Application of an online ion-chromatography-based instrument for gradient flux measurements of speciated nitrogen and sulfur

The dry component of total nitrogen and sulfur atmospheric deposition remains uncertain. The lack of measurements of sufficient chemical speciation and temporal extent make it difficult to develop accurate mass budgets and sufficient process level detail is not available to improve current air–surface exchange models. Over the past decade, significant advances have been made in the development of continuous air sampling measurement techniques, resulting with instruments of sufficient sensitivity and temporal resolution to directly quantify air–surface exchange of nitrogen and sulfur compounds. However, their applicability is generally restricted to only one or a few of the compounds within the deposition budget. Here, the performance of the Monitor for AeRosols and GAses in ambient air (MARGA 2S), a commercially available online ion-chromatography-based analyzer is characterized for the first time as applied for air–surface exchange measurements of HNO3, NH3, NH+4 , NO−3 , SO2 and SO 2− 4 . Analytical accuracy and precision are assessed under field conditions. Chemical concentrations gradient precision are determined at the same sampling site. Flux uncertainty measured by the aerodynamic gradient method is determined for a representative 3-week period in fall 2012 over a grass field. Analytical precision and chemical concentration gradient precision were found to compare favorably in comparison to previous studies. During the 3-week period, percentages of hourly chemical concentration gradients greater than the corresponding chemical concentration gradient detection limit were 86, 42, 82, 73, 74 and 69 % for NH3, NH+4 , HNO3, NO − 3 , SO2 and SO 2− 4 , respectively. As expected, percentages were lowest for aerosol species, owing to their relatively low deposition velocities and correspondingly smaller gradients relative to gas phase species. Relative hourly median flux uncertainties were 31, 121, 42, 43, 67 and 56 % for NH3, NH+4 , HNO3, NO − 3 , SO2 and SO2− 4 , respectively. Flux uncertainty is dominated by uncertainty in the chemical concentrations gradients during the day but uncertainty in the chemical concentration gradients and transfer velocity are of the same order at night. Results show the instrument is sufficiently precise for flux gradient applications.


Introduction
Development of risk assessments and mitigation strategies such as critical load frameworks (Burns et al., 2008) to protect ecosystems from nutrient and acidic deposition requires accurate speciated deposition budgets of nitrogen (N) and sulfur (S) compounds.In the United States, wet deposition has been well characterized by the National Atmospheric Deposition Program (NADP).The US Environmental Protection Agency's (US EPA's) Clean Air Status and Trends Network (CASTNet) was established in 1991 to characterize temporal and spatial trends in atmospheric concentrations and dry deposition of select N and S compounds in rural locations.Air concentrations of sulfur dioxide (SO 2 ), nitric acid (HNO 3 ) ammonium aerosol (NH + 4 ), nitrate aerosol (NO − 3 ) and sulfate aerosol (SO 2− 4 ) are measured on a weekly timescale using a filter pack (Sickles et al., 1999), from which dry deposition fluxes are estimated using a multi-layer resistance model.While NADP and CASTNet are, in combination, very useful for estimating deposition of some compounds, the N budget derived from these measurements is incomplete, particularly the dry deposition fraction.For example, fluxes of ammonia (NH 3 ) are not quantified.Furthermore, CASTNet dry deposition is not directly measured, but rather it is estimated using a resistance model.This model, and others used within regional chemical transport models such as the Community Multi-scale Air Quality Model (CMAQ), have not been rigorously evaluated across the range of chemical, meteorological and canopy characteristics of ecosystems for which deposition budgets are urgently needed.Thus, the dry component of total N and S deposition remains uncertain due to a lack of measurements of sufficient chemical speciation and temporal extent to develop complete annual mass budgets or of sufficient process level detail to improve current air-surface exchange models.
Over the past decade, significant advances have been made in the development of continuous air sampling measurement techniques with sufficient sensitivity and temporal resolution to directly quantify air-surface exchange of N and S compounds.With respect to N, these include bulk measurements of groups of compounds, such as fast chemiluminescence with thermal conversion for total reactive nitrogen ( N r ) (Marx et al., 2012), thermal dissociationlaser-induced fluorescence (TD-LIF) for total peroxy nitrates ( PNs) and total alkyl and multifunctional alkyl nitrates ( ANs) (Farmer et al., 2006).More selective methods for specific compounds have also emerged, including chemical ionization mass spectrometry (Sintermann et al., 2011) and tunable diode laser spectroscopy (Whitehead et al., 2008) for NH 3 , thermal dissociation-chemical ionization mass spectrometry for peroxyacetyl nitrate, peroxypropionyl nitrate and peroxymethacryloyl nitrate (Wolfe et al., 2009), and aerosol mass spectrometry for inorganic particles (Farmer et al., 2011;Nemitz et al., 2008).These methods are sufficiently fast such that fluxes may be quantified by the eddy covariance (EC) technique.However, their applicability is generally restricted to only one or a few of the compounds within the deposition budget.
NH 3 and HNO 3 , which are thought to together dominate the N deposition budget in many areas (Dennis et al., 2013), are difficult to measure due to their tendency to stick to surfaces within the sampling and analytical components of online measurement systems.For this reason, wet chemical techniques such as the Gradient of Aerosols and Gases Online Register (GRAEGOR) (Thomas et al., 2009;Wolff et al., 2010), Ammonia Measurement by ANnular Denuder sampling with online Analysis (AMANDA) (Wyers et al., 1993) and GRadient Ammonia High Accuracy Monitor (GRA-HAM) (Kruit at al., 2007) systems are the preferred methods for air-surface exchange applications.These systems are configured such that the air sample travels only a short distance (∼ 0.1 m) before diffusion into solution within a wet rotating denuder.Opportunity for loss to surfaces within the sampling system are therefore minimized.A secondary benefit of the wet chemical techniques is that the use of ion-chromatography or flow injection analysis allows for simultaneous measurement of multiple compounds, thereby minimizing the bias introduced by constructing deposition budgets from multiple measurement systems.For the NH 3 -HNO 3 -NH 4 NO 3 system, simultaneous measurement of gas and aerosol components is essential to assess potential errors in fluxes related to aerosol thermodynamic instability (Wolff et al., 2010;Nemitz et al., 2004).Furthermore, simultaneous measurement of S and N compounds allows for examination of co-deposition effects between SO 2 and NH 3 related to surface acidity (Erisman and Wyers, 1993) as well as the degree of ammonium sulfate aerosol neutralization.While wet chemical techniques meet the rigorous precision and accuracy requirements of air-surface exchange applications, their temporal resolution is on the order of 30 min to 1 h.In contrast to the direct EC technique, in which air concentrations are measured at 10 Hz or faster using a single concentration measurement, fluxes must be quantified using the aerodynamic gradient method (AGM), which uses gradient concentrations at a 30 to 60 min temporal average.Furthermore using gradient concentration measurements requires additional experiments to determine the precision associated with using two sampling collection devices.
The Monitor for AeRosols and GAses in ambient air (MARGA, Metrohm-Applikon, the Netherlands) is a commercially available online ion-chromatography-based analyzer that semi-continuously measures gases and soluble ions in aerosols (ten Brink et al., 2007;Makkonen et al., 2012;Rumsey et al., 2014) The MARGA is quasi-similar to the GRAEGOR system described by Thomas et al. (2009) and Wolff et al. (2010), which has been used for flux measurements.The major difference between the MARGA 2S and GRAEGOR systems is that the MARGA employs ion chromatography for analysis of both anions and cations whereas the GRAEGOR employs ion chromatography for anions and flow injection analysis for cations.The MARGA also employs mass flow control to regulate air sampling flow rates as opposed to control by critical orifice, as in the GRAE-GOR.Another difference between the GRAEGOR and the MARGA that may influence the performance of the instruments is the integration of instrument control and chromatography in the MARGA software, which includes real-time instrument performance and data quality indicators for air and liquid flows, sample collection device conditions and chromatography.The performance of the GRAEGOR in measuring air-surface fluxes has been described by Thomas et al. (2009) and Wolff et al. (2010); however, there has been no evaluation of the performance of the MARGA in measuring air-surface fluxes.Furthermore, neither the Thomas et al. (2009) nor Wolff et al. (2010) studies assessed the performance of the GRAEGOR for S compounds in comparison to empirical gradient flux data.
In this study, the performance of the MARGA in measuring gradient flux of speciated N and S is evaluated and described for the first time.This study uses a MARGA 2S system, which is different from the MARGA 1S system described by Rumsey et al. (2014) in two key ways.First, the 2S system employs two sampling boxes interfaced to a single analytical system.The two sampling boxes in this case are positioned at two heights above the terrestrial surface to simultaneously measure the vertical concentration gradient.Second, the MARGA 2S, as configured for this work, draws the air sample through a much shorter length of tubing (30 cm) relative to the 1S configuration described by Rumsey et al. (2014).
The objective of this paper is to comprehensively evaluate and describe the performance of the MARGA in the measurement for air-surface exchange measurements of HNO 3 , NH 3 , NH + 4 , NO − 3 , SO 2 and SO 2− 4 .This requires two sets of experiments: one set to describe the performance of the MARGA as an analytical instrument and another to determine the performance of the MARGA as a gradient flux system.The analytical performance of the instrument is assessed by determining the accuracy, precision and analytical detection limit of the instrument using liquid standards in field conditions.To assess the performance of the MARGA as a gradient flux system, the precision of the concentration gradient which can also be defined as the gradient detection limit is determined in field conditions.The concentration gradient precision (uncertainty) and overall flux uncertainty (concentration gradient uncertainty + transfer velocity uncertainty) are then examined for a representative 3-week period over an unfertilized grass surface during the fall of 2012.A companion paper focusing on the air-surface exchange processes of individual compounds over a longer period of study is forthcoming.

Study site
Measurements were conducted in an unfertilized 15 ha grass field in the Blackwood Division of Duke Forest, Orange County, North Carolina, USA (35.58 • N, 79.05 • W).Vegetation is primarily tall fescue (Festuca arundinacea Shreb.), with less common species consisting of a mixture of C3 and C4 grasses, herbs and forbs (Fluxnet, 2014).The field is generally cut twice per year, once in summer and fall, and the clippings are removed for use as animal feed at local farms.

Description of MARGA gradient system
As previously mentioned, the MARGA is a commercially available online ion-chromatography-based analyzer that semi-continuously measures gases and soluble ions in aerosols.The 2S version used in this study employs two sampling boxes interfaced to a single analytical system.The two sampling boxes (SB1 and SB2) are positioned at two heights above the surface to measure simultaneous concentration gradients from which the vertical chemical fluxes are calculated.Air is sampled through a short length (30 cm, 0.5 outer diameter) of PFA Teflon tubing with a coarse Teflon screen over the inlet to exclude large material such as insects and entrained vegetation.
Each sample box contains a wet rotating denuder (WRD) and steam jet aerosol collector (SJAC).The sample air first flows (as laminar flow) into the WRD (Wyers et al., 1993;Keuken et al., 1998), which rotates continuously so that the walls of the denuder are coated with absorption solution (double de-ionized water with 10 ppm hydrogen peroxide), ensuring that the gases diffuse into the liquid film.The level of the bulk liquid within the WRD is kept constant using a level sensor and pump connected to the absorbance solution.Particles pass through the WRD and are collected directly downstream in the SJAC (Khlystov et al., 1995).Within the SJAC, a supersaturated environment is created in which particles grow by deliquescence, allowing them to be collected by inertial separation.The supersaturated environment is created using a temperature-controlled steamer continuously supplied with absorbance solution.Air is drawn through the WRD and SJAC at 16.7 L min −1 using a vacuum pump (KNF model N840FT.18,KNF Neuberger, Inc., Trenton, NJ) and mass flow controller (Brooks Smart Mass Flow Controller, Brooks Instrument, Hatfield, PA).The liquid samples from the WRD and SJAC are collected in a syringe pump module located in the detector box.The syringe pump module consists of three sets of syringes: one set for the WRD, another for the SJAC and a third for the internal standard.The syringe pumps operate in tandem such that, while a set of samples is being collected, the set collected during the previous hour is being analyzed.Prior to analysis, each sample (volume = 25 mL) is mixed with an internal standard (LiBr) solution, which uses two smaller syringes (volume = 2.5 mL).Further information on the internal standard, the absorption solution and other chemical solutions used for MARGA ion-chromatography system is included in Sect.S2.2 in the Supplement.The samples are analyzed using cation and anion ion conductivity detectors (IC, Metrohm USA, Inc., Riverview, FL, USA).For the cation chromatography, the MARGA uses a 500 µL injection loop and a Metrosep C4 150 mm column (Metrohm USA, Inc.) in conjunction with a methanesulfonic acid eluent.For the anion chromatography, the MARGA uses a 250 µL injection loop and a Metrosep A Supp-10 75 mm column (Metrohm USA, Inc.) in conjunction with an eluent containing sodium carbonate monohydrate and sodium bicarbonate anhydrous.
Software integrated within the MARGA calculates atmospheric concentrations based on air sample flow rate, syringe speed during injection (relatively constant) and ion concentrations (corrected for internal standard) in the collected solutions.These results, as well as the anion and cation chromatograms and various hardware parameters, are recorded by the MARGA software.

Accuracy
To verify the analytical accuracy of the MARGA as controlled by the internal LiBr standard and to assess potential contamination in the liquid solutions and in the liquid flow path of the MARGA system, an experiment was conducted during field deployment using a liquid blank and four liquid external standards with different concentrations.Furthermore, the relationship between the expected and observed external standard concentrations as well as the blank concentrations were used to adjust the raw concentration data prior to flux calculations.Both the blanks and the external standards experiments were conducted with the air pumps disconnected and denuder inlets sealed.A blank was assessed using the absorption solution for a period over 24 h.The external standard test was conducted by replacing the absorption solution with a known liquid standard containing SO 2− 4 , NH + 4 , NO − 3 , Na + and K + .Although Na + and K + atmospheric concentrations are the not the focus of this particular study, the analytical performance of the MARGA for these compounds is included to give additional information on the performance of the MARGA for a range of compounds.The ranges of concentrations for the external standard were chosen to represent the typical ambient concentrations observed at the study site.Additional information on the external standard liquid solutions is provided in Sect.S2.3.1 in the Supplement.The external standard experiments were conducted for a minimum of 12 h.
It is acknowledged that the liquid external standards used to determine accuracy do not take into account all uncertainties associated with the MARGA measurement system.In this study, it is assumed that the performance of the WRD and SJAC for collecting gases and aerosols is similar to that reported by Keuken et al. (1988), Wyers et al. (1993) and Khylstov et al. (1995), respectively.The inlet associated with the MARGA sampling system may also affect measurement accuracy, particularly for "sticky" gases such as NH 3 and HNO 3 .In this study, however, possible inlet effects were minimized by using a short length (30 cm) of PFA Teflon tubing.Cross-sensitivity of the WRD in measuring dinitrogen pentoxide (N 2 O 5 ) as NO − 3 during the nighttime as reported by Phillips at al. (2013) may also affect the accuracy of NO − 3 measurements.In an analysis on the MARGA instrument, Phillips et al. (2013) determined that on average N 2 O 5 contributed 17 % of measured nighttime HNO 3 at a sampling site near Frankfurt, Germany.The magnitude of N 2 O 5 concentration varies for different geographic locations and is influenced by nitric oxide (NO) concentration, biogenic volatile organic compound concentrations and air temperature (Phillips et al., 2013).During the period of study presented here, the influence of the artifact on HNO 3 fluxes is likely small as a result of the majority of flux occurring during the daytime (owing to diurnal patterns in the momen-tum flux; see Fig. S6 in the Supplement) and also due to the HNO 3 concentration, which is < 0.15 µg m −3 on average at night.Though the N 2 O 5 artifact was not quantified in the current study, its potential importance for sites in the southeast US invites future investigation.

Analytical detection limit
The detection limit of an analytical instrument is defined as the lowest concentration that can be determined to be statistically different from a blank at a certain level of statistical confidence.In this study, the MARGA detection limit is calculated using a method from Currie (1999): where σ o is the standard deviation of the distribution of concentration when the concentration is just above the detection limit, v is the degrees of freedom, α is the level of statistical significance and t is the Student's t statistic.The analytical detection limits of the MARGA were calculated using an observed liquid concentration after being adjusted for the internal standard.
The detection limit was determined by combining data from all analytical channels (in this study, there are four different channels including denuder and SJAC samples from both sample boxes) into a single data set.From this single data set, the standard deviation and number of analyses are used to determine the detection limit.However, using this approach means that the standard deviation may reflect a combination of random error plus systematic error between channels.To investigate this possibility, the detection limit methodology was conducted in conjunction with the Dunn's test (Dunn, 1964) and the Brown-Forsythe test (Brown and Forsythe, 1974) to compare differences across channels.Additional information on the detection limit methodology, which includes descriptions of the Dunn's test and Brown-Forsythe test methodologies, is provided in Sect.S2.3.2 in the Supplement.
When quantifying the detection limit using an external standard, the aim is to use a concentration that is slightly above the detection limit as variance may increase with increasing concentration.Therefore an appropriate external standard level was selected for each compound.In addition, two different solutions used to determine SO 2− 4 and Na + detection limits allow an opportunity to investigate the relationship between concentration level and variance and thus its potential impact on the detection limit.

Aerodynamic gradient method
Air-surface exchange fluxes were quantified using the AGM.The AGM is based on the application of Fick's Law to turbulent diffusion, which relates the flux of heat, mass and mo-mentum to the vertical gradient and turbulent transfer coefficient (eddy diffusivity) of the particular scalar of interest, in this case air concentration (C).Following Thomas et al. (2009), which is an adaptation from Thom (1975), the flux may be expressed as where u * is friction velocity, calculated from the momentum flux measured by EC, and C * is the concentration scale calculated as Here ψ H is the integrated stability function for sensible heat (Thom, 1975), z 1 and z 2 are the measurement heights above ground between which the concentration gradient ( C) is measured, L is the Monin-Obukhov length calculated from the EC-derived sensible heat flux, k is the von Karman constant (k = 0.41) and d is the zero plane displacement height, which is determined by canopy height using the relationship provided by Stanhill (1969).During the period for which fluxes are presented, average grass height within the field was 1.2 m and gradient sampling heights were 1.25 and 4.0 m above ground.AGM fluxes were calculated from hourly average concentration gradients and hourly EC momentum and sensible heat fluxes measured above the canopy.EC momentum and sensible heat fluxes were measured with a sonic anemometer (R.M. Young model 81000V, Traverse City, MI) placed approximately 2.6 m above the ground.EC fluxes were calculated offline from 10 hz data using SAS (SAS Institute, Cary, NC) software following standard approaches for EC.Hourly average concentration gradients were based on adjusted air concentration data.Air concentration measurements were adjusted using the internal LiBr standard, external liquid standard calibrations and air flow quality control (QC) checks.Air flow rates were independently measured at the denuder inlet at least weekly using a National Institute of Standards and Technology (NIST)-traceable primary standard (Bios DryCal DC-Lite flowmeter, Mesa Laboratories, Inc., Lakewood, CO).If the measured airflow rate was > 5 % different from the target airflow rate of 16.7 L min −1 , the MARGA mass flow controllers were recalibrated using the MARGA calibration feature, which consists of a three-point calibration at 0, 15 and 18 L min −1 .The MARGA software continuously records the air flow rate, which is used to calculate air concentrations from liquid concentrations.Air concentrations were also adjusted for the systematic difference in gas and aerosol concentration measurement between the sampling boxes, during co-located sampling in which the two sample boxes are positioned sideby-side.The systematic difference is determined by plotting the concentrations during the colocation against each other and calculating the slope and intercept by orthogonal least squares regression (Wolff et al., 2010).Slope and intercepts significantly different from 1 and 0, respectively, indicate systematic error between the two boxes, which, if present, is subsequently removed prior to calculation of the concentration gradient.

Flux uncertainty and concentration gradient uncertainty
The flux uncertainty σ F is calculated as (Wolff et al., 2010) where F is the flux, C is the concentration gradient and σ C is the precision (uncertainty) of the concentration gradient and v tr and σ v tr are the transfer velocity and precision (uncertainty) of the transfer velocity, respectively.Equation ( 4) is used to assess the uncertainty in the measured fluxes and to quantify the relative contributions of uncertainty in chemical and meteorological measurements.In addition, each observation may be assigned a data quality indicator as being above or below the flux detection limit.v tr is taken as The transfer velocity and thus the uncertainty in the transfer velocity is a function of friction velocity (u * ), the measurement (z 1 and z 2 ) and displacement (d) heights and the integrated stability function at each height (ψ H ), which is a function of u * , the sensible heat flux (H ), buoyancy parameter (g/T ), air density (ρ), specific heat (c p ) and von Karman's constant (k) (e.g., Arya et al., 2001).As noted by Wolff et al. (2010), there are no published estimates of the full uncertainty in v tr .Here we approximate the uncertainty in the transfer velocity (σ v tr ) by calculating v tr using measurements of L and u * from six co-located R.M. Young model 81000V sonic anemometers.The standard deviation of this population (n = 6) of measurements of v tr represents a lower limit of its transfer velocity uncertainty (σ v tr ), as uncertainty in d and ψ H is not explicitly considered.In this case, the precision of v tr was quantified as the standard deviation of the residuals of orthogonal least squares regression of v tr from individual sonic anemometers against the mean for the corresponding hourly observation.The v tr precision experiment was conducted adjacent to the MARGA measurement location and comprises 110 hourly observations.The uncertainty of the gradient concentration is also quantified during co-location tests.Again, the concentrations from the two sample boxes are regressed against each using a slope and intercept from orthogonal least squares regression.The residuals of the regression represent the random error of the concentration difference between the two boxes.The standard deviation of the residuals provides a measure of the precision of the denuder and SJAC measurements for individual analytes, which also represents the precision of the concentration gradient (σ C ), which can also be defined as the gradient detection limit following Wolff et al. (2010).The relationship between precision and concentration is quantified by regressing σ C on the average concentration (C) between the two boxes.Precision is expected to be a function of concentration.Therefore, concentration gradient precision was calculated at three different co-location events (June-July, August and October 2012) during the sampling periods in order to have a wide range of concentrations.
Air-surface exchange fluxes and their associated concentration gradient uncertainty and flux uncertainty were determined over 3-week representative period (23 September-14 October 2012) at the sampling site.

Ancillary measurements
A variety of meteorological parameters and surface characteristics were measured during sampling.The influence of these factors on air-surface exchange flux will be examined in the companion paper.In this paper, the meteorological parameters, wind speed, air temperature and global radiation will be presented to provide basic information on meteorological conditions during the 3-week representative period.Wind speed and air temperature were measured using the sonic anemometer (R.M. Young model 81000V, Traverse City, MI) at a height of 2.6 m.Global radiation was measured using the REBS Q7.1 net radiometer (Campbell Scientific, Logan, UT).Other surface characteristics reported in this paper include canopy height, which was measured manually and leaf area index (LAI).Single-sided LAI was measured by destructive (prior to canopy closure) and optical methods (LICOR model LAI-2000, LICOR Biosciences, Lincoln, NE) 3 Results and discussion

Accuracy
In the following analysis, results from liquid standard tests are expressed as equivalent air concentration unless otherwise noted.Also, though liquid standards obviously only contain dissolved ionic forms (i.e., NO − 3 , NH + 4 , SO 2− 4 ), results are reported for both SJAC and denuder samples, adjusting for molecular weight to express denuder results in equivalent gas phase concentration (i.e., HNO 3 , NH 3 , SO 2 ).The results of the liquid blank are provided in Table S1 in the Supplement.Only SO 2− 4 had a significant blank (value > 0.001 µg m −3 ).Further analysis by an independent IC system has confirmed that both the absorption solution and the MARGA system components contribute to the SO 2− 4 blank.The relationship between the expected and observed (re-sponse) concentrations of the external liquid standards was investigated by regression analysis (see Fig. S1 in the Supplement).For NH + 4 , NH 3 and K + there is good agreement between expected and observed concentrations, with the linear regression slopes for all compounds between 1.00 and 1.04 and offsets close to 0 (< 0.006).For Na + , the external standard response was not as accurate, producing slope values of 0.90 for both SB1 and SB2 and offsets of 0.013 and 0.011 for SB1 and SB2, respectively.This appears to be related to peak integration characteristics but is currently under investigation.
For the sulfur compounds (SO 2− 4 and SO 2 ) associated with anion IC detection, excellent regression slopes were also observed (1.00); however, offsets (intercepts) can be observed using linear regression, which are not reflected in the blank.These offsets range from 0.09 to 0.13 µg m −3 for SO 2 and SO 2− 4 .Linear regression analysis of NO − 3 and HNO 3 produced good regression slopes, ranging from 1.06 to 1.07, and similarly offsets that are not reflected in the blank, ranging from 0.05 to 0.06 µg m −3 (see Fig. S2).Further investigation of the difference between expected and observed concentrations for NO − 3 and HNO 3 at individual external standard levels show that the difference (observed concentration minus expected concentration) at the lowest concentration external standard (equivalent expected air concentrations are 0.131/0.126(SB1/SB2) and 0.133/0.128(SB1/SB2) µg m −3 for NO − 3 and HNO 3 , respectively) is considerably smaller than for the other higher external standard concentrations (see Table S2 in the Supplement).Therefore, a nonlinear (quadratic) standard curve was fitted which produced slightly higher r 2 values and lower offset values in comparison to the linear fit (see Fig. S1).Thus it is hypothesized that a nonlinear response occurs at low NO − 3 concentrations.It is proposed that a similar nonlinear behavior at low concentrations may also exist for SO 2− 4 and SO 2 .However, in this experiment, the lowest SO 2− 4 and SO 2 external standard concentrations (expected equivalent air concentrations are 0.476/0.461(SB1/SB2) and 0.318/0.308(SB1/SB2) µg m −3 for SO 2− 4 and SO 2 , respectively) may have been too large to observe this nonlinearity as use of a quadratic standard curve on the SO 2− 4 and SO 2 data did not reduce the intercept relative to linear regression.More recent results (not shown), however, support the presence of nonlinearity in SO 2− 4 and SO 2 responses at low concentrations.These results suggest that the response is nonlinear below a SO 2− 4 concentration of 0.27 µg m −3 (equivalent to a SO 2 concentration of 0.18 µg m −3 ).Therefore, it was concluded that it was more appropriate to not adjust concentrations for the linear regression slope and offset below these concentration values and to only subtract the experimentally determined blank.For HNO 3 and NO − 3 , quadratic standard curves are used for external standard adjustments and linear functions are used for the other remaining compounds.

Analytical detection limit
A summary of the detection limit analysis for each analyte is provided in Table 1.Detection limits in Table 1 were determined by incorporating data from all four channels for each analyte.Calculated detection limits were low for all compounds ranging from (in equivalent air concentration) 0.020 µg m −3 for NH 3 to 0.064 µg m −3 for SO 2− 4 .In summary, the results of Dunn's test and Brown-Forsythe test indicate that the sampling components of the MARGA are influencing the detection limit of all the compounds except K + (see Sect.S3.1.2ain the Supplement).Therefore, the influence of systematic difference among channels was examined by calculating the detection limit for individual channels using Eq. ( 1), then averaging the four detection limits (see Table S4).Using this methodology, detection limits were lower for all the compounds that had been identified as having a significant difference in median channel concentration or channel concentration variance (all compounds except K + ) with the exception of the 1.75 µg L −1 Na + standard, which was approximately the same.The largest decrease in detection limit was for the 19.47 µg L −1 SO 2− 4 standard, which decreased by 0.009 µg m −3 .The average decrease in detection limit for the compounds was 0.004 µg m −3 .For K + , the only compound that was determined to have no significant difference in median channel concentration or channel concentration variance, the detection limit was slightly higher than for the previous method (0.038 µg m −3 ) with a value of 0.040 µg m −3 .The NH + 4 , NO − 3 , HNO 3 , SO 2 and SO 2− 4 detection limits from this study (using detection limits from Table 1) are lower than those determined under field conditions in the Thomas et al. (2009) and Wolff et al. (2010) studies, which used the GRAEGOR system (see Table S5).NH 3 detection limits determined in this study are lower than those reported by Thomas et al. (2009) and Wolff et al. (2010) at a grassland site but are similar to Wolff et al. (2010) at a forest site.This aforementioned comparison takes into account differences in the methodology used for determining detection limits.The lower detection limits observed in this study may be attributed to differences in temperature related detector stability, stability of liquid flow rates or other factors.The detection limit values as well as additional information on adjusting detection limits for different methodologies is provided in the Supplement in Sect.S3.1.2c.

Concentration gradient precision (gradient detection limit)
As previously described, the concentration gradient precision, which can also be defined as the gradient detection limit is the standard deviation of the residuals of the orthogonal least squares regression of SB1 (y) vs. SB2 (x) following Wolff et al. (2010).Scatter plots of SB1 vs. SB2 concentrations measured during three colocation experiments in June-July, August and October 2012 are included in Fig. S3 in the Supplement.The three colocation experiments consist of approximately 89, 138 and 73 hourly observations, respectively.
Results of the orthogonal least squares analysis by colocation period are summarized in Table S6 provided in the Supplement.Slopes are within ±0.06 of unity with the exception of NO − 3 and HNO 3 during the first period, in which SB1 was lower than SB2 by ≈ 15 %.The reason for this underestimation is not obvious.A low bias of SB1 relative to SB2 for both NH 3 and HNO 3 may indicate an effect of inlet tubing condition.Routine visual observation of the SB inlet tubing indicates that SB1, which is positioned closed to the ground, tends to accumulate dust more rapidly than SB2.As the colocation experiments are meant to serve as a QC measure for  a N is number of observations for all three colocation experiments.b C ave is average air concentration during co-location.c σ C is the standard deviation of the orthogonal least squares residuals (i.e., gradient detection limit).d σ C is the standard deviation of the air concentration.e C max is the maximum air concentration.f C min is the minimum air concentration.g σ C /C (%) is the gradient detection limit expressed as a percentage of the average air concentration.
fluxes measured during the period prior to colocation, inlets are replaced after, rather than before, colocation experiments.Thus, the bias observed during colocation period 1 may reflect a dirtier inlet on SB1.This would not, however, explain the bias for NO − 3 aerosol unless the loss of NH 3 and HNO 3 in the inlet promoted NH 4 NO 3 dissociation.
Results of the combined colocation experiments are summarized in Table 2.In general, concentrations during the three experiments were low, < 0.65 µg m −3 , with the exception of SO 2− 4 .The gradient detection limit, defined as the standard deviation of the residuals of the orthogonal least squares regression of SB1 vs. SB2 concentrations (σ C ), ranges from 0.02 µg m −3 for NO − 3 to 0.049 µg m −3 for SO 2 .The residual standard deviations determined in this study, which assume a Gaussian distribution, are considerably lower than the Gaussian standard deviations (i.e., gradient detection limits) determined by Wolff et al. (2010) for NH 3 , NH + 4 , HNO 3 and NO − 3 , which range from 0.093 µg m −3 for HNO 3 at a forest site to 0.44 µg m −3 for NO − 3 at a grassland site.Expressed as a percentage of the average concentration during co-location, the gradient detection limit (σ C /C ave ) ranges from 2.1 % for SO 2− 4 to 9.0 % for NH 3 .Unfortunately, a direct comparison of σ C expressed as a percentage of average concentration between this study and Wolff et al. (2010) study cannot be made as the average concentration during the co-location experiments is not reported by Wolff et al. (2010).
When comparing gradient detection limits, it is important to consider the relationship between concentration gradient precision and concentration.As discussed by Wolff et al. (2010), for some species the standard deviation of the orthogonal least squares residuals tends to increase with air concentration.Thus, the gradient detection limit varies with air concentration.The relationship between gradient detection limit and air concentration observed during our experiments is provided in Fig. S4.For this analysis, orthogonal least squares residuals were combined for the three coloca-tion experiments and sorted into seven bins defined by air concentration (see Fig. S5).Within each bin, which individually contained ≈ 42 observations, the standard deviation of the residuals and corresponding bin average concentration were calculated.With the exception of NO − 3 , all species show an increase in gradient precision with increasing concentration.In most cases, it appears that relationship between gradient precision and concentration weakens as concentrations increase.Consistent with Wolff et al. (2010), our results suggest that for some compounds, and most likely including NO − 3 , the relationship between precision and air concentration should be considered when calculating gradient and flux detection limits at the hourly timescale.The lack of relationship observed for NO − 3 may be due to a relatively narrow range of low concentration observed during the colocation experiments.It is likely that this precision/concentration relationship is a general feature of the measurement system and would likely be present over a larger range of NO − 3 concentrations.Similar to Wolff et al. (2010), empirical functions relating gradient precision and concentration were used in this study.These were derived using the plot between bin residual standard deviation and concentration (see Fig. S4) to predict a gradient detection limit for each hourly observation based on corresponding air concentration.The relationship between gradient precision and concentration was determined using regression and are presented in Fig. S4 in the Supplement.In this study, median relative gradient detection limit uncertainty (σ C / C) was 19.6 % for NH 3 , 90.4 % for NH + 4 , 29.6 % for HNO 3 and 29.2 % for NO − 3 .These are all lower than the equivalent median relative gradient detection limit uncertainty values (σ C / C) reported by Wolff et al. (2010) at a grassland site, which were 36.3, 129.6, 40.1 and 49.4 % for NH 3 , NH + 4 , HNO 3 and NO − 3 , respectively.Thomas et al. (2009) used a different methodology to determine concentration gradient precision that is not comparable to the methodology used in this study.

Flux uncertainty
Of 504 possible hourly observations (during a period of 3 weeks), there were ≈ 445/380 gradient measurements and ≈ 410/360 flux measurements for gas/aerosol compounds, respectively.During this period, canopy height was approximately 1.2 m with a single-sided LAI of about 3.5 m 2 m −2 .Example time series of meteorology, air concentrations and fluxes are given in Fig. S6 in the Supplement.Summary statistics for select meteorological variables, air concentrations and fluxes are provided in Table S7.Hourly air concentrations of N compounds ranged from near 0 to ≈ 2.0 µg m −3 , with mean concentrations ranging from 0.3 µg m −3 for HNO 3 to 0.7 µg m −3 for NH + 4 .While HNO 3 , NH 3 and NO − 3 showed distinct diurnal patterns with midday peaks, NH + 4 did not.Relative to N compounds, SO 2 and SO 2− 4 exhibited a wider range of hourly concentrations from near 0 up to 8.8 and 4.3 µg m −3 , respectively.SO 2 displayed a distinct recurring diurnal pattern of peak concentration during the day while SO 2− 4 temporally correlated with NH + 4 .Average concentrations of SO 2 and SO 2− 4 were 0.5 and 1.9 µg m −3 , respectively.
Over the period of fluxes analyzed, air temperatures generally ranged from 5 to 10 • C at night to a maximum near 25 • C during the day.Wind speed and u * ranged from near 0 at night to daytime maxima of ≈ 1.5 to 2.0 and 0.25 to 0.3 m s −1 , respectively.Fluxes of all compounds followed the diurnal profile of friction velocity, with peak fluxes during the daytime and fluxes near 0 at night.With the exception of NH 3 , all fluxes on average were directed toward the grass canopy.NH 3 showed a distinct diurnal profile of emissions during the day and fluxes near 0 or slightly negative overnight.As previously mentioned, a companion paper focusing on the air-surface exchange processes of individual compounds over a longer sampling period is forthcoming.
As mentioned, the concentrations used to determine concentrations gradients and thus fluxes were adjusted for the systematic difference between concentration measurements during co-location sampling.An analysis of the influence of the co-location concentration adjustment on calculated fluxes using the 3 weeks of flux values presented in this manuscript as an example is provided in Sect.S3.2.2b in the Supplement.
Individually, percentages of hourly chemical concentration gradients larger than the corresponding gradient detection limit were 86, 42, 82, 72, 74 and 69 % for NH 3 , NH + 4 , HNO 3 , NO − 3 , SO 2 and SO 2− 4 , respectively.As expected, percentages were lowest for aerosol species, owing to their relatively low deposition velocities and correspondingly smaller gradients relative to gas phase species.The majority of concentration gradients exceeded the gradient detection limit.
Patterns of flux uncertainty are summarized in Figs. 1 and 2. Overall uncertainty in the transfer velocity (σ v tr ) was estimated as 0.0041 m s −1 (n = 660), which is applied as σ v tr /ν tr to estimate the hourly fractional or per-  during the day, these very large uncertainties characterize only a minor fraction of the overall flux to the ecosystem.
Total flux uncertainty is summarized in Fig. 2. When both day and night periods are considered, median total flux uncertainties are 31, 121, 42, 43, 67 and 56 % for NH 3 , NH + 4 , HNO 3 , NO − 3 , SO 2 and SO 2− 4 .Considering only concentration gradients above the gradient detection limit reduces the median uncertainties to 28, 69, 37, 41, 56 and 50 %, respectively.Flux uncertainties for N compounds are generally similar to those reported by Wolff et al. (2010).However, when comparing flux uncertainties between studies it should be acknowledged that the transfer velocity uncertainty will vary from site to site depending on meteorological conditions.Furthermore, the methodology for determining the transfer velocity uncertainty could be different, as it is between this study and the Wolff et al. (2010) study.When only daytime concentration gradients above the detection limit are considered, the uncertainties further reduce to 21, 64, 29, 31, 51 and 44 %.

Conclusions
This paper presents for the first time an assessment of the performance of a MARGA 2S instrument as applied for the measurement of air-surface exchange of N and S compounds.Analytical detection limits were low for all compounds ranging from 0.02 g m −3 for NH 3 to 0.064 g m −3 for SO 2− 4 .The NH + 4 , NO − 3 , HNO 3 , SO 2 and SO 2− 4 detection limits reported in this study are lower than those determined under field conditions in the Thomas et al. (2009) and Wolff et al. (2010) studies, both of which used the GRAE-GOR system.Three colocation experiments were conducted to determine concentration gradient precision.Concentration gradient precision ranged from 0.02 µg m −3 for NO − 3 to 0.049 µg m −3 for SO 2 .Chemical concentration gradients determined in this study compare favorably to those determined by Wolff et al. (2010).Over a period of 3 weeks in early fall 2012, we find that the majority of chemical gradients exceed the corresponding detection limit and are therefore distinct from 0. Over the range of meteorological conditions observed, median flux uncertainty ranges from ≈ 31 % for NH 3 to ≈ 121 % for NH + 4 .Flux uncertainties reported here for N compounds are similar to those of the GRAEGOR as reported by Wolff et al. (2010).
While the characteristics of the analytical system reported here should be generally applicable to the MARGA 2S, the assessment of gradient precision and flux uncertainty will vary to some extent for different meteorological and atmospheric chemical conditions, though not dramatically.Overall, we find that the flux uncertainties are similar to other wet chemical methods and that the instrument is sufficiently precise for flux gradient applications.It is recommended that colocation experiments be conducted regularly for long-term deployments (e.g., monthly) or for each short-term intensive deployment to properly account not only for any short-term systematic bias that may develop between the two sample boxes but also to assess the relationship between concentration gradient precision and concentration.A companion paper focusing on the air-surface exchange processes of individual compounds over a longer period of study at our site is forthcoming.
The Supplement related to this article is available online at doi:10.5194/amt-9-2581-2016-supplement.
I. C. Rumsey and J. T. Walker: Application of an online ion-chromatography-based instrument 2.3 Analytical experiments

Figure 1 .
Figure 1.Diurnal profiles of uncertainty in chemical concentration gradients and transfer velocity expressed as a percentage of the corresponding concentration gradient ( C) and transfer velocity (ν tr ).Data points represent median value for the corresponding hour.

Figure 2 .
Figure 2. Summary of flux error (Eq.4) expressed as a median percentage of the flux magnitude.Data are summarized as all data, all fluxes in which the chemical gradient exceeds the gradient detection limit, all daytime data and daytime data in which the chemical gradient exceeds the gradient detection limit.Error bars represent interquartile range.

Table 1 .
Detection limit (liquid and air equivalent) results incorporating data from all four channels for each analyte.

Table 2 .
Summary of colocation results.