Dry nitrogen deposition estimates over a forest experiencing free air CO2 enrichment

The quantification of atmospheric additions of nitrogen (N) to an ecosystem is often desirable, but difficult for many locations including many ecological manipulation experiments. Ideal methodologies for the complete and chemically speciated quantification of dry N deposition (e.g. tunable diode laser absorption spectrometer, thermal‐dissociation laser‐induced fluorescence) are expensive, technologically challenging to maintain, and rarely colocated with important global change manipulation experiments. Here we present an alternative method for obtaining an approximation of total N deposition using short‐term eddy flux and concentration measurements, annual regional concentration estimates, and modeling for the Duke Experimental Forest. The motivation for generating estimates for this location was to inform the long‐term elevated CO2 experiment conducted at Duke Forest. We estimated the total annual atmospheric N deposition to the forest to be 13.7 kg N ha−1. Of this total, ∼58% was in dry‐deposited forms. Surprisingly, and contrary to some previous predictions, nitric acid (HNO3) was not the dominant portion of the total dry‐deposited N, implying strongly that other forms of gaseous N like organic peroxy and alkyl nitrate compounds are a significant portion of the total flux. Furthermore, this study sheds light on the completeness of the estimates derived from Clean Air Status and Trends Network (CASTNet) and other dry deposition networks. CASTNet does not measure organic forms of dry deposition. In fact, CASTNet only quantifies HNO3 in the gas phase and may significantly underestimate total N deposition in many environments.


Introduction
The primary inputs of available nitrogen (N) to ecosystems (excluding the direct addition of fertilizers) are biological fixation by organisms, N-containing compounds in precipitation, and the direct removal of gases or particles from the atmosphere. Biological fixation and the N contained in precipitation (i.e. wet deposition) are routinely monitored and estimated at many ecological research sites including those aimed at understanding the influence of global change. However, the deposition of gases and particulates (i.e. dry deposition) is rarely reasonably constrained at such sites. The importance of this 'dry' depositional pathway has been recognized for over 50 years (Meetham, 1950), but technological con-straints have made direct measurements challenging. Dry deposition is an important pathway for ammonia (Sutton et al., 1995), nitrogen dioxide (Duyzer et al., 1995;Duyzer et al., 2004), and nitric acid (HNO 3 ) (Dollard et al., 1987). Therefore, quantifying the dry deposition of these compounds is often imperative to close the N budget of a particular ecosystem.
A multitude of studies are currently underway investigating the influences of global change and, in most, the quantification of dry-deposited N is imperative for understanding ecosystem function. Therefore, the motivation for constraining dry N deposition estimates is high in the ecological community. In recent years, new technologies have been applied to directly and continuously measure the dry deposition of N. For example, Horii et al. (2005) and Farmer et al. (2006) have reported total reactive N (NO y ) and HNO 3 flux estimates using tunable diode laser absorption spectrometery and thermal-dissociation laser-induced fluorescence to pine-forest ecosystems in California. These methods show great promise. However, they are expensive, technologically challenging to maintain, and, because of these constraints, are unlikely to be colocated with many important global change manipulation experiments. Here, we present an alternative method for obtaining an approximation of total N deposition using short-term eddy flux and concentration measurements, annual regional concentration estimates, and modeling. We use the Duke Experimental Forest as a case study and as an exemplar global change experiment site.
There are currently several large-scale experiments examining the influence of elevated CO 2 on terrestrial ecosystem function [e.g. free air CO 2 enrichment (FACE) experiments conducted at Duke Forest, on the Nevada Test Site, Rhinelander, and Illinois]. Most investigations to date show an increase in woody biomass under elevated atmospheric CO 2 (e.g. Bazzaz & Miao, 1993;Curtis & Wang, 1998;Zak et al., 2000;Hamilton et al., 2002) and a potential role of woody ecosystems in sequestering carbon in the future (Schimel et al., 2001). However, the growth response to elevated CO 2 in woody plants appears, in many systems, to be constrained by the availability of soil nutrients, most notably N (Pan et al., 1998;Luo et al., 1999;Zak et al., 2000;Oren et al., 2001). Although there is some debate as to what ecosystems will respond to added N (e.g. Ollinger et al., 2002), these two observations underscore the need to estimate both endogenous cycling of N within the ecosystem and all of the exogenous inputs and losses if we are to predict future function. The most challenging of these estimates is the dry deposition of gases and particles. Measurement methodologies are expensive and technologically intensive. A method that could be broadly applied at a reasonable price would be beneficial to global change research. To that end, the work reported here examines the atmospheric input of gaseous and particulate N to a forested system using a combination of empirical measurements and modeling to generate a constrained estimate of dry N deposition to Duke Forest, an important ecological global change research site currently receiving elevated levels of CO 2 .
The measurements we present here were made as part of the Chemical Emission, Loss, Transformation and Interactions within Canopies (CELTIC) study conducted at the Duke Forest FACE site during the summer of 2003. The overarching goal of CELTIC was to develop a quantitative understanding of the processes controlling the exchange of gas and aerosol carbon and N between the atmosphere and vegetated canopies. The CELTIC study addressed emission and deposition of volatile organic carbon and N, as well as the formation and deposition of secondary organic and inorganic aerosol, through measurements and modeling from leaf to canopy scales. The work presented here focuses primarily on the N flux measurements made as part of the CELTIC experiment.
Estimates of dry deposition of reactive N to individual plants or entire ecosystems have been made using dynamic and open-topped chambers (Gessler et al., 2000;Sparks et al., 2001) and by measuring precipitation throughfall after it passes through the plant canopy (Lovett et al., 2000). However, these methods preclude deposition to nonplant surfaces, chemical transformations within or above the canopy, and attempts to scale chamber measurements to whole-system fluxes are associated with a high degree of uncertainty (Hanson et al., 1989). The best regional estimates of dry N deposition come from the EPA-sponsored Clean Air Status and Trends Network (CASTNet) sites. However, CASTNet sites only measure a subset of dry-deposited N (particulate NO 3 À and NH 4 1 and gas-phase HNO 3 ) and do not quantify several important species of gaseous N (e.g. NH 3 , and major components of total NO y including NO 2 , N 2 O 5 , HONO, PANs, and alkyl nitrates). In addition, CASTNet sites are relatively rare and are often not colocated near global change experiments. Micrometeorological techniques, including eddy covariance, resistance analogies, and Bowen ratio methods, have been used to estimate total dry deposition (e.g. Meyers et al., 1996;Munger et al., 1996;de Miguel & Bilbao, 1999;Nemitz et al., 2001). However, these studies often focus only on part of the total dry deposition based on a molecule of interest and do not quantify the entire flux and they are rarely conducted at research sites examining global change.
Our goal with this work was to provide an estimate for total N deposition in the gas and particle phase to an experimental forest where CO 2 is artificially elevated. This estimate is extremely useful in the context of evaluating the N input and, hence, N dynamics of this system. Although specific to this location to make it relevant to an important global change manipulation experiment, the results and techniques presented are generalizable to any forested system. Secondarily, we were interested in partitioning the potential inputs by chemical species and quantifying the seasonal variation in dry N input to the system. Our approach was to use direct flux measurements of NO y and concentration measurements of HNO 3 , NH 3 , particulate NH 4 1 , and NO 3 À in conjunction with long-term measurements at nearby North Carolina Division of Air Quality (NC DAQ, 2005) and CASTNet (2005) sites to infer an annual input of N to this system from the atmosphere.

Site description
Flux measurements were conducted at the Duke Forest FACE site between July 11 and 25, 2003 [day of year (DOY) 191-206] within one of the control plots (i.e. no CO 2 enhancement). The site has been described in depth previously, both in terms of its physical and biological properties, and its suitability for measurement of ecosystem level fluxes by eddy covariance (Katul & Albertson, 1999). Therefore, we will present only a brief description here. The forest consists of a Loblolly pine (Pinus taeda) plantation with an average canopy height (h c ) of 17 m and an LAI of 3.5 m 2 m À2 (Katul & Albertson, 1999). The understory consists of a mixture of hardwood species dominated by sweetgum (Liquidambar styraciflua). Sweetgum individuals were typically 10 m in height, although some individual trees extended into the upper canopy. Vegetation was homogeneous in the primary wind direction (southwest) for $ 1 km.

Long-term data sites
Data from four NC DAQ long-term air quality stations and one CASTNet station were used to scale depositional data for Duke forest. Two of the NC DAQ sites are in urban areas and two are located in more rural settings. In addition, NH 3 data from a coastal site operated by the UNC Institute of Marine Sciences were also used to help quantify gaseous NH 3 fluxes. Table 1 gives an overview of the sites and distances from the primary Duke Forest field site where flux measurements were made.
Eddy covariance fluxes of NO y A detailed description of the NO y flux measurements and data analysis has been given in Turnipseed et al. (2006) and only a brief description will be presented here. Flux measurements were made at a height of 26 m. Three-dimensional wind velocities (u, v, w) were measured at 10 Hz by a sonic anemometer (ATI-K, Advanced Technologies Inc., Boulder, CO, USA). NO y fluxes were measured similar to the method described by Munger et al. (1996). Sample air at 2000 sccm was drawn into a heated catalytic converter containing a 60 cm piece of 1 4 00 gold tube at 295 1C. A small flow of H 2 (25 sccm) was added just before the heated region to help reduce oxidized N species to NO (Fahey et al., 1985). The flow was then passed to an NO analyzer (Ecophysics CLD 770 AL ppt, Duernten, Switzerland). The CLD 770 AL ppt generates a excess source of ozone, reacts it with the sample flow, and detects light from the chemiluminescent reaction between NO and O 3 : Light emission was detected by a red-sensitive photomultiplier tube and the subsequent current pulses were amplified and counted directly by a PC and recorded at 10 Hz. The analyzer avoids interferences using a prereaction chamber where O 3 is used to consume NO before entering the reaction chamber. Because the rate of reaction with NO is very fast and the reaction with most interference compounds is relatively slower, two sequential runs (the first utilizing the prereaction cell and second going directly to the sample cell without prereaction) allows for the quantification of NO independent of interferences. The analyzer was calibrated daily via standard addition by injecting a small flow (1-10 sccm) of NO standard (1.9 ppm NO/N 2 ) at the inlet. These tests (conducted upon each calibration) also indicated an adequately fast response (to0.5 s, 1/e). The conversion efficiencies and losses of NO y within our system are discussed more fully in Turnipseed et al. (2006). Simply stated, it is likely that there was some loss of HNO 3 within the initial inlet, although the exact amount of loss under the experimental conditions was difficult to quantify. Therefore, it is possible that our NO y fluxes may be underestimated by up to $ 35%. Before calculation of fluxes, wind coordinates were rotated such that v ¼ w ¼ 0 (Kaimal & Finnigan, 1994), and scalar time series were despiked, linearly detrended, and shifted in time relative to w to account for lags induced by the inlet line. These lags were determined via cross correlation between the vertical wind velocity, w, and NO y . Fluxes were then calculated from the covariance between vertical wind velocity (w) and total NO y density.
Measurements were taken at a height of 2 m above the forest floor from July 10 to 16 and at 22 m, just above the canopy, from July 16 to 24. Sample air was drawn via a pump through the ADS (URG Corporation, Chapel Hill, NC, USA) through the following components in series: an aluminum Teflon-coated cyclone (10 Lpm, o2.5 mm cutoff), two annular denuder tubes (30 mm OD Â 242 mm length), a two-stage Teflon filter pack, and a third denuder tube. The Teflon filter pack contains a Teflon filter (47 mm diameter) followed by a nylon filter. The nylon filter and third denuder capture HNO 3 and NH 3 , respectively, which are produced when NH 4 NO 3 is lost by volatilization from the primary Teflon filter. Denuders and filters were extracted with 10 and 5 mL of deionized water, respectively, and analyzed by ion chromatography (Dionex Corporation, Sunnyvale, CA, USA). Measurement uncertainties, expressed as the median coefficient of variation from triplicate samples, were 24%, 11%, 11%, and 27% for NH 3 , NH 4 1 , HNO 3 , and NO 3 À , respectively.
During CELTIC, the denuder/filterpack system was not deployed in a configuration that allowed for the direct determination of fluxes. This is primarily due to the difficulty in achieving acceptable precision at short sampling times required by traditional micrometeorological flux measurement techniques. We have, therefore, estimated fluxes of NH 3 , NH 4 1 , HNO 3 , and NO 3 À using a combination of measurements and modeling, which is a common approach when only routine chemical and meteorological data are available (Meyers et al., 1998;US EPA, 2004). Only chemical and meteorological measurements taken above the canopy were used to estimate dry deposition fluxes from the model simulation. For each chemical species, hourly deposition fluxes (F) were calculated by applying 12-h inte-grated air concentrations (C) to hourly deposition velocities (V d ) (i.e. F 5 CV d ).
Particulate NO 3 À and NH 4 1 deposition velocities were estimated using a version of Slinn's (1982) model, modified to account for particle speciation and surface wetness (Ruijgrok et al., 1997). Deposition rates for HNO 3 and NH 3 were calculated using the classic resistance analogy (Hicks et al., 1987) in which the deposition velocity is viewed as the reciprocal of the sum of resistances to deposition due to turbulence (R a , calculated via Garland, 1977), diffusion across the quasilaminar sublayer of air surrounding the receptor elements (R b , calculated via Meyers et al., 1989), and surface uptake processes that collectively determine the canopy resistance (R c ). For HNO 3 , R c is a function of the dynamic stomatal resistance (Monteith & Unsworth, 1990) and static cuticular (R w ) and ground (R g ) resistances. Based on the approaches of Brook et al. (1999) and Meyers et al. (1998), R w and R g are both set to a value of 30 s m À1 . Forests act as both sources and sinks of NH 3 , depending on the ambient NH 3 concentration, leaf and soil N status, and other environmental factors (Duyzer et al., 1992;Wyers & Erisman, 1998;Pryor et al., 2001). For this reason, the bidirectional canopy-scale NH 3 flux was calculated using a canopy compensation point model developed by Sutton et al. (1998). This is similar to the resistance model used for HNO 3 except that it includes a stomatal compensation point and soil emission. The stomatal compensation point (mg NH 3 m À3 ) was calculated according to Nemitz et al. (2000) as a function of temperature and the leaf emission potential G s 5 [NH 4 1 ]/[H 1 ]. In this study we used G s 5 250, which is well within the range of values reported for low N systems (Flechard & Fowler, 1998;Schjoerring et al., 1998;Milford et al., 2001). Finally, we specified a static soil emission flux of 5.0 ng NH 3 m À2 s À1 based on flux measurements from other unfertilized soils in North Carolina (Walker et al., 2002).

Concentration and flux measurements
During the time period of the CELTIC experiment, NO y fluxes and concentrations were monitored over Duke Forest (Fig. 1). In general, NO y concentrations were highest in the early morning and late evening after 18:00 hours (averages $ 6.5-7.5 ppbv), likely due to a combination of local traffic patterns and lower boundary layer heights during these periods. Average daytime NO y fluxes tended to be between À20 and À25 ng N m À2 s À1 (Fig. 1b). In addition, NO y deposition reached a maximum at midmorning and was substantially lower during the night-time hours due to decreased turbulent mixing. On average, about 0.75 mg N m À2 was deposited daily. This is in reasonable agreement with the daily NO y flux estimates of 0.35-0.70 mg N m À2 day À1 reported by Munger et al. (1996)    Modeled HNO 3 , NO 3 À , and NH 4 1 fluxes (Fig. 3) are consistent with the observed concentration gradients, and correspond to net deposition fluxes of 2.05, 1.43, and 6.85 ng N m À2 s À1 , respectively, over the entire study period. Modeled deposition fluxes of these species are highest during the late morning and early afternoon, reflecting the combination of higher concentrations and more effective turbulent mixing during daytime hours. As described above, HNO 3 deposition was estimated independently of NO y during the CELTIC experiment. Gas phase HNO 3 deposition is a subset of the total NO y flux and comparisons of the two measurements can be used to estimate the proportion of the total NO y flux that is gaseous HNO 3 . During the CELTIC measurement period, the weekly sum of deposited NO y was 5.25 mg N m À2 compared with 1.24 mg N m À2 of HNO 3 (estimated from direct concentration measurements and resistance modeling). Therefore, HNO 3 constituted $ 24% of the NO y flux, which agrees with other recent studies (Sparks et al., 2003;Turnipseed et al., 2006) suggesting organic nitrate and peroxynitrates may represent a large fraction of the total dry-deposited oxidized N and contradicting the common notion that HNO 3 dominates total N deposition at most sites (Williams et al., 1997).
Ammonia concentrations tended to be similar above and below the canopy. This pattern suggests a small net flux or the presence of an NH 3 source in the lower canopy, most likely soil and decaying organic matter. Emissions from these sources are expected to increase with temperature when soil mineralization and litter decomposition rates are highest. The temperature dependence of these sources, in combination with the diurnal cycle of dew at the forest floor, most likely explains the higher daytime concentrations of NH 3 observed below the canopy (Fig. 2b). In contrast, the observed higher concentrations of NH 3 above the canopy at night may reflect the influence of boundary layer depth (Fig. 2b). In both cases, however, the differences in average day and night-time concentrations are small. Over the entire study period, the canopy compensation point model yielded a net downward flux of 5.76 ng NH 3 -N m À2 s À1 in spite of a consistent pattern of emission during daytime hours. During periods of emission, the ambient concentration of NH 3 was below the predicted canopy compensation point. This diurnal pattern of bidirectional exchange is consistent with the relatively low concentrations of NH 3 observed during the experiment (0.23-1.15 mg m À3 ), which were always close to the predicted canopy compensation point (0.01-1.6 mg m À3 ).

Scaling to annual estimates
Because our measurements covered a relatively short period of time ( $ 15 days), annual deposition was estimated using representative data from the nearest long-term measurement sites. Given the significant uncertainties inherent in such an exercise, we present deposition calculations for individual chemical species or total NO y using multiple approaches (scenarios) in an effort to provide a range of estimates.
Annual NO y fluxes Figure 4a shows NO y concentrations measured at several sites across North Carolina. Although there is considerable day-to-day variability between sites, trends on longer time scales (many days to weeks) are well correlated. This suggests that regional scale meteorology plays a large role in governing mean trace gas concentrations across North Carolina and supports our contention that concentrations measured at these various sites over the annual time scale will be relevant to the Duke site. Figure 4b shows NO y concentrations measured at rural and urban NC DAQ measurement sites and measurements made at Duke Forest during CELTIC. Some correlations among sites can be seen and our measurements tend to fall within the range observed at the rural NC DAQ sites (12-h running averages correlated to CELTIC measurements; r 2 5 0.56). This is illustrated in Fig. 4c showing the average diel profile for NO y concentration for the month of July at the various NC DAQ sites and our measured data at Duke Forest (between DOY 194 and 205, July 13-24). Again, the Duke Forest measurements consistently overlap with the rural site estimates. These diel profiles point out another important aspect of congruency between sites. Even though larger scale meteorology appears to govern the mean NO y concentration, all sites, regardless of whether they were situated in urban or rural areas, tended to show similar diel patterns (Fig. 2c). The observed concentration peaks in the morning and evening are consistent with low boundary layer heights and significant local NO x emission (e.g. traffic patterns). The general consistency between NO y concentration measurements from the rural long-term sites and the Duke Forest measurements allowed us to scale our limited dataset to an annual deposition estimate. The scaling strategy was to apply the correlation between the real-time measurements made concurrently in the two locations to predict concentrations at Duke Forest during other times of the year.
The primary uncertainty associated with this extrapolation strategy is that NO y consists of multiple chemical species that have different seasonal cycles, deposition characteristics, and sources, and the NO y deposition is the net flux of their sum. The majority of NO y in moderately polluted environments such as North Carolina consists of NO x (NO 1 NO 2 , 40-80%, Williams et al., 1998). NO is emitted from soils with essentially no uptake by vegetation (Hereid & Monson, 2001). NO 2 is also produced by soils, but is primarily produced in urban and suburban areas by the burning of fossil fuels. NO 2 can be either emitted or taken up by vegetation depending upon the ambient concentration (Sparks et al., 2001). The remaining components of NO y consist of more oxidized forms of N such as organic nitrates, peroxynitrates, and HNO 3 . It has been suggested that a majority of the dry N flux to most vegetated surfaces is due to HNO 3 vapor although it makes up only 2-10% of the total NO y composition (Williams et al., 1997). However, this study, other measurements made during CELTIC (Turnipseed et al., 2006), and other recent work (Sparks et al., 2003) suggest that peroxynitrates deposit more readily to vegetation than was once thought and can make a significant contribution to the total NO y flux. Given the potential variation in total dry N driven by the compositional differences of NO y outlined above, we have used four scenarios devised to encompass the possible range of NO y -deposition values. These scenarios are described below and the annual sums are shown in Table 2. The results across the four scenarios represent an upper and lower bound on the potential annual dry deposition of N to Duke Forest.
Scenario (1) was based on the measurements at Duke Forest indicating that midday NO y fluxes show a linear relationship with concentration, giving a constant deposition velocity (slope) of 0.9 cm s À1 (R 2 5 0.69, N 5 163). This relationship does not hold for night-time data. At night when turbulent transport limited deposition, V d (NO y ) varied with turbulence (denoted by the friction velocity, u*). Average [NO y ] from the two rural sites was combined with friction velocity and solar radiation information from the Duke Forest AmeriFlux site to estimate the total flux of deposited N to the surface. The shortcomings of scenario (1) are that it assumes that all species of NO y contribute equally to the deposition flux and/or the NO y composition remains the same regardless of season. Neither of these assumptions is likely to be valid. Photochemical conversion of NO x to more oxidized N species (e.g. HNO 3 ) occurs more readily during the summer (i.e. higher light levels and temperatures lead to more rapid photochemical conversion). However, because anthropogenic emissions of NO y (consisting mostly of NO/NO 2 from combustion) remain relatively constant throughout the year, a seasonal cycle in the NO y partitioning is typically observed with NO x constituting a higher fraction of NO y during winter periods. Furthermore, NO y concentrations are often higher at the surface during winter due to lower boundary layers (see Fig. 4a and b). These larger wintertime concentrations and the assumption of a constant year-long deposition velocity for NO y yields an overestimate of N deposition using this analysis, but may be viewed as an upper limit to total NO y deposition.
Scenario (2) builds on the observations presented by Munger et al. (1996). Their long-term measurements over Harvard Forest (a deciduous forest located in Massachusetts, USA) suggested similar fluxes of NO y (but not deposition velocity) regardless of season. Therefore, this scenario assumes the average diurnal flux measured during the July measurement period at Duke is applicable throughout the year (i.e. the distribution of chemical compounds within NO y is fixed throughout the year). This is obviously a simplification of the mechanism because it does not require any knowledge of annual trends in concentrations.
Scenarios (3) and (4)  where i represents the week number relative to the fraction observed during week 28). Scenario (4) was generated in a similar fashion, except that the average weekly flux was scaled by the ratio of the total amount of HNO 3 1 NO 3 À for a given week relative to that on week 28, (NO 3 À 1 HNO 3 ) i /(NO 3 À 1 HNO 3 ) i 5 28 . CELTIC NO y measurements spanned week 28 of 2003. Annual NO y sums for all four scenarios are given in Table 2. As described above, scenario (1) likely leads to an upper limit of 5.4 kg N ha À1 yr À1 . Conversely, scenario (2) will likely lead to a lower limit as it does not take into account both increased NO y (from rural NC DAQ stations) and HNO 3 (from CASTNet sites) concentrations during winter, which should lead to higher deposition rates. Scenario (3) may also be negatively biased as the highly oxidized fraction of NO y decreases in winter. Because scenario (4) is scaled directly to HNO 3 1 NO 3 À concentrations, it should provide the most reliable estimate of the seasonal changes in NO y flux and, thus, its result of 4.34 kg N ha À1 yr À1 is likely to be the most accurate estimate.
Annual HNO 3 , NO 3 À , and NH 4 1 fluxes To estimate HNO 3 deposition independently from NO y , we again utilized four scenarios to provide an upper and lower bound for HNO 3 deposition to Duke Forest. The first scenario takes the measured Candor CASTNet concentration as a proxy for Duke Forest and calculates HNO 3 fluxes using a resistance-based multilayer model (MLM; Meyers et al., 1998) to produce weekly estimates of dry deposition as described by Clarke et al. (1997). A comparison of concentration estimates using ADS during the CELTIC study at Duke Forest and filter packs from the Candor CASTNet site suggested a possible bias in the partitioning between HNO 3 and NO 3 À (Table 3) in the CASTNet data. Revolatilization of NH 4 NO 3 from the Teflon particulate filter within the CASTNet filter pack system may impart a negative bias to NO 3 À and a corresponding positive bias in the HNO 3 (which is more readily deposited) collected, leaving the sum of HNO 3 1 NO 3 À unaffected (Table 3; Harrison & Kitto, 1990;Sickles, 1999). Under scenario (2), CASTNet HNO 3 concentrations were first adjusted for potential positive bias resulting from NH 4 NO 3 volatilization on the collection filter using the ISORROPIA gas-aerosol equilibrium model (Nenes et al., 1998), estimated NH 3 concentrations ( Fig. 5; Walker et al., 2004), and meteorological data from the CASTNet site. Candor CASTNet deposition velocities derived from the MLM were then applied to these adjusted weekly concentrations to estimate deposition. Under scenarios (3) and (4), deposition velocities were derived from a single-layer resistance model used during CELTIC with bulk canopy resistances (R c ) of 15 and 0 s m À1 , respectively, using Duke Forest meteorology. These deposition velocities were then applied to the adjusted Candor HNO 3 concentrations to yield weekly deposition estimates for Duke Forest. In each of the four scenarios, hourly fluxes were calculated by applying weekly average concentrations to hourly deposition velocities.
HNO 3 scenarios (1) and (2) yielded annual deposition rates to Duke Forest of 1.56 and 1.38 kg N ha À1 , respectively ( Table 2), suggesting that correction of CASTNet filter pack measurements for NH 4 NO 3 loss at the Candor site had a relatively small effect on annual deposition estimates. The average deposition velocity for scenarios (1) and (2) was 1.3 cm s À1 . Scenarios (3) and (4) yielded annual deposition rates of 1.95 and 2.9 kg N ha À1 , respectively, which illustrates the importance of estimating the bulk canopy resistance (R c ) term accurately. Changes in the canopy resistance (R c 5 0-15 s m À1 ) can lead to substantial differences in the deposition estimate that will become larger when extrapolated to annual or longer time scales. Average deposition velocities for scenarios (3) and (4) were 1.8 and 2.7 cm s À1 , respectively. Scenario (3) likely produces the most accurate estimate of annual deposition, because this approach is consistent with recent measurements of R c 40 (Tarnay et al., 2002;Nemitz et al., 2004) and utilizes Duke Forest meteorology.
Particulate NO 3 À and NH 4 1 dry deposition rates were each estimated using the same three scenarios. Scenario (1) used the unadjusted CASTNet deposition from the Candor site as a direct estimate. Similar to the estimate of HNO 3 deposition, scenario (2) used concentrations that were first adjusted for the possible bias resulting from NH 4 NO 3 volatilization on the CASTNet filter pack measurements. CASTNet deposition velocities for the Candor site were then used with the adjusted concentrations to estimate weekly deposition at Duke Forest. Scenario (3) used the same corrected concentration measurements as scenario (2), but deposition velocities were derived from the modified version of Slinn's model (Ruijgrok et al., 1997) used during CELTIC with Duke Forest meteorology. These deposition velocities were then applied to the adjusted Candor CASTNet concentrations to estimate weekly deposition at Duke Forest. In each scenario, hourly fluxes were calculated by applying weekly average concentrations to hourly deposition velocities.
Scenarios (1) and (2) yielded particulate NO 3 À annual deposition rates of 0.043 and 0.058 kg N ha À1 , respec- tively (Table 2). Adjusting the filter pack concentration measurements for potential NH 4 NO 3 bias had a significant influence on the calculated annual flux for this chemical species (an increase of 430%). However, the relatively low deposition velocities estimated using MLM under scenarios (1) and (2) (Gallagher et al., 1997) and is likely the more robust estimate of true deposition. The annual flux of particulate NO 3 À calculated under scenario (3) (Flechard & Fowler, 1998;Schjoerring et al., 1998;Milford et al., 2001) and soil emissions (Walker et al., 2002) were based on values previously reported for unfertilized systems. Over the range of values used in this analysis, G s has a larger influence on the annual flux estimate than soil emission rate (Table 2). Under scenario (1) (low G s ), the calculated annual deposition rate was 2.07 kg N ha À1 , while scenario (3) (high G s ) yielded a net annual emission flux of 0.18 kg N ha À1 . The lower and upper soil emission rates used for scenarios (4) and (5) yielded a range of annual deposition rates between 1.9 and 1.22 kg N ha À1 . Given that Duke Forest is an unfertilized plantation forest and there are no available direct measurements, scenario (2) (i.e. intermediate values of G s and soil emission) is likely the most reasonable estimate. Scenario (2) produced an annual deposition rate of 1.59 kg N ha À1 . Figure 6 shows the range of weekly deposition estimates for each chemical species under the scaling scenarios described above. During the winter months, dry deposition under all scenarios is dominated by NO y . Interestingly, under most scenario runs, HNO 3 accounts for a relatively small portion of the total NO y addition (32-67% of NO y ). This is important for two reasons. First, it implies strongly that other forms of NO y (most likely organic peroxy and alkyl nitrate compounds) are a significant portion of the total flux. Second, if organic forms of NO y are a large portion of the flux, it casts doubt on the completeness of estimates derived from CASTNet and other dry deposition assessments. CASTNet does not measure organic forms of dry deposition. In fact, CASTNet only quantifies HNO 3 in the gas phase and may significantly underestimate total dry N deposition in many environments.

Annual estimates and seasonal trends
In general, modeled fluxes of individual chemical species tended to follow the seasonal pattern of ambient concentration instead of being limited by such things as plant or stomatal activity. HNO 3 concentrations were highest during winter and spring, primarily in response to higher NO x emissions. Particulate NO 3 À concentrations were also highest during the colder months when HNO 3 was more abundant and low temperatures shift the equilibrium of NH 4 NO 3 toward the aerosol phase. Particulate NH 4 1 is primarily associated with SO 4 À2 at this site. Thus, higher concentrations of NH 4 1 during spring and summer result from the temperature dependence of NH 3 emissions from agricultural and biogenic sources, as well as seasonality in SO 2 oxidation rates. Ammonia deposition was estimated to be highest during the summer when higher ambient concentra-  tions and lower stomatal and cuticular resistances tend to reduce the overall canopy compensation point. Deposition rates were highest at night when stomatal exchange ceases and surface moisture promotes rapid cuticular uptake. Lower deposition rates or periods of emission occur during the day when surfaces are dry and the stomatal compensation point increases in response to temperature. This diurnal pattern is consistent across seasons, indicating that ambient NH 3 concentrations are always near the canopy compensation point. Figure 7 describes the cumulative weekly dry deposition of NO y , NH 3 , particulate NO 3 À , and particulate NH 4 1 over the course of 2003. The central line is our best estimate of total deposition arrived at by combining what we felt were the most robust scaling scenarios for each chemical species. The gray area around the best estimate line represents the range of estimates from all scenarios. Using the respective 'best estimate' scenarios for each chemical species, dry deposition of NO y , NH 3 , NO 3 À , and NH 4 1 during 2003 totaled approximately 7.90 kg N ha À1 . Upper and lower limits on annual dry N deposition were estimated from the combination of scenarios for individual chemical species, which yielded the highest (9.55 kg N ha À1 ) and lowest annual deposition rates (2.63 kg N ha À1 ). While wet N deposition shows strong seasonality, peaking in spring and summer, the accumulation of N from dry deposition rates early in the year result from higher NO 2 concentrations during the winter months. The combination of our best estimate of dry deposition and measured wet deposition (NO 3 À and NH 4 1 ; no measurement of organic N) resulted in an N deposition to Duke Forest of 13.7 kg N ha À1 during 2003 (Fig. 8, Table 4). The seasonal variability of total deposition was primarily driven by the wet fraction with the highest wet deposition of NO 3 À and NH 4 1 occurring during the spring and summer. On an annual basis, dry deposition accounted for approximately 58% of total N deposition ( Table 4).
The results presented here have three important implications. First, it shows that reasonably constrained estimates of dry N deposition can be generated for sites of ecological interest such as a FACE site using shortterm flux measurements and regional concentration estimates. Second, the monitoring of gaseous HNO 3 and particulate nitrate and ammonia as is done across the CASTNet network potentially significantly underestimates total dry N deposition. Third, significant dry deposition of N occurs during the winter and monitoring only during the growing season captures only a portion of the annual flux. These observations taken together strongly suggest that current strategies for estimating dry deposition of N potentially suffer from significant limitations.
Recent modeling efforts (Ollinger et al., 2002;Thornton et al., 2002) have suggested exogenous N inputs will likely have significant implications for future forest production, growth, and, ultimately, carbon sequestration. Our results suggest that modeling based upon parameterizations from national deposition networks (e.g. NADP and CASTNet) only may significantly underestimate dry N deposition. Therefore, a priority should be placed on generating robust estimates of dry N deposition including organic forms and ammonia.