Water Pollution and Its Causes in the Tuojiang River Basin, China: An Artiﬁcial Neural Network Analysis

: This work aimed to assess the water quality of the Tuojiang River Basin in recent years to provide a better understanding of its current pollution situation, and the potential pollution risks and causes. Water quality parameters such as dissolved oxygen (DO), ammonia–nitrogen (NH 3 -N), total phosphorus (TP), the permanganate index (CODMn), ﬁve-day biochemical oxygen demand (BOD5), pH, and concentrations of various heavy metals were measured in the Tuojiang River, according to the national standards of the People’s Republic of China. Samples were collected between 2012 to 2018 at 11 national monitoring sites in the Tuojiang River Basin. The overall water pollution situation was evaluated with back propagation artiﬁcial neural network (BP-ANN) analysis. The pollution causes were analyzed considering both industrial wastewater discharge in the upper reaches and the current pollution situation. We found potential risks of excessive NH 3 -N, TP, Cd, Hg, and Pb concentrations in the Tuojiang River Basin. Moreover, corresponding water pollution control suggestions were given.


Introduction
With the rapid urban and industrial development, water pollution has gradually become a serious issue [1][2][3][4][5]. Many water-related diseases such as Minamata disease and bone pain disease have emerged together with other water pollutants, causing severe harm to human life and health [6]. Generally, water pollution can be classified as organic matter pollution [7], excessive nutrients, mainly nitrogen and phosphorus [8,9], and heavy metal pollution. In particular, the rapid economic development in Southwest China has led to a series of imbalances between the natural environment and industrial development [10]. According to the Environment Bulletin of Sichuan Province in 2019 [11], there are issues of excessive cadmium, lead, and mercury in the Tuojiang River Basin. The conflict between water resource conservation and industrial development in the Tuojiang River Basin is a typical contradiction between human beings and nature. To control the problems and risks associated with excess heavy metals in the Tuojiang River Basin, it is necessary to understand the pollution causes and potential risks to the Tuojiang River Basin's water resources.
Environmental pollution can be identified from environmental variables using single factor and multi-factor analysis methods. A representative of the single-factor method is the Nemerow pollution index [12]. However, single-factor analyses are associated with many problems, mainly because they conduct an isolated evaluation of various impact factors, which may not accurately represent a comprehensive, multiple-stressor water quality situation [13]. Multi-factor analysis methods include the fuzzy analysis method [14][15][16][17][18], the grey system analysis method [19], the hierarchical analysis method [20], the ecological footprint method [21], and the artificial neural network (ANN) analysis method [22]. Multi-factor analyses can provide comprehensive water quality information by assigning can learn autonomously, reduce the error, and make the output accurate and idealized. Therefore, BP-ANN has become a hot topic in many research fields.
This paper aimed to use BP-ANN to comprehensively analyze the water quality in the Tuojiang River Basin. The temporal and spatial analysis of the measured variables and the output data of the BP-ANN were used to assess the overall pollution situation and to identify the pollution index that represents the greatest impact on the environment. By analyzing the factors of wastewater discharge in the upper reaches of the river and phosphorus pollution by industrial and mining operations, the river's self-purification, and seasonal variation in pollutants in the Tuojiang River Basin, this paper assessed the pollution causes in this basin in the last ten years. This paper's results and conclusions may be useful for the strategic planning of water resource utilization and ecologicalenvironmental development for the cities and counties in the Chengdu Plain in the future.

Data Acquisition
Water quality data were collected from 11 sampling sites along the Tuojiang River ( Figure 1). These sites were all national first-class water quality monitoring sites. The sites were densely distributed in the upper and lower reaches of the river, but sparse in the middle reaches. The sections between the Jinghu and Shiqiao sites in the upper reaches belong to Deyang and Chengdu's municipalities urban divisions and are characterized by substantial domestic sewage and industrial wastewater inputs. The sections from Chengdu to Ziyang are characterized by rural land use, with very few villages. The water quality in these sections is relatively stable, so fewer sampling sites have been established in the middle reaches. The sections from Neijiang to Luzhou in the lower reaches, where the Tuojiang River is about to converge with the Yangtze River, are densely populated. The pollutant loads of Tuojiang River represent a significant impact on the water quality of the Yangtze River. autonomously, reduce the error, and make the output accurate and idealized. Therefore, BP-ANN has become a hot topic in many research fields. This paper aimed to use BP-ANN to comprehensively analyze the water quality in the Tuojiang River Basin. The temporal and spatial analysis of the measured variables and the output data of the BP-ANN were used to assess the overall pollution situation and to identify the pollution index that represents the greatest impact on the environment. By analyzing the factors of wastewater discharge in the upper reaches of the river and phosphorus pollution by industrial and mining operations, the river's self-purification, and seasonal variation in pollutants in the Tuojiang River Basin, this paper assessed the pollution causes in this basin in the last ten years. This paper's results and conclusions may be useful for the strategic planning of water resource utilization and ecological-environmental development for the cities and counties in the Chengdu Plain in the future.

Data Acquisition
Water quality data were collected from 11 sampling sites along the Tuojiang River ( Figure 1). These sites were all national first-class water quality monitoring sites. The sites were densely distributed in the upper and lower reaches of the river, but sparse in the middle reaches. The sections between the Jinghu and Shiqiao sites in the upper reaches belong to Deyang and Chengdu's municipalities urban divisions and are characterized by substantial domestic sewage and industrial wastewater inputs. The sections from Chengdu to Ziyang are characterized by rural land use, with very few villages. The water quality in these sections is relatively stable, so fewer sampling sites have been established in the middle reaches. The sections from Neijiang to Luzhou in the lower reaches, where the Tuojiang River is about to converge with the Yangtze River, are densely populated. The pollutant loads of Tuojiang River represent a significant impact on the water quality of the Yangtze River.  total phosphorus (TP), five-day biochemical oxygen demand (BOD 5 ), the permanganate index for chemical oxygen demand (CODMn) as well as copper, zinc, arsenic, cadmium, hexavalent chromium, mercury, and lead. Sampling sections were selected to avoid local pollution sources and other factors that could cause water pollution spikes. The sampling process was in strict accordance with the surface water-related monitoring technical specifications and national standard methods. The water samples were quickly brought back to the laboratory for analysis and determination within 24 h after sampling.

Back Propagation (BP) Artificial Neural Network Analysis
In previous water quality assessments using BP-ANN in Chinese rivers, the concentration limits of the five water quality classes of the Chinese national surface water environmental quality standard were used as the input eigenvalues for sample learning. A disadvantage of this approach is that the upper limit of water quality class I was used as an evaluation standard, and therefore could not accurately represent water quality better than class I. Thus, the lower limits of class I were added to the learning process to solve this problem. The evaluation sample dataset is shown in Table 1. A BP-ANN model with 12 input nodes, six hidden layers, and one output node was then established in the MATLAB software (version 2016a) ( Figure 2). The random assignment of initial learning parameters, initial weights, and thresholds was carried out using the rand function, and water quality standards (Table 1) were used as learning samples. The learning time was set to 10 5 , while the error limit was 10 −8 .     After 444 learning and training cycles, the accuracy requirement of e < 10 −8 was reached with a total time consumption of less than 1 s (Figure 2A). Other training parameters are shown in Figure 2B-D. The trained sample output is shown in Table 2. After learning and training, the BP-ANN analyzed the annual average variation and the average value of each monitoring site using the data from the 11 monitoring sites from 2012 to 2018. The comprehensive water quality analysis results are shown in Tables 3 and 4.

Sensitivity Analysis
According to Tables 3 and 4, we obtained Tables 5 and 6 by means of BP-ANN. In order to figure out which independent variables involved in the BP-ANN evaluation had a greater contribution to water quality, we adopted the network sensitivity analysis to achieve this goal.  Considering Table 5 (2012 to 2017 as the training sample and 2018 for sensitivity  analysis) as an orthogonal test table, we obtained the preliminary sample data. The 12 independent variables involved are denoted as X1 to X12. Then, all preliminary data samples need to be normalized according to the formula below: Y presents the data, which has been normalized. The normalized results are shown in Table 7 and the final sensitivity analysis results are shown in Table 8. According to Table 8, heavy metals have a significant impact on water quality, especially As and Hg.

Analysis of BP-ANN, Temporal, and Spatial Pollution Distribution
According to Table 5, the comprehensive water quality score decreased from 2012 to 2018. By 2018, the average water quality score of the Tuojiang River Basin was less than 2, reaching class I. Compared with 2012, the comprehensive water quality score in 2018 was 0.65% lower, and the average annual water quality score was 0.093% lower. Table 6 shows that the water quality of monitoring sites A and B reached class I, while all other monitoring sites had class II. From the upper reaches, beginning at site A, to the lower reaches at site K, the water quality score increased by 0.99%, and the average water quality score of each monitoring site was 0.09% higher than that of the previous station. The water quality score of sites B and C increased most, indicating the most severe pollution (0.5%). However, the water quality scores decreased from stations D to E, F to G, and H to I, which may be attributed to the improvement of water quality due to self-purification in the Tuojiang River Basin's middle reaches.
According to Table 9, each monitoring site's water quality scores decreased year by year. Most monitoring sites reached water quality class I in 2018, except for sites H and K. All the monitoring sites of Tuojiang River reached water quality class III. The water quality of monitoring site A always remained at class I, while site B's water quality assessment remained between 1.999 and 2.002. The water quality score of sites A and B in 2018 decreased by 0.02% and 0.2%, respectively, compared with 2012. Therefore, the water quality of monitoring sites A and B can be considered stable. Since 2012, the water quality score of sites C to K (excluding E) gradually decreased. By 2018, the water quality scores of these sites had decreased by 0.86%, 1.99%, 0.72%, 0.75%, 0.75%, 0.83%, 1.10%, and 1.17%. The water quality of site E increased initially and then decreased between 2012 and 2018. The value reached a maximum value in 2015, and the final water quality score increased by 0.015%. According to Figure 3, the average concentration of DO at the four monitoring sites A, B, C, and D in the upper reaches remained stable at 8 mg·L −1 . In contrast, the DO concentrations in the lower reaches decreased gradually. The concentrations of stations H, I, J, and K were low and ranged between 5.5 and 6.5 mg·L −1 . The pH value of the Tuojiang River was between 7.5 and 8.6, while the minimum value of 7.5 was found at sites D and E, and the highest value of 8.6 was observed at sites G, H, and I. COD Mn remained between 2.5 mg·L −1 and 5.5 mg·L −1 . The minimum value occurred at site A and the maximum values at sites J and K in the lower reaches. There was a risk of exceeding the 6 mg·L −1 class III limit for COD Mn . BOD 5 fluctuated between 2.0 mg·L −1 and 5.0 mg·L −1 . The spatial pattern of BOD 5 was contrary to that of COD Mn . The monitoring sites A, B, and C in the upper reaches had the maximum values of BOD 5 , while the concentrations in the lower reaches were lower. There was no risk of BOD 5 exceeding the class III limit (<20 ppm). The highest concentrations of NH 3 -N were found at sites E and F in the middle reaches and exceeded the class III limit of 1.0 ppm. The concentrations of NH 3 -N in the lower reaches were lower than those at sites A and B, which indicated that NH 3 -N discharge was concentrated in the middle and upper reaches of the Tuojiang River. The concentration range of NH 3 -N was 0.25-3.0 mg·L −1 . The TP values exceeded the water quality class III limit of 0.2 mg·L −1 at all stations. Maximum values of 0.3 mg·L −1 occurred at sites B, C, D, and E in the middle and upper reaches. The lowest TP concentrations were found at sites A, I, and J, but also exceeded 0.2 mg·L −1 .
According to Figure 4, Cu concentrations were low in the upper reaches (0.005 mg·L −1 ), and higher concentrations were observed in the middle and lower reaches (0.03 mg·L −1 ). All these values were far below the concentration limit of 1.0 mg·L −1 . The maximum concentration of Zn was 0.045 mg·L −1 , observed at sites C, D, and E. The minimum Zn concentration of 0.005 mg·L −1 at sites A, I, and J was far below the 1.0 mg·L −1 class III concentration limit. The maximum concentration of As was found at sites A, B, C, D, and E in the upper reaches with an average concentration of 0.004 mg·L −1 , while the low concentration of As occurred at sites G, H, and I with an average concentration of 0.0005 mg·L −1 . All these values were far below the class III limit of 0.05 ppm. Maximum Hg concentrations were recorded in the upper reaches of the Tuojiang River between sites A and E. According to Figure 4, there may exist a risk of high Hg concentrations in the sampling sites in southwest Deyang. The average concentration of Hg in the upper reaches was 0.00008 mg·L −1 , which was close to the limit of 0.0001 mg·L −1 , indicating the risk of exceeding this standard. The minimum value of 0.00001 mg·L −1 was found at sites in the middle and lower reaches. According to Figure 4, there was a risk of high Cd concentration throughout the Tuojiang River Basin, especially for the cities of Chengdu and Deyang, with average concentrations of >0.004 mg·L −1 , close to the limit of 0.005 mg·L −1 . Cr concentrations were about 0.01 mg·L −1 throughout the Tuojiang River Basin, which was much lower than the limit of 0.05 mg·L −1 for class III water quality. High Pb concentrations were observed in the middle and lower reaches, sites F to K, with an average concentration of 0.02 mg·L −1 . There was a slight risk of exceeding the class III limit in the lower reaches. The minimum value of 0.005 mg·L −1 at sites in the upper reaches was lower than the limit of 0.05 mg·L −1 .   In summary, according to the BP-ANN based evaluation and temporal and spatial analysis, there exists the risks of (a) COD Mn and Hg concentrations higher than class III limits in the upper reaches, (b) excessive Cd concentrations in the entire Tuojiang River basin, (c) excessive Pb in the lower reaches, (d) excessive NH 3 -N in the region between the Deyang and Chengdu sites in the upper reaches, and (e) excessive TP concentrations in the entire Tuojiang River Basin.

Effects of Industrial Phosphorus Pollution
Shifang City, located in the upper reaches of Tuojiang River and southwest of Deyang City, is one of the six major phosphate mining sites in China. The phosphorus production accounts for more than 90% of the gross product of Sichuan Province. The proven reserves in Shifang City amount to 170 million tons, and there are more than 40 phosphorus related enterprises. The annual output of small-scale enterprises is more than 10,000 tons, while the production of large-scale enterprises is about 400,000 tons. The application of phosphate fertilizer and the phosphorus industry may be potential sources of the observed excessive heavy metal concentrations compared with national standards.
Human activities greatly contribute to heavy metals in water and soil. Non-ferrous metal mining, smelting, electroplating, glass manufacturing, the battery industry, printing, and dyeing discharge cadmium, mercury, lead, and other heavy metals. The heavy metal discharge caused by human activities has exceeded natural emissions, and currently, human activities are the main source of heavy metal pollution [40]. Zhang et al. [41] found that the phosphorus industry was the main factor that caused excessive Cd and Pb concentrations in the environment. There are many mining areas in the upper reaches of the Tuojiang River, and their effect on water quality should not be underestimated, as shown in Table 10. The phosphorus industry's development in the upper Tuojiang River Basin can also explain the entire basin's excessive TP concentrations.

Sediment Settlement
The bioavailability and toxicity of heavy metals in water bodies mainly depend on the activity of free heavy metal ions, that is, the fraction of soluble or exchangeable metals in water, rather than the total mass of such heavy metals [42][43][44]. Most heavy metals in water and soil exist in an insoluble state, and the pH value controls their solubility. A decrease in the pH value can lead to the dissolution and release of heavy metals combined with carbonate and hydroxide. It can also increase the release of adsorbed heavy metals. The availability of cadmium in water, soil, or sediment, namely its chemical forms and adsorption-desorption behavior, are regulated mainly by environmental pH. With increases in environmental pH, negative charges in sediment colloids increase, while the competitive effect of H + is weakened. Thus, heavy metals are bound more firmly, mostly in the form of insoluble hydroxide or carbonate and phosphate. As a result, cadmium, mercury, and lead's effectiveness are greatly reduced [45]. The Tuojiang River Basin is about 600 km long. The widest part of the river is tens of meters, and the depth can reach tens of meters. The river sediment is rich in nutrients. Combined, the concentration variations of heavy metals including cadmium and mercury, and high pH values cause an effective reduction of heavy metal concentrations due to sedimentation.

Seasonal Variations
As mentioned previously, the pH value and carbonate ligands are among the major factors affecting the speciation of heavy metals in the water body. Figure 5 show the variation of pH in the Tuojiang River between 2012 and 2018 and the total hardness in four seasons from 2015 to 2018.
the Deyang and Chengdu region as well as further data analysis.
As shown in Figure 7, water hardness is most affected by summer, then by winter, and finally by autumn. In addition, significant changes in the standard deviation of the six upstream monitoring stations may be linked to high urbanization (near Chengdu), so the water quantity varied greatly. The downstream water flows through rural and suburban areas, so the water quality and quantity gradually become stable.
According to seasonal variation of pH, it is noteworthy to find summer has a different change trend from other seasons. Namely, most mentoring sites have a rising trend in summer. Meanwhile, summer also has the greatest impact on the overall water hardness. At least, we could learn that for water quality control, summer is a special season, which is worth more attention.  The pH values of the six sampling sites in the lower reaches of the Tuojiang River showed little variation among seasons (Figure 6), while the water quality of the five sampling sites in the upper reaches varied seasonally. The pH values of Shiqiao and Sanhuangmiao were most significantly affected by the seasons. The water quality of Shiqiao and Sanhuangmiao was affected by exogenous pollutants, which is consistent with industrial wastewater and domestic sewage discharge from the cities of Qingjiang to Sanhuangmiao. Total hardness values of the river were similar in spring and winter as well as in summer and autumn ( Figure 5). Variations in total water hardness may be affected by water temperature or other contaminants [46,47]. Water quality is complex and variable. A variety of inorganic salt ions and heavy metals may combine with ligands and settle, affecting the total hardness of the river water. According to a previous study at Jinghu, the best water quality monitoring section was throughout the whole Tuojiang River Basin. However, in Figure 5, Jinghu's average total hardness was the highest among the five sampling sites in the upper reaches. The unusual trend of total hardness variation in the upper reaches of the Tuojiang River needs further study by sampling the river in the Deyang and Chengdu region as well as further data analysis.
As shown in Figure 7, water hardness is most affected by summer, then by winter, and finally by autumn. In addition, significant changes in the standard deviation of the six upstream monitoring stations may be linked to high urbanization (near Chengdu), so the water quantity varied greatly. The downstream water flows through rural and suburban areas, so the water quality and quantity gradually become stable.   According to seasonal variation of pH, it is noteworthy to find summer has a different change trend from other seasons. Namely, most mentoring sites have a rising trend in summer. Meanwhile, summer also has the greatest impact on the overall water hardness. At least, we could learn that for water quality control, summer is a special season, which is worth more attention.

Conclusions
In this work, 11 national monitoring sites were sampled from 2012 to 2018 in the Tuojiang River Basin in the eastern Chengdu Plain. The BP-ANN method was used to determine the comprehensive water quality and analyze the temporal and spatial variations of water quality variables including DO, COD Mn , BOD 5 , NH 3 -N, TP, and various heavy metals such as copper, zinc, arsenic, mercury, cadmium, hexavalent chromium, and lead. The temporal and spatial analysis (Figures 3 and 4) suggested pollution and risks of excessive COD Mn , NH 3 -N, TP, cadmium, mercury, and lead concentrations compared to the national standards, and network sensitivity analysis revealed that for heavy metals, most change in the water quality was most sensitive to heavy metal content, especially As and Hg.
Combining factors such as industrial phosphorus production and mining in the upper reaches, seasonal water quality variations, speciation of heavy metals in the water and sediment, and potential plant absorption and enrichment, this study explored the temporal and spatial trends of water quality and various analytical indicators. It appears that the phosphorus industry in the upper reaches of the river is the main reason for the excessive contents of TP and heavy metals in the river water. The reduction in the concentrations of pollutants (e.g., mercury and cadmium) in the middle and lower river reaches could be due to self-purification and the enrichment of plants along the Tuojiang River.
For further investigation on water quality, we think that further study should strengthen sampling and monitoring of sediment and aquatic plants to prove our analysis results as well as pay special attention to the concentration of cadmium, mercury, and lead in the sediment. Furthermore, how pH and hardness impact on metal content in water is worth more attention. To control the water pollution of the Tuojiang River, concerned agencies should plant a large number of green plants along this river and reduce upstream industrial emissions.
Potential limitations of this study are due to the fact that some data (such as those from 2012) were obtained very early, and there are inevitably minor measurement errors. In this study, water quality analyses from 2012 to 2018 provided specific data and method references for future studies of the whole basin or parts of the Tuojiang River Basin. This study may also offer new perspectives to investigations in the broader area of water quality.