Net ozone production and its relationship to NOx and VOCs in the marine boundary layer around the Arabian Peninsula

Strongly enhanced tropospheric ozone mixing ratios have been reported in the Arabian Basin, a region with intense solar radiation and high concentrations of ozone precursors such as nitrogen oxides and volatile organic compounds. To analyze photochemical ozone production in the marine boundary layer (MBL) around the Arabian Peninsula, we use ship-borne observations of NO, NO2, O3, OH, HO2, HCHO, actinic flux, water vapor, pressure and temperature obtained 15 during the summer 2017 Air Quality and Climate in the Arabian Basin (AQABA) campaign, compare them to simulation results of the ECHAM-MESSy atmospheric chemistry (EMAC) general circulation model. Net ozone production rates (NOPR) were greatest over the Gulf of Oman, the Northern Red Sea and the Arabian Gulf with median values of 14 ppbv day -1 , 16 ppbv day -1 and 28 ppbv day -1 , respectively. NOPR over the Mediterranean, the Southern Red Sea and the Arabian Sea did not significantly deviate from zero; however, results for the Arabian Sea indicate weak net ozone production of 5 20 ppbv day -1 , and net ozone destruction over the Mediterranean and the Southern Red Sea with -2 ppbv day -1 and -4 ppbv day -1 , respectively. Constrained by measured HCHO/NO2-ratios, our photochemistry calculations show that net ozone production in the MBL around the Arabian Peninsula occurs mostly in a transition regime between NOxand VOC-limitation with a tendency towards NOx-limitation except over the Northern Red Sea and the Oman Gulf. 25 https://doi.org/10.5194/acp-2019-1031 Preprint. Discussion started: 3 December 2019 c © Author(s) 2019. CC BY 4.0 License.


Introduction 25
Revenues from exploitation of the great oil reserves in the states of and around the Arabian Peninsula have propelled remarkable economic development associated with industrialization and urbanization. Strong population growth and anthropogenic emissions of gases and particulates in the last few decades have resulted in the Middle East becoming a hotspot for air pollution and associated health effects, while it is also one of the regions worldwide where climate change is particularly rapid (Lelieveld et al., 2016a). Unique meteorological conditions such as intense solar radiation, high temperatures and aridity, 30 as well as strong anthropogenic emissions of volatile organic compounds (VOCs) and NOx (= NO + NO2) by on-and off-shore petrochemical industries, dense ship traffic, fossil energy production for air conditioning and desalination, and urban development are expected to further intensify in the future and contribute to photochemical ozone production (Lelieveld et al, 2009;Krotkov et al., 2016;Pfannerstill et al., 2019). Understanding the sources and sinks of NOx and other ozone precursors on and around the Arabian Peninsula is therefore of major importance for atmospheric chemistry studies, including the 35 investigation of net ozone production rates (NOPR) (Monks et al., 2015;Reed et al., 2016;Bozem et al., 2017).
NOx plays a central role in atmospheric photochemistry (Nakamura et al., 2003;Tuzson et al., 2013;Reed et al., 2016). It is the primary precursor for tropospheric ozone (O3), secondary organic aerosols and photochemical smog in urban areas (Hollaway et al., 2012;Javed et al., 2019). Main ground-based sources of NO and NO2 are fossil fuel combustion and to a lesser extent bacterial processes in soils, and both lightning and aircraft emissions in the upper troposphere (Nakamura et al., 40 2003;Miyazaki et al., 2017;Javed et al., 2019). Transport of NOx in the atmosphere is relatively limited due to its short lifetime of a few hours (Reed et al., 2016). It is removed from the troposphere mainly by conversion to HNO3 (via reaction with OH) during the day, or the formation of N2O5 (in the reaction of NO2 with NO3 at night-time), which also leads to formation of nitric acid by heterogeneous hydrolysis on aerosol surfaces (Crutzen, 1973;Liu et al., 2016;Reed et al., 2016). Ultimately, the deposition of HNO3 constitutes the major loss process of NOx from the atmosphere. Ozone is a secondary pollutant that is 45 photochemically formed in the troposphere from its precursors NOx and VOCs (Bozem et al., 2017;Jaffe et al., 2018). It is an important greenhouse gas, an atmospheric oxidant and the most important primary precursor for OH Monks et al., 2015;Bozem et al., 2017). O3 in the planetary boundary layer causes health damage, notably respiratory diseases, and reduces crop yields (Monks et al., 2015;Jaffe et al., 2018).
NOx and O3 mixing ratios in the troposphere vary from less than 20 pptv and 10 ppbv, respectively, for pristine conditions such 50 as the remote marine boundary layer (MBL) up to mixing ratios of several hundreds of ppbv in regions with heavy automobile traffic and in international shipping lanes (for NOx) and downwind of urbanized areas (for O3) (Reed et al., 2016;Jaffe et al., 2018). Low NOx environments such as the clean MBL and the lower free troposphere are considered net ozone destruction regimes whereas the upper troposphere and areas with anthropogenic emissions of ozone precursors are regions of net ozone production (Klonecki and Levy, 1997;Bozem et al., 2017). Measurements performed in the the Houston Ship Channel revealed In the last decade much effort has been successfully devoted to the mitigation of NOx emissions over Europe and America, and levels of reactive nitrogen trace gases have decreased (Miyazaki et al., 2017). But in Asia, India and the Middle East, NOx emissions have substantially increased during the last decade so that the global NOx burden has essentially remained constant (Miyazaki et al., 2017). NOx emissions by ocean-going vessels have attracted considerable attention as they are reported to 60 account for 15 % of the global NOx emission burden (Celik et al., 2019). Model calculations suggest that the Arabian Gulf, with an estimated annual NOx emission density of about one ton km -2 from ship traffic, is among the regions with highest NOx emission densities worldwide (Johansson et al., 2017). Although NOx emissions in the Red Sea and Arabian Sea areas were reported to be three and five times smaller than for the Arabian Gulf, respectively, these values are still 50-100 times larger than the emission density reported for the South Pacific Ocean, for example (Johansson et al., 2017). 65 In the present study, we characterize photochemical NOPR in the MBL around the Arabian Peninsula. In Sect. 2, the campaign, instrument description, data processing and a description of the methods used in this study is presented. In Sect. 3, mixing ratios of nitrogen oxides and ozone around the Arabian Peninsula are reported. Based on concurrent measurements of HOx, actinic flux, temperature and pressure, noontime RO2 mixing ratios are estimated and used to calculate NOPR in the different regions around the Arabian Peninsula. Observation-based analysis of HCHO/NO2-ratios will be used to distinguish between 70 NOx-or VOC-limited chemistry in the particular regions. A comparison of the results with data retrieved from the 3D global circulation model EMAC is also included.

AQABA campaign
The AQABA ship campaign (Air Quality and Climate in the Arabian Basin) investigated the chemical composition of the 75 MBL around the Arabian Peninsula. From late June to early September 2017, the Kommandor Iona Research and Survey Vessel sailed from Toulon (France) to Kuwait and back in order to perform gas-phase and particle measurements in the region.
The gas-phase and aerosol measurement instrumentation was housed in five laboratory containers on the front deck. A 6 m high, 20 cm diameter cylindrical stainless steel common inlet was installed on the front deck of the vessel to sample air at a total mass flow rate of 10,000 SLM. NO and NO2 chemiluminescence measurements were obtained at a total bypass flow rate 80 of 28.5 SLM sampling air from the common inlet with a residence time in the tubing of ~3 s. HCHO, NO2 cavity ring-down spectroscopy and O3 measurements were obtained with similar bypass systems sampling air from the common inlet. H2O vapor was measured on the top of the ship mast in the front. The OH and HO2 detection units were placed on the prow to allow for inlets with residence times less than 10 ms.
The Kommandor Iona left Malta in late June 2017 traversing the Mediterranean Basin, the Suez Canal and the Northern Red 4 and the Gulf of Oman. Kuwait at the northern end of the Arabian Gulf marked the turning point of the ship cruise where, during a second 3-day stop-over, scientific staff was exchanged. The Kommandor Iona started the second leg on 03 August 2017 arriving in Toulon (France) in early September 2017 without any further stops. Figure 1 shows the ship's route subdivided 90 into six different regimes.
Figure 1: Ship cruises during both legs and color-coded subdivision into six different regimes. The following abbreviations will be used: AG for Arabian Gulf (purple), OG for Oman Gulf (dark blue), AS for Arabian Sea (blue), SRS for Southern Red Sea (green), NRS for Northern Red Sea (yellow), M for Mediterranean (red).

95
To enhance the statistical significance of our results and due to comparable signatures of the NOx and O3 measurements in the northern part of the Red Sea, the Suez Gulf and the Suez Canal, we have combined these regions which are represented by the 'Northern Red Sea' (NRS). For the same reasons we have merged the Gulf of Aden with the Arabian Sea (AS). See supplementary Table ST1 for the range of latitudinal and longitudinal coordinates of the different regions and supplementary  100   Table ST2 for a detailed day to day description of the route.

Measurements of nitrogen oxides during AQABA
Chemiluminescent detection of NO and NO2 is a widely applied method to quantify mixing ratios from the ppmv down to the low pptv range (Nakamura et al., 2003;Pollack et al., 2011;Hosaynali Beygi et al., 2011;Reed et al., 2016). During AQABA 105 we deployed a compact, robust and commercially available two-channel chemiluminescence instrument CLD 790 SR (ECO Physics AG, Dürnten, Switzerland) that has been optimized for in situ field measurements during the last decade (Hosaynali Beygi et al., 2011). The measurement principle of the CLD is based on the addition of O3 to NO to produce stoichiometric quantities of excited state NO2 * that will emit an infrared photon ( > 600 nm) forming the chemiluminescent detection principle for NO (Drummond et al., 1985;Reed et al., 2016). Both channels feature an identical layout and were operated at a 110 mass flow of 1.5 SLM during AQABA. One channel of the CLD (NOc-channel) has additionally been equipped with a LED solid state photolytic converter (Droplet Measurement Techniques, Boulder, Colorado) installed upstream of the O3 addition to selectively photolyze NO2 to NO, which is subsequently measured. In this section, we will concentrate on modifications made prior to the campaign and especially on operational conditions of the photolytic converter during the campaign. Further details on the measurement principle are described elsewhere (Pollack et al., 2011;Hosaynali Beygi et al., 2011;Reed et al., 115 2016).
During AQABA, the cylindrical photolytic converter (length 14 cm, volume ~ 0.079 l) was operated at a constant pressure of 95 hPa yielding a residence time of ~ 0.3 s. The photolytic NO2 converter features a set of 200 UV LED units attached to each end of the converter. The emission profile of the UV LED units was characterized in laboratory measurements to peak at 398 nm with a Full Width at Half Maximum (FWHM) of 16 nm. The UV-induced positive bias in the NO2-measurement due to 120 photolysis of BrONO2, HONO, NO3 and ClNO2 to produce NO was estimated at 6.1 %, 2.8 %, 2.7 % and 1.2 %, respectively, based on the absorption cross sections from the MPI-Mainz UV/VIS Spectral Atlas of Gaseous Molecules (Keller-Rudek et al., 2013). These values represent upper limits for the interference of the respective NOy compound as the respective molecular quantum yield was estimated conservatively at 1. Note that the values represent percent interferences if the interferent had the same concentration as NO2. Due to small daytime concentrations of these molecules in the MBL, a UV-induced bias was 125 neglected for the observations in this study. To limit wall loss of NO2, the inner cavity surface is made of PTFE (polytetrafluoroethylene), which may potentially provide a reservoir (via surface adsorption) for NOy that can thermally dissociate to increase the background signal of the NO2 measurement (Reed et al., 2016). The conversion efficiency of the photolytic NO2 conversion was estimated by gas phase titration (SYCOS K-GPT-DLR, ansyco, Karlsruhe, Germany) several times before, during and after the campaign at (29.4 ± 0.9) % allowing the calculation of NO2 concentrations by [NO 2 ] = 130 During AQABA, regular dry zero-air measurements as well as NO and NO2 calibrations were performed autonomously over a 10 minute period every 6 hours to accurately quantify the instrumental background and to correct for sensitivity drifts. An autonomous cycle of '2 min zero air measurements -2 min NO calibration -2 min zero air measurement -2 min NO2 135 calibration -2 min zero air measurement' was implemented. Continuous flows NO and NO2 calibration gases were added to the synthetic airflow or directed to a pump by switching solenoid valves. The NO calibration standard (1.954 ± 0.039 ppmv NO in N2, Air Liquide, Germany) used during the campaign was compared to a primary standard (5.004 ± 0.025) ppmv (NPL, Teddington, UK) after the campaign yielding an effective NO mixing ratio of (2.060 ± 0.057) ppmv in the NO calibration gas. Zero air measurements and NO calibrations were performed with a total flow of 3.44 SLM achieving an 140 overflow of 0.44 SLM to guarantee ambient air free standard measurements. The calibration gas was added at 4.5 sccm to the zero air flow. During AQABA, NO calibrations at 2.5 ppbv were achieved. During the first leg of the campaign, zero air was sampled from a bottle (Westfalen AG, Germany), whereas during the second leg zero air was generated from a zero air generator (Air Purifier CAP 180, acuraLine). Zero air measurements generated with the zero air generator were statistically not significantly different from those achieved by a bottle. To correctly account for the photomultiplier background and 145 chemical interferences due to reactions of ozone with ambient alkenes additional pre-chamber measurements were performed every 5 minutes as well as at the beginning of zero air measurements and calibrations for 25 s each. This correction is removing a large fraction of the interference signal from alkenes. However, in regions where alkene concentrations are strongly varying in time and magnitude, the CLD is prone to enhanced backgrounds due to the interference of alkenes with ozone in the instrument. A schematic setup of the two-channel CLD instrument is given in Figure 2. 150 The total measurement uncertainty (TMU) in the NO data is 5.5 % at a 5 min integration time and a confidence level of 1 . 155 The limit of detection in the NO channel was estimated as the full width at half maximum of the frequency distribution of all zero air measurements obtained during the campaign to be 9 pptv at a 5 min integration time and a confidence level of 1 . The TMU in NO2 has been estimated by means of the largest error possible from error propagation at ∆NO 2 = √∆NO 2 + ∆NO c 2 + ∆ 2 = √6% 2 + 6% 2 + 3% 2 = 9 % at a confidence level of 1σ and an integration time of five minutes. As the zero air measurements in the NO2 channel produced 160 an increased background affected by memory effects after exposure to high NOx levels e.g. during measurements of stack emissions, the NO2 raw data were initially processed without converter background subtraction. As we therefore expect the CLD NO2 data to be offset due to not being initially background corrected, the converter background was estimated at 112 pptv from the centre of a Gaussian fit representing the difference of 1-minute averaged CLD NO2 and concurrent cavity ring-down spectroscopy (CRDS) NO2 measurements for data points below 10 ppbv. Setting the threshold for calculating the 165 difference of the two concurrent data sets to 10 ppbv is somewhat arbitrary, however, changing this limit to 5 ppbv or 20 ppbv does not significantly vary the estimated offset of the CLD NO2 data. The offset correction of 112 pptv was taken as the ultimate absolute measurement uncertainty of the CLD NO2 measurement. Further corrections of to the final CLD data include residence time corrections as well as corrections for NO and O3 losses and the subsequent formation of NO2 in the sampling line (Ryerson et al., 2000). Both NO and NO2 CLD data have also been corrected for nonlinearities for concentrations higher 170 than 55 ppbv, as experienced during probing of stack emissions.

Further measurements used in this study
An extensive set of concurrent measurements providing mixing ratios of O3, NO2, HCHO, OH, HO2, absolute humidity and actinic flux, temperature and pressure data obtained during AQABA was used in this study. Ozone was measured with an 175 absorption photometer (Model 202 Ozone Monitor, 2B Technologies, Boulder, Colorado) based on the well-established absorption of the mercury line in the Hartley band at 254 nm (Viallon et al., 2015). Eliminating water and particle interferences during sampling was achieved via sampling through a nafion tube and a Teflon filter. The ozone monitor was zeroed ten times during the campaign. NO2 was further measured by cavity ring-down spectroscopy (Sobanski et al., 2016) and used for correcting the instrumental background of the CLD NO2 data, as described above (the correction was taken as the ultimate 180 absolute measurement uncertainty in the CLD NO2 data). Note that in this study we will use the NO2 CLD data rather than the NO2 CRDS data as the temporal coverage of the CLD NO2 data over the course of the campaign is about 60 % compared to about 35 % for the cavity ring-down measurement. Formaldehyde (HCHO) was measured with an Aerolaser 4021 (AERO-LASER GmbH, Garmisch-Partenkirchen, Germany), which is a fully automatized monitor based on the Hantzsch technique (Kormann et al., 2003). H2O measurements were obtained using a cavity ring-down spectroscopy monitor (PICARRO G2401, 185 Santa Clara, California) supervised by Laboratoire des Sciences du Climat et de l'Environnement (LSCE) (Kwok et al., 2015).
Measurements of OH and HO2 were performed with the custom-built HydrOxyl Radical measurement Unit based on fluorescence Spectroscopy (HORUS) instrument based on laser-induced fluorescence (LIF) spectroscopy of the OH molecule and NO titration of HO2 to OH followed by LIF spectroscopy detection of the OH molecule (Martinez et al., 2010;Regelin et al., 2013). HO2 data used in this study is still preliminary due to not yet corrected interference of organic peroxy radicals RO2. 190 The largest uncertainty due to interference by contribution of RO2 is 7 % or 3 pptv whichever is higher. The 1 sigma accuracy of both OH and HO2 is 20 %. The uncertainty in the OH data is here estimated as the 1 sigma accuracy of the data set at 20 %, whereas the uncertainty in HO2 is estimated at √20 % 2 + 7 % 2 ≈ 21 %. Wavelength resolved down-welling actinic flux was measured with a spectral radiometer (model CCD Spectroradiometer 85237). The j-values for NO2 and O3 were not corrected for upwelling UV radiation and were estimated to have a ~ 10 % measurement uncertainty (Meusel et al., 2016). The radiometer 195 was installed 10 m above sea level, respectively 5 m above the front deck surface. Decreases in sensitivity due to sensor contamination with e.g. sea-spray were corrected with a linear interpolation between two (daily) cleaning events. Temperature and pressure measurements were performed with the Shipborne European Common Automatic Weather Station (EUCAWS), a weather station specifically designed for ships. The weather station incorporates sensors, processing units, satellite positioning and communication systems in one device and is implemented and coordinated by the European National 200 Meteorological Service EUMETNET. Table 1 lists the measurement methods and the TMU for each observation. The Kommandor Iona Research and Survey Vessel sailed whenever possible with the wind coming from the bow to avoid 205 contamination by stack emissions. However, based on the relative wind direction, the variability in NO as well as the temporal evolution of NOx, SO2, and O3 sections of data in which the air mass was contaminated by the ship's stack were identified. All data used here to calculate RO2 and NOPR have been filtered to remove contaminated air masses. Altogether, 21 % of the sampling time was potentially contaminated by the ship exhaust of the KI of which 87 % occurred on the first leg. During the second leg the ship sailed against the wind and most of the data was free of stack contamination. Our analysis is based on a 5-210 minute running mean for each data set, whereby only averages that have been calculated at a temporal coverage greater than of peroxy radicals RO2 and NOPR around the Arabian Peninsula we have removed data measured during the stop-overs in Jeddah (11 July to 13 July), Kuwait (31 July to 03 August) and during bunkering at Fujairah City (06 August, 07:00 -15:00 UTC). Due to HOx data being available from 18 July 2017 onward, we have limited the net ozone production analysis to the period after this date. 220

Methods
The so-called NOx-O3-null cycle represents a rapid daytime cycling between NO, NO2 and O3. Solar UV radiation photolyzes NO2 to NO and O( 3 P) (R1) which will reform O3 in the subsequent reaction with molecular oxygen O2 (R2) (Leighton, 1961).
NO and O3 react to form NO2 and O2 (R3). R1, R2 and R3 constitute a so called null cycle which establishes photostationary steady state (PSS) for both NOx and O3 in mid latitudes during noon time on a time scale of ~100 s (Thornton et al., 2002;225 Mannschreck et al., 2004).
Deviations from expected NO/NO2-ratios at low NOx generally refer to missing oxidants converting NO to NO2 (Hosaynali Beygi et al., 2011;Reed et al., 2016) or to a measurement error due to an instrumental background or a positive interference from thermal labile NOx reservoir species (Reed et al., 2016;Silvern et al., 2018). In the present study we include HO2 and 240 RiO2 into the production term for NO2.
Assuming that the temperature-dependent rate coefficient for the reaction of each particular peroxy radical RiO2 with NO equals the rate NO+HO 2 for Reaction R4 (Hauglustaine et al., 1996;Cantrell et al., 1997;Thornton et al., 2002), we can combine HO2 and the sum of all organic peroxy radicals RiO2 to the entity RO2 that can be estimated using the steady state 245 equation . ( However, the steady state assumption is not valid if the sampled air parcel is affected by fresh emissions or fast changes in the actinic flux (Thornton et al., 2002). After sampling a fresh emission e.g. a ship plume, for which NOx went up typically to values of several tens of ppbv with simultaneous titration in O3, we assume that PSS is re-established on a time scale of 2 250 minutes (Thornton et al., 2002;Mannschreck et al., 2004). To best approximate PSS in our analysis we have restricted the estimation of RO2 on time frames ± 2 h around noontime for which we expect the smallest relative changes in the actinic flux.
Noontime for each day was determined as the centre of a Gaussian fit that was applied to the actinic flux data. We applied a Gaussian Fit to the actinic flux data as this fitting method is sufficient to estimate the centre of the diurnal actinic flux. To further limit the effect of periods for which PSS is not fulfilled, we use the median instead of the average that is often 255 disproportionately biased by strong NOx sources nearby. See supplementary Tables ST3, ST5 and ST7 for detailed statistics and a further motivation on regional averages and median values. See supplementary Figure S1 for a detailed illustration of the calculation of the fraction of the noontime integral.
A further part of the analysis will be the investigation of NOPR. Ozone production is initiated by reactions that produce HOx, for which primary production is from the photolysis of ozone, formaldehyde, nitrous acid (HONO) and hydrogen peroxide 260 (H2O2) (Thornton et al., 2002;Lu et al., 2010;Hens et al., 2014;Mallik et al., 2018). The production of ozone can be approximated by the rate of oxidation of NO with RO2 (HO2 + ΣiRiO2) to form NO2 that will rapidly form O3 (R1-R2) (Bozem et al., 2017). For RO2 we use the result from Eq. 3 that incorporates HO2 and the sum of all further peroxy radicals ∑ R i O 2 (Parrish et al., 1986;Thornton et al., 2002).
Photochemical O3 loss is mainly due to photolysis (λ < 340 nm) in the presence of water vapor and the reactions of ozone with OH and HO2 (Bozem et al., 2017).
was (10.6 ± 2.2) % during AQABA with a quasi linear dependence on water concentrations. The error in α is mainly determined by the error of H2O at 5 %. Furthermore, ozone is lost due to reactions with alkenes (R12) and halogen radicals 275 We find that the loss rate is dominated by the photolysis of ozone with subsequent reaction of O( 1 D) with H2O, was 60 -80 % of the total loss rate, followed by the reaction of O3 with HO2, which makes up 10 -30 % (note that the uncertainty in HO2 280 radical concentrations mentioned above has no significant influence on the total O3 loss rate, due to its small contribution).
The remaining fraction (10-30 %) is due to the reaction of O3 with OH. The reaction of ozone with ethene is on average 0.005 -0.01 ppbv h -1 and therefore generally less than 2 % of the total ozone loss rate . The reaction of O3 with all alkenes will hence be neglected. Halogen radicals were not measured during AQABA and will not be incorporated into our study. Based on oxidative pairs, Bourtsoukidis et al. (2019) have classified the majority of their samples collected 285 during AQABA by an OH/Cl-ratio of 200:1. As measured daytime OH concentrations were of the order of 5 • 10 6 molecule cm -3 , the estimate would yield a Cl concentration of 2.5 • 10 4 molecule cm -3 , which would decrease the estimated diurnal net ozone production rates by roughly 0.2 ppbv day -1 over the Arabian Sea and at most 0.6 ppbv day -1 over the other regions, which does not substantially alter the here presented results. The noontime chemical ozone loss rate can be summarized by 290 NOPR presented in this study is finally calculated as the difference of Eq. 4 and Eq. 6.
Under the assumption of constant chemical composition for a given day, the NOPR is expected to have a diel cycle following the measured actinic flux. Hence integrating the estimated NOPR over the course of a day based on the particular fractional noontime integral of j(NO2) will yield a diurnal value for NOPR. A detailed calculation of the diurnal fractional integrals is 295 given in the supplementary Figure S1. Note that all reaction rate constants used are from the IUPAC Task Force on Atmospheric Chemistry Chemical Kinetic Data Evaluation (Atkinson et al., 2004). Indications whether a chemical regime is NOx-limited or VOC-limited can be derived from the ratio of HCHO to NO2. Former studies have derived HCHO/NO2-ratios from satellite measurements to establish whether ozone production is NOx-limited or VOCs-limited. The results indicate NOxlimitation for HCHO/NO2 > 2 and prevailing VOC-limitation for HCHO/NO2 < 1 (Duncan et al., 2010). 300

ECHAM/MESSy Atmospheric Chemistry (EMAC) model
EMAC is a 3D general circulation model that includes a variety of sub-models to describe numerous processes in the troposphere, their interaction with oceans and land surfaces and incorporates anthropogenic influences. Here we use the second development cycle of the Modular Earth Submodel System (MESSy2) (Jöckel et al., 2010) and ECHAM5 (Röckner et al., 2006) which is the fifth generation European Centre Hamburg general circulation model in the T106L31 resolution 305 (corresponding to a quadratic grid of roughly 1.1° and 1.1°). The model has 31 vertical pressure levels and involves the complex organic chemistry mechanism MOM (Mainz Organic Mechanism) as presented by Sander et al. (2019) that includes further developments of the version used by Lelieveld et al. (2016b). Here we use the lowest pressure level in a terrain following coordinates (equivalent to the surface level) and simulations of NO, NO2, O3, OH, HO2, j(NO2) and j(O 1 D). The sum of peroxy radicals was estimated as the sum of all radicals R i O 2 with less than four carbon atoms. Net ozone production based on data 310 retrieved from EMAC was estimated as

NOx and O3 in the MBL around the Arabian Peninsula 315
During AQABA NOx mixing ratios varied over five orders of magnitude with lowest values of less than 50 pptv observed in relatively pristine regions and highest values of several hundred ppbv found in the vicinity of megacities or nearby passing ships. Ozone mixing ratios ranged from values of less than 20 ppbv, detected over the Arabian Sea, to more than 150 ppbv during episodes of severe pollution. Figures 3a) and 3b) show distributions of NOx measured during the first and second leg of the campaign (range from 0.1 ppbv to 20 ppbv) while Figure 3c) and 3d) show corresponding ozone mixing ratios covering 320 a range from 20 ppbv to 100 ppbv, respectively. A classification of the different regions based on Box-Whisker-Plots, including the 25-75-percentile interval (box) and whiskers for the 10-90-percentile interval, is shown in Figure 4 and Figure 5 for NOx and O3, respectively. As average NOx is often influenced by fresh, localized emissions, we have included the median (black bar) instead of the average in the Box-Whisker-Plot for NOx, which is less sensitive to extreme values. For O3, although the difference between median and mean is mostly negligible, we also use the median in Figure 5. NOx and O3 averages, medians, 325 standard deviations, 1 st and 3 rd quantiles and the number of data points quantified per region are given in the supplementary Table ST3. See supplementary Figure S4 for OH and preliminary HO2 mixing ratios around the Arabian Peninsula.
Supplementary Figure S5 shows the variation of the absolute humidity around the Arabian Peninsula.

330
scaled O3 mixing ratios (linear scale) c) during the first and d) during the second leg. Note that both NOx and O3 has been filtered for own stack contamination.

340
Overall, we find that NOx mixing ratios over the Northern Red Sea, the Gulf of Oman and the Arabian Gulf are approximately one order of magnitude higher than in the other three regions (Southern Red Sea, Arabian Sea, Mediterranean). NOx medians over the Arabian Gulf, the Northern Red Sea and the Gulf of Oman are 1.26 ppbv, 1.76 ppbv and 2.74 ppbv, respectively. Lower median NOx mixing ratios were measured over the Southern Red Sea (0.46 ppbv), the Mediterranean (0.25 ppbv) and the 345 Arabian Sea (0.19 ppbv). With respect to observed O3 mixing ratios, the Arabian Sea is the only region representing remote MBL conditions with lowest median and average O3 of 21.5 ppbv and 22.5 ppbv respectively, followed by the Gulf of Oman where median and mean O3 were 31.5 ppbv and 34 ppbv, respectively. The low O3 mixing ratios over the Arabian Sea were accompanied by the smallest variability (whisker-interval: 15.1 ppbv). Although observing highest NOx over the Oman Gulf, O3 observed over the Oman Gulf was amongst the lowest detected throughout the whole campaign, which can be partly 350 explained by the fact that high NOx lead to low ozone production or even net ozone destruction. However, a significantly larger whisker-interval of observed ozone of 31.4 ppbv over the Gulf of Oman indicates increasing amounts of pollution and advection from the Arabian Gulf where extreme events of ozone were observed several times during the campaign with maximum mixing ratios of up to 170 ppbv when wind was coming from Kuwait/Iraq. Please note that during the second leg wind was coming from Iran (Pfannerstill et al., 2019). The whisker-interval over the Arabian Gulf was 100.9 ppbv, more than six times higher than that over the Arabian Sea. Reasons for large variations of both NOx and O3 over the Arabian Gulf were a multitude of point sources as well as a change in the observed wind direction with air masses coming from Iraq/Kuwait area during the first leg and air masses coming from Iran during the second leg (Pfannerstill et al., 2019). Over the Mediterranean, the Northern Red Sea and the Southern Red Sea, median ozone was 61.5 ppbv, 64.2 ppbv and 46.9 ppbv, respectively. The whisker-intervals over the Northern Red Sea and the Southern Red Sea were 44.2 ppbv and 31.6 ppbv, respectively. Air masses over the 360 Mediterranean were characterized as photochemically aged due to their impact by northerly winds (Etesians) which bring processed/oxidized air from eastern Europe (Turkey, Greece) to the Mediterranean area (Derstroff et al., 2017;Pfannerstill et al., 2019). This photochemical ageing/oxidation over the Mediterranean leads to a rather small whisker-interval of 18.7 ppbv in ozone. In summary, median NOx over the Oman Gulf was 56 % and 117 % higher than over the Northern Red Sea and the Arabian Gulf, respectively. However, the highest NOx average was measured over the Northern Red Sea at 4.69 ppbv, similar 365 to the values observed over the Oman Gulf (4.16 ppbv) and the Arabian Gulf (3.65 ppbv). Note that highest NOx mixing ratios over the Oman Gulf and over the Northern Red Sea are not always associated with high O3 mixing ratios. We find that average ozone was highest over the Arabian Gulf with 74 ppbv followed by the Northern Red Sea region (63.4 ppbv). The average ozone mixing ratio over the Oman Gulf was 34 ppbv, which corresponds to 46 % of the value observed over the Arabian Gulf. Due to a number of large pollution sources in the region around the Arabian Peninsula such as passing ships, highly urbanized areas as well as on-and off-shore petrochemical processing, NOx levels were rarely as low as those found in remote locations such as over the South Atlantic (Fischer et al., 2015) where NOx levels may be under 20 pptv. Apart for a few occasions where NOx was below 50 pptv for short periods (Arabian Sea, the Southern Red Sea and the Mediterranean), NOx levels during 375 AQABA generally ranged from 100 pptv up to several ppbv. The campaign NOx median of 0.65 ppbv and mean value of (2.51 ± 5.84) ppbv is comparable to urban sites (Kleinman et al., 2005). A detailed emission density analysis performed by Johansson et al. (2017) shows that NOx emissions on and around the Arabian Peninsula are amongst the highest worldwide, which could explain the rather high NOx level in the MBL around the peninsula (Johansson et al., 2017;Pfannerstill et al., 2019). O3 mixing ratios measured during AQABA were also very variable with O3 mixing ratios ranging between less than 20 ppbv in the remote 380 MBL (Fischer et al., 2015) to 60-70 ppbv in the Mediterranean (consistent with previous ship-based measurements in the region (Kouvarakis et al., 2002) and as high as 150 ppbv measured over the Arabian Gulf region. The latter are consistent with O3 mixing ratios reported from regions influenced by oil and gas processing (Pfannerstill et al., 2019) and shipping lanes such as the Houston Ship Channel (Mazzuca et al., 2016). Figure 4 also shows that the general trend for NOx mixing ratios in the different regions is widely reproduced by the EMAC 385 model. We find that the median NOx(model)/NOx(measurement)-ratio of all five minute averaged data points of the whole campaign is 0.91, indicating that the model underestimates NOx by roughly 10 %. The average ratio and its standard deviation are significantly larger at 2.57 and 5.71, respectively, indicating that single modeled data points strongly exceed the measurements, especially during periods of low in situ NOx (see supplementary Figure S6). Particularly over the Arabian Sea and the Southern Red Sea, the model generally simulates NOx mixing ratios higher than 100 and 200 pptv, respectively while 390 the measurements indicate mixing ratios of less than 50 pptv for certain periods. Furthermore, as expected, the model is not able to reproduce point sources such as passing ships for which we observe a significant underestimation of the measured NOx.
For ozone we find that the median O3(model)/O3(measurement)-ratio throughout the campaign is 1.23, indicating that over the course of the campaign the model overestimates O3 by about 23 %. This could partly be related to the same limitation, i.e. the inability of the model to resolve point sources in which O3 is locally reduced due to titration by NO. While the model is in 395 rather good agreement with the measurements over the Mediterranean, the Northern Red Sea and Southern Red Sea, large deviations are found over the Arabian Sea and the Oman Gulf, where the model overestimation with respect to the regional median is 63 % and 75 %, respectively. A possible explanation for the overestimation of both ozone and NOx in pristine regions such as over the Arabian Sea and the Oman Gulf could be related to the model resolution of 1.1° x 1.1°. Interpolation of model simulations along the Kommandor Iona ship track close to the coast at this resolution will most likely incorporate contributions 400 from nearby land areas, affected by anthropogenic emissions. See supplementary Table ST3 and Table ST4 for further information and Figure S6 and S7 for additional scatterplots of measured and simulated regional median NOx and O3, respectively.

Estimation of RO2 around the Arabian Peninsula
Noontime RO2 was estimated based on Eq. 3. As the steady state assumption will not hold for air masses originating from 405 fresh emissions (times to acquire steady state estimated from the inverse sum of the loss and production terms for NO2 typically ranged from 1-2 minutes during AQABA) and for fast changes in the actinic flux, we have calculated Box-Whisker-Plots for ± 2 h around noontime for which we expect relatively minor changes in the actinic flux ( Figure 6). The noontime of each day was approximated by applying a Gaussian fit routine to the measured j(NO2) values whereas j(NO2) values being less than 10 -3 s -1 were neglected. Due to the availability of OH and HO2 data from 18 July 2017 onwards, we have limited the analysis to 410 this period. Note that there are no noontime RO2 estimates from 18 July to 21 July due to contamination by the ship exhaust and on 24 August 2017 due to missing data. The black bar in Figure 6 indicates the median value, with the Box-interval marking the 25-and 75-percentile and the whisker showing the 10-and 90-percentile. Figure 7 shows summarized regional trends of the RO2 estimates for measured and simulated data.  We find median noontime RO2 mixing ratios over the Mediterranean, the Northern Red Sea, the Southern Red Sea, the Arabian Sea and Oman Gulf of 16 pptv, 28 pptv, 15 pptv, 33 pptv and 22 pptv, respectively, with each respective 75-percentile RO2 being equal or less than 54 pptv. Based on the total measurement uncertainties of the measured quantities in Eq. 3, the uncertainty in 425 RO2 is estimated by means of the largest error possible at 15 %. ∆RO 2 = √∆NO 2 + ∆NO 2 2 + ∆O 3 2 + ∆ (NO 2 ) 2 = √6% 2 + 9% 2 + 2% 2 + 10% 2 ≈ 15 % Note that our calculation assumes that errors in the used rate coefficients are negligible. Only over the Arabian Gulf, the RO2 estimate yields a median noontime mixing ratio of 73 pptv accompanied by the largest variations in the box-interval of the whole campaign. While the box-interval of the RO2 estimate in the other regions is 25-57 pptv, the box-interval over the 430 Arabian Gulf is significantly higher at 165 pptv. Negative values for all regions are regularly found in the vicinity of fresh emissions and air masses not in photochemical equilibrium. The elevated 90-percentile over the Arabian Sea is due to high RO2 estimates during the first leg on 22 and 23 July.
Estimated RO2 mixing ratios based on measured tracer data are in general agreement with previous studies performed in marine boundary layer environments which report maximum mixing ratios between 30 and 55 pptv around noontime (Hernandez et 435 al., 2001). As peroxy radicals are short-lived molecules generated from the oxidation of VOCs, enhanced RO2 concentrations observed over the Arabian Gulf are most likely due to high VOC emissions from intense oil and gas activities in the region Pfannerstill et al., 2019). However high HO2 and RO2 can also occur in aged air masses with low NOx and VOCs but still significant O3 (and perhaps HCHO whose photolysis would then yield peroxy radicals). Bourtsoukidis et al. report that spatial volume mixing ratios of ethane and propane over the Arabian Gulf were about a factor of 10-15 times 440 higher than over the Arabian Sea and the Southern Red Sea . We find that the median noontime RO2(measurement estimate)/HO2(measurement)-ratio throughout the whole campaign is 1.88. Note that during single days, HO2 may be higher than the RO2 estimate, which is within the uncertainty of the RO2 estimate.
EMAC modelled, median noontime RO2 mixing ratios estimated as the sum of simulated HO2 and all simulated peroxy radicals with less than four carbon molecules are 41 pptv, 46 pptv, 38 pptv, 41 pptv, 50 pptv and 49 pptv over the Mediterranean, the 445 Northern Red Sea, the Southern Red Sea, the Arabian Sea, the Oman Gulf and the Arabian Gulf, respectively. The observation based RO2 estimate yields 16 pptv, 28 pptv, 15 pptv, 33 pptv, 22 pptv and 73 pptv respectively. We find that the median point by point RO2(model)/RO2(measurement estimate)-ratio from 18 July onward is 1.05 so that, on average, the model overestimates the measurement by 5 %. Please note that the observational variability is much higher than the modeled one and that the median of 1.05 is accompanied by a larger average (1.84) and a large variability (42.51). See supplementary Table ST5 and ST6 for 450 further information and Figure S8 for an additional scatterplot of measured and simulated regional median RO2.

Net ozone production rates around the Arabian Peninsula
In the following, net ozone production rates (at noon) are calculated based on Eq. 7 for the different regions. These noontime values are scaled to diurnal production rates ( Figure 8). As photochemical net ozone destruction is in good approximation linear with actinic flux j(NO2) and as on average (46.1 ± 2.8) % of the total j(NO2) occurred ± 2h around noon, the median 455 noontime NOPR estimate was multiplied by 4/0.461 ≈ 8.68 to obtain a diurnal value. The error in the total actinic flux located ± 2h around noon is estimated from the standard deviation of the best estimate of 0.461 at ∆ ≈ 6 %. Due to contamination by the own ship exhaust and due to the availability of OH and HO2 data only from 18 July 2017 onwards, we have limited the analysis to the period from 22 July 2017 to 31 August 2017. A comparison of NOPR estimated based on measured and simulated data for the different regions is shown in Figure 9. A break-down of the different terms of Eq. 7 in the six regions is 460 included in the supplementary Figures S10-S13. Over the Mediterranean and the Southern Red Sea, NOPR values do not significantly deviate from zero (production equals loss) within the atmospheric variability. Based on the total measurement uncertainties of the measured quantities in Eq. 3, the systematic error in NOPR is estimated from error propagation by means of the largest error possible at 38 %. ∆NOPR = √∆NO 2 + ∆NO 2 2 + ∆O 3 2 + ∆ (NO 2 ) 2 + ∆ (O 1 D) 2 + ∆ 2 + ∆RO 2 2 + ∆OH 2 + ∆HO 2 2 + ∆ 2 = √6% 2 + 9% 2 + 2% 2 + 10% 2 + 10% 2 + 5% 2 + 15% 2 + 20% 2 + 21% 2 + 6% 2 ≈ 38 %. 475 The best estimate indicates slight net ozone destruction for the Mediterranean and Southern Red Sea (-1 ppb day -1 ) and (-4 ppb day -1 ) respectively, and slight net production for the Arabian Sea (5 ppb day -1 ), which is significantly positive within the variability of the box-interval. Variations in NOPR calculated as the width of the 25-75-percentile-box yield comparable values of 9-11 ppb day -1 for these three regions. Substantial net ozone production was inferred over the Oman Gulf, the Northern Red Sea, and the Arabian Gulf with the respective median values being 16 ppb day -1 , 16 ppb day -1 and 32 ppb day -1 , respectively. 480 Especially over the Red Sea we find a strong latitudinal gradient in net ozone production rates with higher values towards the northern end, while slight net ozone destruction of -4 ppb day -1 is reported over the southern part.
NOPR estimates for the Oman Gulf, the Northern Red Sea and the Arabian Gulf are comparable to results reported for dense traffic shipping routes such as the Houston Ship Channel with NOPR of a few tens of ppb h -1 for periods of severe pollution (Zhou et al., 2014). Similar net ozone production rates have been reported for regions of Beijing in summer 2006 (Lu et al., 485 2010). For regions with low anthropogenic influence such as the Southern Red Sea and the Arabian Sea we estimate net ozone production that does not differ significantly from zero. This is due to the rather low NOx mixing ratios in the clean marine boundary layer (Bozem et al., 2017). Note that we calculated net ozone destruction only for a few days over the Southern Red Sea and the Arabian Sea, indicating that the marine boundary layer around the Arabian Peninsula is rarely free from anthropogenic influence owing to the multitude of on-and off-shore anthropogenic activities. 490 We find that model-calculated estimates of NOPR reproduce the trends observed for NOPR calculated from in situ measurements except over the Mediterranean and the Southern Red Sea. Although EMAC predicts high ozone levels over the Arabian Sea, it also reports the lowest NOPR in this region. On the other side, the large overestimation of the model-calculated estimate NOPR against the one based on measured tracer data over the Mediterranean and over the Southern Red Sea could be linked to NOx being overestimated in the model in these regions. In the model, pollution emissions, especially over the 495 Oman Gulf and the Arabian Gulf, seem to be averaged over a large (1.1° grid size) region. High background concentrations of ozone precursors hence contribute to net ozone production rates that compare to conditions observed in the Houston case (Zhou et al., 2014). Even in the more pristine regions such as over the Southern Red Sea and the Arabian Sea, the model is not able to reproduce net ozone destruction, which is consistent with the fact the ozone is generally too high and that NOx levels below 0.1 ppbv are not found in the model. See supplementary Table ST7 and ST8 for further information and supplementary Figure  500 S9 for an additional scatterplot of measured and simulated regional NOPR.

VOC-and NOx-sensitivity
Ozone is photochemically formed when the precursors NOx and VOCs are abundant in the presence of sunlight (Bozem et al., 2017;Jaffe et al., 2018). In order to determine whether a chemical system is NOx-or VOC-limited or in a transition between those two regimes, one has to estimate the total amount of OH reactivity towards VOCs and towards NOx. Therefore the 505 VOC/NOx-ratio is an important indicator of the behavior of NOx, VOCs and O3 in a system. Since it is not feasible to precisely define all ambient VOCs (could be thousands), formaldehyde mixing ratios have been used as a proxy for the OH reactivity towards VOCs since it is a short-lived oxidation product of many VOCs that is often positively correlated with peroxy radicals (Sillman et al., 1995;Duncan et al., 2010). Sillman et al. first used afternoon concentrations of indicator species such as HCHO and total reactive nitrogen (NOy) to determine the sensitivity of ozone production to VOCs or NOx (Sillman et al., 1995). Their 510 approach was later successfully transferred to space-based satellite observations by using the ratio of tropospheric columns of HCHO and NO2 to determine the sensitivity of ozone production (Martin et al., 2004). Here we use HCHO/NO2-ratios (referred to as "Ratio") deduced by Duncan et al. as indicators for the sensitivity of ozone production to NOx-and VOC-limitations in megacities in the United States with large amounts of anthropogenic NOx and VOC emissions (Duncan et al., 2010). The Ratio is an indicator of surface photochemistry as most of the atmospheric column of HCHO and NO2 is located in the planetary 515 boundary layer (Duncan et al., 2010). Duncan et al. have derived NOx-limited ozone production regimes for HCHO/NO2 > 2 and VOC-limited ozone production for HCHO/NO2 < 1 (Duncan et al., 2010). For 1 < HCHO/NO2 < 2 both NOx and VOC emission reductions may lead to a reduction in ozone. Figure 10 shows the Box-Whisker-Plot classification of the HCHO/NO2ratio of the different regions during noontime.

525
Median HCHO/NO2-ratios of 5, 7.7, 9.4 and 9.3 over the Mediterranean, the Southern Red Sea, the Arabian Sea and the Arabian Gulf respectively indicate tendencies towards NOx-limited regimes. In a previous study based on measured OH reactivity, Pfannerstill et al. classified these regions as being mostly in a transition between NOx-and VOC-limitation, with a tendency towards NOx-limitation (2019). Median HCHO/NO2-ratios of 1.4 and 2.2 estimated over the Northern Red Sea and the Oman Gulf signify tendencies towards VOC-limitation. However, none of the medians of the six regions falls below the 530 VOC-limit deduced by Duncan et al. (2010).
Over the Red Sea we find a latitudinal gradient in the HCHO/NO2-ratio, similar to the gradients for NOx and NOPR. Due to very low NOx over the Southern Red Sea, O3 production is NOx-limited, changing into a more VOC-limited regime over the Northern Red Sea. Ozone production over the Mediterranean was classified as rather NOx-limited, however partly being in the transition regime between NOx-and VOC-limitation., which can be explained by measurements obtained on 29 August 2017 535 when laying at anchor in front of Malta with a multitude of (NOx)-emissions from nearby situated vessels. Average noontime NOx on that particular day was about three times as large as the regional average noontime NOx observed over the whole Mediterranean area. NOx limitation is also inferred for the relatively clean Arabian Sea and the polluted Arabian Gulf atmosphere. Note that a further increase in NOx-emissions from increased shipping in the Arabian Gulf may initially lead to higher ozone production. However, a further increase in NOx might eventually lead to a change from NOx-to VOC-sensitivity 540 and a decrease in ozone production for this region, as observed for the Oman Gulf (median HCHO/NO2-ratio of 2.2 and average O3 of 34 ppbv). See supplementary Table ST9 for detailed statistics on regional HCHO/NO2-ratios.

Conclusion
In situ observations of NO, NO2, O3, HCHO, OH, HO2, absolute humidity, actinic flux, temperature and pressure were carried 545 out in the marine boundary layer around the Arabian Peninsula during the AQABA ship campaign from late June to early September 2017. Concentration ranges of both NOx and O3 clearly showed anthropogenic influence in the MBL. NOx was highest over the Arabian Gulf, the Northern Red Sea and the Oman Gulf. Lowest NOx was observed over the Arabian Sea and over the Southern Red Sea during the second leg. O3 mixing ratios were highest over the Arabian Gulf. We observed a latitudinal gradient in O3 concentrations with higher values towards the northern part of the Red Sea. Although comparable O3 550 averages were measured over the Northern Red Sea and over the Mediterranean, lower variability over the Mediterranean towards the end of August 2017 indicates photochemically more extensively aged air masses. The lowest regional O3 mixing ratio average was detected over the Arabian Sea, which is broadly comparable to remote marine boundary layer conditions in the Northern Hemisphere.
Noontime RO2 estimates based on deviations from the Leighton Ratio yield median values around the Arabian Peninsula 555 amount to 15 -33 pptv for all regions except over the Arabian Gulf where the median is 73 pptv. The uncertainty due to the missing up-welling actinic flux portion is expected to be insignificant. Furthermore, we estimated noontime and diurnal NOPR based on Eq. 6 and the integral over the actinic flux. Highest diurnal NOPR were observed over the Oman Gulf, the Northern Red Sea and the Arabian Gulf with median values of 16 ppbv day -1 , 16 ppbv day -1 and 32 ppbv day -1 , respectively, which is in agreement with previous studies that predicted net photochemical O3 formation conditions in the region. Net ozone destruction 560 was only observed for a few days with clean conditions over the Arabian Sea and the Southern Red Sea. Based on HCHO/NO2ratios our analysis suggests tendencies towards NOx-limitation over the Mediterranean, the Southern Red Sea, the Arabian Sea and the Arabian Gulf and VOC-limitation over the Northern Red Sea and the Oman Gulf, which reproduces the trends observed by Pfannerstill et al. (2019).
NOx results from the atmospheric chemistrygeneral circulation model EMAC underestimate the measurement data by 10 % 565 whereas median modeled O3 overestimates the measurement by 23 %, the latter being related to limitations in model resolution in coastal proximity and near shipping lanes. Although EMAC generally reproduces regional NOx and O3 medians, the scatter when comparing both data sets is large. NOx is generally too low as it does not resolve local point sources and too high for clean regions. Lowest NOx of less than 0.1 ppbv found in the in situ measurements is not reproduced by the model as emissions are averaged over a large area (1.1°). Median noontime RO2 retrieved from the EMAC model are ~ 5 % higher than RO2 570 estimates based on measurement data, however, the RO2 sum deduced from EMAC is sometimes about a factor of 2 higher than the regional RO2 estimate based on the Leighton Ratio and measured tracer data. NOPR estimates based on modeled data reproduce the tendencies derived from the measurements very well. However, the model does not reproduce observed net ozone destruction along some clean parts of the ship cruise.

Data availability 575
Data used in this study is available to all scientists agreeing to the AQABA protocol at https://doi.org/10.5281/zenodo.3693988.

Author contributions
IT, HF and JL designed the study. UP and IT performed the CLD NO and NO2 measurements and processed the data. JC and PE performed the O3 measurements, JS performed the actinic flux measurements. JS performed cavity ring-down spectroscopy 580 measurements of NO2. DD and BH performed the HCHO measurements. HH, MM, RR, ST performed the OH and HO2 measurements. J-DP was responsible for the H2O measurements. Model simulations were made by AP. All authors have contributed to writing this manuscript

Competing interests
The authors declare no conflict of interest. 585