References

Abstract. Aquifer denitrification is among the most poorly constrained fluxes in global and regional nitrogen budgets. The few direct measurements of denitrification in groundwaters provide limited information about its spatial and temporal variability, particularly at the scale of whole aquifers. Uncertainty in estimates of denitrification may also lead to underestimates of its effect on isotopic signatures of inorganic N, and thereby confound the inference of N source from these data. In this study, our objectives are to quantify the magnitude and variability of denitrification in the Upper Floridan Aquifer (UFA) and evaluate its effect on N isotopic signatures at the regional scale. Using dual noble gas tracers (Ne, Ar) to generate physical predictions of N2 gas concentrations for 112 observations from 61 UFA springs, we show that excess (i.e. denitrification-derived) N2 is highly variable in space and inversely correlated with dissolved oxygen (O2). Negative relationships between O2 and δ15NNO3 across a larger dataset of 113 springs, well-constrained isotopic fractionation coefficients, and strong 15N:18O covariation further support inferences of denitrification in this uniquely organic-matter-poor system. Despite relatively low average rates, denitrification accounted for 32 % of estimated aquifer N inputs across all sampled UFA springs. Back-calculations of source δ15NNO3 based on denitrification progression suggest that isotopically-enriched nitrate (NO3–) in many springs of the UFA reflects groundwater denitrification rather than urban- or animal-derived inputs.


Introduction
Anthropogenic increases in reactive nitrogen (N) availability have wide-ranging consequences including eutrophication of aquatic systems, acidification of soils and surface waters, loss of biodiversity, and facilitation of disease transmission (Vitousek, 1994;Galloway et al., 2003;Smith and Schindler, 2009).Denitrification, which reduces NO − 3 to N 2 gas, mitigates this enrichment by returning N to long-residence-time atmospheric pools, and is an important component of the nitrogen cycle at local, regional, and global scales (David et al., 2006;Seitzinger et al., 2006;Townsend and Davidson, 2006;Schlesinger, 2009;Sigman et al., 2009).Although denitrification was once thought to occur only via the oxidation of simple organic compounds, more recent work has demonstrated that NO − 3 reduction can involve multiple electron donors and end products (Burgin and Hamilton, 2007).Patchy and ephemeral distribution, diverse reaction modes, and challenges of direct measurement of N 2 all contribute to persistent high uncertainty in local, regional, and global estimates of denitrification (Davidson and Seitzinger, 2006;Groffman et al., 2009).
Aquifer denitrification is a potentially large component of regional and global nitrogen (N) budgets, with a recent global estimate of 44 Tg N yr −1 (16 % of land-based annual N inputs; Seitzinger et al., 2006).However, existing estimates are based on simple models and are extremely poorly constrained (range of estimates from Seitzinger et al., (2006): 0-138 Tg N yr −1 ), in large part due to the limited number and J. B. Heffernan et al.: Denitrification and inference of nitrogen sources spatio-temporal extent of available direct measurements of denitrification.Additional uncertainty arises because existing studies of groundwater N concentrations and denitrification are potentially biased by preferential study of aquifers with high N loading and high rates of denitrification (Green et al., 2008;Schlesinger, 2009).Moreover, measures of denitrification based on nitrate loss appear to provide much higher estimates than those based on direct measurement of N 2 gas accumulation (Green et al., 2008).Despite their limited numbers, directly-measured aquifer denitrification rates nonetheless span several orders of magnitude, and associated reductions in NO − 3 range from negligible to complete across aquifer systems (Green et al., 2008).The extent to which these outcomes vary in space and time within individual aquifers is poorly understood.
Estimation of denitrification from N 2 gas concentrations in groundwaters requires determination of physical parameters (recharge temperature [T rec ] and excess air [A ex ]) that influence the quantity and composition of dissolved gases (Vogel et al., 1981;Wilson and McNeill, 1997;Cey et al., 2009).Recharge temperature, rather than ambient temperature at the time of sample collection, is what determines the solubility of atmospheric gases at the time when infiltrating groundwater loses contact with the atmosphere.Depending on the seasonality of temperature, precipitation, and infiltration, as well as other factors, recharge temperatures can diverge substantially from mean annual air temperatures (Hall et al., 2005;Castro et al., 2007).Dissolution of excess air occurs when bubbles of atmospheric air are entrained beneath the saturated zone; supersaturation of gases with respect to surface conditions is enabled by hydrostatic pressure at depth.Direct simultaneous estimation of these parameters requires two tracers (typically noble gases; Feast et al., 1998;Cey et al., 2009), but most measurements of N 2 are made only in conjunction with Ar (Kana et al., 1994).Studies that estimate denitrification via direct measurement of N 2 thus typically rely on assumed constant values of either recharge temperature or excess air entrainment to estimate biologically-derived N 2 (e.g. Green et al., 2008).Since recharge temperature and excess air entrainment can vary at broad scales due to variation in climate and geological structure, assessment of denitrification at regional or broader scales requires estimation of these physical parameters for each study site.
Across diverse ecosystems, availability of organic matter is the primary driver of denitrification (Taylor and Townsend, 2010).Within ecosystems, spatial and temporal variability in the concentration of organic matter and nitrate and anoxic conditions produce heterogeneous mosaics of denitrifying activity (e.g.Harms and Grimm, 2008).In groundwater environments with strong directional flow, denitrification and other redox processes can follow distinctive spatial patterns reflecting the gradual downstream depletion of electron donors and acceptors (Chapelle et al., 1995;Hedin et al., 1998;Tarits et al., 2006).However, a growing body of research suggests that denitrification in most aquifers depends on matrix-derived, solid-phase electron donors (e.g.Fe 2+ , H 2 S) rather than surface-derived solutes (Green et al., 2008;Schwientek et al., 2008;Zhang et al., 2009;Torrento et al., 2010Torrento et al., , 2011)).As a result, concentrations of dissolved organic matter and other electron donors may be a poor indicator of denitrification rates across aquifers, and spatial patterns within aquifers may reflect the distribution of these reactants within the aquifer matrix rather than substrate depletion along advective flowpaths.
Efforts to understand and manage N enrichment of aquatic ecosystems have relied heavily on the distinctive isotopic signatures of potential sources (Kendall, 1998;Kendall et al., 2007), particularly the difference in δ 15 N NO3 between inorganic fertilizers (typically 0-3 ‰) and organic N pools (i.e.animal and human waste; typically 9-12 ‰).More recently, methodological developments that permit determination of both N and O allow greater separation of sources with overlapping δ 15 N signatures (e.g.atmospheric deposition and synthetic fertilizer).Biogeochemical reactions such as denitrification and assimilation can alter these isotopic signatures during transport, potentially confounding N source inference (Kendall et al., 2007).Despite pleas for caution (Bedard-Haughn et al., 2003), observed δ 15 N NO3 are commonly used to infer N sources and guide management and policy related to point and non-point inputs (Fogg et al., 1998;USGS, 2003;Harrington et al., 2010).While the potential effect of denitrification on isotope signatures is widely acknowledged, few studies to date have quantified its influence on source inference at the scale of a regional aquifer (but see McMahon and Bohlke 2006).
In addition to their utility in separating potentially confounded N sources, dual isotopic tracers (δ 15 N NO3 and δ 18 O NO3 ) of NO − 3 can also be used to infer nitrogen transformations.Although insufficient to directly estimate rates of denitrification, coupled enrichment of δ 15 N NO3 and δ 18 O NO3 is now widely used to infer the occurrence of fractionating processes (Burns et al., 2009).Among these are recent studies that suggest denitrification in the Upper Floridan Aquifer and other karst groundwater systems (Panno et al., 2001;Hackley et al., 2007;Albertin et al., 2011).One potential value of dual isotopic measurements is the ability to partition removal into its component processes (e.g.assimilation, denitrification) based on the ratio of 15 N: 18 O enrichment.Theoretical and laboratory studies have suggested that denitrification results in 2:1 fractionation of 15 N: 18 O (Aravena and Robertson, 1998;Lehmann et al., 2003), but other studies have recently suggested a 1:1 ratio (Granger et al., 2008), in which case dual isotopes would be unable to distinguish between assimilation and denitrification.
In this study, our objectives are (1) to quantify the magnitude and variability of denitrification at the regional scale in a karstic groundwater system (the Upper Floridan Aquifer [UFA]), and (2) to assess the influence of denitrification on isotopic signatures of nitrate in the UFA and its influence on apparent N sources.To these ends, we measured dissolved gases and other biogeochemical characteristics of 35 Florida springs, conducted a 3 year study (quarterly-monthly samples) of 6 springs that feed the Ichetucknee River, and assembled published data describing dissolved gas concentration, nutrient chemistry, and/or nitrate isotope composition from over 100 additional springs.From a subset of 31 of these springs for which dual noble gas tracers (Ne and Ar) were available, we derived statistical predictors of excess air entrainment.These data enable an extensive and robust assessment of denitrification and its influence on nitrate isotopic composition at the scale of the entire UFA.

Study system
The karstic Upper Floridan Aquifer (UFA) supports the highest density of large natural artesian springs in the world (Fig. 1), and is a major regional economic resource (Notholt et al., 1989;Miller, 1990;Bonn and Bell, 2003;Bonn, 2004).Throughout parts of northern Florida, the UFA is confined by low-permeability, high-clay deposits that preclude infiltration to the UFA except via sinkholes and fractures; these confining layers are largely absent in the central-western portion of the state (Scott et al., 2004).Springs are concentrated along drainage features, especially near boundaries of confining layers.Land use throughout the study region includes variable mixtures of row crop agriculture, urban and suburban development, and secondary forest (Katz et al., 2001).
Geochemistry of the UFA can generally be characterized as a mixture of two end members (Toth and Katz, 2006;Knowles et al., 2010).Older water, characteristic of matrix porosity and deep flowpaths, is generally anoxic, low in NO − 3 , and enriched in calcium; younger water characteristic of conduits and shallower flowpaths is generally oxic, enriched in NO − 3 , and sometimes subsaturated in calcium (Toth and Katz, 2006;Knowles et al., 2010).Over event-driven and decadal timescales, the contribution of these water sources can vary considerably among springs as changes in flow drive exchange between primary and secondary porosity (i.e. the limestone matrix and karst conduits; Martin and Dean, 2001;Heffernan et al., 2010a, b).Except during runoff and backflow events that deliver organic-matter-rich waters to conduits (Gulley et al., 2011), dissolved organic carbon (DOC) levels in UFA springs are among the lowest measured globally (Duarte et al., 2010).In conjunction with oxic conditions of many springs, low DOC concentrations undoubtedly contribute to the prevailing assumption that denitrification is negligible in this system (Katz, 2004).
NO − 3 concentrations in the UFA and its springs have risen dramatically over the past half-century (e.g.Upchurch et al., 2007), and springs discharge accounts for a large proportion of the N load to estuarine and coastal waters (Pittman et al., 1997).Despite the perceived vulnerability of the UFA to nutrient enrichment, river export accounts for only a moderate proportion of N inputs to North Florida landscapes, and considerable proportions of inputs remain unaccounted for (Katz et al., 2009).Landscape-scale mass balance generally suggests inorganic fertilizer as the primary source of N enrichment (Katz et al., 2009), but isotopic studies (that assumed negligible denitrification) have indicated a greater role of organic N from animal or human waste water (Katz et al., 2001).

Sample collection and analysis
Between June and September 2010, we sampled 33 Floridan Aquifer springs that varied in size, surficial hydrogeology, and NO − 3 and O 2 concentrations.At each spring, we measured O 2 , temperature, specific conductance, and pH from spring vents using a YSI 556 sonde equipped with an optical or Clark O 2 probe.Water samples for laboratory analyses were collected using a peristaltic pump with a 5 m weighted intake tube placed as near as possible to the spring vent.We collected 3 replicate samples for nutrient and isotopic analyses in acid-washed pre-rinsed polyethylene bottles.During the synoptic survey, we collected 5 replicate field samples for dissolved gas analysis by flushing 300 ml BOD bottles 3 times, sealing with glass stopper, and capping with waterfilled plastic caps to minimize exchange with atmosphere and to prevent stoppers from becoming dislodged during transport.Dissolved gas samples were stored under ice water until analysis within 36 hours; water samples were frozen until analysis.
We measured dissolved N 2 and Ar using a Membrane Inlet Mass Spectrometer (MIMS: Kana et al., 1994) within 36 hours of collection, over which period our storage protocol exhibited negligible atmospheric contamination.The membrane inlet mass spectrometer was equipped with a copper reduction column heated to 600 • C to remove O 2 and reduce interference with N 2 measurements (Eyre et al., 2002).Standards for N 2 and Ar concentration consisted of atmosphereequilibrated deionized water in 1 L spherical vessels incubated and stirred in high-precision water baths (±0.01 • C) at their respective temperatures (10, 15, and 20 • C) for at least 24 hours prior to analysis.Gas concentrations in each standard were calculated using temperature-solubility formulas without salinity correction (Hamme and Emerson, 2004).Signal strength for samples and standards was determined as the mean value of the 1st minute following signal stabilization.To account for instrument drift, we ran complete standard curves every 6-8 samples and applied interpolated parameter values from adjacent standard curves (r 2 range: 0.997-1.00;mean r 2 = 0.9997) to estimate gas concentrations in each sample.A fourth standard equilibrated with pure N 2 gas served as an external source QC.Coefficients of variation for field replicates ranged from 0.22-2.27% (mean: 0.80 %; median: 0.49 %).We measured nitrate concentrations (expressed in this paper in mg N L −1 ) in samples from the synoptic survey and Ichetucknee River springs time series using secondderivative UV spectroscopy (APHA et al., 2005) using an Aquamate UV-Vis spectrometer.Isotopic composition of nitrate (δ 15 N NO3 , δ 18 O NO3 ) was measured using the bacterial denitrifier method (Sigman et al., 2001;Casciotti et al., 2002) in the Department of Geological Sciences at the University of Florida (2007Florida ( -2009) ) or the UC-Riverside Facility for Isotope Ratio Mass Spectrometry (2010).
Previously-collected data both increased spatial coverage and in many cases provided repeated measurements of springs included in our synoptic survey (Fig. 1).Measurements of Ne, Ar, N 2 , O 2 , NO − 3 , and nitrate isotopes (δ 15 N NO3 , δ 18 O NO3 ) spanning from 1997 to 2008 were obtained from published articles and agency reports, or directly from researchers when dissolved gas concentration or other data were not reported directly (see supplemental materials).Thirty-six archival observations (from 31 springs) included Ne in addition to Ar, O 2 , and N 2 (and in 23 cases δ 15 N NO3 ).In all, we assembled 112 observations of dissolved gas concentrations (O 2 , Ar, N 2 ) from 62 distinct spring vents, of which 58 included both δ 15 N NO3 and δ 18 O NO3 , and 34 others included δ 15 N NO3 but not δ 18 O NO3 .Excluding the repeated measurements of the Ichetucknee River springs in 2008-2009, our data included 166 observations of δ 15 N NO3 and O 2 and 204 total observations of δ 15 N NO3 .Of the 113 springs represented in the isotope data set, 14 had 4 or more instances of concurrent measurements of both O 2 and δ 15 N NO3 .Observations were drawn from springs in each major drainage in North and Central Florida (Fig. 1), and with discharges ranging from <0.01 m 3 s −1 to 11 m 3 s −1 .Because noble gas concentration data are used to assess contamination of 3 H/ 3 He samples, we were also able to obtain data describing 3 H/ 3 He water age for 36 springs from the same sources that provided Ne data (see Supplement).
All measurements of Ne from prior studies were determined by mass spectrometry at the Lamont-Doherty Earth Observatory Noble Gas Laboratory at Columbia University.Nitrate (as NO 3 -N) was generally measured using the cadmium reduction method (Wood et al., 1967).N 2 and Ar from previously published studies were measured using gas chromatography.We are unaware of studies that directly compare GC and MIMS methods for measurement of N 2 and Ar, but in springs with repeated measurements that included N 2 from both methods, we found no evidence of bias between historic data and our new observations.Recent (2007 and later) measurements of δ 15 N NO3 were generally conducted using the bacterial denitrifier method and included δ 18 O NO3 (Sigman et al., 2001;Casciotti et al., 2002).For data prior to 2007, δ 15 N NO3 was measured via combustion and mass spectrometry (Kendall and Grim, 1990).

Springshed characterization
To determine hydrogeologic predictors of variation among springs in excess air entrainment and recharge temperature, we characterized each spring by latitude, long-term mean discharge, and springshed hydrogeology as measured by aquifer vulnerability to surface contamination (Arthur et al., 2007).We collected discharge records for each spring from online databases of the United States Geological Survey National Water Information System (http: //waterdata.usgs.gov/nwis),Southwest Florida Water Management District (http://www.swfwmd.state.fl.us/data/), and the St. John's River Water Management District (http: //www.sjrwmd.com/toolsGISdata/index.html) where available, since these records were generally the most complete.Where continuous records were unavailable, we used the mean of discrete measurements from published studies and agency reports as our estimate of mean long-term discharge.Since discharge variability of Floridan Aquifer springs is extremely low, use of these more limited data to quantify longterm mean discharge is unlikely to have introduced significant error in subsequent analyses.
Where available, we used previously delineated boundaries (http://www.dep.state.fl.us/geology/programs/ hydrogeology/hydro resources.htm)to characterize springshed hydrogeology, and to estimate springshed size and location for un-delineated springs.The relationship between discharge and springshed area was determined for those springs with previously delineated springsheds (A = Q × 134.9, where A is springshed area in km 2 and Q is discharge in m 3 s −1 ; n = 14, r = +0.79,p < 0.001).For springs without a delineated springshed, we estimated the contributing area based on their period-of-record discharge.We assumed each springshed was circular, and estimated the springshed orientation based on the regional drainage network such that the springshed was located with one edge at the spring vent, and the rest up-gradient from the closest spring-fed river.
We used the Floridan Aquifer Vulnerability Assessment (FAVA) as a metric of springshed hydrogeologic characteristics (Arthur et al., 2007).This measure quantifies the intrinsic contamination risk of the Upper Florida Aquifer (UFA) based on local hydrogeologic conditions such as aquifer chemistry, surface soil permeability, surface elevation, subsurface stratigraphy, presence of karst features (e.g.sinkholes) at the surface, thickness of a the intermediate aquifer system that regulates hydraulic confinement of the UFA, and the potentiometric head difference between the surface and UFA (Arthur et al., 2007).Based on these characteristics, regions of the UFA have been designated as "less vulnerable", "vulnerable", and "most vulnerable".We used the areal coverage of these categories as our descriptor of surface-aquifer connectivity.Because none of the springsheds in this study contained more than 3 % of their area in the less vulnerable category, the proportions of "vulnerable" and 'most vulnerable' were highly correlated.We therefore used only the fraction of each springshed area delineated as most vulnerable, typically more than 75 % of the area, as a predictor in our model of excess air entrainment.

Estimation of excess air, recharge temperature, and excess N 2
The concentration of N 2 gas in groundwater is influenced by physical conditions during infiltration (specifically, recharge temperature [T rec ] and excess air entraintment [A ex ]) as well as biological production of N 2 .We therefore calculated the magnitude of denitrification ([N 2 ] den ) for each sampling date and time as the difference between observed N 2 concentrations and the concentration predicted by physical processes ([N 2 ] phys ): We determined concentrations expected based on physical processes, in turn, as the sume of the equilibrium N 2 concentration at recharge temperature plus additional N 2 dissolved as excess air: where k N2 is a unit conversion factor (from µmol kg −1 to mg L −1 ; 0.028); (N 2 ) Trec is the concentration of N 2 at recharge temperature (in µmol kg −1 ); P N2 is the partial pressure of N 2 in the atmosphere (0.78084) and A ex is the mass of excess air (in mg L −1 ).We used the equations of Hamme and Emmerson (2004) to determine equilibrium N 2 concentration ([N 2 ] Trec ) for a given recharge temperature: where A 0 -A 3 are compound-specific solubility constants (Table 1), and T S is determined as: We used Ne and Ar concentrations to determine recharge temperature [T rec ] and excess air [A ex ] in the subset (n = 36) of springs for which measurements of both gases were available.We estimated these parameters for each observation by simultaneously solving the following equations using the Solver function in Microsoft Excel: where k Ne and k Ar are coefficients for unit conversion of Ne and Ar from nmol kg −1 (for Ne, k = 0.02) or µmol kg −1 (for Ar, k = 0.04) to mg L −1 ; P Ne and P Ar are the mass proportion of Ne (1.81810 −5 ) and Ar (9.3410 −3 ) in the atmosphere; and [Ne] Trec and [Ar] Trec are the equilibrium concentrations (Hamme and Emerson, 2004) of those gases at the recharge temperature as determined by: where T S is a function of T rec (Eq.4) and values for the A 0 -A 3 are given in Table 1.Among this set of springs with Ne data, estimated T rec ranged from 15-22 • C, was overwhelmingly determined by Ar rather than Ne, and varied significantly as a function of latitude (Fig. 2a, b, d).The observed latitudinal variation in T rec is much greater than variation in mean annual air temperature, potentially reflecting regional variation in timing of precipitation and thus temperature of infiltrating water (Schmidt et al., 2001).
A ex ranged from 1.0 to 2.7 ml L −1 and was overwhelmingly determined by Ne rather than Ar (Fig. 2a, b).Based on multiple regression analysis, mean discharge over the period of record (Q POR ) and springshed vulnerability were strong predictors of excess air (Fig. 2c).Palm spring, whose springshed had no land in the 'most vulnerable' category and was the only value less than 50 %, was excluded from this analysis.We used this statistical relationship to estimate A ex in springs for which Ne data were not available.Statistical estimates of A ex constrained solutions for recharge temperature based on Ar concentrations alone (Eqs.6, 8); T rec and A ex were then used to determine equilibrium N 2 concentrations (Eqs.3, 4) and denitrification-dervied N 2 (Eq.1).
To test the hypothesis that [N 2 ] den reflects the magnitude of denitrification, we used regression analyses to evaluate the relationship between [N 2 ] den and dissolved O 2 .We used both linear and logarithmic forms to predict [N 2 ] den from O 2 , for the entire data set and for the subset of observations in which A ex was calculated from Ne data, rather than estimated statistically.We also analyzed the relationship between mean dissolved O 2 and mean [N 2 ] den from the subset of springs for which 3 or more observations were available.For the subset of springs for which 3 H/ 3 He ages were available, we estimated zero-order denitrification rates as the concentration of excess N 2 divided by water age (Green et al., 2008).Because we lack data for discrete geochemical end-members within each spring, we are unable to quantitatively address the effects of mixing (Green et al., 2010) on our estimates of denitrification rates or O 2 thresholds.
To evaluate the relative precision and accuracy of [N 2 ] den estimates based on Ne and statistically modeled excess air, we calculated the mean and standard deviation of [N 2 ] den estimates for all springs with O 2 greater than 2 mg L −1 .Bias in estimates would cause divergence of the mean from zero, assuming that denitrification is negligible in these oxic springs (Bohlke et al., 2002;Green et al., 2008).

Denitrification progression and isotopic fractionation
We indirectly evaluated the relationship between denitrification progression and δ 15 N NO3 via analysis of relationships between dissolved O 2 and δ 15 N NO3 both within and across springs, reasoning that springs with lower dissolved O 2 would have greater depletion of NO − 3 pools by denitrification than springs with higher O 2 .We used both linear and logarithmic regression equations to evaluate dissolved O 2 as a predictor of δ 15 N NO3 across all observations and excluding observations from the Ichetucknee time series collected between July 2007 and November 2009.Inclusion of the entire Ichetucknee data set had a minimal influence on regression parameters, so only the results from the complete (global) data set are reported here.In addition to this global analysis, we used linear regression to evaluate relationships between dissolved O 2 and δ 15 N NO3 within springs for which 3 or more observations were available.We then used regression analysis to evaluate how the strength (as measured by the correlation coefficient (r)) and slope of these within-spring relationships varied as a function of mean dissolved O 2 .This analysis allowed us to evaluate the contribution of variation within and among springs to patterns seen across all observations.
We directly evaluated the relationship between denitrification progression and isotopic composition of NO − 3 by determining the fractionation coefficient ( 15 ε) for δ 15 N NO3 from a cross-system analysis that included springs with dissolved gases from both our synoptic survey and previously reported data, and a separate analysis from the Ichetucknee Springs time series (of which most dates did not include dissolved gas measurements).These analyses required estimates of initial NO − 3 concentration ([NO − 3 ] init ) at the time of recharge, which we estimated using different approaches for springs with dissolved gas data and for the Ichetucknee Springs time series.For analysis of data from the synoptic survey and previous observations that included dissolved gases, we calculated [NO − 3 ] init as the sum of [NO − 3 ] obs and [N 2 ] den (all in units of mg N L −1 ).This estimate would include nitrate derived from nitrification in the vadose zone or UFA as part of [NO − 3 ] init , and assumes that denitrification is the only sink for NO − 3 (i.e. that assimilation, dissimilatory nitrate reduction to ammonium [DNRA], etc. are negligible) as indicated by concentrations of ammonium and particulate and dissolved organic nitrogen that are typically below detection limits at spring vents.Effects of these processes on δ 15 N NO3 are also assumed to be zero.
Estimates of [NO − 3 ] init allow determination of the progression of denitrification.For each observation, we calculated the proportion of nitrate remaining from the original pool (NO − 3 ) R as: where [NO − 3 ] obs is measured concentration, and [NO − 3 ] init is the initial concentration.We used linear regression to determine the isotopic enrichment factor ( 15 ε) for δ 15 N NO3 (assuming Rayleigh distillation kinetics) as the slope of the relationship between δ 15 N NO3 and ln([NO − 3 ] R ) (Mariotti, 1986;Bohlke et al., 2002;Green et al., 2008).We excluded springs with NO − 3 concentrations below 0.05 mg L −1 (Juniper, Silver Glen, and Alexander Springs) from this analysis due to the high variability of [N 2 ] den estimates relative to these lower concentrations.
For the Ichetucknee Springs time series, we calculated [NO − 3 ] R for each spring and sampling date by assuming that [NO − 3 ] init for all springs was equal to [NO − 3 ] obs in the Ichetucknee Headspring on the same date.The first assumption implicit in this analysis is that denitrification rates in the Ichetucknee Headspring are negligible.High O 2 concentrations (mean ± SD: 4.1 ± 0.2 mg O 2 L −1 ), low values of [N 2 ] den (which averaged 0.32 mg N L −1 and represented minimal (<30 %) depletion of the estimated original nitrate pool), and the low and temporally stable δ 15 N NO3 (mean ± SD: 3.6 ± 0.3, n = 16) observed in the Ichetucknee Headspring all support this assumption.The second implicit assumption, that springsheds of the Ichetucknee springs receive equivalent areal rates of N inputs from sources with identical isotopic signatures, is based on the relative homogeneity of land use in the Ichetucknee springshed, and the predominance of fertilizer application to improved pasture as a www.biogeosciences.net/9/1671/2012/Biogeosciences, 9, 1671-1690, 2012 source of N to the watershed (Katz et al., 2009).The third assumption, that variation among springs of the Ichetucknee is driven by denitrification, is supported by strong correlations between dissolved O 2 and NO − 3 within and across these systems (Fig. 3a).
To further evaluate the latter two assumptions, we used [NO − 3 ] init and [NO − 3 ] R values for the springs of the Ichetucknee River on three dates when [N 2 ] den estimates were available.First, we compared [NO − 3 ] init from the Ichetucknee Headspring with the mean value of other springs on the same date.The similarity and covariation of these values (Table 2) is consistent with the assumption that all springs in the Ichetucknee System receive similar N loads.In addition, we assessed the correlation between alternative estimates of [NO − 3 ] R , namely estimates calculated from [N 2 ] den and NO − 3 from each spring and those estimated from the differences in NO − 3 concentration between each spring and the Ichetucknee Headspring (Fig. 3a).The relationship between these estimates (Fig. 3b) suggests that the NO − 3 difference approach used in the analysis of the Ichetucknee Springs time series provides a more conservative estimate of the progression of denitrification than those determined using [N 2 ] den , which is to be expected if denitrification has also reduced NO − 3 concentrations to a small degree in the Ichetucknee Headspring.More importantly, the correlation between these estimates is consistent with the assumption that variation in NO − 3 concentration both within and among the springs of the Ichetucknee River is driven at least in part by differences in the progression of denitrification along the flowpaths that contribute to these springs.
We used fractionation factors ( 15 ε) from our analyses, along with individual measures of denitrification progression ([NO 3 ] R ) to estimate the original isotopic composition of source N for each observation, according to: This calculation assumes that fractionation is constant in space and time within the UFA, and ignores potential mixing effects (Green et al., 2010).To assess the effects of denitrification at the regional scale, we compared the distributions of observed δ 15 N and estimated δ 15 N source , using the reference values reported by Heaton (1986), whose distinguish synthetic fertilizer, organic waste, and their mixtures as <6 ‰ for synthetic sources, >9 ‰ for animal-derived sources, and intermediate values for mixtures.Despite the potential for isotopic composition of these sources to vary in space of and time (Kendall 1998), the reliance on these values by policy makers in Florida makes them important benchmarks.

Results
For the vast majority of our observations, N 2 concentrations exceeded values predicted from recharge temperature and excess air (51 of 61 springs; 94 of 112 observations).[N 2 ] den ranged from −0.7 to 3.5 mg N 2 L −1 (median: 0.67 mg N 2 L −1 ; mean ± SD: 0.82 ± 0.83 mg N 2 L −1 ), and was inversely correlated with O 2 (Fig. 4a).Among springs with Ne data, this relationship exhibited a sharp break at ca. 2 mg O 2 L −1 , above which [N 2 ] den averaged 0.003 mg N 2 L −1 (± 0.32; 2 SE); below 2 mg O 2 L −1 , [N 2 ] den averaged 1.5 mg N 2 L −1 (±0.33; 2 SE).Among all   springs, this threshold was less distinct, and a linear relationship was a better fit than a logarithmic relationship.Among the 16 springs with 3 or more observations (max = 5) of [N 2 ] den , over 90 % of the total variation in [N 2 ] den occurred among rather than within springs (ANOVA; Fig. 5a, Table 3).Standard deviations within springs for [N 2 ] den ranged from 0.05 to 0.65 mg N 2 L −1 and averaged 0.31 mg L −1 .Among the same set of springs, over 88 % of total variation in [NO − 3 ] R occurred among springs (ANOVA; Fig. 5b, Table 3).Standard deviations within springs for [NO − 3 ] R ranged from <1 % to 29 % and averaged 10.4 %.For both variables, variation among springs was strongly correlated with mean dissolved O 2 from the same set of observations.However, variation in [N 2 ] den and [NO − 3 ] R within springs was not correlated with variation within springs in dissolved O 2 , presumably due to low sample size Fluxes of [N 2 ] den from UFA springs were comparable to but uncorrelated with those of NO − 3 , and the proportion of NO − 3 removed by denitrification varied among springs from 0 to as high as 97 % (mean ± 2 SE: 34 ± 9 %) among springs.Denitrification removed more than 75 % of N inputs in 8 of 61 springs, and more than 50 % in 20 of 61.Compared to this spatial heterogeneity, temporal variation in [N 2 ] den among springs was low (Fig. 5, Table 3).We estimate that denitrification reduced total flow-weighted NO − 3 flux from sampled UFA springs by 32 %, with uncertainty in this estimate primarily driven by the representativeness of our sample of springs.Volumetric denitrification rates calculated from [N 2 ] den and 3 H/ 3 He water age (Green et al., 2008) averaged 5.13 µmol m −3 d −1 , with a maximum rate of 24.1 µmol m −3 d −1 .Aggregate (i.e.flow-weighted) areal denitrification for all springsheds was 1.22 kg ha −1 yr −1 , with ca.20 % of springsheds exceeding the estimated global average for groundwater denitrification (3.49kg ha −1 yr −1 ; Seitzinger et al. 2006).
Across 292 observations from 103 springs, δ 15 N NO3 ranged from −0.3 to 23.9 ‰, was inversely correlated with O 2 , and varied more among low-O 2 (<2 mg L −1 ) than high-O 2 springs (Fig. 4b).Among springs with 3 or more observations (max = 18) of δ 15 N NO3 , 86 % of the total variation in δ 15 N NO3 in our data set was accounted for by variation among springs (ANOVA; Fig. 6, Table 3).Standard deviations of δ 15 N NO3 within individual springs ranged from 0.1 to 5.2 ‰ and averaged 1.4 ‰.Unlike [N 2 ] den and [NO − 3 ] R , variation within springs in δ 15 N NO3 was also correlated with variation in dissolved O 2 , particularly for low O 2 springs, which had strong, steeply negative relationships between temporal variation in δ 15 N NO3 and dissolved O 2 (Fig. 7).
The progression of denitrification, as indicated by [NO − 3 ] R , was a strong predictor of δ 15 N NO3 in both the synoptic survey and Ichetucknee River time series (Fig. 8a,  b).Estimated fractionation coefficients ( 15 ε) were similar between the Ichetucknee River time series (7.37) and the synoptic survey (7.49), though slightly less so when springs without Ne data were considered (5.27).Both data sets also exhibited strong relationships between nitrate δ 15 N NO3 and δ 18 O NO3 ; however, the slope of this relationship differed substantially between the synoptic cross-system survey (1.8:1) and the Ichetucknee River time series (1:1; Fig. 8c, d).For the Ichetucknee River springs, relationships between δ 15 N NO3 and δ 18 O NO3 within sampling dates were consistently near unity (Table 4).
Estimates of source δ 15 N NO3 from denitrification progression and observed δ 15 N NO3 values suggest that denitrification alters δ 15 N NO3 at the regional scale.Among springs with estimates of [N 2 ] den , nearly 20 % of observed δ 15 N NO3 values were greater than 9 ‰, and more than 50 % were greater than 6 ‰ (Fig. 9c), values used in Florida and elsewhere to delineate inorganic and organic sources and mixtures thereof (Katz, 2004;Bohlke, 2002).Estimated δ 15 N NO3 of source N (Fig. 9d) were much lower, with only 3.3 % of observations estimated to have original source δ 15 N NO3 greater than 9 ‰, and 31 % greater than 6 ‰.Within the Ichetucknee River time series, differences between the distribution of observed  and estimated source δ 15 N NO3 were even greater (Fig. 9e, f).Across all observations of δ 15 N NO3 , values greater than 6 ‰ were common (>33 %), but were rare among springs with DO greater than 3 mg L −1 (Fig. 9a, b).In all three data sets, estimated source δ 15 N NO3 was both lower on average and much less variable than spring water.

Evidence for denitrification in the Upper Floridan Aquifer
Relationships among denitrification-derived N 2 ([N 2 ] den ), O 2 , and δ 15 N NO3 and δ 18 O NO3 all support the widespread occurrence and significance of denitrification in the Upper Floridan Aquifer.In springs with low dissolved O 2 , N 2 concentrations exceeded those predicted by physical processes (as measured by noble gas tracers), but closely matched those predictions in high O 2 springs (Fig. 4a).The negative relationship between O 2 concentrations and [N 2 ] den provides clear evidence of both the accuracy (0.003 mg N 2 L −1 ) and precision (±0.32 mg N 2 L −1 ; 2 SE) of our approach and the occurrence of denitrification in hypoxic portions of the UFA.Like [N 2 ] den , δ 15 N NO3 was inversely related to dissolved O 2 , with high values observed almost exclusively below 2 mg O 2 L −1 .δ 15 N NO3 values for springs with O 2 greater than 3 mg L −1 were generally consistent with nitrogen derived from predominantly but not exclusively inorganic sources (Figs. 4b,9b).
Relationships between the progression of denitrification and δ 15 N NO3 (Fig. 8a, b) support both the inference of denitrification in the UFA and the hypothesis that variation in isotopic signatures is primarily driven by denitrification rather than differential contribution from organic and inorganic sources.Fractionation coefficients ( 15 ε) derived from both the cross-spring analysis and (5.27-7.49) the Ichetucknee River springs time series (7.37) are within the lower range of values reported for other aquifers (Mariotti, 1986;Bohlke et al., 2002;Green et al., 2008), and other marine and freshwater environments (Sigman et al., 2005;Granger et al., 2008), but are lower than some recent experimental values (Knoller et al., 2011).These relatively low values may indicate diffusion-constrained NO − 3 limitation of denitrification in the UFA (Sebilo et al., 2003), but could also be an artefact of mixing between distinct water sources (Green et al., 2010).Covariation between δ 15 N NO3 and δ 18 O NO3 confirms the inference of denitrification in the UFA and its influence on isotopic composition of NO − 3 at the regional scale.Across springs and over time within the springs of the Ichetucknee River, we observed strong relationships between δ 15 N NO3 and δ 18 O NO3, although the slopes of these relationships differed between the synoptic survey and the temporally intensive study of the Ichetucknee system.The 1:1.7 relationship across sites (Fig. 8c) is consistent with theoretical and empirical studies showing 1:2 enrichment by denitrification (Lehmann et al., 2003;Aravena and Robertson, 1998).In contrast, covariation within the springs of the Ichetucknee River exhibited slopes near 1:1 on each date and across all dates (Fig. 8d, Table 4).A recent study of isotope dynamics within the Ichetucknee River itself yielded similar 1:1 fractionation ratio for 18 O: 15 N associated with denitrification (Cohen et al., 2012).These relationships, both from studies with independently-constrained, direct estimates of denitrification, are similar to fractionation ratios obtained from laboratory experiments (Granger et al., 2008) and marine systems (Sigman et al., 2005).The divergence of fractionation during denitrification between freshwater and marine ecosystems has been attributed to taxonomic or environmental differences between these systems, but our observations suggest factors varying among watersheds can alter these relationships.Regardless of these differences, strong covariation among δ 15 N NO3 and δ 18 O NO3 provide additional evidence that denitrification drives variation in nitrate isotopic composition in the Floridan Aquifer.
Temporal patterns within individual springs provide a final line of support for denitrification as a driver of NO − 3 concentrations and isotopic composition.Within the springs of the Ichetucknee River, O 2 concentrations were positively correlated with NO − 3 concentrations (Fig. 3a).Across all springs, O 2 concentrations were negatively correlated with δ 15 N NO3 , and negative relationships were also observed over time within low-O 2 springs.The apparent absence of O 2driven variation within springs for [N 2 ] den and [NO − 3 ] R is most likely attributable to low power in our data set for those variables and lower precision in estimates of [N 2 ] den (and thus [NO − 3 ] R ) than for δ 15 N NO3 .Given the strength of observed relationships across springs between denitrification progression and δ 15 N NO3 , the most likely explanation for within-spring δ 15 N NO3 -O 2 relationships (Fig. 7) is that variation in both O 2 and δ 15 N NO3 reflect differential contributions of young and old groundwater (Toth and Katz, 2006).In the former, isotopic signatures are likely unaltered by denitrification; in the latter, depletion of nitrate by denitrification under anoxic conditions would enrich δ 15 N NO3 signatures (Kendall et al., 2007).www.biogeosciences.net/9/1671/2012/Biogeosciences, 9, 1671-1690, 2012 A plausible alternative hypothesis is that O 2 − δ 15 N NO3 relationships reflect the concurrent influence of humanor animal-derived effluent on dissolved O 2 (via increased BOD) and δ 15 N NO3 .However, if coincident BOD and δ 15 Nenriched NO − 3 inputs were responsible for these patterns, then high-O 2 springs would exhibit covariation between O 2 and δ 15 N NO3 , as is observed in low-O 2 springs.Thus, while alternative explanations might plausibly explain some of the pairwise correlations observed in this study (e.g.organic waste inputs as a driver of both O 2 and δ 15 N NO3 ), the convergence of multiple lines of evidence clearly indicates that groundwater denitrification is a significant process in north Florida, as both a sink for N inputs and as a driver of variation in isotope signatures.

Magnitude and mechanisms of nitrate reduction
An important feature of this study is that springs integrate upgradient N transformations over considerable spatial domains and over the entire duration of water residence in the subsurface.In combination with the large spatial extent of our study springs, this feature of springs enables relatively simple scaling of groundwater denitrification to springshed and regional scales.Studies of aquifer denitrification generally address denitrification along individual flowpaths with relatively small spatial footprints, and as such may not reflect the distribution of rates or residence times within the larger aquifer.
At the scale of the UFA, denitrification appears to be a significant sink for N leaching to the Upper Floridan Aquifer, removing approximately 32 % of the total (flow-weighted) N discharging from sampled springs.Average volumetric rates derived from [N 2 ] den were toward the low end of values obtained from direct measurement of N 2 , but were within the range reported for agriculturally enriched aquifers with even higher NO − 3 concentrations (Green et al., 2008).Nonethless, aggregate areal rates of denitrification (122 kg km 2 yr −1 ) are comparable to the estimated global average for aquifer denitrification (Seitzinger et al., 2006).These averages, however, integrate a high degree of variation among springs.Average areal rates for individual springs ranged from less than 0 to as high as 5300 kg km 2 yr −1 , and the depletion of NO − 3 load from 0 to more than 97 %.Thus within the Floridan Aquifer we observe variation in denitrification comparable to that observed globally (Seitzinger et al., 2006).Significant spatial heterogeneity of denitrification has been observed in other aquifers, but none to our knowledge have measured denitrification directly over the spatial and temporal extent found in this study.Our results suggest that measurements along individual groundwater flowpaths are unlikely to be applicable to entire regional aquifer systems.
The occurrrence and significance of denitrification in the organic-matter-poor Floridan Aquifer is superficially suprising, given the importance of organic matter supply as a constraint on denitrification across diverse ecosystem types.
However, several recent studies have found isotopic evidence for denitrification in karstic aquifers, including those in southwestern Illinois, USA (Panno et al., 2001), southern Germany (Einsiedl and Mayer, 2006), and the Floridan Aquifer (Albertin et al., 2011).One feature of karst aquifers that may facilitate denitrification is large difference in hydraulic conductivity, and thus water age and chemistry, between karstic aquifer matrices and conduits.McCallum et al. (2008) showed that mixing of groundwaters of different ages, NO − 3 concentrations, and redox potentials can promote aquifer denitrification, and such oxic-anoxic interfaces are widely recognized as locations of significant biogeochemical reactivity in surface waters (Dahm et al., 1998) and groundwaters (McMahon, 2001).Similar phenomena might be occurring throughout the considerable surface area of the conduit-matrix interface in karstic systems, and its occurrence and distribution might vary sufficiently to generate the observed differences in denitrification among UFA springs.
The source and character of electron donors that fuel nitrate reduction in the UFA are not known.Low DOC concentrations in UFA springs could imply that denitrification is fueled by some other source, but availability of labile DOC is likely to be higher in water entering the aquifer than in water discharging from springs.Runoff entering sinkholes provide one pathway for relatively labile carbon to enter karstic aquifers, in either dissolved or particulate form.In some cases backflow from C-rich surface waters influences the chemistry of springs discharge (Gulley et al., 2011), and anthropogenic carbon from septic, municipal, or agricultural waste might also provide labile C to the UFA.That DOC from surface soils and vegetation is degraded after delivery to the UFA is supported by the inverse relationship observed between DOC and spring discharge (Duarte et al., 2010), since smaller springs are less likely to be connected to older flowpaths.Nonetheless, DOC concentrations in virtually all UFA springs are relatively low.It remains to be determined whether the magnitude of DOC inputs and oxidation could account for some, most, or all of the O 2 and NO − 3 depletion that occurs along UFA flowpaths.
The importance of electron donors other than organic C for denitrification in the UFA remains unclear, but has been demonstrated through correlational and experimental studies in other aquifers, including some karstic systems.Although the carbonate Floridan Aquifer matrix itself is generally poor in minerals that might serve as terminal electron donors, Florida springs exhibit significant variation in mineral chemistry (Scott et al., 2004), and many springsheds include the clay-rich Hawthorn Formation (Wicks and Herman, 1994).If such alternative electron donors drive variation in denitrification among UFA springs, then concentrations of solutes such as Fe or SO 2− 4 should be correlated with denitrification.The role of different nitrate reduction pathways, as well as broader drivers such as hydrogeology, geochemistry, and land use as factors that influence denitrification in the UFA, are beyond the scope of this paper, but clearly worthy of further investigation.

Uncertainty in estimates of denitrification
Use of dual noble gas tracers (Ne, Ar) to estimate recharge temperature and excess air produced estimates that were more precise and more accurate than those derived from statistical modeling of excess air.Among springs with O 2 greater than 2 mg L −1 and thus presumably negligible denitrification, [N 2 ] den estimates based on dual tracers averaged 0.003 mg N L −1 (0.1 µmol N 2 L −1 ) with a standard deviation of 0.32 mg N L −1 (11.6 µmol N 2 L −1 ).For observations from high O 2 springs where Ne data were unavailable, [N 2 ] den estimates based on statistically modeled excess air averaged 0.32 mg N 2 L −1 (11.6 µmol N 2 L −1 ) with a standard deviation of 16.8 µmol N 2 L −1 .For [N 2 ] den , [NO − 3 ] R , and δ 15 N NO3 , spatial variation among springs was large compared to temporal variation within springs.
The stronger relationship between denitrification progression and δ 15 N NO3 for springs with Ne data vs. all springs provides additional evidence for greater precision of these estimates.Greater bias and lower precision of these estimates most likely reflects the variability of excess air entrainment over time among springs, but may also reflect introduction of excess air during sampling, an artifact for which our statistical approach does not account.Nonetheless, uncertainty of our direct and statistically-derived estimates of denitrification compare favorably with bias (5 µmol N 2 L −1 ) and precision (SD = 22 µmol N 2 L −1 ) in a previous study of denitrification in agricultural aquifers, in which limited spatial extent permitted assumptions of constant recharge temperature within regions, and calculation of excess air from Ar concentrations (Green et al., 2008).The relatively high precision and minimal bias of [N 2 ] den estimates in this study illustrate both the value of dual noble gas tracers and the utility of statistical modeling of physical processes where direct measurements are unavailable.Similar approaches will likely be necessary and useful in evaluating the spatial heterogeneity of denitrification in other aquifers.
Mixing of water of different ages and histories can have substantial effects on apparent rates, thresholds, and fractionation factors of denitrification in groundwaters (Green et al., 2010).Because our areal rates are a mass flux (the product of [N 2 ] den and discharge over springshed area), rather than an inferred volumetric rate (e.g. Green et al., 2008), we believe that these estimates are robust to mixing effects.However, both the O 2 thresholds that support denitrification and isotopic changes resulting from NO 3 pool depletion are nonlinear and thus potentially subject to mixing effects.Unfortunately, because we lack geochemical end-members within the aquifer, we are unable to quantify these mixing effects on our estimates of these parameters.Qualitatively, the effects of mixing would be to increase the apparent O 2 threshold that supports denitrification, and to reduce the slope of the relationship between denitrification progression and δ 15 N-NO 3 (Green et al., 2008).If these mixing effects are significant, then the actual source δ 15 N-NO3 of springs NO − 3 would be lower, and thus even more distinct from observed values.Thus, our estimates of the effect of denitrification on regional-scale isotope signatures are in all likelihood conservative.
Our approach for estimation of denitrification does not spatially isolate the signal of denitrification occurring within the UFA from that occurring in saturated soil horizons, sinkhole lakes and wetlands, or in overlying geologic strata (e.g.Hackley et al., 2007;Tihansky et al., 1997).Excess N 2 produced in such environments, after the last contact of a water parcel with the atmosphere, could certainly contribute to observed accumulation of N 2 in springs and groundwaters within the UFA.However, the signal of recent recharge in the UFA tends to be oxic, nitrate-rich water (Arthur et al., 2007), suggesting that reduced surface environments contribute minimally to inflow of excess N 2 .Moreover, the strong correlations among gaseous and isotopic signatures of denitrification might not be expected if surface environments with the potential for atmospheric exposure were the predominant location of NO − 3 reduction.One important implication of the low within-spring variance in [N 2 ] den and [NO − 3 ] R (Fig. 5) is that uncertainty in our regional estimate of the magnitude of denitrification and its effect on N loads delivered to surface waters is largely influenced by whether or not sampled springs are representative, rather than by uncertainty of estimates within sampled springs.Our population of springs almost certainly overrepresents large springs, since we include more than half of the first magnitude springs in northern Florida, but we found no relationship between spring size and excess N 2 (data not shown).It is unclear whether our study oversampled N rich or N poor springs, since the distribution of NO − 3 concentrations in small springs is not known.A second source of uncertainty is the magnitude of diffuse groundwater discharge from the UFA, and the comparability of the chemistry of this discharge to that of springs.On an areal basis, we have almost certainly underestimated denitrification in the UFA because estimates only include discharge from one spring over the area of its springshed.Incorporation of excess N 2 fluxes from springs with overlapping springsheds and from diffuse groundwater efflux would increase areal estimates of denitrification, but the magnitude of this bias is not known.It is also unclear whether these unmeasured hydrologic flowpaths have excess N 2 concentrations comparable to those of the measured springs.In light of these uncertainties, our estimate of denitrification at the scale of the North Florida portion of the UFA should be viewed as a first approximation.

Implications for N source inference
Concurrent measurements of dissolved gases, nitrate concentrations and isotopes enabled direct estimation of the absolute www.biogeosciences.net/9/1671/2012/Biogeosciences, 9, 1671-1690, 2012 magnitude and relative progression of denitrification, and their relationship to δ 15 N NO3 enrichment (Fig. 8).The strong negative relationships between the size of residual nitrate pools and δ 15 N NO3 not only provide evidence for the occurrence of denitrification, but also clearly indicate that denitrification exerts a significant influence on nitrate isotopic composition in the UFA (particularly where dissolved O 2 concentrations are <2 mg L −1 ).These relationships, in turn, allowed us to estimate the isotopic composition of the original source NO − 3 for each spring.These estimates suggest that δ 15 N NO3 of nitrate discharging from UFA springs may in some cases differ substantially from the isotopic signature of the original N source.
Observed variation in δ 15 N NO3 among UFA springs was considerable, ranging from values near zero to more than 20 ‰. δ 15 N NO3 distributions for springs with denitrification estimates and within the Ichetucknee River springs were similar to that of all springs, but the former was biased towards heavier δ 15 N NO3 values, and the latter had a smaller range of values and relatively fewer high values.Nonetheless, the similarity of these distributions suggests that interpretation of our source estimates should be applicable to the broader population of springs.In all three data sets, the largest subset of springs had δ 15 N NO3 less than 6 ‰, but in each case more than one-third of springs had δ 15 N NO3 greater than 6 ‰.Observed values in springs with N 2−den estimates were higher than the larger data set that included all springs; more than 50 % of N 2−den springs had δ 15 N NO3 values greater than 6 ‰.
Estimated source δ 15 N NO3 signatures from springs with dissolved gas data and from the Ichetucknee River springs time series had distributions that differed from observed values in two important respects.First, mean and median values for estimated sources were lower by ca. 2 ‰ and 1.5 ‰, respectively.Second, the frequency of extremely high δ 15 N NO3 values was much lower.Significantly, these estimated source values had distributions similar to those of all springs with O 2 > 3 mg L −1 , where denitrification is presumably negligible.Together, these observations strongly suggest that most δ 15 N NO3 values in UFA springs somewhat overestimate the contribution of organic sources, and in particular that very high values overwhelmingly reflect fractionation resulting from nitrate removal by denitrification, rather than large contributions from organic N sources.
To date, denitrification has largely been assumed to be negligible in the Upper Floridan Aquifer, and as a result, elevated δ 15 N NO3 values have been interpreted as indicating a significant contribution from organic sources.Despite mass balance studies indicating fertilizer application as the dominant N source to springsheds, policy and management efforts have largely focused on reducing N inputs from septic tanks, and agricultural and municipal waste (Loper et al., 2005;Dederkorkut, 2005;Mattson et al., 2006), largely on the basis of enriched δ 15 N NO3 signatures.Our data suggest that interpretation of δ 15 N NO3 values must account for fractionating N transformations within the aquifer.
As one example of the effects of fractionation, Wekiwa springs, near Orlando FL, has a heavily urbanized catchment and consistently elevated δ 15 N NO3 values (mean of 4 observations from 2001-2010: 11.5 ± 3.1 ‰), which would initially suggest that elevated NO − 3 concentrations in that system are primarily derived from organic sources.However, δ 18 O NO3 are also highly enriched (12.4 ‰), and measurements of N 2−den are consistently high (mean of 4 observations from 2001-2010: 3.1 ± 0.4 mg N 2 L −1 ).We estimate that denitrification within the aquifer typically removes ca.75 % of NO − 3 before discharge from Wekiwa springs, and that the original source of nitrate in Wekiwa Springs had a δ 15 N NO3 value ca.6.3 ‰.While individual estimates of source δ 15 N NO3 should be viewed with caution, it seems likely that despite its urban setting, N enrichment of Wekiwa springs is due primarily to inorganic fertilizers, with contributions from organic sources.In contrast, Wakulla Springs, near Tallahassee, FL, has somewhat enriched δ 15 N NO3 (6.4-7.9 ‰) but no excess N 2 .Moreover, hydrologic tracer studies have demonstrated direct connections between the Tallahassee municipal waste sprayfield and Wakulla Springs (Kincaid et al., 2005).Thus, it seems safe to conclude that elevated δ 15 N NO3 values in Wakulla springs do in fact reflect the isotopic signature of N sources, which include significant contributions from organic sources.However, our observations suggest that this is the exception rather than the rule.Absent direct evidence for substantial organic sources for a specific spring, efforts to reduce N loading to the UFA should focus on fertilizer inputs.

Conclusions
The importance of denitrification for N fluxes and isotopic composition in the UFA has important implications both for management of North Florida landscapes and for broader understanding of groundwater denitrification.Methodologically, this study illustrates the value of multiple lines of inference for assessing denitrification, which are strengthened by direct estimates of the physical processes that influence N 2 concentration using multiple tracers.Significant spatial and temporal variability of denitrification within the UFA suggests that improving regional and global estimates of denitrification will require more extensive measurements in other aquifers.The variability of denitrification in the Upper Floridan Aquifer has implications not only for regional estimates of N removal, but also for values and variability of isotopic signatures of residual nitrate pools at the regional scale.Accurate assessments of the contribution of various sources of N enrichment, in North Florida and elsewhere, must account for the influence of denitrification on N isotope ratios.

Fig. 1 .
Fig. 1.Geographic distribution of Florida springs and observations sets used in this study.Panels illustrate (A) distribution of study sites (closed symbols) in comparison to distribution of all named springs, (B) distribution of study springs with and without isotopic measurements, and (C) distribution of study sites without dissolved gas data, with O 2 , Ar, and N 2 , and with additional observation of Ne.

Fig. 2 .
Fig. 2. Geochemical indicators (A,B) and springshed predictors (C,D) of excess air (A exc ) and recharge temperature (T rec ) in Florida springs.(A) Ne concentrations are the overwhelming determinant of A exc estimates, and only weakly correlated with T rec ; (B) Ar concentrations had relatively small influence on A exc estimates, but a high degree of influence over estimates of T rec .These relationships permit statistical determination of excess air and robust estimation of recharge temperature for springs without Ne measurements.(C) A exc across springs was best predicted by the combination of aquifer vulnerability (V aq ) and spring size as measured by mean historic discharge (Q por ).(D) T rec decreased with increasing latitude, a relationship that is clearer among springs for which Ne data allowed simultaneous direct estimation of A exc .

Fig. 3 .
Fig. 3. Denitrification as a driver of nitrate concentration in the springs of the Ichetucknee River as indicated by (A) correlations between dissolved O 2 and NO − 3 and (B) correlations between estimates of residual nitrate pools [NO −3 ] R from differences between the Ichetucknee Headspring and from direct measures of denitrification-derived N 2 ([N 2 ] den ).

Fig. 4 .
Fig. 4. Dissolved oxygen as a predictor of (A) excess N 2 and (B) δ 15 N NO3 in Florida springs.In (A), closed symbols indicate measurements of denitrification-derived N 2 ([N 2 ] den ) based on direct estimation of excess air and recharge temperature via Ne and Ar; best fit for Ne springs is given by the solid line.Open symbols in (A) indicate [N 2 ] den measurements based on modeled excess air and estimation of recharge temperature from Ar; best fit for all [N 2 ] den data are shown by the dashed line.In (B), open symbols indicate data from springs with 3 or fewer observations; closed symbols indicate data from springs with 4 or more observations.Lines in (B) are best-fit linear regressions for individual springs with four or more observations of O 2 and δ 15 N NO3 .Springs with high O 2 exhibit little variation in δ 15 N NO3 , but low O 2 springs exhibit higher variability in δ 15 N NO3 that is linked to variation in oxygen concentrations.Best-fit parameters for individual springs and their relationship to mean dissolved O 2 is shown in Fig. 7.

Fig. 5 .
Fig. 5. Variation in the magnitude of denitrification [N 2 ] den ; panel (A) and residual nitrate pools [NO − 3 ] R ; panel (B) within and among springs of the Upper Floridan Aquifer.Variation among springs accounts for more than 90 % of total variation in [N 2 ] den .Full results of analysis of variance are given in Table3.Data shown are from all springs with 3 or more estimates of [N 2 ] den .

Fig. 6 .
Fig. 6.Variation in nitrate isotopic composition (δ 15 N NO3 ) within and among springs of the Upper Floridan Aquifer.Variation among springs accounts for more than 90 % of total variation in δ 15 N NO3 .Full results of analysis of variance are given in Table3.Data shown are from all springs with 3 or more measurements of δ 15 N NO3 .

Fig. 7 .
Fig. 7. Parameters of within-spring relationships between dissolved O 2 and δ 15 N NO3 as a function of spring mean dissolved oxygen.Large negative values of both (A) correlation coefficient and (B) slope in low-O 2 springs, and their absence in higher-O 2 systems, suggest that isotopically-enriched nitrate pools are associated with old, deeply anoxic flowpaths where denitrification would be most likely to occur.Open symbols indicate springs with 3 observations of O 2 and δ 15 N NO3 .Closed symbols indicated springs with 4 or more observations.

Fig. 8 .
Fig.8.Effects of denitrification on isotopic composition of nitrate in Florida springs.Variation in δ 15 N NO3 was strongly correlated with denitrification progression (A) as estimated from excess N 2 and observed nitrate concentrations across 61 springs, and (B) as estimated from differences between the Ichetucknee Headsprings and other springs in the Ichetucknee River.Positive correlation between δ 15 N NO3 and δ 18 O NO3 (C) across springs and (D) over time within the Ichetucknee system are also consistent with denitrification rather than variation in source as a driver of δ 15 N NO3 .

Fig. 9 .
Fig. 9. Implications of fractionation by denitrification for inference of N sources to Florida springs.The distribution of δ 15 N NO3 across all observations (A), among springs sampled for dissolved gases in this study (C) and from the Ichetucknee time series (E) all suggest meaningful contributions of organic sources (one third to one half of springs).However, δ 15 N NO3 values in high DO springs (B) and source δ 15 N NO3 as back-calculated from isotopic enrichment factor and denitrification progression (D, F) suggest inorganic fertilizers and soil N (from mineralized OM) as the predominant source in the overwhelming majority of springs.

Table 1 .
Parameter values for determination of solubilitytemperature relationships for Ne, Ar, and N 2 gas.

Table 2 .
Alternative estimates of initial NO − 3 concentration ([NO − 3 ] init ) and the proportion of [NO − 3 ] R remaining in springs of the Ichetucknee River for dates when direct estimates of denitrification ([N 2 ] den ) are available.All values are in mg N L −1 .
Calculated as the ratio of [NO − 3 ] in each spring relative to [NO − 3 ] init for that spring on the same sampling date.c Calculated as the sum of [NO − 3 ] and [N 2 ] den for each sampling date.d Calculated from all springs other than the Ichtetucknee Headspring for which data are available on each date.
a Calculated as the ratio of [NO − 3 ] in each spring relative to Headspring [NO − 3 ] on the same sampling date.b

Table 3 .
Results of analysis of variance (ANOVA) to evaluate within-vs.among-spring variation in the concentration of denitrification-derived N 2 ([N 2 ] den ), proportional size of residual nitrate pool ([NO − 3 ] R ), and isotopic signature of nitrate (δ 15 N NO3 ).