Volume 32 Issue 4
Jul.  2022
Turn off MathJax
Article Contents

SHI Jinhao, JIN Ri, ZHU Weihong, 2022. Quantification of Effects of Natural Geographical Factors and Landscape Patterns on Non-point Source Pollution in Watershed Based on Geodetector: Burhatong River Basin, Northeast China as An Example. Chinese Geographical Science, 32(4): 707−723 doi:  10.1007/s11769-022-1295-z
Citation: SHI Jinhao, JIN Ri, ZHU Weihong, 2022. Quantification of Effects of Natural Geographical Factors and Landscape Patterns on Non-point Source Pollution in Watershed Based on Geodetector: Burhatong River Basin, Northeast China as An Example. Chinese Geographical Science, 32(4): 707−723 doi:  10.1007/s11769-022-1295-z

Quantification of Effects of Natural Geographical Factors and Landscape Patterns on Non-point Source Pollution in Watershed Based on Geodetector: Burhatong River Basin, Northeast China as An Example

doi: 10.1007/s11769-022-1295-z
Funds:  Under the auspices of the National Key R&D Program (No. 2019YFC0409104), the National Natural Science Foundation of China (No. 41830643), the National Science and Technology Basic Resources Survey Project (No. 2019FY101703)
More Information
  • Corresponding author: JIN Ri. E-mail: jinri0322@ybu.edu.cn; ZHU Weihong. E-mail: whzhu@ybu.edu.cn
  • Received Date: 2021-10-21
  • Accepted Date: 2022-01-23
  • Rev Recd Date: 2021-12-23
  • Publish Date: 2022-07-05
  • Changes in natural geographic features and landscape patterns directly influence the hydrology and non-point source pollution processes in the watershed; however, to slow down non-point source pollution, it is necessary to distinguish their effects. But the non-point source pollution process is interactional as a result of multiple factors, and the collinearity between multiple independent variables limits our ability of reason diagnosis. Thus, taking the Burhatong River Basin, Northeast China as an example, the methods of hydrological simulation, geographic detectors, and redundancy analysis have been combined to determine the impact of natural geographic features and landscape patterns on non-point source pollution in the watershed. The Soil & Water Assessment Tool (SWAT) has been adopted to simulate the spatial and temporal distribution characteristics of total nitrogen and total phosphorus in the watershed. The results show that the proportions of agricultural land and forest area and the location-weighted landscape contrast index (LWLI) are the main indicators influencing the rivers total nitrogen and total phosphorus. The interaction of these indicators with natural geographic features and landscape configuration indicators also significantly influences the changes in total nitrogen (TN) and total phosphorus (TP). Natural geographical features and landscape patterns have different comprehensive effects on non-point source pollution in the dry and wet seasons. TN and TP loads are affected mainly by the change in landscape pattern, especially in the wet season. Although the ecological restoration program has improved forest coverage, the purification effect of increased forest coverage on the water quality in the watershed may be offset by the negative impact of increased forest fragmentation. The high concentration and complexity of farmland patches increase the risk of non-point source pollution spread to a certain extent.
  • 加载中
  • [1] Ai L, Shi Z H, Yin W et al., 2015. Spatial and seasonal patterns in stream water contamination across mountainous watersheds: linkage with landscape characteristics. Journal of Hydrology, 523: 398–408. doi:  10.1016/j.jhydrol.2015.01.082
    [2] Alberti M, Booth D, Hill K et al., 2007. The impact of urban patterns on aquatic ecosystems: an empirical analysis in Puget lowland sub-basins. Landscape and Urban Planning, 80(4): 345–361. doi:  10.1016/j.landurbplan.2006.08.001
    [3] Alnahit A O, Mishra A K, Khan A A, 2020. Quantifying climate, streamflow, and watershed control on water quality across Southeastern US watersheds. Science of the Total Environment, 739: 139945. doi:  10.1016/j.scitotenv.2020.139945
    [4] Arnold J G, Allen P M, 1996. Estimating hydrologic budgets for three Illinois watersheds. Journal of Hydrology, 176(1-4): 57–77. doi:  10.1016/0022-1694(95)02782-3
    [5] Cai Y M, Feng M Q, Zhang T, 2021. Review of distribution of nitrogen and phosphorus in riparian zones of Chinese inland water bodies. Acta Ecologica Sinica. in press. doi: 10.1016/j. chnaes. 2021.09. 010
    [6] Chen Liding, Tian Hiuying, Fu Bojie et al., 2009. Development of a new index for integrating landscape patterns with ecological processes at watershed scale. Chinese Geographical Science, 19(1): 37–45. doi:  10.1007/s11769-009-0037-9
    [7] Clément F, Ruiz J, Rodríguez M A et al., 2017. Landscape diversity and forest edge density regulate stream water quality in agricultural catchments. Ecological Indicators, 72: 627–639. doi:  10.1016/j.ecolind.2016.09.001
    [8] de Mello K, Taniwaki R H, de Paula F R et al., 2020. Multiscale land use impacts on water quality: assessment, planning, and future perspectives in Brazil. Journal of Environmental Management, 270: 110879. doi:  10.1016/j.jenvman.2020.110879
    [9] Fan Y P, Fang C L, 2020. A comprehensive insight into water pollution and driving forces in Western China—case study of Qinghai. Journal of Cleaner Production, 274: 123950. doi:  10.1016/j.jclepro.2020.123950
    [10] Gao J B, Wang H, 2019. Temporal analysis on quantitative attribution of karst soil erosion: a case study of a peak-cluster depression basin in Southwest China. Catena, 172: 369–377. doi:  10.1016/j.catena.2018.08.035
    [11] Gémesi Z, Downing J A, Cruse R M et al., 2011. Effects of watershed configuration and composition on downstream lake water quality. Journal of Environmental Quality, 40(2): 517–527. doi:  10.2134/jeq2010.0133
    [12] Ierodiaconou D, Laurenson L, Leblanc M et al., 2005. The consequences of land use change on nutrient exports: a regional scale assessment in south-west Victoria, Australia. Journal of Environmental Management, 74(4): 305–316. doi:  10.1016/j.jenvman.2004.09.010
    [13] Jin Shuqin, Wu Yan, 2014. Is agricultural non-point source the primary cause of water pollution?—a test based on data from the Huai River Basin. China Rural Economy, (9): 71–81. (in Chinese)
    [14] Khatri N, Tyagi S, 2015. Influences of natural and anthropogenic factors on surface and groundwater quality in rural and urban areas. Frontiers in Life Science, 8(1): 23–39. doi:  10.1080/21553769.2014.933716
    [15] Lee S W, Hwang S J, Lee S B et al., 2009. Landscape ecological approach to the relationships of land use patterns in watersheds to water quality characteristics. Landscape and Urban Planning, 92(2): 80–89. doi:  10.1016/j.landurbplan.2009.02.008
    [16] Leip A, Billen G, Garnier J et al., 2015. Impacts of European livestock production: nitrogen, sulphur, phosphorus and greenhouse gas emissions, land-use, water eutrophication and biodiversity. Environmental Research Letters, 10(11): 115004. doi:  10.1088/1748-9326/10/11/115004
    [17] Li C W, Zhang H Y, Hao Y H et al., 2020. Characterizing the heterogeneous correlations between the landscape patterns and seasonal variations of total nitrogen and total phosphorus in a peri-urban watershed. Environmental Science and Pollution Research, 27(27): 34067–34077. doi:  10.1007/s11356-020-09441-5
    [18] Li L, Gou M M, Wang N et al., 2021a. Landscape configuration mediates hydrology and nonpoint source pollution under climate change and agricultural expansion. Ecological Indicators, 129: 107959. doi:  10.1016/j.ecolind.2021.107959
    [19] Li W H, Cheng X J, Zheng Y et al., 2021b. Response of non-point source pollution to landscape pattern: case study in mountain-rural region, China. Environmental Science and Pollution Research, 28(13): 16602–16615. doi:  10.1007/s11356-020-12196-8
    [20] Liu J, Shen Z Y, Chen L, 2018. Assessing how spatial variations of land use pattern affect water quality across a typical urbanized watershed in Beijing, China. Landscape and Urban Planning, 176: 51–63. doi:  10.1016/j.landurbplan.2018.04.006
    [21] Liu X W, Wang H Y, Zhou J M et al., 2016. Effect of N fertilization pattern on rice yield, N use efficiency and fertilizer–N fate in the Yangtze River Basin. China. PLoS One, 11(11): e0166002. doi:  10.1371/journal.pone.0166002
    [22] Liu Yi’na, Kong Lingqiao, Xiao Yi et al., 2019. Relationships between landscape pattern and ecosystem water purification service in the Yangtze River Basin. Acta Ecologica Sinica, 39(3): 844–852. (in Chinese)
    [23] Nash J E, Sutcliffe J V, 1970. River flow forecasting through conceptual models part I—a discussion of principles. Journal of hydrology, 10(3): 282–290. doi:  10.1016/0022-1694(70)90255-6
    [24] Ouyang W, Skidmore A K, Toxopeus A G et al., 2010. Long-term vegetation landscape pattern with non-point source nutrient pollution in upper stream of Yellow River basin. Journal of Hydrology, 389(3−4): 373–380. doi:  10.1016/j.jhydrol.2010.06.020
    [25] Ouyang W, Jiao W, Li X M et al., 2016. Long-term agricultural non-point source pollution loading dynamics and correlation with outlet sediment geochemistry. Journal of Hydrology, 540: 379–385. doi:  10.1016/j.jhydrol.2016.06.043
    [26] Pérez-Gutiérrez J D, Paz J O, Tagert M L M, 2017. Seasonal water quality changes in on-farm water storage systems in a south-central U. S. agricultural watershed. Agricultural Water Management, 187: 131–139. doi:  10.1016/j.agwat.2017.03.014
    [27] Prasad R, Hochmuth G J, 2016. Environmental nitrogen losses from commercial crop production systems in the Suwannee River basin of Florida. PLoS One, 11(12): e0167558. doi:  10.1371/journal.pone.0167558
    [28] Rabalais N N, Turner R E, Díaz R J et al., 2009. Global change and eutrophication of coastal waters. ICES Journal of Marine Science, 66(7): 1528–1537. doi:  10.1093/icesjms/fsp047
    [29] Ricart A M, Dalmau A, Pérez M et al., 2015. Effects of landscape configuration on the exchange of materials in seagrass ecosystems. Marine Ecology Progress Series, 532: 89–100. doi:  10.3354/meps11384
    [30] Ritter L, Solomon K, Sibley P et al., 2002. Sources, pathways, and relative risks of contaminants in surface water and groundwater: a perspective prepared for the Walkerton inquiry. Journal of Toxicology and Environmental Health, Part A, 65(1): 1–142. doi:  10.1080/152873902753338572
    [31] Rodrigues V, Estrany J, Ranzini M et al. , 2018. Effects of land use and seasonality on stream water quality in a small tropical catchment: the headwater of Córrego Água Limpa, São Paulo (Brazil). Science of the Total Environment, 622–623: 1553–1561. doi:  10.1016/j.scitotenv.2017.10.028
    [32] Roth N E, Allan J D, Erickson D L, 1996. Landscape influences on stream biotic integrity assessed at multiple spatial scales. Landscape Ecology, 11(3): 141–156. doi:  10.1007/BF02447513
    [33] Rutledge J M, Chow-Fraser P, 2019. Landscape characteristics driving spatial variation in total phosphorus and sediment loading from sub-watersheds of the Nottawasaga River, Ontario. Journal of Environmental Management, 234: 357–366. doi:  10.1016/j.jenvman.2018.12.114
    [34] Santhi C, Arnold J G, Williams J R et al., 2001. Validation of the SWAT model on a large river basin with point and nonpoint sources. Journal of the American Water Resources Association, 37(5): 1169–1188. doi:  10.1111/j.1752-1688.2001.tb03630.x
    [35] Shen Z Y, Hou X S, Li W et al., 2014. Relating landscape characteristics to non-point source pollution in a typical urbanized watershed in the municipality of Beijing. Landscape and Urban Planning, 123: 96–107. doi:  10.1016/j.landurbplan.2013.12.007
    [36] Shi Jinhao, Zhu Weihong, Tian Le et al., 2021. Using SWAP to study the changes in nonpoint source pollution in Burhatong River Basin. Journal of Irrigation and Drainage, 40(4): 130–136. (in Chinese)
    [37] Shrestha S, Bhatta B, Shrestha M et al., 2018. Integrated assessment of the climate and landuse change impact on hydrology and water quality in the Songkhram River Basin, Thailand. Science of the Total Environment, 643: 1610–1622. doi:  10.1016/j.scitotenv.2018.06.306
    [38] Singh A P, Ghosh S K, Sharma P, 2007. Water quality management of a stretch of river Yamuna: an interactive fuzzy multi-objective approach. Water Resources Management, 21(2): 515–532. doi:  10.1007/s11269-006-9028-0
    [39] Sun Y W, Guo Q H, Liu J et al., 2014. Scale effects on spatially varying relationships between urban landscape patterns and water quality. Environmental Management, 54(2): 272–287. doi:  10.1007/s00267-014-0287-x
    [40] Tang J X, Li Y M, Cui S H et al., 2020. Linking land-use change, landscape patterns, and ecosystem services in a coastal watershed of southeastern China. Global Ecology and Conservation, 23: e01177. doi:  10.1016/j.gecco.2020.e01177
    [41] ter Braak C J F, Smilauer P, 2012. Canoco Reference Manual and User’s Guide: Software for Ordination, Version 5.0. Ithaca USA: Microcomputer Power.
    [42] Uuemaa E, Roosaare J, Mander Ü, 2005. Scale dependence of landscape metrics and their indicatory value for nutrient and organic matter losses from catchments. Ecological Indicators, 5(4): 350–369. doi:  10.1016/j.ecolind.2005.03.009
    [43] Vilmin L, Mogollón J M, Beusen A H W et al., 2018. Forms and subannual variability of nitrogen and phosphorus loading to global river networks over the 20th century. Global and Planetary Change, 163: 67–85. doi:  10.1016/j.gloplacha.2018.02.007
    [44] Wang Cheng, Guo Zhongling, 2006. Analysis on land-use pattern in Buerhatong River Basin. Chinese Journal of Eco-Agriculture, 14(4): 231–234. (in Chinese)
    [45] Wang G Q, Wu B B, Zhang L et al., 2014. Role of soil erodibility in affecting available nitrogen and phosphorus losses under simulated rainfall. Journal of Hydrology, 514: 180–191. doi:  10.1016/j.jhydrol.2014.04.028
    [46] Wang Jie, Li Peng, Gao Haidong et al., 2018. Preliminary study on the relationship between land use/landscape index and water quality in the upper reaches of Danjiang River. Research of Soil and Water Conservation, 25(6): 383–389. (in Chinese)
    [47] Wang Jinfeng, Xu Chengdong, 2017. Geodetector: principle and prospective. Acta Geographica Sinica, 72(1): 116–134. (in Chinese)
    [48] Wang X Y, 2006. Management of agricultural nonpoint source pollution in China: current status and challenges. Water Science and Technology, 53(2): 1–9. doi:  10.2166/wst.2006.033
    [49] Wu J H, Lu J, 2019. Landscape patterns regulate non-point source nutrient pollution in an agricultural watershed. Science of the Total Environment, 669: 377–388. doi:  10.1016/j.scitotenv.2019.03.014
    [50] Wu J H, Lu J, 2021. Spatial scale effects of landscape metrics on stream water quality and their seasonal changes. Water Research, 191: 116811. doi:  10.1016/j.watres.2021.116811
    [51] Xiang W N, 1996. GIS-based riparian buffer analysis: injecting geographic information into landscape planning. Landscape and Urban Planning, 34(1): 1–10. doi:  10.1016/0169-2046(95)00206-5
    [52] Yanbian Bureau of Statistics, 2017. Yanbian Statistical Yearbook. Yanji: Yanbian Bureau of Statistics. (in Chinese)
    [53] Yang Linzhang, Wu Yonghong, 2018. Agricultural non-point source pollution prevention and control and water environmental protection. Proceedings of the Chinese Academy of Sciences, 33(2): 168–176. (in Chinese)
    [54] Zhang J, Li S Y, Dong R Z et al., 2019. Influences of land use metrics at multi-spatial scales on seasonal water quality: a case study of river systems in the Three Gorges Reservoir Area, China. Journal of Cleaner Production, 206: 76–85. doi:  10.1016/j.jclepro.2018.09.179
    [55] Zhao T Q, Xu H S, He Y X et al., 2009. Agricultural non-point nitrogen pollution control function of different vegetation types in riparian wetlands: a case study in the Yellow River wetland in China. Journal of Environmental Sciences, 21(7): 933–939. doi:  10.1016/S1001-0742(08)62364-5
    [56] Zhou Chenghu, Cheng Weiming, Qian Jinkai et al., 2009. Research on the classification system of digital land geomorphology of 1: 1000000 in China. Journal of Geo-Information Science, 11(6): 707–724. (in Chinese)
  • 加载中
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Figures(6)  / Tables(4)

Article Metrics

Article views(374) PDF downloads(48) Cited by()

Proportional views
Related

Quantification of Effects of Natural Geographical Factors and Landscape Patterns on Non-point Source Pollution in Watershed Based on Geodetector: Burhatong River Basin, Northeast China as An Example

doi: 10.1007/s11769-022-1295-z
Funds:  Under the auspices of the National Key R&D Program (No. 2019YFC0409104), the National Natural Science Foundation of China (No. 41830643), the National Science and Technology Basic Resources Survey Project (No. 2019FY101703)
    Corresponding author: JIN Ri. E-mail: jinri0322@ybu.edu.cn;  ZHU Weihong. E-mail: whzhu@ybu.edu.cn

Abstract: Changes in natural geographic features and landscape patterns directly influence the hydrology and non-point source pollution processes in the watershed; however, to slow down non-point source pollution, it is necessary to distinguish their effects. But the non-point source pollution process is interactional as a result of multiple factors, and the collinearity between multiple independent variables limits our ability of reason diagnosis. Thus, taking the Burhatong River Basin, Northeast China as an example, the methods of hydrological simulation, geographic detectors, and redundancy analysis have been combined to determine the impact of natural geographic features and landscape patterns on non-point source pollution in the watershed. The Soil & Water Assessment Tool (SWAT) has been adopted to simulate the spatial and temporal distribution characteristics of total nitrogen and total phosphorus in the watershed. The results show that the proportions of agricultural land and forest area and the location-weighted landscape contrast index (LWLI) are the main indicators influencing the rivers total nitrogen and total phosphorus. The interaction of these indicators with natural geographic features and landscape configuration indicators also significantly influences the changes in total nitrogen (TN) and total phosphorus (TP). Natural geographical features and landscape patterns have different comprehensive effects on non-point source pollution in the dry and wet seasons. TN and TP loads are affected mainly by the change in landscape pattern, especially in the wet season. Although the ecological restoration program has improved forest coverage, the purification effect of increased forest coverage on the water quality in the watershed may be offset by the negative impact of increased forest fragmentation. The high concentration and complexity of farmland patches increase the risk of non-point source pollution spread to a certain extent.

SHI Jinhao, JIN Ri, ZHU Weihong, 2022. Quantification of Effects of Natural Geographical Factors and Landscape Patterns on Non-point Source Pollution in Watershed Based on Geodetector: Burhatong River Basin, Northeast China as An Example. Chinese Geographical Science, 32(4): 707−723 doi:  10.1007/s11769-022-1295-z
Citation: SHI Jinhao, JIN Ri, ZHU Weihong, 2022. Quantification of Effects of Natural Geographical Factors and Landscape Patterns on Non-point Source Pollution in Watershed Based on Geodetector: Burhatong River Basin, Northeast China as An Example. Chinese Geographical Science, 32(4): 707−723 doi:  10.1007/s11769-022-1295-z
    • In recent years, the rapid development of urbanization and industrialization has brought not only economic benefits, but also wide resource shortages and serious environmental problems; water environmental problems have become one of the most serious and common problems (Alberti et al., 2007; Fan and Fang, 2020). Among them, point source pollution and non-point source pollution (NPS) have a particularly prominent impact on the water environment (Singh et al., 2007; Chen et al., 2009). As the largest proportion and most widely distributed freshwater resources of the world’s multiple water sources, rivers and lakes are threatened by various forms of pollution. The discharge of industrial wastewater and domestic sewage has frequently proved the main factor of point source pollution (Singh et al., 2007; Pérez-Gutiérrez et al., 2017; Shrestha et al., 2018). However, compared with point source pollution, the water environment pollution caused by NPS has the space universality, time uncertainty, and delay (Ai et al., 2015; Rodrigues et al., 2018; Tang et al., 2020). The unique character of NPS makes it more difficult to identify, monitor, and evaluate (Li et al., 2021b). NPS causes can be attributed to the interaction of natural and human factors such as climate change, vegetation characteristics, land use, and pollutant discharge (Ai et al., 2015; Rutledge and Chow-Fraser, 2019). On one hand, NPS pollution is caused by rainfall or snowmelt generating runoff on the ground (Jin and Wu, 2014); as runoff moves, it will take away natural and artificial pollutants and eventually deposit them in lakes and rivers, wetlands, coastal waters, and groundwater systems. During the process, topography, vegetation, and soil have an important impact on NPS pollution (Gao and Wang, 2019; Wu and Lu, 2021). When the topographic slope is large, along with the rainfall, soil particles, nitrogen and phosphorus pollutants, and heavy metals are more likely to be lost or washed into the water and are thereby harmful to water quality (Ritter et al., 2002). Different vegetations have different absorbing capacities for pollutants (Zhao et al., 2009); and forest and grassland can reduce the entry of pollutants from runoff into water, having a purifying effect (Cai et al., 2021). The higher the degree of ground hardening, the easier for pollutants to be lost with rainfall and surface runoff (Li et al., 2021a), and these common physiographic factors have different degrees of influence on NPS in different watersheds. On the other hand, frequent human activities may also cause nitrogen and phosphorus to be discharged into the water environment, leading to eutrophication and deterioration of water quality (Rabalais et al., 2009; Leip et al., 2015). For example, the significant increase in crop fertilization raised the risk of nitrogen load and aggravated NPS pollution in the watershed (Liu et al., 2016; Prasad and Hochmuth, 2016).

      The production and use of chemical fertilizers in China ranks first globally, but the use efficiency of chemical fertilizers and pesticides is low (Wang, 2006). Most of the residual chemical fertilizers and pesticides enter the water body through surface runoff under the action of rainfall and irrigation, a critical factor in NPS. Rural sewage includes rural domestic sewage and rural production wastewater. The untreated direct discharge of rural domestic sewage and production wastewater is also the main cause of agricultural NPS. Through lack of management and planning, most rural areas have no sewage collection and treatment system and no garbage collection and treatment system. This scattered domestic sewage or landfill leachate directly enters rivers and farmland ecosystems, forming a large-scale, low-concentration NPS load (Yang and Wu, 2018). Residual bait and excrement produced in livestock and poultry farming and aquaculture are also the main types of agricultural non-point source pollutants. Livestock and poultry manure has become a major polluter. Two-thirds of China’s large-scale farming still lacks anti-pollution facilities. About 3.8 billion t of livestock and poultry manure are produced every year, and the treatment rate is less than 50%, of which 1/4 of the total amount entering water bodies is an important source of NPS (Yang and Wu, 2018).

      Moreover, population growth and its related urbanization and socioeconomic development lead to land use changes, which influences the hydrologic process, leading to changes in the form and intensity of NPS pollution (Ierodiaconou et al., 2005; Ouyang et al., 2016). The landscape pattern includes mainly landscape composition (proportion of different land use modes) and landscape configuration (spatial structure of different land use) (Wu and Lu, 2021), which determine the storage, exchange, and movement of materials among different landscape patches and affect the supply of ecosystem services (Shen et al., 2014). Rivers are also an essential part of the watershed landscape, and NPS pollution changes are closely relevant to the watershed landscape pattern. Many scholars have confirmed the importance of landscape pattern changes to river hydrology and water quality at different spatial scales (Uuemaa et al., 2005; Gémesi et al., 2011; Sun et al., 2014; Li et al., 2021b). However, given the complexity of surface water environmental processes, the influence of differences in natural geographic features on NPS pollution is usually misunderstood as being caused by land-use changes (Wu and Lu, 2021). Therefore, it is necessary to consider the influence of three indicators—natural geographic features, landscape composition and landscape configuration—on NPS pollution simultaneously.

      Methods such as conventional linear regression methods, multiple stepwise regression, and redundancy analysis can be used to assess the factors that contribute to changes in river water quality (Lee et al., 2009; Wu and Lu, 2019; Zhang et al., 2019). These methods are all based on statistical data, and some key environmental factors may be excluded from the study due to the possibility of collinearity between the data. Furthermore, these methods are difficult to assess the impact of spatial variation and the interaction of influencing factors on NPS pollution, which can not be resolved by traditional linear statistical methods. However, the geographic detection method (Wang and Xu, 2017) can explain the driving force behind the spatial differentiation of elements through a set of statistical methods and detect the explanatory power of two-factor interactions. The method can detect numerical and qualitative data, so it has been widely applied in the fields of human medicine, social economy, environmental science, and ecological landscape. The purpose of the study is to 1) evaluate the spatial and temporal distribution characteristics of NPS pollution in the watershed, 2) use geographic detectors to study the interaction of natural geographic features and landscape patterns on NPS pollution at different spatiotemporal scales in the watershed, and 3) explore the comprehensive effects of different influencing factors on NPS pollution through redundancy analysis. In order to achieve the above goals, we take the Burhatong River Basin in Northeast China as an example, and combines the methods of hydrological simulation, geographic detectors and redundancy analysis to determine the impact of natural geographical features and landscape patterns on non-point source pollution in the watershed. The methods and results can provide reference for clarifying the main factors and comprehensive effects of non-point source pollution control in the watershed, as well as clear ideas and beneficial views for landscape planners and administrators to improve the non-point source pollution of the watershed on a practical level.

    • Chinese and foreign economists name the development of the Tumen River, the Three Gorges of the Yangtze River, and Shanghai Pudong as China’s three cross-century projects (Shi et al., 2021), which play important strategic roles in the economic development of Northeast Asia. The Tumen River Basin is an important corridor that transverses the three countries of China, North Korea, and Russia, and the health of its river ecosystem has great significance to the maintenance of the management and sustainable development of the water environment of the transboundary basin. The Burhatong River Basin (BRB) is one of the important tributaries of the Tumen River (Fig. 1), with a total river length of about 172 km, an area of about 7064 km2. With 113 000 hm2 of arable land, a total cultivated land area comprising 16% of the whole watershed, and 28 500 hm2 of paddy fields, accounting for 25.2% of the cultivated area (Shi et al., 2021), which mostly concentrated in the valley basin of the Burhatong River, it is the famous rice-producing area in Jilin Province, Northeast China. Because of its early development history and reclamation rate of 20% to 30% (Shi et al., 2021), most of the land in the river valley has been developed into paddy fields and urban land, common for the dry fields and orchards reclaimed on the bilateral slopes of the river valley. Influenced by the policy of returning farmland to forests, the original forest resource area is slowly increasing, but the degree of forest fragmentation is also increasing, and the area of soil erosion accounts for 39% to 49% of the total area of the basin (Wang and Guo, 2006), increasing the risk of NPS in the basin.

      Figure 1.  Location of the Burhatong River Basin in Northeast China and location distribution of hydrological stations

    • As required by the NPS pollution simulation and geographic detector in the watershed, the remote sensing data required in the paper were mainly meteorological data, digital elevation model (DEM), soil data, land use data, vegetation cover data (FVC), landform type data, and hydrological data (Table 1). The precipitation data were from the China Meteorological Science Data Sharing Service Network (http://data.cma.cn/). The meteorological data of three stations in the study area were processed using professional meteorological interpolation software ANUSPLIN (Australian National University, Australia) to obtain raster data with a resolution of 30 m, divided into two phases, dry season and wet season. DEM data, the basic data for constructing small watersheds and slopes in the study area, were from the Resource and Environmental Science Data Center of the Chinese Academy of Sciences (http://www.resdc.cn/). The soil data were from the Cold and Arid Regions Science Data Center (http://bdc.casnw.net/), based on the World Soil Database China Soil Data Set (V1.1), with a resolution of 1000 m. In the study, based on the 2016 Landsat series of satellite remote sensing images of the BRB, the imaging was conducted using remote sensing interpretation, supported by ArcGIS10.1 (Esri, America) and Cognition902 software (Definiens Imaging, Germany); the object-oriented classification method and the artificial visual interpretation method, in combination, were used for land use of information extraction. From the TM (Thematic Mapper) image, the land use was divided into seven categories: farmland, forest, grassland, urban land (residential, commercial, and industrial land), wetland, water area, and bare land. The vegetation cover data (FVC) come from the Geographic Information Monitoring Cloud Platform (http://www.dsac.cn) with a time resolution rate for April and August and a spatial resolution of 30 m. The geomorphological data were from the Resource and Environment Data Center of the Chinese Academy of Sciences (http://www.resdc.cn). According to the geomorphological classification system of Zhou et al. (2009), the geomorphological data were merged based on the undulations of basic geomorphic form type in the study area and finally divided into five categories of plains, platforms, hills, small undulating mountains, and medium undulating mountains. The hydrological site data were used mainly to verify the subsequent the SWAT from the monthly monitoring data of the local environmental protection bureau. The Mopanshan site is located at the total outlet of the basin, and the lower Yanjixia site is located at the lower reaches of the basin (Fig. 1). The meteorological data required for the SWAT model were from the China Atmospheric Assimilation Driven Dataset (CMADS—LV1.0, http://www.cmads.org/), with a time scale from 1979 to 2018. The data include precipitation, temperature, wind speed, solar radiation, and relative humidity required by the SWAT model.

      DatasetsData sourceResolution
      DEM Resource and Environmental Science Data Center of the Chinese Academy of Sciences (http://www.resdc.cn/) 30 m
      Vegetation cover data (FVC) Geographic Information Monitoring Cloud Platform (http://www.dsac.cn) 30 m
      Precipitation China Meteorological Science Data Sharing Service Network (http://data.cma.cn/) 30 m
      Geomorphological Resource and Environment Data Center of the Chinese Academy of Sciences (http://www.resdc.cn) 1 km
      Soil type Cold and Arid Regions Science Data Center (http://bdc.casnw.net/) 1 km
      CMADS China Institute of Water Resources and Hydropower Research of the Chinese Academy of Sciences (http://www.cmads.org/)
      Notes: DEM, digital elevation model; CMADS, China Atmospheric Assimilation DrivenDataset

      Table 1.  Data needed for the Soil & Water Assessment Tool (SWAT) model building and geodetector studies

    • SWAT was adopted to simulate the watershed NPS pollution load, a watershed-scale model developed by the United States Department of Agriculture to assess and predict the impact of land use, land management practice, and climate change on hydrology and water quality (Arnold and Allen, 1996). The SWAT database input was finished in the extension module ArcSWAT 2012 (United States Department of Agriculture, America) of ArcMap10.2 (Esri, America). The data required to build SWAT model is detailed in the data source(Table1). In addition, the data of agricultural management measures in the watershed were from the Yanbian Prefecture Statistical Yearbook (Yanbian Bureau of Statistics, 2017).

      First, the SWAT model was used to extract the water system data from the watershed according to the watershed DEM elevation data, set the minimum area threshold of the sub-watershed, manually add the Mopanshan station as the outlet of the entire watershed, and then divide the entire watershed into 50 sub-watersheds (named No. 1 to No. 50). After these data were used to construct the SWAT model, SWAT-CUP was adopted for sensitivity analysis after screening out 23 parameters that were relatively sensitive to runoff and nitrogen and phosphorus. First, the monthly runoff data from the Mopanshan site from 2000 to 2006 were used for hydrological calibration and verification. Second, the monthly total phosphorus from 2015 to 2018 and the total nitrogen data of the Yanjixia lower site from 2017 to 2018 were used for water quality calibration and verification. To judge the simulation effect of the model, the determination coefficient R2 and the Nash efficiency coefficient (NSE) were selected to evaluate the model’s accuracy (Nash and Sutcliffe, 1970).

    • The land use data of study area in 2016 were converted from vector data to raster data; then, the sub-watersheds divided by SWAT were cut into 50 pieces of raster data, and finally, the landscape pattern index was calculated using Fragstats 4.2 software. In the study, PLAND (percentage of landscape), PD (patch density), ED (edge density), and AI (aggregation index) were selected at the level of patch type. Table 2 briefly summarizes the function and role of each indicator. Moreover, in the study, the landscape pattern evaluation method based on the ecological process of the source and sink was also calculated, namely, the location-weighted landscape contrast index (LWLI). The index was calculated using Eqs. 1 and 2 (Chen et al., 2009):

      IndexFunction and role
      Percentage of landscape (PLAND) Percentage of landscape composed of the corresponding patch type (class)
      Patch density (PD) Quantifies the number of patches per unit area
      Edge density (ED) Reflects the degree of landscape fragmentation
      Aggregation index (AI) Aggregation degree of different types of patches in response to landscape

      Table 2.  Functions and functions of main indicators of landscape pattern

      $$ \begin{split} {\text{LWLI }} ' =&\sum\limits_{i{\text{ = }}1}^m {A_i} \times W_{{i}} \times AP_{{i}} / \\ &\left[ {\sum\limits_{{{i = }}1}^m {A_i} \times W_i \times AP_i{\text{ + }} \times \sum\limits_{j{\text{ = }}1}^n {S_{ j}} \times W_{ j} \times AP_{ j}} \right] \end{split} $$ (1)
      $$ {\text{LWLI }} = {{{\rm{LWLI}}'_{\rm{distance}}}} \times {{{\rm{LWLI}}'_{\rm{elevation}}}} / {{{\rm{LWLI}}'_{\rm{slope}}}} $$ (2)

      where Ai and Sj refer to the areas of the ith ‘source’ and jth ‘sink’ landscapes; Wi and Wj are the weights for the ‘source’ and ‘sink’ landscapes; APi and APj refer to the percentages of the i-source and j-type landscapes; and m and n are the numbers of ‘source’ and ‘sink’ landscape types, respectively. LWLI'distance, LWLI'elevation, and LWLI'slope are the values of LWLI with respect to the distance, relative elevation, and slope gradient, respectively. We categorized forest, wetland, and grassland as ‘sink’ landscapes and farm, urban, and bare as ‘source’ landscapes. The weights for the forest, grassland, farmland, urban, wetland, and bare were assigned as 0.8, 0.6, 0.4, 1.0, 0.6, and 0.5, respectively, with reference to Li et al. (2020).

    • Geodetector is a set of statistical methods that reveal the driving forces of spatial differentiation of various elements (Wang and Xu, 2017). The basic idea is that if the spatial distribution of the independent variable is similar to that of the dependent variable, then the independent variable has an important influence on the dependent variale; the higher the similarity, the greater the influence. In the study, NPS pollution (TN and TP) was used as the dependent variable, and Geodetector was used to determine the main factors affecting its spatial differentiation. For influencing factors, the natural geographical features (soil, geomorphological, vegetation coverage, elevation, slope, rainfall) and land use pattern (landscape composition and landscape configuration) were comprehensively considered, given that the top three land use types in the study area were forestland, farmland, and urban land. Thus, in the land use pattern analysis, the three types of land use composition and their landscape configuration were mainly considered, with 19 factors in total finally selected as independent variables. The factor detector in Geodetector can detect the spatial differentiation of dependent variables and clarify the degree of interpretation of the spatial differentiation of NPS pollution by different influencing factors, and its differentiation size is determined by comparing the intra-layer variance of the NPS pollution with the total inter-layer variance, represented by value q:

      $$ q=1-\frac{{\displaystyle \sum _{h=1}^{L}{N}_{h}}{\sigma }_{h}^{2}}{\text{N}{\sigma }^{2}}=1-\frac{S S W}{S S T} $$ (3)

      where, h (= 1, 2, 3…L) is the number of classifications; Nh and N are the number of sampling units in layer h and the whole region; and σh2 and σ2 are the variances of the layer h and whole region, respectively. SST and SSW are the total quadratic sum and the inner quadratic sum. The statistic q is the monotonic function of the hierarchical spatial heterogeneity and the strength of q ∈ [0, 1]. As the stratification heterogeneity strength increases, its value also increases, which indicates that the influencing factor has a more significant effect on NPS pollution. The interactive detector Geodetector can test whether two (or more) factors (X1, X2) interact on the response variable by comparing q values (Gao and Wang, 2019). Table 3 lists the interaction between the two factors.

      DescriptionInteraction
      q(X1X2) < Min(q(X1), q(X2)) Weakened, nonlinear
      Min(q(X1), q(X2)) < q(X1X2) < Max(q(X1), q(X2)) Weakened, single factor nonlinear
      q(X1X2) > Max(q(X1), q(X2)) Enhanced, double factors
      q(X1X2) = q(X1) + q(X2) Independent
      q(X1X2) > q(X1) + q(X2) Enhanced, nonlinear

      Table 3.  Types of interaction between the two factors as defined in Geodetector

    • First, detrended correspondence analysis (DCA) on NPS pollution was conducted to determine whether to use a linear model or a unimodal model. The DCA results show that the longest gradient lengths of the four sorting axes were all less than 3; therefore, redundant analysis was used to identify the impact of influencing factors on NPS pollution. Redundancy analysis (RDA) is now widely used to determine the relationship between environmental factors and ecological indicators. CANOCO5.0 software (Microcomputor Power, America) was used to analyze the relationship between NPS pollution and impact factors.

    • The sequence uncertainty fitting version 2 algorithm was used to determine the results of the model, and the monthly runoff data of the Mopanshan site from 2000 to 2006 were used to calibrate and verify the hydrological process of the model. The total phosphorus data and the total nitrogen data from 2017 to 2018 were calibrated and verified for water quality. The determination coefficient R2 and the Nash efficiency coefficient NSE were selected to evaluate the suitability of the simulation effect. Fig. 2 shows the calibration and verification results. The R2 and NES of the runoff during the calibration period reached 0.84 and 0.83, respectively, and 0.78 and 0.69 respectively during the verification period. The R2 and NES of the total phosphorus during the calibration period were 0.76 and 0.72, respectively, and 0.78 and 0.69 respectively during the verification period. The R2 and NES of total nitrogen in the calibration period reached 0.88 and 0.86, respectively, and 0.89 and 0.78, respectively, in the verification period. In general, when R2 > 0.60 and NSE > 0.50, the model can have a good simulation effect (Santhi et al., 2001). On the whole, the SWAT model constructed in the study meets the requirements of simulation accuracy and can well simulate the basin’s hydrology and water quality.

      Figure 2.  Calibration and verification results of monthly runoff at Mopanshan station (a) and total phosphorus (TP, b) and nitrogen (TN, c) at Yanjixia station in the Burhatong River Basin, Northeast China

      The SWAT model was used to simulate the output intensity of total nitrogen and phosphorus in each sub-field because of the significant difference in rainfall in the region’s dry season (March, April, May) and wet season (July, August, September), so the monthly average output intensity of total nitrogen and phosphorus in the dry and wet seasons from 2016 to 2018 were extracted to compare the seasonal differences in the NPS pollution of the basin. Fig. 3 shows the results: in the dry season, the areas with high total phosphorus output load were mainly concentrated in the lower reaches of the basin, and the output intensity of No. 29, No. 30, and No. 33 subwatershed all exceeded 0.1 kg/ha. The pollution output load of total nitrogen in the dry season had strong spatial heterogeneity, reaching the maximum in the eastern plains, and the total phosphorus output intensity of the No. 19, No. 20, No. 22, No. 29, and No. 30 of sub-basins all exceeded 2.0 kg/ha, which was mainly because these sub-basins are all located in plain areas, the agricultural and urban land occupy a relatively high proportion, and the agricultural runoff and domestic wastewater cause high total nitrogen pollution. The wet season was the peak period of nitrogen and phosphorus output: the output intensity of nitrogen and phosphorus in the basin increases significantly, and the TP and TN output loads of No. 48 and No. 49 in the south of the basin increase significantly. Total phosphorus increased from 0.1 kg/ha to 0.6 kg/ha, and total nitrogen increased from 0.1 kg/ha to 3.0 kg/ha, which was caused by a combination of rainfall, topography and human activities. Under the action of rainfall for the steep terrain, soil erosion was relatively serious, leading to more NPS pollution. Vilmin et al. (2018) pointed out that an increase in surface runoff processes can lead to the loss of nitrogen and phosphorus, and excess nitrogen and phosphorus can enter rivers with runoff and sediment. The plain area in the eastern part of the basin was the main polluted area in the entire basin, and these sub-basins had relatively high pollution in the dry and wet seasons and require long-term treatment. However, in the wet season, the basin administrators should also propose corresponding countermeasures to cope with the problems of significant increase in NPS pollution of the southern part of the basin.

      Figure 3.  Total phosphorus and nitrogen export loads in the dry and wet seasons of the Burhatong River Basin, Northeast China

    • The value q obtained by Geodetector reflects the importance of natural geographical features or land use patterns in the dry and wet seasons on the spatial heterogeneity of NPS pollution. In the dry season, rainfall, farmland, and farmland concentration had the highest explanation of spatial heterogeneity of total phosphorus (Fig. 4a), which were 71.5%, 42.0%, and 37.8%, respectively: significantly higher than other influencing factors. The explanation rate of forest and urban land were 34.2% and 22.6%, and that of LWLI of characterization source-sink landscape process index on the spatial heterogeneity of total phosphorus also reached 35.2%. For total phosphorus, the explanatory power of forest fragmentation and elevation were 60.8% and 54.4% (Fig. 4b), and the forest, farmland, urban land, slope, forest patch aggregation, and farmland patch shape complexity had an explanation rate of over 30.0% for total nitrogen space of heterogeneity; generally, in the dry season, the explanation rate for the spatial heterogeneity of NPS pollution by natural geographical features and landscape composition was relatively high, while that of landscape configuration was generally low. In the wet season, the impact for influencing factors with explanatory power over 70.0% on the spatial heterogeneity of total phosphorus decreased successively, in order of complexity, for farmland patch (79.1%), proportion of farmland (78.3%), degree of forest patch aggregation (76.8%), forest patch shape complexity (75.6%), LWLI (75.4%), and proportion of forest (74.1%) (Fig. 4a). The explanatory power of vegetation coverage for total phosphorus also increased from 30.2% in the dry season to 62.7%. The explanation rate for landforms and soil types for NPS pollution is generally low in both dry and wet seasons, not exceeding 20.0%. For the total nitrogen in the wet season, the proportion of farmland and LWLI had the highest explanatory power, namely 72.3% and 68.0%. Compared with the proportion of forest, the spatial configuration of forest landscape had a greater influence on the spatial heterogeneity of total nitrogen in the wet season, and the explanatory power for forest patch complexity and aggregation was greater than 65.0%. In general, most of the natural geographical feature factors and landscape spatial configuration had higher explanatory power for NPS pollution in the wet season than in the dry season.

      Figure 4.  Different impact factors (q) on the spatial heterogeneity of total phosphorus (a) and nitrogen (b) pollution in dry and wet seasons. 1PD represents the fragmentation degree of forest patches, 1ED represents the complexity of the shape of forest patches, 1AI represents the aggregation degree of forest patches, 2PD represents the fragmentation degree of farmland patches, and 2ED represents the complexity of the patch shape of the farmland, 2AI represents the aggregation degree of farmland patches, 3PD represents the fragmentation degree of urban land patches, 3ED represents the complexity of the shape of urban land patches, and 3AI represents the aggregation degree of urban land patches, rainfall represents the amount of rainfall, soil represents soil type, slope represents the magnitude of the slope, forest represents the proportion of forest land, geomorphological represents different geomorphological types, urban represents the proportion of urban land, DEM represents terrain height, LWLI represents the location-weighted landscape contrast index, farmland represents the proportion of farmland and FVC represents Fractional Vegetation Cover

      From the above, the degree of influence of a single factor on the distribution of NPS pollution was analyzed, but in the actual process, the complex interaction among multiple factors jointly determines the spatial pattern of NPS pollution. Multifactor interaction detection also proved that the influence of factor interaction on the spatial distribution pattern of NPS pollution was higher than the influence degree of a single factor, which was manifested as two-factor enhancement and nonlinear enhancement (Fig. 5). In the dry season, the interaction between rainfall and other factors had strong explanatory power for the spatial distribution of total phosphorus, exceeding 80.0% (Fig. 5a), among which the interaction between rainfall and farmland agglomeration was the strongest, reaching 86.6%. For more information related to the original, if checking other translation information, you must enter the corresponding original, and the explanatory power of farmland fragmentation for total phosphorus in the dry season was only 17.6%, but its interaction with other influencing factors was 60.0% to 90.0%, and the interaction between farmland fragmentation and rainfall was 84.9%. For the total nitrogen in the dry season, the interaction between farmland patch complexity and DEM had high influence with explanatory power 82.4%. Forest land fragmentation and the interaction between DEM and other influencing factors had high explanatory power for the total nitrogen in the dry season with 60.0% to 85.0%. In the wet season, the influencing factors of more than 90.0% of the interaction’s explanatory power on the spatial distribution of total phosphorus were farmland fragmentation and farmland patch complexity (94.3%), LWLI and forest patch complexity (92.1%), LWLI and forest degree of aggregation (90.4%), and proportion of farmland and farmland patch complexity (90.3%). For the spatial distribution of total nitrogen, the interaction between farmland fragmentation and farmland patch complexity was the most significant, reaching 90.1%. The explanatory power of the interaction between the source-sink landscape index LWLI and forest fragmentation was 88.9%, ranking second among all interactive detection results. Generally, the interaction among various impact factors in the wet season was significantly higher than in the dry season.

      Figure 5.  Interaction detection of influencing factors on spatial distribution of total phosphorus pollution in dry season (a), wet season (b) and of total nitrogen pollution in dry season (c), wet season (d). 1PD represents the fragmentation degree of forest patches, 1ED represents the complexity of the shape of forest patches, 1AI represents the aggregation degree of forest patches, 2PD represents the fragmentation degree of farmland patches, and 2ED represents the complexity of the patch shape of the farmland, 2AI represents the aggregation degree of farmland patches, 3PD represents the fragmentation degree of urban land patches, 3ED represents the complexity of the shape of urban land patches, and 3AI represents the aggregation degree of urban land patches, X represents geomorphological types, rainfall represents the amount of rainfall, soil represents soil type, slope represents the magnitude of the slope, forest represents the proportion of forest land, urban represents the proportion of urban land, DEM represents terrain height, LWLI represents the location-weighted landscape contrast index, farmland represents the pro-portion of farmland and FVC represents Fractional Vegetation Cover

    • To eliminate the collinearity among the influencing factors, RDA was performed twice. First, all influencing factors were used as explanatory variables; in the process, variables were selected based on the test results of the sig significant test, importance test, and variance inflation factor test. The second RDA was performed by the influencing factors obtained after the first RDA. In the analysis, a Monte Carlo permutation test was adopted to test the significance (Braak and Smilauer, 2012), and RDA was further applied to clarify and determine the comprehensive effect of influencing factors on NPS pollution. In the dry season, the overall explanation rate of influencing factors on NPS pollution changes was 83.4% (Table 4), among which the contribution rate of DEM for nitrogen and phosphorus pollution changes was 54.1% (Table 4), and the relationship between DEM and TP, TN was negatively correlated (Fig. 6). The contribution rate of urban proportion for nitrogen and phosphorus pollution changes was 17.0%, ranking second, positively correlated with nitrogen and phosphorus pollution (Fig. 6). The proportion of farmland, degree of farmland aggregation, and degree of farmland patch fragmentation had a negative impact on nitrogen and phosphorus pollution, all positively correlated with nitrogen and phosphorus pollution (Fig. 6). The proportion of forest land, degree of forest land accumulation, and complexity of forest land patches had a positive impact on nitrogen and phosphorus pollution, negatively correlated with NPS pollution. In the wet season, the overall explanation rate of influencing factors on NPS pollution changes was 85.3% (Table 4), and the key influencing factors became the proportion of farmland (57.7%) and LWLI (10.5%) (Table 4); both were positively correlated with nitrogen and phosphorus pollution (Fig. 6). During the period, high vegetation coverage had a positive effect on the purification of nitrogen and phosphorus pollution, negatively correlated with pollutants. Under the leaching and scouring action of rainfall runoff, pollutants in the atmosphere, ground, and soil will diffusely enter into surface water and groundwater to cause water environment pollution; therefore, rainfall was positively correlated with NPS pollution.

      SeasonExplained variation / %Pseudo-FP valueKey metrics
      Axis1Axis2All axes
      Dry78.74.783.415.50.002DEM / 54.1%, Urban /17.0%
      Wet75.13.685.317.90.002Farmland /57.7%, LWLI /10.5%

      Table 4.  Percentage of non-point source pollution changes explained by influencing factors in the Burhatong River Basin

      Figure 6.  Redundancy analysis (RDA) shows the correlation between non-point source pollution (solid line) and impact factors (solid line) in different seasons, 1PD represents the fragmentation degree of forest patches, 1ED represents the complexity of the shape of forest patches, 1AI represents the aggregation degree of forest patches, 2ED represents the complexity of the patch shape of the farmland, 2AI represents the aggregation degree of farmland patches, 3ED represents the complexity of the shape of urban land patches, rainfall represents the amount of rainfall, slope represents the magnitude of the slope, forest represents the proportion of forest land, urban represents the proportion of urban land, DEM represents terrain height, LWLI represents the location-weighted landscape contrast index, farmland represents the proportion of farmland and FVC represents Fractional Vegetation Cover. If the included angle between the arrow of each Influencing Factors and the arrow of non-point source pollution were less than 90 degrees, they were in direct ratio to each other; if greater than 90 degrees, it indicated negative correlation; and if it were equal to 90 degrees, there was no correlation

    • On the time scale, the peak of nitrogen and phosphorus loss load in the Burhatong river basin occurs mainly from June to August. In general, rainfall and nitrogen and phosphorus loss were positively correlated, as shown by the increase of nitrogen and phosphorus load with the increase of runoff (Wang et al., 2014). The No. 48 and No. 49 subwatersheds in the south of the basin had a combined effect of rainfall, DEM, and human activities. The output load of total phosphorus and total nitrogen increased significantly. In the dry season, the areas with higher TP and TN output loads were mainly concentrated in the plains downstream of the watershed, which was the main distribution area of cities, farmland and other infrastructure, and agricultural runoff and domestic wastewater lead to higher pollution. Although the subwatershed No. 19 is located in the downstream area of agricultural land and urban land, the pollution output load of the sub-watershed No. 19 was smaller than that of No. 20 and No. 24. Existing studies had shown that the spatial locations of the ‘source’ and ‘sink’ landscapes of the watershed were different, and their impact on the water environment was also different (Shi et al., 2021). The urban land and farmland of the No. 19 subbasin are located in the upstream area, and the downstream area is a large area of forest, which has played a better role in purifying the NPS of the subwatershed. It also reflects the importance of the spatial location of the ‘source’ and ‘sink’ landscape on the water environment (Liu et al., 2018). From the spatial distribution characteristics of NPS pollution in the watershed, the key control areas for non-point source pollution prevention and control were the southern area, with higher slopes, and the eastern area, where farmland and towns are concentrated.

      There was a large difference in the spatial distribution and seasonal change of NPS pollution in the basin, indicating strong spatial heterogeneity. Natural geographical features and land use patterns can have a significant impact on the spatial distribution of TN and TP (Fig. 4). The correlation of the spatial heterogeneity cannot simply be analyzed by traditional statistical analysis methods based on global statistics. When analyzing the relation between influencing factors and NPS pollution, at different spatiotemporal scales, the dominant influencing factors that affect the spatiotemporal scale distribution of NPS pollution were different (Li et al., 2020), and the spatial heterogeneity must be considered using methods such as the Geodetector method used in the study.

    • NPS pollution in the basin was a comprehensive reflection of multi-scale environmental factors. It was not only influenced by natural geographical features; the land use pattern was also considered the main cause of regional NPS pollution. NPS pollution was caused by concurrent multiple factors, and the coupling mechanism of multiple factors and the contribution degree of determining every factor were still a challenge. In the study, the Geodetector method was used to clarify the main control factor producing spatial differentiation of nitrogen and phosphorus pollution. Wang and Xu. (2017) believed that the consistency of the spatial distribution between two geographic variables is more difficult to obtain than the linear correlation between the two variables. Therefore, compared with Pearson correlation analysis, Geodetector provides stronger statistics to reveal the causality between independent variables and dependent variables (Wang and Xu, 2017). Either single factor analysis or factor interaction reveals the dominant factors of NPS pollution in basin at different spatiotemporal scales at a deeper level. DEM and forest fragmentation are the main factors controlling the spatial differentiation of total nitrogen in the dry season; in plains and platforms with small topographical undulations, land use type was the main impact factor after precipitation, with contribution rate of over 30.0% (Fig. 4). The eastern region was the main distribution area of cities, arable land, and other infrastructure and the main implementation area for future new urbanization and rural revitalization policies; and the spatial distribution of land use types will also change accordingly. Thus, in the process of future urban and economic development, the intensive use of land and reasonable planning should be paid special attention to minimize its impact on NPS pollution. Rainfall variation changed the flow path time, peak flow, and basin flow, as well as the transformation and transport characteristics of NPS pollutants (Shrestha et al., 2018). In the wet season, we found that most of the landscape configuration indicators have a significant impact on the distribution characteristics of NPS pollution, mainly because it determines the water flow and nutrient and energy exchange among different landscape patches; such influence was more obvious in the wet season (Ricart et al., 2015). Landscape patterns affect the hydrological and chemical runoff processes that carry pollutants and nutrients into streams, rivers, reservoirs, and lakes, while the proportions of certain land-use types determine the pollutant and nutrient loads in the watershed. Thus, landscape configuration may characterize hydrological and chemical runoff processes to some extent by accelerating or decreasing runoff velocities with the same proportion of certain landuse types (Xiang, 1996).

      The interaction test results show that the combined influence of any two factors was significantly higher than that of a single factor, mainly manifested as nonlinear interaction enhancement (q(X1X2) > q(X1) + q(X2)) and two-factor interaction enhancement (Table 3), and the dominant interaction of nitrogen and phosphorus pollution differs in different seasons. During the dry season, both forest fragmentation and the interaction of rainfall with other factors were at higher strengths, and when more precipitation was present, reducing forest fragmentation suppressed the increase in nitrate loading in the baseflow (Li et al., 2021a). Especially in the BRB where forests were dominant patches, understanding the impact of the interaction between forest fragmentation and other factors on NPS can provide new ideas for watershed managers to efficiently control NPS. High fragmentation of forests impairs their ecosystem functions such as hydrological regulation and water purification (de Mello et al., 2020). The interaction of forest fragmentation and negative factors such as rainfall and elevation will increase the risk of loss of non-point source pollution (Li et al., 2021a). The results once again prove that it is more effective to evaluate the comprehensive effect of multiple influencing factors on non-point source pollution based on geodetector.

    • Although geodetectors were effective for obtaining the key factors influencing the spatial heterogeneity of NPS pollution in the watershed, we still want to determine the combined effects of these key factors through further analysis, and RDA was used to analyze the correlation between the impact factors and NPS pollution. The results show that the land use pattern was the main factor affecting nitrogen and phosphorus pollution, especially in the rainy season. Although the dominant influence of natural geographic features (rainfall, slope, etc.) on NPS pollution in the watershed had become a relatively broad view, our research results show that the influence of landscape patterns on NPS pollution exceeds that of natural factors such as precipitation and elevation. The area with higher rainfall and greater slope is more likely to cause NPS pollution, and the general view has been confirmed by most studies (Shen et al., 2014; Alnahit et al., 2020), but in this study, it was found by RDA that rainfall and DEM are negatively correlated with TN and TP in the dry season (Fig. 6), which may be due to the large accumulation of cultivated land and urban land in plain areas with low rainfall. In these areas with frequent human activities, a large amount of domestic waste, the discharge of industrial wastewater, and the chemical fertilizers and pesticides used in agricultural production enter the river through the impervious surface of the city, which leads to more serious water pollution (Khatri and Tyagi, 2015), and the general effect of rainfall and DEM on NPS pollution were obscured by the land use pattern. Our findings support the view of Li et al. (2021a), who point out that land use/land cover changes may alter and submerge the impact of climate change on ecosystems. Different types of landscape composition can change the hydrological cycle process through evapotranspiration, interception, infiltration, and absorption to influence the migration of terrestrial nutrients. Forests have always been considered directly related to agricultural NPS (Ouyang et al., 2010), and its vegetation cover can reduce soil erosion, functioning as reducing storm runoff, decreasing soil erosion, and adsorbing pollutants, and can effectively reduce surface runoff carrying nutrients into rivers, reducing water quality deterioration in a certain degree (Wang et al., 2018).

      Studies had shown (Roth et al., 1996) that for larger basins, land cover percentage alone can explain most of the changes in water quality; however, for smaller basins, landscape configuration is more critical. Therefore, determining the contribution of landscape composition or configuration on the contribution of NPS pollution was particularly important for controlling NPS pollution in basins. Studies had shown that the lower the degree of fragmentation of farmland patches, the better the connectivity of patch corridors. The farmland pollutants were prone to form NPS (Wang et al., 2018), so the high degree of farmland aggregation indicates that the degree of farmland fragmentation in the basin was low, which was an important cause of total phosphorus pollution in the dry season. The patch complexity of farmland was also positively correlated with NPS pollution in the dry and wet seasons. Moreover, the interactive results of the map detector show that the patch complexity and fragmentation of farmlands have 90.1% and 94.3% of explanation power for total nitrogen and total phosphorus in the wet season. The increase in farmland ED was accompanied by the increase in PLAND, which seems to expand the diffusion interface of agricultural sources and promotes the loss of nutrients. The finding was consistent with the results of other studies: namely, agricultural landscapes with more complex shapes impose a relatively great nutrient load on rivers (Lee et al., 2009; Li et al., 2021a). On the contrary, the more complete and complex the forest, the more its nutrient load interception (Clément et al., 2017), which was also confirmed by RDA analysis. Although the proportion of forest land was negatively correlated with NPS pollution, indicating that the increase in forest land cover contributes to water purification, the conclusion was often limited to the analysis of landscape composition and pattern. In the basin landscape pattern of patches, taking forests with pollutant reduction effects as the advantage, it was most sensitive to the degree of landscape fragmentation (Liu et al., 2019). The more fragmented the forest landscape, the more scattered the patches are, and the less conducive to processes such as inhibition of soil erosion, runoff, resulting in the loss of pollutant elements. In the study, it was also found through redundant analysis that the increased fragmentation of forest landscape patches increases the risk of NPS in the basin. In 2002, the process of returning farmland to forests was fully launched in the basin, transform cropland into ecological or economic forests to improve water efficiency and control soil erosion, and reduce NPS pollution in the basin; however, the negative effects of forest fragmentation may had negated the projects positive effects.

      Different spatial locations for the ‘source’ and ‘sink’ landscapes in the basin had different impacts on the water environment (Liu et al., 2018); most of the above landscape indicators had been introduced to characterize the landscape pattern, although the studies using landscape pattern to analyze related ecological process were very few; to quantitatively evaluate the role of the basin landscape pattern in the ecological process, the relationship between LWLI and the NPS pollution were analyzed in this study. The results indicate that LWLI can be used to assess the risk of nutrient loss and soil erosion. It is possible to use LWLI to provide a scientific basis for landscape pattern design and planning; although we realize that in such a rapidly urbanized basin, it is unrealistic to reduce the proportion of agricultural and urban land on a large scale, we can still adjust the spatial location of ‘source’ and ‘sink’ landscape to mitigate NPS pollution.

    • NPS pollution in the basin was influenced by natural geographical features and land use patterns. In the study, the Burhatong River Basin, Northeast China was taken as an example, based on the SWAT and Geodetector models, the indicators of natural geographic features, landscape composition, and landscape configuration on the impact and comprehensive effects of the spatial heterogeneity of NPS pollution were comprehensively analyzed. The main conclusions drawn in the research were as follows.

      1) the SWAT model can better simulate the spatiotemporal distribution of NPS pollution in the BRB. In the dry season, the areas with high total phosphorus and total nitrogen output loads were mainly concentrated in the plains of the lower reaches of the basin, with cities, arable land, and other infrastructure mainly distributed, and agricultural runoff and domestic wastewater leading to high water pollution. In the wet season, the peak period of nitrogen and phosphorus output, under the combined effects of rainfall, DEM, and human activities in the No. 48 and No. 49 sub-basins of the southern part of the basin, the output load of total phosphorus and nitrogen increase significantly. The key control areas for NPS prevention and control are in the southern area, with high slopes, and the eastern area, where farmland and towns were concentrated.

      2) Geodetectors had revealed the dominant factors of NPS pollution in basins at different spatiotemporal scales, and both the single factor analysis and factor interaction further explain the spatial heterogeneity of NPS pollution. The impact of landscape composition and landscape configuration indicators on NPS pollution was more significant in the wet season, and the natural geographic characteristic indicators of the basin were less significant than the previous two types of indicators. The results of the interaction test show the importance of considering the multifactor interaction when studying NPS pollution in the watershed.

      3) The proportion of forest land was negatively correlated with nitrogen and phosphorus load. The higher the fragmentation of forest patches, the more serious the loss of NPS in the watershed; the proportion of farmland is positively correlated with nitrogen and phosphorus load, and the agglomeration and complexity of farmland patches was high, to a certain extent, increasing the risk of NPS spreading. When planning land use and optimizing landscape patterns in the area, the landscape composition pattern and spatial pattern should be considered comprehensively to better improve the NPS of the basin.

Reference (56)

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return