Nitrate source apportionment in a subtropical watershed using Bayesian model

Science of the Total Environment 463–464 (2013) 340–347

Contents lists available at ScienceDirect

Science of the Total Environment journal homepage: www.elsevier.com/locate/scitotenv

Nitrate source apportionment in a subtropical watershed using Bayesian model Liping Yang a, Jiangpei Han a, Jianlong Xue a, Lingzao Zeng a, Jiachun Shi a,⁎, Laosheng Wu a,⁎, Yonghai Jiang b a College of Environmental and Natural Resource Sciences, Zhejiang Provincial Key Laboratory of Subtropical Soil and Plant Nutrition, Zhejiang University, Hangzhou, 310058, People's Republic of China b State Key Laboratory of Environmental Criteria and Risk Assessment, Chinese Research Academy of Environmental Sciences, Beijing, 100012, People's Republic of China

H I G H L I G H T S • • • •

Nitrate concentration in water displayed significant temporal variation in a subtropical watershed. Chemical and isotopic characteristics were combined for nitrate source identification. Bayesian model stable isotope analysis in R (SIAR) was used for nitrate source apportionment. The merits and uncertainties of SIAR were discussed.

a r t i c l e

i n f o

Article history: Received 12 January 2013 Received in revised form 4 June 2013 Accepted 6 June 2013 Available online 29 June 2013 Keywords: Temporal variation Source apportionment Nitrogen isotope Oxygen isotope Stable isotope analysis in R (SIAR)

a b s t r a c t Nitrate (NO− 3 ) pollution in aquatic system is a worldwide problem. The temporal distribution pattern and sources of nitrate are of great concern for water quality. The nitrogen (N) cycling processes in a subtropical watershed located in Changxing County, Zhejiang Province, China were greatly influenced by the temporal variations of precipitation and temperature during the study period (September 2011 to July 2012). The −1 ) and the lowhighest NO− 3 concentration in water was in May (wet season, mean ± SD = 17.45 ± 9.50 mg L est concentration occurred in December (dry season, mean ± SD = 10.54 ± 6.28 mg L−1). Nevertheless, no − water sample in the study area exceeds the WHO drinking water limit of 50 mg L−1 NO− 3 . Four sources of NO3 (atmospheric deposition, AD; soil N, SN; synthetic fertilizer, SF; manure & sewage, M&S) were identified using − 2− 2+ , K+, Mg2+, Na+, dissolved oxygen (DO)] and both hydrochemical characteristics [Cl−, NO− 3 , HCO3 , SO4 , Ca 18 − dual isotope approach (δ15N–NO− 3 and δ O–NO3 ). Both chemical and isotopic characteristics indicated that denitrification was not the main N cycling process in the study area. Using a Bayesian model (stable isotope analysis in R, SIAR), the contribution of each source was apportioned. Source apportionment results showed that source contributions differed significantly between the dry and wet season, AD and M&S contributed more in December than in May. In contrast, SN and SF contributed more NO− 3 to water in May than that in December. M&S and SF were the major contributors in December and May, respectively. Moreover, the shortcomings and uncertainties of SIAR were discussed to provide implications for future works. With the assessment of temporal variation and sources of NO− 3 , better agricultural management practices and sewage disposal programs can be implemented to sustain water quality in subtropical watersheds. © 2013 Elsevier B.V. All rights reserved.

1. Introduction Nitrate (NO− 3 ) pollution in aquatic systems has become a worldwide problem in recent decades. High NO− 3 concentration in groundwater threatens human health (Burow et al., 2010; Di and Cameron, 2002; Macilwain, 1995), while excessive nitrogen can cause deterioration of surface water quality, resulting in eutrophication and algal bloom (Diaz, 2001). In China, serious eutrophication was observed in many agriculture intensive regions such as the Taihu Lake (Guo, 2007). The ⁎ Corresponding authors. E-mail addresses: [email protected] (J. Shi), [email protected] (L. Wu). 0048-9697/$ – see front matter © 2013 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.scitotenv.2013.06.021

groundwater in some areas in this region has exceeded the drinking water quality limit of 50 mg L−1 NO− 3 set by the World Health Organization (Chen et al., 2010; WHO, 2011; Zhu et al., 2003). Hexi Reservoir watershed is a drinking water source protection area that is located in the upstream of Taihu Lake, with forest to be the dominant land use type, mixed with agricultural and residential areas. Previous investigations indicate that the water quality there so far can meet the drinking water quality standard, but the increasing NO− 3 concentration in the reservoir causes a great concern for the future use as the drinking water source (Xie, 2005). Thus, it is of great importance to understand the nitrogen pollution characteristics and NO− 3 sources for sustaining drinking water supply in the watershed.

L. Yang et al. / Science of the Total Environment 463–464 (2013) 340–347

Nitrate source apportionment method using dual isotope approach (δ15N and δ18O) is based on the concept that different sources of NO− 3 have different stable isotope signatures. For instance, δ15N–NO− 3 from atmospheric deposition (AD) ranges from −5‰ to +3‰ (Kendall and McDonnell, 1998); δ15N of soil extracted N (SN) ranges from +3‰ to +8‰ (Kendall and McDonnell, 1998; Singleton et al., 2007); δ15N of synthetic fertilizer (SF) is around 0‰ since N in SF is from atmospheric N2, of which the δ15N is about 0‰. Manure (M&S) always display close ranges of δ15N, although a recent review indicated that some chemical markers (antibiotics and others) can be used to discrim− inate manure derived NO− 3 from sewage derived NO3 (Fenech et al., 2012). Nevertheless, due to the similarity of the chemical composition and isotope signatures of these two sources, they were often combined and treated as one source, and the δ15N of combined source was about +10‰ ~ +25‰ (Kendall and McDonnell, 1998). Because N in these sources are mostly in reduced forms, NO− 3 derived from SN, SF or M&S sources always go through the nitrification process, after which the δ18O in NO− 3 will display the isotope signature of nitrification process rather than the original oxygen isotope signature (Kendall and McDonnell, 1998; Mayer et al., 2001; Mengis et al., 2001). Thus, δ18O in NO− 3 derived from nitrification is often −5‰ ~ +15‰, which is 18 quite different from δ18O in NO− 3 from AD, of which the δ O is from +25‰ to +75‰. These differences in the sources make it possible to use isotope range comparison (Chen et al., 2009; Li et al., 2010) or linear mixing models (Deutsch et al., 2006; Voss et al., 2006) to track NO− 3 sources and apportion the contribution of each source. However, due to the wide ranges of δ15N and δ18O in the NO− 3 sources and the fractionation of δ15N and δ18O during the transport and transformation processes, the apportionment efficiency of these works based on the dual isotope approach cannot be fully assessed. For most of the isotope mixing models, they can only apportion a single contribution for each NO− 3 source (Savard et al., 2010; Voss et al., 2006). Recently, a Bayesian model for stable isotope analysis ran under an open source statistical software R called stable isotope analysis in R (SIAR) was developed and used to apportion the latent sources of the receptor water samples (Parnell et al., 2010). This model was first constructed to solve source apportionment problems in food web among different nutrition levels, for instance, to calculate the proportion of different food sources contributing to herbivore. Xue et al. (2009; 2012) then employed it for apportioning contributions of different sources to NO− 3 in water bodies. The biggest advantage of the Bayesian method is that it can give complete posterior distributions for source contributions, and many statistics (the variance, mean value, quintiles) can be drawn from these posterior distributions. The model can overcome some chronic problems that other simple mixing models cannot handle. For example, it can apportion more than three sources when only two isotopes are used. It also considers the ranges and distributions of the sources instead of using a single signature to represent one source as in the previous works. In addition, the deviation of stable isotopes caused by fractionation, which may significantly change the original signature of the isotopes during transport and transformation processes, can be easily incorporated into this model (see Section 2.5 for the details). The objectives of this study in the Hexi Reservoir watershed were − to (1) investigate the temporal variation of NO− 3 , (2) identify the NO3 sources using hydrochemical and isotopic characteristics, and (3) apply source apportionment model SIAR to estimate the proportion of each source to NO− 3 in the watershed. 2. Material and methods 2.1. Description of the study area Hexi Reservoir watershed (119°37′–119°49′E, 31°02′–31°11′N) is situated in Changxing County, Zhejiang Province, a tributary to Taihu Lake in East China (Fig. 1). The water holding capacity of the reservoir

341

is 111 million cubic meters and it supplies drinking water for more than 380 thousand people in and nearby the county. There are three streams flowing into the reservoir. The watershed is mostly river valley plain and low hills. It has a subtropical monsoon climate with annual mean air temperature of approximately 15.6 °C. August is the hottest month and January is the coldest month. The mean annual precipitation is approximately 1309 mm with nearly 75% of the precipitation falling during the period of March–September. The dominant bedrocks are sandstone and limestone and the main soil type is paddy soil. The watershed is a low-density residential area with scattered agricultural and industrial activities, and the headwater area is dominated by bamboo forest. The total area of the watershed is 235 km2, among which 76% is forest, 12% is for agriculture and cultivated land, while 4% is residential area that is mainly situated along the three inflow streams. For the agricultural land use, chemical fertilizers (mainly urea, ammonium fertilizer or NPK compound fertilizer but seldom in NO− 3 form) are typically applied at the rate of about 500 kg N ha−1 yr−1 for rice–rape rotation system (Xing et al., 2002). SF is also applied in May shortly before rice is transplanted into the field. Manure (mainly animal waste) is applied to the field. Most residents use on-site septic tanks to dispose human excretion, which can leach to the groundwater (Aravena et al., 2005; Krapac et al., 2002; Toetz, 2006). Because the N isotope signature of sewage is similar to N isotope signature of manure, these two sources were treated as one source (expressed by M&S) in this study (Kendall and McDonnell, 1998; Widory et al., 2004). 2.2. Sampling Water samples were collected from upstream reservoirs (3 reservoirs, R1, R2, and R3) and downstream Hexi Reservoir (3 sites in the Hexi Reservoir, R4, R5, R6), see Fig. 1. Along the main streams in the watershed, 28 surface water samples (river water, SW1–SW28) and 14 groundwater samples (wells for drinking or irrigation, GW1–GW14) were collected (Fig. 1). All samples were collected bimonthly approximately (totally 7 times for the study period) from September 2011 to July 2012. Rainwater (4 samples) was also collected for chemical and isotope analysis, because it represented AD source of NO− 3 in the watershed. Other nitrate sources (SN, SF, and M&S) were not measured due to logistic reasons; instead, relevant data from other literatures were used in this study (see details in Fig. 4). 2.3. Chemical and isotope analysis Several water quality parameters including pH, temperature (T), and dissolved oxygen (DO) were determined in the field using a pre-calibrated handheld multiparameter meter (YSI, USA), see Table 1. Data for temperature and DO were missing for the month of December 2011 due to the YSI instrument breakdown. After water samples were taken back into the laboratory, a portion of each sample was filtered through 0.45 μm cellulose-acetate filter paper for chemical and isotope analysis. For chemical parameter analysis, water samples were stored in high-density polyethylene bottles at 4 °C to minimize micro− 2− biological activity. The concentrations of Cl−, NO− 3 , HCO3 , and SO4 were analyzed using a Dionex ICS-2000 ion chromatography system (Dionex, USA). The concentrations of major cations Ca2+, K+, Mg2+, Na+ were determined by inductively coupled plasma spectrometer iCAP6300 (Thermo Fisher Scientific, USA). To determine isotopes, filtered water samples and rainwater samples were frozen in 10 ml polyethylene bottles with airtight 18 − caps. Then δ15N–NO− 3 and δ O–NO3 were analyzed using the denitrifier method at the Facility for Isotope Ratio Mass Spectrometry, University of California, Riverside. The determination procedure and main principles (Revesz et al., 2007) are as follows. Stable nitrogen isotope-amount ratios δ15N and stable oxygen isotope-amount ratios δ18O are described as the relative difference in the ratio of 15N (18O)

342

L. Yang et al. / Science of the Total Environment 463–464 (2013) 340–347

Fig. 1. Study area and sampling sites (GW, groundwater; SW, surface water; RW, reservoir water; GWI, groundwater for isotope analysis; SWI, surface water for isotope analysis; RWI, reservoir water for isotope analysis).

Table 1 Basic description of chemical parameters in the study watershed from September 2011 to July 2012. Water type SW

GW

RW

Descriptive parameters Sampling number Mean SD Sampling number Mean SD Sampling number Mean SD

pH 196 8.03 0.59 98 6.82 0.43 42 7.70 0.63

T (°C)

DO (mg L−1)

NO− 3 (mg L−1)

168 18.5 5.50 84 18.2 4.46 36 19.2 7.04

168 9.66 2.48 84 6.64 2.18 36 8.72 2.19

196 11.78 4.20 98 19.85 9.50 42 7.77 3.88

SW, surface water; GW, groundwater; RW, reservoir water; SD, standard deviation; T, temperature; DO, dissolved oxygen.

to the amount of 14N (16O) of samples with respect to the reference materials and calculated by the following equations:

15

nX

18

nX

δ N¼

δ O¼

−



15



18

−

       14 15 14 N =nX N −nref N nref N

nref 15 N nref 14 N

       O =nX 16 O −nref 18 O nref 16 O

nref 18 O =nref 16 O

NO3 →NO2 →NO→1=2N2 O

ð1Þ

ð2Þ

ð3Þ

L. Yang et al. / Science of the Total Environment 463–464 (2013) 340–347

where nX(iE)/nX(jE) is the ratio of the isotope amounts in the sample X and nref(iE)/nref(jE) is the ratio of the isotope amounts in the reference material. A positive δ15N (δ18O) value indicates that the sample is more enriched for isotope 15N (18O) than the reference and vice versa. Values of δ15N and δ18O have been reported in parts per thousand (‰) in this work. The reference materials for δ15N and δ18O are atmospheric nitrogen gas (N2) and Vienna Standard Mean Ocean Water (VSMOW), respectively, which have a δ15N value and δ18O value of 0‰. The precision and accuracy were less than ± 0.25‰ and less than ± 0.5‰, respectively, for the δ15N and δ18O measurements (Revesz et al., 2007). The δ15N and δ18O of dissolved NO− 3 in all water samples were analyzed by converting NO− 3 to nitrous oxide (N2O) by a culture of denitrifying bacteria (Pseudomonas aureofaciens) (Eq. 3). As this bacterium lacks N2O reductive activity, N2O cannot be further converted to N2. Then the generated N2O was analyzed by mass spectrometry after some extraction and purifying steps. 2.4. Multivariate statistics Cluster analysis (CA) for temporal clustering was done using average linkage (between groups) with squared Euclidean distance interval. To test the distribution normality, one-sample K–S test was applied to − NO− 3 concentration. Homogeneity of variance of NO3 was also tested. Due to the non-normal distribution of NO− or heterogeneity of variance 3 of NO− 3 , Kruskal–Wallis test was applied to test the difference among clusters generated by CA. All the data analyses were done using “Statistical Package for the Social Sciences Software-SPSS 16.0 for Windows” (Norusis, 2008). 2.5. Source appointment mixing model (SIAR) Stable isotope analysis in R (SIAR) deals with a general situation where data comprise N measurements on J isotopes with K sources (Parnell et al., 2010). It is a fully Bayesian probabilistic approach based on the mass balance of isotopes. The contributions of different sources are treated as random variables whose statistics can be fully characterized by probability distribution functions. The Dirichlet distribution is considered as the prior distribution of source contributions, which forces the sum of source contributions to be one (Parnell et al., 2010; Massoudieh and Kayhanian, 2012). Once isotope measurements are obtained, the updated information about the contributions is included in the posterior distributions, which can be calculated by combining the prior distribution and likelihood. Since it is difficult to obtain the analytical forms of the posterior distributions, SIAR generates a large quantity of realizations for the proportional contribution of sources through Markov Chain Monte Carlo (MCMC). The properties of posterior probability distributions of the source contributions (e.g., the mean, standard deviation, and quartiles et al.) can be obtained via analyzing these generated realizations. The system model used in SIAR can be expressed as follows:

X ij ¼

  k ∑k¼1 P k qjk Sjk þ C jk k

∑k¼1 P k qjk

þ εij

ð4Þ

  2 Sjk ∼ N μ jk ; ωjk   2 C jk ∼ N λjk ; τjk   2 εij ∼ N O; σ j : where Xij refers to observed isotope value j (i.e., δ15N and δ18O values in Eqs. (1) and (2)) of the water sample i; sjk refers to source value k on isotope j, normally distributed with mean μjk and variance ω2jk; cjk refers

343

to fractionation factor for isotope j on source k, normally distributed with mean λjk and variance τ2jk; pk refers to proportional contribution of source k, which is estimated by SIAR; qjk refers to concentration of isotope j in source k; and εij refers to residual error, describing additional inter-observation variance not described by the model, σ2j is the residual variance and is estimated by the model. A detailed description of this model can be found in Hopkins and Ferguson (2012), Moore and Semmens (2008), and Parnell et al. (2010). − 18 In this study, SIAR was used to integrate δ15N–NO− 3 and δ O–NO3 values to estimate the contributions of four predefined sources (AD, SN, SF, and M&S) for the two seasons (December for the dry season and May for the wet season). In SIAR, the prior distribution of source contributions can be chosen as vague or informative based on available information. In our study, the vague prior was used, i.e., each source contribution used the same prior mean and variance. Since the source terms, sjk, are modeled as Gaussian random variables in SIAR. In this study, the mean and variance of each source were either calculated from local data or cited from literatures (as described in Fig. 4). The fractionation factors for all sources [i.e., cjk in Eq. (4)] were set to be zero as described in Section 3.4. Measured isotope data from three reservoir water samples, nine river water samples and five groundwater samples were used as inputs in SIAR. 3. Results and discussion 3.1. Hydrochemical characteristics in the study area Basic descriptions of chemical parameters of the water samples are shown in Table 1. The surface river (SW) and reservoir water (RW) were generally alkaline while groundwater (GW) was neutral to weak acidic. The mean temperatures of the three types of water ranged from 8.9 °C (in February) to 25.2 °C (in July). DO concentrations in the SW and RW were higher than in the GW. All the three types of water had much higher mean concentrations of DO than the appropriate limit (less than 1–2 mg L−1) for denitrification (Rivett et al., 2008). The NO− 3 concentration in GW was generally higher than that in SW, and the RW displayed lowest NO− 3 concentration. In the entire watershed, the highest NO− 3 concentration of 43.50 mg L−1 was found in GW in May, which is close to the 50 mg L−1 NO− 3 concentration limit for drinking water (WHO, 2011). 3.2. Temporal variation of NO− 3 in the study period Temporal trend is evidently noticeable for NO− 3 in SW, GW, and RW (Fig. 2). The highest mean concentration was in May for GW, in February for SW and in March for RW, while the lowest concentrations were observed in November or December for all the three types of water. This can be attributed to the application of nitrogen fertilizers to rape from early February, and fertilization of SF and manure to rice crops. Another reason is that our study area is located in a subtropical monsoon climate region. Under such climate, intense weathering of soil parent materials is expected in warm and moist seasons. From May on, this region is dominated by mould rains, which accelerates the NO− 3 runoff into SW and leaching into the GW. Therefore, high NO− 3 concentrations were observed in all the three types of water in May. Cluster analysis (CA) was carried out to classify the temporal patterns using NO− 3 data from the entire watershed (Fig. 3). The CA classified the 7 sampling months into four clusters. November and December were in cluster 1, September and July were in cluster 2, and February and March were in cluster 3, while May was in a single group of cluster 4. Due to most variables in the original datasets were not normally distributed, Kruskal–Wallis test was conducted to check whether there were significant differences among the classified groups. The results showed that except for cluster 3 and cluster 4, the differences between each two of the four clusters were significant

344

L. Yang et al. / Science of the Total Environment 463–464 (2013) 340–347

were lower than the value of local rainfall (1.8) in the study watershed, indicating that there existed anthropogenic pollution sources of Cl− besides natural rainfall. These Cl− sources included chemical fertilizers as well as sewage effluent (Jiang et al., 2009). Since the chemical fertilizers used in the study area typically did not contain Cl−, sewage effluent was considered as the main Cl− source. M&S have been found high in Cl− − − and SO2− 4 (Yao et al., 2007), and low in NO3 /Cl ratios (Liu et al., 2006). − − Thus the lower NO3 /Cl ratio of the water samples in December (1.08 on average) than that in May (2.62 on average) indicated that M&S contributed more to NO− 3 in December than in May. 3.4. Nitrate source identification using isotope range comparison

Fig. 2. Seasonal variations of NO− 3 in SW (surface water), GW (groundwater), and RW (reservoir water). Each error bar represents 1 standard deviation.

(p b 0.01). The temporal variation pattern of NO− 3 was attributed mainly to the hydrological factors. Nevertheless, other factors such as differences in NO− 3 release from various sources might also contribute to this NO− 3 variation pattern, since not all wet months or dry months were grouped into the same cluster (e.g., February and December were both in the dry season but they were not grouped into the same cluster). Because cluster 1 (November and December) and cluster 4 (May) represented the largest difference in NO− 3 concentrations during the study period, December and May were selected to track the sources of NO− 3 using hydrochemical characteristics and isotope signatures. Then the Bayesian method was used to apportion the contribution of each source to NO− 3 concentration for the water samples from the three types of water. 3.3. Nitrate source identification using hydrochemical characteristics Concentrations of major anions and cations are useful information in NO− 3 source tracking (Min et al., 2002; Spruill et al., 2002), which were also used for NO− 3 source tracking in this study (Table 2). As the bedrocks in this area are limestone and sandstone, the higher concentrations of Ca2+, Mg2 + and HCO− 3 in GW than in SW and RW in summer showed a sign of faster dissolution of carbonate minerals (Jiang et al., 2009), which resulted in the relatively high abundance of the three ions in the GW during the warm season. For most of the eight ions except for NO− 3 , the ion concentrations showed higher or equal values in December than in May in the three types of water. If weathering of soil parent materials was the dominant source type for all ions, these ions should have higher concentrations under suitable environmental conditions in May, not in December as the data showed. The unanticipated higher concentrations of these ions in December indicate that there were contributions from other sources besides soil in May and December. Chloride (Cl−) is a useful indicator for contamination source tracking. Most of the water samples showed that the Na+/Cl− molar ratio values

Ranges of δ15N and δ18O for the four sources of NO− 3 in the study area are shown by boxes in Fig. 4. To calculate δ18O–NO− 3 derived from oxidation of reduced N forms, the rule of two-thirds of the O in NO− 3 from water and one-third from atmospheric O was used (Böttcher et al., 1990; Kendall and McDonnell, 1998). The δ18O value in the atmospheric O was + 23.5‰, and the δ18O values in water ranged from −11.8‰ to +2.1‰, which were calculated using δ18O in precipitation water from a nearby International Atomic Energy Association (IAEA) station in Nanjing (IAEA, 1992). Thus, the δ18O–NO− 3 derived from nitrification of reduced N forms ranged from −0.2‰ to +9.1‰. This range is close to but narrower than the values in other studies (Mayer et al., 2001; Williard et al., 2001). As shown in Fig. 4, there are relatively wide ranges for δ15N–NO− 3 values of N sources. The range of δ15N–NO− 3 values for AD was −1‰ to + 3‰. The range of δ15N–NO− 3 values for SN was + 3‰ to + 8‰. The SF for the agriculture land use in our study area were mainly urea and ammonium, while nitrate N form was seldom used. By taking into the consideration of nitrogen fertilizer types applied in China (Zhang, 2006), the ratio 7:3 of urea-N: ammonium-N was used to cal15 − culate δ15N–NO− 3 , which resulted in the δ N–NO3 range of − 3.8‰ to 15 − + 1.4‰ for the SF. The δ N–NO3 values for M&S ranged from + 8‰ to + 20‰. The measured δ18O–NO− 3 values ranged from + 5.5‰ to + 9.2‰ in the SW and from +0.2‰ to +7.6‰ in the GW (Fig. 4). These relatively depleted values suggested that nitrification process was the dominant 18 − source of the NO− 3 in these two water types. In the RW, the δ O–NO3 values were from +6.3‰ to +23.0‰. The higher values of δ18O–NO− 3 indicated that NO− 3 in the RW was more affected by AD than in the 15 − SW and GW. Both δ18O–NO− 3 and δ N–NO3 displayed higher values in December than in May, which was attributed to the greater contribution of sources with high δ15N and high δ18O values, and/or higher denitrification rate in December (Chen et al., 2009). 15 − No significant correlation of δ18O–NO− 3 and δ N–NO3 was observed 18 15 − in the SW, but the correlation between δ O–NO3 and δ N–NO− 3 in GW was significant (r2 = 0.6482, p b 0.01, n = 10). Nevertheless, the slope of 0.36 when δ18O–NO− 3 is plotted as the dependent variable against δ15N–NO− 3 cannot be explained as a result of denitrification, since the isotope enrichment factor value for O in NO− 3 should be about 1/2 for that of N in denitrification (Aravena and Robertson, 2005). At the mean time, the DO concentration (mean value of all three types of water, 8.66 mg L−1 O2) was much higher than the O2 concentration suitable for denitrification (b1–2 mg L−1), which also denied an obvious denitrification process (Rivett et al., 2008). 3.5. Nitrate source apportionment using SIAR

Fig. 3. Dendrogram generated from the cluster analysis of NO− 3 concentrations in Hexi Reservoir watershed.

The posterior distributions of the contributions of the four NO− 3 sources for sampling site S2 in winter are shown in Fig. 5. The posterior probability distribution of the M&S source contribution was symmetric and close to normal distribution. The mean, median and maximum a posteriori probability (MAP) estimations of the source contributions were close. However, the posterior probability distributions of the AD, SN, and SF contributions were not symmetrically

L. Yang et al. / Science of the Total Environment 463–464 (2013) 340–347

345

Table 2 Summary of major anion and cation concentrations in SW, GW and RW (all units are in mg L−1). Major ions

SW GW RW

Mean SD Mean SD Mean SD

NO− 3

Cl−

HCO− 3

SO2− 4

Ca2+

K+

Mg2+

Na+

Dec

May

Dec

May

Dec

May

Dec

May

Dec

May

Dec

May

Dec

May

Dec

May

12.27 7.87 15.98 14.39 5.44 2.52

5.68 1.00 11.79 6.73 4.91 2.14

9.11 5.11 15.28 7.03 3.94 1.82

14.03 2.22 31.54 17.28 9.38 2.81

81.76 26.07 77.65 37.25 81.79 65.06

55.85 18.40 72.41 36.12 46.51 38.71

48.21 32.37 46.23 26.91 35.26 12.83

33.43 10.40 47.57 25.15 35.18 13.52

44.55 17.21 32.09 15.58 22.35 18.16

25.04 8.89 33.24 14.07 20.10 14.01

5.79 2.83 3.94 1.63 1.93 1.62

3.30 2.08 7.11 5.48 1.49 1.09

5.13 1.54 5.34 3.60 3.41 1.29

3.42 0.87 4.97 3.43 3.16 1.00

10.42 6.25 10.48 7.50 4.51 2.64

5.03 2.05 8.40 4.94 4.53 2.24

SW, surface water; GW, groundwater; RW, reservoir water; SD, standard deviation; Dec, December.

distributed. For the source contributions of SF, the mean value (0.12) and median value (0.10) of the posterior probability distribution differed substantially from the MAP value (0.01) in December (Fig. 5). In SIAR, the Dirichlet distribution treats each source input as independent and requires the mean proportions for each NO− 3 source to sum to unity (Parnell et al., 2010). Therefore, it is straightforward to use the posterior mean values as the estimations of source contributions. In contrast, if other estimations are employed (e.g., the MAP estimation, the medians of posterior realizations), it is not guaranteed that the sum of estimation values equals to one. Take Fig. 5 for example, the sum of the posterior mean values of the four sources in December is 1, while the sum of the median values or the sum of MAP values is less than 1 in this study, due to the unsymmetrical distribution of the SN and SF posterior probability distribution. In our study, the mean values of posterior contribution were used for further analysis. From the mean values, we can infer that source contributions varied significantly among different seasons and different water types (Table 3). For the entire watershed, M&S and AD contributed more NO− 3 in December than in May, while SF and SN contributed more NO− 3 in May than in December. In December, M&S contributed most while in May SF was the major contributor to NO− 3 in the entire watershed. As a whole, AD contributed least to NO− 3 in the entire watershed, but it contributed substantially more to the RW than to the SW and GW in both May and December. This is attributed to the fact that the RW samples accumulated more NO− 3 from the AD source than SW and GW. The total contribution of natural sources (AD + SN) to RW was 48% in winter and 41% in summer. These values were higher than the

40

30

Density

20

5

10

0

AD SN SF M&S

10

δ18O (‰)

50

15

GW-May GW-Dec SW-May SW-Dec RW-May RW-Dec Rain AD SN SF M&S

60

contributions of the natural sources (AD + SN) to the SW and GW. Using Kruskal–Wallis test, the posterior mean values of SN, SF, and M&S were identified to have significant (p b 0.01) differences in their NO− 3 contributions to water in summer and winter. In general, the source contributions estimated by SIAR were reasonable and in agreement with previous analysis by hydrochemical characteristics. However, some unsolved problems require future investigations and improvements. In this study, the mean and variance of sources were mostly calculated from literatures. Site-specific and more accurate isotope information of N and O isotopes for different sources from the study area is needed for future studies. Secondly, as SIAR assumes that the sources are normally distributed. Therefore, if the actual source isotope values were not normally distributed, it can lead to biased results. To cope with this violation of assumption, SIAR should be expanded to deal with different distributions. However, this requires non-trivial recoding work (Parnell et al., 2010). In addition, as shown in Fig. 5, the posterior distributions of source contributions have very wide ranges, which imply large uncertainty in the apportionment results. Use of more accurate information of the source values (i.e., smaller variance ω2jk), if possible, can reduce the uncertainty. The apportionment result based on the single-value estimation (posterior means) should be used with caution (Parnell et al., 2010; von Toussaint, 2011). Finally, the results were based on the interpretation of two sets of water samples, one from dry season (December 2011) and one from wet season (May 2012), for they represented the greatest differences in climate conditions, crop types, and nitrate concentrations in the water samples. The samples collected from the other five times were not used for isotope analysis. Use of the sparse samples (twice a year) may not be able to reflect the dynamic nature of the source contributions. Further study regarding the nitrate source apportionment should

-5

0

5

10

15

20

Fig. 4. Isotope ranges of potential nitrate sources and samples. The ranges of δ15N and δ18O for AD were calculated from the rain samples in the study area; the δ18O ranges of SN, SF, and M&S were calculated using H2O and O2 used in nitrification process, the H2O was taken from an IAEA station near the study area. The ranges of δ15N of SN, SF and M&S were calculated from Cao et al. (1991), Cey et al. (1999), Heaton (1986), Kendall and McDonnell (1998), Kreitler (1979), Liu et al. (2006) and Widory et al. (2004). (AD, atmospheric deposition; SF, synthetic fertilizer; SN, soil N; M&S, manure and sewage).

0

δ15N (‰)

0.0

0.2

0.4

0.6

0.8

1.0

Proportion Fig. 5. The posterior distributions of the contributions of the four NO− 3 sources of sampling site S2 in December.

346

L. Yang et al. / Science of the Total Environment 463–464 (2013) 340–347

Table 3 Posterior mean values of source contributions apportioned using SIAR. Water type

Dry season (December) AD

SW(Mean ± SD) GW(Mean ± SD) RW(Mean ± SD) WW(Mean ± SD)

0.07 0.03 0.21 0.09

Wet season (May)

SN ± ± ± ±

0.07 0.02 0.14 0.09

0.25 0.21 0.27 0.24

SF ± ± ± ±

0.05 0.15 0.04 0.09

0.13 0.15 0.23 0.15

M&S ± ± ± ±

0.07 0.12 0.05 0.09

0.54 0.61 0.29 0.52

AD ± ± ± ±

0.15 0.25 0.12 0.21

0.05 0.02 0.12 0.05

SN ± ± ± ±

0.03 0.01 0.06 0.05

0.32 0.34 0.29 0.32

SF ± ± ± ±

0.02 0.02 0.02 0.02

0.39 0.33 0.41 0.37

M&S ± ± ± ±

0.06 0.09 0.11 0.08

0.24 0.31 0.18 0.25

± ± ± ±

0.08 0.11 0.14 0.1

AD, atmospheric deposition; SN, soil N; SF, synthetic fertilizer; M&S, manure and sewage; SW surface water; GW, groundwater; RW, reservoir water; WW, whole watershed; SD standard deviation.

consider for longer period of isotope analysis to demonstrate the dynamic change of nitrate source contributions.

4. Conclusions Hexi Reservoir watershed in East China is a representative watershed in the subtropical area of monsoon climate that receives nitrate from mixing sources of natural and anthropogenic inputs. This study evaluated the temporal pattern and sources of NO− 3 pollution in the watershed. Our study shows that the watershed displayed most significant variations in NO− 3 concentration in December and May (representing dry season and wet season, respectively). In this study, hydrochemical characteristics and dual isotope approach 18 − (δ15N–NO− 3 and δ O–NO3 ) were combined for nitrate source identification followed by the Bayesian model (SIAR) for quantitative nitrate source apportionment. Despite of some uncertainties, SIAR performed reasonably well in estimating the contributions from each of the sources to the receptor water in December and May. It was identified that M&S and SF were the largest contributors to nitrate in the watershed in December and in May, respectively. In the entire watershed, M&S contributed an average of 52% of the nitrate in December, SF contributed an average of 37% of nitrate in May. The findings from this research may help to develop better nitrogen management practices in a watershed scale. The result of high nitrate concentration (close to WHO threshold for drinking water) in some of the sampling sites suggests that the government should urgently setup water quality monitoring networks to monitor real-time nitrate concentrations. The planners and policy makers can take innovative and effective planning policies and management programs according to the nitrate source contributions. For example, using fertilizers/ manure according to the soil nutrients and crop demands in different growing periods can be implemented to reduce the non-point pollution caused by excessive N application. For controlling the sewage source, local government should strengthen the control of sanitary sewage by constructing wastewater disposal systems and using new technologies to increase the capacity of domestic wastewater treatment. In addition, environmental awareness of local people should be aroused to better implement these programs to sustain safe drinking water supply in the watershed.

Acknowledgements This work was jointly supported by the National Natural Science Foundation of China (41271470, 41071144), National High-tech R&D Program of China (863 Program) (2012AA062603), National Key Technologies R&D Program of China (2012BAD15B04), Zhejiang Provincial Natural Science Foundation of China (LY13D010001), and the Key SQT Innovation Team of Zhejiang Province for AgroProducts Standards and Testing Technology (2010R50028). The authors would like to thank the Center for Conservation Biology (University of California, Riverside, USA) for their help with the stable isotope analyses.

References Aravena R, Robertson WD. Use of multiple isotope tracers to evaluate denitrification in ground water: study of nitrate from a large‐flux septic system plume. Ground Water 2005;36:975–82. Aravena R, Evans M, Cherry JA. Stable isotopes of oxygen and nitrogen in source identification of nitrate from septic systems. Ground Water 2005;31:180–6. Böttcher J, Strebel O, Voerkelius S, Schmidt HL. Using isotope fractionation of nitrate-nitrogen and nitrate-oxygen for evaluation of microbial denitrification in a sandy aquifer. J Hydrol 1990;114:413–24. Burow KR, Nolan BT, Rupert MG, Dubrovsky NM. Nitrate in groundwater of the United States, 1991–2003. Environ Sci Technol 2010;44:4988–97. Cao Y, Sun G, Xing G, Xu H. Natural abundance of 15N in main N-containing chemical fertilizers of China. Pedosphere 1991;1:377–82. Cey EE, Rudolph DL, Aravena R, Parkin G. Role of the riparian zone in controlling the distribution and fate of agricultural nitrogen near a small stream in southern Ontario. J Contam Hydrol 1999;37:45–67. Chen F, Jia G, Chen J. Nitrate sources and watershed denitrification inferred from nitrate dual isotopes in the Beijiang River, south China. Biogeochemistry 2009;94:163–74. Chen X, Wo F, Chen C, Fang K. Seasonal changes in the concentrations of nitrogen and phosphorus in farmland drainage and groundwater of the Taihu Lake region of China. Environ Monit Assess 2010;169:159–68. Deutsch B, Mewes M, Liskow I, Voss M. Quantification of diffuse nitrate inputs into a small river system using stable isotopes of oxygen and nitrogen in nitrate. Org Geochem 2006;37:1333–42. Di H, Cameron K. Nitrate leaching in temperate agroecosystems: sources, factors and mitigating strategies. Nutr Cycl Agroecosyst 2002;64:237–56. Diaz RJ. Overview of hypoxia around the world. J Environ Qual 2001;30:275–81. Fenech C, Rock L, Nolan K, Tobin J, Morrissey A. The potential for a suite of isotope and chemical markers to differentiate sources of nitrate contamination: a review. Water Res 2012;46:2023–41. Guo L. Doing battle with the green monster of Taihu Lake. Science 2007;317:1166. Heaton T. Isotopic studies of nitrogen pollution in the hydrosphere and atmosphere: a review. Chem Geol 1986;59:87–102. Hopkins JB, Ferguson JM. Estimating the diets of animals using stable isotopes and a comprehensive Bayesian mixing model. PLoS One 2012;7:e28478. IAEA. Statistical treatment of environmental isotope data in precipitation. 2nd ed. Vienna: International Atomic Energy Agency; 1992. Jiang Y, Wu Y, Yuan D. Human impacts on karst groundwater contamination deduced by coupled nitrogen with strontium isotopes in the Nandong underground river system in Yunan, China. Environ Sci Technol 2009;43:7676–83. Kendall C, McDonnell JJ. Isotope tracers in catchment hydrology. 1st ed. Amsterdam: Elsevier; 1998. Krapac I, Dey W, Roy W, Smyth C, Storment E, Sargent S, et al. Impacts of swine manure pits on groundwater quality. Environ Pollut 2002;120:475–92. Kreitler CW. Nitrogen-isotope ratio studies of soils and groundwater nitrate from alluvial fan aquifers in Texas. J Hydrol 1979;42:147–70. Li SL, Liu CQ, Li J, Liu X, Chetelat B, Wang B, et al. Assessment of the sources of nitrate in the Changjiang River, China using a nitrogen and oxygen isotopic approach. Environ Sci Technol 2010;44:1573–8. Liu CQ, Li SL, Lang YC, Xiao HY. Using δ15N- and δ18O-values to identify nitrate sources in karst ground water, Guiyang, Southwest China. Environ Sci Technol 2006;40:6928–33. Macilwain C. US report raises fears over nitrate levels in water. Nature 1995;377:4. Massoudieh A, Kayhanian M. A Bayesian chemical mass balance method for surface water contaminant source apportionment. J Environ Eng 2012;139:250–60. Mayer B, Bollwerk SM, Mansfeldt T, Hütter B, Veizer J. The oxygen isotope composition of nitrate generated by nitrification in acid forest floors. Geochim Cosmochim Acta 2001;65:2743–56. Mengis M, Walther U, Bernasconi SM, Wehrli B. Limitations of using δ18O for the source identification of nitrate in agricultural soils. Environ Sci Technol 2001;35:1840–4. Min J-H, Yun S-T, Kim K, Kim H-S, Hahn J, Lee K-S. Nitrate contamination of alluvial groundwaters in the Nakdong River basin, Korea. Geosci J 2002;6:35–46. Moore JW, Semmens BX. Incorporating uncertainty and prior information into stable isotope mixing models. Ecol Lett 2008;11:470–80. Norusis M. SPSS 16.0 guide to data analysis. 2nd ed. New York: Prentice Hall Press; 2008. Parnell AC, Inger R, Bearhop S, Jackson AL. Source partitioning using stable isotopes: coping with too much variation. PLoS One 2010;5:e9672. Revesz KM, Casciotti KL, Coplen TB. Determination of the [delta](15N/14N) and [delta](18O/16O) nitrate in water. RSIL Lab Code 2900: US Geological Survey; 2007.

L. Yang et al. / Science of the Total Environment 463–464 (2013) 340–347 Rivett MO, Buss SR, Morgan P, Smith JW, Bemment CD. Nitrate attenuation in groundwater: a review of biogeochemical controlling processes. Water Res 2008;42:4215–32. Savard MM, Somers G, Smirnoff A, Paradis D, Bochove E, Liao S. Nitrate isotopes unveil distinct seasonal N-sources and the critical role of crop residues in groundwater contamination. J Hydrol 2010;381:134–41. Singleton M, Esser B, Moran J, Hudson G, McNab W, Harter T. Saturated zone denitrification: potential for natural attenuation of nitrate contamination in shallow groundwater under dairy operations. Environ Sci Technol 2007;41:759–65. Spruill TB, Showers WJ, Howe SS. Application of classification-tree methods to identify nitrate sources in ground water. J Environ Qual 2002;31:1538–49. Toetz D. Nitrate in ground and surface waters in the vicinity of a concentrated animal feeding operation. Arch Hydrobiol 2006;166:67–77. von Toussaint U. Bayesian inference in physics. Rev Mod Phys 2011;83:943–99. Voss M, Deutsch B, Elmgren R, Humborg C, Kuuppo P, Pastuszak M, et al. Source identification of nitrate by means of isotopic tracers in the Baltic Sea catchments. Biogeosciences 2006;3:663–76. WHO. Guidelines for drinking-water quality. 4th ed. Geneva: World Health Organization; 2011. Widory D, Kloppmann W, Chery L, Bonnin J, Rochdi H, Guinamant J-L. Nitrate in groundwater: an isotopic multi-tracer approach. J Contam Hydrol 2004;72: 165–88.

347

Williard KW, DeWalle DR, Edwards PJ, Sharpe WE. 18O isotopic separation of stream nitrate sources in mid-Appalachian forested watersheds. J Hydrol 2001;252:174–88. Xie H. Geo-chemical abnormal features and preliminary assessment on environment quality in Meishan basin of the Changxing county, Zhejiang Province. Chinese J Geol Hazard Control 2005;16:124–9. [in Chinese]. Xing G, Cao Y, Shi S, Sun G, Du L, Zhu J. Denitrification in underground saturated soil in a rice paddy region. Soil Biol Biochem 2002;34:1593–8. Xue D, Botte J, De Baets B, Accoe F, Nestler A, Taylor P, et al. Present limitations and future prospects of stable isotope methods for nitrate source identification in surface - and groundwater. Water Res 2009;43:1159–70. Xue D, De Baets B, Van Cleemput O, Hennessy C, Berglund M, Boeckx P. Use of a Bayesian isotope mixing model to estimate proportional contributions of multiple nitrate sources in surface water. Environ Pollut 2012;161:43–9. Yao LX, Li GL, Tu SH, Sulewski G, He ZH. Salinity of animal manure and potential risk of secondary soil salinization through successive manure application. Sci Total Environ 2007;383:106–14. Zhang R. Production and market information of nitrogen fertilizers in China in 2005. Econ Anal Petrol Chem Ind 2006;10:10–3. [in Chinese]. Zhu J, Liu G, Han Y, Zhang Y, Xing G. Nitrate distribution and denitrification in the saturated zone of paddy field under rice/wheat rotation. Chemosphere 2003;50:725–32.