Optimal control of nonpoint source pollution in the Bahe River Basin, Northwest China, based on the SWAT model

This study conducts in the Bahe River Basin, an agricultural basin in Northwest China. We use the Soil and Water Assessment Tool (SWAT) model to identify the spatial distribution characteristics of non-point source (NPS) pollution and determine the critical source areas (CSA). Then the relationship between landscape pattern and NPS pollution is analyzed by spearman correlation analysis and redundancy analysis (RDA). On this basis, we set up eight landscape management practices in the CSA and evaluate their reduction effects on NPS pollution loads. The results show that the spatial distribution of nitrogen and phosphorus loss intensity has a certain correlation with rainfall and runoff, and the correlation between phosphorus loss intensity and sediment loss intensity is more significant. The NPS pollution load is closely related to the landscape pattern of the river basin, and is affected by the fragmentation, aggregation and complexity of the landscape. Farmland, forest land, and grassland are the main landscape components of the river basin. Farmland is the main source of NPS pollution, whereas forest land and grassland can effectively inhibit the output of NPS pollution, and the reduction effect of forest land is significantly better than that of grassland. The largest patch index (LPI), landscape shape index (LSI), patch density (PD) are the main landscape factors that affect the output of NPS pollution load. Among all the scenarios, the reduction effect of returning farmland to forest land in slopes above 15° is the best, and the reduction rates of total nitrogen (TN) and total phosphorus (TP) loads have reached about 25%. This study provides some reference for the management of NPS pollution in the Bahe River Basin and other similar basins.

Keywords: Bahe River Basin; Landscape pattern; NPS pollution; Optimized control; SWAT model

Surface water and groundwater are the two important components in the hydrological cycle (Li et al. [20]). The research object of this paper is the non-point source (NPS) pollution caused by surface runoff under the effect of rainfall, which is affected by a variety of factors: hydro-meteorological factors (Rainfall, rainfall intensity, and rainfall duration), soil type factors (soil structure and physicochemical properties), terrain factors (slope, slope length, and slope type), biological factors (vegetation canopy interception, root fixation, and litter interception), and human activities (Du et al. [9]; Ferreira et al. [10]). In ecological landscape science, it is generally believed that eco-hydrological process and watershed landscape pattern are closely related to each other. Landscape pattern plays a key role in hydrological cycle process and then affects the generation, migration, and transformation of NPS pollution (Zhang et al. [33]). Artificial landscapes such as farmland and construction land have replaced natural landscapes such as grassland and forest land, which causing the imbalance in the proportion of "source" and "sink" landscape. In particular, the change of landscape pattern caused by land use is the main cause of NPS pollution (Mehdi et al. [22]). For example, Cui et al. ([8]) found that land use/landscape pattern can affect the water quality through water environmental pressure and water environment bearing capacity in the upper reaches of Xiaoxia Bridge section of the Huangshui River Basin. Therefore, it is important to determine the quantitative relationship between landscape pattern and water quality and then to further understand the hydrological cycle and NPS pollution (Wu and Lu [28]).

In recent years, a large number of models, such as SWAT, Generalized Watershed Loading Function (GWLF), Spatially Referenced Regressions On Watershed attributes (SPARROW), and Hydrological Simulation Program-Fortran (HSPF), have been widely used and become powerful tools for scholars to study NPS pollution (Banadkooki et al. [4]; Panagopoulos et al. [25]). SWAT is most widely used to estimate the output of NPS pollution and assess the effects of climate change and management measures on NPS pollution (Arabinda Sharma and Tiwari [3]; Himanshu et al. [15]). Coupling the eco-hydrological process with the landscape pattern index; combining the model with the eco-landscape method; and managing the proportion, quantity, and space-time allocation of landscape elements can improve the stability of the landscape structure and achieve the goal of NPS pollution control (Zhang et al. [31]).

Bahe River is the primary tributary of Weihe River, a tributary of the Yellow River. And it is also one of the main water sources in Xi'an, Shaanxi Province in Northwest China. However, due to the acceleration of urbanization, the water quality in the basin is deteriorated seriously. Meanwhile, under the call of "Yellow River Basin ecological protection and high-quality development" in China, the NPS pollution control work of the Bahe River Basin is imminent. Previous studies in the Bahe River Basin mainly focused on the characteristics of runoff and NPS pollution, the analysis of the total amount control of pollutants into river based on pollution carrying capacity, the impacts of climate change and land use change on NPS pollution and the evaluation of best management practices (BMPs) (Zhang and Zhang [32]; Nie et al. [24]; Bi et al. [5]). Few studies have involved how landscape pattern affects NPS pollution. Based on this, the Bahe River Basin is selected as the study area. The SWAT model is used to simulate the NPS pollution load on the watershed scale. The landscape pattern index, spearman correlation analysis, and RDA are applied to analyze the relationship between watershed landscape factors and NPS pollution loads at the landscape level. And then eight landscape management practices are set up in the critical source areas (CSA) to evaluate the reduction effect on NPS pollution loads. This paper aims to provide theoretical basis for the optimal control scheme of NPS pollution in the Bahe River Basin and other similar basins.

The Bahe River Basin is located in Shaanxi Province in Northwest China (Fig. 1). The total length of the river is 104 km and the drainage area is 2581 km2. The study area belongs to the warm temperate semi-humid continental monsoon climate. In spring, the temperature rises rapidly and the temperature range varies greatly. In summer, the solar radiation is strong and rainstorm occurs frequently. In autumn, the climate is mild, humid, and mostly rainy. In winter, the solar radiation is weak, cold, and dry. The runoff in the basin varies greatly from year to year and the annual distribution is uneven. The annual runoff is mainly concentrated in flood season from July to October. The distribution of sediment is similar to that of runoff. The inter-annual sediment transport varies greatly. And the annual sediment transport mainly concentrates from July to October with the total sediment transport is as high as 1.83 million tons, accounting for 84.43% of the annual average sediment transport. Therefore, the variation of runoff and sediment in the Bahe River Basin has the characteristics of "large runoff and more sediment, small runoff and less sediment." The study area is a typical agricultural basin in Northwest China, with a large area of farmland, grassland, and forest land (Fig. 2). The water quality of Bahe River is polluted in varying degrees from upstream to downstream. The overall pollution trend is gradually increasing along the flow direction. The main pollutants are chemical oxygen demand (COD), ammonia nitrogen (NH4-N), petroleum, fecal coliform, and TP.

Graph: Fig. 1 Location map of the Bahe River Basin

Graph: Fig. 2 Land use map of the Bahe River Basin

The dataset of the SWAT model includes spatial data and attribute data, as is shown in Table 1. Spatial data mainly involve digital elevation model (DEM), land use map, and soil type map. Attribute data include soil attribute data, land use data, hydro-meteorological data, and agricultural management data. Among them, the meteorological data adopts the China Meteorological Assimilation Driving Datasets for the SWAT model (CMADS V1.0), which is developed based on China Land Data Assimilation System (CLDAS) and provide high resolution and quality meteorological data for researchers. CMADS has greatly reduced the uncertainty of meteorological input data and has achieved good simulation results in China.

Table 1 Bahe River Basin dataset.

This paper chooses Krasovky_1940_Alber projection to transform all spatial data in a unified coordinate system. The threshold value is set to 49 km2, and the whole basin is divided into 27 sub-basins. Then the basin is divided into 1049 HRUs by setting the minimum threshold ratio of land use area, soil area, and slope area as 1, 1, and 0%. After the meteorological data and agricultural management data are imported, the model can be simulated.

It is necessary to assess the uncertainty of parameters in the construction of SWAT model (Yan et al. [30]). The Sequential Uncertainty fitting (SUFI-2) algorithm of SWAT-CUP tool is used to analyze the sensitivity and uncertainty of parameters from three aspects (Abbaspour et al. [1]): runoff, sediment, and water quality. The finalized sensitivity parameters and the calibrated values are shown in Table 2.

Table 2 The sensitivity parameters and their calibration values

The Nash-Sutcliffe efficiency coefficient (ENS), relative error (Re), and the determination coefficient (R2) are selected as the criteria for the accuracy evaluation of the SWAT model. The calculation formulas are as follows (Eqs. (1), (2), and (3) (Freer et al. [11])):

Graph

2 $$R_e=\frac{\sum _{i=1}^n\left(O_i-S_i\right)}{\sum _{i=1}^n{S}_i}\times 100\%$$

Graph

3 $$R^2={\left\{\frac{\sum _{i=1}^n\left(O_i-\overline{O}\right)\left(S_i-\overline{O}\right)}{{\left[\sum _{i=1}^n{\left(O_i-\overline{O}\right)}^2\right]}^{0.5}{\left[\sum _{i=1}^n{\left(S_i-\overline{S}\right)}^2\right]}^{0.5}}\right\}}^2$$

Graph

where n is the total number of runoff data series, Oi is the observed measured data, Si is the simulated data of the SWAT model, ‾O is the average of observed measured data, and‾S is the average value of simulated data of the SWAT model.

Landscape pattern refers to the spatial arrangement and combination of Landscape elements, including the types, numbers, spatial distribution, and configuration of the landscape components. It is a comprehensive expression of landscape heterogeneity in space. In the landscape pattern analysis software FRAGSTATS v4.2, landscape indicators are displayed at three levels: patches, classes, and landscapes. This study selects six indicators at the landscape level: area ratio (R), aggregation index (AI), patch density (PD), largest patch index (LPI), edge density (ED), and landscape shape index (LSI).

Related analysis measures the correlation between numerical variables and calculates the degree of correlation. The commonly used correlation coefficients are shown in Table 3. This study uses IBM SPSS Statistics 25 to analyze the correlation between landscape pattern index and NPS pollution load. Since individual indicators do not satisfy the normal distribution, spearman correlation analysis is selected.

Table 3 The applicable conditions of the three correlation coefficients

First of all, it is necessary to perform detrended correspondence analysis (DCA) on water quality data and then determine whether to choose linear model or unimodal model (Wu and Lu [28]). If the maximum gradient of the four axes is less than 3, it is more appropriate to choose the unimodal model (canonical correspondence analysis, CCA); if it is less than 3, the liner model (redundancy analysis, RDA) should be selected; if it is between 3 and 4, then both are reasonable (Petr and Jan [26]).

Since the maximum gradient of the four axes is less than 3, Canoco 5.0 is used to analyze the relationship between landscape pattern and NPS pollution process by RDA. This method takes 27 sub-basins as samples, takes NPS pollution loads of each sub-basin as species variable and the landscape pattern indexes at landscape level as environmental factor.

The warm-up period of the model is set as 2 years (2008 and 2009), and the simulation period is set from 2010 to 2016. The FLOW_OUT, SED_OUT, NH4-N_OUT, TN_OUT, and TP_OUT fields in the Rch file in SWAT Output are selected for parameter calibration. The runoff and sediment are calibrated using regular monthly data from 2010 to 2012, and the verification period is from 2013 to 2014 and 2016. The water quality is calibrated in 2015, and verification period is 2016.

When R2>0.6 and ENS>0.5, the results are acceptable. If the Re of runoff is ±25% and the Re of sediment and pollutants is ±40%, the simulation results are satisfied (Moriasi et al. [23]; Li and Fang [17]). The calibration and verification results are shown in Fig. 3 and Table 4, which show that the three evaluation indicators are within the parameter range. Therefore, the SWAT model is suitable for the Bahe River Basin.

Graph: Fig. 3 Comparison of the observed and simulated for monthly calibration and verification: (a) Runoff; (b) Sediment; (c) NH4-N; (d) TN; (e) TP

Table 4 Applicability evaluation of SWAT model

The distribution of NPS pollution has a strong spatial character, and it is closely related to rainfall, soil properties, land use type, and topography distribution in the study area. Based on the output files of 27 sub-basins from 2010 to 2016, this paper extracts PRECIP (rainfall), WYLD (runoff), SYLD (sediment), ORGN (organic nitrogen), NSURQ (nitrate), ORGP (organic phosphorus), SEDP (mineral phosphorus), and SOLP (soluble phosphorus) in the Sub file in SWATOutput and links the parameter values in ArcGIS to study the spatial distribution characteristics of NPS pollution loads from 2010 to 2016.

Figure 4(a) shows that the spatial distribution of rainfall in the basin decreases with the topography from south to north. The area with the highest rainfall is mainly in the upper reaches of the basin, including Wangchuan, Lanqiao, Muhuguan, and Gepaizhen, with an average rainfall of about 640 mm. The areas with the least rainfall are mainly concentrated in Xiqu and Maduwang in Baqiao District, with an average annual precipitation of about 440 mm. By comparing Fig. 4(a) and Fig. 4(b), it is found that there is a positive correlation between runoff and rainfall in the basin. The rainfall of sub-basins 24, 25, 27 is large and the slope is above 20°, so the corresponding runoff is also large. As can be seen from Fig. 4(c), the distribution of sediment in the Bahe River basin is extremely uneven. There is a certain similarity between the distribution of sediment and runoff, but there are some differences. For example, in sub-basins 3 and 12, rainfall and runoff are low, while the sediment output is large. The main reason is that the soil type in this area is mainly cinnamon soil, which is highly erodible. The spatial distribution of sediment output is related to soil type, rainfall, and other factors.

Graph: Fig. 4 Spatial distribution map of annual NPS pollution output: (a) PRECIP; (b) WYLD; (c) SYLD (d) ORGN; (e) NSURQ; (f) ORGP; (g) SEDP; (h) SOLP

The output of ORGN load is related to the content of organic nitrogen in soil, soil erodibility, and rainfall intensity. The distribution of ORGN loss intensity is basically the same as that of sediment (Fig. 4(d)), and it is mainly concentrated in the sub-basin 3, 12, 14, 19, and 24. These sub-basins are mainly dominated by farmland with large amount of fertilizer. Therefore, the nitrogen content entering the river channel with the soil erosion is relatively high. The output of NSURQ load is related to nitrate nitrogen content and soil erosion. The loss intensity of NSURQ in sub-basins 6, 16, and 13 is relatively large (Fig. 4(e)) because of the heavy rainfall, high proportion of farmland, and the large amount of nitrate in the soil. It can be seen from Fig. 4(f) and Fig. 4(g) that the distribution of ORGP and SEDP loss intensity is basically consistent with sediment and is greatly affected by soil erosion. Phosphorus is chemically inactive, not easily soluble in water and is generally adsorbed on the surface of the soil. Therefore, under the action of rainfall runoff, ORGP and SEDP are transported into the river with sediment, while SOLP is transported into the river with surface runoff.

In conclusion, there is a certain correlation between rainfall and the loss intensity of the nitrogen and phosphorus in each sub-basin of the Bahe River Basin. However, the loss intensity of ORGN, ORGP, and SEDP is more significant correlated with sediment loss intensity and the coefficient reaches about 0.9. Therefore, controlling soil erosion is an effective way to reduce the nutrient load in the Bahe River Basin. Although there is a close relationship between the spatial heterogeneity of rainfall and the spatial difference of pollutants, rainfall is not necessarily the decisive factor (Li et al. [18]). In some sub-basins, even if the rainfall is not heavy, the loss intensity of NPS pollution load is very large under the joint action of the land use type, slope and slope length, etc.

Based on the simulation result of the SWAT model, the study uses the unit area load index method to identify the CSA in the basin (Giri et al. [13]). Taking TN and TP as indicators, the natural crack point classification method is used to divide the nitrogen and phosphorus loss intensity into five levels (Table 5). The calculation formulas of TN and TP are as follows (Eqs. (4) and (5)):

4 $$\mathit{TN}=\mathit{\text{ORGN}}+\mathit{\text{NSURQ}}+\mathit{LAT}‐Q‐{\mathit{NO}}_3+{\mathit{\text{GWNO}}}_3$$

Graph

5 $$\mathit{TP}=\mathit{\text{ORGN}}+\mathit{\text{SEDP}}+\mathit{\text{SOLP}}$$

Graph

where LAT-Q-NO3 is the lateral flow nitrate and GWNO3 is the groundwater nitrate. Unit: kg/ha.

Table 5 Classification of evaluation indexes

The sub-basins with high output of NPS pollution load per unit area are regarded as priority areas for NPS control, that is, the CSA. The classifications of loss intensity of TN and TP are shown in Fig. 5.

Graph: Fig. 5 Classification of NPS loss intensity: (a) TN; (b) TP

The loss intensity distribution of TN and TP loads in the Bahe River Basin is basically the same, mainly concentrated in the middle reaches of the basin. The output of TN and TP loads in sub-basins 3, 12, 14, 16, 19, and 24 is serious, and the area of the six sub-basins accounts for 27.76% of the whole basin. And the output of TN and TP loads accounts for 38.8 and 57.92%, respectively, which is a relatively high proportion. Therefore, this area is identified as the CSA of NPS pollution. Analysis of the reasons: (1) the land use types in this area is mainly farmland, accounting for 47.83% of the Bahe River Basin. The long-term crop cultivation and the accumulation of fertilizer application amount year by year, which aggravate the nitrogen and phosphorus pollution (Srinivas et al. [27]). (2) The livestock breeding industry is relatively developed. According to the results of the second pollution source census in Shaanxi Province, the number of large-scale farms in Guanzhong area accounts for 69% of Shaanxi Province. Most of the pollutants produced by livestock and poultry are piled directly in the open air for composting and enter the water with rainfall, causing serious NPS pollution.

The proportion of land use area in each sub-basin is shown in Fig. 6, and it is found that the composition of land use types varies greatly. In the 27 sub-basins, the proportion of farmland, forest land and grassland is 32, 35, and 21%, respectively. Urban land is mostly distributed in the lower reaches of the basin where the terrain is relatively gentle. The proportion of the water area and towns in each sub-basin is mostly below 10%, which has a low influence on the entire pollution process. Therefore, only the area of farmland (RA), forest land (RF), and grassland (RG) is considered in the subsequent statistical analysis.

Graph: Fig. 6 Proportion distribution of landscape area in the Bahe River Basin

At the landscape level (Fig. 7), the difference in AI values among sub-basins is not obvious, which indicates that the degree of landscape aggregation in the study area is comparable. However, the other four indicators (PD, LPI, ED, and LSI) are quite different among the sub-basins, which indicate that the degree of landscape fragmentation and the complexity of patch shape are good examples of the pattern research. The larger the PD value is and the smaller the LPI value is, the higher the fragmentation degree of the landscape is. The larger the ED and LSI values are, the more complex the patch shape of the landscape is (Geng et al. [12]). The differences in landscape composition and landscape index values ​​of the 27 sub-basins are evident, indicating the diversity of landscape elements and the complexity of landscape patterns. These results reflect the agricultural production pattern in northwestern China under special terrain conditions, thereby constituting the best research sample for the gradient of landscape pattern in the area.

Graph: Fig. 7 Landscape pattern metrics measured at the landscape level in 27 sub-basins: (a) PD; (b) LPI; (c) ED; (d) LSI; (e) AI

Under the influence of human activities, the land use types of the river basin are significantly different, and the output of NPS pollution load on different landscape types is also different to some extent.

There is a positive correlation between RA and the NPS pollution load. RA has a low correlation with nitrogen load but a strong correlation with phosphorus load at the significant level of P < 0.01. Most scholars believe that farmland is an important factor causing NPS pollution. This study also confirms that farmland is the main source of pollution in the basin. Due to the application of pesticides and chemical fertilizers in the agricultural planting, the residual nitrogen and phosphorus on the surface are easily lost with rainfall and runoff. In particular, phosphorus is mostly adsorbed on the surface of sediment in the form of particles and RA is significantly related to ORGP and SEDP loads (Table 6). RA is not significantly correlated with ORGN and NSURQ loads. It may be due to the livestock and poultry industry in this basin is relatively developed, and its impact on nitrogen is more obvious than that of land use. RF is positively correlated with ORGN, NSURQ, and SOLP loads but negatively correlated with ORGP and SEDP loads, and the correlation is not significant (Table 6). It is generally believed that the vegetation coverage rate of forest land is high, which can effectively prevent soil erosion and intercept NPS pollution. However, in the study area, forest land is mostly distributed in mountainous areas with slopes above 25° and is dominated by arbor forests with canopy closure greater than 30%. Under the combined action of slope, slope length, and other factors, the interception of forest land on NPS pollution may be weakened. RG in the study area is positively correlated with the ORGN, NSURQ, and SOLP loads (Table 6). The possible reason is that the average proportion of grassland in the study area is relatively small. Due to the impact of the surrounding construction land instead of the natural landscape pattern, the interception and absorption of pollutants by grassland are weakened, which increases the output of NPS pollution and covers the relationship between grassland and water quality (Xu et al. [29]). It is similar to Jiulong River Basin and Huntai River Basin in China (Huang et al. [16]; Li et al. [19]).

Table 6 Spearman correlation analysis

*, at the 0.05 level (two-tailed), the correlation is significant **, at the 0.01 level (two-tailed), the correlation is significant

As shown in Table 6, LPI and LSI have the most significant impact on NPS pollution load (P < 0.01). LSI is positively correlated with NPS pollution load, while LPI is negatively correlated with NPS pollution load. LSI represents the complexity of patch shape of different land use types, and LPI represents the fragmentation of landscape. The more complex the patch shape is, the lower the degree of landscape fragmentation is, indicating that the study area is greatly affected by human activities, the integrity of the initial landscape pattern is destroyed, and the role of dominant patches is weakened, which is more unfavorable to the control of NPS pollution. PD is significantly negatively correlated with the ORGN, NSURQ, and SOLP loads (P < 0.05), which is inconsistent with the conclusions of related studies (Liu et al. [21]). It is possibly related to the scale effect of the study area. For example, PD has a significant positive effect on TN load in the 100 m buffer zone, while a negative effect in the 200–300 m buffer zone (Chi [7]). Therefore, the impact of landscape pattern on water quality indicators of Bahe River Basin under different spatial scales still needs to be further studied. ED and AI are weakly correlated with NPS pollution load and have certain uncertainties. The reason may be that the landscape indexes do not consider the migration process of NPS pollution, which leads to the insignificant correlation between the landscape indexes and NPS pollution load (Huang et al. [16]).

In RDA analysis (Fig. 8), the cosine of the angle between the arrows approximates the correlation coefficient between the variables. The analysis shows that the result is basically consistent with the correlation obtained by spearman correlation analysis. The length of the arrow is used to compare the magnitude of the influence between variables. Therefore, LSI, LPI and PD are the key factors for the prediction of water quality in the Bahe River Basin.

Graph: Fig. 8 Landscape horizontal pattern and pollution load based on RDA

On the basis of the conclusions in Section 3.3, landscape management practices, such as returning farmland to forest land or grassland, can be implemented to reduce the NPS pollution load in the river basin. The national water source protection policy indicates that all slopes above 15° in the basin should be returned to forest land. Under the condition of meeting the drinking water quality standards, a small amount of farmland suitable for farming should be allowed in the slope below 15°, with no unused land and good vegetation coverage (Li et al. [18]). Therefore, the following eight landscape management practices are set up to simulate the reduction effect on NPS pollution load, as shown in Table 7. Table 8 presents the simulation results.

Table 7 NPS pollution control scheme

Table 8 NPS pollution control results

Table 8 shows the following:

• The reduction rate of NPS pollution load is (a)>(b), (a)>(g), (h)>(b), (c)>(d), (e)>(f), indicating that the forest land reduction effect is better than the grassland reduction effect.
• The reduction rate of NPS pollution load is (a)>(c), (b)>(d), (e)>(c), (f)>(d), indicating that the larger the area of the land use type is, the smaller the PD is, and the greater the AI is; hence the higher the reduction rate of NPS pollution is.
• The reduction rate of NPS pollution load is (h)>(g), indicating that under the condition of a certain area of returning farmland to forest land or grassland, the practices for farmland to forest land in areas with large patch areas (Fig. 9) are suitable for reducing NPS pollution.
• Among all the scenarios, the reduction rates of TN and TP load in the first scenario reaches about 25%, and the overall control effect is optimal.
• According to the statistical calculation of SWAT model simulation results, the annual average sediment transport of the study area from 2010 to 2016 is up to 706,500 t, which shows that the phosphorus load adsorbed on the sediment surface is high. Therefore, it can be predicted that the reduction rate of phosphorus load will be higher than that of nitrogen load under the practices of returning farmland to forest land. This provides a reference for the specific control of a certain type of pollutants in the river basin.

Graph: Fig. 9 Slope distribution in the Bahe River Basin

The SWAT model has good applicability in NPS pollution prediction and control. However, there are also some limitations. First, in China, due to the large topographic changes, distinct seasons, and the great influence of human activities, the parameter accuracy of the mechanism model is highly required (Hao et al. [14]). Although this paper achieves a better simulation effect by adjusting the parameters, the uncertainty of parameters and other factors still have a great impact on the simulation accuracy. Therefore, it is important to establish scientific, reasonable, and robust model parameters to ensure the credibility of model predictions and to avoid implementing faulty or ineffective watershed management (Ahmadi et al. [2]). Second, the accuracy of the model depends on the integrity of the input data. However, the databases of the model are based on the USA, and it is difficult for foreign users to find the soil type and land use type corresponding to the study area. Therefore, this paper establishes the soil type database of the Bahe River Basin. However, it is difficult to apply SWAT model in the basin with incomplete land use and soil data, and the applicability will be limited to a certain extent. Finally, the simulated objects of SWAT model are typical pollutants such as nitrogen and phosphorus. In recent years, emerging contaminants have become a hot issue for scholars at home and abroad. How to understand the migration and transformation mechanism of emerging contaminants and coupling them with SWAT model as a new module is also a direction of future research.

The spatial distribution of NPS pollution in the Bahe River Basin shows that the loss intensity of nitrogen and phosphorus loads is correlated with the distribution of rainfall and runoff to a certain extent, while the loss intensity of phosphorus load is more significantly correlated with the loss intensity of sediment. The sub-basins 3, 12, 14, 16, 19, and 24 are the CSA of NPS pollution. The landscape pattern of the river basin is closely related to the NPS pollution load, and the land use area can better predict the NPS pollution load of the watershed. Farmland, forest land and grassland are the main landscape components of the basin and the main land use types that affect the output of NPS pollution. Farmland is the main source of pollution, while forest land and grassland can effectively control the output of NPS pollution to a certain extent and the reduction effect of forest land is significantly better than that of grassland. Among the landscape indicators, LPI and LSI are significantly correlated with NPS pollution load, while ED and AI are weakly correlated with NPS pollution load. PD is negatively correlated with the output load of ORGN, NSURQ, and SOLP at the level of P < 0.05 but is not significantly correlated with the load of ORGP and SEDP. Among the eight landscape management practices, the reduction effect of returning farmland to forest land in slopes above 15° is the best, and the reduction rates of TN and TP loads have reached about 25%. Therefore, the maximum degree of returning farmland to forest land can most effectively control the NPS pollution under the condition of satisfy the natural conditions. In this study, only landscape management practices are performed in the Bahe River Basin. In related research, the management practices also involve engineering and non-engineering management practices such as fertilization, soil and water conservation, contour planting and vegetation filtration zones. Therefore, we should take into account both environmental benefits and cost-effectiveness in order to find the BMPs in the river basin (Caddis et al. [6]).

We gratefully thank all the members of the research group on Non-point Source Pollution Control and Sponge City of the State Key Laboratory of Eco-hydraulics in Northwest Arid Region of China for their efforts.

JL and JX contributed to the study conception and design. Material preparation, data collection, and analysis were performed by SL and GH. The first draft of the manuscript was written by SL, JL, and GH. All authors read and approved the final manuscript.

The study is financially supported by the key research and development project of Shaanxi Province (2019ZDLSF06-01) and the National Natural Science Foundation of China (51879215).

Not applicable

Not applicable

The datasets generated and analyzed during the current study are not publicly available due [REASON WHY DATA ARE NOT PUBLIC] but are available from the corresponding author on reasonable request.

The authors declare that they have no competing interests.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.