Introduction

Assessing gaseous nitrogen (N) losses via denitrification from soils and watersheds remains a major challenge in our understanding of ecosystem N dynamics (Groffman 2012; Kulkarni et al. 2008; Yang et al. 2011). An improved understanding of the spatial and temporal controls on denitrification rates, and their importance in ecosystem N budgets, is critical for addressing problems linked to the cascading impacts of reactive nitrogen in our environment (Galloway et al. 2003). One important method for assessing spatial and temporal patterns in denitrification exploits the variation in the natural abundance isotopic composition (δ15N and δ18O) of nitrate (NO3 −) and/or δ15N of bulk soils to constrain denitrification losses using δ15N mass balance (Bai and Houlton 2009; Houlton et al. 2006; Houlton and Bai 2009). Recent studies have expanded on this approach by combining δ15N of NO3 − with ∆17O, a sensitive tracer of atmospheric NO3 − (Michalski et al. 2004), to derive gross NO3 − production (Riha et al. 2014) and denitrification at the watershed scale (Fang et al. 2015). Natural abundance isotopes clearly represent a powerful and minimally invasive tool to assess patterns and controls on denitrification. However, several uncertainties remain with regards to the interpretation of natural abundance NO3 − isotope measurements in the context of denitrification.

The degree to which NO3 − stable isotopes can record patterns of denitrification in surface soil horizons (i.e., to a depth of tens of cm), as opposed to deeper subsurface or groundwater environments, has received little attention. It has long been known that significant hotspots of denitrification can occur in anaerobic microsites in otherwise aerobic soils (Parkin et al. 1987). Houlton et al. (2006) found strong evidence for denitrification in wet surface soils from humid tropical forests using natural abundance N isotopes. In contrast, Fang et al. (2015) used surface soil NO3 − isotope compositions as inputs to models of watershed-scale denitrification, assuming that nitrification and denitrification were spatially segregated between surface and subsurface soils, respectively. It remains unclear how surface soil NO3 − samples might best be collected and interpreted to inform ecosystem-scale isotope models and our broader understanding of denitrification.

Different types of samples, such as soil salt extractions or soil water samples, could differ substantially in their capacity to record NO3 − isotope effects from denitrification due to differences in water residence time and spatial distribution in the soil—an important scale dependence that we highlight in this manuscript. Water has a continuum of mobility in soil due to physical and chemical interactions with soil constituents, which can be quantified according to water potential. Soil extractions capture most of the inorganic nitrogen (including N interacting electrostatically with soil exchange sites) in a sampled soil volume, whereas water collected in vadose zone lysimeters under minimal applied tension may reflect N in comparatively mobile water pools that have a shorter residence time in the soil matrix. Water pools associated with soil mineral and organic surfaces (i.e., less mobile pools) often differ greatly in H2O isotope composition relative to highly mobile water in soil macropores, reflecting differences in water pool transit time and thus their capacity to record terrestrial biophysical processes (Evaristo et al. 2015; Good et al. 2015). We hypothesize that these water pools might also differ in their capacity to record N biogeochemical processing via stable isotopes. Both soil extractions and lysimeter water collected under various tensions have been used in ecosystem-scale studies involving NO3 − isotopes, but they have seldom been directly compared. Perhaps even more importantly, differences in NO3 − stable isotopes among samples have not typically been compared with canonical indices of denitrification, such as potential denitrification enzyme activity.

Comparisons of NO3 − isotope composition with measured potential denitrification activities could help address possible ambiguities in isotope interpretation. The use of NO3 − isotopes to infer the occurrence of denitrification relies on kinetic fractionation from denitrifying bacteria, which increases the δ15N and δ18O of residual NO3 − (Mariotti et al. 1981). Linear relationships between δ18O and δ15N of NO3 − with slopes between 0.5 and 1 are assumed to reflect kinetic fractionation by denitrifiers, a relationship that appears to hold across soils, groundwater, and aquatic ecosystems (Cohen et al. 2012; Granger et al. 2008; Houlton et al. 2006; Lehmann et al. 2003; Sigman et al. 2005). This assumption is plausible given that assimilatory N fractionation by plants and microbes, another explanation for NO3 − isotope enrichment, is typically minor in N-limited ecosystems (Evans 2001; Granger et al. 2010). However, co-occurring fractionation from other N-cycling processes, such as nitrification (Mariotti et al. 1981), or N inputs with differing isotope compositions, could potentially obscure isotope enrichment from denitrification.

In surface soils with heterogeneous O2 availability, nitrification (an aerobic process) and denitrification (an anaerobic process) can potentially co-occur. In open systems, nitrifying bacteria tend to decrease the δ15N of soil NO3 − relative to NH4 + by 14–38 ‰ (Casciotti et al. 2003; Mariotti et al. 1981). Open-system fractionation by denitrifying bacteria spans a similar range, increasing δ15N of the residual NO3 − pool by between 5 and 29 ‰ (Granger et al. 2008; Mariotti et al. 1981). In cases where NH4 + or NO3 − are fully consumed in soil microsites (i.e., closed systems), it is also possible that fractionation is partially or not expressed during either nitrification or denitrification (Kendall et al. 2007). However, even in cases where fractionation from nitrification and denitrification are both expressed at the microsite (i.e., µm) scale, it is possible that their effects could be mutually obscured over larger spatial scales (i.e., cm) characteristic of soil samples due to their opposite effects on δ15N of NO3 −. This situation potentially creates a challenge to interpretation by obscuring the original isotope composition of NO3 − inputs, a necessary parameter for quantitative isotope models (e.g., Fang et al. 2015), and diluting the pool of δ15N-enriched NO3 − that is typically used to indicate the presence of denitrification (Billy et al. 2010; Houlton et al. 2006; Wexler et al. 2014). Comparing traditional indices of denitrification and related biogeochemical parameters with NO3 − isotopes could help to decipher if isotope overprinting from nitrification and denitrification can be important in surface soil environments.

Combining measurements of δ15N and δ18O of NO3 − with δ15N of N sources and δ18O of soil water could also assist in identifying the occurrence of denitrification in surface soils. The δ15N of nitrified NO3 − depends on both the δ15N of NH4 + sources and the extent of fractionation during nitrification. Ammonium typically reflects the δ15N of the plant litter, soil organic matter (SOM), or fertilizer from which it is derived, with minimal (~1 ‰) fractionation (Kendall et al. 2007). Atmospheric deposition is another potentially important NH4 + source. A representative range of δ15N in these NH4 + sources that is specifically applicable to the present study is indicated by the vertical shaded rectangle in Fig. 1a. Nitrification of NH4 + fractionates δ15N substantially in open systems (Casciotti et al. 2003; Mariotti et al. 1981), with potential values indicated by the “nitrification” vertical shaded rectangle in Fig. 1a. Nitrification also imparts a characteristic δ18O composition to NO3 −. Oxygen in nitrified NO3 − is dominantly (i.e., ≥2/3) derived from soil water, with a variable contribution from atmospheric O2 (23.5 ‰) that depends on the degree of in situ O exchange with water (Kool et al. 2011; Mayer et al. 2001). This range is indicated by the nitrification box in Fig. 1a. Subsequent denitrification of this NO3 − pool would presumably be reflected by ~1:1 increases in the δ15N and δ18O of nitrified NO3 − (indicated by the denitrification arrow in Fig. 1a) which could ultimately return δ15N of NO3 − to values similar to or greater than the ecosystem N sources (Fig. 1a). However, δ18O of NO3 − would then remain substantially elevated relative to δ18O of soil H2O, which is inconsistent with NO3 − derived from nitrification. Mixing of soil NO3 − with atmospheric inputs is another important consideration. Atmospheric NO3 − has very high δ18O (i.e., 60–80 ‰) as a consequence of atmospheric oxidative processes (Kendall et al. 2007), which could also increase the δ18O of soil NO3 − relative to the products of nitrification, as indicated by the vertical mixing arrow in Fig. 1a.

Fig. 1
figure 1

a Conceptual dual isotope plot (δ18O and δ15N) showing the measured variation in δ15N of SOM, litter, and fertilizer as a vertical rectangle and the measured variation in precipitation δ18O as a horizontal rectangle. The predicted isotope composition of nitrified NO3 − with complete expression of kinetic isotope effects on δ15N is bounded by a black rectangle. This incorporates potential N fractionation during nitrification of an NH4 + source with δ15N within the observed range of SOM, litter, and fertilizer, and NO3 − δ18O ranging from complete incorporation of soil water to a 1/3 contribution from atmospheric O2 with δ18O = 23.5 ‰ (Kool et al. 2011; Mayer et al. 2001). The dotted line represents an illustrative trend of kinetic denitrification isotope effects with a δ18O/δ15N slope of 1, beginning at an arbitrary point in δ18O/δ15N space. The vertical arrow indicates potential impacts of mixing with atmospheric NO3 − enriched in δ18O. b Measurements of NO3 − isotopes in soil KCl extractions (black symbols) and lysimeter water samples (open symbols). Triangles and circles represent the lawn and meadow sites, respectively. The dashed line represents a linear regression of δ18O and δ15N values in soil KCl extractions from both sites (slope = 1.04 ± 0.12, R 2 = 0.64, p < 0.0001)

Full size image

To assess relationships between NO3 − stable isotope composition and microsite-scale denitrification in surface soils, we designed a study involving two nearby sites (a riparian meadow and a managed urban lawn) in a semi-arid montane ecosystem where denitrification likely represents an important fate of N during spring snowmelt. In temperate ecosystems with a seasonal snowpack, snowmelt often represents a period of maximum soil moisture and potential N loss, when soil biogeochemical processes are especially important in attenuating N inputs to streams (Brooks and Williams 1999; Zak et al. 1990). Our sites provided a natural gradient in denitrification potential, allowing us to assess their relationship with NO3 − isotopes while maintaining similar climate and soil physical/chemical characteristics. Importantly, these sites also had similar surface soil (0–15 cm) NH4 + pools, similar plant litter δ15N, and relatively similar bulk soil δ15N (within 1.5 ‰), implying that the initial δ15N of NH4 + and NO3 − pools produced by mineralization and nitrification were also likely to be similar between sites.

Methods

We sampled soils from two sites in the Red Butte Creek watershed in Salt Lake City, Utah, USA: a natural riparian meadow with herbaceous vegetation (Dawson and Ehleringer 1991) and a managed lawn on the University of Utah campus. Both sites received >4 kg N ha−1 year−1 in atmospheric deposition (Hall et al., J Geophys Res Biogeosci, in revision). The lawn received ~100 kg ha−1 year−1 of fertilizer N as a urea blend (32–5–7 ratio of N, P, and K) applied during late spring, and ammonium sulfate (20–0–0) applied during October. Fertilizer was obtained from the Intermountain Farmers Association (Salt Lake City, UT, USA). Soils were sampled following spring snowmelt (February and March 2014 at the lawn and meadow, respectively), approximately 5 months after the last fertilizer application to the lawn. Snowmelt typically represents the period of maximum soil moisture in semi-arid montane ecosystems.

The meadow soil was a loamy Mollisol with a predominant rooting depth above 35 cm, and the lawn soil was a sandy loam Inceptisol derived from local alluvium added on top of rocks. Bulk density and pH were similar between sites (Table 1). We sampled the meadow at depths of 0–5, 5–15, and 15–35 cm with a 6-cm diameter auger (n = 9 for each depth); the lawn could only be sampled to 15 cm due to an impenetrable rock layer (n = 12 for each depth). Tension lysimeters (Prenart Super Quartz, Frederiksberg, Denmark) were installed at 15 and 35 cm in the lawn and meadow, respectively, at the bottom of the main rooting zone at each site. Lysimeters were installed approximately 5 months prior to sampling for this study. Lysimeter water was sampled by applying a vacuum of −28 kPa relative to ambient pressure (86 kPa). This sampling pressure was chosen because its absolute value is similar to but less than field capacity, typically defined as −30 kPa for loamy soils (Schaetzl and Anderson 2005). Field capacity is an estimate of the residual soil water remaining after macropores have drained. Thus, our lysimeters captured water at potentials above field capacity, or a comparatively “mobile” water pool. Others have used tension lysimeters to sample soil water under much more negative water potential (−70 kPa) than we used here, but they still judged these samples to reflect water capable of significant advection (Castellano et al. 2013). We stress that the concepts of “mobile” and “immobile” water do not represent a dichotomy but rather an operationally defined continuum. Because our lysimeters were sampled under tension, they contain some fraction of soil-associated water differing in composition from water that instantaneously drains through soils during melt or precipitation events (Landon et al. 1999). Our operational separation between more and less mobile water pools, respectively sampled in lysimeters and soil extractions, is thus conservative. Water samples were collected throughout the period of snowmelt (total n = 186; Jan–April 2014).

Table 1 Soil biogeochemical characteristics by depth for the lawn and meadow sites. Standard errors are shown in parentheses
Full size table

Oxygen (O2) sensors (Apogee Instruments, Logan, UT, USA) were installed at depths of 10 cm in 4.8 cm diameter polyvinylchloride pipes with sealed caps (equilibration chambers), where the bottom lip of the chamber was pressed into undisturbed soil (Hall et al. 2013). Soil O2 content was recorded on a datalogger at 15-min intervals. Here, we report O2 during the period of NO3 − sample collection. All of the above measurements were made in the context of a snow removal experiment at each site (Hall et al. 2016a), but given the absence of treatment effects on any biogeochemical response variable, we do not report treatment identities for ease of interpretation. Precipitation was collected on an event basis during the study period for analysis of NO3 − and NH4 + concentrations and isotope composition. Lysimeter and precipitation samples were filtered through pre-combusted and rinsed Whatman GF/F filters and frozen until analysis by ion chromatography.

We measured potential denitrification rates in shaken slurries of soil samples within 24 h of sampling using the acetylene inhibition method (Groffman et al. 1999), using an incubation temperature of 4 °C to approximate field conditions and 25 °C to facilitate comparisons with other studies. Potential denitrification rates are an appropriate metric to compare with the isotope composition of NO3 − pools because they are tightly linked to the abundance of bacterial denitrification functional genes, which integrate recent process activity in the soil environment (Petersen et al. 2012). Soil subsamples were extracted in 2 M potassium chloride (KCl) solutions for analysis of NH4 + and NO3 − concentrations and isotope composition. We measured δ15N and δ18O of NO3 − in soil extractions (n = 50), lysimeter water samples (n = 31), and precipitation samples (n = 8) using Pseudomonas aureofaciens and the denitrifier method (Bell and Sickman 2014; Casciotti et al. 2002) to generate N2O for analysis by isotope ratio mass spectrometry at the Stable Isotope Ratio Facility for Environmental Research at the University of Utah. When analyzing the soil extractions using the denitrifier method, we corrected for the presence of trace NO3 − in KCl and any physiological effects on the denitrifying bacteria by analyzing reference materials prepared in the same matrix (Bell and Sickman 2014). Samples were normalized to δ notation, relative to atmospheric N2 for N and VSMOW for O, using a combination of the international NO3 − reference materials USGS 34, 35, and 32. Precision and accuracy were 0.6 and 0.7 ‰ (respectively) for δ15N, and 0.6 and 0.3 ‰ (respectively) for δ18O, assessed by repeated analysis of IAEANO3 (for δ15N) and USGS 32 (for δ18O) treated as unknowns. Sample reproducibility was 0.3 ‰, based on duplicate analyses of 1/8 of the samples. To characterize the δ15N of fertilizer inputs to the lawn, three subsamples were analyzed from each of two production batches of each fertilizer type (ammonium sulfate and urea). Bulk soil δ15N from each site and sampled depth increment as well as surface litter (n = 9 plots per site) were analyzed on samples dried at 60 °C. Soil and fertilizer δ15N were measured via combustion on an elemental analyzer coupled to an isotope ratio mass spectrometer (Thermo MAT 253, Waltham, MA, USA). A subset of lysimeter water samples from the meadow site (n = 17) was analyzed for δ18O of H2O by isotope ratio infrared spectroscopy (Picarro L-2130i, Santa Clara, CA, USA).

A subset of precipitation samples (n = 15) was analyzed for δ15N of NH4 + using a modified NH3 diffusion method (Holmes et al. 1998). Briefly, 30 ml of the sample were added to a 60-ml HDPE bottle with 1.5 g sodium chloride. A glass-fiber filter acidified with 30 µl of 4 M phosphoric acid was pressed between Teflon tape and added to the bottle. Magnesium oxide (90 mg) was added to volatilize NH3, and the bottle was immediately capped and incubated for 7 days on an orbital shaker/incubator at 40 °C to allow NH3 to be completely trapped as NH4 + on the acidified filter. Filters were analyzed for δ15N (precision < 0.2 ‰) by combustion as described above. Analysis of ammonium sulfate solutions with known δ15N values verified a lack of fractionation during diffusion.

Soil subsamples were also extracted in the field immediately after sampling with 0.5 M hydrochloric acid for analysis of reduced and oxidized iron [Fe(II) and Fe(III), respectively] as a redox indicator (Hall et al. 2013). The presence of Fe(II) implies the presence of anaerobic microsites where dissimilatory Fe reduction exceeds Fe(II) oxidation by O2. The ratio of Fe(II)/FeHCl reflects the proportion of HCl-extractable Fe that has been reduced, facilitating comparisons among sites that may differ in total HCl-extractable Fe content (Hall and Silver 2015). Thermodynamic principles imply that NO3 − should have been consumed via denitrification in microsites where Fe(II) is present (Chapelle et al. 1995).

We analyzed linear relationships between δ15N and δ18O of NO3 − both within and among sites and for each separate sample type (soil extractions vs. lysimeter water) using ANCOVA. Mean differences in δ15N and δ18O composition among sites were assessed using MANOVA (i.e., ANOVA testing two response variables simultaneously). Differences in other response variables among sites were assessed with ANOVA. Relationships between δ15N of NO3 − and potential denitrification and Fe(II)/FeHCl were highly nonlinear, so we fitted trends with a two-component piecewise linear regression using the SiZer package in R (Sonderegger 2012).

Results

While both sites typically had near-atmospheric O2 concentrations in the bulk soil throughout the period of measurement, they also showed evidence of redox heterogeneity and the presence of reducing conditions in soil microsites, as reflected by Fe(II)/FeHCl ratios (Table 1). The lawn had much higher Fe(II)/FeHCl relative to the meadow (p < 0.0001), coinciding with increased potential denitrification rates. Denitrification potential was 16-fold greater in lawn soils at 4 °C (p < 0.0001) and sixfold greater at 25 °C (p < 0.001; Table 1) when considering all depth increments combined. Nitrate concentrations in soil extractions were generally similar between sites, although lysimeter NO3 − was significantly greater in the lawn (Table 1).

Bulk soil δ15N was similar between sites in 0–5 cm soil, but was slightly (1.5 ‰) and significantly (p < 0.0001) greater in lawn 5–15 cm soil (Table 1). Despite the fact that NH4 + concentrations significantly differed between sites at a given depth increment (p < 0.001), total KCl-extractable NH4 + mass to 15 cm was similar between the meadow and lawn, measuring 0.056 (0.013) and 0.044 (0.003) kg NH4 +–N ha−1, respectively. At the lawn, comparison of the KCl-extractable NH4 + pool with annual NH4 + fertilizer inputs indicated that >99.9 % of the previous year’s fertilizer application was assimilated, or otherwise transformed or exported, prior to measurement. Surface litter δ15N was also similar between the meadow (−0.5 ± 0.1 ‰) and lawn (0.1 ± 0.4 ‰). Fertilizer δ15N was highly consistent within and between production batches and was relatively similar between the two fertilizer types applied here. Two separate batches of ammonium sulfate δ15N measured −0.9 ± 0.2 and −1.0 ± 0.3 ‰, and the urea blend δ15N was −1.7 ± 0.1 and −1.6 ± 0.3 ‰. Values of δ15N of NH4 + in bulk atmospheric deposition measured −1.2 ± 1.5 ‰, similar to fertilizer.

Trends in δ18O and δ15N of NO3 − in soil extractions from both sites considered together were consistent with linear isotopic effects from the denitrification of NO3 − sources with similar initial isotope values, reflected by a slope of 1.04 ± 0.12 (R 2 = 0.64, p < 0.0001; Fig. 1b). The slope of this relationship did not differ significantly between sites. However, the relationship between δ18O and δ15N was insignificant when assessing samples from the lawn alone (slope = 0.62, R 2 = 0.10, p = 0.21), and was much stronger in the meadow (slope = 0.72 ± 0.13, p < 0.0001, R 2 = 0.55). Nevertheless, the lawn extractions had significantly greater (more enriched) NO3 − δ18O and δ15N values than the meadow (p < 0.0001; Fig. 1b; Table 2). Nitrate δ15N in soil extractions also had a weakly significant site by depth interaction (p = 0.02), where values did not differ by depth in the lawn but were significantly greater in meadow 0–5 cm samples than the deeper samples.

Table 2 Nitrate isotope compositions in soil KCl extractions and lysimeter samples from the lawn and meadow
Full size table

In contrast to the soil extractions, δ18O and δ15N of lysimeter NO3 − were not significantly correlated, and lysimeter NO3 − isotope composition did not differ between sites (Fig. 1b; Table 2). Furthermore, lysimeter samples significantly differed in NO3 − isotope composition when compared with soil extractions from the same site (MANOVA, p < 0.0001; Fig. 1b). Values of δ18O in precipitation NO3 − (81.9 ± 4.0 ‰) were substantially greater than any of the soil extractions or soil lysimeter samples, while δ15N of precipitation NO3 − (2.3 ± 0.7 ‰; Table 2) fell within the bounds of soil extractions and lysimeter samples. Neither the soil extractions nor lysimeter samples displayed significant relationships between NO3 − isotope composition and NO3 − concentrations (data not shown).

Nitrate δ15N showed distinct trends with potential denitrification and Fe(II)/FeHCl at both sites: significant linear trends at the meadow (p < 0.001 and 0.03, respectively), but no relationship at the lawn. Combined, these data suggested significant threshold relationships between δ15N, potential denitrification, and Fe(II)/FeHCl as determined by piecewise linear regressions (Fig. 2). Nitrate δ15N initially increased sharply with potential denitrification (slope = 0.75, between 0.10 and 0.90 with 95 % confidence) and Fe(II)/FeHCl (slope = 170, between 60 and 550 with 95 % confidence), but subsequently plateaued to slopes indistinguishable from zero for both response variables (Fig. 2).

Fig. 2
figure 2

a Relationship between δ15N of NO3 − in soil extractions and potential denitrification rate; the gray line represents a piecewise linear regression as described in the text. b Relationships between δ15N of NO3 − in soil extractions and the Fe(II)/FeHCl ratio (a redox indicator, described in the text and in Hall and Silver 2015); the gray line represents a piecewise linear regression

Full size image

Discussion

We found evidence that denitrification was likely occurring during snowmelt in surface soil horizons at both sites, as reflected by substantial rates of potential denitrification activity combined with the presence of anaerobic microsites as reflected by Fe(II)/FeHCl. Although high potential denitrification rates indicate the presence of denitrifying organisms, they do not definitively indicate when denitrification may have occurred. In these ecosystems, increased soil moisture during snowmelt likely generated anaerobic microsites that promoted denitrification, given that very low soil moisture (<10 % by volume) prevailed for several months prior at each site until snowmelt began (Hall et al. 2016a). The importance of denitrification in cold soils during snowmelt has been well established in other studies (Brooks et al. 2011), and our measured potential denitrification activities at 25 °C were similar to rates from wetland and riparian soils in other studies (Roach and Grimm 2011). Yet, different NO3 − samples (soil extractions vs. lysimeter water) and sites (the lawn and meadow) varied dramatically in their relationships between δ15N and δ18O and proxies for denitrification.

Trends in NO3 − isotopes in soil extractions and proxies for denitrification

Plotting δ18O versus δ15N of NO3 − in soil extractions from both sites yielded a slope similar to 1, consistent with the upper bound of slopes invoked as evidence for denitrification in previous field studies in terrestrial, groundwater, and marine environments (Cohen et al. 2012; Houlton et al. 2006; Lehmann et al. 2003; Sigman et al. 2005), and similar to laboratory experiments with denitrifying bacteria (Granger et al. 2008). However, δ15N of NO3 − remained lower than bulk soil N in most samples, despite the fact that denitrification had most likely occurred. These trends are consistent with a large initial isotope fractionation from nitrification (Casciotti et al. 2003) which was subsequently impacted by denitrification to a variable extent among samples. A large fractionation from nitrification is consistent with the observed presence of excess NH4 + at both sites relative to NO3 − (Table 1), and comparatively slow enzymatic catalysis in cold soils following snowmelt. A large isotopic effect of nitrification was also recently observed in New Zealand pasture soils during winter (Wells et al. 2015).

Our interpretations of δ15N of NO3 − are enhanced by considering potential variation in δ15N of the NH4 + from which NO3 − is derived. We argue that there were not likely to be systematic differences in δ15N of NH4 + between sites, according to the following reasoning. The major potential sources of NH4 + at these sites during the period of sampling include mineralization of litter and SOM, residual fertilizer at the lawn, and atmospheric deposition. Most importantly for our study, these potential NH4 + sources had relatively similar δ15N compositions in comparison with the observed variation in δ15N of NO3 −, and tended to be similar between sites. First, mineralized NH4 + has similar δ15N (±1 ‰) relative to the organic matter (e.g., litter or SOM) from which it was derived (Kendall et al. 2007), and differences in litter and SOM δ15N pools between the two sites were small (<0.5 and 1.5 ‰, respectively). Multiple litter and SOM pools mineralize NH4 + at different rates, so it is likely that the measured differences between litter and bulk soil δ15N (−0.5 to 5.8 ‰) approximate the range of variation in δ15N of mineralized NH4 + at these sites. Mean fertilizer (−0.9 to 0.1 ‰) and atmospheric deposition (2.3 ‰) δ15N fell close to or within this range. We suggest that mineralization of litter and SOM (as opposed to fertilizer) were especially important sources of NH4 + in both the meadow and the lawn during the period of sampling. This interpretation was supported by (1) the fact that total extractable NH4 + stocks were equivalent between lawn and meadow despite fertilization; (2) the very small extractable pool of NH4 + relative to annual fertilizer inputs, indicative of rapid cycling of mineral N where >99.9 % of fertilizer N inputs were assimilated or lost via gaseous or leaching pathways prior to sampling; (3) the fact that 6 months had elapsed since fertilizer application, providing ample opportunity for assimilation or other transformations. Even if residual fertilizer did represent a significant component of soil NH4 +, it is unlikely to lead to substantial differences in δ15N of NH4 + between sites, given that fertilizer δ15N was very similar to litter δ15N at both sites.

Secondly, it is plausible to assume that initial δ15N values of nitrified NO3 − may also have been similar between sites and depths. This assumption is reasonable given overall similarities in NH4 + concentrations, bulk soil N concentration and δ15N, microbial biomass N, pH, bulk density, and climate between the lawn and meadow (Hall et al. 2016a). Under this assumption, the lawn showed greater enrichment of δ15N and δ18O subsequent to nitrification, implying a greater fraction of NO3 − loss to denitrification relative to the meadow, in agreement with the denitrification potential assays. Trends in δ15N of NO3 − with depth in the lawn also corresponded with the denitrification potential assays: δ15N of NO3 − was greater in 0–5 cm soils, which had greater potential denitrification rates. Previous studies have similarly demonstrated greater denitrification potential in surface than subsurface soil horizons (Groffman et al. 2002) and have reported high denitrification rates in lawns (Raciti et al. 2011). Both factors reflect the importance of plant C and N inputs (and potentially fertilizer additions) as controls on denitrification in surface soils, as opposed to O2 diffusion limitation in the subsurface. However, if we had assessed relationships between δ18O and δ15N of NO3 − at the lawn site alone without the context of the meadow samples with much lower denitrification rates, this conclusion would have been obscured. The meadow samples revealed the likely importance of fractionation during nitrification, and without accounting for this fractionation we would have had little isotopic evidence for denitrification at the lawn—even though denitrification potential assays and Fe(II) measurements strongly suggested it had occurred. This highlights a critical limitation on the use of NO3 − isotopes to infer site-specific or watershed-scale patterns of denitrification, as recently implemented elsewhere.

Contrasting NO3 − isotope compositions in soil extractions and lysimeter water

In contrast to the linear trend in the soil extractions, δ18O and δ15N values in soil lysimeter NO3 − exhibited significant scatter, likely reflecting multiple fractionating processes and mixing with atmospheric NO3 − that was enriched in 18O. The NO3 − isotope disparities between lysimeters and soil extractions may also reflect the redox heterogeneity of terrestrial soils (Sexstone et al. 1985). Measurements of Fe(II)/FeHCl in soil extractions suggested the presence of reducing conditions in soil microsites, whereas the soil O2 sensors indicated highly aerobic conditions at the scale of the bulk soil atmosphere in both sites (Table 1). Differences in dual isotope (δ18O/δ15N) slopes and δ15N and δ18O values between the soil extractions and the lysimeters support our hypothesis that soil extractions and lysimeter samples differ in their capacity to record ecosystem-scale N biogeochemical cycling via stable isotopes. A plausible explanation for this pattern is that soil extractions sampled the anaerobic microsites inside soil aggregates where denitrification can occur (Parkin et al. 1987), whereas lysimeters largely reflected the transport of N that was nitrified in the aerobic macropore environment or in surface litter.

This distinction may be critical in interpreting patterns of denitrification at the watershed scale, where the NO3 − isotopic composition of stream water is often assumed to integrate gross nitrification or denitrification across entire watersheds (Fang et al. 2015; Riha et al. 2014). Our data suggest that this assumption does not necessarily hold, given that relatively mobile lysimeter water and soil extractions from the same sites differed substantially in their NO3 − isotope compositions. Rather, as supported by recent syntheses of H2O isotope dynamics, these samples may reflect two “water worlds:” poorly mobile water accessible to plants and microbes, and comparatively mobile water that dominates groundwater recharge and stream discharge (Evaristo et al. 2015; Good et al. 2015). Thus, stream or groundwater NO3 − isotope composition may not necessarily reflect denitrification in surface soil microsites, which are a dominant zone of potential denitrification activity at the ecosystem scale (Groffman et al. 2002). As a consequence, we suspect that model estimates based on stream NO3 − isotopes might systematically underestimate watershed-scale denitrification in some ecosystems. However, even in studies aimed at estimating riparian denitrification alone, our data imply that characterizing the initial isotope composition of NO3 − in relatively mobile soil water is a critical, if nontrivial, step in interpreting stream NO3 − isotopes. Our data show that total soil NO3 − and NO3 − in relatively mobile water can differ substantially, and misspecification of the initial isotope composition of mobile soil NO3 − could variably lead to under- or overestimation of denitrification.

Implications for interpreting δ15N in models and environmental samples

Importantly, dual isotope trends in soil KCl-extractable NO3 − from both sites combined were consistent with denitrification isotope effects (Fig. 1) and were related to potential denitrification rates (Fig. 2) despite the fact that δ15N values of NO3 − were typically less than bulk soil δ15N. This finding has interesting implications for models that estimate denitrification at the ecosystem scale using isotopic measurements of NO3 − (Billy et al. 2010; Fang et al. 2015). Our data suggest that substantial denitrification might occur even in situations where δ15N values of NO3 − do not display substantial enrichment relative to N inputs or other ecosystem pools, due to the potentially large initial N isotope fractionation associated with nitrification. Thus, isotope models which assume that δ15N of surface soil NO3 − predominantly represents the product of nitrification (Fang et al. 2015) would tend to underestimate the relative importance of denitrification in the case where large fractionation effects of nitrification were partially counteracted by denitrification. In the latter study (ibid.), correlations between δ15N and δ18O of NO3 − were not significant in surface soil KCl extractions, leading the authors to conclude that NO3 − dynamics were spatially separated in these sites, with nitrification dominating in near-surface soils and denitrification prevailing in deeper soils.

The very weak linear trends between δ15N and δ18O we observed at the lawn site alone (as opposed to considering all samples together), combined with our process-level measurements, show that denitrification is not necessarily well recorded by site-specific trends in NO3 − isotope composition in surface soils. Rather, δ15N of NO3 − showed a threshold relationship with potential denitrification rates and reducing conditions—strong trends at low rates that became a null relationship at higher rates. This may reflect closed-system kinetics where NO3 − is mostly consumed via denitrification in soil microsites, leading to an underexpression of kinetic isotope effects, as noted previously in both soils and sediments (Houlton et al. 2006; Sebilo et al. 2003). Thus, pairwise trends between δ15N and δ18O of NO3 − and their relationships with potential denitrification only became strongly significant when considering a very broad gradient in redox variability and potential denitrification activity (which spanned three orders of magnitude among samples and sites, Fig. 2). Moreover, our results are consistent with a recent global-scale analysis of soil δ15N, which found that much of the variability in bulk soil δ15N previously attributed to denitrification might instead be explained by clay content and preferential retention of microbially processed organic matter (Craine et al. 2015). Differences in clay content between our sites (with loam and sandy loam texture, respectively) could contribute to the fact that bulk soil δ15N did not reflect the large site differences in denitrification potential and δ15N of soil extracted NO3 −.

In summary, recent analytical and conceptual advances related to NO3 − isotopes provide powerful opportunities for exploring controls on N sources and biogeochemical processing over multiple spatial scales (Fang et al. 2015; Houlton et al. 2006; Kendall et al. 2007; Michalski et al. 2004; Riha et al. 2014). We showed that surface soils characterized by denitrification at the microsite scale in a heterogeneous redox environment can exhibit NO3 − isotope trends consistent with denitrification when examining samples exhibiting a broad range of potential denitrification rates. However, these patterns are highly dependent on spatial scale. They vary between mobile water and total extractable soil NO3 −, and depend on whether comparisons are conducted within or among sites. They also depend on assumptions about NH4 + isotope composition, nitrification isotope effects, and initial NO3 − isotope composition prior to any denitrification. Thus, we suggest that NO3 − isotope data should be interpreted with careful consideration of boundary conditions, sample types, and hydrologic connectivity between sampled pools before using them as parameter inputs or response variables in N cycling models.