Risk assessment of non-point source pollution in karst reservoirs based on ‘source–sink’ landscape theory

The water pollution in reservoirs comes from both point source pollution and non-point source pollution. In the past, point source pollution caused more severe harm, manifested as a series of emergency threats to drinking water safety due to the direct discharge of pollutants from industrial wastewater, urban domestic sewage, and poultry and livestock excreta into water bodies. The point source pollution around drinking water sources can be effectively addressed through joint efforts from the government and society. However, fundamental improvement in water quality through controlling point source pollution alone is insufficient. Non-point source pollution has become the main factor threatening the safety of the surface and underground drinking water (Xia et al. 2020). Compared with point source pollution, non-point source pollution has a larger spatiotemporal scale, greater uncertainty, and a more complex composition and formation process and is more difficult to monitor (Yang et al. 2016). Therefore, the assessment and control of non-point source pollution are tough. Based on ‘China's first national pollution source survey report’ (2009), the contributions of non-point source pollution to the total nitrogen and total phosphorus in rural areas are approximately 57 and 67%, respectively (Yi et al. 2020). Nitrogen and phosphorus are the main factors affecting the eutrophication of water bodies (Jabbar & Grote 2019). Therefore, the source, migration, and transfer cost of nitrogen and phosphorus in non-point source pollutants are analyzed to effectively identify the critical pollution source areas so that limited resources can be allocated to critical source areas to achieve the maximum control effect. This process is essential to pollution control and management and is of great significance to realizing drinking water safety in cities and towns.

The mechanisms, related models, pollutant loads, and control measures of non-point source pollution have been intensely studied (Yang et al. 2016; Zhao et al. 2019; Feng et al. 2020; Wan et al. 2021; Xue et al. 2022). Past studies on non-point source pollutant loads and the loss of nitrogen and phosphorus have mainly focused on the following aspects (Wei et al. 2019; Wang et al. 2020a; Huang et al. 2022): ①Observation and experimental studies. This method requires the construction of a runoff field in a small-area watershed and regular monitoring of the non-point source pollutant load in the watershed. In general, the loss characteristics of non-point source pollutants are more realistically reflected by the measured data. Thus, the accuracy of the pollutant load estimated by this method is high. However, this method is labor intensive and expensive and requires an extended monitoring period, making it unsuitable for particular areas (Yang et al. 2016). ②Impact of landscape change on the water environment in the watershed. Many scholars believe that the landscape pattern in a watershed has a significant impact on the hydrological characteristics and water quality and that a decision-making basis for water resource management can be obtained by investigating the response relationship between the structure of a landscape and its water quality (Zhang et al. 2019; Wan et al. 2021; Wu & Lu 2021; Wang et al. 2021, 2022). For example, Shen et al. (2015) used multiple stepwise regression analysis and redundancy analysis to explore the quantitative association between landscape metrics, at both the landscape and class levels, and water quality in the highly urbanized Beiyun River Watershed. The result indicated that the landscape composition was found to explain 46.9% of the variation in water quality. Zhou et al. (2016) developed a self-organizing map (SOM)-based approach to explore the relationship between land use and water quality in the Minjiang River Watershed, Southeast China. The results demonstrated how PS pollution weakens the land use-water quality correlation. ③Non-point source pollution model. The non-point source pollution model is an objective description of the surface migration of pollutants that calculates the load of pollutants entering the watershed to thereby identify key source areas, assess the risk of pollution, and investigate the relevant mechanisms. Non-point source pollution models mainly include empirical and process-based models (Bai et al. 2020). Due to a lack of understanding about non-point source pollution in the early stage as well as a lack of monitoring data, empirical models such as the export coefficient model and the Soil Conservation Service (SCS) runoff curve number model have mainly been used. The export coefficient model is based on a ‘unit-load estimation’ approach; the core of this approach is to calculate the number of pollutants produced per unit (person, livestock, or unit land area) and thus further estimate the potential amount of non-point source pollutants generated within the study area. This model has been widely used because of its simple structure and the ease of data access (Cheng et al. 2018). For example, Wang et al. (2020b) used the dynamic export coefficient model to simulate non-point source pollution in a small catchment of the Three Gorge Reservoir Region in China and achieved satisfactory results. In addition, there are some standards integrating empirical models and statistical and deterministic methods. For example, Jiang et al. (2013) calculated the spatial distribution of non-point source pollution in the Jiulong River estuary in China based on remote sensing technology. Yang (2012) explored the quantitative relationship between landscape features and nitrogen load using a regression model. Mechanistic models are mainly used to simulate the generation and migration of non-point source pollutants and their impact on the environment. It is believed that mechanistic models can effectively address the fact that empirical models do not consider the mechanism or spatial variability (Wang et al. 2016b). Mechanistic models, such as the soil and water assessment tool (SWAT) (Shen et al. 2015), stormwater management model (SWMM) (Aryal et al. 2016), hydrological simulation program – FORTRAN (HSPF) (Stern et al. 2016), annualized agricultural non-point source (AnnAGNPS) (Wang et al. 2019), long-term hydrologic impact analysis (L-THIA) (Zhang et al. 2011), source-precipitation-landscape model (SPLM) (Cheng et al. 2019), and integrated model of non-point source pollution processes (IMPULSE) (Li et al. 2014), have a stronger explanatory power for pollutant migration and a generally high simulation accuracy.

However, the above models also have some drawbacks (Bai et al. 2020). Although the export coefficient models are simple and easy to operate, they ignore the complex migration routes of non-point source pollutants and thus have poor universality. The empirical models, generally based on statistical analysis, apply to small watersheds with a simple internal structure. However, they ignore the migration and transformation of pollutants and thus cannot explain the mechanisms. In addition, they are usually not extendable. The mechanistic models require a large volume of high-accuracy data, can obtain high-accuracy simulation through standardized calibration and verification, and are easily extendable. However, mechanistic models involve many parameters and require an in-depth understanding of the mechanisms and a large volume of data. The best ways to understand and quantitatively assess the uncertainty in the simulation results of non-point source pollution remain a research focus.

Therefore, a non-point source pollution risk assessment method was developed in this study based on the ‘source–sink’ landscape theory using Baihua Lake, a karst plateau deep-water reservoir in Guiyang City, as the study area. This method takes into account the load and transport of non-point source pollutants. First, the landscape types were classified and adjusted based on Gaofen-1 remotely sensed satellite imagery data and Google Images to obtain the ‘source–sink’ landscape distributions of the Baihua Lake watershed. Second, the digital elevation model (DEM) data (1:50,000) were used for hydrological analysis to extract the sub-watersheds of Baihua Lake, which were used as assessment units, and then the ‘source–sink’ landscape pollutant load location-weighted landscape contrast index (LCI) and non-point source pollution risk index (NSPRI) were constructed based on the nitrogen and phosphorus pollution factors of different landscape types and the factors (slope, distance from a water source, and vegetation cover) influencing the pollutant transport pathways to quantitatively identify the regions at risk for non-point source pollution in the Baihua Lake watershed. Finally, the assessment results were compared with the water quality data from sampling sites in recent years to verify the accuracy of the risk assessment results. The results of this study provide new insights for the management of non-point source pollution and the optimization of landscapes in drinking water source areas.

Baihua Lake is located at 106°27′–106°34′E and 26°35′–26°42′N. It is located between Zhuchang Township and Baihuahu Township in Guiyang City, Guizhou Province, approximately 22 km from the urban area, in the middle reaches of the Maotiao River, which is the largest tributary of the Wujiang River (Figure 1). It is a typical artificial karst plateau deep-water reservoir. The lake covers an area of approximately 14.5 km², with a total capacity of 1.82×108 m³, a maximum depth of 45 m, and a mean depth of 10.8 m. It has an elongated shape, extending along the southwest-northeast direction. The study area is in a karst mountainous region, with complex terrain, considerable topographic variation in the reservoir, and a steep watershed slope. This area has a subtropical monsoon humid climate, with an annual mean temperature of approximately 14 °C and annual precipitation of approximately 1,175 mm. The main functions of Baihua Lake include flood control, power generation, domestic water supply provision, industrial water supply provision, irrigation water supply provision, and tourist accommodation. The water in Baihua Lake mainly comes from the discharge of the upstream Hongfenghu Lake, the Nanmen River in the Baiyun District, the Maixi and Limu Rivers in the Guanshanhu District, and the Dongmenqiao and Laoma Rivers in Qingzhen City. The lake mainly supplies water to the Baiyun Waterworks, with an annual water supply of approximately 42 million tons, and is of great importance to the drinking water supply of Guiyang City.

Landsat-8 Operational Land Imager (OLI) data (3 March 2018) derived from the geospatial data platform GSCloud (http://www.gscloud.cn/) were used to determine the vegetation cover in 2018. Gaofen-1 is equipped with two 2 m Pan/8 m multispectral cameras and a four 16 m multispectral medium-resolution and wide-field camera set. Gaofen-1 satellite data (7 January 2020) were used for the classification of the landscapes. The DEM data (1:50,000) were used for sub-watershed division, water system extraction, and slope factor calculation. The water quality data came from years of actual monitoring and were used for comparative analysis of the assessment results, The sampling points are set at Maixi River (MXH), Huaqiao (HQ), Yanjiaozhai (YJZ), Guilv Waterworks (GLSC) and Dam (DB), the monitoring time is from 2017 to 2019, once a year in April, August and November. The monitored water quality indicators mainly include total phosphorus and total nitrogen. The total phosphorus water quality index is measured by flow injection-ammonium molybdate spectrophotometry, and the total nitrogen index is measured by alkaline potassium persulfate digestion UV spectrophotometry. The land use data were classified based on Gaofen-1 remotely sensed satellite imagery data and corrected by a field survey based on Google Images (Meng et al. 2018; Song et al. 2018).

The land use data were adjusted and corrected using ArcGIS software. The land use was divided into eight types, including orchard or tea plantation, construction land, dryland or irrigated land, paddy fields, forestland, unused land, water, and grassland. In total, 348 sub-watersheds were extracted by hydrological analysis of the DEM data by merging homogenous units, requiring each sub-watershed to have an area of more than 50,000 m², and ignoring small islands in the lake. The fractional vegetation cover (FVC) index was calculated using the dimidiate pixel model based on geometrically and atmospherically corrected Landsat-8 OLI data images (Ding et al. 2016).

Source landscapes are the types of landscapes that promote process development, while sink landscapes are the types of landscapes that delay or prevent process development (Wang et al. 2018). Source and sink landscapes are process specific. A source landscape of one process may be a sink landscape of another process, and the key to determining whether a landscape is a source landscape or a sink landscape is to judge whether the landscape promotes or hinders the evolution of an ecological process. Different types of source or sink landscapes make different contributions to the same ecological process. According to the ‘source–sink’ theory, during the formation of non-point source pollution, some landscape types play a source role because they generate pollutants, some landscape types play a ‘sink’ role because they absorb pollutants, and other landscape types play a role in pollutant transport. If the source and sink landscapes in a region reach equilibrium in terms of their spatial distribution and form a reasonable spatial distribution pattern, few non-point source pollutants are generated in this region. In contrast, a large number of pollutants may be generated if the spatial distribution of the landscapes in the region is unreasonable; for example, when the source landscapes are distributed close to water bodies, and the sink landscapes are distributed far from water bodies (Wang et al. 2020a).

The composition and spatial configuration of land use play an essential role in the formation and migration of pollutants. In particular, the release, absorption, and retention of pollutants by various land use types in the migration path significantly affect the water quality (Haruna et al. 2018). The forestland retains, filters, and adsorbs particles entering the river through infiltration and biological absorption of nitrogen and phosphorus by plant roots and thus improves the water quality. Therefore, the forestland can be categorized as a sink landscape that retains pollutants (Rehman et al. 2018). The grassland effectively reduces nutrient transport by slowing surface runoff and negatively correlates with water quality. Overall, the grassland exhibits a ‘sink’ effect (Sun et al. 2018). The unused land has no significant correlation with the water quality index; its influence on the water quality index is not definite and can be either positive or negative. In addition, the unused land has little effect on the overall assessment results due to its small area; it constitutes only a proportion (0.1%) of the study area and exhibits low interception efficiency and capacity for nitrogen and phosphorus pollution sources. Therefore, it is regarded as a corridor patch with no ‘source–sink’ effect. In the wet and dry seasons, areas of water have different influences on pollutants. In addition, the areas of water outside Baihua Lake are small in the study area. Therefore, the areas of water are also considered to be a corridor patch. The construction land is positively correlated with the water quality index, and the representative industrial and domestic pollution sources contribute significantly to the nitrogen and phosphorus input to the water. Therefore, the construction land is considered to be a source landscape that emits pollutants. The agricultural non-point source pollution is mainly caused by the loss of soil nutrients such as nitrogen and phosphorus in rainwater runoff on farmland surfaces. Orchard or tea plantations, paddy fields, and dryland or irrigated land may exhibit varying discharge coefficients under different farming methods and different amounts of fertilizers. They are source landscapes that emit pollutants (Cheng et al. 2019). Based on the categorization of these landscapes, the ‘source–sink’ landscape distribution pattern of the Baihua Lake watershed is ultimately determined (Figure 2).

In these formulae, and are the pollutant loads of total nitrogen and total phosphorus, respectively; i and j are the numbers of the types of source and sink landscapes, respectively; and are the weights of the total nitrogen and total phosphorus discharged by source landscape i, respectively; and are the weights of the total nitrogen absorption and total phosphorus absorption by sink landscape j, respectively; and are the proportions of the sub-watershed constituted by source landscape i and sink landscape j, respectively.

The weights of the discharge (absorption) of major pollutants (nitrogen and phosphorus) by the source landscapes and sink landscapes were calculated mainly based the nitrogen and phosphorus discharge coefficients in the First National Survey of Pollution Sources – National Handbook of Pollutant Loss Factors for Agricultural Pollution Sources, statistical yearbooks, and other relevant studies (Ongley et al. 2010; Liu et al. 2016; Zhang et al. 2018). The results are shown in Table 1.

According to the source–sink landscape theory (Jiang et al. 2013), the LCI values >0 indicate that the contribution of source landscapes outweighs sink landscapes and are likely to cause NPS pollution via nutrient discharges. Conversely, LCI values < 0 indicate a sink with balanced ecological processes and a lower NPS pollution risk. An LCI of 0 indicates a dynamic balance between pollutant export and retention. The larger is the LCI value, the greater is the intensity of the source and the smaller the intensity of the sink, and vice versa (Xue et al. 2022). The LCI values for subbasins in the study area were classified into six risk levels using the natural breakpoint method, a statistical method commonly used in studies of grading and classification which can maximize the differences between categories based on the numerical distribution (Xiao et al. 2020). Then, the resulted levels are employed to identify the spatial variations in the land use types related to the risks from NPS pollution.

In the formula, is the NSPRI of sub-watershed m; is the LCI of sub-watershed m; and are the mean slope of sub-watershed m and the maximum slope of the sub-watersheds of Baihua Lake, respectively; and are the surface distance from the geometric center of sub-watershed m to the reservoir and the surface distance from the geometric center of the sub-watershed farthest from the reservoir, respectively; and are the mean vegetation cover of sub-watershed m and the maximum mean vegetation cover of the sub-watersheds, respectively.

The spatial distribution of the source–sink pollutant load in the Baihua Lake watershed was obtained by using ArcGIS to calculate the proportions of the source and sink landscape types in the 348 sub-watersheds of Baihua Lake and using the LCI formula to calculate the pollutant load of each sub-watershed (Figure 3). The results show that there are a total of 132 sub-watersheds (37.93%) with LCI_NP values greater than 0 in the entire Baihua Lake watershed; these watersheds are distributed in Yanshanhong Township and Yeya Township in the eastern part of the Baihua Lake watershed and the Qinglong Subdistrict in the southern part of the Baihua Lake watershed, with construction land and farmland as the main landscape types. This region has a high pollutant load, especially in the Qinglong subdistrict of Qingzhen City. A large number of sub-watersheds with high pollutant loads in this region are located in the upper reaches of Baihua Lake and close to the reservoir. Therefore, the potential risk is high. There are 216 sub-watersheds (62.07%) with LCI_NP values less than 0; these watersheds are distributed in Zhanjie Township in the western part of the Baihua Lake watershed and Baihuahu Township and the northern Jinhua Township in the northwestern part of the Baihua Lake watershed.

Based on the natural breakpoint method, the LCI_NP values are categorized in descending order and can be roughly divided into six categories of extremely strong source effect, strong source effect, relatively strong source effect, relatively strong sink effect, strong sink effect, and extremely strong sink effect. There are a total of 37 sub-watersheds (10.63%) with an extremely strong source effect in the Baihua Lake watershed, with an LCI_NP range of (4.5, 8.1); these watersheds are distributed in the urban built-up area, Dongmenqiao Village, Liyutang Village, Liangjiazhai Village, Maolishan Village, Heinishao Village, and Shiguan Village in the Qinglong Subdistrict, Xiaojing Village and Jinlong Village in Yeya Township, and Chengguan Village, Xiamai Village, and Jianshan Village in the Yanshanhong Subdistrict. This region is mainly a constructed region with an aggregated population. There are a total of 62 sub-watersheds (17.82%) with strong source effects, with an LCI_NP range of (1.5, 4.5); these watersheds are mainly distributed in the eastern and southern regions with mixed landscapes of construction land and farmland, including Bianpo Village, Pingyuanshao Village, and Liangshuijing Village in the Qinglong Subdistrict, Heguan Village, Gaozhi Village, and Xiapu Village in the central Jinhua Township, Shitou Village in the eastern Zhuchang Township, and Shangmai Village and Xiamai Village in the northwestern Yeya Township. There are 33 sub-watersheds (9.48%) with a relatively strong source effect, with an LCI_NP range of (0, 1.5), which are scattered in the eastern part of the Baihua Lake watershed. There are 83 sub-watersheds (23.85%) with an extremely strong sink effect, with an LCI_NP range of (−2.5, 0), which are mainly distributed in Sanbao Village, Maoli Village, Wuliqu Forest Farm, Yungui Village, Santun Village, and Shicao Village in Baihuahu Township. There are 69 sub-watersheds (19.83%) with a relatively strong sink effect, with an LCI_NP range of (−4.5, −2.5); these watersheds are mainly distributed in Baihuahu Township and Zhanjie Township. There are a total of 64 sub-watersheds (18.39%) with a strong sink effect, with an LCI_NP range of (−6.1, −4.5). In general, the ‘source–sink’ pollutant load is higher in the eastern and southeastern parts and lower in the western and northwestern parts of the Baihua Lake watershed. This is generally consistent with the spatial distribution pattern of source and sink landscape types.

Based on the assessment results of the ‘source–sink’ non-point source pollutant load in the Baihua Lake watershed, the non-point source pollution risk distribution in the Baihua Lake watershed was obtained by considering the major factors (vegetation cover, slope, and distance from the reservoir) influencing the pollutant transport route and by calculating the NSPRI (Figure 4). The risk value of the Baihua Lake sub-watershed was categorized into six levels based on the natural breakpoint method using ArcGIS software, i.e., extremely low risk, low risk, relatively low risk, moderate risk, high risk, and extremely high risk. The area and proportions of the sub-watersheds with different risk levels were statistically analyzed (Table 2).

Overall, the distribution of regions with a risk of non-point source pollution basically overlaps with the ‘source–sink’ pollutant load distribution in the Baihua Lake watershed, with slight local differences in the influences of transport pathways on the sub-watersheds. There are 132 moderate-risk, high-risk, and extremely high-risk sub-watersheds, accounting for 37.93% of the total number of sub-watersheds. The 17 (4.89%) extremely high-risk sub-watersheds are mainly distributed in the upper reaches of Baihua Lake, including Liyutang Village and Jiancheng District of Qingzhen City. This region mainly includes residential areas for urban and rural residents and a vocational education area. Although the slope of this region is slight, the volume of discharged domestic sewage is significant due to the high population density. In addition, this region is close to water bodies, so the risk of pollutant migration is high, greatly threatening the water quality of the upstream regions of Baihua Lake. The 44 (12.64%) high-risk sub-watersheds are mainly distributed in the eastern and southern parts of the Baihua Lake watershed. This high-risk region is still dominated by construction land. The 71 (20.40%) moderate-risk sub-watersheds are mainly distributed in the Baihua Lake watershed along the northeast–southeast direction. This moderate-risk region is adjacent to the high-risk region, with mixed landscape types of construction land and farmland (the area of farmland is greater than the construction land area). The numbers of sub-watersheds at relatively low risk, low risk, and extremely low risk are 85 (24.43%), 79 (22.70%), and 52 (14.94%), respectively. The relatively low-risk, low-risk, and extremely low-risk regions are characterized by sub-watersheds with large mean slopes. However, the relatively low-risk, low-risk, and extremely low-risk regions have a low degree of human interference and high proportions of forestland and grassland, and they are dominated by sink landscapes. Therefore, the risk of non-point source pollution in these regions is small.

Most of the exogenous pollutants in Baihua Lake enter ditches and rivers through surface runoff and then enter Baihua Lake through ditches and rivers. As shown by the assessment results of the non-point source pollution risk in the Baihua Lake sub-watersheds, the extremely high-risk, high-risk, and relatively high-risk sub-watersheds are mainly distributed in the Qinglong Subdistrict, Jinhua Township, Zhuchang Township, Yeya Township, and Yanshanhong Township. The non-point source pollutants in Yeya Township, Zhuchang Township, and Yanshanhong Township are mainly discharged through the Maixi River, near the Maixi River water quality sampling site. The non-point source pollutants in the sub-watersheds in the risk zone of Jinhua Township and the Qinglong Subdistrict are discharged through the Maicheng River and the Dongmenqiao River. The outlets of the two rivers are close to the Huaqiao water quality sampling site. The non-point source pollutants in some sub-watersheds in the risk zones of Zhanjie Township are discharged through the Changchong River. The results of the non-point source pollution risk in the Baihua Lake watershed were comparatively analyzed based on the statistics of the total nitrogen and total phosphorus concentrations at five water quality sampling sites in the Maixi River, Huaqiao, Yanjiaozhai, Guilv Waterworks, and the dam over the past three years (Figures 5 and 6).

As shown by the analysis results, the concentrations of total nitrogen and total phosphorus at the Huaqiao water quality sampling site are much higher than those at other water quality sampling sites and are highest in April each year. In addition, the frequencies of the sudden occurrence of abnormal total nitrogen and total phosphorus concentrations are high at the Huaqiao water quality sampling site, indicating that this region is susceptible to sudden water pollution. The mean concentrations of total nitrogen and total phosphorus are high at the Yanjiaozhai water quality sampling site. Although the risk of non-point source pollution in the sub-watersheds in this region is not high, there are many agritainment (Nongjiale) activities in the surrounding area, and these sub-watersheds are close to the reservoir. Therefore, these sub-watersheds have an enormous impact on the total nitrogen and total phosphorus concentrations. The mean values of the total nitrogen and total phosphorus concentrations in the Maixi River water quality sampling site are ranked fourth, and the mean values of the total nitrogen and total phosphorus concentrations in the dam water quality sampling site are ranked fifth. The ranking of the mean values of the total nitrogen and total phosphorus concentrations at the five water quality sampling sites is similar to that based on the results of the non-point source pollution risk assessment, but there is still a certain difference between the two.

The water quality monitoring results at the Huaqiao water quality sampling site are the same as those of the risk assessment, and the mean values of total nitrogen and total phosphorus concentrations are both highest at this location. In a field survey, it was observed that the water at the Huaqiao sampling site mainly comes from the upstream Changchong River, Dongmenqiao River, and Limu River and that the downstream water of the Changchong River is dark and smells terrible, with slow water flow and garbage floating on the water. The urban construction in the upper reaches of the Changchong River is intense. Although there is the Yunmeng Xiaozhen sewage treatment plant, the sewage discharged from other upstream areas is not treated. There is a wetland park in the lower reaches of the Dongmenqiao River and a sewage treatment plant near the outlet of the Dongmenqiao River, where the water flow is low, and there is a large amount of duckweed and Alternanthera philoxeroides (Mart.) covering the water surface, indicating that the water body in this area is rich in nutrients. The water downstream of the Limu River is yellowish-green, with a large amount of duckweed and household garbage on the surface, as well as farmland in some of the surrounding areas. The domestic sewage from the upstream rural settlements and the surface runoff from the farmland converges in the downstream area of the Limu River. The monitoring results at the dam water quality sampling site are consistent with the assessment results, the mean values of total nitrogen and total phosphorus concentrations are low, and there are no algae or duckweed on the water surface.

The assessment results for the Yanjiaozhai water quality sampling site are not consistent with the monitoring results at the Yanjiaozhai water quality sampling site. In theory, the sub-watersheds upstream of the Yanjiaozhai water quality sampling site are mostly forestland with a high degree of vegetation cover, with an apparent ‘sink’ effect, but the mean values of the total nitrogen and total phosphorus concentrations in this area are high. In contrast, a field survey showed many agritainment (Nongjiale) activities in the surrounding area and that these sub-watersheds are close to the waters. Although there are sewage treatment plants and sewage treatment stations in villages along the lake, not all domestic sewage is collected. Therefore, the sewage treatment is not very effective. Relatively high-risk and high-risk sub-watersheds, including Yeya Township, Yanshanhong Township, and Zhuchang Township, are upstream of the Maixi River water quality sampling site. Theoretically, the total nitrogen and total phosphorus concentrations at the water quality sampling site are high. However, the mean values of the measured total nitrogen and total phosphorus concentrations are not high. A field survey showed that the water surface of the lower reaches of Maixi River is relatively broad and that vegetation grows in the water. Therefore, the downstream lake surface can absorb and precipitate the pollutants (Huang et al. 2017). Guilv Waterworks is in the first-grade water source protection area in Baihua Lake, many residents have moved out of this area, and the pollution is strictly controlled. In theory, the impact of domestic sewage is small. However, the mean values of monitored total nitrogen and total phosphorus concentrations are not small. The specific causes need to be further investigated.

First, for regions with a relatively high risk, high risk, and extremely high risk for non-point source pollution, we should adjust the farming methods, strengthen the treatment and management of discharged domestic sewage to the channels and banks of the Changchong River, Dongmenqiao River, Limu River, and Maixi River, and appropriately increase the area of sink landscapes. Additionally, we can introduce plants that are effective in the retention and absorption of nitrogen and phosphorus, such as Eichhornia crassipes, Phragmites australis, and Typha orientalis (Lu et al. 2018), to reduce nitrogen and phosphorus inputs to the reservoir and construct wetlands and centralized sewage treatment systems to strictly control the pollution that enters the lake (Nandakumar et al. 2019). Second, the collection and centralized treatment and management of discharged domestic sewage should be strengthened in regions near Baihua Lake, where there are many agritainment activities and densely distributed villages. Vegetation buffer zones or forest and grass filter strips should be established along the shore to filter and retain inflows from farmland, and farming on sloped land near the reservoir's shore should be prohibited (Qiu et al. 2019).

From the perspective of the ‘source–sink’ landscape theory, based on the differences in the discharge coefficients between different landscape types, this paper used the LCI to identify the pollutant loads of total nitrogen and total phosphorus in each sub-watershed of Baihua Lake and used the NSPRI to assess the risk of pollution in the Baihua Lake watershed by considering the factors influencing transport such as vegetation cover, distance from the reservoir, and slope. The high reliability of the assessment results was confirmed by comparing them with the measured water quality data and field survey results. Therefore, this method is worthy of promotion. More influencing factors should be considered in future studies to obtain more accurate assessment results. For example, the types of landscapes near the reservoir have prominent influences on water quality, so they can be assessed separately. Additionally, although the volume of discharged sewage in urban built-up areas is high, there are sewage treatment plants in most of these areas. In contrast, the scattered distribution of sewage discharge sites in rural residential areas makes it more challenging to manage sewage in these areas. Therefore, the pollutant discharge coefficients of urban built-up areas and rural residential areas should be considered separately. In addition, whether the concentration of pollutants decreases with increasing transport distance needs to be further investigated. In future studies, more water quality sampling sites will be added, and more influencing factors will be considered to make the assessment model more comprehensive and applicable.

• (1)

  The pollutant load in the Baihua Lake watershed was quantitatively identified in this paper based on the ‘source–sink’ landscape theory by using 348 natural sub-watersheds as the basic units. The NSPRI was constructed to assess the pollution risk in the Baihua Lake sub-watershed. The reliability and accuracy of this method were verified by comparing the assessment results with the measured water quality data and field survey results.

• (2)

  A total of 17 sub-watersheds at extremely high risk for non-point source pollution and that require close attention are mainly distributed in the Qinglong Subdistrict and Zhanjie Township in the upper reaches of Baihua Lake, where construction land and farmland are the dominant landscapes.

• (3)

  The scattered rural settlements along the shore of the reservoir have a large impact on the water quality of the reservoir. Therefore, it is necessary to strengthen the management and control of pollutant discharge and farming management along the reservoir's shore.

• (4)

  The rivers corresponding to relatively high-risk, high-risk, and extremely high-risk sub-watersheds are the Dongmenqiao River, Limu River, Changchong River, and Maixi River. The government should actively adopt effective measures targeting these key tributaries, such as farming management, the collection and centralized management of sewage in residential areas, river dredging, and the construction of wetlands.

We would like to thank the anonymous reviewers of the manuscript. This research was supported by the Natural Science Foundation of Guizhou Province of China (Grant numbers [2018]1418), the Guizhou Provincial Science and Technology Projects (Grant numbers [2018]2806, and [2020]4Y132), the National Natural Science Foundation of China (Grant numbers [U1612441]), the Guizhou High-Level Innovative Talents Training Program – Hundred Talent Level (Grant numbers [2016]5674), and the Youth Fund of the Guizhou Academy of Sciences (Grant numbers Qiankeyuan J [2018]11 and [2021]25).

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.