The Oujiang River is located in the south of Zhejiang Province, China. The basin ranges from 118°45′E - 121°00′E and 27°28′N - 28°48′N in China with drainage area of 18,100 km², mainly including the Lishui City and Wenzhou City. The main river has a length 388 km and an elevation drop of 1.3 km. The average annual flow at the estuary is approximately 612 m³/s. The main tributaries of the Oujiang River are distributed in the form of dendrites. The area is dominated by in industrial and agricultural land, and urban settlements. In addition, as an important water source and sewage discharge channel in the area, it plays a pivotal role in economic development and people’s lives. In recent years, the continuous development of the economy and the acceleration of industrialization have brought more and more serious water pollution to the Oujiang River.

Water samples were collected from free-flowing reaches (FFRs) between January 13-20, 2021 and June 10-17, 2021. A total of 26 surface water samples were collected in the whole basin of the Oujiang River, including 18 from the main stream and 8 from the main tributaries ( Supplementary Figure S1 ). In addition, water samples, manure, and soil samples from pollution sources were collected. A total of 50 enterprise factory samples, including paper mills (Pr), chemical plants (Cl), stainless steel plant (SS), electroplating factories (Plating), leather factories (Fc), Slaughterhouse (Sr), detergent factories (Dt) and breweries (Br). Additionally, a total of 50 manure samples were collected from farming production activities, including chicken (C), pig (PigM), duck (D), and goat farms (G), and additional tail water samples were collected from piggeries (PigW). Meanwhile, the tail water generated by human life was collected through the pipe network to the sewage treatment setup kind, and a total of 30 samples of tail water from sewage treatment plants (WWTPs) and 6 samples of rural domestic sewage treatment terminals (RSTT) were collected. In addition, agricultural cultivation is an important source of nitrogen pollution, and a total of six rice samples (RP), six JiaoBai samples (JBW) and six vegetable samples (V) were collected, as well as 12 natural soil samples (Soil). And 6 tea soil samples were collected in both farmland(TeaF) and hilly areas (TeaU). Meanwhile, 6 tailwater samples were collected from aquaculture plants (Ac).

The pH, water temperature (T, ˚C), electrical conductivity (EC), dissolved oxygen (DO), and oxidation-reduction potential (ORP/Eh, mV) were measured in situ using a multiparameter portable meter (HACH40d, USA) supplied by Hach Company (Loveland, CO, USA). All water samples were collected at 0.5 m below the water surface using two polyethylene bottles and preserved at 4˚C. They were then taken back to the laboratory and analyzed within 24 hours. Concentrations of chloride (Cl⁻), nitrate (NO₃ ⁻), nitrite (NO₂ ⁻), and ammonium (NH₄ ⁺) were analyzed using ionchromatography (Dionex ICS-600). The δ¹⁵N-NO₃ ⁻ and δ¹⁸O-NO₃ ⁻ values were obtained using the denitrifier method, by which NO₃ ⁻ was converted into nitrous oxide by Pseudomonas aureofaciens, and then detected using an isotope ratio mass spectrometer (IRMS,Isoprime100, Cheadle, UK) after concentration and purification by the trace gas system (Liu et al., 2018).

The total nitrogen point sources within the study area included industrial wastewater, urban domestic sewage, and effluent from large-scale livestock and poultry farms, totaling 593 sites (( Supplementary Figure S2 ). Point source pollution refers to pollutants that originate from fixed discharge locations and are distributed in a point-like fashion. Generally, this includes industrial wastewater, urban domestic sewage, and emissions from large-scale livestock and poultry operations.This study calculated the total nitrogen emissions from point sources using statistical methods based on data from the 2022 Environmental Statistics, Statistical Yearbook, and the Fixed Pollution Source Automatic Monitoring Information Management Platform.Considering the loss due to evaporation, sedimentation, and other transformation processes that pollutants undergo after being discharged into water bodies, the load that ultimately impacts the water bodies differs from the initial emissions. The method of using river entry correction coefficients is employed to quantify the amount of pollutants that actually enter the river (Dumortier et al., 2019).

Using ArcGIS and BIGEMAP for point source latitude and longitude measurement correction, and based on the distance to the river combined with the National Water Environment Capacity Determination Technical Guide, the river entry coefficients for this study area were determined, as shown in Table 1 .

Non-point source pollution originates from dispersed sources, and its emission pathways are complex, often displaying randomness over time and uncertainty in space. Generally, this includes sources such as rural domestic sewage, dispersed livestock and poultry farming, agricultural runoff, and urban stormwater runoff (Fan et al., 2024).

This study employed the modified Jones export coefficient model to calculate the load of non-point source pollution (Equation 1) (Johnes, 1996):

L is the pollutant emission load in the study area, t/a;i is the population, livestock population, and soil type in the study area, i =1,2,3, ···, n; ω i is the conversion coefficient, such as domestic sewage or fertilizer application into the river coefficient/livestock and poultry farming sheep, cattle, rabbits converted to pigs conversion ratio, etc., dimensionless; E i is the ith species of livestock and poultry/soil type/unit population and other export coefficients, the unit of the above three are kg/(head or only · a), t/(km²·a), kg/(person · a), mainly refer to the National Technical Guidelines on Approval of the Capacity of Water Environments, Manual on Methods and Coefficients for Accounting for Emissions from Agricultural Sources, Manual on Production and Emission Coefficients for Domestic Pollution Sources, Emission Standards for Pollutants from Livestock and Poultry Farming (DB 33/593-2005), and related watershed background literature and field research to match research results information acquisition. A i is the number of livestock and poultry of the ith species/area of land class/population in heads or animals, km², and people, respectively; The second term of the equation is the rainfall runoff input pollution load calculation, φ j is the jth field precipitation correction coefficient; P j is the field rainfall of the jth precipitation event affecting surface runoff, mm; E M C j is the average concentration of the jth precipitation event, mg/L; S is the runoff catchment area, km². The values of each export coefficient are determined as shown in Table 2 .

Event Mean Concentration EMC in the above equation means the surface runoff pollution from precipitation at a site divided by the total amount of runoff under that site, calculated by the formula (Equation 2):

c j is the pollutant concentration in time period j, mg/L; v j is the runoff volume in time period j, m³.

Surface runoff from precipitation processes, especially heavy rainfall, impacts surface pollution systems by transporting pollutants, such as sediment and litter, into them (Wang et al., 2018).

The DNA from filter membrane was extracted with the Qiagen DNeasy PowerWater Kit. DNA concentration and purity were monitored on 1% agarose gels. Based on Qubit quantification, DNA was diluted to 10 ng/µL with sterile water. The V3-V4 region of the microbial 16S rRNA gene was amplified using the universal primers 341F (5′-CCTAYGGGRBGCASCAG-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3′). An equal volume of 1X loading buffer containing GelRed^® was mixed with PCR products, and electrophoresis was performed on a 2% agarose gel. Target bands (~550 bp) were excised and purified using the Qiagen Gel Extraction Kit (Qiagen, Germany). Purified amplicons were pooled in equimolar ratios and sequencing libraries were constructed using the NEBNext^® Ultra™ II Q5^® Master Mix (M0544) following the manufacturer’s protocol. The library quality was assessed using the Qubit^® 2.0 Fluorometer and Agilent Bioanalyzer 2100 system with High Sensitivity DNA chips confirming a peak size of ~550 bp. Each PCR batch included negative and positive controls. Finally, libraries were sequenced on an Illumina NovaSeq platform, generating 250 bp paired-end reads that were merged using FLASH v1.2.11.

The 16S rRNA gene sequence data were processed by using two pipe-lines, LotuS (Hildebrand et al., 2014) and QIIME v1.9.1 as described previously. In brief, the paired-end reads were quality trimmed, merged, and clustered into operational taxonomic units (OTUs) with a threshold of ≥ 97% similarity. The reads of each OTU were classified using the Ribosomal Database Project classifier (80% confidence) based on the SILVA database v132. Quantitative analysis was performed using thespecies sequence abundance data from each sampling site. In order to standardize the uneven sequencing effort, all samples were randomly subsampled to the smallest library sizes with 32000 reads for microeukaryotic, respectively.

In the present study, bacteria from 24 fecal samples, 120 wastewater samples and 35 soil samples were defined as a potential source library and the bacteria from 36 surface water samples were defined as sink. The community-based MST analysis was performed using SourceTracker v. 1.0.1 in RStudio v. 0.99.896 with default parameters. Briefly, SourceTracker was run with the parameters of rarefaction depth 1000, burn-in 100, restart 10, alpha 0.001 and beta 0.01, which has been demonstrated to be high sensitivity, specificity and accuracy (Henry et al., 2016; Li et al., 2018). We calculated a diversity and b diversity using the ‘vegan’ package (v 2.6-4) in R. Non-metric multidimensional scaling analysis (NMDS), similarity analysis, α diversity index calculation, and mantel text analysis were performed using the R platform to determine the significance of differences in microbial community composition at different periods and the effects of environmental factors on microbial communities. Within the SPSS 22.0 platform, analyses were performed using the Kruskal-Wallis test and Spearman correlation.

Based on the data and coefficient corrections, the results shown in Figure 1 . 7 indicate that in 2022, the point source emissions in the study area were primarily from urban domestic sources, amounting to 1.29×10^8 tons and accounting for 85.72% of the total. Industrial wastewater emissions were the second largest, totaling 1.63×10^7 tons and accounting for 10.79%. Emissions from other sources such as livestock and poultry farming were minimal.In the emissions of wastewater pollutants, the highest point source emissions come from domestic sources, followed by industrial sources. Among these, COD (Chemical Oxygen Demand) emissions account for the majority, totaling In the COD emissions, domestic sources contribute 96.78%, industrial sources 3.16%, and livestock and poultry farming sources 0.06%. The total NH3-N emissions across the entire watershed amount to 12,100 tons, with domestic sources accounting for 98.42%, industrial sources 1.48%, and livestock and poultry farming sources 0.10%.

The total TN (Total Nitrogen) emissions across the entire watershed amount to 2,870 tons, with domestic sources contributing 2,730 tons, accounting for 95.46%, industrial sources 127 tons, accounting for 4.42%, and livestock and poultry farming sources 3.3 tons, accounting for 0.12%. TP (Total Phosphorus) emissions are relatively lower, with domestic sources emitting 161 tons, industrial sources 2.7 tons, and livestock and poultry farming sources 0.04 tons.

According to statistics from the “Zhejiang Province Pollution Source Automatic Monitoring Information Management Platform,” the average concentrations and flow emissions data of COD, NH₃-N, TN, and TP from 2016 to 2022 were used to calculate the pollution load emissions from significant industrial enterprises in the study area. The results are displayed in Figure 2 . Over time, except for NH₃-N, the trend of pollutant emissions has generally increased and then decreased in recent years, reaching a peak in 2019 and 2020. In contrast, NH₃-N emissions have shown an overall declining trend. TN emissions peaked in 2019 and 2020 at 1834.9 and 1825.6 tons per year, respectively, with the lowest value recorded in 2016 at 170.3 tons per year. The emissions have also declined in recent years. Spatially, pollutant emissions are concentrated in the upper and middle reaches of the watershed, with tributaries primarily focused on the upper tributary, Songyin Creek. Notably, TN load emissions at some points along Songyin Creek exceed those in the main river, with the highest value recorded at 256.5 tons per year ( Figure 1 ).

To facilitate the identification of spatial distribution characteristics of pollutants, the natural breaks method was used to reclassify the non-point source pollution load within the study area. The resulting spatial distribution is shown in Figure 3 . The total annual loads of non-point source COD, NH3-N, TN, and TP are 10,689.857 tons, 535.781 tons, 2,699.555 tons, and 131.197 tons, respectively. High COD load areas are primarily located in the upstream tributaries of Songyin Creek and the mid-lower reaches of the main river, predominantly from rural domestic discharges, accounting for 54.13% of the load. The distribution characteristics of NH3-N loads are similar to those of COD, with rural domestic sources accounting for 55.36% of the NH3-N load. TN emissions are concentrated in the midstream tributaries, where, as previously mentioned, various types of agricultural pollutants have significant contributions. These midstream tributaries are covered with various types of forests, such as forest lands, open woodlands, shrub lands, and forests with varying degrees of cover. The load entering the river from these sources accounts for 82.01% of the total, while rural domestic sources contribute a smaller proportion, at 15.52%. This indicates that the amount of TN entering the river is heavily influenced by rainfall runoff, with TN wash-off amounting to 20.723 tons per year. This may be associated with relatively high background concentrations of TN in the study area, as analyzed in section 2.2.1, where TN concentrations have been high across different years. This underscores the necessity of establishing regulatory limits for TN loads in the study area. The TP load in the study area ranges from 0.147 to 6.323 tons per year, with the main type of pollution source being agricultural activities. The proportions of the other two types of non-point source pollution are relatively equal.

Based on the calculation results of the export coefficient model, potential nitrogen pollution sources were screened out to provide scientific basis for further MST of nitrogen pollution sources.

The results of total nitrogen, ammonia nitrogen and nitrate nitrogen concentrations measured in the source wastewater are shown in Figure 4 . TN concentrations in the WWTP effluent ranged from 6.25 to 12.3 mg/L, with an average concentration of 8.98 mg/L; NO₃-N concentrations ranged from 0.14 to 8.45 mg/L, with an average concentration of 4.62 mg/L; NH₃-N concentrations ranged from 0.04 to 0.334 mg/L, with an average concentration of 0.03 mg/L. TN concentrations in the RSTT effluent ranged from 10.3 to 20.5 mg/L, with an average concentration of 16.4 mg/L; NO₃-N concentrations ranged from 0.9 to 3.08 mg/L, with an average concentration of 1.99 mg/L; NH₃-N concentrations ranged from 8.92 to 18.78 mg/L, with an average concentration of 13.85 mg/L. In addition, the average concentration of total nitrogen in the effluents of enterprises in the sectoral industries is high. For example, the stainless steel industry has a total nitrogen concentration of 12.4 mg/L, the electroplating industry has a total nitrogen concentration of 28.4 mg/L, the brewery has a total nitrogen concentration of 11.5 mg/L, the slaughterhouse has a total nitrogen concentration of 30.2 mg/L, and the chemical industry has a total nitrogen concentration of 55.05 mg/L. However, the total nitrogen concentration of the tailwater from the washing industry (0.47 mg/L) is low. The total nitrogen concentration of 7.990 mg/L in the tailwater of the agricultural crop Zizania field is a significant source that cannot be ignored. It is worth noting that livestock farming, which is an important human economic activity, should be given priority attention as the total nitrogen concentration in the tailwater of pig farms reached 526 mg/L.

Overall, 211558 OTUs were obtained from 180 samples by 16S rRNA gene amplicon sequencing, respectively. As shown in Figure 1a , Seven typical sources of pollution are included. At the phylum level, all contaminants consisted mainly of Proteobacteria, Firmicutes, Bacteroidota, Actinobacteriota and Acidobacteriota, with relative abundances greater than 50% ( Figure 7a ). At the genus level, there were significant differences in the composition of the flora from different sources of contamination ( Figure 7b ). Among the source categories of corporate plant pollution, the dominant genera in stainless steel plants are Rhodobacter (26.19%) and Brachymonas (17.8%). The advantageous bacterial genera of the brewery is Aquicella (13.63%).The advantageous microbial community of slaughterhouses is Leeia (26.20%). The excellent microbial community of the synthetic leather factory is unclassified_Anaerolineae (12.57%) and Nitrospira (5.61%). The advantageous microbial communities of cleaning agents, soap, and toothpaste factories are uncultured_soil_bacterium (7.49%). The excellent microbial community of leather substrate factory is Hydrogenophaga (5.77%). The advantageous microbial community of the paper mill is unclassified_Saccharimon (16.34%). The advantageous microbial community of electroplating enterprises is unclassified_Gemmatimo (2.06%).

The dominant bacterial group in the wastewater of pig farms is Comamonas (3.62%), and Acinetobacter (20.27%) is present in the manure. The dominant microbial community in chicken manure is Acinetobacter (19.73%). The dominant bacterial group in chicken manure is Acinetobacter (18.30%), in sheep manure is Halomonas (7.76%), and in duck manure is Brachybacterium (12.70%). The dominant microbial community in aquaculture effluent is hgcI_clade (8.45%) and Limnohabitans (11.49%). The composition characteristics of fecal microbiota differ from the dominant microbial community of chicken manure samples collected in the Chaohu Lake Basin in other studies (chicken manure, Escherichia-Shigella), while pig manure samples are similar (pig manure, Acinetobacter) (Qi et al., 2023). Indicating regional differences in microbial community characteristics.

The dominant microbial community in natural soil is unclassified_Bacteria (11.38%~13.86%) and unclassified_Vicinamibacter (4.63%~5.34%). The dominant microbial communities in tea, rice, and vegetables are unclassified_Bacteria (5.52%~5.96%). The dominant bacterial community in the tail water of Jiaobai field is hgcI_clade (10.16%), and the dominant bacterial community in the sediment is uncultured_soil_bacterium (4.40%). The dominant microbial communities in sewage treatment plants are mainly Methylophilus, Lachnospiraceae_NK4A13 Bacteroides, and Bacteroides.

Pollution sources can be categorized into four major groups: Manure, Sewage, Agricultural Planting, and Soil. Additionally, we further distinguished specific pollution sources within these four categories. Using contribution networks, we elucidated the direct and indirect interactions among coexisting microbial taxa in environmental samples, documenting the co-occurrence patterns within complex and diverse communities. Based on this, we constructed a microbial fingerprinting map for each pollution source.

The machine learning classification program SourceTracker v1.0.1 was used to identify potential major input sources of bacterial communities associated with Nitrogen sources present in surface water. During the dry season, the proportion of nitrogen-contaminated surface water from Sewage sources is 47.3% ( Figures 10a, b ), from Manure sources 16.6%, from Agricultural sources is 8.5% and from natural Soil sources is 15.3%. However, there is a marked variation in the wet season, with a significantly higher proportion of contributions from agricultural (20.5%) and natural soil sources (27.8%). The percentage of Sewage sources is 22.6%, despite relatively high water use in the summer months. SourceTracker might underestimate the impact of unidentified sources, particularly for receivers with additional unknown sources. Environmental features that were not considered in the source tracing process, such as the effects of amoebae in sediments and the hydraulic relationship between streams and sediments, are designated as “unknown” (Chen et al., 2023). Previous studies found that SourceTracker could effectively predict manure sources in lakes when considering native taxa, and a higher level of false positive predictions occurred when not considering native taxa (Brown et al., 2019). Moreover, SourceTracker shows good pollution identification and distribution capabilities for different types of industrial pollution sources (Peng et al., 2024). Some studies have shown that agricultural areas can also experience fecal contamination, making it a major source of nitrogen pollution (Zimmer-Faust et al., 2025). In addition, long-term fertilizer application will change soil properties, and transport and leaching during rainfall will contribute agricultural sources that can be well calculated by the sourcetracker model (Zhang et al., 2024). Thus, microbial traceability is able to quantify the contribution of pollution sources well in watersheds with multiple complex nitrogen pollution sources. The inability of nitrogen and oxygen isotopes to differentiate between sewage and fecal sources was solved to improve the efficiency and accuracy of traceability.

Overall, microbial tracing models have better potential for identifying nitrogen pollution sources, and can identify more pollution sources compared to nitrogen oxygen isotope tracing methods. The microbiological tracing results indicate that industrial wastewater from aquaculture, machinery, chemical, textile and other industries is the main source of surface water pollution in the region. These results indicate that the export coefficient model (ECM) can serve as a screening process before microbial traceability, eliminating irrelevant sources and improving traceability efficiency. In addition, Huang et al. (2024) proposed a novel microbial source tracking tool, SourceID-NMF, which uses algorithms to identify microbial sources contributing to target samples. In future research, it can be compared with the SourceTracker model to optimize a more precise source-tracking method.