Towards a global interpretation of dual nitrate isotopes in surface waters

Modern anthropogenic activities have significantly increased nitrate (NO3) concentrations in surface waters. Stable isotopes (δN and δO) in NO3 offer a tool to deconvolute some of the human-made changes in the nitrogen cycle. They are often graphically illustrated on a template designed to identify different sources of NO3 and denitrification. In the two decades since this template was developed, δNand δO-NO3 have been measured in a variety of ecosystems and through the nitrogen cycle. However, its interpretation is often fuzzy or complex. This default is no longer helpful because it does not describe surface water ecosystems well and biases researchers towards denitrification as the NO3 removal pathway, even in well oxygenated systems where denitrification is likely to have little to no influence on the nitrogen cycle. We propose a different scheme to encourage a better understanding of the nitrogen cycle and interpretation of NO3 isotopes. We use a mechanistic understanding of NO3 formation to place bounds on the oxygen isotope axis and provide a means to adjust for different environmental water isotope values, so data from multiple sites and times of year can be appropriately compared. We demonstrate that any interpretation of our example datasets (Canada, Kenya, United Kingdom) show clear evidence of denitrification or a mixture of NO3 sources simply because many data points fall outside of arbitrary boxes which cannot be supported once the range of potential δONO3 values has been considered.


Background
The schematic biplot figure was originally designed for interpreting groundwater data where NO3isotope values of different NO3sources are preserved except by (chemo)denitrification (e.g., Böttcher et al. 1990;Aravena et al. 1993;Aravena and Robertson 1998). Some researchers identified that forests receiving a lot of nitrogen deposition export NO3in streams and this NO3does not retain the atmospheric deposition isotope values (e.g., Spoelstra et al. 2001;Pardo et al. 2004). This was early evidence that measured NO3isotopes in surface water showed that they should be carefully used for source identification because of various biological alterations along their flowpath. As method improvements allowed more NO3isotope data to be generated, a schematic figure that recognized biotic and abiotic processing of NO3between its sources and sampling point needed to be developed. Knowledge of isotope fractionation during NO3production and consumption was summarized in Kendall (1998) yet, despite the many figures in this chapter, one figure described as "simplified" has become the ubiquitous interpretation scheme. This figure visually summarizes a compilation of NO3isotope data with boxes by "dominant sources of nitrate" and encourages researchers to think only about one process, denitrification, although this process may be uncommon in well oxygenated lake surfaces or streams and rivers. In this way, we need a better schematic figure that explicitly recognizes the differences between NO3sources and processes that produce and consume NO3 -.
The "nitrogen axis" had been used as the primary differentiator between sources. However, given the wide range of possible δ 15 N values in manure/sewage and soils (e.g., 30‰ range in soil alone, Craine et al. 2015), and the obvious fact that nitrogen will be biologically cycled in those systems, source identification cannot be done with boxes on a figure. Moreover a system with three NO3sources and only one measurement, δ 15 N, is underdetermined. Measuring locally appropriate sources of nitrogen as potential initial δ 15 N values is the appropriate way to constrain this axis instead of relying on the broad assumption that a single set of boxes, derived from a limited number of measurements, are globally appropriate (Bateman and Kelly 2007). Without locally appropriate values, the borders between NO3sources become very blurred on the δ 15 N-NO3axis (e.g., Kendall et al. 2015) and this provides no useful resolution in the measured surface water data and no direct ability to identify sources.
In some cases, nitrogen from fertilizers and legumes will be mixed into the soil nitrogen pool (e.g., Oelmann et al. 2007) before NO3is exported to surface waters (e.g., Deutsch et al. 2006). In such cases the exported δ 15 N-NO3values will be controlled largely by the soil nitrogen pool and land-use history, rather than a single year of precipitation and fertilizer input (e.g., Loo et al. 2017). In this scenario the soil nitrogen averages all of its nitrogen inputs and NO3subsequently exported from the soil to surface water maintains this average unless there is direct input of isotopically district NO3to the surface waters. Hence the large overlap in the NO3sources boxes that does not contribute to source identification (e.g., Kendall et al. 2015).
The "oxygen axis" has groups that can be defined a priori: (i) high δ 18 O values from NO3produced in the atmosphere where the δ 18 O value depends strongly on latitude (Michalski et al. 2012); and (ii) low δ 18 O values where the δ 18 O value depends strongly on the δ 18 O of H2O where the NO3is formed (Snider et al. 2010). The δ 18 O value of NO3produced by autotrophic and heterotrophic nitrification can be bounded in two ways. First, canonical two-step nitrification (from NH4 + to NH2OH to NO2to NO3 -) adds one O atom from O2 in the first step and one O atom from H2O in each of the next two steps (Hollocher et al. 1981;Andersson et al. 1983;Aleem et al. 1965;Hollocher 1984;DiSpirito and Hooper 1986). Isotope fractionation during these steps occurs but is not always expressed, such as when NO2is fully consumed Casciotti et al. 2010;Snider et al. 2010). Abiotic equilibrium of oxygen may occur between H2O and NO2and increase the δ 18 O value of the NO2 - (Casciotti et al. 2007). In surface soils, the pore gas δ 18 O-O2 value is very likely near the atmospheric value of +23.5‰ (vs SMOW). However, in productive aquatic ecosystems, the diel variability of δ 18 O-O2 values can be large (e.g., 26‰ range in Gammons et al. 2011, 23‰ range in Venkiteswaran et al. 2015, 18‰ range in Hotchkiss and Hall, Jr 2014, 14‰ range in Wassenaar et al. 2010, and 13‰ range in Parker et al. 2005) though this range can be estimated by one set of diel samples during the most productive part of the year and analyzed via a variety of techniques (e.g., Barth et al. 2004;Wassenaar and Koehler 1999)

Site descriptions:
To highlight the need to include nitrogen cycling in surfaces waters into our working interpretation of NO3isotopes, we selected six rivers from Canada, Kenya, and the United Kingdom each with different climate regions, seasonal variation in flow, and δ 18 O-H2O values.
The Grand River, Ontario, Canada is the largest river draining into the Canadian side of Lake Erie. There are five cities, 30 wastewater treatment plants, and extensive modern agriculture along the 300km river in its 6800km 2 basin (Venkiteswaran et al. 2015). Climate is humid continental with a warm summer (Köppen-Geiger classification Dfb), average temperature is around 9°C and mean precipitation is 915mm. Samples were collected weekly to monthly from March 2015 to March 2016 from three sites: two sites upstream of the first major city and first large wastewater treatment plant and one below two cities and two large wastewater treatment plants. These sites offer the opportunity to sample from the river largely affected by diffuse non-point sources and after two large point sources (Hood et al. 2014;Venkiteswaran et al. 2018). All sites are in the middle of the Grand River and were sampled at baseflow.
The Nzoia, Nyando, Sondu Rivers drain from Kenya into the east side of Lake Victoria. Kenyan drainage comprises 40% of the inflows to Lake Victoria (COWI 2002) and is therefore a significant source of the increasing nutrient concentrations in the lake (Juma et al. 2014). Eight sites on the Nzoia River, 11 sites on the Nyando River, and five sites in the Sondu River were sampled from January to April 2015. Sampling sites were selected based on access to the river and upstream land use. Climate in western Kenya is tropical rainforest and tropical monsoon (Köppen-Geiger classifications Af and Am).
The UK study sites compare nitrogen sources from peri-urban and rural river floodplains. Climate is maritime (Köppen-Geiger classification Cfb). Site 1 focuses on a peri-urban section of the River Thames in the vicinity of the city of Oxford in the southern UK. The mean annual flow of the Thames upstream of the study area is 18.48 m 3 /s (Marsh and Hannaford, 2008). The baseflow index for the river at this location is 0.67, reflecting the influence of influent groundwater, sourced from the limestone aquifers located in the headwaters, and the extensive floodplain gravel aquifers. During the summer a significant component of flow is supported by effluent from Wastewater Treatment Works (WwTW) (Bowes et al., 2010). Five sites upstream and downstream of a WwTW were selected along the Thames and sampled in April and September 2016 for NO3isotopes at steady-state flow. Site 2 is on the River Lambourn in Berkshire. Chalk streams such as this are widespread across southern England (Allen et al., 2010). They are characterised by a high baseflow index (>0.9) and a shallow hyporheic zone. The primary source of nitrogen therefore comes from NO3in groundwater due to fertilizer use. Samples where collected at steady-state flow.

Methods:
Canadian samples for NO3isotopes were collected in HDPE bottles and filtered in the field to 0.45µm. Samples were kept cold and dark until returned to the lab where they were frozen until analysed. Samples for H2O isotopes were collected in HDPE bottles without headspace. Canadian analyses were performed at the Environmental Isotope Laboratory at the University of Waterloo. NO3isotope samples were analysed via the chemical denitrifier method where NO3is reduced to N2O with cadmium and sodium azide (McIlvin and Altabet 2005). The resultant N2O gas was analysed on an IsoPrime continuous flow isotope ratio mass spectrometer (now Elementar, Cheadle Hulme, UK) with a precision of ±0.3‰ for δ 15 N-NO3and ±0.5‰ for δ 18 O-NO3 -. Water isotopes were measured on a a Los Gatos (Los Gatos Research, San Jose, USA) water isotope analyser with a precision of ±0.2‰ for δ 18 O-H2O.
Kenyan samples were filtered to 0.45μm and s m and stored below 4°C in 1L HDPE bottles. Kenyan analyses were performed at the Ghent University Stable Isotope Facility (UGent-SIF). NO3isotopes were analysed by the bacterial denitrification method (Xue et al., 2009) and the resulting N2O gas analyzed with a SerCon trace gas preparation unit coupled to a SerCon 20-20 isotope ratio mass spectrometer (SerCon, Crewe, UK).
UK samples were also filtered to 0.45 μm and s m and stored below 4°C in 1L HDPE bottles. Isotope preparation and analysis for UK samples was carried out at the NERC Isotope Geosciences Laboratory (Keyworth, UK). NO3was separated on anion resins and prepared as AgNO3 using the method of Silva et al. (2000) and δ 15 N analysed by combustion in a Flash EA coupled to a Delta Plus XL mass spectrometer (ThermoFinnigan, Bremen, Germany) with precision (1 SD) typically <0.8‰. δ 18 O was analysed by thermal conversion to CO gas at 1400°C in a TC-EA online to a Delta Plus XL mass spectrometer with precision (1 SD) typically <1.2‰.

Results and Discussion:
On the traditional biplot, our data from Canada, Kenya, and the United Kingdom fall in a wide swath ( Figure 1A). Data from each country has a wider range of δ 15 N-NO3values than δ 18 O-NO3values. Additionally, data from each country has a positive relationship between δ 18 O-NO3and δ 15 N-NO3 -(2tailed parametric p<0.006 for each country). But this relationship also contains seasonal changes in ambient δ 18 O-H2O values, temperature, and nitrogen sources and processes that confound direct This means that without additional independent information, there are several possible explanations for the data that are more complex than simply assigning a source of NO3based on the δ 15 N values or assigning a single process based on a simplistic pattern in the δ 18 O-vs 3and δ 15 N-NO3values. For example, varying contributions of the δ 18 O-H2O values, two or more sources of nitrogen, uptake and release of varying amounts of ammonium and NO3 -, and denitrification in varying combinations may have produced the observed patterns in our data. It is critical to avoid wrongly invoking denitrification as the primary explanation for individual points on the traditional biplot as this risks suggesting nitrogen removal from the ecosystem when other explanations for the data need to be considered.
Certainly, any interpretation that our data show clear evidence of denitrification or a mixture of NO3sources because many data points fall outside of arbitrary boxes with the traditional δ 18 O axis (Fig.  1A) cannot be supported once the range of potential δ 18 O-NO3values has been considered (Fig. 1B). Moreover, almost all measured δ 18 O-NO3values fall within the range of expected δ 18 O-NO3values based on nitrification with variable amount of H2O exchange (Fig. 1B). Thus, the theoretical range of δ 18 O-NO3values should be generated for each field site rather than a single catch-all approach. Globally, δ 18 O-H2O values of surface water vary widely along a meteoric water line, but they can be predicted by latitude and databases such as waterisotopes.org though direct measurement is much simpler than NO3isotopes. Additionally, to make δ 18 O-NO3data comparable between seasons and sites, δ 18 O-NO3data should be displayed vs the δ 18 O-H2O value from the same sample (i.e., same location and time) rather than vs SMOW. This is akin to the way δ 18 O-PO4 3values are plotted relative to their temperature-specific equilibrium point with δ 18 O-H2O (e.g., Davies et al. 2014, Paytan et al. 2002 in order to remove the influence of difference δ 18 O-H2O values ( Figure 1B). Here the differences in δ 18 O-NO3values between countries is much reduced and most δ 18 O-NO3values are near the upper-end of the δ 18 O-NO3values predicted from microbial transformation of nitrogen. There is a positive relationship between δ 18 O-NO3and δ 15 N-NO3in the Kenya and UK data (p<10 -4 ) but not Canada (p>0.4).
Some variability due to watershed size and seasonality can also be considered with this approach. First, as watershed size increases above a river sampling point the average duration the nitrogen spends in the watershed increases and thus the likelihood that the sampled NO3had been assimilated and released multiple times approaches 100%. Second, initial δ 18 O-NO3values entirely depend on the ambient δ 18 O-H2O and δ 18 O-O2 at the time of nitrification and not the δ 18 O value of the NO3added to the watershed at some point upstream if the nitrogen has been cycled at least once. Thus changes in δ 18 O-H2O between seasons or throughout watersheds are accounted for by reporting δ 18 O-NO3relative to the H2O. The implication here is that identifying the source of the NO3cannot be done with δ 18 O-NO3values. Increases in δ 15 N-and δ 18 O-NO3values, which are often interpreted as evidence of denitrification with closed-system assumptions (e.g., Böttcher et al. 1990), cannot be uniquely separated from multiple processes that recycle nitrogen in surface waters. Necessarily, this requires us to move beyond looking only for denitrification in our δ 15 N-and δ 18 O-NO3data and towards how multiple processes and sources interact to produce the values measured in surface waters. Likely, this will ultimately require development of process-based NO3isotope models for surface waters and will be informed by measurements of other nitrogen species, transformation processes and associated isotope enrichment factors (e.g., Venkiteswaran et al. 2018).
Only once the appropriate range of initial δ 18 O-NO3values has been determined, can processes such nitrification, denitrification, and NO3assimilation be considered. Here, the δ 15 N-and δ 18 O-NO3values in the environment will be pulled in multiple directions at the same time. The magnitude of change depends on multiple factors that are difficult or impossible to statically display in a biplot: (1) mineralization of organic nitrogen and subsequent nitrification may decrease δ 15 N-and δ 18 O-NO3values depending on if there is a difference between the δ 15 N value of organic nitrogen and NO3and the δ 18 O contributions of O2 and H2O; (2) ammonia and NO3uptake and release by riverine periphyton and macrophytes may have differing impacts since isotope fractionation during ammonia uptake is non-linearly dependant on concentration (Fogel and Cifuentes 1993;Hoch et al. 1992) and denitrification in riparian zones and anoxic river and lake sediments may increase δ 15 N-and δ 18 O-NO3values if there is residual NO3to measure. In all cases, changes in the δ 15 N-and δ 18 O-NO3values are more complex than a single arrow for denitrification suggests (Kendall 1998). A recent review has summarised the modelling approaches and isotope fractionation factors necessary to interpret measured δ 15 N-and δ 18 O-NO3values in soils (Denk et al. 2017). With this process-based understanding it is clear that a single vector or slope on a biplot for denitrification is inappropriate for surface waters.

Summary and Conclusions:
In order to move beyond the simple source apportionment assumptions commonly made in NO3isotope biplots and to explicitly acknowledge that there are a variety of processes that alter the δ 15 Nand δ 18 O-NO3values in situ we therefore recommend:    (Venkiteswaran et al. 2015). Thus the δ 18 O value of newly producted NO3in these rivers may cycle through these ranges on a diel basis. Here, data are more clearly expressed relative to the appropriate environmental conditions that recognize that nitrogen is biologically cycled and will be largely imprinted with the ambient δ 18 O-H2O value with a minor contribution from the variable δ 18 O-O2 value. A parsimonious interpretation here is that many data from Kenya and the UK exhibit the range of known contributions of the δ 18 O-H2O values, i.e., from two-thirds to one. Most Canadian and some Kenyan and UK data approach the theoretical maximum δ 18 O-NO3before a requirement of denitrification must be considered.