Analysis of Runoff Changes and Their Driving Forces in the Minjiang River Basin (Chengdu Section) in the Last 30 Years

Abstract

Surface runoff is a key component of the hydrological cycle and is essential for water resource management and water ecological balance in river basins. It is important to accurately reveal the spatial and temporal dynamics of regional surface runoff over long time scales and to quantify the impacts of climate change and human activities on surface runoff changes for sustainable water resources management and utilization. In this study, the Minjiang River Basin (Chengdu section) was selected, which has significant natural and anthropogenic variations, and a comprehensive analysis of runoff and its drivers will help to formulate an effective regional water resource management strategy. We mainly used SWAT to simulate the monthly-scale runoff in the Chengdu section of the Minjiang River Basin from 1990 to 2019 and combined SWAT-CUP to perform sensitivity analysis on the model parameters and Partial Least Squares Structural Equation Modeling (PLS-SEM) to quantitatively analyze the main drivers of the changes in surface runoff. The results show that the average multi-year runoff in the Minjiang River Basin (Chengdu section) ranges from 628.96 to 1088.46 mm, with an average value of 834.13 mm, and that the overall annual runoff in the past 30 years shows a fluctuating tendency. The goodness-of-fit of the PLS-SEM model is 0.507; the validity and reliability assessment indicated that the model was reasonable, and its results showed that economic and landscape factors had significant negative impacts on runoff changes, while natural factors had positive impacts on runoff changes, with path coefficients of −0.210, −0.131, and 0.367, respectively. Meanwhile, this study also identified two potential indirect impact pathways, i.e., the economic factors had an indirect negative impact on runoff by changing the distribution of landscapes, and the natural factors had indirect negative impacts on runoff by influencing economic activities, reflecting the complex interactions among economic activities, landscape distribution, and natural factors in influencing surface runoff. This study provides a research framework and methodology for quantitatively modeling surface runoff and the analysis of influencing factors in watersheds, contributing to a deeper scientific understanding of long-term runoff changes and the contribution of their drivers.

1. Introduction

Surface runoff, as a key element of the natural water cycle, plays an indispensable role in maintaining the stability of the global ecosystem, not only carrying the water and nutrients necessary for the functioning of the ecosystem but also profoundly affecting the overall health of the ecosystem [1,2]. As a core link in the water cycle, surface water flows through the precipitation transformation mechanism, nourishing freshwater ecosystems such as rivers and lakes and constituting a basic resource to support irrigated agriculture, industrial production, and urban water supply [3]. However, in the macro context of global climate change, the dynamic characteristics of surface runoff are undergoing significant and complex changes, which are impacting ecological security, water resources management strategies, and sustainable development pathways in multiple dimensions [4]. Specifically, a reduction in surface runoff or a change in its pattern could exacerbate the drought stress on ecosystems and affect biodiversity, while threatening the security of human society’s water supply and causing knock-on effects on agricultural production, economic development, and the lives of residents [5]. In order to respond effectively to these challenges, China is systematically addressing water resources, water ecology, water environment, and water disasters on the basis of river basins as a whole, with the aim of realizing harmonious coexistence between human beings and water [6,7].

Climate change and human activities, as important components of the changing environment, have significant impacts on hydrological processes, the spatial and temporal distribution of water resources quantity and quality, and bring about hydrological effects that are of deep concern [8]. The hydrological processes in the watershed are complex and variable, resulting in surface runoff with obvious spatial and temporal distribution characteristics, which are jointly influenced by natural and social factors [9,10,11]. On the one hand, climate change climates alter the amount, frequency, and pattern of rainfall, increase extreme weather events, and affect vegetation, water evapotranspiration, and soil moisture, thereby altering surface runoff volume, peaks, and fluctuation rates [12,13,14]. On the other hand, human activities such as urbanization that change the type and structure of the landscape also affect the distribution and use of surface precipitation; especially the increase in impermeable surfaces promotes evapotranspiration of precipitation and surface runoff and reduces infiltration [15], which is more pronounced in agricultural and urban watersheds [9]. This also leads to an increase in the conflict between water supply and demand, which should be given adequate attention by the management of watershed resources. Therefore, identifying the main drivers of runoff change is key to understanding changes in available water resources.

Current hydrologic simulation methods: commonly used models to simulate surface runoff include the Soil and Water Conservation Service Curve Numbers (SCS-CN) model, the Stormwater Management Model (SWMM), and the SWAT model. Among them, the SCS-CN model is simple in structure and easy to operate but lacks physical significance and is difficult to reflect the spatial heterogeneity of surface runoff [16,17,18]. The SWMM model is powerful and flexible but requires a large number of parameters and measurements to be calibrated and validated [19]. The SWAT model is suitable for simulating hydrologic events of hydrologic events and long time series at the basin scale. It is widely used in different regions of the basin and complex environments [20,21,22]. Therefore, the use of the SWAT model can provide an important source of data on water resources and the hydrological cycle in watersheds in regions where hydrological observation data are lacking. However, some studies have also argued that high input data requirements such as those of the SWAT model and large uncertainties in the model (e.g., input uncertainty, parameter uncertainty, and structural uncertainty) can lead to large errors or even distortions [23]. These factors limit the applicability of hydrologic modeling methods for assessing the impacts of climate and land use or cover change on the hydrologic cycle [24]. The PLS-SEM method is a structural equation modeling approach that allows for the consideration of relationships between observed and latent variables, as well as causal relationships between latent variables [25,26], and is an important method for identifying the impacts of climate change and human activities on runoff.

The Minjiang River is a first-class tributary of the Yangtze River and the main water system of the Chengdu Plain, sustaining the regional water supply and the health of the water environment. The Minjiang River Basin (Chengdu section) has been greatly affected by climate change, landscape changes, and urbanization in recent years, which has led to an imbalance between the supply and demand of water resources, as well as the unfair distribution and inefficient utilization of water resources. This has, in turn, decreased the security and restricted the sustainable development of the region. The main objectives of this study were (1) to reveal the spatial and temporal distribution characteristics of runoff in the watershed by simulating accurate surface runoff data with the SWAT model and (2) to reveal the main driving factors affecting runoff and the interrelationships among the factors using the PLS-SEM model. The results of the study will help to optimize water management strategies in the basin, improve the efficiency of water resource use, and protect the water environment.

2. Materials and Methods

2.1. Study Site

The Minjiang River Basin is a first-class tributary of the upper reaches of the Yangtze River. The territory of Chengdu City enters the basin from the Waijiang Gate of Dujiangyan City in the northwest and flows out from Dongheba of Dengshuang Township in Xinjin County in the southeast (Figure 1). This area of the basin has a total length of 81 km and a watershed area of 5462.69 km². The upstream to downstream of the Longxi River, Baisha River, West River, and South River are rich in water resources. It is at a latitude of 30°05′–31°15′ North and a longitude of 103°00′–104°00′ East, located in the western Sichuan Plateau to the Sichuan Basin Transitional Interchange Zone, with a subtropical humid monsoon climate [27]; the terrain is high in the west and low in the east, with medium-high mountainous areas in the west, and plains in the east. The elevation is between 341 and 4020 m, and with the increase in altitude the heat is reduced; the emergence of warm winter and cool west of the two climate types coexist in the pattern. The basin has four distinct seasons and abundant heat, with an annual average temperature of 16–17.5 °C, and an annual average active cumulative temperature of ≥10 °C of 4700–5300 °C; the summers are hot and rainy, with abundant rainfall, and the annual average precipitation is about 975.89–1124.5 mm, which mainly concentrates in June–September, accounting for about 60% of the total precipitation of the year [28,29]. Soil types in the watershed are diverse, mainly including purple soil, yellow-brown soil, and rice soil [30]. In terms of land use types, the main land use types in the west are forest land and grassland, while cultivated land dominates in the east.

2.2. Data Sources

The study data included land use, topographic, climatic, soil, hydrological, and socio-economic data (

Table 1. Data types and sources.

Data Types	Specific Figures	Data Sources	Data Description
Topographic data	Elevation maps, Slope maps, Topographic humidity index maps	Geospatial Data Cloud (http://www.gscloud.cn/, accessed on 7 July 2023)	30 m, slope and Topographic Wetness Index were extracted from elevation
Land use data	1990, 1995, 2005, 2015, and 2019	Pixel Information Expert Engine, accessed on 10 July 2023	30 m
Soil data	Soil organic carbon content, organic matter content, clay content, sand content, and silt content data	Harmonized World Soil Database, accessed on 15 July 2023	1000 m
Climate data	/	National Meteorological Data Centre (http://data.cma.cn/, accessed on 17 July 2023)	Daily
Hydrological data	/	Sichuan Hydrological Statistical Yearbook (accessed on 18 July 2023)	Monthly runoff
Socio-economic data	Gross Domestic Product (GDP), population density	Centre for Resource and Environmental Sciences and Data, Chinese Academy of Sciences (https://www.resdc.cn/, accessed on 20 July 2023)	1000 m, where the 1990 GDP is interpolated through the statistical yearbook

). Land use data were classified into cropland, forest, grassland, watershed, construction land, and bare land. The watershed soil database was constructed from soil data that used the FAO-90 classification standard in the World Soil Database (HWSD). Meteorological data included day-by-day rainfall, air temperature, evaporation, relative humidity, and wind speed, which were measured by the three meteorological stations in the study area from 1979 to 2019 and simulated using the weather generator simulation to construct the meteorological database. Hydrological data included the measured monthly runoff data from Chengdu Xinchang Station as the evaluation and validation data of the model.

2.3. Study Framework

In this study, the SWAT model was used to simulate and analyze the runoff data of the study area over the last 30 years to reveal its spatial and temporal distribution characteristics. Additionally, these natural, climatic, social, and landscape factors were selected, and the PLS-SEM model was used to quantitatively assess the relationship between each driver and runoff changes. The specific research framework is shown in Figure 2.

2.4. Study Methods

2.4.1. SWAT Model Construction and SUFI-2 Algorithm

The SWAT (Soil and Water Assessment Tool) model is a distributed hydrological model based on process and physical principles that is capable of simulating hydrological changes over a long time series in a watershed [31,32,33]. In this study, the ArcSWAT2012 extension module was used for the simulation of hydrological processes in the study basin [34]. Based on the hydrological and DEM data, the optimal threshold of the catchment area was determined to be 150 hm² after several experiments, which divided the study area into 89 sub-catchments. The study area was divided into 826 hydrological response units (HRUs) by setting the thresholds for land use area, soil area, and slope at 10%, 15%, and 10%, respectively.

The SUFI-2 uncertainty analysis algorithm in SWAT-CUP was used for model parameter estimation and uncertainty analysis. The algorithm uses the Nash efficiency coefficient (NES) and the coefficient of determination (R²) to evaluate the fit of the model [35,36]; that is, the model simulation results are better when R² ≥ 0.6 and NSE > 0.5.

In this study, the model rate period was set as 1978–1981, of which 1978 was the warm-up period to reduce the effect of the initial conditions on model accuracy. The model validation period was 1982–1985, of which 1982 was the warm-up period. The SWAT-CUP software was chosen to evaluate and validate the model with six runs of 500 iterations each to obtain the best fit.

2.4.2. Mann–Kendall Test and Theil–Sen Slope

Both parametric and non-parametric trend tests can be used to test for significant trends in time series. The parametric trend test requires that the data are independent and follow a normal distribution, while the non-parametric trend test only requires that the data are independent [37]. The Theil–Sen median trend analysis was combined with the Mann–Kendall test to detect trends in long time-series data [38]. The Mann–Kendall test statistic S is calculated as follows:

$$S=\sum _{i=1}^{n-1}\sum _{j=i+1}^{n}sgn (x_j-x_i)$$

(1)

Note that if S > 0, later observations in the time series tend to be larger than those that occur earlier in the time series, while the opposite is true if S < 0.

sgn (x_i − x_j) is the sign function as follows:

$$sgn(x_j-x_i)=\left\{\begin{array}{c}1 (x_j-x_i)>0\\ 0 (x_j-x_i)=0\\ -1 (x_j-x_i)<0\end{array}\right.$$

(2)

The variance of S is expressed as:

$$Var(S)={\displaystyle \frac{n(n-1)(2n+5)}{18}}$$

(3)

where n is the number of samples. When n > 10, the standard normal statistical variable Z is expressed as follows:

$$\mathrm{Z}=\left\{\begin{array}{c}{\displaystyle \frac{\mathrm{S}-1}{\sqrt{\mathrm{V}\mathrm{a}\mathrm{r}\left(\mathrm{S}\right)}}} S>0\\ 0 S=0\\ {\displaystyle \frac{\mathrm{S}+1}{\sqrt{\mathrm{V}\mathrm{a}\mathrm{r}\left(\mathrm{S}\right)}}} S<0\end{array}\right.$$

(4)

where for a given significance level α, if $\left|Z\right|\ge Z_{1-\alpha /2}$, then the original hypothesis is unacceptable, i.e., there is a significant upward or downward trend in the time series data at the significance test level. A positive value of Z indicates an upward trend, a negative value indicates a downward trend, and the absolute value of Z is greater than or equal to 1.65, 1.96, and 2.58, which means that it passes the significance test with confidence level of 90%, 95%, and 99%, respectively.

The Theil–Sen slope is calculated as follows:

$$slope=\mathrm{M}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{n}\left({\displaystyle \frac{X_j-X_i}{j-i}}\right){\forall }_j>i$$

(5)

Here, 1 < j < i < n; when slope > 0, the time-series data shows an upward trend; when slope < 0, the time-series data show a downward trend.

In this study, linear trend regression analyses were performed on the average runoff volume from 1990–2019, as well as on precipitation, air temperature, and evaporation. The Theil–Sen median trend analysis and Mann–Kendall test were used to analyze the mean annual runoff in 89 sub-basins, and spatial interpolation was used to obtain the trend in the mean annual runoff at the image metric scale for the whole region from 1990 to 2019. According to the calculation of the Theil–Sen median trend analysis slope value, the trend was divided into two categories: upward trend (slope > 0) and downward trend (slope < 0). The confidence level of the significance test in the Mann–Kendall test was selected to be 95%, and the result was finally classified into significant change (|Z| > 1.96) and non-significant change (|Z| < 1.96).

2.4.3. Relative Importance Analysis

Relative importance analysis is a method used to assess the extent to which multiple explanatory variables influence the response variable, helping us to understand the contribution of each variable in the model. In order to determine the relative contribution of drivers such as topography and geomorphology, soil, climate, economy, and landscape to changes in surface runoff, we used a generalized linear model to fit surface runoff and the associated drivers and chose “lmg” to quantify the relative contribution of each driver to surface runoff in order to quantify the relative importance of these drivers [39].

Among the landscape factors, two types, connectivity indicators and fragmentation, were selected to reflect the changes in landscape structure, of which the connectivity indicators include AREA_MN (Patch Area_Mean), COHESION (Patch Cohesion Index), PLADJ (Percentage of Landscape), AI (Aggregation Index), and LPI (Largest Patch Index), and the fragmentation indicators include DIVISION (Landscape Division Index), SHDI (Shannon’s Diversity Index), and SPLIT (Splitting Index). Fragstats 4.2 software was used to calculate the values of each landscape indicator (https://fragstats.org/index.php/documentation, accessed on 26 June 2024), and then the relative importance analysis was carried out using the “relaimpo” package in R language (https://cran.r-project.org/web/packages/relaimpo/index.html, accessed on 26 June 2024).

2.4.4. The Partial Least Squares Structural Equation Modeling (PLS-SEM) Construction

The PLS-SEM model uses partial least squares regression to estimate the parameters and is not constrained by data normality, sample size, model complexity, or the type of measurement. Variables in the PLS-SEM model were categorized into latent and observed variables. Variables that cannot be directly measured are called latent variables, and observed variables are variables that can be directly measured. The model consists of two parts: the measurement model and the structural model, where the measurement model describes the relationship between latent and observed variables. The structural model describes the relationship between the latent variables. To ensure the reliability and validity of the model, a series of evaluation metrics is usually adopted to assess the reliability, validity, and predictive ability of the PLS-SEM model. The unidimensionality of the observed variables was evaluated using Dillon–Goldstein’s rho, with values > 0.7 indicating that the unidimensionality test was passed. The internal consistency reliability was assessed using composite reliability (CR), which usually has a value > 0.7. The average variance extracted (AVE) was used to verify convergent validity and was required to be greater than 0.5. The loading represented the correlation between the latent variable and its observed variable, with values > 0.6 considered to be acceptable. The overall fit integrates the fit of the measurement model and the structure, and greater than 0.5 indicates that the model is fitted with high precision [40,41,42].

In this study, the PLS-SEM model was used to analyze the complex relationships between the drivers and to evaluate the intrinsic factors affecting changes in surface runoff. The establishment of the PLS-SEM model consisted of the following steps: (1) identifying obvious and potential variables, making assumptions about the relationships between variables based on a priori knowledge or literature inquiries, and constructing a conceptual model based on these assumptions; (2) converting the conceptual model into a mathematical model using the PLS-SEM model; and (3) fitting the model, checking the effect of the fit, and—if the model was not well fitted—eliminating variables based on the results and re-correcting the model until the model results reached the optimal level. In this study, natural factors, climatic factors, economic factors, and landscape factors were selected as four latent variables in the model, including seventeen apparent variable factors. The description of each variable is shown in

Table 2. Description of latent variables driving runoff variability.

Latent Variables	Apparent Variables	Explanation
Natural factors	Relative elevation	T = Tmax − Tmin (Difference between the maximum and minimum elevation values within the hydrological response cell), Unit: m
	Relative slope	S = Smax − Smin(Difference between the maximum and minimum slope values within the hydrological response cell), Unit:°
	Topographic Wetness index	TWI = ln(α/tanβ) Where: α represents the unit grid catchment area; β represents the slope
	Soil pH	Soil potential of hydrogen
	Clay contents	Percentage of clay in soil, Unit: %
	Sand contents	Percentage of sand in soil, Unit: %
	Silt contents	Percentage of silt in soil, Unit: %
	Soil organic matter	Carbon-containing organic compounds in soil, Unit: %
	Soil organic carbon	Humus formed through microbial action, Unit: %
Climatic factors	Average annual precipitation	Multi-year average precipitation, 1 km resolution, Unit: mm
	Average annual temperature	Multi-year average temperature, 1 km resolution, Unit: °C
	Average annual evaporation	Multi-year average evaporation, 1 km resolution, Unit: mm
Socio-economic factors	Gross Domestic Product (GDP)	Based on spatialisation of national sub-county GDP statistics, Unit: millions of yuan.
	Population density	Based on spatialisation of national sub-county demographics, Unit: persons.
Landscape factors	Landscape change	Raster values with no landscape change were recorded as 0 and those with landscape change were recorded as 1. The number of rasters with landscape change was counted as a proportion of each hydrological response cell.
	Landscape connectivity	Landscape connectivity indicators were constructed based on the relative importance of each of these indicators: Connectivity = α1AREA_MN + α2COHESION + α3PLADJ + α4AI + α5LPI Where: Connectivity represents landscape connectivity, and α1 − α5 represent the relative importance share of each metric, respectively.
	Landscape fragmentation	A landscape fragmentation index was constructed based on the relative importance of each of the above indicators: Fragmentation = β1DIVISION + β2SHDI + β3SPLIT Where: Fragmentation represents the degree of landscape fragmentation, and β1-β3 represent the relative importance of each indicator.

.

The following hypotheses were formulated, considering the characteristics of the region’s natural and social environment: 1. Surface runoff is directly or indirectly influenced by several natural, climatic, economic, and landscape factors. 2. Natural conditions determine the level of the economy, which in turn has a feedback effect on climate change and thus influences changes in surface runoff. 3. Topographic and climatic factors are the main factors influencing the level of economy, and they further influence the amount of surface runoff through the mediating effect of the level of economy. 4. The landscape factor is a combination of natural, climatic, and economic factors, which reflects the degree of human use and modification of the natural environment and also impacts surface runoff. To test these hypotheses, a conceptual model was constructed and the relationship between the factors was graphically presented, as shown in the model framework in Figure 3.

In this study, each hydrological response unit was taken as a sample set and its sample value was taken as the raster mean value of each hydrological response unit. Invalid or missing samples were excluded, and a total of 349 valid samples were obtained. Based on these samples, a PLS-SEM model of the Minjiang River Basin (Chengdu section) was constructed using the ‘plspm’ package in R language.

3. Results

3.1. The Construction of the SWAT Model

In this study, a t-test was used to determine the sensitivity of each parameter, 23 parameters related to runoff were selected, and the results of the sensitivity analysis are shown in

Table 3. Parameter sensitivity analysis results.

Parameters	t-Stat	p-Value	Parameter Optimum	Sensitivity	Connotation
v__SFTMP.bsn	0.04	0.97	4.16	NO	Snowfall temperature/°C
v__SMTMP.bsn	−0.44	0.66	−7.72	NO	Base temperature for snowmelt/°C
v__SMFMX.bsn	0.17	0.87	−1.62	NO	Snowmelt factor on 21 June (mm/(°C·day))
v__TIMP.bsn	0.25	0.80	0.23	NO	Snow temperature lag factor
v__SURLAG.bsn	−1.30	0.20	20.16	YES	Surface runoff hysteresis factor
v__TLAPS.sub	0.58	0.57	10.74	NO	direct temperature decrease (°C/km)
v__SLSUBBSN.hru	−2.76	0.01	43.67	YES	Average slope length (m)
v__HRU_SLP.hru	−0.24	0.81	1.49	NO	Average specific gravity (m/m)
v__CANMX.hru	−0.65	0.52	73.21	NO	Maximum canopy retention (mm)
v__ESCO.hru	−0.40	0.69	0.43	NO	Soil evaporation compensation factor
v__EPCO.hru	1.16	0.24	1.14	YES	Plant uptake compensation factor
v__OV_N.hru	1.41	0.16	0.12	YES	Values of Manning’s coefficient n for diffuse flow on slopes
r__CN2.mgt	5.95	0.00	0.45	YES	Initial SCS runoff curve number for moisture condition II
r__BIOMIX.mgt	1.20	0.23	−0.80	YES	Biomixing efficiency
v__CH_COV1.rte	−0.11	0.91	0.32	NO	Channel erosion factor
v__CH_COV2.rte	0.43	0.67	0.99	NO	Stream cover factor
v__GW_DELAY.gw	−1.42	0.16	−428.75	YES	Time delay in groundwater (day)
v__ALPHA_BF.gw	0.64	0.52	0.66	NO	Baseflow α factor (day)
v__GWQMN.gw	7.94	0.00	2260.49	YES	Water level threshold of the shallow aquifer required for return flow to occur (mm)
v__GW_REVAP.gw	2.12	0.03	0.13	YES	Groundwater revap coefficients
v__REVAPMN.gw	0.80	0.42	522.34	NO	Water level threshold of the shallow aquifer required for revap or infiltration into the deep aquifer to occur (mm)
v__RCHRG_DP.gw	6058	0.00	0.23	YES	Permeability coefficient of deep aquifers
a__SOL_AWC.sol	−1.33	0.18	0.11	YES	Effective water content of the soil layer (mm/mm)

Note: ① v_: replace; r_: multiply by (1 + x); a_:add; ② t-value indicates the size of the sensitivity of the parameter: the larger the t-value, the more sensitive the parameter; ③ p-value indicates the significance of the t-value: the smaller the p-value, the more significant the parameter on behalf of the sensitivity.

. The results show that different parameters have different sensitivities to runoff, with the top three sensitivities being GWQMN, RCHRG_DP, and CN2, sequentially. The parameters SURLAG, SLSUBBSN, OV_N, BIOMIX, GW_DELAY, and SOL_AWC were also more sensitive to runoff, while the other parameters were not.

The rate and validation of the month-by-month runoff from the Xinchang hydrological station were investigated according to the 11 parameters with the highest sensitivity, including GWQMN, RCHRG_DP, CN2, etc. The results of the fitting results show that (Figure 4) the NSE in the validation period was 0.71 and the R2 was 0.74; the rating period was 0.83 and the R2 was 0.84. This indicates that the SWAT model effectively simulated the monthly runoff of the Minjiang River Basin, Chengdu.

3.2. Runoff Simulation Analysis

3.2.1. Inter-Annual Variability of Runoff

The average runoff in the Minjiang River Basin (Chengdu section) from 1990 to 2019 was 628.96–1088.46 mm, with a multi-year average of 834.13 mm. The inter-annual fluctuation of runoff is obvious, in which the runoff showed a non-significant decreasing trend with a rate of change of −8.959 mm·a⁻¹ from 1990 to 1999, and a significant increasing trend with a rate of change of 23.921 mm·a⁻¹ and 11.135 mm·a⁻¹ from 2000 to 2009 and from 2010 to 2019. The mean annual temperature and mean annual evaporation in the watershed showed a significant upward trend, with mean linear rates of increase of 0.026°C·a⁻¹ and 1.141 mm·a⁻¹, respectively; the trend of mean annual precipitation was not significant, with a mean linear rate of increase of 2.172 mm·a⁻¹. Meanwhile, the trends of runoff volume changes in different seasons were not different, in which the abundant and dry water periods showed an increasing trend, with an average linear growth rate of 0.804 mm·a⁻¹ and 0.539 mm·a⁻¹, respectively; the flat water period showed a decreasing trend, with an average linear decreasing rate of 0.126 mm·a⁻¹ (Figure 5).

3.2.2. Characteristics of the Spatial Distribution of Runoff

The results showed that the spatial distribution of runoff in the Minjiang River Basin (Chengdu section) is differentiated (Figure 6), There was a general trend of “high–low–high” in the northwest–southeast direction. Sub-catchments 2, 18, and 74 are located close to the river and have high runoff; in contrast, sub-catchments 24, 32, and 50 are in the gently sloping area of the agricultural land, where agricultural irrigation is the main mode of water use, and their runoff is relatively low.

In terms of runoff trends (Figure 7), runoff in the southern and northern parts of the basin showed an increasing trend, and runoff in the central part of the basin showed a decreasing trend, but none of the changes were significant. As can be seen in

Table 4. Minjiang River Basin (Chengdu section) runoff trend statistics.

Slope	Z_Value	Trends in Runoff	Area/km2	Proportion of Study Area/%
Slope > 0	|Z| > 1.96	Significant increase	52.25	0.96
Slope > 0	|Z| < 1.96	No significant increase	3440.44	63.01
Slope < 0	|Z| < 1.96	No significant decrease	1917.2	35.11
Slope < 0	|Z| > 1.96	Significant decrease	50.02	0.92

, the largest proportion of area in the basin is occupied by areas of non-significant increase, followed by areas of non-significant decrease, and a smaller proportion of area is occupied by areas of significant increase and significant decrease.

3.3. Relative Importance of Drivers

An analysis of the relative importance of the driving factors shows (Figure 8) that each factor contributes to surface runoff to varying degrees. Among the topographic factors, relative elevation is the most important factor, with a relative importance of 44.92%. The relative importance of the soil factors, chalk content, gravel content, and clay content were 31.75%, 31.47%, and 20.78%, respectively. Among the climatic factors, mean annual temperature has the greatest influence on surface runoff changes, with a high relative importance of 69.85%; among the socio-economic factors, population density and GDP contribute more, with a relative importance of 52.87% and 47.13%, respectively; and among the landscape factors, AI and SHDI are the most important indicators, with a relative importance of 33.57% and 90.08%, respectively.

3.4. The PLS-SEM Model Assessment

3.4.1. Assessment of Measurement Models

The PLS-SEM model needed to be evaluated for both the measurement model and the structural model. Measurement models provide relationships between latent variables and the indicators they define. To ensure the accuracy of the model results, the measurement model needed to be assessed using a range of parameters: the unidimensionality test of indicators, and the confidence and validity of the model. The variance inflation factor (VIF) was used to detect multicollinearity among the apparent variables. In the present study, VIF < 10 was used to indicate that there was no serious covariance among the variables and, therefore, VIF > 10 indicated serious multicollinearity among the variables. The results of the measurement model evaluation are shown in

Table 5. PLS-SEM measurement model assessment results.

Latent Variable	AVE	CR	DG.rho	Loading	t-Test	Observed Variable	Test Results
Natural factors	0.619	0.882	0.887	0.859	2.604	Relative elevation	accept
				0.927	2.565	Relative slope	accept
				0.863	2.546	Topographic humidity index	accept
				0.148	1.329	Soil PH	reject
				0.045	0.434	Clay contents	reject
				0.695	2.498	Sand contents	accept
				0.703	2.478	Silt contents	accept
				0.369	2.060	Organic matter contents	reject
				0.029	0.435	Organic carbon contents	reject
Climatic factors	0.618	0.841	0.813	0.855	2.214	Average annual precipitation	accept
				0.644	2.357	Average annual temperature	accept
				0.960	2.316	Average annual evaporation	accept
Socio-economic factors	0.696	0.976	0.984	0.982	91.722	GDP	accept
				0.985	135.008	Population density	accept
Landscape factors	0.968	0.671	0.827	0.912	20.478	Landscape change	accept
				0.748	6.471	Landscape connectivity	accept
				0.277	2.392	Landscape fragmentation	reject
Runoff	1.000	1.000	1.000	1.000		Runoff	accept

. Five observed variables—pH, organic matter content in the soil, organic carbon content, clay content, and landscape fragmentation—were excluded and significance tests were passed between the remaining latent and observed variables.

3.4.2. Assessment of Structural Models

The interactions between latent variables were analyzed using a structural model, and the degree of association between them was indicated by path coefficients. Ten direct pathways, such as nature → climate and climate → surface runoff, were defined in the initial stage of modeling (

Table 6. Results of hypothetical path test for the PLS-SEM model.

Direct Paths	Path Coefficient	T-Value	p-Value	Test Results
Socio-economic—>Runoff	−0.210	4.251	0.000	significant
Socio-economic—>Landscape	0.446	8.223	0.000	significant
Socio-economic—>Climate	0.219	2.018	0.044	significant
Landscape—>Runoff	−0.131	2.164	0.030	significant
Climate—>Runoff	−0.102	1.266	0.205	insignificant
Climate—>Landscape	0.004	0.042	0.966	insignificant
Nature—>Socio-economic	−0.268	1.734	0.083	insignificant
Nature—>Runoff	0.367	2.901	0.037	significant
Nature—>Landscape	0.075	0.879	0.879	insignificant
Nature—>Climate	−0.620	2.200	0.028	significant

). According to the results of the model (Figure 8), economic and landscape factors had significant negative effects on surface runoff changes, with path coefficients of −0.210 (p = 0.000) and −0.131 (p = 0.03), respectively. Natural factors had positive effects on surface runoff changes, with a path coefficient of 0.367 (p = 0.037). Although climatic factors had a negative effect on surface runoff changes, their path coefficient was only −0.102 (p = 0.205), which was not significant.

In addition to the direct pathways, several indirect pathways were identified; these reflect the transmission of impacts between different factors. Economic factors can have an indirect negative effect on runoff changes by changing landscape distribution, while natural factors could have an indirect negative effect on surface runoff by influencing economic activities (Figure 9).

4. Discussion

Surface runoff is an important part of the hydrological cycle, and it is important to analyze the change in surface runoff and its main driving factors to elucidate the hydrological response mechanism. Compared with previous studies, which mostly focused on global [43], China [44,45], or large catchment scales [46,47,48,49], this study focuses on the catchment intervals with greater intensity of human activities. By simulating the dynamic change of surface runoff, factors such as surface, climate, human activities, and landscape are taken into account, and quantitative methods are adopted. The complex causal relationship between surface runoff change and its driving factors was discussed, and some indirect ways of influencing surface runoff change were revealed to provide a new perspective for understanding and predicting surface runoff change.

4.1. Spatial and Temporal Variability of Runoff

The average annual runoff in the Minjiang River Basin (Chengdu section) varied from 628.96 to 1088.46 mm, with a multi-year average of 834.13 mm. This was essentially consistent with the results of Zhang [50] and Zhai [51]. The ‘high–low–high’ pattern of runoff from the northwest to the southeast of the basin was mainly influenced by the following factors: the upstream confluence area being a mountainous area, with a large channel slope drop and a high proportion of surface runoff, leading to an increase in runoff; the midstream area having a small channel slope drop and a low proportion of surface runoff, resulting in a decrease in runoff; and the confluence of several tributaries, such as the Xijiang River and the Linxi River, leading to a further increase in runoff [52]. Simultaneously, subject to the combined constraints of natural and social factors, the areas usually distributed along the river channels were doubly recharged by incoming water and precipitation and their runoff was larger, while the areas with flat topography and well-developed agriculture were depleted by factors such as evaporation and artificial water abstraction, which led to less runoff.

4.2. Analysis of the Drivers of Runoff Change

4.2.1. Significant Positive Impact of Natural Factors

The Minjiang River Basin (Chengdu section) has a wide range of elevation changes (341–4020 m) and a significant difference in topography between the east and west, leading to a significant positive relationship between surface factors and surface runoff. The higher the relative elevation and slope, and the more undulating the terrain, the more likely it was for precipitation and flowing water to form runoff rather than infiltrate the soil [53,54]. The topographic moisture index reflects the degree of wetness and runoff potential of a surface, with a larger index indicating that the point is more likely to accumulate moisture and precipitation is more likely to be converted to runoff than evaporated [55]. In the study area, anthropogenic soils are present in the highest proportion, especially at low altitudes where the scale of agricultural land is large and long-term operations such as ploughing and irrigation have impacted the physical properties of the soils and led to loose, sparse and chalky soils. This in turn has reduced soil porosity, pore connectivity and—consequently—has led to decreased water containment capacity. Therefore, more precipitation results in surface runoff.

4.2.2. Significant Negative Impacts of Human Activities

This study showed that population density and GDP were important drivers of changes in surface runoff. The resident population of Chengdu City increased by 3.5 times between 1990 and 2019, the population density increased from 42.5 to 147.9 people km⁻² and the GDP also increased 264-fold. These changes have put enormous pressure on the Minjiang River Basin, which Chengdu relies on as one of the city’s main water sources and an important source of water for agriculture. With population growth and the expansion of agricultural land, the Minjiang River Basin is facing water shortages and pollution problems. Simultaneously, the implementation of a series of water conservancy projects, such as dams, river embankments and reservoirs, has altered the natural water cycle of the basin and affected its hydrological situation.

4.2.3. Significant Negative Effects of Landscape Factors

Landscape factors directly affect the generation of and changes in surface runoff [56]. Landscape types can be classified into ‘source’ and ‘sink’ landscapes according to their contribution to and consumption of surface runoff. Cropland, woodland and grassland are usually considered sink landscapes and impervious surfaces as source landscapes. The area of source and sink landscapes in this watershed in 1990 was 4881.77 and 477.98 km², and in 2019 was 4520.95 and 828.91 km², respectively. This was indicative of the significant urbanisation of the region and an increase in the area of impervious surfaces, which reduced the infiltration capacity of rainwater and increased the runoff coefficient. Additionally, landscape connectivity affects the flow and distribution of water between patches, thereby influencing surface runoff volume and peaks [57,58]. In the Minjiang River Basin in Chengdu, large areas of the original continuous landscape have been cut up in the last 30 years due to the construction of infrastructure (roads, industrial and mining, irrigation canals, etc.) and land use changes; this has resulted in increased landscape fragmentation and homogeneity, and decreased landscape diversity and connectivity. These effects also have an impact on surface runoff.

4.2.4. Non-Significant Effects of Climatic Factors

In the quantitative analysis of the effect of climate factors on surface runoff, we found that the path coefficient of the climate factor was −0.102, but it did not pass the significance test, i.e., the effect of climate on surface runoff was not significant. Although climatic factors are widely recognized as one of the key factors driving changes in surface runoff in the theoretical framework and are expected to have a significant impact on the study area [59], the specific findings of this study reveal the limitations of climatic impacts in the watershed. This non-significant result may be attributed to a combination of several complex factors. First, the relatively small geographic scale of the study watershed may limit the ability of climatic factors to contribute significantly to surface runoff on a larger spatial scale. Second, the climatic factors themselves exhibited a high degree of heterogeneity in their spatial distribution, i.e., the climatic characteristics of different geographic locations varied significantly, and this heterogeneity may have weakened the overall influence of climatic factors on surface runoff in a given watershed. Furthermore, the intensity of changes in climatic factors during the study period was relatively small and failed to trigger the threshold for significant changes in surface runoff. It is worth noting that although the current study concludes that the climate effect is not significant, this does not mean that the relationship between climate factors and surface runoff can be completely ignored. On the contrary, it suggests that we need to further explore the specific mechanisms of climate factors at different time scales, spatial scales, and intensities of change in future studies. However, there are many challenges in acquiring and analyzing climate research data. On the one hand, long time series of climate data are crucial for revealing the long-term trend and cyclical pattern of climate change, but the existing data are often limited by the lack of spatial and temporal resolution, making it difficult to accurately characterize the climate at the regional scale. On the other hand, many river basins did not have a well-established meteorological observation system at an early stage, resulting in a serious lack of historical data and impaired data continuity, which greatly limits the development of climate change research on a continuous time scale.

5. Conclusions

In this study, the Minjiang River Basin (Chengdu section) was taken as a case study. The monthly-scale runoff process from 1990 to 2019 was simulated using the SWAT model and the PLS-SEM model was used to explore the impacts of nature, climate, landscape, and society on surface runoff changes, as well as the interactions among the factors. The results showed that the average runoff in the Minjiang River Basin in Chengdu was 628.96–1088.46 mm, with an average value of 834.13 mm. The runoff showed a non-significant upward trend, with an average linear growth rate of 1.247 mm·a⁻¹ and an overall spatial distribution pattern of ‘high–low–high’ from the northwest to the southeast. The main factors affecting the changes in surface runoff were relative elevation, powder content in soil, mean annual air temperature, population density, and SHDI. The PLS-SEM model analysis showed that socio-economic and landscape factors had significant negative effects on surface runoff, surface factors had significant positive effects on surface runoff, and climatic factors did not have significant effects on surface runoff. Additionally, economic factors had an indirect negative effect on runoff by changing the landscape distribution, and natural factors had an indirect negative effect on runoff by influencing economic activities. Runoff changes in the Minjiang River Basin (Chengdu section) are the result of a combination of natural and socio-economic factors. Therefore, it is necessary to comprehensively consider the influence of various factors and rationally optimize water resource management measures to achieve the sustainable use of water resources.