Identifying the critical areas and primary sources for agricultural non-point source pollution management of an emigrant town within the Three Gorges reservoir area

Agricultural non-point source pollution is threatening water environmental health of the Three Gorges reservoir. However, current studies for precision management of the agricultural non-point source pollution within this area are still limited. The objective of this study was identifying the critical areas and primary sources of agricultural non-point source pollution for precision management. Firstly, the inventory analysis approach was used to estimate the discharge amount of total nitrogen (TN), total phosphorus (TP), and chemical oxygen demand (COD) from farmland fertilizer, crop residues, livestock breeding, and daily activities. Afterwards, the deviation standardization method was applied to evaluate the emission intensity of TN, TP, and COD, as well as calculating the comprehensive pollution index (CPI) of each village, based on which the critical areas for agricultural non-point source pollution management could be distinguished. Moreover, the equivalence pollution load method was conducted to identify the primary pollution sources within each critical zone. The above methods were implemented to an emigrant town within the Three Gorges reservoir area named Gufu. Results showed that agricultural non-point source pollution in Gufu town has been alleviated to a certain extent since 2016. Nevertheless, in four areas of the town (i.e., Longzhu, Fuzi, Shendu, and Maicang), the agricultural non-point source pollution still deserved attention and improvement. For the mentioned critical areas, farmland fertilizer and livestock breeding were the primary sources causing agricultural non-point source pollution. The emission amount of TN and TP from farmland fertilizer accounted for 60% and 48% of the total, respectively. And those from livestock breeding were 29% and 46%. Our research could provide definite targets to relieve agricultural non-point source pollution, which had great significance to protect water environment while coordinating regional economic growth after emigrant resettlement.

quality of the aquatic environment, and the pollution sources are mainly concentrated in nutrients such as nitrogen and phosphorus (Zhang & Xu, 2011;Ghebremichael et al., 2013;Hou et al., 2022). As affected by numerous external factors such as climate, topography, hydrology, and land use patterns, the NPS pollution embodies large randomness and temporal-spatial variations of the pollution load, which greatly increases the difficulty and complexity in NPS pollution management (Chen et al., 2022;Wang et al., 2020). It is rather difficult to remain a long-lasting effect of the NPS pollution management. For example, the Chesapeake Bay watershed has gone through the implementation of the best management practices (BMPs) for decades, but the improvement of clean water quality in many tributaries was faintness (Ator et al., 2020;Keisman et al., 2018). Moreover, Lintern et al. (2020) conducted a literature review representing over 90 sites across Europe, Asia, and North America, most studies indicated no improvement of the water quality responded to the implementation of BMPs.
Management of NPS pollution could still offer a strong basis for protecting regional water environment and ecological security (Hoang et al., 2019;Rong et al., 2021). However, nutrient levels from different regions of a watershed can vary substantially. The NPS pollution in some regions might be significantly severe than that in the other regions even in the same sub-basin (Niraula et al., 2013;Zuo et al., 2022). The overall control of the NPS pollution is often difficult and costly, and it is also unnecessary to implement comprehensive governance in the whole river basin (Chen et al., 2023;Wang et al., 2016). To address the NPS pollution management situation, critical area, which is responsible for disproportionately high levels of the NPS pollution to the other areas of a watershed, was first proposed by some researchers (Dickinson et al., 1990;Mostaghimi et al., 1997). Afterwards, numerous studies have been launched on targeting the critical areas for achieving water quality goals more efficiently, as its potential advantages in terms of both costs and environmental benefits (Behera & Panda, 2006;Fleming et al., 2018;Ramesh et al., 2020;Shortle & Horan, 2001;Zuo et al., 2022). It has been proved that the NPS critical areas identified prior to watershed management could indeed improve the efficiency of NPS pollution reduction. Giri et al. (2012) evaluated the targeting methods for BMPs implementation of the Saginaw River watershed. They agreed that identification of critical source areas was most effective at TN and TP reduction, probably because it selected out the maximum area of high priority for BMPs implementation. Li et al. (2019) proposed a framework of delineating lake buffer zones include critical source areas for NPS pollution control, resulting a minimum buffer range that meeting both environmental requirements and economic needs. Ding et al. (2020) designed five BMPs scenarios for environmental and cost-benefit analyses of the NPS pollution based on critical pollution source areas identification, which showed a 63.4% and 62.6% reduction efficiency of the average TN and TP, respectively.
Presently, methods of identification of the critical areas can be divided into the following three categories roughly: the empirical formula method, the multi-factor comprehensive analysis method, as well as the mechanism model method (Chen et al., 2022;Grossweiler et al., 2021;Shen et al., 2015). The empirical formula method was easy to be applied, whereas with the limitation in quantifying the pollution loads in terms of absolute values (Ramesh et al., 2020;Sharpley et al., 2011). Taking this limit into consideration, the multi-factor comprehensive analysis method has been confirmed to grasp the nettle. For example, Guo et al. (2004) combined the agricultural non-point pollution potential index method with the geographic information system to recognize the critical areas when the basic data was not fully accessible. Karst-Riddoch (2014) developed a method based on the USLE and export coefficient modeling, to assess the P loading and identify hot spot areas of the P loading. However, the multi-factor comprehensive analysis method was mostly source or availability-based approach, which did not consider transport factors (Ramesh et al., 2020). But the critical source areas were affected by source and transport factors simultaneously. Therefore, mechanism processes of the pollution loading were incorporated into the models to support the identification of the critical source areas in water pollution studies (Shoemaker et al., 2005). The models were usually hydrological and water quality (HWQ) models including the Water Assessment Tool (SWAT) model (Young et al., 1989;Villeneuve et al., 1998;Arnold et al., 1998;Bingner & Theurer, 2001;US EPA, 2019). The facts showed that the SWAT model was the most widely applied model (Ramesh et al., 2020). For example, Ghebremichael et al. (2010) applied the SWAT model to find that about 80% of TP loss occurred within only 24% of the whole watershed area. Dong et al. (2018) incorporated the SWAT model and genetic algorithm to reveal the response of water quality to critical source area identification, and results showed that almost 85% of diffused TP originated from 30% of the watershed area.
Although substantial research efforts have been made to develop techniques for critical source areas identification, it remains challenging to identify the critical source areas for data-poor regions. In this research, an emigrant town, Gufu, located within the Three Gorges reservoir area was selected as a typical case study area. Firstly, the inventory analysis approach was used to estimate the discharge amount of pollutants from different sources. Afterwards, the deviation standardization method was applied to calculate the comprehensive pollution index (CPI) of the region, based on which the critical areas for agricultural non-point source pollution management could be distinguished. Eventually, the equivalence pollution load method was conducted to identify the primary pollution sources within each critical zone. Our research could provide definite targets to relieve agricultural non-point source pollution, which had great significance to protect water environment while coordinating regional economic growth after emigrant resettlement.

Study area
Gufu is an emigrant town which has been developed to ensure the construction of the Three Gorges dam. It is in the upstream of the Three Gorges Reservoir, extending an area of about 447 km 2 (Fig. 1). There are 10 villages in Gufu town named Longchi, Shenduhe, Maicang, Gudong, Xianshui, Zhongyangya, Pingshui, Longzhu, Fuzi, and Beidouping, respectively. The region has a typical subtropical monsoon climate with annual mean air temperature of 15.3 °C. The annual precipitation distributes non-evenly among seasons from 800 to 1200 mm. Wheat, rice, maize, coarse cereals, broad beans, and soybeans are the main crops there. The natural river system is abundant in Gufu town, Pingshuihe, Xianshuihe, and Gufuhe are the three main rivers flow through the region. According to the results of water quality monitoring, Gufuhe, a big brunch of the input tributary to the Three Gorges Reservoir, mirrored deplorable water quality inferior to class V. The effluent of the rivers in Gufu town flowing into the tributaries to the Three Gorges Reservoir tends to intensify the eutrophication risk very likely. Therefore, it is extremely important and essential to ascertain the critical areas and primary sources of the pollution, for the sake of making direct actions to protect the water environment.

Inventory analysis method
The inventory analysis approach was utilized to determine the emission of agricultural non-point source pollution within the Gufu town from 2016 to 2021. There was negligible use of agricultural pesticides and plastic film within Gufu town. Beidouping was the only hamlet in the town had aquaculture ponds, where a rather tiny amount of grass carp is breeding. Therefore, the inventory analysis method divided the sources of agricultural non-point source pollution into the following four categories: farmland fertilizer, crop residues, livestock breeding, and daily activities (Table 1). Based on the previous water quality monitoring, the main pollutants were identified as TN, TP, and COD. Detailed calculation equations for each pollution source were attached subsequently.
1. The amount of pollutants from farmland fertilizers: According to field investigation, the most commonly used fertilizers in Gufu town were nitrogen, phosphate, and compound fertilizers. The amount of N and P from farmland fertilizers were calculated as followed: where E F,i is the annual emission of pollutant i produced by farmland fertilizers; i is the type of pollutants including TN and TP; A j is the annual application amount of fertilizers; j is the type of chemical fertilizer including nitrogen, phosphate and compound fertilizer; C f ,j is the proportion of N or P in the chemical fertilizer; R j is the loss coefficient of pollutants of the chemical fertilizer, which approximately equal to 20% and 7% in Hubei Province, respectively.
2. The amount of pollutants from crop residues: the crop residues in our study refers to the waste of rice, wheat, maize, beans, potatoes, oil crops, and vegetables. Annual amount of pollutants from crop residues could be determined as follows: where E C,i is the annual emission of pollutant i generated by crop residues; i is the type of pollutants including TN and TP; Y j is the annual yield of the crop j; j is the type of crop including rice, wheat, maize, beans, potatoes, oil crops, and vegetables; Sr j is the ratio of straw produced by crop j to its yield ( Table 2); C c,j is the content of N and P in straw from crop j (Table 3); R i,c is the loss coefficient of pollutants of the crop straw.
3. The amount of pollutants from livestock breeding: Livestock excrement was the primary source of pollution in our study. The amount of pollut- ants from livestock breeding had a direct bearing on the number of livestock, the excreta coefficient of different animals, as well as the loss coefficient of pollutants from livestock excrement, which could be described as follows: where E L,i is the annual emission of pollutant i generated by livestock breeding; i is the type of pollutants including TN, TP, and COD; N j is the number of livestock; j is the type of livestock in Gufu town including cattle, pig, sheep and live poultry; C L,j is the annual excreta coefficient of livestock j; R C,i is the annual loss coefficient of pollutants of livestock breeding (Table 4).

The amount of pollutants from daily activities:
The indiscriminate discharge of untreated domestic sewage containing high TN, TP, or COD posed potential risk of non-point source pollution to surface water and degraded the surrounding water environment. The amount of pollutants from the above daily activities could be quantified as follows: where E D,i is the annual emission of pollutant i produced by daily activities; i is the type of pollutants including TN, TP, and COD; P r is the amount of rural population in Gufu town; I i is the rural residents' average per capita domestic wastewater per day; R D,i is the removal rates of pollutants by sewage treatment; C D,i is the rural residents' average per capita domestic garbage per day (Table 5).

Deviation standardization method
According to the amounts of pollutants calculated through the inventory analysis method as mentioned above, the pollution-emission intensity under each inventory category was determined by the following equation: where X ij is the pollution-emission intensity under category i of village j, kg/(hm −2 ·a); E i is the annual amount of pollutants from category i; i is the inventory category of the pollution sources including farmland fertilizer, crop residues, livestock breeding, and daily activities; F is the area of the village j. Afterwards, the pollution-emission intensity under each inventory category was processed by the deviation standardization method: where Y ij is the standardized value of the pollutionemission intensity under category i of village j; X ij is the pollution-discharging intensity under category i of village j, kg/(hm −2 ·a); X min is the minimum value of the pollution-discharging intensity; X max is the maximum value of the pollution-discharging intensity.
And then, the standardized values of the pollutionemission intensity were added together to get a comprehensive pollution index (CPI) for each village, based on which the pollution status of all the villages were ranked. Equation of the comprehensive pollution index was described below: standardized value of the pollution-emission intensity under category i of village j.

Equivalence pollution load method
The equivalence pollution load method is used to identify the primary pollution sources in each village. In the "Inventory analysis method" section, the amount of pollutants from each inventory category have been calculated, the equivalence pollution load of each pollutant could be determined as follows: where P ij is the equivalent pollution load of pollutant i of village j, m 3 /a; Q ij is the average annual emission of pollutant i of village j, t/a; C i is the emission standard of pollution i to surface water bodies, which could be referred to the Surface Water Environmental Quality Standard (GB3838-2002); K ij is the equivalent pollution load ratio of pollutant i of village j.

Results and discussions
Temporal change characteristics of the annual average pollutant emissions Figure 2 depicts, based on the inventory analysis method, the temporal change characteristics of the annual average pollutant emissions. It was evident that the yearly emissions of TN and TP from farmland   (Fig. 2a, b). Further analysis found significant decreases of the application amounts of nitrogen, phosphate fertilizer, and compound fertilizers from 2019 to 2020, which could perfectly explain the remarkable reductions of the TN and TP emissions within these 2 years. Specifically, the nitrogen fertilizer application rate in 2020 was reduced to half of that in 2019, while the phosphorus fertilizer application rate was decreased to even less than half that of the previous year (Fig. 3). The large-scale outbreak of COVID-2019 could be the deep causes for this phenomenon.
In addition, the annual emissions of TN, TP, and COD from livestock breeding were also decreasing, owing to the continual declining of the livestock numbers (Fig. 4). The highest decline rates of TN, TP, and COD emissions were 24.83%, 24.99%, and 28.19%, respectively (Fig. 2). However, the decline rates of livestock breeding sources were much lower than that of the farmland fertilizer sources. Therefore, livestock breeding sources have replaced farmland fertilizer sources to become the primary contributor of TP pollution since 2020 (Fig. 2b). The annual emissions of TN, TP, and COD from crop residues and daily activities showed small changes (Fig. 2), the main reason could be attributed to the slow growth of both rural population and straw yield.
The yearly emissions of TN and TP exhibited a "decline-smooth" trend, while the annual emission of COD exhibited a "decline-smooth-decline" trend. But on the whole, all these three pollutants displayed a distinct downward tendency, which indicated certain appreciation for the agricultural non-point source pollution measurements of the Gufu town. Furthermore, although the annual emission of COD was the largest, the variation of emission amount was the least and did not go above 21% (Fig. 5). The emissions of TN and TP were largely affected by the application amounts of farmland fertilizers, the yearly emissions the 10 villages. The large difference in cultivated land area could be the main factor for this phenomenon. For example, the annual emissions of pollutants from Shendu were at low level, but the cultivated land area of the village was less than 500 hm 2 , which could also result in a relatively high level of the pollution-emission intensity. In general, the annual emissions and pollution-emission intensities of TN and TP had dropped significantly since 2016, with relatively large declines occurred in villages such as Fuzi and Longzhu. As for the annual emission and pollution-emission intensity of COD, most villages showed a negative trend except Maicang, with greater declines occurred in villages such as Beidouping, Fuzi, and Longzhu. In terms of the annual TN and TP emissions, the first three villages were followed by Pingshui, Xianshui, and Gudong. As for the annual COD emission, Xianshui, Pingshui, and Gudong were also top villages that made main contributions. However, the COD emission from Beidouping was equally significant, probably because Beidouping was the only village had aquaculture ponds. The spatial distribution of pollution-emission intensities of TN and TP was basically consistent with that of the annual pollutants' emissions. But small arable land area and excessive applying fertilizer led to high pollution-emission intensities of TN and TP in villages such as Longzhu and Fuzi. The cultivated land area of Longzhu and Fuzi was less than 1000 hm 2 , while the fertilization application rate of the two villages was on the top. As consequences, the pollution-emission intensity of Longzhu and Fuzi toped the list of all the 10 villages. Although certain regulation measures have been matched to these villages, the control effects were somewhat different form anticipation. Therefore, more refined approaches need to be established to improve the pollution management level, of which identifying the critical areas and primary sources of agricultural non-point source pollution would be worth trying.
Identifying the critical areas for agricultural non-point source pollution management As has been described in the "Deviation standardization method" section, the comprehensive pollution index (CPI) was presented for the identification of critical areas. Table 6 gives a concrete result of the CPI of TN, TP, and COD of all the 10 villages. The ranking of CPI was consistent with the results of pollution-emission intensities. The range of CPI was between 0.0024 and 0.8291, corresponding respectively to Longchi and Longzhu. Except for Longzhu, villages such as Shendu, Fuzi, and Maicang also showed higher CPI values than the others, which were 0.6967, 0.3440, and 0.3187, respectively. Therefore, Longzhu, Shendu, Fuzi, and Maicang were identified as the critical areas for agricultural non-point source pollution management afterwards. For the four critical areas, this paper would make deep analysis on the primary sources of agricultural non-point source pollution, so as to provide a targeted and refined approach to improve the pollution management level. Identify the primary sources of agricultural non-point source pollution

Identifying the main pollutants
As has been described in the "Equivalence pollution load method" section, the equivalence pollution load method was adopted to analyze the primary sources of agricultural non-point source pollution. Figure 7 depicts the equivalence pollution load of TN, TP, and COD in each village. Results indicated that the equivalence pollution load was not the same among all the villages. Specifically, in villages such as Longchi, Shendu, Maicang, Fuzi, Gudong, Xianshui, and Longyangya, the equivalence pollution load was TN > TP > COD. While in villages such as Beidouping, Longzhu, and Pingshui, the equivalence pollution load was TP > TN > COD. Fertilizer application rate was the key factor to decide whether TN was greater than TP. According to the Agricultural Economics Annual Report of Gufu town, the nitrogen application rates were much higher than that of the phosphate fertilizer in many villages except Beidouping, Longzhu, and Pingshui. The difference between nitrogen and phosphate application rates reached to a maximum value of about 79 t in 2016, and was gradually narrowing since 2019.
From an overall perspective of view, the equivalence pollution load of TN and TP was very approximate. The ratio of equivalence pollution load among TN, TP, and COD were about 50%, 47%, and 3%, respectively. Therefore, it could be concluded that TN and TP were the main pollutants of not only the four critical areas, but also the whole Gufu town.

Identifying the primary sources of the main pollutants
In the "Inventory analysis method" section, the inventory analysis method has divided the agricultural nonpoint source pollution sources of our study into the following four categories: farmland fertilizer, crop residues, livestock breeding, and daily activities. The equivalence pollution load ratio of the four pollution sources were shown in Table 7. Results indicated that farmland fertilizer and livestock breeding were the uppermost sources of the agricultural nonpoint source pollution, considering their high equivalence pollution load ratios of about 53% and 38%,  respectively. By contraries, both the daily activities and crop straw had low equivalence pollution load ratios, which were 7% and 2%, respectively. As for the 4 critical areas has been identified in the "Identifying the critical areas for agricultural non-point source pollution management" section, the primary source of pollution in Longzhu and Fuzi was farmland fertilizer, while in Shendu and Maicang, the main sources of pollution were farmland fertilizer and livestock breeding. As for the identified critical areas, Longzhu, Fuzi, Shendu, and Maicang, it would be better to implement certain measures such as water-saving irrigation and water-fertilizer integration, for the aim of continuously reducing the main pollutants without influencing the crop yield. Furthermore, livestock breeding in such areas also demands much attention. For the purpose of pollution reduction, solid-liquid separation, and anaerobic digestion of the animal manures, as well as returning the fecal waste to field are preferential choices. The contribution of primary pollution sources to main pollutants is shown in Fig. 8. It could be concluded that the major contributor to TN was the farmland fertilizer (60%), followed by livestock breeding (29%), daily activities (8%), and crop residues (3%). Similarly to TP, the contribution of farmland fertilizer took up about 48% of the total, followed by livestock breeding (46%), daily activities (4%), and crop residues (2%). Although the use of farmland fertilizer has dropped by about 49.08% from 2016 to 2021, it was still one of the primary sources of TN and TP. As to COD, most of the yearly average COD emission was originated from livestock breeding (75%), followed by daily activities (19%), and crop residues (6%). Although the daily activity was not the main pollution sources of TN and TP, it was very influential to the annual COD emission. Considering the rising population of the study area, in which the topography is mountainous and very complex. The establishment of a decentralized quality separation treatment system and subsurface artificial wetlands is recommended in areas with low-income level, complex terrain, dispersed population, and limited water resources. In places with favorable economic conditions, flat topography, dense populations, and good water use conditions, decentralized mixed sewage treatment systems and stabilization ponds are advised to be built.

Conclusions
In this research, we took an emigrant town, Gufu, within the Three Gorges reservoir area as a case study area. The inventory analysis method, the deviation standardization method, as well as the equivalence pollution load method were coupled to identify the critical areas and primary sources for precision management of the agricultural non-point source pollution, producing the following results.
(i) The annual emissions and pollution-emission intensities of TN, TP, and COD showed a significant downward trend since 2016. Among the four classified pollution sources (i.e., farmland fertilizer, crop residues, livestock breeding, and daily activities), farmland fertilizer and livestock breeding had greater impact on the pollutant emissions.
(ii) On the basis of the temporal and spatial characteristics of the pollution emissions, combine with the comprehensive pollution index (CPI) of TN, TP, and COD of all the 10 villages, Longzhu, Shendu, Fuzi, and Maicang were identified as the critical areas for agricultural non-point source pollution management.
(iii) The equivalence pollution load ratio of different pollution sources indicated that TN and TP were the main pollutants of the four critical areas. The primary source of pollution in Longzhu and Fuzi was farmland fertilizer, while in Shendu and Maicang, the main sources of pollution were farmland fertilizer and livestock breeding.
Author contribution Wen Xu and Ling Liu wrote the original draft; Shijiang Zhu and Aihua Sun edited the manuscript text; Hao Wang and Zhiyu Ding made data analysis and prepared the pictures. All authors reviewed the manuscript.
Funding This work was supported by the Youth Fund from the National Natural Science Foundation of China (grant number: 52000120); the Key Scientific Research Projects of Water Conservancy in Hubei Province (grant numbers: HBSLKY201919 and HBSLKY202124).

Data availability
The data that support the findings of this study are available from the corresponding author, [zhusj iang@ aliyun. com], upon reasonable request.

Declarations
Ethics approval and consent to participate All authors have read, understood, and have complied as applicable with the statement on "ethical responsibilities of authors" as found in the instructions for authors and are aware that with minor exceptions, no changes can be made to authorship once the paper is submitted.