Hydrological response and soil water soluble antimony and arsenic transport under climate change using SWAT modeling

Abstract

Climate change affects hydrological processes globally, impacting the water-soluble forms of metals such as antimony (Sb) and arsenic (As) in various soils. This study developed an enhanced SWAT-based framework incorporating soil experiment data and coupled it with SSP scenarios (SSP126, SSP245, SSP370, SSP585) to project future hydrological processes, and then assessed water-soluble proportions of Sb and As distribution under climate change based on cluster analysis results of soil, precipitation, and streamflow classification. The model, incorporating measured values, accurately simulated the streamflow and sediment loads of the study area during the historical period. Under the climate scenarios, the study area exhibited trends of fluctuating precipitation and rising temperatures; compared to the reference period (2008–2014), the total streamflow generally increased across all scenarios. Experimental results revealed that Sb and As concentrations exceeded standards, with pH and soil organic matter showing significant correlations with the water-soluble proportions of these metals. Analyzing the combined results from soil, precipitation, and streamflow cluster analyses indicated that a higher percentage of water-soluble Sb and a higher percentage of water-soluble As are likely to be distributed in the subbasins along the main river in the east, potentially increasing over time. These results offer a scientific foundation for water resource management and pollutant mitigation in mining watershed areas.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-25715-9.

Introduction

Mining activities increase concentrations of heavy metals or metalloids in soil^1,2. These metals primarily migrate in dissolved and adsorbed forms during their surface movement³. The water-soluble state of metals, transported via streamflow, has the potential to pollute large areas and pose risks to human health⁴. The Intergovernmental Panel on Climate Change (IPCC) report states that the average global surface temperature between 2001 and 2020 was 0.84–1.10 °C higher compared to the period from 1850 to 1900⁵. The ongoing climate change is anticipated to strengthen the hydrological cycle, influencing precipitation, evapotranspiration, and streamflow, and thereby altering the water balance in river basins^6,7. This will further affect the concentration and migration of water-soluble metals in mining area⁸. Examining the hydrological response to climate change at the basin level and analyzing the distribution of water-soluble metals under different climate scenarios are crucial for effective pollutant management and protection in river basins⁹.

The Duliujiang basin in Southwest China is known for its abundant antimony (Sb) resources, with mining activities leading to severe metalloid pollution¹⁰. Previous studies indicate that Sb and arsenic (As) are the primary metalloid pollutants in this watershed, posing potential risks to human health and the environment^11,12. The environmental risk associated with these metals is strongly linked to their mobility in the soil. Sb and As, being metalloids from the same group in the periodic table, often coexist in natural environments and exhibit similar geochemical behaviors due to their identical electronic configurations¹³. The water-soluble fraction of Sb and As both in soil, though a small part of the total metal content, is considered bioavailable and significantly impacts ecological restoration and human health^11,14. Research shows that this fraction is influenced by soil physicochemical properties such as pH values^15–17 and organic matter content¹⁸. Therefore, assessing the proportion of water-soluble Sb and As in the soil requires analyzing the soil’s physicochemical properties. To evaluate the proportion and distribution of water-soluble Sb and As in the soil of the mining watershed under climate change, it is essential to consider the meteorological and hydrological conditions of the watersheds, as well as the influence of varying soil characteristics on the water-soluble fraction of metals. This will help construct a framework for studying the mobility of metals in the watershed under climate change.

Simulating watershed hydrological processes quantifies the dynamic shifts in precipitation and streamflow compared to a baseline year, considering climate change impacts^19–21. The Soil and Water Assessment Tool (SWAT), a distributed hydrological model, is commonly used for this purpose. It integrates meteorological and surface elements²² and can be calibrated and validated to adjust model outputs to closely match observed results in the study area²³. The SWAT model has been applied in numerous watersheds^24,25. To examine the effects of future climate change on hydrological processes, data from General Circulation Models (GCMs) are incorporated into the SWAT model²⁶. GCMs, widely used to assess the effects of future climate impacts, provide insights into past climate trends and future scenarios based on certain emission assumptions^27,28, serving as input data for SWAT^29,30. This allows further analysis of the impacts on other environmental aspects in the study area³¹, such as the proportion of water-soluble metals, enhancing watershed assessment and management^32–34.

To address the future distribution of the water-soluble fraction of metals such as Sb and As in mining area watersheds under climate change, this study utilizes CMIP6 data and the SWAT model to simulate and predict streamflow in the study area, analyzing its spatiotemporal variation characteristics. Based on these simulations, the relationship between water-soluble Sb, As, and soil physicochemical properties will be analyzed using experiment results. The objectives of this study are to: (1) Improve the soil database of the study area based on soil sampling experiment results, thereby enhancing the SWAT simulation outcomes. Subsequently, calibrate and validate the improved SWAT model using historical data.; (2) Use the SWAT model to simulate the hydrological processes in the study area under four Shared Socioeconomic Pathways (SSP) scenarios (SSP126, SSP245, SSP370, and SSP585). Analyze the hydrological process responses and their spatiotemporal variability in the near-term (2020–2040), medium-term (2041–2070), and long-term (2071–2100); (3) Analyze the distribution of water-soluble fraction of Sb and As in each subbasin of the study area over a long-time scale based on model simulations and experiment results. Offer scientific support for the management of metalloids in mining areas under climate change conditions.

Materials and methods

Study area

The research region encompasses the middle and lower reaches of the Duliujiang Basin, situated in the southwestern region of China. The Duliujiang River is a significant tributary of the Pearl River system. The river channel in the study area is wide and turbulent, with a shallow riverbed composed of large boulders, providing a natural environment for studying the transport of water-soluble metals such as Sb and As. The Shihuichang Station and Yongwei Station are representative hydrological stations within the basin, located at 108.54°E, 25.89°N and 109.29°E, 25.70°N, respectively. The annual average temperature of the basin ranges from 18 to 20 °C, with annual precipitation between 1,400 and 1,800 mm, predominantly occurring from April to August. The high temperature and precipitation facilitate the growth of subtropical and tropical crops. The terrain is highly undulating, with slopes ranging from 0 to 68.7° and elevations between 203 and 3,025 m, with higher elevations in the northeast and southwest and lower elevations in the central part. The study area is located in the southern non-ferrous metal belt of China, encompassing Sandu, Rongjiang, and Congjiang, primarily consisting of clastic rocks and flowing through the Shihuichang Station and Yongwei Station. According to historical data from Yongwei Station, the sediment content in the area is low, with an average annual sediment content of about 0.4 kg m^− 3. Soil samples were collected along the main river channel of the basin, as shown in Fig. 1.

Fig. 1.

(a) Geographic location of the study watershed, with black lines indicating the watershed boundaries, blue lines representing the streamflow, red pins marking the sampling points, green stars indicating the hydrological stations, and green triangles for the meteorological stations; (b) Digital elevation model (DEM) of the watershed, with colors representing different geographic elevations; (c) Land use types in the watershed, with colors representing different types of land use; and (d) Soil types in the watershed, with colors representing different soil types. Abbreviations for land use and soil types are provided in Table S1 in the Supporting Information. Map created using ArcGIS 10.0 (https://www.esri.com).

SWAT inputs and uncertainty analysis

In this study, we used the SWAT model, developed by the United States Department of Agriculture and Texas A&M University in the mid-1990s^35,36. The watershed was divided into 307 subbasins. Based on 15 land use types, 15 soil types, and 2 slope classes (Table 1), the 307 subbasins provide a total of 1,405 Hydrologic Response Units (HRUs).

Table 1.

List of input and validation data, their resolution, and sources.

Data	Resolution	Source
Digital elevation model	30 m × 30 m	http://www.gscloud.cn
Land use map	1 km × 1 km	Data Center for Resources and Environmental Sciences
Soil map	1 km × 1 km	Data Center for Resources and Environmental Sciences
Streamflow (2010–2017)	Daily	Hydrological Yearbook of the Pearl River Basin
Sediment load (2010–2017)	Daily	Hydrological Yearbook of the Pearl River Basin
Current climate data	Daily	China Meteorological Assimilation Driving Datasets (CMADS) (http://www.cmads.org/)
Climate scenarios SSP (126, 245, 370, 585)	Daily	Coupled Model Intercomparison Project Phase 6 (CMIP6)

To calibrate and validate hydrological models, we utilized monthly flow data from the Shihuichang and Yongwei hydrological stations in the Duliujiang Basin over a continuous nine-year period from 2009 to 2017, sourced from the Hydrological Yearbook (Table 1). During the parameter calibration step, the widely used SUFI-2 (Sequential Uncertainty Fitting Version 2) algorithm^37,38 in SWAT-CUP was selected for iterative calibration, validation, and sensitivity analysis, upon which parameter adjustments were based. To assess the accuracy of the model outputs, this study employed three inherent statistical metrics: the coefficient of determination (R²), Nash-Sutcliffe efficiency (NSE)³⁹, and observation standard deviation ratio (RSR)²³. The calculation methods for these evaluation metrics are as follows:

1

2

3

(i = 1, 2, ⋯n) is the observed daily streamflow or sediment loads, (i = 1, 2, .n) is the simulated daily streamflow or sediment loads, is the mean of the observed daily streamflow or sediment loads, is the mean of the simulated daily streamflow or sediment loads. According to the classification criteria established by Moriasi et al.⁴⁰ and Santhi et al.⁴¹, evaluated on a monthly scale, streamflow and sediments simulations should satisfy: NSE > 0.5, R² > 0.65, and RSR ≤ 0.7, which are followed in this study.

Sample collection and analysis

After collecting soil samples, they are air-dried, ground, and sieved through a 0.149 mm polyethylene sieve. Following acid digestion, the concentration of Sb and As in the soil are quantitatively analyzed using plasma mass spectrometry/spectroscopy, with detection limits meeting the requirements of the “GB15618-1995/2008 Environmental Quality Standards for Soils”⁴². Further measurements of the water-soluble concentrations of Sb and As are conducted by placing 1.0 g of soil and 10 mL of water into a centrifuge tube, which is then shaken at room temperature for 2 h using an oscillator. The sample is subsequently centrifuged at 3000 rpm for 30 min, after which the filtered supernatant is collected and analyzed to determine the concentrations of water-soluble Sb and As. The physicochemical properties of the soil analyzed in this study include pH, soil organic matter (SOM), and soil particle size. To measure the soil particle size, 0.5 g of soil sample is placed in a 500 mL beaker, and within 24 h, 10 mL of 30% H₂O₂ is added to remove organic matter. This is followed by the addition of 5 mL of 10% HCl for 10 min to eliminate CaCO₃. Residual H₂O₂ and HCl are washed away with ultrapure water. The beaker is then filled with ultrapure water and left to stand for 12 h. The supernatant is removed, and 10 mL of 0.05 mol L^{− 1} (NaPO₃)₆ is added before measuring the soil particle size using a laser particle size analyzer. The stirrer speed is set at 2500 rpm with a light-blocking range of 10–15%. Soil pH is determined by mixing 10 g of soil with 25 mL of deionized water. The mixture is shaken at room temperature for 30 min and then allowed to settle for 2 h before the pH is measured using a pH meter. SOM is quantified using a dichromate-sulfuric acid solution in an electric sand bath to oxidize organic carbon in the soil. The remaining dichromate is titrated with a standard ferrous sulfate solution, and silica is used as an additive for reagent blank calibration⁴³. All experiments include duplicates and blanks to ensure quality control.

K-means cluster analysis

Cluster analysis is an unsupervised algorithm used in data mining for clustering samples⁴⁴. It groups samples with similar characteristics into different categories, ensuring that all samples within the same group exhibit relatively similar features⁴⁵. In geochemical research, a series of soil samples are collected, and near-term studies have utilized the K-means method to classify these samples based on their distinct characteristics^46,47.

4

where is the objective function, is the cluster, is the number of samples in the cluster, and the distance function = represents the distance between each sample point and the center of mass in the cluster. The center of mass can be calculated based on the following function:

5

After standardizing the data, three classes were designated for cluster analysis.

Future climate change

This study utilizes the CMIP6 models⁴⁸ to simulate future hydrological processes under four different societal development pathways (SSP126, SSP245, SSP370, and SSP585), estimating streamflow using three climate variables: daily precipitation, minimum temperature, and maximum temperature. To reduce model uncertainty, this study employs five climate models from CMIP6 in our research (Table 2). To assess the suitability of this model data, the reference period data (2008–2014) from this scenario model is compared with historical observed values. The daily climate data from the historical observations are derived from the CMADS dataset (Table 1), including precipitation, temperature, wind speed, radiation, and humidity, which has been widely applied in studies related to the SWAT model. The study area currently includes 61 CMADS stations, as shown in Fig. 1.

Table 2.

Information about the GCMs used in this study.

Development Institute	GCM
Commonwealth Scientific and Industrial Research Organization (Australia)	ACCESS-CM2
Commonwealth Scientific and Industrial Research Organization (Australia)	ACCESS-ESM1-5
Atmosphere and Ocean Research Institute, The University of Tokyo (Japan)	MIROC6
Meteorological Research Institute (Japan)	MRI-ESM2–0
The NUIST Earth System (China)	NESM3

Bias correction method

The bilinear interpolation method was used to interpolate to the location of the meteorological stations within the study area⁴⁹. Considering the Multi-Model Ensemble (MME) shows less uncertainty. The MME mean of five climate models was therefore employed to use in the SWAT model⁵⁰. Then, the multi-model ensemble data were corrected by the Quantile mapping (QM) method⁷. QM is a bias correction technique that modifies the cumulative distribution function (CDF) of model outputs to align with observed data, using a non-parametric approach⁵¹. This method is widely used in the post-processing of downscaled simulated precipitation and temperature data^52,53. In this study, QM is applied to three climatic variables: precipitation (Pr), minimum temperature (Tasmin), and maximum temperature (Tasmax), fitting them against observed data from the historical period (2004–2014) to reduce the biases of the GCMs. The traditional quantile mapping method is defined as follows⁵⁴:

6

where represents the in-situ climatic variable, represents the GCM output, is the CDF of , and is the inverse of the CDF of . The corrected factors are then used to adjust the biases in the predicted data for each station for the period 2015–2100. This study implements the QM method to perform bias correction on the three climatic variables: Pr, Tasmin, and Tasmax.

Figure 2 describes the methodological steps used in this study. Initially, observed data on pH, SOM, and soil particle size by subbasin were input to construct the SWAT model for the study area, followed by calibration and validation using SWAT-CUP 2019. Subsequently, data from the MME, after bias correction, were input into the model to simulate future hydrological processes in the study area. Finally, the precipitation, streamflow, and soil data for different periods were superimposed on the classification results from the cluster analysis to identify the spatial distribution of the water-soluble proportions of Sb and As in the study area over different periods.

Fig. 2.

Overview of the research framework, where blue boxes represent the inputs/experimental content of the model, orange boxes represent the research methodology used, gray boxes represent the scenarios and time scales simulated, and green boxes represent the outputs.

Result

SWAT model improved and hydrological response

Improvement of the SWAT model soil database

The sampled soils were analyzed for pH, SOM and soil particle size, and the results are shown in Table S4. Soil particle size is determined as volume percentages within the range of 0.02–2000 μm, divided into clay (0–2 μm), silt (2–20 μm), and sand (20–2000 μm)⁵⁵. The soil composition of the study area is shown in Fig. S2, indicating a predominance of sandy loam, silty loam, and silt, with silt accounting for a higher proportion overall. Higher SOM and clay content are distributed in the predominant yellow soil, while higher pH values are found in the yellow-red soil, mainly in the western part of the study area. Parameters of the soil database for the SWAT model include pH value, SOM, and soil particle size. Considering that the simulation of SWAT-based metal migration is evaluated at the subbasin level, the measured soil physicochemical properties are further applied to the subbasin where the study is located, followed by simulation with the SWAT model.

The improved soil database was applied to the SWAT model, and the SWAT simulation results were compared before and after the soil database improvement. As shown in Table 3, the R², NSE, and RSR values for streamflow results using the improved soil database demonstrated improvements compared to historical observed data from 2010 to 2015. The NSE and RSR for sediment load also showed significant enhancement, indicating improved prediction accuracy. This indicates that dynamic soil data updates enhance the accuracy of the SWAT model.

Table 3.

Comparison of simulation results and measured values of hydrological processes in the study area before and after soil database improvement (2010–2015).

Hydrological station	Hydrological processes	Before updating soil data			After updating soil data
		R 2	NSE	RSR	R 2	NSE	RSR
Shihuichang	Streamflow	0.88	0.69	0.56	0.93	0.84	0.40
Yongwei	Streamflow	0.86	0.43	0.76	0.94	0.65	0.59
	Sediment loads	0.51	-73.10	8.61	0.35	-1.99	1.73

Calibration and verification of SWAT model

The bias correction of Pr, Tasmax, and Tasmin was conducted for the 61 CMADS stations in the study area using the QM method. The results of the calibration and validation periods were compared. The first CMADS Station, before bias correction, the simulated Pr significantly overestimated measured precipitation for most of the calibration period. After bias correction, the results are shown in Fig. 3. During the calibration period, the simulated Pr, Tasmax, and Tasmin demonstrate pronounced improvements. The improvement in simulations during the validation period was slightly less significant than that observed during the calibration period. Although the results during the validation phase were slightly less favorable compared to the calibration phase, the biases in underestimation and overestimation were significantly reduced. As illustrated in Fig. 4, compared to Tasmax and Tasmin, the bias correction effect on Pr in July and August is significant. Overall, the bias-corrected Pr, Tmax, and Tmin data exhibited some improvement. Thus, the bias-corrected data derived via the QM method can be reliably utilized to facilitate the ensuing SWAT model simulations.

Fig. 3.

Bias correction results of the probability distribution of monthly precipitation (Pr) in the calibration period (a), monthly maximum temperature (Tasmax) in the calibration period (b), monthly minimum temperature (Tasmin) in the calibration period (c); Bias correction results of the probability distribution of monthly precipitation (Pr) in the validation period (d), monthly maximum temperature (Tasmax) in the validation period (e), monthly minimum temperature (Tasmin) in the validation period (f).

Fig. 4.

Bias correction results of monthly precipitation (Pr) in the calibration period (a), monthly maximum temperature (Tasmax) in the calibration period (b), monthly minimum temperature (Tasmin) in the calibration period (c); Bias correction results of monthly precipitation (Pr) in the validation period (d), monthly maximum temperature (Tasmax) in the validation period (e), monthly minimum temperature (Tasmin) in the validation period (f).

The performance of five individual CMIP6 models and their multi-model ensemble (MME) for Pr, Tasmax, and Tasmin at one meteorological station was visually assessed using standardized Taylor diagrams based on three metrics: correlation coefficient, relative standard deviation, and normalized root mean square error. Points closer to the observation point indicate better overall model performance. The results for Tasmax, Tasmin, and Pr during the 2008–2014 period from the five individual CMIP6 models and the MME are shown in Fig. 5. The MME demonstrates good performance across all three meteorological variables. As shown in Fig. 5, the MME results after QM bias correction are closer to the observations than individual models, improving the simulation performance to some extent. Notably, Pr shows significant improvement under the MME: the coefficient of determination (R²) for individual models ranged from approximately 0.54 to 0.96, while the MME achieved an R² of 0.99. Similarly, the Nash-Sutcliffe efficiency (NSE) for individual models ranged from about − 0.89 to 0.81, whereas the MME attained an NSE of 0.95. For Tasmax and Tasmin, both R² and NSE exceeded 0.98 under the MME.

Fig. 5.

Taylor diagram of simulated and observed precipitation, maximum temperature, and minimum temperature in the weather station.

The simulation period is divided into a warm-up period (2008–2009), a calibration period (2010–2015), and a validation period (2016–2017), with the calibration results presented in Table 4. The best simulation results for the calibration and validation periods, including the monthly average observed and simulated streamflow and sediment loads, are shown in Fig. 6. Figure 6a displays the streamflow results from the Shihuichang station using the SWAT model during both periods. Since sediment load data were lacking, only streamflow data were calibrated at this station. The streamflow simulation results at the Shihuichang station performed well across all evaluation metrics during both periods. Figure 6(b) shows the streamflow results at the Yongwei station, which, compared to the Shihuichang station, had poorer performance in the NSE, but overall, the fit between simulated and observed streamflow values during the calibration and validation periods was good. Additionally, the streamflow at both stations showed a significant correlation with precipitation, a finding also noted in previous studies³⁰. Over the entire simulation period, the average observed streamflow at the Shihuichang station was 138.01 m³/s, and the simulated average was130.38 m³/s; the average observed streamflow at the Yongwei station was 301.22 m³/s, and the simulated average was 208.11 m³/s. Figure 6c displays the sediment loads simulation results at the Yongwei station, where the calibration period showed better simulation results compared to the validation period, but the accuracy was lower than that of the streamflow simulation. Notably, the sediment simulation results met the existing standards and allowed for further analysis, which is a situation also observed in previous studies^56,57. This might stem from employing the Modified Universal Soil Loss Equation (MUSLE) to compute sediment loads, making the sediment loads simulation results also dependent on factors such as slope, slope length, and land cultivation practices. The best-fit regression lines for the three hydrological parameters at both stations are plotted in Fig. 6d, showing that the R² values for streamflow at both the Shihuichang and Yongwei stations exceeded 0.9, indicating a good fit, whereas the fit for sediment at the Yongwei Station was slightly lower, reaching 0.62. In summary, the calibrated SWAT model meets the accuracy requirements for simulating streamflow indicated in the Duliujiang Basin and can be used for subsequent analyses.

Table 4.

Calibration results for the hydrological station parameters in the study area.

Parameter	Calibration in Shihuichang station	Final value	Calibration in Yongwei Station	Final value
Streamflow	R__CN2.mgt	0.195	R__CN2.mgt	0.0875
	V__ALPHA_BF.gw	0.275	V__ALPHA_BF.gw	0.2325
	V__GW_DELAY.gw	77.5	V__GW_DELAY.gw	228.75
	V__GWQMN.gw	425	V__GWQMN.gw	4812.5
	V__GW_REVAP.gw	0.170	V__GW_REVAP.gw	0.14915
	V__RCHRG_DP.gw	0.285	V__RCHRG_DP.gw	0.7875
	V__REVAPMN.gw	162.5	V__REVAPMN.gw	351.25
	R__SOL_AWC(.).sol	-0.415	R__SOL_AWC(.).sol	0.0125
	V__CH_N2.rte	0.0204	V__CH_N2.rte	0.001748
	V__CH_K2.rte	122.50	V__CH_K2.rte	1.250998
	R__SOL_K(.).sol	0.265	R__SOL_K(.).sol	0.0975
Sediment			V__CH_COV1.rte	0.149875
			V__CH_COV2.rte	0.657158
			V__ADJ_PKR.bsn	0.72875
			V__SPCON.bsn	0.007847
			V__SPEXP.bsn	1.0325
			V__USLE_K(.).sol	0.430625

Fig. 6.

Observed versus simulated monthly streamflow in study area. (a) Observed and simulated monthly streamflow at Shihuichang station. (b) Observed and simulated monthly streamflow at Yongwei station. (c) Observed and simulated monthly sediment loads at Yongwei station. In figures (a–c), the calibration and validation phases list the corresponding R², NSE, and RSR results. (d) Best fit results for the hydrological process.

Future climate change

The research period is divided into four intervals: 2015–2040 (Near-term), 2041–2070 (Medium-term), and 2071–2100 (Long-term). Using the QM method to correct the bias of Pr, Tasmax, and Tasmin, the results are shown in Fig. 7. According to Fig. 7b1,c1, the future Tasmax and Tasmin under four climate scenarios show a rising trend with similar patterns of change, while Fig. 7a1 indicates that Pr exhibits a fluctuating trend. The rate of change for the future three periods is shown in (Fig. 7a2,b2,c2). Among four SSP scenarios in the future periods, Pr has increased by 0.24 to 0.68 from near-term to medium-term, and by 0.18 to 0.60 from medium-term to long-term, overall maintaining a fluctuating state. Tasmin has increased by 0.86 to 1.90 from near-term to medium-term, and by 0.06 to 1.17 from medium-term to long-term, showing an overall upward trend. Tasmax has increased by 1.99 to 3.30 from near-term to medium-term, and by 0.24 to 2.99 from medium-term to long-term, indicating an overall upward trend as well. Tasmin and Tasmax exhibit similar growth patterns, with growth rates of approximately 10% in the near-term periods, 10% in the medium-term periods, and 5% in the long-term periods. The increased rate order of climate scenarios with Pr being ranked as SSP585 > SSP370 > SSP126 > SSP245, while Tasmin and Tasmax are ranked as SSP585 > SSP370 > SSP245 > SSP126.

Fig. 7.

Changes in climate indicators under various scenarios. (a1), (b1), and (c1) represent the trends in precipitation (Pr), maximum temperature (Tasmax), and minimum temperature (Tasmin) under the SSP126, SSP245, SSP370, and SSP585 scenarios from 2015 to 2100; (a2), (b2), and (c2) represent the values of Pr, Tasmax, and Tasmin in the near-term, medium-term, and long-term under different scenarios.

Future hydrological response

Using climate data (Tasmax, Tasmin, Pr) corrected by the QM method for the years 2015–2100 and a SWAT model that has been calibrated and validated, future Yongwei hydrological station variables were simulated to enable a prospective assessment of environmental dynamics. As shown in Fig. 8a, the average streamflow values for the near-term, medium-term, and long-term are 147.51, 182.11, and 217.13 m³/s, respectively, indicating a gradual increase in average streamflow over time. The ranking of average streamflow values under different climate scenarios for each period is as follows: for the near-term periods, SSP126 > SSP245 > SSP585 > SSP370; for the medium-term periods, SSP126 > SSP585 > SSP245 > SSP370, and for the long-term periods, SSP585 > SSP126 > SSP370 > SSP245. Regarding the monthly changes in streamflow, as depicted in (Fig. 8b), all four scenarios show a significant increase in streamflow during May to July and a decrease in January and February in the future. Regarding the spatial patterns of streamflow, the main river channels are relatively greater than those in the tributaries (Fig. 8c).

Fig. 8.

Comparison of annual distribution of streamflow under different future climate scenarios: (a) Annual streamflow from 2020 to 2100 under four scenarios: SSP126, SSP245, SSP370, SSP585; (b) Comparison of interannual distribution of streamflow under different future climate scenarios; (c) The streamflow with the subbasin under four scenarios: SSP126, SSP245, SSP370, SSP585. Map created using ArcGIS 10.0 (https://www.esri.com).

Sensitivity analysis of metalloid migration under climate change

Soil cluster analysis

This study measured the concentration of soil heavy metal (loid)s (Table S2) and the water-soluble heavy metal (loid)s (Table S3) at sampling locations within the study area. The experimental results revealed that the concentrations of Sb and As in the soil exceeded the standard. Particularly, at agricultural land use, the exceedance rate of Sb reached 50%, and for As was about 15%⁴³ (Fig. S1a). The exceedance rates for the water-soluble proportions of Sb and As were 40 and 13%, respectively (Fig. S1b)⁵⁸. This indicates that the soils in the study area, under significant streamflow erosion, possess potential pollution risks related to leaching, which could pose health risks to the surrounding population, necessitating long-term monitoring. The non-carcinogenic health risk associated with centralized drinking water sources was assessed using hazard quotients in the present study. The hazard quotients were calculated as the ratio between the exposure concentration and the reference dose (RfD)^59,60. The calculated parameters are listed in Table S5-1, S5-2, and S5-3. The results are shown in Fig. S4. It is evident that Sb exhibits a higher non-carcinogenic health risk upstream compared to As. It is noteworthy that children experience higher non-carcinogenic health risks from As and Sb compared to adults.

Considering the relationship between soil pH and SOM with the proportions of water-soluble Sb and As, a cluster analysis of these four parameters was conducted across all sampling sites to discern patterns in the proportions of water-soluble Sb and As corresponding to different soil properties. The results showed that the sampling sites were divided into three clusters, as detailed in Table 5. The cluster analysis revealed significant differences in SOM, pH values, percentage of clay, and the proportions of water-soluble Sb and As among the three clusters. Cluster 1, containing 9 sampling sites, was characterized by low SOM and low pH, with the lowest proportions of water-soluble Sb and As. Cluster 2, containing 5 sampling sites, had medium SOM, medium pH, and high clay fraction, with a higher proportion of water-soluble Sb and a lower proportion of water-soluble As. Cluster 3, with 13 sampling sites, featured low SOM and low clay fraction, and displayed the same trend of water-soluble Sb and As as Cluster 3. Cluster 4, containing two sampling sites, was characterized by high SOM and high pH, with higher proportions of water-soluble Sb and As.

Table 5.

Clustering results of soil physicochemical properties and proportions of Water-Soluble state Sb.

Name		Cluster_1 Cluster_1		Cluster_2	Cluster_3	Cluster_4		F	P
		Low SOM / low pH		Medium SOM / Medium pH /High clay fraction	Low SOM / low clay fraction	High SOM / high pH
SOM		2.09 ± 0.94		4.54 ± 1.37	2.20 ± 1.07	6.42 ± 1.74		13.352	0.000**
pH		5.99 ± 1.22		6.60 ± 0.81	7.65 ± 0.36	7.80 ± 0.92		8.227	0.001**
Clay fraction		12.62 ± 6.31		15.80 ± 5.24	5.88 ± 4.08	8.81 ± 1.06		5.995	0.003**
Percentage of water-soluble Sb		0.27 ± 0.14		0.81 ± 0.09	0.87 ± 0.24	0.94 ± 0.14		20.53	0.000**
Percentage of water-soluble As		0.06 ± 0.04		0.33 ± 0.16	0.32 ± 0.25	1.37 ± 0.60		19.28	0.000**

*p < 0.05, **p < 0.01.

Further analysis of the clustering results reveals that the proportions of water-soluble Sb and As generally show a positive correlation with SOM and pH. Previous studies have indicated that SOM generates negative surface charges as the pH increases, enhancing the repulsion of Sb oxyanions and leading to an increase in the water-soluble Sb⁶¹. As shown in Fig. S3, the average proportion of water-soluble Sb and As in Cluster Low SOM / Low pH is the lowest among the four groups. Research has demonstrated that pH can affect the speciation of heavy metals; when pH decreases, ions with similar charges repel each other, causing minerals, soil particles, and SOM to carry positive charges, thereby facilitating the adsorption of anions⁶². Considering that Sb and As mainly exist in the form of oxyanions in soil⁶³, these metals are more likely to be adsorbed at lower pH values, resulting in a lower proportion in the water-soluble fraction. Cluster High SOM / High pH has the highest average proportion of water-soluble Sb and As among the four groups, characterized by higher SOM content and pH. Based on the above analysis, the soil in the study area is further classified by subbasin into Cluster Low SOM / Low pH, Cluster Medium SOM / Medium pH /High Clay Fraction, Cluster Low SOM / Low Clay Fraction and Cluster High SOM / High pH, as shown in Fig. 9. It is evident that soils of Cluster Medium SOM / Medium pH /High Clay Fraction are predominantly found in the western part of the study area, the soils of Cluster Low SOM / Low pH are predominantly found in the eastern part of the study area, while soils of High SOM / High pH and Low SOM / Low Clay Fraction are interspersed in the central parts. Therefore, soils with a higher proportion of water-soluble Sb and As are more prevalent in the central part of the study area.

Fig. 9.

The soil in the study area is reclassified based on cluster analysis results, with each subbasin as a unit. The soil characteristics of the four clusters include Low SOM /Low pH Cluster, Medium SOM / Medium pH /High Clay Fraction Cluster, Low SOM / Low Clay Fraction Cluster, and High SOM / High pH Cluster. Map created using ArcGIS 10.0 (https://www.esri.com).

Classification of hydrological and meteorological factors

Under varying precipitation conditions, changes in soil pH and SOM are anticipated, with studies indicating that increased precipitation can raise soil pH and cause slight losses of SOM. These changes subsequently affect the water-soluble fraction of metals such as Sb and As in the soil. Future near-term, medium-term, and long-term precipitation data from CMADS stations in the study area were processed using Kriging interpolation after QM correction to obtain the total precipitation across the study area. Each sub-basin was then analyzed for its interpolated precipitation results. Based on the annual precipitation amounts for different periods, the subbasins were classified into: Cluster 1: (> 1300 mm), Cluster 2: (1200–1300 mm), Cluster 3: (< 1200 mm). The classification of streamflow in the subbasins for each period is shown in Fig. 10. From the distribution of increased precipitation in subbasins, the majority are located in the western region of the study area; over time, precipitation gradually increases, which may be a significant factor leading to increased future basin streamflow. From the near-term to medium-term, precipitation has significantly risen in the southern parts of the study area, while from the medium-term to long-term, the western areas exhibit a notable increase in precipitation.

Fig. 10.

Future precipitation classification results for subbasins in different periods, where Cluster 1 has an annual precipitation of > 1300 mm, Cluster 2 has an annual precipitation of (1200–1300) mm, and Cluster 3 has an annual precipitation of (< 1200) mm. (a) Near-term (2015–2040), (b) Medium-term (2041–2070), (c) Long-term (2071–2100). Map created using ArcGIS 10.0 (https://www.esri.com).

To further summarize the future period streamflow in the study area, it’s noted that the relative change rate of flow in each subbasin under different scenarios is about 2%, with the maximum change rate not exceeding 9%. Accordingly, the streamflow under various scenarios is averaged and categorized into three clusters: Cluster 1 (> 10 m³/s), Cluster 2 (1–10 m³/s), Cluster 3 (0–1 m³/s). The classification of streamflow in subbasins across different periods is shown in Fig. 11. It is evident that subbasins near the main river channel will continue to experience higher streamflow volumes in the future. Additionally, due to the overall increase in streamflow, many subbasins originally in Cluster 3 have transitioned to Cluster 2. Spatially, most of these subbasins are located in the western part of the study area, which corresponds to regions with increased precipitation. During the transition from the near-term to the medium-term and from the medium-term to the long-term, some tributaries have experienced an increase in streamflow.

Fig. 11.

Future streamflow classification of subbasins in different periods, where Cluster 1 (> 10 m³/s), Cluster 2 (1–10 m³/s), and Cluster 3 (0–1 m³/s). (a) Near-term (2020–2040), (b) Medium-term (2041–2070), (c) Long-term (2071–2100). Map created using ArcGIS 10.0 (https://www.esri.com).

Distribution of water-soluble Sb and as under climate change

To evaluate the distribution of water-soluble fractions of Sb and As under climate change, subbasins were categorized and scored based on three classifications: Soil properties : Group “Low SOM / Low pH” was assigned 1 point, “Medium SOM / Medium pH / High Clay Fraction” and “Low SOM / Low Clay Fraction” 2 points, and “High SOM / High pH” 3 points. Precipitation patterns : Cluster 1 received 3 points, Cluster 2 2 points, and Cluster 3 1 point. Streamflow regimes (Sect. 3.2.3): Cluster 1 was assigned 3 points, Cluster 2 2 points, and Cluster 3 1 point. Assuming invariant soil physicochemical properties, scores for each subbasin were aggregated across near-term (2020–2040), mid-term (2041–2070), and long-term (2071–2100) periods. Four weighting schemes were applied: equal weights (2:2:2) and three sensitivity scenarios emphasizing one indicator (weights 1:1:4). Total subbasin scores ranged from 6 to 18 points. Figure 10 details the resulting distributions of water-soluble Sb and As fractions across the study area for each period, including the full future horizon (2015–2100). This structured scoring framework identifies regions potentially at higher risk of water-soluble metal pollution under evolving climatic conditions.

Figure 12 displays the distribution of the percentage of water-soluble Sb and the percentage of water-soluble As across the study area under different weighting schemes. The subplots (a), (b), and (c) represent the near-term (2015–2040), medium-term (2041–2070), and long-term (2071–2100) periods, respectively. The subplots 1, 2, 3, and 4 denote the weighting schemes: equal weighting scheme, soil properties weight prioritized scheme, precipitation weight prioritized scheme, and streamflow weight prioritized scheme, resulting in 12 subplots (a1-c4). Subplots a1, b1, and c1 (equal weighting) show that subbasins along the main river channels in the central and eastern study area exhibit a higher percentage of water-soluble Sb and percentage of water-soluble As compared to the western region. This is primarily attributed to the lower proportion of Low SOM and Low pH soils and higher streamflow in the eastern main river subbasins. Temporally, significant precipitation changes are projected to drive a gradual increase in the percentage of water-soluble Sb and the percentage of water-soluble As in the western region from the medium-term to the long-term. Further examination of subplots a2-b2-c2, a3-b3-c3, and a4-b4-c4 reveals that, regardless of the weighting scheme, subbasins along the eastern main river channels consistently exhibit significantly higher percentages of water-soluble Sb and percentages of water-soluble As. In contrast, the western main river subbasins show a clearer influence of precipitation. Overall, the geographical spatial trend indicates a progressive increase in the percentage of water-soluble Sb and the percentage of water-soluble As from north to south and west to east across the study area. Temporally, the overall percentage of water-soluble Sb and the percentage of water-soluble As gradually increase over time. The influence of soil properties, precipitation, and streamflow underscores the complex interplay between climate change and pollutant dynamics within the watershed.

Fig. 12.

The possibility distribution scores of percentage of water-soluble Sb and percentage of water-soluble As for future periods, where (a1) (b1) (c1) represent the scores in near-term (2015–2040), medium-term (2041–2070), and long-term (2071–2100) periods under equal weighting scheme; (a2) (b2) (c2) represent the scores in near-term (2015–2040), medium-term (2041–2070), and long-term (2071–2100) periods under soil properties weight prioritized scheme; (a3) (b3) (c3) represent the scores in near-term (2015–2040), medium-term (2041–2070), and long-term (2071–2100) periods under precipitation weight prioritized scheme; (a4) (b4) (c4) represent the scores in near-term (2015–2040), medium-term (2041–2070), and long-term (2071–2100) periods under streamflow weight prioritized scheme. Note: Map created using ArcGIS 10.0 (https://www.esri.com).

Conclusion

Sb and As may pose risks to both human health and the environment, and there is a strong correlation between the risk these metals to the environment and their mobility in soil. This study focuses on the Duliujiang basin in Southwest China, employing the SWAT model combined with experimental results, and using the QM method with temperature and precipitation data under five CMIP6 climate models and four SSP scenarios to construct a framework for assessing the water-soluble proportions of Sb and As. The primary conclusions are as follows:

The improved SWAT model performed satisfactorily in simulating basin streamflow and sediment. Projections from 2015 to 2100 indicate rising temperatures and a gradual increase in streamflow over time. Experimental results revealed that both agricultural and industrial areas exceeded standards for Sb and As, with Sb contamination being particularly severe. K-means cluster analysis of soils identified four distinct categories based on relationships between water-soluble Sb/As proportions and soil pH, particle size, and SOM: High pH/High SOM; Medium SOM/Medium pH/High Clay Fraction; Low SOM/Low Clay Fraction; and High SOM/High pH. This analysis further demonstrated that high pH and high SOM positively influence the percentage of water-soluble Sb and the percentage of water-soluble As. By integrating soil clustering with projected precipitation and streamflow patterns, the study found that the percentage of water-soluble Sb and the percentage of water-soluble As are expected to increase from west to east across subbasins, reaching the highest concentrations in the eastern main river channel. Over time, both the percentage of water-soluble Sb and the percentage of water-soluble As are projected to increase throughout the study area, driven by the combined effects of soil properties, streamflow, and precipitation.

This research, utilizing an improved model to construct a framework for assessing the water-soluble proportions of Sb and As in mining areas, is significant for the prevention and control of pollution migration risks under climate change. However, the study has limitations, as highlighted by Mengistu et al.⁶⁴, including data, model settings, and parameter uncertainties in the SWAT model. Despite these recognized limitations and uncertainties, the model’s soil database was improved through sampling experiments, achieving satisfactory simulation results, thus providing reliability for analyzing hydrological process changes in the basin under climate change. Another limitation is that future climate change scenarios, under conditions of precipitation, will lead to increased soil pH and loss of SOM, which will respectively have positive and negative impacts on the water-soluble proportions of Sb and As. Considering that changes in pH are more significant than those in SOM and that pH has a more pronounced effect on the proportion of water-soluble Sb and As, areas with heavy precipitation are identified as favorable for increasing the proportion of water-soluble Sb and As.

This study significantly enhances understanding of hydrological processes and the water-soluble proportion of Sb and As in mining areas under climate change. Importantly, it highlights the critical link between hydrological processes and the water-soluble proportion of Sb and As in mining areas, emphasizing the dominant role of soil pH and SOM. Notably, the evident changes in hydrological processes in the study area under the backdrop of climate change will inevitably lead to the migration of pollutants, necessitating analysis and assessment of potential pollution areas. This research provides information for the management of metals in mining areas under climate change and for health assessments of mining area basins.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Chujie Bu: Writing original draft, Software, Methodology, Formal analysis, Conceptualization. Pan Wu: Supervision, Conceptualization, Methodology, Funding acquisition, Validation. Jie Niu: Writing review & editing, Validation. Yonglin Chen: Writing review. Jiangxun Huang: Investigation. Bo Liu: Investigation.

Data availability

Data will be made available by contacting the corresponding author.pwu@gzu.edu.cn (Pan Wu).

Declarations

Competing interests

The authors declare no competing interests.