Lake Biwa, located in the central part of the Japanese islands, is the largest freshwater lake in Japan (Fig. 2). It has a surface area of 670.4 km2, a total catchment area of 3174 km2, and annual precipitation of around 2000 mm. More than 120 streams flow into the lake, but the Seta River (no. 33 in Fig. 2b) at the southern end of the lake, also known as the Yodo River, is the only natural outflow. The average residence time of water in the lake is 5.5 years.

Similar to many lakes throughout the world, Lake Biwa has experienced eutrophication in the past. Urbanisation near the lake, beginning in the 1960s, particularly on the southern and eastern shores, likely caused an increase in nutrient loading. Blooms of Uroglena americana and cyanobacteria have occurred since 1977 and 1983 respectively (Hsieh et al., 2011). To clarify the pathways and sources of nitrate that were fed into the lake, the stable isotopic compositions (δ15N and δ18O) of dissolved nitrate were determined in the major streams flowing into the lake (Ohte et al., 2010). Based on the δ15N values of nitrate showing positive correlation with the population densities of each catchment area, it was concluded that sewage effluent was the dominant source contributing to the increase in the δ15N values of nitrate.

In this study, stream water samples were collected near the mouths of 33 inflow streams and 1 outflow river (Seta River) of Lake Biwa (Table 1; Fig. 2b) during base flow periods four times in 2013: on 15 March, 17 June, 5 August, and 21 October, except for stream nos. 3 and 28 in June, which became dry arroyos at that time. The catchments of the studied inflow streams occupied 70 % of the entire Lake Biwa basin area. The streams were selected to cover those in which the concentrations and stable isotope compositions of nitrate, δ15N and δ18O, had already been determined in 2004–2006 (Ohte et al., 2010). The categories of locations classified by Ohte et al. (2010) were also used in this study to classify the location of each stream (Table 1). Either a bucket or dipper was used to collect samples as far from the bank as possible. Each sample was transferred into a dark polyethylene bottle that was pre-rinsed at least twice with the sample itself and subsequently stored in a refrigerator. Then, the samples were filtered through a pre-combusted Whatman GF/F filter with a 0.7 µm pore size within a few hours after collection, and the filtrate was stored in a different dark polyethylene bottle at 4 ∘C until analysis.

In this study, we defined the sampling number n, where n= 1, 2, 3, and 4, which represent the sampling in March, June, August, and October respectively. In addition, we defined one more hypothetical sampling number (n= 5) set just 1 year later than the n= 1 date. Please note that there are no data for sampling n= 5. Furthermore, we rated the intervals between n= 1 and n= 2, n= 2 and n= 3, n= 3 and n= 4, and n= 4 and n= 5 as spring, summer, autumn, and winter respectively for the streams in this study.

The Sado-seki National Acid Rain Monitoring Station (38∘14′59′′ N, 138∘24′00′′ E) was established on Sado Island (Fig. 2a), at 110 m above sea level, as a monitoring observatory of the Acid Deposition Monitoring Network in East Asia (EANET) representing the central Japan area (EANET, 2014). Samples of wet deposition were taken at the station by using standard methods for evaluating acid deposition in Japan for the 3 Japanese financial years (FYs) from April 2009 to March 2012. An automatic wet deposition sampler (US-420, Ogasawara) was used in the collection. All of the deposition samples were stored in 1 L polyethylene bottles under refrigeration until daily recovery. After measuring the volume (i.e. precipitation rate), conductivity, and pH, the recovered samples were filtered through a 0.2 µm pore-size membrane filter (Dismic-25CS, ADVANTEC) and stored in a refrigerator until analysis.

The annual wet deposition rate of nitrate was 19.3 mmol m-2 yr-1 for FY2009, from April 2009 to March 2010; 28.0 mmol m-2 yr-1 for FY2010, from April 2010 to March 2011; 27.0 mmol m-2 yr-1 for FY2011, from April 2011 to March 2012; and 24.5 mmol m-2 yr-1 on average from FY2009 to 2011 (EANET, 2014). The annual wet deposition rate of NH4+ was 17.1 mmol m-2 yr-1 on average from FY2009 to 2011 (EANET, 2014).

The concentrations of nitrate (NO3-) and nitrite (NO2-) in each filtrate sample were measured by ion chromatography (Prominence HIC-SP, Shimadzu, Japan) within a few days (stream water samples) and within 2 weeks (wet deposition samples) after each sampling. The error (standard error of the mean) in the determined concentrations of nitrate was ±3 %. The δ18O values of H2O in the samples were analysed using the cavity ring-down spectroscopy method by employing an L2120-i instrument (Picarro Inc., Santa Clara, CA, USA) equipped with an A0211 vaporiser and auto sampler; the error (standard error of the mean) in this method was ±0.1 ‰. Both Vienna Standard Mean Ocean Water (VSMOW) and Standard Light Antarctic Precipitation (SLAP) were used to calibrate the values to the international scale.

To determine the stable isotopic compositions, nitrate in each filtrate sample was chemically converted to N2O by using a method originally developed to determine the 15N / 14N and 18O / 16O ratios of seawater and freshwater nitrate (McIlvin and Altabet, 2005) and that was later modified (Tsunogai et al., 2008; Konno et al., 2010; Yamazaki et al., 2011). In brief, the procedures were as follows. Approximately 10 mL of each sample solution was pipetted into a vial with a septum cap. Then, 0.5 g of spongy cadmium was added, followed by 150 µL of a 1 M NaHCO3 solution. The sample was then shaken for 18–24 h at a rate of 2 cycles s-1. Then, the sample solution was decanted into a different vial with a septum cap. After purging the solution using high purity helium, 0.4 mL of the azide/acetic acid buffer was added. After 45 min, the solution was made basic by adding 0.2 mL of 6 M NaOH.

Then, the stable isotopic compositions (δ15N, δ18O, and Δ17O) of N2O in each vial were determined by using a continuous-flow isotope ratio mass spectrometry (CF-IRMS) system at Nagoya University. The analytical procedures using the CF-IRMS system were the same as those detailed in previous research (Komatsu et al., 2008; Hirota et al., 2010). The obtained values of δ15N, δ18O, and Δ17O for N2O derived from the nitrate in each sample were compared with those derived from our local laboratory nitrate standards that had been calibrated using the internationally distributed isotope reference materials to calibrate the values of the sample nitrate to an international scale and to correct for both the isotope fractionation during the chemical conversion to N2O and the progress of oxygen isotope exchange between the nitrate-derived reaction intermediate and water (ca. 20 %). In this study, we adopted the internal standard method (Nakagawa et al., 2013; Tsunogai et al., 2014) for the calibrations of sample nitrate. All values in this paper are expressed relative to air (for nitrogen) and VSMOW (for oxygen).

To determine whether samples were deteriorated or contaminated during storage and whether the conversion rate from nitrate to N2O was sufficient, concentrations of nitrate in the samples were determined each time we analysed isotopic compositions using CF-IRMS based on the N2O+ or O2+ outputs. We adopted the δ15N, δ18O, or Δ17O values only when concentrations measured by CF-IRMS correlated with those measured by ion chromatography just after the sampling within a difference of 10 %. About 10 % of the whole isotope analyses showed conversion efficiencies lower than this criterion. Nitrate in these samples was converted to N2O again and re-analysed for the stable isotopic compositions. None of the samples showed significant nitrate deterioration or nitrate contamination during storage.

We repeated the analyses of the δ15N, δ18O, and Δ17O values of nitrate for each sample at least three times to attain high precision. Most of the samples had a nitrate concentration of more than 5.0 µmol L-1, which corresponded to a nitrate quantity greater than 50 nmol in a 10 mL sample. This amount was sufficient for determining the δ15N, δ18O, and Δ17O values with high precision. For cases of nitrate concentration less than 5.0 µmol L-1, the sample volume was increased to 30 mL and the number of analyses was also increased. Thus, all isotopic data presented in this study have an error (standard error of the mean) better than ±0.2 ‰ for δ15N, ±0.3 ‰ for δ18O, and ±0.1 ‰ for Δ17O.

Nitrite (NO2-) in the samples interferes with the final N2O produced from nitrate (NO3-) because the chemical method also converts NO2- to N2O (McIlvin and Altabet, 2005). Therefore, it is sometimes necessary to correct for the contribution of NO2--derived N2O to determine the stable isotopic compositions of the sample nitrate accurately. However, in this study, more than 90 % of the samples analysed for stable isotopic compositions had NO2- concentrations lower than the detection limit (0.05 µ mol L-1). Even for the samples with NO2- concentrations higher than the detection limit, the NO2- / NO3- ratio was less than 1 %. Thus, in this study, the results were used with no correction.

To clarify the chemical and isotopic characteristics of each stream, we determined both the flow-weighted annual average concentration (C‾total) and the flow-weighted annual average δ15N, δ18O, and Δ17O values (δ‾) of nitrate for each stream assuming the same flow rate, the same nitrate concentration, and the same isotopic compositions for the interval until the next observation by using Eqs. (3), (4), and (5): q=∑n=14fn⋅Δtn,C‾total=∑n=14Cn⋅fn⋅Δtnq,δ‾=∑n=14δn⋅Cn⋅fn⋅Δtn∑n=14Cn⋅fn⋅Δtn, where Cn and δn denote the concentration (Ctotal in Eq. 2) and isotopic values (δ15N, δ18O, or Δ17O) of nitrate in each stream during each observation n respectively; fn denotes the flow rate of each stream during each observation n; and Δtn denotes the time interval between the observation n and the next observation n+1. When possible, we used the flow rate of each stream that was determined monthly by the Shiga Prefecture (Shiga Prefecture, 2015) for fn. For small streams with no data for flow rate (n= 13), we used a small and stable flow rate of 0.1 m3 s-1 for fn.

To exclude the contribution of NO3-(atm) from the δ15N and δ18O values of nitrate (total) and to clarify the sources and behaviour of NO3-(re) by using both δ15N and δ18O as tracers, we estimated the end-member δ15N and δ18O values of the remineralised nitrate portion, δ15Nre and δ18Ore, by excluding the contribution of NO3-(atm) in nitrate (total; Tsunogai et al., 2010, 2011, 2014; Dejwakh et al., 2012; Liu et al., 2013; Riha et al., 2014) by using Eqs. (6) and (7): δ15Nre=Ctotal⋅δ15N-Catm⋅δ15NatmCtotal-Catm,δ18Ore=Ctotal⋅δ18O-Catm⋅δ18OatmCtotal-Catm, where Catm and Ctotal denote the concentrations of NO3-(atm) and nitrate (total) in each water sample respectively, and δ15Natm, δ18Oatm, and Δ17Oatm denote the δ15N, δ18O, and Δ17O values of NO3-(atm) in each sample respectively. The actual values of δ15Natm, δ18Oatm, and Δ17Oatm used in this study were determined from the δ15N, δ18O, and Δ17O values of nitrate in the wet deposition samples and are reported in Sect. 3.1 along with the ranges of errors. Please note that the errors in the estimated values of δ15Nre and δ18Ore become larger in accordance with an increase in the Catm / Ctotal ratio due to the propagation law of errors, even if the errors in the values of δ15Natm and δ18Oatm are the same.

Because we used the power law shown in Eq. (1) for the definition of Δ17O, the Δ17O values are different from those based on the linear definition (Michalski et al., 2002). Please note that our Δ17O values would be (0.1 ± 0.1) ‰ higher for the stream water nitrate and (0.9 ± 0.3) ‰ higher for the atmospheric nitrate if we had used the linear definition for calculation.

Compared with Δ17O values based on the linear definition, Δ17O values based on the power law definition are more stable during mass-dependent isotope fractionation processes, so we rated the Δ17O values of nitrate as always stable irrespective of any biogeochemical partial removal processes subsequent to deposition, such as assimilation or denitrification. However, Δ17O values based on the power law definition are not conserved during mixing processes between fractions with different Δ17O values, so the Catm / Ctotal ratios estimated using Eq. (2) are somewhat deviated from the actual Catm / Ctotal ratios in the samples. However, in this study, the extent of the deviations of the Catm / Ctotal ratios of the stream nitrate was less than 0.15 %, so we disregard this effect in the discussion.

The δ15N, δ18O, and Δ17O of atmospheric nitrate NO3-(atm) are shown in Fig. 3a–c respectively as a function of sampling day (local time, UT + 09:00), and the daily depositional flux of NO3-(atm) when each of the wet deposition samples was taken is also shown in Fig. 3c. The daily depositional flux of NO3-(atm) was calculated from the nitrate concentration and the daily precipitation (Table S2 in the Supplement).

The atmospheric nitrate at the Sado-seki monitoring station showed large 17O excesses with Δ17O values from +18.6 to +32.4 ‰. Moreover, a clear normal correlation between Δ17O and δ18O was shown (r2= 0.878; Fig. 4). A similar trend was reported for atmospheric nitrate aerosols collected for a 1-year period in La Jolla, California (32.7∘ N, 117.2∘ W; Michalski et al., 2003), and similar results also have been obtained in other areas of the world (Kaiser et al., 2007; Morin et al., 2009). Michalski et al. (2003) interpreted that the linear correlation corresponds to the mixing line between tropospheric ozone and tropospheric H2O and thus tropospheric OH radicals, with Δ17O = 0 ‰ and δ18O =-5 ‰. However, the NO3-(atm) data obtained at the Sado-seki monitoring station showed a somewhat different trend in the Δ17O–δ18O plot between summer, from May to October, and winter, from November to April (Fig. 4). Although the line fitted to the summer data showed a slope of 2.21 ± 0.22 and an intercept of (+19.7 ± 5.1) ‰ in the Δ17O–δ18O plot, that of the winter data showed a statistically significant larger slope of 2.89 ± 0.38 and a smaller intercept of (+3.0 ± 9.2) ‰; all errors were in the 2σ range. Although the winter data included an intercept of -5 ‰ reported by Michalski et al. (2003) as the end-member δ18O value of the tropospheric OH radical within the possible error range, the intercept of summer data deviated strongly from the value. Because the monitoring station is located in the Asian monsoon area, the major air mass that arrived at the station was different seasonally: Pacific air originated from south-east was dominant in summer, whereas continental air originated from the north-west was dominant in winter. The present results imply seasonal and regional changes in the δ18O / Δ17O ratios of tropospheric ozone and the OH radical.

On the basis of both the temporal variation in the daily depositional flux of NO3-(atm), shown in Fig. 3c, and the Δ17O value, we estimated the monthly average Δ17O value of NO3-(atm) (Δ17O(m)) deposited at the Sado-seki monitoring station for each month (m) from April 2009 to March 2012 by using Δ17O(m)=∑kCk⋅Vk⋅Δ17Ok∑kCk⋅Vk, where Ck denotes the concentration of nitrate in each wet deposition sample and Vk denotes the total water volume of each wet deposition sample. Then, we estimated the annual and the 3-year average Δ17O values of NO3-(atm) (Δ17Oavg) as +25.5 ‰ for FY2009, +27.2 ‰ for FY2010, +25.7 ‰ for FY2011, and +26.3 ‰ for the 3 years by using Δ17Oavg=∑mDm⋅Δ17Om∑mDm, where D(m) denotes the monthly wet deposition rate of nitrate at the Sado-seki monitoring station determined by EANET (EANET, 2014). Because no wet deposition sample for measuring stable isotopes was taken in May 2009 or March 2012, we used the Δ17O(m) values of May 2010 and March 2011 respectively for these values. Substituting Δ17O with δ15N (δ18O) in Eqs. (8) and (9), we estimated δ15Navg (δ18Oavg) as -4.4 ‰ (+78.5 ‰) for FY2009, -3.8 ‰ (+81.8 ‰) for FY2010, -4.4 ‰ (+78.6 ‰) for FY2011, and -4.2 ‰ (+79.8 ‰) for the 3 years.

To apply the Δ17Oavg values for the 3 years obtained at the Sado-seki monitoring station as Δ17O of NO3-(atm) deposited on the studied watershed (i.e. Δ17Oatm in Eq. 2), additional corrections could be needed because the Δ17O value of NO3-(atm) is a function of the NOx oxidation channels in the atmosphere, which shift depending on the intensity of sunlight, temperature, and oxidant levels (e.g. Michalski et al., 2003; Morin et al., 2008; Kunasek et al., 2008; Alexander et al., 2009; Morin et al., 2012; Savarino et al., 2013). The latitudinal difference between the Sado-seki monitoring station (38∘15′ N, 138∘24′ E; Fig. 2) and the watershed studied (35∘15′ N, 136∘5′ E; Fig. 2) could change the intensity of sunlight and thus the NOx oxidation channel. Moreover, Tsunogai et al. (2010) reported that nitrate in polluted air masses derived directly from megacities in winter showed slightly larger Δ17O values than nitrate in background air masses in the same seasons owing likely to the relative increase in the reaction via NO3 radicals within the entire NO3-(atm) production channel to produce NO3-(atm) in the polluted air mass. The annual average Δ17O value determined in this study was lowest in FY2009 when the deposition rate of nitrate was the smallest, at 19.3 mmol m-2 yr-1 (EANET, 2014.), whereas the annual average Δ17O value was highest in FY2010 when the deposition rate was the largest, at 28.0 mmol m-2 yr-1, within the 3 years of observation. These results also imply that the difference in the deposition rate of nitrate must be also corrected to apply the Δ17Oavg value to Δ17Oatm.

Nevertheless, both the annual average and the seasonal variation range of Δ17O correlated strongly with those determined at the Rishiri monitoring station (45∘07′11′′ N, 141∘12′33′′ E; Fig. 2) at +26.2 ‰ in FY2008 (Tsunogai et al., 2010), where the wet deposition rate of NO3-(atm) was an average of 40 % smaller than that at the Sado-seki monitoring station from 2000 to 2013 (EANET, 2014). Moreover, the values also coincided with those reported for midlatitudes, such as at La Jolla, at 33∘ N (Michalski et al., 2003), and at Princeton, at 40∘ N (Kaiser et al., 2007). We concluded that by allowing 1 ‰ of error, the standard deviation of the Δ17Oavg values determined at the four different monitoring stations located within the same midlatitude range in the past, the obtained Δ17Oavg value of NO3-(atm) can be considered representative for middle latitudes worldwide, including the Lake Biwa watershed basin.

In addition, the actual Δ17Oatm values of NO3-(atm) in each stream water sample can differ from the Δ17Oavg owing to the seasonal variation in the Δ17O values of NO3-(atm). In correcting for the seasonal variation, however, it is not adequate to use the Δ17O values determined for the seasons of sampling as Δ17Oatm in Eq. (2), because the residence time of water is longer than a few months for most of the catchments in Japan with a humid temperate climate (Takimoto et al., 1994; Kabeya et al., 2007). That is, the nitrate in base flow stream water had been stored previously in subsurface runoff and groundwater, for which seasonal Δ17O changes have not been found thus far (Tsunogai et al., 2010; Nakagawa et al., 2013). We concluded that by allowing an additional 1.8 ‰ of error, the standard deviation of the 6-month moving average of atmospheric nitrate determined at the Sado monitoring station in this study, the obtained Δ17Oavg value of NO3-(atm) represented those eluted from the Lake Biwa watershed basin.

In summary, we used the 3-year average Δ17O value of NO3-(atm) obtained at the Sado-seki monitoring station in this study (Δ17Oavg=+26.3 ‰) as the Δ17Oatm in Eq. (2) to estimate Catm in the streams of the Lake Biwa watershed basin by allowing an error range of 3 ‰, considering the factor changes of Δ17Oatm from Δ17Oavg described above. About 65 % of all of the Δ17O data of NO3-(atm) obtained at the Sado-seki monitoring station were included in this range of (+26.3 ± 3.0) ‰.

In the case of the δ15N and δ18O of NO3-(atm) in each stream water sample (i.e. δ15Natm and δ18Oatm in Eqs. 6 and 7), the values differed further from δ15Navg and δ18Oavg owing to isotopic fractionation during partial removal subsequent to deposition. As a result, while using the 3-year average values of δ15N (δ15Navg= -4.2 ‰) and δ18O (δ18Oavg=+79.8 ‰) as δ15Natm and δ18Oatm, we assumed much a larger error range in the values, i.e. ±10 ‰ for both δ15N and δ18O, twice the enrichment factor during assimilation of nitrate. Because the concentration of atmospheric nitrate would be reduced to e-2 of the original value if δ15N and δ18O values increased +10 ‰ from their original values through assimilation, it might be difficult to detect atmospheric nitrate within total nitrate. Of course, this estimation is less reliable, and we further discuss the appropriateness of these error ranges in Sect. 3.3. Because of the small Catm / Ctotal ratios of stream water of generally less than 7 % (Sect. 3.2), the error propagated to δ15Nre and δ18Ore was generally small, less than 1 and 2 ‰ respectively for most of the data presented in this study (Table S5).

The concentrations (Ctotal) and the δ15N, δ18O, and Δ17O values of nitrate in the stream water samples determined for each observation (n= 1 2, 3, and 4) are presented in Fig. 5a–d. The annual average concentration (C‾total) and the annual average δ15N, δ18O, and Δ17O values (δ15N‾, δ18O‾, and Δ17O‾ respectively) in each stream estimated by using Eqs. (3), (4), and (5) are shown in the figure as black bars. In this figure, each stream was plotted on the x axis in the order of location, beginning from stream no. 31, which lies south-west of all of the streams (Fig. 2), and proceeding in a clockwise direction. The errors are comparable to the sizes of the symbols in Fig. 5a–d. The spatially continuous variation of the values of δ15N‾, δ18O‾, and δ17O‾ imply that the values may represent land use changes in each catchment area.

Although the Δ17O values presented significant spatial and temporal variation from +0.0 to +6.8 ‰, the range of the Δ17O‾ values from +0.5 to +3.1 ‰ was typical for nitrate in natural stream water (Michalski et al., 2004; Tsunogai et al., 2010, 2014; Liu et al., 2013; Rose et al., 2015). These results correspond to mole fractions of unprocessed NO3-(atm) to total nitrate from (1.8 ± 0.3) to (11.8 ± 1.3) %, obtained by using Eq. (2).

By using the concentration (Ctotal) and the Δ17O values of nitrate, the NO3-(atm) concentration (Catm) was calculated using Eq. (2) and is plotted in Fig. 5e. In addition, the annual average concentration of NO3-(atm) (C‾atm) in stream nitrate was calculated and is also presented in Fig. 5e as black bars. Please note that the errors in Catm and C‾atm are not presented in Fig. 5e (see Table S4 for the respective ranges of errors).

To verify possible secular changes (i.e. long-term non-periodic variation), the estimated C‾total,δ15N‾, and δ18O‾ for each stream were compared with those determined by Ohte et al. (2010), in which annual average concentration and annual average δ15N and δ18O values of nitrate (total) were determined for the same streams in 2004 to 2006. Although both concentrations and δ15N and δ18O values in the streams showed significant spatial and temporal variations during 2013, as presented in Fig. 5, the annual average values almost correlated with the values determined in 2004 to 2006 (Fig. S1 in the Supplement). The average differences from the values determined in the streams in 2004 to 2006 were +5.3 µmol L-1 for C‾total, +0.6 ‰ for δ15N‾, and +1.6 ‰ for δ18O‾, whereas the standard deviation ranges of the differences were 14.9 µmol L-1 for C‾total, 1.6 ‰ for δ15N‾, and 2.1 ‰ for δ18O‾. That is, the differences from the values determined in 2004 to 2006 were smaller than their standard deviation ranges, so the differences were not significant. We concluded that secular changes were minimal for nitrate in the streams, at least for the most recent 10-year period of observations.

One of the features in the spatial variation shown in Fig. 5 is the positive correlation between Δ17O and δ18O. As is clearly represented by the relationship between Δ17O‾ and δ18O‾ (Fig. 6), these values showed linear correlation with an r2 value of 0.88. Because NO3-(atm) is characterised by highly elevated values of both Δ17O and δ18O (Figs. 3 and 4), changes in the mole fraction of unprocessed NO3-(atm) within the total nitrate pool must be responsible for the positive correlation between Δ17O‾ and δ18O‾ for nitrate in the streams.

The slope value of the least-squares-fitted line between Δ17O‾ and δ18O‾ (Fig. 6) also supports this hypothesis. By extrapolating the least-squares-fitted line to the region of NO3-(atm) with a Δ17O value of +26.3 ‰, we obtained δ18O = (+86 ± 7) ‰, which also corresponds with the average δ18O value of NO3-(atm) of +79.8 ‰ obtained in Sect. 3.1. We concluded that the δ18O‾ values also primarily reflect the mole fraction of NO3-(atm) within nitrate.

Without Δ17O data and without some assumptions, it was difficult to decide the major factor controlling the δ18O values of the stream nitrate. However, by adding Δ17O data, as presented above, it became apparent that the changes in the mole fraction of unprocessed NO3-(atm) within the total nitrate pool were primarily responsible for the δ18O variation between the streams. These results further support our hypothesis, presented in Sect. 1, that interpretations on the stable isotopic compositions of nitrate (δ15N and δ18O) made without Δ17O values can often be misleading. When using stable isotopic compositions, particularly the δ18O value, of nitrate in freshwater environments to trace its sources and fate, the determination of Δ17O values is essential.

By extrapolating the linear correlation between Δ17O‾ and δ18O‾ to Δ17O‾ = 0 ‰, we obtained the δ18O value of (-2.9 ± 1.2) ‰ as the average δ18O value of the remineralised portion of nitrate (NO3-(re)) in the streams. Although the δ18O value was substantially 18O depleted compared with that produced through microbial nitrification in soil during in vitro incubation experiments in past studies (Mayer et al., 2001; Burns and Kendall, 2002; Spoelstra et al., 2007), it correlated strongly with the δ18O values of NO3-(re) determined recently by using the linear relationship between Δ17O and δ18O of nitrate eluted from forested watersheds, such as NO3-(re) in the groundwater of cool-temperate forested watersheds at (-4.2 ± 2.4) ‰, where the δ18O(H2O) was around -13 ‰ (Tsunogai et al., 2010), and NO3-(re) in stream water in a cool-temperate forested watershed at (-3.6 ± 0.7) ‰, where the δ18O(H2O) was around -11 ‰ (Tsunogai et al., 2014). Moreover, the δ18O value of NO3-(re) obtained in this study, (-2.9 ± 1.2) ‰, is close to the possible lowermost δ18O value of NO3-(re) (Buchwald et al., 2012) produced through microbial nitrification under H2O of (-7.8 ± 1.0) ‰ (the average and the standard deviation of the δ18O values of H2O of the streams; Table S3). Furthermore, the δ18O value of NO3-(re) correlates strongly with that obtained through in vitro incubation experiments in recent studies that simulated temperate forest soils (Fang et al., 2012). We concluded that the δ18O value of NO3-(re) produced through nitrification in the temperate watershed with a δ18O(H2O) value of (-7.8 ± 1.0) ‰ was (-2.9 ± 1.2) ‰ and that we should use such a low δ18O value for the NO3-(re) produced through nitrification in the watershed. Understanding the relationship between Δ17O and δ18O of nitrate shown in Fig. 6 is highly useful for determining the δ18O value of NO3-(re) in each watershed (Tsunogai et al., 2010).

Although the Δ17O values of nitrate were stable during partial biogeochemical processing, such as partial removal through assimilation or denitrification, the δ18O values of nitrate could vary through the isotopic fractionation processes within each catchment area. Nevertheless, the δ18O‾ values of nitrate in the streams plotted on the mixing line between the NO3-(atm) that had been deposited in the watershed and NO3-(re) with δ18O and Δ17O values close to those produced through nitrification in the catchments. Thus, we concluded that the range of isotopic fractionations owing to partial removal through assimilation or denitrification subsequent to deposition of NO3-(atm) or production of NO3-(re) within each catchment area was generally small for the major portion of nitrate eluted from the watershed. If such isotopic fractionations were significant for the portion of NO3-(atm) in total nitrate, the data should plot on the 18O-enriched side of Fig. 6, especially for those data enriched in NO3-(atm) (i.e. those showing high Δ17O values). This result also supports our assumption in Sect. 3.1 that the actual δ15N and δ18O values of NO3-(atm) in each stream water sample (δ15Natm and δ18Oatm in Eqs. 6 and 7) correlate with the δ15Navg and δ18Oavg estimated at the Sado-seki monitoring station within an error of ±10 ‰.

To trace the source of the 18O-depleted NO3-(re) eluted from the watershed into the lake, the annual average δ15N and δ18O values of the remineralised portion of nitrate (δ15N‾re and δ18O‾re) in each inflow stream were estimated using Eqs. (6) and (7) and are plotted as a function of population density in Fig. 7c and d respectively. The original δ18O‾ is also in Fig. 7d. Because of the large δ18O differences of about 80 ‰ between the nitrate in streams and the NO3-(atm), the δ18Ore values were a few ‰ lower than each original δ18O value in total nitrate (Fig. 7d). On the contrary, because of the small δ15N differences of less than 15 ‰ between the total nitrate in streams and NO3-(atm), as well as the small Catm / Ctotal ratios in the streams, most of the δ15Nre values showed small deviations of less than 1 ‰ from each corresponding original δ15N value in the total nitrate in most of the streams, so the original δ15N‾ values are not presented in Fig. 7c. Although the annual average δ18O‾re values were low and almost uniform, from -4.0 to -0.1 ‰, as implied in the linear correlation between Δ17O‾ and δ18O‾ in Fig. 6, δ15N‾re showed larger variation from +1.7 to +10.9 ‰.

Moreover, δ15N‾re showed positive linear correlation with the population density in logarithmic scale (r2= 0.64, p < 0.001; Fig. 7c). A similar trend was reported for the δ15N values of total nitrate (= NO3-(atm) + NO3-(re)) in past studies in this watershed (Ohte et al., 2010) and others (Mayer et al., 2002). We further verified that the remineralised portion of nitrate (NO3-(re)) was responsible for the positive correlation between the δ15N values of total nitrate and population density that has been found often in various streams in the world.

Both the concentrations and the isotopic compositions shown in Fig. 7 clearly demonstrate that most portions of the nitrate eluted from the catchments with lower population densities of less than 100 km-2, showing δ15N values of (+4.4 ± 1.8) ‰ and δ18O values of about (-2.3 ± 0.9) ‰, were produced through nitrification in naturally occurring soil organic matter (Kendall et al., 1995; Ohte et al., 2010). In the latter half of this section, we discuss the source of the 15N-enriched NO3-(re) eluted from the catchments with higher population densities of more than 1000 km-2, showing δ15N values of (+9.2 ± 1.3) ‰ or more and δ18O values of about (-2.2 ± 1.1) ‰.

Denitrification in riverbed sediments adjacent to riparian zones or groundwater bodies (McMahon and Böhlke, 1996) can increase the δ15N value of stream nitrate. However, if such post-production alternation were responsible for the 15N enrichment of NO3-(re) and thus the total nitrate, the values of δ18Ore in addition to those of δ15Nre would be increased (Granger et al., 2008). Moreover, the absolute concentration of NO3-(atm) (Catm) would decrease in accordance with the progress of denitrification, but the observed Catm was almost uniform irrespective of population density. Of course, the initial Catm could vary between the streams. However, to explain the observed uniform Catm, unrealistic assumptions are needed for Catm, such as the initial Catm being higher in accordance with higher population density. The low and uniform δ18Ore values (Fig. 7d) as well as the uniform Catm irrespective of population density (Fig. 7b) imply that denitrification in riverbed sediments was minor for the nitrate in the streams. Rather, the NO3-(re) must be enriched in 15N from its initial production through nitrification within the catchments with high population densities. In addition, the small differences in δ18O values of NO3-(re) between those values irrespective of the population densities in the catchment area (Fig. 7d) imply that the essential parameters for determining the δ18O values of nitrate during nitrification, such as the δ18O values of H2O and the pH of soils (Buchwald et al., 2012; Fang et al., 2012), should be similar among them.

Based on the δ15N values of total nitrate eluted from catchments with high population densities, as well as the positive correlation between the δ15N values of total nitrate and the population densities, Ohte et al. (2010) proposed sewage effluent as the dominant source contributing to the increase in the δ15N values of total nitrate eluted from such catchments. The δ15N and δ18O values of NO3-(re) newly estimated in this study, (+9.2 ± 1.3) ‰ or more and (-2.2 ± 1.1) ‰ respectively, also imply that the dominant source contributing to the increase in the δ15N values of total nitrate had been produced through nitrification in which the source N of the nitrate had already been enriched in 15N. Although the δ15N and δ18O values of total nitrate in sewage effluent determined in past studies (Aravena et al., 1993; Widory et al., 2005; Wankel et al., 2006; Xue et al., 2009) were a few per mille higher than the δ15N and δ18O values of NO3-(re) eluted from the high population density catchments, δ15Nre= (+9.2 ± 1.3) ‰ and δ18Ore= (-2.2 ± 1.1) ‰, the slight deviations in the reported δ15N and δ18O values from our results can be explained by several factors, such as (1) a slight contribution of NO3-(atm) and (2) the progress of denitrification subsequent to production. We concluded that sewage effluent was the most probable pollution source of nitrate to explain the observed concentrations and isotopic compositions of nitrate eluted from the catchments with high population densities, particularly for those of more than 1000 km-2.

Although the annual average values of Δ17O and δ18O in each stream, Δ17O‾ and δ18O‾ respectively, showed linear correlation, as presented in Fig. 6, the same results were not always attained for those in each season. Particularly for those values obtained during June and August (i.e. summer), some of the streams showed significant deviations in δ18O of more than a few per mille from the hypothetical mixing line between NO3-(atm) (Δ17O =+26.3 and δ18O =+79.8 ‰) and NO3-(re) (Δ17O = 0 and δ18O =-2.9  ‰; Fig. 8). Even though the values of Δ17O and δ18O of NO3-(atm) showed seasonal variation, as presented in Figs. 3 and 4, the large deviations from the mixing line could not be explained based on the seasonal changes in NO3-(atm). Rather, we must assume some seasonal changes in the biogeochemical nitrogen cycles within each catchment area to explain the relationship because, unlike the Δ17O values, the δ18O values of nitrate could vary during biogeochemical processing within each catchment area. As a result, we can evaluate the seasonal changes in the biogeochemical processing within each catchment area by using the seasonal changes in the relationship between Δ17O and δ18O shown in Fig. 8.

The increases in the number of data deviated from the hypothetical mixing line by more than a few per mille from the hypothetical mixing line in June and August, especially those plotted on the 18O-enriched region (i.e. vertically upward direction in the figures), imply that partial nitrate removal through assimilation or denitrification was active within each catchment area in these months. The spatial differences in the 18O enrichment also support this hypothesis. As presented in Fig. 8 by the orange squares, 18O enrichment was common in samples obtained from the southern streams with high population densities in their catchment areas. We can assume elevated loading of both nutrients and organic matter of anthropogenic origin in these catchments, both of which naturally enhance assimilation and denitrification.

In contrast, most samples obtained during March and October were distributed on the hypothetical mixing line between NO3-(atm) and NO3-(re), as presented in Fig. 8. We concluded that the range of isotopic fractionation subsequent to production, such as partial removal through assimilation or denitrification, in winter was generally small for the major portion of nitrate eluted from the watershed and fed into the lake. Therefore, the annual average values (i.e. δ18O‾ and Δ17O‾) of the streams distributed on the hypothetical mixing line, as shown in Fig. 6, because the nitrate influx in winter occupied a major portion of the annual nitrate influx. Active removal of nitrate from the streams through denitrification/assimilation in summer was also responsible for the small relative importance of nitrate influx into the lake in summer. In conclusion, the relationship between Δ17O and δ18O of nitrate eluted from a catchment area is a useful indicator for evaluating the biogeochemical processing within the catchment area, including the seasonal change.

By using the δ18O values of nitrate as a tracer, Ohte et al. (2010) found that the mole fractions of unprocessed NO3-(atm) within the total nitrate pool were high in the northern streams of the watershed in winter, from November to late April. Our present results shown in Fig. 5 further verify the past results by adding more robust evidence through the use of the Δ17O tracer for NO3-(atm).

Based on the high accumulation rate of snow in the catchment zones of the northern streams, Ohte et al. (2010) concluded that high loading of unprocessed NO3-(atm) via snow in the catchment zones increased the stored unprocessed NO3-(atm) in the snowpack, which was subsequently released into the streams during the melting seasons. This process enhanced the mole fraction of unprocessed NO3-(atm) within the total nitrate pool during the melting season, as was also reported for streams worldwide (Kendall et al., 1995; Ohte et al., 2004, 2010; Piatek et al., 2005; Pellerin et al., 2012; Tsunogai et al., 2014). However, the contribution of nitrate from anthropogenic sources could be smaller in this northern area because of lower population densities in the catchments (Table 1). Because a major portion of the possible anthropogenic nitrate in the catchments must be occupied by NO3-(re) (Ohte et al., 2010), a lower NO3-(re) supply from anthropogenic sources in each catchment area could elevate the mole fraction of unprocessed NO3-(atm) within the total nitrate pool, even if the absolute concentration of NO3-(atm) (Catm) were uniform in the streams.

To determine the Catm variability among the streams, the Catm values estimated in this study were plotted as a function of population density, as shown in Fig. 7b. The Catm was almost uniform at (2.3 ± 1.1) µmol L-1 irrespective of changes in the population density of the catchment areas. However, a clear Ctotal enrichment trend was noted in accordance with increasing population density of the catchments (Fig. 7a). Similar Ctotal enrichment trends have been reported in previous studies (Ohte et al., 2010).

The northern streams such as nos. 3, 4, and 5 were enriched in Catm, showing annual average Catm of (5.3 ± 1.1), (2.9 ± 0.7), and (4.3 ± 0.9) µmol L-1 respectively and Δ17O values of +3.1, +1.9, and +2.9 ‰ respectively. These results support the previous observation of the streams determined by using the δ18O tracer. Similar Catm enrichment of about 3 µmol L-1 or more, however, was also found in streams in other areas, such as nos. 14 (Catm= (3.3 ± 1.0) µmol L-1), 25 ((3.2 ± 0.8) µmol L-1), and 21 ((4.2 ± 0.9) µmol L-1), but these streams showed lower Δ17O values of +0.9, +1.5, and +2.0 ‰ respectively and thus low mole fractions of unprocessed NO3-(atm) within total nitrate. We concluded that the difference in the addition of anthropogenic nitrate composed of NO3-(re) in the catchments was primarily responsible for the difference in the mole fraction of unprocessed NO3-(atm) within the total nitrate pool, as well as the Ctotal variation in accordance with the population densities of the catchment areas, as illustrated in Fig. 1. That is, a small contribution of anthropogenic nitrate in the catchments of the northern rivers was primarily responsible for the low Ctotal and thus the high mole fraction of unprocessed NO3-(atm) within the total nitrate pool, or the Catm / Ctotal ratio, in the northern streams of the watershed.

Although the difference in the accumulation rate of snow between each catchment zone was not the major factor controlling the Catm / Ctotal ratios, the concentrated release of NO3-(atm) stored in the snowpack during the melting seasons should be one of the important factors determining the Catm variation among the streams. Most of the Catm-enriched streams, such as nos. 3, 4, 5, 14, and 25, originated from a forested catchment at a high elevation of more than 800 m above sea level; thus, we can anticipate heavy snowpack in winter in each headwater region. Moreover, the maximum Catm values in these streams were found in March, which is the season of snowmelt (Fig. 5). On the contrary, most of the Catm-depleted streams, such as nos. 29 ((0.9 ± 0.3) µmol L-1), 19 ((0.8 ± 0.3) µmol L-1), 23 ((0.5 ± 0.2) µmol L-1), and 30 ((0.6 ± 0.2) µmol L-1), originated from low elevations with urban and suburban catchment areas (Table 1). As a result, the concentrated release of stored NO3-(atm) in the snowpack to the forest floor in the catchment zone during the melting seasons is strongly responsible for the Catm enrichment of some of the streams, particularly that in the streams during the month of March, as presented in Fig. 1.

The only exception is stream no. 21, located in the southernmost part of the watershed, which showed a high annual average Catm of (4.2 ± 0.9) µmol L-1. This small stream originates from a low elevation of about 200 m and has a small catchment area of 4 km2. In addition, although the other Catm-enriched streams showed the maximum Catm in March, Catm in stream no. 21 was highest in August, showing an extraordinarily high value of more than 10 µmol L-1. It is unlikely that NO3-(atm) stored in the snowpack in winter was the major source of NO3-(atm) in this stream.

The catchment zone of stream no. 21 had the highest population density of the catchments of the streams studied (Table 1). About one-third of the catchment includes residential areas. Artificial drainage systems in urban or residential areas and agricultural lands in humid temperate regions are usually designed to drain rainwater efficiently into streams (Takimoto et al., 1994). As a result, a significant portion of the NO3-(atm) deposited into the catchment area was deposited onto paved surfaces and was then drained directly into the stream via storm sewers without penetrating the ground. Thus, no interaction with soils occurred, as presented in Fig. 1. Because biogeochemical interactions within soils are the major sink for NO3-(atm) and thus for 17O excess of nitrate (Nakagawa et al., 2013; Tsunogai et al., 2014), the development of such drainage systems in urban/suburban areas is largely responsible for the high Catm in stream no. 21. Similar bypassing effects of NO3-(atm) from soil contact by paved surfaces have been suggested in urban/suburban watersheds by using δ18O values of nitrate as a tracer (Burns et al., 2009; Kaushal et al., 2011). We further verified that the drainage systems in urban/suburban catchments changed the fate of the NO3-(atm) deposited onto land to some extent.

The observed uniform Catm irrespective of population density and headwater elevation shown in Fig. 7 implies that the influences of snowpacks and paved surfaces were still minor in determining the Catm values in the streams. Rather, the observed stable Catm implies that most of the NO3-(atm) in the streams had been stored in groundwater/subsurface runoff in the watershed, which had similar Catm concentrations, and then gushed to the surface at the respective headwater zones with various elevations and land uses, as presented in Fig. 1.

When using the δ18O tracer, it was difficult to determine the precise absolute concentration of NO3-(atm) (Catm) in each stream water, as presented in this study, and to determine whether the absolute concentration of NO3-(atm) was stable among the streams. However, by using the Δ17O values, we can determine the precise Catm in each stream for each season and thus clarify the fate of NO3-(atm).

The concentrations and δ15N, δ18O, and Δ17O values of nitrate in the outflow river (Seta River; no. 33) are also presented in Fig. 5. In a manner similar to the inflow streams (i.e. by using Eqs. 3 to 5), we estimated the annual average concentration of total nitrate in the outflow river (C‾total) to be 13.3 µmol L-1, the annual average δ15N value (δ15N‾) to be +13.1 ‰, the annual average δ18O value (δ18O‾) to be +1.5 ‰, and the annual average Δ17O value (Δ17O‾) to be +0.9 ‰, as presented in Fig. 5. Moreover, in a manner similar to that used for the inflow streams (i.e. by using Eqs. 2, 6, and 7), we estimated the annual average concentration of NO3-(atm) in the outflow river (C‾atm) to be (0.4 ± 0.1) µmol L-1 (Fig. 5e), the annual average δ15Nre value (δ15N‾re) to be (+13.7 ± 0.6) ‰, and the annual average δ18Ore value (δ18O‾re) to be (-1.2 ± 0.9) ‰ (Table S5). Similar to those for inflows, the Δ17O values were typical for nitrate in natural stream waters.

The striking features of the outflow in comparison with the inflows were the depletions in the outflow of both Ctotal and Catm as well as the enrichment in 15N (Fig. 5). Because the denitrification/assimilation processes remove both nitrate and NO3-(atm) and preferentially consume 14N during the removal, the process of denitrification/assimilation in the lake water column can be strongly responsible for the removal of both nitrate and NO3-(atm) and for the 15N enrichment of nitrate in the outflow compared with the inflow. If this were the case in Lake Biwa, the total nitrate efflux would have to be smaller than the total nitrate influx. To verify this hypothesis quantitatively and to evaluate the influences of the stagnant flow in the lake on nitrate, we estimated the total influx through all of the inflow streams for nitrate and NO3-(atm), ΔNin, and ΔAin respectively and the total efflux for nitrate and NO3-(atm), ΔNout, and ΔAout respectively, as well as the flow-weighted average δ15N, δ18O, and Δ17O values of all inflows and outflows to discuss their changes in the lake.

The ΔNin and ΔAin in each interval between the observation n and the next observation n+1 (i.e. each season) and the flow-weighted average δ15N, δ18O, and Δ17O values of the inflows (δ(n)) during each interval between the observation n and the next observation n+1 were determined by using the following equations, assuming the same flow rate, the same nitrate concentration, and the same isotopic compositions for the interval until the next observation: α=Qin∑iqi,ΔNinn=∑iCi⋅fi⋅Δtn⋅α,δn=∑iδi⋅Ci⋅fi⋅Δtn∑iCi⋅fi⋅Δtn,ΔAinn=ΔNinn⋅Δ17OinnΔ17Oatm,ΔNin=∑n=14ΔNinn,ΔAin=∑n=14ΔAinn,δ=∑n=14δn⋅ΔNinn∑n=14ΔNinn, where Qin denotes the annual gross influx of water into the lake; Ci and δi denote the concentration and isotopic values (δ15N, δ18O, or Δ17O) of nitrate in each stream i during each observation n respectively; fi denotes the flow rate of each stream i during each observation n; and Δtn denotes the time interval between the observation n and the next observation n+1.