiyangdian and its sub-basins. From 1969 to 2018, the differences in runoff simulated with 2015 land use versus 1980 land use were consistently positive, ranging mainly from 0.5 to 1.3 mm, except in 1996 (Figure 8). The observed land use changes by 2015, characterized by urban expansion and reductions in farmland, forest, and grassland areas, resulted in increased runoff.

From the sub-watershed perspective, runoff changes predominantly increased in three sub-watersheds except for the SRS. The TRS experienced the most significant increase. Influenced by the 2015 land use changes, the runoff in the SRS decreased by an average of 0.3 mm. The TRS exhibited the highest increase in runoff, averaging 1.7 mm. Runoff in DRS and JRS raised by 0.8 mm and 0.7 mm, respectively, similar to the overall basin trend.

3.4. Runoff Response to Different Land Use Change Scenarios

Simulation of runoff changes in the upstream mountainous basin of Baiyangdian under three land use development scenarios from 2020 to 2030 was conducted using the calibrated SWAT+ model. The trends in average annual total runoff changes were not significant across the scenarios, though all showed an increase (Figure 9a). Compared to the DLU, the average annual total runoff increased the most in the FP scenario (10.9%), followed by the ND (8.9%) and EP (7.7%) scenarios. Monthly total runoff changes showed the most pronounced differences from July to August (Figure 9b), with the ND scenario showing the highest runoff totals. The scenario EP had the greatest effect on reducing the runoff peak compared to the ND and FP scenarios.

Figure 10 presents the monthly total runoff of the four major sub-basins in the upstream mountainous area of Baiyangdian and its monthly changes compared to the DLU under different scenarios. Runoff changes in each sub-basin exhibited distinct seasonal differentiation across the three land use development scenarios. During the summer flood season (May to October), runoff in each basin increased substantially with rising precipitation. From November to April, the runoff remained relatively low in SRS and DRS, while TRS and JRS experienced significant changes. Monthly runoff changes exhibited noticeable scenario variability across all sub-basins under different land use development scenarios. Scenario differences in monthly runoff in SRS were most pronounced from September to December compared to the remaining months. The runoff in TRS was highest and increased the most in EP from March to April and October to November, while it was the lowest in ND. Additionally, the rate of runoff increase from June to August in EP was significantly lower than in ND and FP. The disparity in runoff changes across various scenarios in the JRS was most apparent from May to August. Runoff variations displayed an upward trend relative to the DLU scenario, with EP showing the least amount of increase. In the DRS, the most significant runoff changes occurred from May to August under different land use development scenarios, with runoff notably lower in EP compared to ND and FP.

Figure 11 presents the monthly water yield for each land use type, aiming to explore the impact of land use changes on basin runoff more thoroughly. Substantial disparities in water yield were observed across different land use types. However, for each land use type, the alterations in water yield were remarkably alike across different scenarios. Urban land exhibited the highest annual water yield values, notably from June to September, contributing the most to overall water yield. Grassland, farmland, bare land, and forest exhibited similar patterns in water yield, peaking in August. Grassland had the highest water yield, while forest had the lowest. Minimal water yield was observed for all land use types during the dry months (November to April). In terms of annual variation in water yield, urban land demonstrated the most significant changes over the year, whereas forest maintained consistently low levels.

4. Discussion

4.1. Impacts of Different Land Use Scenarios on Runoff

Numerous studies have demonstrated that shifts in spatial patterns of land use can influence the basin runoff [41,72,73,74,75,76]. Historical land use changes from 1980 to 2015 in the upstream mountainous region of Baiyangdian showed increased urban construction land and decreased farmland and grassland areas (Figure 4). According to Han et al. [77], a notable increase in construction land has occurred in the Baiyangdian Basin since the start of the 21st century. This increase indicated that land use changes in the upstream mountainous area of Baiyangdian were consistent with trends observed throughout the entire basin, reflecting the broader impact of socio-economic development. These land use changes resulted in a runoff increase for the whole basin, although the impact was not significant (Figure 8). Lyu et al. [8] found that urban expansion significantly increased runoff in the Fenhe Basin. This result indicated that the effect of urbanization on runoff varies with different basin characteristics. Additionally, runoff in the SRS generally decreased, while runoff in other sub-basins increased (Figure 8). Grassland loss and forest rise might be contributing to the runoff decrease in SRS (Figure 5), a pattern consistent with findings by Xu et al. [78], who reported similar vegetation shifts leading to reduced runoff in forested sub-basins. Additionally, the expansion of urban land in SRS might partially offset the reduction in runoff.

This study further established various land use development scenarios based on historical urbanization trends. The monthly runoff changes in these scenarios closely matched the precipitation distribution, highlighting precipitation as the main factor influencing runoff variations in the basin [79,80,81,82,83]. In this study, the impacts of climate change under different scenarios were controlled using the same meteorological data. Therefore, the discrepancies in runoff changes across different scenarios only came from the different land use spatial patterns. At both annual and monthly scales, the scenario EP yielded the lowest runoff, whereas FP generated the highest (Figure 9). In the scenario EP, the substantial increase in forest resulted in a slowdown in surface flow velocity and an increase in surface soil organic matter, which ultimately increased precipitation infiltration and reduced surface runoff [84,85,86,87]. At the same time, the runoff peaks were attenuated due to the retention of precipitation by the forest canopy, surface litter layer, and enhanced soil infiltration, while water sources conserved by the forest were able to recharge runoff during the dry season [88,89,90,91,92,93]. This could be reflected in the findings of this study (Figure 9 and Figure 10). In the scenario FP, due to repeated cultivation of farmland, simple vegetation, and low coverage, it was easily affected by soil erosion [75,94]. Consequently, the farmland increase could result in higher runoff. Additionally, urban development reduced vegetation cover and increased impermeable surfaces, which decreased soil infiltration and water storage capacity, further increasing surface runoff [95,96,97]. Meanwhile, according to the changes in water yield by land use type (Figure 11), it could be predicted that the larger the urban expansion in the upstream mountainous basin of Baiyangdian, the larger the runoff, especially from June to September, with abundant precipitation. Consequently, the ND scenario exhibited the highest monthly runoff peaks.

4.2. Implications and Policies for Different Sub-Basins

The differences among the four main sub-watersheds in the upstream mountainous area of Baiyangdian under various land use development scenarios might stem from the varying proportions of land use types in each scenario. Among these, DRS exhibited notable differences in monthly runoff changes compared to the other sub-basins (Figure 10). Forest comprised more than 65% of DRS. Under the scenario EP, forest increase was greater than in other sub-basins (Figure 7). Consequently, its role in regulating peak runoff was the most significant. Additionally, precipitation differences across sub-basins also affected runoff production. In sub-basins with abundant precipitation, such as DRS and JRS (Figure 3), urban expansion was expected to positively influence runoff increases (Figure 11), but the potential threat of flooding should be considered. Therefore, these factors exerted a complex influence on sub-basin runoff responses. To address the challenges posed by the complex impact of urbanization-driven land use changes on runoff, implementing targeted management strategies is crucial. For forested sub-basins like DRS, priority should be given to preserving and expanding forest cover to sustain its runoff-regulating benefits [98]. In contrast, for areas such as JRS, which are experiencing urban expansion and increased runoff, flood mitigation measures should be emphasized [99]. These include enhancing drainage infrastructure, incorporating permeable surfaces, and applying sustainable urban planning practices [95,100]. By tailoring countermeasures to the specific characteristics and challenges of each sub-basin, these strategies will not only protect water resources to ensure downstream water supply but also promote sustainable land use practices and enhance overall watershed resilience.

4.3. Limitations and Uncertainties

Distributed hydrological models possess certain uncertainties resulting from their multi-parameter nature, varied input data, and the model complexities construction process [101,102,103]. With only seven climate stations available in this study and their uneven distribution, this is also a source of errors between simulated and measured streamflow [104,105]. The varying topography, land use types, and soil types of each sub-basin also increase the model uncertainty [60,106]. Therefore, to reduce model uncertainty, a multi-site calibration method was employed during the calibration period, which ultimately achieved good simulation performance (Figure 3). In simulating land use spatial distribution, the limited number and quality of driving factors chosen for the CLUE-S introduced a certain degree of uncertainty into the constructed model. Additionally, the Markov chain was employed to forecast future land use demand, which, to some extent, ignored the systematic and stochastic nature of future land use changes [107].

Although climate and land use change are closely related [108,109,110], this study did not consider their interaction effect when analyzing the impact of land use on runoff change. The long-term changes in land use are influenced by climate change through its effects on natural vegetation growth and human activities, including water use [111,112,113,114]. Meanwhile, land use changes can influence local temperature fluctuations by modifying the surface albedo [115,116,117]. Changes in land use type also impact greenhouse gas emissions, therefore affecting climate change [118,119]. Therefore, a complex and changeable relationship exists between climate and land use. Accordingly, future research should consider their interaction more comprehensively.

5. Conclusions

By integrating the SWAT+ with the CLUE-S, this study examined runoff changes in the upstream mountainous area of Baiyangdian across three land use evolution scenarios within an urbanization context. The primary research findings are summarized as follows:

(1)

In the upstream mountainous basin of Baiyangdian, SWAT+ was well applicable, with NSE values exceeding 0.7 for all hydrological stations. The simulated land use data generated by the CLUE-S model demonstrated high spatial distribution consistency with real data, achieving a Kappa coefficient of 0.83.

(2)

Urban land expansion drove changes in land use in ND, FP, and EP, while the FP and EP decelerated this expansion. Additionally, the scenario EP effectively controlled the reduction of forest area in the basin.

(3)

The annual runoff in the upstream basin of Baiyangdian showed an upward trend under the influence of historical urbanization trends in land use. The simulated annual runoff changes under different land use scenarios also exhibited an upward trend, with no significant differences observed among the three scenarios. However, significant seasonal disparities in runoff changes were observed under different scenarios. Generally, the ND and FP scenarios promoted runoff production from May to October, while the EP scenario had a significant reduction effect on the monthly runoff peak from July to August, particularly in the DRS. These differences were mainly attributed to the different contributions of various land use types to water yield.

These findings support local water resource management in the upstream mountainous area of Baiyangdian and provide a scientific basis for future water supply decisions. Beyond this region, the study offers valuable insights applicable to similar watershed challenges globally, especially in areas undergoing rapid urbanization. The methodologies and conclusions can inform ecological security and sustainable development strategies, including in Xiong’an New Area. Additionally, this study contributed to the field of hydrology by demonstrating the importance of integrating multiple land use scenarios in runoff modeling, particularly in the context of urbanization.