Simultaneous Detection of Ozone and Nitrogen Dioxide by Oxygen Anion Chemical Ionization Mass Spectrometry: A Fast Time Response Sensor Suitable for Eddy Covariance Measurements

We report on the development, characterization, and field deployment of a fast time response sensor for measuring ozone (O3) and nitrogen dioxide (NO2) concentrations utilizing chemical ionization time-of-flight mass spectrometry (CIToFMS) with oxygen anion (O ) reagent ion chemistry. We demonstrate that the oxygen anion chemical ionization mass spectrometer (Ox-CIMS) is highly sensitive to both O3 (180 ions s-1 pptv-1) and NO2 (97 ions s-1 pptv-1), corresponding to 10 detection limits (3σ, 1 s averages) of 13 and 9.9 pptv, respectively. In both cases, the detection threshold is limited by the magnitude and variability in the background determination. The short-term precision (1 s averages) is better than 0.3% at 10 ppbv O3 and 4% at 10 pptv NO2. We demonstrate that the sensitivity of the O3 measurement to fluctuations in ambient water vapor and carbon dioxide is negligible for typical conditions encountered in the troposphere. The application of the Ox-CIMS to the measurement of O3 vertical fluxes over the coastal ocean, via eddy covariance (EC), was tested during summer 2018 at 15 Scripps Pier, La Jolla CA. The observed mean ozone deposition velocity (vd(O3)) was 0.011 cm s-1 with a campaign ensemble limit of detection (LOD) of 0.0042 cm s-1 at the 95% confidence level, from each 27-minute sampling period LOD. The campaign mean and one standard deviation range of O3 mixing ratios were 38.9 ± 12.3 ppbv. Several fast ozone titration events from local NO emissions were sampled where unit conversion of O3 to NO2 was observed, highlighting instrument utility as a total odd oxygen (Ox = O3 + NO2) sensor. The demonstrated precision, sensitivity, and time resolution of this instrument 20 highlight its potential for direct measurements of O3 ocean–atmosphere and biosphere–atmosphere exchange from both stationary and mobile sampling platforms.


Introduction
The deposition of O3 to the ocean surface is a significant component of the tropospheric ozone budget. Global chemical transport model studies that explicitly treat O3 deposition, indicate that approximately one-third of total ozone dry deposition 25 is to water surfaces (Ganzeveld et al., 2009). However, the magnitude of total annual global ozone deposition to ocean surfaces is highly sensitive to the deposition velocity parameterization used, with model estimates ranging from 95 to 360 Tg yr -1 (Ganzeveld et al., 2009;Luhar et al., 2017). Several common global chemical transport models including GEOS-Chem (Bey et al., 2001), MOZART-4 (Emmons et al., 2010), and CAM-chem (Lamarque et al., 2012) apply a globally uniform deposition velocity (vd) that ranges between 0.01-0.05 cm s -1 depending on the model. In comparison to terrestrial measurements, where 2 O3 dry deposition velocities are relatively fast (>0.1 cm s -1 , (Zhang et al., 2003), there is a paucity of direct observations of ozone deposition to the ocean surface necessary to constrain atmospheric models. Previous studies of O3 deposition to water surfaces have been made from coastal towers (Gallagher et al., 2001), aircraft (Faloona et al., 2005;Kawa and Pearson, 1989;Lenschow et al., 1981), underway research vessels (Helmig et al., 2012), and in the laboratory (McKay et al., 1992), with observed vd(O3) ranging between 0.01 and 0.15 cm s -1 There is only one reported study of O3 deposition to freshwater, which 35 showed vd(O3) of 0.01 cm s -1 (Wesely et al., 1981).. Measured deposition rates to snow and ice vary widely, with most observations of vd(O3) from 0 to 0.2 cm s -1 , while models suggest vd(O3) from 0 to 0.01 cm s -1 . Reactions of O3 with iodide and dissolved organic compounds (DOC) in the ocean are known to play a controlling role in setting vd (O3) and may explain some of the variability in observations (Chang et al., 2004;Ganzeveld et al., 2009). However, these quantities have not typically been measured during field studies of vd(O3). To date there is no consensus on whether measured ocean O3 40 deposition velocities show a wind speed dependence . The most comprehensive dataset is from Helmig et al., (2012), which reported a deposition velocity range of 0.009 -0.034 cm s -1 from 1700 hours of observation over five research cruises. This dataset showed variability of vd(O3) with wind speed (U10) and sea-surface temperature (SST), highlighting the need for further field observations as constraints for model parameterizations.
The small magnitude of O3 ocean-atmosphere vertical fluxes presents a significant analytical challenge for existing 45 ozone sensors used in eddy covariance (EC) analyses. Driven in part by stringent sensor requirements for EC techniques, significant uncertainties in the magnitude and variability of ozone deposition to water surfaces remain. In contrast, O3 vertical fluxes to terrestrial surfaces are 10 to 100 times faster than to water surfaces, significantly loosening sensor precision requirements. Nonetheless, significant variability in vd(O3) exists between surface types (e.g. soil vs. leaf) (Wesely and Hicks, 2000). Terrestrial deposition velocities also show strong diel and seasonal variability due to factors such as stomatal opening 50 and within-canopy chemistry (Fares et al., 2010;Fowler et al., 2001;Kurpius and Goldstein, 2003). Highly accurate and precise measurements of O3 are required to correctly model the response of vd(O3) to each of these factors. While terrestrial and ocean exchange studies have substantial differences in experimental design, a sensor suitable for ocean-atmosphere ozone deposition measurements via EC is expected to be highly capable of biosphere-atmosphere measurements due to the significantly larger deposition rates and similar accuracy requirements. 55 Eddy covariance measurements typically require fast (1-10 Hz), high precision sensors in order to resolve covariance on the timescales of the fastest atmospheric turbulent eddies. Due to this constraint, standard O3 monitoring instruments which utilize UV-absorption detection do not have suitable time response or precision for EC measurements and ozone flux measurements have primarily utilized fast response chemiluminescence sensors. Chemiluminescence detectors can use either gas-phase, dry, or wet reagents for detection with important differences between them (Muller et al., 2010). Gas-phase 60 chemiluminescence sensors are typically based on the reaction of O3 with nitric oxide (NO) to form an excited state NO * which then relaxes to the ground state, emitting a photon that can be detected. This method has well understood reaction kinetics and allows for high sensitivity detection on the order of 2.8 counts s -1 pptv -1 (Bariteau et al., 2010;Pearson, 1990). A practical disadvantage to this technique is the necessity of a compressed cylinder of NO which is highly toxic. Wet chemiluminescence techniques are used less, as they exhibit generally lower sensitivity than dry chemiluminescence sensors and can be limited by issues in the liquid flow (Keronen et al., 2003).
Dry chemiluminescence sensors have the simplest operation and have seen the most regular use for EC studies (Güsten et al., 1992;Tuovinen et al., 2004). However, dry chemiluminescence sensor discs require conditioning with high ozone (up to 400 ppbv for several hours) before operation, are known to degrade over time, and have high variability in sensitivity between sensor discs (Weinheimer, 2007). These factors have led to limitations in long term stability and to 70 uncertainty in calibration factors for dry chemiluminescence sensors, resulting in uncertainty in the accuracy of the flux measurement (Muller et al., 2010). Muller et al.(2010), also reported a comparison of two identical co-located dry chemiluminescence sensors with half-hourly flux values differing by up to a factor of two and a mean hourly flux difference ranging from 0 to 23% between sensors. Recently Zahn et al., (2012) reported the development of a commercial dry chemiluminescence ozone detector capable of fast (>10 Hz) measurements with high sensitivity (∼9 counts s -1 pptv -1 ) suitable 75 for EC or mobile platform sampling. However, they also report issues of short-and long-term drift and variability between sensor discs. These accuracy and drift concerns have driven an interest in the development of a new, stable and fast ozone sensor suitable for EC measurements from both stationary and mobile sampling platforms.
In addition to the inherently small magnitude of vd(O3), the fast chemical titration of O3 by NO (R1) often complicates the interpretation of vd(O3) measurements. Surface emissions of NO result in a high bias in the measured deposition velocity 80 when the titration reaction (R1) is fast relative to the transport time to the height of the sensor.
Surface NO emissions from both biogenic and anthropogenic sources are widespread, with ocean emissions on the order of 1 × 10 8 molecules cm -2 s -1 (Zafiriou and McFarland, 1981) and soil emissions ranging from 5 × 10 9 to 2 × 10 11 molecules cm -2 s -1 (Yienger and Levy, 2004). These emissions correspond to a positive bias in the observed vd(O3) dry deposition rate on the 85 order of 5% in the marine atmosphere (discussed in section 3.7.1) and up to 50% in a forested site (Dorsey et al., 2004).
Simultaneous flux detection of O3 with one or both of NO or NO2 is commonly used to address this flux divergence problem (Finco et al., 2018;Stella et al., 2013). However, these studies typically require separate sensors for O3 and NOx which can introduce additional sources of uncertainty. Related challenges of fast O3 titration exists for quantification of O3 from mobile platforms where there is dynamic sampling of different airmasses with potentially differing O3-NO-NO2 steady-state 90 conditions.
In what follows, we describe the characterization and first field observations of a novel oxygen anion chemical ionization time-of-flight mass spectrometer (Ox-CIMS) sensor for O3 and NO2. Over the past two decades, chemical ionization mass spectrometry (CIMS) techniques have emerged as sensitive, selective, and accurate detection methods for a diverse suite of reactive trace gases (Huey, 2007). Successful application of CIMS for EC flux measurements have been demonstrated from 95 many sampling platforms including ground sites (Kim et al., 2014;Nguyen et al., 2015), aircraft , and underway research vessels (Blomquist et al., 2010;Kim et al., 2017;Yang et al., 2013) employing a variety of reagent ion chemistry systems. Here we demonstrate the suitability of the Ox-CIMS for EC flux measurements and provide detailed laboratory characterization of the instrument.

Chemical-ionization time-of-flight mass spectrometer
A complete description of the CI-ToFMS instrument (Aerodyne Research Inc., TOFWERK AG) can be found in Bertram et al. (2011). In what follows we highlight significant differences in the operation of the instrument from what is discussed in Bertram et al., (2011). Oxygen anions are generated by passing an 11:1 volumetric blend of Ultrahigh Purity (UHP) N2 and O2 gas (both Airgas 5.0 grade) through a polonium-210 α-particle source (NRD, P-2021 Ionizer). This N2:O2 volume ratio was 105 found empirically to maximize total reagent ion signal in our instrument while minimizing background signal at the O3 detection product (CO ,-60 m/Q). Further discussion of the reagent ion chemistry and precursor concentration can be found in sections 2.2 and 2.8. The reagent ion stream then mixes with ambient air in an ion-molecule reaction (IMR) chamber held at 95 mbar where product ions were generated. Further discussion of the dependence of instrument sensitivity on IMR pressure can be found in section 2.6. At this pressure, the residence time in the IMR is estimated to be on the order of 100 ms. Product 110 ions then pass into three differentially pumped chambers before reaching the ToF mass analyzer. Ions first move from the IMR to a collisional dissociation chamber (CDC) held at 2 mbar which houses a short-segmented RF-only quadrupole ion guide.
Field strengths in the IMR and CDC were tuned to be as soft as possible to preserve the transmission of weakly bound clusters while still maintaining acceptable total ion signals (ion optic potentials are listed in Table S1). Ions then sequentially pass into a second RF-only quadrupole chamber held at 1.4 × 10 -2 mbar and a final chamber containing focusing optics which prepare 115 the ion beam for entry into the compact ToF mass analyzer (CToF, TOFWERK AG and Aerodyne Research Inc.). The mass resolving power (M/ΔM) of the instrument as configured for these experiments was greater than 900 at -60 m/Q. All ion count rates reported here are for unit mass resolution integrated peak areas. In this work extraction frequencies of 75 kHz were used, resulting in mass spectra from 27-327 -m/Q. All mass spectra were saved at 10 Hz for analysis.

Oxygen Anion Chemistry 120
Oxygen anion (O ) reagent ion chemistry has been investigated previously for its use in the detection of nitric acid and more recently hydrogen peroxide (Huey, 1996;O'Sullivan et al., 2018;Vermeuel et al., 2019). Oxygen anion chemistry has also been used for chemical analysis of aerosol particles in a thermal desorption instrument, primarily for detection of particle sulfate and nitrate (Voisin et al., 2003). Oxygen anion chemistry has also been used for the detection of SO2 via a multi-step ionization process where CO reagent ions are first generated by the reaction of O with added excess O3 in the presence of 125 CO2. The CO reagent ion then ligand switches with SO2 to form SO which then quickly reacts with ambient O2 to form the primary detected SO product (Porter et al., 2018;Thornton et al., 2002a). Ionization of analytes by oxygen anion reagent ion chemistry proceeds through both charge transfer (R2) and adduct formation (R3).
It is expected that charge transfer from oxygen will occur to any analyte with an electron affinity (E.A.) greater than O2 (0.45 eV, (Ervin et al., 2003)) resulting in a relatively non-specific reagent ion chemistry (see Rienstra-Kiracofe et al., (2002) for a compilation of molecular E.A. values).. Adduct formation is observed when the binding enthalpy of the adduct is larger than that of the oxygen-water adduct and the adduct is stable enough to be preserved through the ion optics. This adduct formation framework is analogous to what has been shown for iodide reagent ion chemistry (Lee et al., 2014). 135 The O reagent ions present in the IMR are expected to have a series of attached water molecules at ambient humidity and the IMR pressure (95 mbar) and electric field strengths used in this study (Bork et al., 2011). The reagent ion is therefore reported as O (H O) for the remainder of this work. In the recorded mass spectra from our instrument, all reagent ion signal is observed as n = 0-1 (i.e., O and O (H O) ) as seen in Fig. 1. Oxygen anion-water clusters larger than n = 1 are likely present in the IMR but H2O evaporates off of the cluster in the CDC before detection due to the lower binding enthalpy of each additional 140 water in O (H O) (Bork et al., 2011) and the high filed strength at the exit of the CDC (Brophy and Farmer, 2016). Variability in the number of attached water molecules (n) as a function of humidity introduces the possibility of a water dependence on the ion chemistry, which is discussed further in Section 2.5.
The detection of ozone (O ) by oxygen anion reagent ion chemistry proceeds via a two-step reaction leading to the formation of a carbonate anion (CO ), which is the final detected product. First, the oxygen anion (O (H O) ) either transfers 145 an electron to ozone forming O (R4a) or forms a stable cluster with ozone (R4b). The ozone anion (either bare or as a cluster with O (H O ) ) then reacts with a neutral CO molecule to form CO (R5a-5b) which is the primary, detected product in the mass spectrometer. The electron affinity of O3 is 2.1 eV (Arnold et al., 1994). O but the magnitude of this signal is less than 1% of the signal of CO during ambient sampling. The proposed mechanism of CO formation is supported by a study using isotopically labelled oxygen to form labelled ozone anions ( O ) in a corona discharge source which then reacted with CO to form the detected product C OO (Ewing and Waltman, 2010). This product supports that a single oxygen is transferred from the ozone anion to carbon dioxide (as in R5a).
The detection of NO2 proceeds directly through a charge transfer reaction with O (H O) to form the detected NO 160 product (R6). This is expected based upon the high E.A. of NO2 (2.27 eV, (Ervin et al., 1988)) compared to O2 (E.A 0.45 eV).
Oxygen anions are expected to be a highly general reagent ion chemistry, showing sensitivity to an array of analytes.
While the focus of this work is on detection of O3 and NO2, detection of hydrogen peroxide, nitric acid, formic acid, sulfur dioxide and other species with the Ox-CIMS has demonstrated good performance (Vermeuel et al., 2019). An example ambient 165 mass spectrum recorded at 1 Hz sampling is shown in Fig. 1, with several major peaks highlighted. Also apparent are an abundance of peaks throughout the spectra with high signal intensity. During ambient observations, over one third of masses from -m/Q 27-327 showed signal intensity greater than 1 x 10 4 counts per second (cps). A larger survey and classification of oxygen anion reagent ion chemistry to utilize this versatility is underway.

Laboratory calibration 170
Laboratory calibrations of the Ox-CIMS were performed to determine instrument sensitivity to O3 and NO2. Ozone was generated by passing UHP Zero Air (ZA, Airgas 5.0 grade) through a mercury lamp UV source (Jelight Co, Irvine CA).
Outflow from the lamp source was diluted in UHP ZA and split between the Ox-CIMS and a factory calibrated 2B POM ozone monitor (2B Technologies) with an accuracy of ± 1.5 ppbv, which served as our reference standard. Ozone concentrations were varied over the range 0-80 ppbv and instrument response was determined to generate a calibration curve. NO2 was 175 delivered from a certified standard cylinder (Scott-Marrin 4.84 ± 0.1 ppmv). The primary NO2 standard was diluted in UHP ZA to span the range of 0-10 ppbv. Dilutions of calibration standards were made in UHP ZA which was humidified to the desired amount by splitting a portion of the flow through a bubbler containing 18 MΩ water. CO2 (Airgas Bone Dry grade) was added to the dilution flow to maintain mixing ratios of 380 ppmv for all calibrations (See Section 2.6). A Vaisala HMP 110 sensor continuously measured relative humidity and temperature inline downstream of the Ox-CIMS and POM inlets. All 180 flows were controlled by mass flow controllers (MKS instruments, 1179C series) with an estimated total uncertainty of 10%.
Example calibration curves for O3 and NO2 are shown in Fig. 2. An overview of instrument sensitivity, limits of detection (LOD), and precision to O3 and NO2 is given in Table 1.

Absolute sensitivity
The absolute sensitivity of the Ox-CIMS for detection of analytes is controlled by the kinetics and thermodynamics of the 185 reagent ion chemistry and the total ion generation and transmission efficiency of the instrument. Under the operational configuration described in Section 2.1, typical reagent ion signal (O + O (H O) ) ranged from 0.8 to 2.2 × 10 7 counts s -1 (Fig. S1). The mean total reagent ion signal over 6 weeks of ambient sampling (Section 3.1) was 1.45 × 10 7 cps. The absolute instrument sensitivity at this reagent ion signal to O3 and NO2 is 180 and 97 cps pptv -1 respectively (at 8 g kg -1 SH). Total instrument count rate is a complex function of instrument design, instrument ion optics tuning, Po-210 source decay, micro Conversely, the reagent ion charge transfer or adduct formation chemistry for a given analyte sets a fundamental limit on sensitivity for a given instrument configuration. Sensitivity values can be normalized by scaling all signals to a fixed total reagent ion signal of 1 × 10 6 cps to isolate the sensitivity component controlled by reagent ion chemistry, separate from changes in instrument performance due to decay in the ion source or other factors. The total reagent ion signal is taken as the 195 sum of the O and O (H O) signals. Sensitivity values through the remainder of the text are reported as either absolute sensitivities in counts per second (cps pptv -1 ) or normalized sensitivities in normalized counts per second (ncps pptv -1 ).
Absolute sensitivity values control instrument limits of detection (LOD) and precision, while normalized sensitivities are used for comparison of calibration factors.

Dependence of instrument sensitivity on specific humidity 200
The dependence of instrument sensitivity on ambient water content was assessed for specific humidity (SH) ranging between 0-16 g kg -1 (approximately 0-80% RH at 25 °C) by triplicate calibrations as shown in Fig. 3. Sensitivity to O3 had no significant dependence on specific humidity over the range 4-16 g kg -. Sensitivity to NO2 has a specific humidity dependence over the range 4-16 g kg -1 , decreasing from 7.9 to 4.6 ncps pptv -1 . A 30% and 45% decline in sensitivity was observed from 0 to 4 g kg -1 for O3 and NO2 respectively. This low humidity range is rarely sampled in the boundary layer over water surfaces but may 205 be significant in some terrestrial or airborne deployments and would require careful calibration. The SH range from 8 to 16 g kg -1 corresponds to approximately 40 to 80% RH at 25 °C which is typical of the humidity range over mid-latitude oceans (Liu et al., 1991).

Dependence on CO2 215
The ionization pathway for detection of O3 with O (H O) reagent ion chemistry differs from typical chemical ionization schemes, in that it involves a two-step reaction of charge transfer to ozone forming O , which then reacts with CO to form the detected CO product (R4-R5). Therefore, we assessed the impact of CO2 mixing ratio in the sample flow on O3 sensitivity as shown in Fig. 4. Calibration curves were generated by diluting ozone in dry UHP N2 and mixing in a flow of variable CO2 (Airgas, Bone Dry Grade) mixing ratios before sampling. At nominally 0 ppmv CO2, the O ionization product (-48 m/Q) was 220 detected with sensitivity of 14 ± 2 ncps pptv -1 and the CO product (-60 m/Q) at 5 ± 1 ncps pptv -1 . For CO2 mixing ratios from within the uncertainty. The presence of a significant fraction (36%) of the CO product with nominally 0 ppmv CO2 suggests the presence of a slight leak rate of CO2 via diffusion through the perfluoroalkoxy alkane (PFA) tubing, or CO2 contamination in the UHP N2 supply. The manufacturer stated upper limit of CO2 in the UHP N2 is 1 ppmv which we take to be the lower 225 limit achievable in our system. A CO2 mixing ratio of only 1 ppmv is still an order of magnitude excess relative to a high end ambient O3 mixing ratio of 100 ppbv. An exponential fit of the O product vs CO2 indicates that O makes up less than 1% of the detected ozone at CO2 mixing ratios greater than 10 ppmv. This suggests ambient samples will always have a substantial excess of CO2 necessary to drive the reaction completely to the CO product. The measured flat response from 60-500 ppmv CO2 indicates that natural variability in ambient CO2 will have negligible impact on ambient measurements of ozone. No other 230 analytes that we have calibrated for with the Ox-CIMS (HCOOH, HNO3, H2O2) have shown a CO2 mixing ratio dependence, suggesting that CO2 may be uniquely involved in the detection of O3 and is not a general feature of the oxygen-anion chemistry.
All other reported laboratory calibrations reported here were performed at CO2 mixing ratios of 380 ppmv and all reported sensitivities are for the CO product. This CO2 dependence also requires careful consideration during instrument background determinations by UHP N2 overflow which is discussed in Section 2.8. 235

Dependence on IMR pressure
Instrument sensitivity to O3 increases with increasing IMR pressure as shown in Fig. 5. The normalized signal of O3 increases by 60% at an IMR pressure of 95 mbar compared to 70 mbar when sampling a constant O3 source of 35 ppbv. IMR pressure was increased in approximately 5 mbar steps, with CDC pressure held constant at 2 mbar, and a three-minute dwell time at each step to ensure signal and pressure were stabilized. Total reagent ion signal did not change significantly over this pressure 240 range. Pressures above 95 mbar were not investigated due to concerns over corresponding increases in CDC pressure with the pinhole and pumping configuration used in this work. There is no evident plateauing in the signal increase over the IMR pressure range investigated here, indicating that further optimization is likely possible by operating at higher IMR pressures.
The increase in sensitivity with IMR pressure could be fit well with an exponential least squares fit, which is plotted in Fig.5.
The physical meaning of the exponential relationship is not clear. The source of the response of sensitivity to pressure is not 245 definitive but can possibly be attributed to the increase in the total number of collisions during the 100 ms residence time in the IMR and the corresponding weakening of those collisions. Higher collisional frequencies also lead to proportionally weaker collisions which could better preserve a weakly bound O (O )(H O) cluster and allow a longer lifetime to react with CO2 before dissociation. The operational IMR pressure of 95 mbar used here was empirically selected to maximize sensitivity to O3 without increasing CDC pressure beyond the desired range. Investigation of higher IMR pressures, up to the operation of 250 an atmospheric pressure interface, has the potential to further increase the instrument sensitivity to O3.

Instrument background and limits of detection
Instrument backgrounds were assessed by periodically overflowing the inlet with UHP N2 during field sampling. Details of the inlet and zeroing conditions used are discussed further in Section 3.1. During N2 overflow, O3 displayed a consistently elevated background on the order of 3.1 x 10 5 cps corresponding to 2.1 x 10 4 ncps, or approximately 1.3 ppbv O3, at a typical 255 total reagent ion signal of 1.45 x 10 7 cps. A representative background determination is shown in Fig. 6. The magnitude of the O3 background was observed to vary with the O2:N2 ratio in the reagent ion precursor flow when sampling a UHP ZA overflow with 380ppm CO2 as shown in Fig. S2. The background O3 count rate was observed to increase from 3.0 x 10 4 to 6.3 x 10 4 ncps as the O2 volume fraction in the reagent ion delivery gas flow (fO2) was increased from 0.05 to 0.4. The dependence of the background O3 signal on suggests that the observed background O3 is formed directly in the alpha ion source and is not 260 from off-gassing of inlet and instrument surfaces. The magnitude of this background O3 does not vary when sampling UHP zero air or N2, further confirming that the background O3 is formed directly in the ion source from the O2 used to generate the reagent ion. An operational of 0.08 (actual volumetric flow ratio O2:N2 of 200:2200 sccm) was selected to balance maximizing the total reagent ion signal while minimizing the O3 ion-source background (3.1 x 10 5 cps). The magnitude of this O3 background was observed to be highly consistent during field sampling at a constant of 0.08 and well resolved from all 265 ambient observations (Fig. S3). The 1σ deviation of the distribution of normalized adjacent differences of O3 signal during background periods gives an upper limit of variability of 9% between adjacent background periods. A variability of 9% corresponds to a difference of 70 pptv between subsequent O3 background determinations. The magnitude of this O3 background is a fundamental limit on the achievable limit of detection.
Because CO2 was not added to the UHP N2 overflow during field sampling, the reaction was not driven fully to the 270 CO product and some O signal at -m/Q 48 was observed during UHP N2 overflow periods as shown in Fig. S4. The magnitude of the O3 signal observed as O was approximately 55% of the CO product (mean 1.2 x 10 4 and 9.6 x 10 3 ncps respectively) during overflow periods. The total sensitivity to O3 as the sum of the O and CO was observed to be constant as a function of CO2 as shown in Fig. 4. We therefore assign equal sensitivity to each O3 detection product and took the sum of signal at O and CO in order to determine the total background O3 concentration. This issue will be corrected in future 275 deployments by the addition of CO2 to the N2 overflow used for backgrounds which will drive the product fully to CO . The mean background of O3 for the full field sampling period was 1.3 ± 0.3 ppbv. The 10 Hz precision of O3 during an individual N2 overflow period was found to be 0.75%, corresponding to 7.5 pptv as shown in Fig. S5. This suggests that variability in the O3 signal from this background source is constant over short timescales and has a negligible impact on instrument precision during ambient sampling. 280 The 10 Hz limit of detection for O3 is 42 pptv for a S/N of 3, and a mean background O3 signal of 2.1 x 10 4 ncps as calculated using Eq. 1, below from Bertram et al., 2011, where Cf is the calibration factor, [x] is the analyte mixing ratio, t is averaging time in seconds, and B is the background count rate. The optimum LOD from the minimum of the Allan variance at an 11 second averaging time is 4.0 pptv (Fig. S6a).

E1 285
The mean background signal during field sampling for NO2 was 3.5 x 10 3 ncps which corresponds to 0.28 ppbv. At this background level, the 10 Hz LOD for NO2 is 26 pptv for a S/N of 3. The optimum LOD for NO2 is 2.3 pptv at an averaging time of 19 seconds, determined from the minimum of the Allan variance (Fig. S6b). The background signal of NO2 is notably above zero indicating either off gassing from inlet walls or a secondary production of NO2 in the instrument. A possible source of this background is from degradation of other species such as nitric acid or alkyl nitrates on the inlet walls. Additional 290 calibration will be necessary to ensure that observed NO2 signal is not a secondary product of other species and we can currently quantify their potential interference on measured NO2.

Reagent ion saturation and secondary ion chemistry
During ambient sampling the ozone signal (as CO detected at -60 m/Q) is of comparable magnitude to the O reagent ion signal as shown in Fig. 1. High analyte concentrations (> 5 ppbv) have been shown previously to result in non-linear calibration 295 curves for unnormalized signals (Bertram et al., 2011;Veres et al., 2008). In our system we do not observe non-linearity in the normalized O3 calibration for our highest concentration calibration point of 80 ppbv despite the CO signal being larger than the O reagent ion (9 x 10 6 cps and 6 x 10 6 cps respectively). The electron affinity (E.A.) of carbonate is from 3.26 (Hunton et al., 1985) to >3.34 eV (Snodgrass et al., 1990) and is significantly higher than that of oxygen (E.A. 0.45 eV), making it unlikely that carbonate is involved in charge transfer reactions when excess O is present. . At high O3 concentrations, the 300 reagent ion signal magnitude is reduced, which necessitates normalizing sensitivities to the 1 x 10 6 cps of reagent ion signal before quantification. For NO2 (E.A. 2.27 eV), the normalized sensitivity showed no dependence on O3 concentrations from 0 to 80 ppbv. Carbonate reagent ion chemistry has been utilized for detection of HNO3 and H2O2 via adduct formation raising additional concern about potential secondary ion chemistry (Reiner et al., 1998). In laboratory calibrations, shown in Fig. S7, introduction of 0 to 40 ppb H2O2 resulted in the titration of the O3 signal of 0.06 ppbv per ppbv H2O2. H2O2 was detected as an 305 adduct with O and not CO , indicating that O reagent ion chemistry is more favorable despite high CO signal intensity.
The Ox-CIMS O3 measurement also compared well (R 2 =0.99) against an EPA AQS O3 monitor over 1 month of ambient sampling where H2O2 and HNO3 concentrations both exceeded 5 ppbv at times (see Section 3.1 for further discussion of field intercomparison), further supporting the CO detection product as a robust indicator of O3 in complex sampling environments.
Ab initio calculations of the binding enthalpies of O and CO reagent ions with H2O, HNO3, H2O2, and CH3OOH 310 were performed with the MP2/aug-cc-pvdz-PP theory and basis set in order to assess the relative favorability of adduct formation between O and CO Adduct formation with O was favorable relative to CO by 2.5 to 17 kcal mol -1 for all analytes that were calculated. All calculated binding enthalpy values are listed in Table S2.

Short-and long-term precision
Short term precision of the instrument was assessed by calculating the normalized difference between adjacent 10 Hz data 315 points over a 27-minute sampling period of a constant ambient analyte concentration via Eq. 2.

E2
The standard deviation of the Gaussian fit of the distribution of normalized adjacent differences (NAD) is a direct measure of the short-term instrument precision (Bertram et al., 2011). The 1σ precision from the NAD distribution for 10 Hz sampling of 38 ppbv ozone is 0.74% (Fig. 7). The 10 Hz precision for sampling of 2.3 ppbv NO2 is 1.1% The short-term precision for both 320 analytes was larger than expected if the noise was driven by counting noise alone (10 Hz counting noise limit for O3 and NO2 at the concentrations used above are 0.12% and 0.63% respectively), indicating that other potential points of optimization in the instrument configuration are required to further improve short-term precision. Notably, the observed noise source appears to be white noise given the Gaussian distribution of the NAD (Thornton et al., 2002b).

325
Short term precision was assessed as a function of count rate by calculating the NAD for all masses in the spectrum over a stable 27-minute sampling period for both 1 Hz and 10 Hz data averaging. From this assessment, precision was observed to improve approximately linearly in a log-log scaling for count rates between 1 x 10 3 and 1 x 10 6 cps (Fig. S8) as expected in the case where counting noise drives instrument precision. Above 1 x 10 6 cps there is an apparent asymptote where precision no longer improves with count rate. The counting noise limited 10 Hz precision for 10 6 and 10 7 cps are 0.32% and 0.1% 330 respectively, while the measured values were 0.75 and 2%. The counting noise limited precision is calculated as √ / where N is the number of counts during the integration time. This precision limit could be driven by an uncharacterized source of white noise in the instrument, including MFC drift, IMR turbulence, ion optic voltage drift, and pump drift. Measurement precision of O3 and NO2 could be improved by a factor of 5 and 2 respectively if this non-counting noise source of white noise was eliminated. 335 In theory, detection limits can be improved by signal averaging to a lower time resolution than the 10 Hz save rate.
Signal-to-noise ratios are expected to improve with the square root of the integration time. At longer timescales, factors including instrument drift become significant, creating a limit on the upper end of averaging time which optimizes signal-tonoise. This was assessed quantitatively by calculation of the Allan variance as shown in Fig. S6 (Werle et al., 1993).
3 Field results and discussion 340

Ozone field calibration and intercomparison
Performance of the Ox-CIMS was compared against a co-located EPA Air Quality System (AQS) O3 monitor (Thermo-Fisher 49i, AQS ID 17-097-1007) over one month of ambient sampling during the Lake Michigan Ozone Study 2017 (LMOS 2017) in Zion, IL (Vermeuel et al., 2019). A regression analysis between the two instruments at one-minute averaging showed strong agreement (R 2 = 0.99) as shown in Fig. 8. Ox-CIMS concentrations were averaged to 1 ppbv bins which was the output data 345 resolution of the EPA data logger system for the (Thermo-Fisher 49i). Error bars are the 1σ standard deviation of each Ox-CIMS bin average. Near one-to-one agreement (slope of 0.99) between instruments lends confidence to the calibration, baselining, and long-term stability of the Ox-CIMS. The Ox-CIMS was located on the roof of a trailer (approx. 5 m above ground) and sampled through a 0.7 m long, 0.925 cm i.d., PFA inlet. The inlet was pumped at flow rate of 18-20 slpm from which the Ox-CIMS subsampled at 1.5 slpm. Temperature and RH were recorded inline downstream of the subsampling point. 350 The Ox-CIMS sampling point was approximately 10 m horizontally from the Thermo-Fisher 49i and both instruments sampled at approximately equal heights. Instrument backgrounds of the Ox-CIMS were determined every 70 minutes by overflowing the inlet with dry UHP N2. Calibration factors were determined by the in-field continuous addition of a C-13 isotopically labelled formic acid standard to the tip of the inlet. Laboratory calibrations of the Ox-CIMS to formic acid and O3 as a function of specific humidity were determined immediately pre-and post-campaign and were used to calculate a humidity dependent 355 sensitivity of O3 relative to formic acid. That relative sensitivity was then used to determine the in-field sensitivity to O3 by scaling field sensitivities of formic acid from the continuous additions. Full details of this deployment and calibration methods are described in Vermeuel et al., (2019). The EPA O3 monitor shows a persistent high bias at low O3 concentrations (<10 ppbv) relative to the Ox-CIMS. This discrepancy could arise from known interferences from water, mercury, and other species in 254 nm UV absorbance detection of ozone (Kleindienst et al., 1993). 360 Instrument sensitivity was assessed by the standard addition of a C-13 isotopically labelled formic acid standard for 3 minutes every 35 minutes at the ambient end of the inlet manifold. Ozone mixing ratios were determined by scaling the humidity 380 dependent sensitivity of O3 from pre-and post-campaign calibrations to the field calibrations of C-13 formic acid. Ambient O3 was also measured at 10 s time-resolution with a 2B technologies Personal Ozone Monitor (POM). The POM had a separate 10 m long, 0.47 cm i.d. PFA sampling line located 12 m from the Ox-CIMS inlet manifold and sonic anemometer. The POM was used as an independent verification of the Ox-CIMS measurement and was not used for calibration.

Backgrounds and inlet residence time 385
Instrument backgrounds were determined every 35 minutes by overflowing the entire inlet manifold with dry UHP N2.
Background and ambient count rates were first converted to concentrations using the laboratory determined humidity dependent sensitivities for O3 and NO2 scaled to the C-13 formic acid standard addition sensitivity. Background concentrations of O3 and NO2 from before and after each 30-minute ambient sampling period were interpolated over the ambient sampling period which was then subtracted from each 10 Hz concentration data point to obtain a background corrected time-series. 390 Background concentrations of O3 had a mean 1.5 ppbv and a drift of 1% between adjacent background periods, determined by the distribution of the NAD of the mean background concentrations.
The signal response of O3 during dry N2 overflows were fit to an exponential decay function to characterize inlet gas response times (Ellis et al., 2010). Best fit estimates for decay time constants for O3 across overflow periods were from 0.2 to 0.44 seconds. NO2 decay responses were fit to a bi-exponential decay to characterize inlet evacuation time ( ) and wall 395 interaction times ( ) (Ellis et al., 2010). for NO2 was determined to be approximately 3.2 s. This suggests a potential interference at the NO2 peak, as NO2 is expected to have minimal wall equilibration, similar to O3. NO2 also shows a continually elevated signal during overflow periods suggesting off gassing from inlet or instrument surfaces. The cause of this slow NO2 decay and elevated background is not clear but could be from degradation of nitric acid or nitrate containing aerosol on the instrument surfaces. 400 The instrument response time (τr) for O3 can be calculated during zeroing periods as the time required for the signal to fall to 1/e of its initial value. The response time of the instrument was calculated for each overflow period during field sampling, with a mean value of 0.28 s. The cutoff frequency (fcut) of the instrument is defined as the frequency where the signal is attenuated by a factor of 1/√2 (Bariteau et al., 2010). The cutoff frequency can also be calculated from τr according to Eq.

405
= E3 The calculated fcut from the measured mean response time was 0.57 Hz. This value suggests that minimal attenuation in the flux signal (cospectra) should be apparent at frequencies less than 0.57 Hz. The instrument response time and thus cutoff frequency are function of the flow rate and sampling line volume. The flow rate of 18-23 slpm was the maximum achievable with the tubing and pumping configuration used here but could be improved in future to minimize tubing interactions and shift 410 fcut towards higher frequencies.

Eddy covariance flux method
The transfer of trace gases across the air−sea interface is a complex function of both atmospheric and oceanic processes, where gas exchange is controlled by turbulence in the atmospheric and water boundary layers, molecular diffusion in the interfacial regions surrounding the air−water interface, and the solubility and chemical reactivity of the gas in the molecular sublayer. 415 The flux (F) of trace gas across the interface is described by Eq. 4, as a function of both the gas-phase (Cg) and liquid phase (Cl) concentrations and the dimensionless gas over liquid Henry's law constant (H), where Kt, the total transfer velocity for the gas (with units cm s -1 ), encompasses all of the chemical and physical processes that govern air−sea gas exchange. Surface chemical reactivity terms to the gas exchange rate are incorporated into the Kt term. For purely depositing species where the water side concentration is negligible, Cl and can be neglected in Eq. 4 and Kt can be reformulated into a deposition velocity (vd) calculated according to Eq. 6, where is the mean gas phase mixing ratio during the flux averaging period. A summary of concentration and flux results for the full deployment period are given in Table 2. 430

General Data Corrections
Several standard eddy covariance data filters and quality control checks were applied before analysis. General filters included: 1.) Wind sector: Only periods of mean onshore flow (true wind direction 200-360°) were used.
2.) Friction velocity: A friction velocity ( * ) threshold was applied to reject periods of low shear driven turbulence (Barr et al., 2013) described further below. 435 3.) Stationarity: Each 27-minute flux period was divided into five even non-overlapping subperiods. Flux periods were rejected if any of the subperiods differed by more than 40% (Foken and Wichura, 1996).
The applied * filter was determined by comparing the observed U* values to U* calculated with the NOAA COARE bulk flux v 3.6 algorithm (Fairall et al., 2011). COARE U* were calculated using measured meterology including wind speed, sea-surface temperature, air temperature, and relative humidity. Flux periods were rejected if the observed U* differed from the 440 calculated U* by more than 50%. The stress relationship of wind-speed to U* is well understood over the ocean. Fixed U* filters of ca. 0.2 m s -1 are used frequently as a default in terrestrial flux studies but would reject nearly all observation periods in this study. The observed friction velocities are consistent with other marine flux studies where surface roughness lengths are significantly smaller than over terrestrial surfaces (Porter et al., 2018). Methods of determining site specific * thresholds typically require long-term data series which were not available here (Papale et al., 2006). Papale et al., (2006), 445 applied a minimum U* threshold of 0.1 m s -1 for forest sites and 0.01 m s -1 for short vegetation sites where typical U* values are lower. The selected * filter rejects an additional 44% of the flux periods remaining after the wind direction filter.. The stationarity criteria rejected a further 100 flux periods, potentially driven by periods of activity on the pier driving changes in the sampled O3. Outliers in vd(O3) and the flux limit of detection were determined and removed for points three scaled median absolute deviations from the median. This outlier filter removed an additional 16 data points. After the wind 450 direction filter and all quality control filters were applied, 73% of flux periods were rejected leaving 246 quality-controlled flux periods. Eddy covariance flux values were calculated using 27-minute time windows. The O3 timeseries was detrended with a linear function prior to the flux calculation. The O3 and vertical wind data were despiked using a mean absolute deviation filter before the eddy covariance flux calculation following Mauder et al., (2013).

Planar Fit Wind Coordinate Rotation 455
Coordinate rotation of the u, v, and w wind components was performed by the planar fit method to remove unintentional tilts in the sonic mounting and account for local flow distortions (Wilczak et al., 2001). Briefly, the mean u, v, and w wind components and the stress tensor were determined for each 15-minute onshore flow period during the full campaign. A linear regression was used to find the best fit of a plane with a coordinate system where the z-axis is perpendicular to the mean streamline. Individual 27-minute flux periods are then rotated such that the x-axis is along the mean wind and ̅ = 0. Vertical 460 wind velocity ( ) in any individual rotation period may be non-zero due to mesoscale motions but for the full campaign is zero. The residual mean vertical velocity in any individual rotation period is subtracted out, so it does not impact the Reynolds averaging.

Lag time shift
The Ox-CIMS signal is delayed relative to the sonic anemometer due to transit time in the inlet line which must be accounted 465 for before calculating the covariance between the vertical wind and analyte concentration. The cross-covariance of the two timeseries were first calculated within a ± 5 s window to determine the lag time of the Ox-CIMS and synchronize with the anemometer. The volumetric evacuation time of the inlet is 1.65 to 2.1 s for the inlet volume and flow rates of 18 to 23 slpm used in this study. Following the method and terminology outlined in Langford et al., (2015), the position of the maximum (MAX) of the cross-covariance is taken as the lag time needed to align the vertical wind and analyte concentration for that flux In low signal-to-noise (SNR) data, the use of the MAX leads to high variability in the determined lag time caused by uncertainty in the position of the peak in the cross-covariance. This results in a systematic high bias on the absolute magnitude of the resulting flux. The position of the maximum of a centered running median (AVG) function of the cross-covariance is an alternative method to determine lag time with less expected bias for low SNR data (Langford et al., 2015;Taipale et al., 2010). 475 Lag times for each O3 flux period determined by the MAX and a 10 point AVG method showed reasonable agreement, with a campaign average lag time from the MAX with a mean of 1.0 seconds and the AVG at 0.7 seconds (Fig. S9). This agreement suggests that a clear peak in the cross-covariance was present for most flux periods leading to a convergence of the two methods. This lag time also shows agreement with the inlet response time of 0.9 s determined during dry N2 overflows. Due to the convergence of the determined lag times around a central value, a prescribed lag time of 0.9 s was used for all reported 480 vd(O3) values. A prescribed lag time has the least bias to extreme values caused by noise, provided that the true lag time is known well (Langford et al., 2015). Deposition velocities were then recalculated with the prescribed lag time of 0.9 s and with the MAX and AVG method over a narrower lag window of ± 3 which is expected to be physically reasonable range for the flow rate and inlet line volume. The mean vd(O3) using the prescribed, MAX, and AVG lag times were 0.013, 0.012, and 0.012 cm s -1 respectively, suggesting the campaign mean value was relatively insensitive to the lag time method. 485

Cospectra and Ogives
The frequency weighted cospectrum of O3' with w' has a well characterized form with exhibited dependence on wind-speed and measuring height (Kaimal et al., 1972). Comparison of observed cospectra shape against the idealized Kaimal cospectra is useful to validate that the observed signal was not significantly attenuated at low or high frequencies. Cospectral averaging is performed by binning frequency into 50 evenly log spaced bins and normalizing the integrated cospectra to 1. The integral 490 of the unnormalized cospectra is the flux for that observation period. The mean wind-speed binned cospectra of sensible heat and O3 appear to match well with the idealized Kaimal cospectra for an unstable boundary layer at sampling height z = 13m (Fig.11).
The ogive is the normalized cumulative distribution of the cospectra, which is used to validate both that no high-frequency 495 attenuation is present and that the flux averaging time is sufficiently long that all frequencies contributing for the flux is captured. Figure 11 shows the averaged cospectra and ogives for O3 and sensible heat flux from the average of two flux averaging periods 14:10 -15:20 on July 20 th . The asymptote to 1 at low frequencies validates that the 27-minute flux averaging time was sufficiently long for this site to capture the largest flux carrying eddies. High pumping rates in sampling line ensured that turbulent flow was always maintained in the line (Reynolds number 3860-4940). Higher Reynolds numbers in the turbulent 500 regime lead to smaller high frequency attenuation (Massman, 1991). The overlap of the idealized Kaimal curve and the observed sensible heat and O3 ogives suggest that high frequency attenuation in the sampling line is minimal above approximately 0.4 Hz, consistent with our calculated fcut of 0.57 Hz. We calculate the high frequency correction transfer function for turbulent attenuation in a tube from Massman, (1991) as a constraint, which is shown in Fig. 11b. This transfer function shows attenuation primarily above 1 Hz and is not sufficient to describe the observed attenuation above 0.4 Hz. This 505 implies that the attenuation observed cannot be explained only as turbulent smearing in the inlet and that other wall interactions are likely present.
Due to the small magnitude of the O3 EC flux there is low signal to noise in the cospectra at high frequency for many of the flux averaging periods. This makes application of cospectra based correction factors challenging and likely to introduce added variance on the signal. We therefore apply an attenuation correction factor following Bariteau et al., (2010). First we calculate 510 the idealized unattenuated Kaimal cospectra (Cwx_k) (Eq. 7) for each flux period, and then apply frequency attenuation to that cospectra by applying a low-pass filter function (H(f)) characterized by τc (Eq. 8). The ratio of the flux of the unattenuated (Fraw) and attenuated (Fatt) cospectra is then taken as the correction factor (Af, Eq.9) to apply to the observed O3 flux (Eq. 10).
Where n is the surface layer normalized frequency defined as n = f z/U, where z is the measurement height and U is the horizontal wind speed.
, is the attenuation corrected O3 flux and , is the original measured flux. This approach has 515 the benefit of applying a single correction factor to the total flux, rather than frequency dependent corrections which might serve to amplify noise at high frequencies. The net impact of this correction factor was an increase of campaign mean vd(O3) of 4%.

E10.
We also calculate the attenuated flux from the model of Horst, (1997) shown in Eq. 11, for a response time ( ) of 0.0.28 s, 525 and a wind speed of 3 m s -1 to be 6%. The general agreement of the Horst and Bariteau attenuation correction factors indicate that the applied correction is reasonable.

E11
Where Fm/Fx is the ratio of the measured flux to the unattenuated flux, U is wind speed, z is measurement height, and nm and α are scaling factors for an unstable boundary layer taken as 0.085 and 7/8 respectively. 530

Uncertainty and flux limit of detection
Variance in the atmospheric O3 signal was estimated by calculating the autocovariance of the signal during a 27-minute flux averaging period (Fig. S11). Uncorrelated white noise only contributes to the first point in the autocovariance spectrum, while autocovariance at longer time shifts represents real atmospheric variance or correlated instrument drift (Blomquist et al., 2010; variance is 35 to 55%. This corresponds to a standard deviation from white noise , noise of 0.4 ppbv. The error in each flux averaging period (LODσ) can be determined by taking the standard deviation of the cross-covariance between vertical wind speed and mass spectrometer signal at lag times significantly longer than the calculated true lag time (Spirig et al., 2005;Wienhold et al., 1995). The random flux error is determined using lag windows of -150 to -180 and 150 540 to 180 s, which are significantly larger than the true lag time from sensor separation of 0.9 s as shown in Figure S12. The selection of the -150 to -180 and 150 to 180 s lag windows is somewhat arbitrary and may still capture organized atmospheric structure that persists over long time periods. We also calculate the root mean squared deviation (LODRMSE) of the crosscovariance over the same lag windows as proposed by Langford et al., (2015), which captures the variance in the crosscovariance in those regions but also accounts for long term offsets from zero in the cross-covariance. The resulting error from 545 the LODσ and LODRMSE methods showed good correlation (Fig S13), with periods where the LODRMSE error is larger. We apply the RMSE method for our reported flux error determination. The final deposition velocity limit of detection was determined for each 27-minute flux averaging period by multiplying the LODRMSE error by 1.96 to give the flux limit-ofdetection at the 95% confidence level. The flux error was then divided by the mean O3 concentration for that averaging period to convert from flux to deposition velocity units. The campaign ensemble flux LODRMSE was 0.0027 cm s -1 , calculated using 550 Eq. 12 following Langford et al.,(2015). A total of 59 out of 246 (24%) flux periods had deposition velocities below the campaign ensemble LOD. These values are still included in the reported mean vd(O3).

Density fluctuation corrections
The Ox-CIMS measures O3 as the apparent mixing ratio relative to moist air, as is true of all CIMS based measurements, which 555 means fluctuations in the density of air due to changes in temperature, pressure, and humidity could introduce a bias in the EC flux measurement (Webb et al., 1980). The temperature and pressure in the Ox-CIMS and sampling lines were both actively controlled during sampling, making density fluctuations from those sources negligible. The long (20 m) inlet sampling line used likely also dampened a substantial portion of the water vapor flux. This has been demonstrated in an EC study utilizing a closed path H2O sensor for EC flux measurements (through an 18 m long, 0.635 cm i.d. inlet, pumped at 18 slpm, comparable 560 to the inlet used in this study) which showed complete attenuation above 0.1 Hz and overall attenuation of ~80% of the H2O (latent heat) flux (Yang et al., 2016). However, without a direct measure of water vapor fluctuations collocated with the Ox-CIMS this is difficult to definitively rule out in our measurement. We therefore calculate a conservative estimate of this correction factor from Eq 45b. in Webb et al., (1980), assuming a latent heat flux of 50 W m -2 and neglecting the sensible heat term which is removed by active heating of the inlet. For a specific humidity of 12 g kg -1 , a temperature of 293 K, a pressure 565 of 1 atm, and an O3 mixing ratio of 40 ppbv, we calculate a flux correction term of 2.6 x 10 9 molecules cm -2 s -1 , which is 20% of our mean measured flux of -1.3 x 10 10 molecules cm -2 s -1 . We expect that the actual density correction for our instrument much smaller given that water vapor fluctuations were likely dampened in the inlet line, and the high latent heat flux used in the calculation (50 W m -2 ). Due to the uncertainty in this correction term for our instrument, we do not add it to our measured flux values and instead use the calculated value above as a conservative constraint on the magnitude. The addition of a Nafion 570 drier on the inlet has been successfully implemented in other O3 flux instruments to fully remove water fluctuations and will be used in future deployments of the Ox-CIMS (Bariteau et al., 2010).

Surface NO emissions
The observed dry deposition velocity of ozone is potentially biased by simultaneous air-sea exchange of nitric oxide (NO). 575 NO is expected to be emitted from the ocean on the order of 1 x 10 8 molecules cm -2 s -1 with dependence on dissolved surface nitrate and solar irradiance (Zafiriou and McFarland, 1981). This NO source near the surface will cause titration of O3 to NO2 resulting in a positive bias for the observed vd(O3). Assuming a maximum NO emission flux of 5 x 10 8 molecules cm -2 s -1 and that all NO reacts with O3 before being advected to the sensor height, the resulting O3 flux bias would be -5 x 10 8 molecules cm -2 s -1 . Our mean case of 40 ppbv O3 and vd (O3) of 0.013 cm s -1 corresponds to a flux of -1.3 x 10 10 molecules cm -2 s -1 . 580 Therefore, the resulting bias in observed vd(O3) from NO emissions is 3.8% or 4.9 x 10 -4 cm s -1 . This value is an upper limit for expected ocean NO emissions and is well within the uncertainty of the observed vd(O3). There is also potential for short term anthropogenic emissions of NO (such as from a boat engine passing by the sensor) to create a flux divergence term. We expect that the combination of signal despiking and the flux stationarity criteria described in Section 3.3 will minimize the impact of this potential divergence term. Despiking will remove most short term (<1 s) emission events and the stationarity 585 criteria will filter out any period where longer term titration events cause large changes in the observed flux within a flux measurement period.

Free troposphere entrainment
The entrainment of O3 enhanced or depleted air in the free troposphere to the marine boundary layer (MBL) creates a potential flux gradient that will contribute to the measured flux values at the near surface measurement height (zo) of 13 m. Lenschow 590 et al., (1982) presented aircraft observations of O3 deposition over the Gulf of Mexico at heights of 15, 60, and 325 m which showed a strong flux gradient term driven by entrainment from the free troposphere. The boundary layer height (zi) during those flights was approximately 1.2 km, suggests a strong flux gradient was present even within the surface layer (approximated as the lowest 10% of the boundary layer). The magnitude of this flux gradient depends on the magnitude of the O3 concentration gradient (ΔC) and the entrainment velocity (we) of air from free troposphere into the MBL. Faloona et al., 595 (2005), reported entrainment velocities from 0.12 to 0.72 cm s -1 and an enhancement in O3 (ΔC) of 20 ppbv in the free troposphere relative to the boundary layer in the summertime eastern subtropical pacific. Using those values and Equations 13and 14 below we calculate the percent fractional error from entrainment on the observed flux for a range of reasonable ΔC and we as shown in Fig. 12 (Blomquist et al., 2010).
Where zi is the boundary layer height, z is the measurement height, and Fi and F0 are the entrainment flux and surface flux respectively. We use the SIO measurement height (z) = 13 m and mean surface flux (Fo) = -5.2 x 10 -3 ppbv m s -1 (from vd = 0.013 cm s -1 and [O3] = 40 ppbv ), and an O3 mixing ratio gradient (ΔC) from -20 to +20 ppbv in the free troposphere relative to the boundary layer. The resulting fractional error in our observed mean surface flux from Scripps Pier using the values from 605 Faloona et al, 2005 (ΔC of +20 ppbv, MBL height of 800m) is 4.4% for we of 0.12 and 33% for 0.72 cm s -1 . This entrainment flux error is clearly significant for marine O3 flux measurements assuming there is a gradient of O3 in the free troposphere relative to boundary layer. This entrainment flux error is independent of the surface flux instrument measurement error and adds a systematic bias on the surface flux measurement. This calculation also makes clear that marine O3 measurements should be made as close to the surface as possible, and that the O3 concentration gradient and entrainment rate should be explicitly 610 measured if possible. We do not have an explicit measure of ΔC, we, or the MBL height so we tentatively assign entrainment error of up to 33% from the maximum values of those parameters reported in Faloona et al. (2005). We emphasize this source of uncertainty is independent of the O3 sensor and is a systematic bias that should be considered in all O3 air-sea exchange determinations.
4 Fast NO2 measurements, eddy covariance and O3 titration 615 Discussion of EC flux results have been limited to O3 because ocean-atmosphere exchange of NO2 is expected to be small and below the limit of detection of our instrument. The potential flux divergence from the reaction of O3 with NO is also below the instrument flux limit of detection as discussed in section 3.6. However, over terrestrial surfaces where NO2 emissions can be large, we expect this instrument would be well suited for measuring NO2 flux. From Equations 15 and 16, following Bariteau et al., (2010) and Lenschow and Kristensen, (1985), we calculate the we calculate an expected flux LOD for the case 620 where counting noise is the controlling term in the flux error. The calculated flux LOD is 4.3 x 10 9 molecules cm s -2 s -1 (1.6 pptv m s -1 ) for an NO2 mixing ratio of 1 ppbv and a friction velocity of 0.2 m s -1 .
Where u* is the friction velocity (m s -1 ), Ca is the gas phase concentration (ppbv), is the instrument sensitivity (cps ppbv -1 ), 625 and is the integral time scale (s). can be further expressed following Eq. 12 where a is constant taken as 0.3 for neutral conditions (Lenschow and Kristensen, 1985), z is the measurement height (here 10 m), and U is the horizontal wind speed at that measurement height (taken as 5 m s -1 ).
Observations of a short duration NO plume from a boat motor starting near our inlet at Scripps Pier highlights the 630 utility of the simultaneous O3 and NO2 detection from this instrument (Fig. 13). Highly localized NO emissions were observed as the titration of O3 and prompt production of NO2. Observed total odd oxygen (Ox = O3 + NO2) was conserved during this titration event, where NO2 and O3 concentrations were determined from independent calibration factors and backgrounds. The 1:1 conversion of Ox from O3 to NO2 shown in Fig. 13b, validates the laboratory generated instrument calibration factors for O3 and NO2. The temporal agreement of the O3 and NO2 signals also demonstrates that both O3 and NO2 are transmitted 635 through the inlet and detected with nearly identical instrument response times. This analysis assumes that there were no direct NO2 emissions during the titration event. A NO2 to NOx emission ratio of 0.08 was observed for ship emissions from diesel motors on inland shipping vessels (Kurtenbach et al., 2016). Without additional knowledge amount the NOx emission source during this event, the observed conservation of total Ox could be partially driven by compensating errors within 10%. This simultaneous detection of both Ox species is likely also well suited for mobile sampling in the presence of dynamic NO 640 emission sources, which challenge other fast ozone measurements. This method would also be well suited for direct measurement of flux divergence in the presence of strong surface NO emission sources.

Conclusions and Outlook
This study demonstrated the utility of oxygen anion chemical ionization mass spectrometry for the fast and sensitive detection of O3 and NO2. Field measurements of O3 dry deposition to the ocean surface from Scripps Pier, La Jolla CA demonstrate that 645 this method has suitable time response, precision, and stability for successful EC measurements. The mean measured vd(O3) with the Ox-CIMS is in within range ofprior studies of O3 ocean-atmosphere exchange. Further optimization and characterization of the Ox-CIMS is ongoing, including efforts to validate the specificity of the NO2 detection, addition of a Nafion drier system, and better background determination methods. While this work has focused primarily on the deposition of O3 to the ocean surface, the demonstrated instrument performance suggests the Ox-CIMS to be highly capable of O3 and 650 NO2 flux measurements in the terrestrial biosphere and urban environments and from mobile platforms.

Author Contributions
GAN, MPV, and THB designed the lab and field experiments and GAN and MPV collected all data. GAN lead the data processing, interpretation and analysis with MPV contributing. GAN prepared the manuscript with contributions from all coauthors. THB supervised all work and contributed to data analysis, writing and editing of the manuscript.

910
Sensitivity is reported at a specific humidity (SH) of 8 g kg -1 which corresponds to 40% RH at 25 °C. All limits of detection (LOD) are for a S/N = 3. The optimum LOD is reported as the LOD at the optimum averaging time determined by the minimum of the Allan variance spectrum. Optimum averaging times were determined to be 11 s for O3 and 19 s for NO2. The reported field comparison (R 2 ) is from a regression of 1-minute bin averaged ozone concentration from the Ox-CIMS with an EPA (Thermo-Fisher 49i) monitor in Zion, Il during four weeks of ambient observation shown in Fig. 7. Table 2. Overview of flux and concentration measurements of O3 and NO2 from Scripps Pier. Concentration ranges are reported for all periods of onshore winds. Flux results are reported only for final quality-controlled flux periods Ozone mean deposition velocity (vd) was well resolved from the campaign ensemble average LOD of 0.0042 cm s -1 . Reported vd LOD is the ensemble mean 940 of the LOD determined by the RMSE method at long lag times for each 27-minute flux period. 39% of quality-controlled flux periods fell below the campaign ensemble LOD. Deposition velocity of NO2 across the air-sea interface is expected to be small (<0.002 cm s -1 ) and was consistently below the LOD of our instrument so no values are reported here.

970
Ozone is detected as at -60 m/Q. NO2 is detected as the charge transfer product ( ) at -46 m/Q. Error bars are the standard deviation in normalized count rate for each measurement point.  980 of sensitivity as and shows that total sensitivity to O3 is conserved as the product distribution shifts with CO2 mixing ratio. Greater than 99% of O3 is observed as at CO2 mixing ratios greater than 60 ppmv. Figure 5. Normalized count rate of (-60 m/Q) ozone detection product as a function of pressure in the IMR during sampling of a constant 35 ppbv O3 source. The exponential fit of the data is shown by the dashed line. Fit parameters are included to allow for 985 calculation of potential sensitivity improvements with further increase in IMR pressure. Figure 6: Representative instrument backgrounding determination for O3 and NO2 where the inlet was rapidly switched from ambient sampling to an overflow with dry UHP N2 indicated by the grey dashed line. O3 response is fit to anexponential decay, 990 plotted as solid lines with a mean response time of 0.28 s.,NO2 is fit to a bi-exponential decay where the initial rapid decay ( ) in attributed to gas evacuation of the inlet line and the second slower decay ( ) is attributed to equilibration with the inlet walls. Best fit estimates for of NO2 from 0.7 to 1.2 seconds. for NO2 was determined to be 3.2 s for this decay period.  1005 binned to 1 ppbv over four weeks of ambient sampling in Zion, Illinois in May-June 2017. The solid black line is the linear leastsquares regression. Error bars represent the standard deviation of each bin. Instrument agreement is strong for O3 greater than 10 ppbv, with an apparent bias in one or both instruments below 10 ppbv.  from the maximum of the covariance to be 0.9 seconds which compares reasonably with the volumetric evacuation time of the inlet of 1.7 to 2.1 seconds. Figure 11. (a) Mean binned frequency weighted cospectra O3 and sensible heat flux with vertical wind from the average of two consecutive flux periods from 14:10 -15:20 local time on July 20th. The Kaimal trace is the idealized cospectra Kaimal et al. (1972) for mean windspeed of 4.4 m s -1 and an unstable atmosphere. The sensible heat trace is inverted, and the observed net sensible heat flux was positive for this period (b) Corresponding ogives for cospectra shown in (a). The M91 Transfer trace is the calculated transfer function for turbulent attenuation in a tube from (Massman, 1991).