2. MATERIALS AND METHODS

2.1 Study area

The Huashan watershed is located in the lower reaches of the Yangtze River basin, with a geographical location of 118°08′05″~118°16′52″E and 32°13′14″~32°18′53″N. The watershed is surrounded by mountains on three sides, which contributes to its high degree of enclosure. The watershed has an elevation of 28 to 350m and an area of 80.13 km2. The watershed has a warm temperate semi-humid monsoon climate with an annual average temperature of 15.2℃. Precipitation is mainly concentrated from May to September, with annual precipitation of about 1043 mm and annual evaporation of about 922 mm. The downstream 16km of the watershed is the Chengxi Reservoir, the urban drinking water source of Chuzhou City, with a storage capacity of 85.3 million m³, providing domestic water for 610,000 residents (Zhang et al., 2020).
The rivers in the basin are fan-shaped and are fed by the Dongyuan Tributary, the Zhongyuan Tributary, the Xiyuan Tributary and the Zhuyuangou Tributary, respectively (Figure 1). The land use types in the watershed are mainly Orchard, Agriculture and Forest, accounting for 51.25% 24.5% and 17.74% respectively. The predominant soil types in the watershed are paddy soil and lime soil, with small amounts of yellow-brown soil and silicate coarse loam. Among them, lime soil and paddy soil account for 57.69% and 26.92%, respectively (Wang et al., 2016).
The Huashan watershed is located at the intersection of the southern end of the North China Land Platform Block and the northern part of the Yangzi-Qiantang quasi-fold belt. The regional mountainous and hilly area is characterized by Cambrian and Ordovician sedimentary limestone, and the parent rocks for soil formation include weathered products of andesite, coarse sandstone, conglomerate, limestone, and other rock types. According to the lithology, groundwater can be classified into two types: carbonate fracture-karst water and porous water in loose rocks. According to the burial conditions, the aquifer can be divided into the unconfined and confined aquifers, and this paper mainly studies unconfined aquifers. The aquifer in the watershed is mainly composed of Quaternary porous media, with lithology mainly consisting of silt clay, silty fine sand, and muddy gravel. The aquifer is relatively thick in the middle, about 10-12 meters, and thinner on both sides, with an average thickness of 7.5 meters (Zhang et al., 2022).

2.2 Sampling and analyses

Sampling was carried out seasonally, from November 2021 to October 2022, in spring, summer, autumn and winter, respectively, and the spatial distribution of sampling sites is shown in Figure 1. When collecting surface water samples, the sampling point was below 0.3-0.5 m from the water surface to above 0.3-0.5 m from the water bottom. Before sampling, the groundwater was pumped for 30 minutes until the conductivity remained constant. The pH, temperature (T), electrical conductivity (EC), oxidation-reduction potential (ORP) and dissolved oxygen (DO) were measured in site using a portable water quality analyzer (HQ30D) that had been pre-calibrated. The analyzed parameters included anions (Cl-, SO42-, NO3-, NO2-) and cations (K+, Na+, Ca2+, Mg2+, NH4+), alkalinity, δD-H2O, δ18O-H2O, δ15N-NO3-and δ18O-NO3-. Samples for analyses of NH4+were filtered by a 0.45 μm syringe-tip filter to avoid NH4+ volatilization and then were acidified with reagent quality HCl to a pH of approximately 4.0. Alkalinity was tested in the field by the Gran titration method. Samples used to measure other indicators were filtered through 0.45 μm membranes on site. After filtration, Samples used for testing anion and cation were collected in 100 ml polyethylene bottles; Samples used to test δD-H2O, δ18O-H2O, δ15N-NO3- and δ18O-NO3- were collected using 100 ml polyethylene bottles and kept cold on site by ice bag. All the sampling bottles used above were washed 3 times in advance with ultrapure water in the laboratory and rinsed 3 times with water samples to be taken during sampling, no headspace was left in the sampling bottles during the sampling process and the bottles were sealed. After sampling, all samples were returned to the laboratory, and the samples for δ15N-NO3- and δ18O-NO3- were frozen immediately and stored at -20°C. The other samples were stored at 4°C. All indicators were tested within two weeks after sampling.
Anions (Cl-, SO42-, NO3-) and cations (K+, Na+, Ca2+, Mg2+) were analyzed using ion chromatography (ICS-900), and the test results were evaluated using the charge balance error, yielding errors within ±5%. NH4+ and NO2- were analyzed using a spectrophotometer (Shimazu UV-2600) with a test accuracy of 0.01 mg/ L; hydrogen and oxygen isotopes were analyzed using a stable isotope mass spectrometer (Isoprime 100) with a testing accuracy of ±0.03‰ and ±0.02‰ for δD-H2O and δ18O-H2O, respectively. The nitrogen and oxygen isotopes of nitrate were analyzed using denitrifying bacteria method, with the bacterial strain being Pseudomonas aureofaciens (ATCC 13985). The test instrument was TraceGas coupled with continuous flow isotope ratio mass spectrometer (IRMS), with USGS32, USGS34, and USGS35 used as standards. The data was calibrated using the two-point correction method.
Figure1 Location of the Huashan watershed and sampling sites

2.3 Source apportionments calculation

The Bayesian isotope mixture model, SIAR, was used to calculate the contribution of different potential pollution sources. Based on the R platform, the SIMMR package was used for analysis and calculation, and the model expression is as follows (Parnell et al., 2010):
(1)
where Xij represents the value of isotope j of mixture i (where i=1, 2, 3, …, N, j=1, 2, 3, …, J), Pk is the estimated proportion of source k, Sjk is the value of isotope j of source k, with mean μjk and standard deviation ωjk, Cjk is the fractionation factor of isotope j of source k (following a normal distribution with mean λjk and variance τjk), and εjk is the residual error of isotope j of mixture k (with mean 0 and standard deviation σj). The parameters of the SIAR model are mainly the number of iterations, the sampling interval and the number of decays, which are set to 500,000, 15 and 5000, respectively.
In this study, the end-member values of δ15N and δ18O from different sources were selected through self-testing in the study area and referencing to nearby regions (Table 1). The end-member values of δ15N and δ18O in atmospheric deposit were obtained through self-testing in the study area, while the δ15N values in soil organic nitrogen, manure and sewage, and nitrogen fertilizer were referenced from another study conducted in a similar geographical and climatic basin (Zhang et al., 2018). The end-member values of δ18O in soil organic nitrogen, manure and sewage, and nitrogen fertilizer were designated as the same values, which had undergone microbial processes and were different from atmospheric deposition sources (Yi et al., 2017). Based on the microbial nitrification model theory (Kendall, 1998), we used the collected samples and the δ18O of oxygen to predict the δ18O-nitrate of nitrification products, as shown in the following formula:
(2)
The δ18O value of atmospheric oxygen is 23.5%. (Aravena and Mayer, 2010)
Table 1 End-members ofδ15N and δ18O inputs for the Bayesian isotope mixing model.

2.4 Uncertainty analysis and Sensitivity analysis

Uncertainty analysis uses probabilistic statistics to evaluate the results of the SIAR model. UI90 is defined as the cumulative frequency distribution of 0.95 minus the cumulative frequency distribution of 0.05 and then divided by 0.9 (Ji et al. 2017). UI90 can represent the intensity of uncertainty under high probability conditions (90%), thus eliminating the effect of small probability (10%) extremes. Uncertainty analysis can quantify the uncertainty of model output, and help to better identify the main sources of pollution.
Sensitivity analysis was performed by adjusting the mean values of δ15N and δ18O for different potential nitrate sources separately. The sensitivity of the nitrate source assignment in the study area to the mean values of δ15N and δ18O from different sources was analyzed by varying the input values. Sensitivity analysis can help optimize the design of sampling schemes and improve the accuracy of model allocation results.