Reactive nitrogen around the Arabian Peninsula and in the Mediterranean Sea during the 2017 AQABA ship campaign

. We present shipborne measurements of NO x ( ≡ NO + NO 2 ) and NO y ( ≡ NO x + gas- and particle-phase organic and inorganic oxides of nitrogen) in summer 2017 as part of the expedition “Air Quality and climate change in the Arabian BAsin” (AQABA). The NO x and NO z ( ≡ NO y NO x ) measurements, made with a thermal dissociation cavity ring-down spectrometer (TD-CRDS), were used to examine the chemical mechanisms involved in the processing of primary NO x emissions and their inﬂuence on the NO y budget

OH + RH(+O 2 ) → RO 2 + H 2 O (R4) RO 2 + NO → RO + NO 2 (R5a) HO 2 + NO → OH + NO 2 (R6) RO 2 + NO 2 + M → RO 2 NO 2 + M (R7) The formation of long-lived organic nitrates (R5b) and especially nitric acid (R3) represent daytime sinks for both NO x and RO x (OH + HO 2 +RO + RO 2 ). At night-time, when the photolysis of NO 2 ceases, NO is sequentially converted to the NO 3 radical (Reaction R8). This radical can also be a source of RONO 2 species through the addition to unsaturated VOCs. NO 3 exists in thermal equilibrium with NO 2 and N 2 O 5 (Reaction R9) and the heterogeneous loss of N 2 O 5 to aqueous surfaces results in transfer of NO y to the particle phase as HNO 3 (Reaction R10) or its loss via deposition. In some (especially marine) environments (Osthoff et al., 2008;Kercher et al., 2009), loss of N 2 O 5 to particles can result in formation of ClNO 2 (R11) which, via photolysis, re-forms NO 2 the next day.
The above reactions illustrate that NO x and VOCs provide the catalyst and fuel for photochemical ozone formation, the efficiency of which is determined by the competition between photolysis of NO 2 to ozone and its conversion to NO z (Day et al., 2003;Wild et al., 2014Wild et al., , 2016Womack et al., 2017). Modelling studies have identified the Arabian Gulf as a hotspot for O 3 pollution and photochemical smog, with O 3 mixing ratios exceeding 100 ppbv (Lelieveld et al., 2009).
The lack of measurements in the Arabian Gulf and the eastern Mediterranean, both of which are expected to be significantly impacted by climate change , preclude accurate prognosis of air quality in these regions and provide the rationale for conducting the AQABA campaign (AQABA: Air Quality and climate change in the Arabian BAsin), in which a large suite of instruments were operated in regions that were influenced by anthropogenic emissions from megacities, petrochemical and shipping activity, and desert dust emissions and through regions that could be classified as maritime background conditions. Emissions from oil exploration provide a complex atmospheric mixture of NO x and anthropogenic VOCs. The presence of desert dust can have a significant impact on the budget of inorganic acids such as HNO 3 . Finally, the overall elevated temperatures and actinic fluxes during AQABA promoted rapid photochemical processing of NO x . We therefore expect a more varied and complex chemistry than found in remote marine locations.
Previous analyses from this campaign focussed on sources and sinks of non-methane hydrocarbons ; the role of OH reactivity in ozone chemistry (Pfannerstill et al., 2019); formation of ClNO 2 (Eger et al., 2019a), ethane, and propane emissions from the Red Sea (Bourtsoukidis et al., 2020); emission factors in ship plumes (Celik et al., 2020); marine emissions of methane sulfonamide ; rates of net O 3 production ; and the abundance of carbonyl compounds.
In this paper we present NO x , NO y , and NO z mixing ratios obtained by a thermal dissociation cavity ring-down spectrometer (TD-CRDS), together with NO and NO 2 mixing ratios from a chemiluminescence detector, a comprehensive set of ancillary measurements, and an analysis of the results in terms of photochemical processing and ageing of air masses, chemical sources of NO x (e.g. from the photolysis of HONO), and the efficiency of ozone formation.
The ozone production efficiency (OPE), a metric used in the analysis of the O 3 formation, quantifies the fractional transformation of primarily emitted NO x to O 3 (Liu et al., 1987;Trainer et al., 1993) and thus reflects the relative importance of competing photochemical processes leading to O 3 and NO z formation from NO x . High values of OPE are favoured by low OH and VOC concentrations, and values exceeding 80 have been reported for remote marine environments. Low single-digit values have been observed in polluted urban environments (Rickard et al., 2002;Wang et al., 2018). The location dependence of the OPE can be further classified with previous observations from the literature. Minimal OPEs in urban environments between 1 and 2 have been reported from the Beijing area (Lin et al., 2011;Ge et al., 2013) and from the USA Sillman, 2000;Nunnermacker et al., 2004). In rural and suburban environments, the OPE can increase to values between 10 and 15, as demonstrated in North America (Olszyna et al., 1994;Roussel et al., 1996;Fried et al., 1997;Ninneman et al., 2017) and in China (Sun et al., 2010). From oceanic samples, OPEs of 65 and 87 were observed on the southeastern coast of the UK (Rickard et al., 2002) and on Sable Island, Canada (Wang et al., 1996). Flights over the western Pacific Ocean found values of 102-246 in the tropical area (latitude 0-18 • N) and of 73-209 further north (18-42 • N) (Davis et al., 1996). For the AQABA campaign, we expect lower OPEs than those observed in remote oceanic locations, due to the variable influx from harbours, coastal pollution, and surrounding ship traffic.

Methods
The AQABA ship campaign followed a route from Toulon in southern France to Kuwait (and back) via the Mediterranean Sea, the Suez Canal, the Red Sea, the Arabian Sea, and the Arabian Gulf (see Fig. 1). Stops were made in Malta, Jeddah, Djibouti, and Fujairah on the first leg (24 June to 30 July 2017) and in Fujairah and Malta on the second leg (2 August to 30 August 2017). Most measurements started in the south-eastern Mediterranean Sea on the first leg and finished ca. halfway between Sicily and Corsica on the second leg. The instruments were located either in air-conditioned research containers aboard, or directly on the deck of, the 73 m long research vessel "Kommandor Iona". Periods during which instrument inlets were contaminated by ship-stack emissions from the ship (identified based on relative wind direction and speed and the variability in measured SO 2 and NO mixing ratios) were excluded from the analysis. This resulted in rejection of 38.4 % of the data points on the first leg, when the wind and ship direction were often similar, and rejection of 1.4 % on the second leg of the campaign, when sailing mainly into the wind.

TD-CRDS instrument for NO x , NO y , and NO z detection
The TD-CRDS instrument, its operating principles, laboratory characterisation, and a validation of the NO x measurements versus an independent chemiluminescence detector (CLD) instrument have recently been presented . The TD-CRDS (located in an air-conditioned research container on the front deck of the vessel) has two separate cavities operating at a wavelength of 405 nm and at subambient pressure (720 to 770 hPa) to prevent condensation of water on inlet lines under humid conditions. One of the cavities is connected to an inlet (perfluoroalkoxy alkane (PFA) tubing) at ambient temperature, and the other cavity is con- nected to a tubular quartz inlet, which was heated to 850 • C to thermally dissociate NO y trace gases to NO or NO 2 . The TD oven was accommodated in an aluminium box on top of the container with the inlet ca. 1.2 m above the container roof. Air samples reached the TD area less than 30 cm behind the tip of the inlet, and we expect negligible inlet losses for NO y species. Inlet lines of the heated and the ambient temperature channel were each overall ca. 4 m long (2 m located inside and 2 m outside the container). In the Red Sea and Arabian Gulf, the inlet heating of the NO y channel was switched off occasionally during the hottest hours of the day to prevent damage to the oven electronics. The campaign data coverage for NO y is 65 %, considering only time periods when the ship was moving. The total uncertainty (at 50 % relative humidity and 1 min integration time) amounts to 11 % + 10 pptv for NO x and to 16 % + 14 pptv for NO z if we disregard the non-quantitative detection of coarse-mode non-refractory nitrate (see below). Detection limits (5 s integration time) during the AQABA campaign were 98 pptv for NO x , 51 pptv for NO y , and 110 pptv for NO z and are higher than those reported for laboratory operation owing to problems with optical alignment due to the motion of the ship. Detection limits are defined as the 2σ standard deviation between consecutive zeroing periods. Under laboratory conditions, NO x detection limits of 40 pptv (1 min average) were obtained ; 6 pptv (40 s) has been achieved with undegraded mirrors (Thieser et al., 2016).
The NO z mixing ratios obtained using the TD-CRDS were calculated from the difference between NO y and NO x measurements and thus contain a contribution from particulate nitrate. Friedrich et al. (2020) have shown that this instru-ment measures ammonium nitrate quantitatively but detects only a fraction (≈ 25 %) of sodium nitrate (NaNO 3 ) of 200-300 nm diameter as NO x . The inefficient detection of some non-refractory nitrate species (e.g. NaNO 3 ) means that the NO y mixing ratios presented below are thus (potentially) lower limits. As NaNO 3 is usually associated with coarsemode aerosol (particle diameter > 1 µm), this also implies that the particle-phase nitrate measured by the TD-CRDS is comparable to that measured by an aerosol mass spectrometer (HR-ToF-AMS; see Sect. 2.3). In marine environments, sea salt aerosol can be the dominant aerosol component (Lewis and Schwartz, 2004). We therefore note that the definition of NO y , in this work, is restricted to non-refractory nitrate particles which can be vaporised by the AMS or in the TD inlet of the CRDS. Nitrate detection by the AMS is further discussed in Sect. 3.2.2 and 3.4.
High loadings of coarse-mode particles are associated with high wind speeds, which were encountered on the first leg passing the Strait of Bab al-Mandab, through the Arabian Sea, and until the Gulf of Oman and on the second leg in the Arabian Sea and in the northern Red Sea. The fractional contribution of coarse-mode particles to the overall mass concentration were derived using data from an optical particle counter (OPC) and via the (PM 10 -PM 1 ) / PM 10 ratio (both PM 1 and PM 10 were measured with the OPC). We see from Fig. S1 that the impact of coarse-mode nitrate may have been largest on both legs in the transitional area between the southern Red Sea and Arabian Sea, where OPC PM 10 mass concentrations exceeded 150 µg m −3 and the coarse-mode fraction was consistently > ca. 90 %.

CLD measurements of NO x
NO and NO 2 were measured with a chemiluminescence detector (CLD 790 SR, ECO PHYSICS, 5 s time resolution) as described in Tadic et al. (2020), with total measurement uncertainties of 6 % (NO) and 23 % (NO 2 ) and detection limits of 22 pptv for NO and 52 pptv for NO 2 , both calculated at a time resolution of 5 s and a confidence interval of 2σ . The CLD detection method is based on the chemiluminescence of electronically excited NO * 2 formed in the reaction of NO with O 3 . Ambient NO 2 is photolytically converted to NO by exposure to UV light from LEDs emitting at wavelengths close to 398 nm. The CLD was calibrated every 6 h using a 2 ppmv NO gas standard.

Other measurements
An overview of the instruments deployed is given in Table 1. Total organic nitrates (ONs) were measured as the sum of peroxy nitrates (PNs, RO 2 NO 2 ) and alkyl nitrates (ANs, RONO 2 ) in a five-channel thermal dissociation cavity-ringdown spectrometer (5C-TD-CRDS; Sobanski et al., 2016). SO 2 and ClNO 2 were measured with a chemical ionisation quadrupole mass spectrometer (CI-QMS) with 15 s time resolution (Eger et al., 2019a, b). The detection limits for SO 2 and ClNO 2 were 38 and 12 pptv, respectively; the total uncertainties were 20 % ± 23 pptv (SO 2 ) and 30 % ± 6 pptv (ClNO 2 ). Particulate-phase nitrate (pNit) and sulfate concentrations in the PM 1 size range were obtained by an aerosol mass spectrometer (Aerodyne HR-ToF-AMS; DeCarlo et al., 2006) with measurement uncertainties of 30 % and 35 %, respectively, for the mass concentrations of NO − 3 and SO 2− 4 . Total aerosol mass concentrations in the PM 1 and PM 10 size ranges were calculated from particle size distributions, detected with an optical particle counter (OPC, Grimm model 1.109; size range: 250 nm to 32 µm) in a 6 s time resolution and with a 35 % uncertainty. Ozone was measured by optical absorption at 253.65 nm in a commercial ozone monitor (2B Technologies model 202) with total measurement uncertainty of 2 % ± 1 ppbv and a detection limit of 3 ppbv (at 10 s integration time). HONO mixing ratios were measured by a longpath absorption photometer (LOPAP; Heland et al., 2001) with a 3-5 pptv detection limit and a measurement uncertainty of 20 %. The path length of the instrument was 1.9 m, and the inlet was also located on the foredeck of the ship at a ca. 5 m distance to the TD-CRDS inlets. A spectral radiometer (Metcon GmbH) measured wavelength-resolved actinic flux, which was converted to photolysis rate constants (J ) for NO 2 , NO 3 and HONO using evaluated quantum yields and cross sections (Burkholder et al., 2015). The overall uncertainty in J is ca. 15 %, which includes calibration accuracy (Bohn et al., 2008) and the neglect of upwelling radiation from the sea surface. OH concentrations were obtained from a custom-built laser-induced fluorescence (LIF) instrument (Martinez et al., 2010;Regelin et al., 2013) with an upper limit total uncertainty of 40 %. Total OH reactivity measurements were performed according to the comparative reactivity method (Sinha et al., 2008), with a 5 min detection limit of 5.4 s −1 and a ca. 50 % total uncertainty, as described in Pfannerstill et al. (2019). HCHO was detected by a commercial instrument (AL4021, Aero-Laser GmbH) according to the Hantzsch method and had a relative uncertainty of 13 % (Stickler et al., 2006). Multi-pass absorption spectroscopy using a quantum cascade laser was used to measure CO mixing ratios with 20 % uncertainty and a limit of detection of 0.6 ppbv (Li et al., 2013).

Meteorological data
Temperature, wind direction, wind speed, and relative humidity were measured by a weather station (Neptune, Sterela), together with the GPS position and velocity of the ship. Back-trajectories were obtained using the HYSPLIT transport and dispersion model (Stein et al., 2015;Rolph et al., 2017). The trajectories were calculated backwards for 48 h from the GPS location of the ship with a starting height of 100 m a.m.s.l., using the Global Data Assimilation System (GDAS1) meteorological model. The back-trajectories were limited to 48 h as this exceeds the lifetimes of both NO x and NO z (see later) and is thus sufficient to indicate potential source regions. Back-trajectories displayed in graphs are considered to be representative for the prevailing atmospheric flow conditions when passing the respective areas along the AQABA ship track.

Results and discussion
In Fig. S2 we show the complete NO x , NO y , and NO z time series from the campaign, averaged from the 5 s raw data time resolution onto a 5 min grid. Periods of contamination by the ship's own exhaust are indicated by grey background colouring. The regional variation in NO x and NO z during the second leg is illustrated in Figs. S3c and 1 which also delineates the campaign into the "Red Sea" (2-16 July and 17-24 August 2017), the "Arabian Sea" (16-24 July and 7-17 August 2017), the "Arabian Gulf" (24-31 July and 3-7 August 2017), and the "Mediterranean Sea" (24-31 August 2017). Altogether, 4.8 % of the NO x measurements during AQABA were below the ca. 100 ppt detection limit of the TD-CRDS instrument, indicating only sporadic occurrence of maritime background conditions. Similar observations were made by Tadic et al. (2020), with only 3.3 % of the NO x data set below 50 pptv in the Arabian Sea, the southern Red Sea, and the eastern Mediterranean. In comparison, NO x mixing ratios below 20 pptv were previously found, for example, over the South Atlantic . The black lines in Fig. 1 represent 2 d back-trajectories (HYS-PLIT; see Sect. 2.4). A similar figure for the first leg is given in Fig. S3. For the Mediterranean Sea, the Red Sea, and the Arabian Gulf, we present an analysis of the lifetimes and sources of NO x and NO z . Chemical sources of NO x , e.g. from the photolysis of HONO or pNit, are discussed in Sect. 3.4. The chemically distinct regions are compared and contrasted in Sects. 3.5 and 4. Dividing the analysis into the three regions helps to highlight the chemically different environments encountered. An analysis of the Arabian Sea region was unfortunately not possible due to a gap in the NO z measurements between 9 and 17 August 2017, caused by instrument failure during heavy seas and winds. The division of the regions was based on the prevalent NO x mixing ratios displayed in Fig. S3c. In contrast to other AQABA publications (Eger et al., 2019a;Pfannerstill et al., 2019;, the Gulf of Oman, and the Suez Channel were included in the Arabian Gulf and the Red Sea regions, respectively, as a clear shift in NO x to mixing ratios below ca. 1 ppbv occurred both upon leaving the Gulf of Oman into the Arabian Sea and upon exiting the Suez Channel to the north towards the Mediterranean Sea. The transitions between the Arabian Gulf and the Gulf of Oman, and between the northern Red Sea and the Suez region are less obviously represented in the NO x levels.

Mediterranean Sea
Owing to unfavourable winds resulting in contamination of the measurements by the ships own exhaust as well as instrument malfunction, very little useable data were obtained by the TD-CRDS during the first leg through the Mediterranean Sea, and we analyse only the data obtained on the return leg (24-31 August 2017). In this period, temperatures varied between 24 and 29 • C with relative humidity between 52 % and 89 % (see Fig. S2). During most of the transit through the Mediterranean Sea, winds were from the north. At the end of the cruise when approaching Sicily, we encountered a shift in wind direction with air arriving from the north-west. Backtrajectories (see Fig. 1) indicate that when sailing through the eastern Mediterranean Sea we encountered air masses that had passed over Turkey; the air we sampled in the central Mediterranean Sea had passed over the Balkan states, and in the western Mediterranean it had passed over Greece and Italy. The trajectories ending at the ships location were persistently located in the boundary layer (height < 1000 m) for the previous 48 h. An exception was the back-trajectory originating from the Black Sea, which was located at a height (above ground level) of up to 1740 m. The back-trajectory passing over the island of Crete was located at a maximum height of 3224 m, which may be the result of orographic uplift caused by the central Cretan mountain range.
3.1.1 NO x NO x mixing ratios were generally low in the Mediterranean Sea (Fig. 2a). One-minute mean and median mixing ratios of NO x as detected by the TD-CRDS were 1.3 and 0.3 ppbv, respectively. For the CLD measurements of NO x , the equivalent values are 1.1 and 0.2 ppbv, respectively. For both instruments, the difference between mean and median values stems from the frequent occurrence of NO x plumes resulting from emissions of nearby ships. The NO x mixing ratios measured by TD-CRDS and CLD were in good agreement (see Friedrich et al., 2020) and the bias of the TD-CRDS to higher values reflects the exclusion of data below the detection limit. A histogram of the NO x measurements made by the CLD is displayed in Fig. 2b), which indicates that 33 % of the NO x data were between 100 and 250 pptv and 24 % were above 1 ppbv. The maximum mixing ratio of NO x in the Mediterranean Sea of 84.7 ppbv was measured in the narrowest part of the Strait of Messina, which is a busy corridor for international shipping with ferry traffic between Italy and Sicily crossing the Kommandor Iona's ship track. This observation highlights the importance of NO x shipping emissions in some parts of the Mediterranean Sea, which we return to later.
Potential non-shipping sources of NO x in this region can be identified via the back-trajectories plotted in Fig. 1. In the eastern part of the Mediterranean Sea, the air masses were influenced by emissions from the heavily populated and industrialised western Turkish coastal area, the island of Crete, and mainland Greece. However, as we show below, the lifetime of NO x is generally less than 6 h, and the greater fraction of any land-based NO x emissions would have undergone oxidation to NO z during the 48 h transport time of the back-trajectory. In the western Mediterranean Sea, the 2 d back-trajectories end above the open ocean.
Our data can be compared to results from previous measurements of NO x in the Mediterranean area. Excluding pollution events, Mallik et al. (2018) report NO and NO 2 levels below 0.05 and 0.25 ppbv, respectively, during the 2014 Cyprus-based CYPHEX (CYprus PHotochemical EXperiment 2014) campaign in the eastern Mediterranean Sea. Plume-like increases in NO x were associated with enhanced SO 2 and related to emissions from shipping (Eger et al., 2019b). During the MINOS (Mediterranean INtensive Oxidant Study) campaign on the island of Crete, median NO 2 mixing ratios between 0.3 ppbv and 0.7 ppbv were reported (Berresheim et al., 2003). The lower mixing ratios were associated with air masses arriving from the western European free troposphere, whereas the higher values were air masses impacted by biomass burning in eastern Europe. In contrast, higher NO 2 mixing ratios (typically between 4 and 6 ppbv excluding plumes) were reported from shipboard measurements in the Aegean Sea (Večeřa et al., 2008). Satellite-based observations of NO 2 vertical column densities over Crete and in the region between Crete and Sicily, were used to derive near-surface NO 2 mixing ratios of up to ∼ 0.4 ppbv (Ladstätter-Weißenmayer et al., 2003. Our NO x measurements are thus broadly consistent with previous measurements in the Mediterranean Sea which indicate mixing ratios of less than 1 ppbv in the absence of recent emissions from ships. The higher mixing ratios reported by Večeřa et al. (2008) are likely to be related to the close proximity of their ship to NO x sources on the European continent and denser ship traffic compared to the more southerly AQABA route through the eastern Mediterranean Sea. Figure 2c and d show a time series and histogram, respectively, of NO z for the Mediterranean Sea. The shape of the distribution indicates that NO z mixing ratios close to the detection limit were rarely measured. The mean (0.8 ppbv), median (0.7 ppbv), maximum (2.8 ppbv), and minimum NO z mixing ratios (< 0.1 ppbv) along with the narrower distribution indicate that, as expected, NO z is significantly less variable than NO x . The ratio of the median mixing ratios NO z / NO y in the Mediterranean Sea is ∼ 0.8, and, concomitantly, that of NO x / NO y is ∼ 0.2. A more detailed analysis of the relative contributions of NO x and NO z to NO y , in which we divide the Mediterranean Sea into seven subregions, is presented in the following paragraphs. The choice of these subregions was based on the existence of homogeneous NO z / NO y ratios over periods of hours to days. This approach enabled us to compare subregions with substantially different chemical regimes along the ship track but does not lend itself to the derivation of a representative NO y budget for the entire region.

NO z
The pie charts in Fig. 3 indicate the regional average contributions (in subregions M1 to M7) of reactive nitrogen species to NO y . The fractional contributions are based on measurements of NO x , NO y , gas-phase organic nitrates (ON), particulate nitrate (pNit), ClNO 2 , and HONO. HNO 3 was not measured directly but calculated from HNO 3 = NO y -(NO x + ON + pNit + ClNO 2 + HONO), where pNit refers to submicron particulate nitrate as measured by the HR-ToF-AMS. Detection of coarse-mode pNit by the TD-CRDS (see Friedrich et al., 2020) would lead to an overestimation of HNO 3 . However, given that the thermal dissociation to NO 2 of NaNO 3 particles with 300 nm diameter is inefficient (∼ 20 %) with this instrument, a significant bias by coarsemode nitrate (e.g. associated with sea salt or mineral dust) appears unlikely. The data available in each subregion did not always cover an entire diurnal cycle, which will have an impact on the fractional contributions of individual NO y species (see differences between day-and night-time chemistry in Sect. 1). We argue, however, that diurnal patterns in NO y are likely overshadowed by the variability of air mass sources. The NO y compositions presented are thus to be considered coarse estimates.
In all subregions, HNO 3 is the dominant component of NO y besides NO x and therefore the most important NO z species. Contrastingly, submicron pNit only contributes between 5.4 % (M6) and 15.5 % (M2) to NO y and ONs between 7.1 % (M4) and 16.9 % (M2). ClNO 2 only constitutes a minor part of NO y with ca. 1 % contribution in all regions where ClNO 2 was measured (M1-M6). The low mixing ratios of ClNO 2 have been attributed to high night-time temperatures and high reactivity of NO 3 which reduce the interaction of N 2 O 5 with chloride-containing particles (Eger et al., 2019a). Elevated HONO mixing ratios (up to 0.3 ppbv) were observed in regions M3 and M6 where its contributions to NO z was 3.8 % and 4.2 %, respectively. As the daytime lifetime of HONO is short (a few minutes) due to its rapid photolysis (Platt et al., 1980), HONO levels up to 0.3 ppbv imply strong sources. Elevated HONO mixing ratios in ship plumes have been observed in previous field measurements (Večeřa et al., 2008;Sun et al., 2020) and could explain the presence of HONO in subregions M3 and M6. Other sources of HONO, summarised in Elshorbany et al. (2012), include heterogeneous or photochemical reactions of NO x and NO z on various surfaces and also the photolysis of particulate nitrate . Figure 3 also plots the NO z / NO y ratio along the ship's track. The highest values with median NO z / NO y > 0.68 were found in regions M2, M4, and M7, reflecting a lack of local NO x sources as confirmed by the back-trajectories. For visual clarity, only the back-trajectories starting at the geographical centres of the respective subregions are displayed in Fig. 3. Back-trajectories starting at the ship's location 4 h before or after confirmed that the air mass origin was very similar. In contrast, the regions designated M3, M5, and M6 are influenced by fresh emissions from land-based sources and are characterised by low NO z / NO y ratios (medians < 0.55), reflecting the higher levels of NO x which contributed 52 % (M3 and M5) and 43 % (M6) to total NO y .

Lifetime and sources of NO x
In the following section, the observations of NO x in the Mediterranean Sea are analysed in terms of its production and loss. Following the considerations in Sect. 1, we com-pare the daytime loss of NO x via the reaction between NO 2 and OH (Reaction R3; expected to dominate over other daytime NO x loss processes in the marine environment) with night-time losses via the reaction between NO 2 and O 3 (Reaction R8): where k NO 2 represents the total loss rate constant (in s −1 ) for NO 2 and is the inverse of the NO 2 lifetime (τ NO 2 ). The first term on the right-hand side of this expression is most important during the day when OH levels were high (up to 1.4 × 10 7 molec. cm −3 at local noon) but relatively unimportant at night. In contrast, the second term on the right is only important at night as the NO 3 product of R8 is rapidly photolysed back to NO x during daytime so that NO x is conserved. By using Eq.
(1) to approximate the NO 2 loss rate constant, we neglect two further processes which can, under some conditions, influence the lifetime of NO 2 . Our approach assumes that the night-time formation of NO 3 leads to the removal of one NO 2 molecule. This approach would be invalid if a significant fraction of NO 3 would be lost via formation (and subsequent heterogeneous loss) of N 2 O 5 . Firstly, we note that formation of N 2 O 5 was hindered during AQABA by the high gas-phase reactivity of NO 3 towards VOCs (Eger et al., 2019a) and that the transfer of N 2 O 5 to the particle phase was hindered by high temperatures. For example, taking an N 2 O 5 uptake coefficient γ N 2 O 5 of 0.03 (as found for polluted marine environments by Aldener et al., 2006) and the median night-time aerosol surface area (ASA) in the Mediterranean Sea of 1.78 × 10 −6 cm 2 cm −3 (Eger et al., 2019a), we estimated a loss rate constant for uptake of N 2 O 5 of 3.5×10 −4 s −1 , which is 2 orders of magnitude lower than the rate constant (4.9 × 10 −2 s −1 ) for thermal decomposition at 25.7 • C (the mean, minimum night-time temperature in the Mediterranean Sea).
We also neglect the loss of NO x via uptake of NO 2 onto black carbon (BC) particles. Using a literature uptake coefficient γ NO 2 of ca. 1 × 10 −4 (Longfellow et al., 1999) and the aforementioned ASA, the first-order loss rate constant for the heterogeneous uptake would be 1.8 × 10 −6 s −1 . Using an O 3 mixing ratio of 63.4 ppbv (equal to night-time median mixing ratio in the Mediterranean Sea), we calculate a firstorder loss rate constant for the reaction of NO 2 and O 3 of 5.5 × 10 −5 s −1 , which implies that > 95 % of total NO 2 loss at night-time NO 2 is due to O 3 . Uptake of NO 2 might, therefore, be relevant for HONO formation (see Sect. 3.4) but does not constitute a relevant loss process for NO x .
In order to fill gaps in the OH data set (daytime data coverage of 71 %), complete diel cycles of OH were generated by scaling measurements of J O 1 D to the OH noon-time maxima. Figure S4 compares the measured OH concentrations with the interpolated trace and shows that the thereby derived OH levels can be considered upper limits. Inserting these values and the measured O 3 concentration into Eq. (1) and using preferred rate coefficients for k 3 and k 8 (IUPAC, 2020) we derive lifetimes (Fig. 4a) of ∼ 2 h at local noon (largest OH levels) and 5-6 h at night. Loss of NO x by deposition may be important in forested regions (Delaria et al., 2018;Delaria and Cohen, 2020) but is expected to be insignificant in a marine environment. The relative importance of day-and night-time losses of NO 2 in the Mediterranean Sea during AQABA was estimated by integrating the two loss terms using the available NO 2 , O 3 , and OH data. Averaged over the 6 d of measurements, 3.71 ppbv of NO x was lost per 12 h day and 1.51 ppbv was lost per 12 h night (Fig. 4c).
Although our conclusion is based on a limited data set, we calculate that the OH-induced daytime loss of NO x is most important in the Mediterranean Sea, reflecting the high levels of OH encountered during AQABA, but we note that night-time losses make a significant contribution. It is very likely that in other seasons with reduced photochemical activity and lower temperatures (which favour the formation of N 2 O 5 which can remove two NO 2 molecules via heterogeneous processes), the night-time losses gain in relative importance. Averaged over the entire data set obtained in the Mediterranean Sea, we calculate a lifetime of NO 2 of 3.9 h. Chemical sources of NO x in the Mediterranean Sea, i.e. from the photolysis of HONO and pNit, as well the reaction of OH and HNO 3 , are discussed in detail in Sect. 3.4.
In the following, we examine the contribution of ship emissions to the NO x budget in the Mediterranean region and especially along the track taken by the Kommandor Iona during the AQABA campaign. In Fig. S5 we plot a time series of NO x and SO 2 data for the transit through the Mediterranean Sea. It is immediately apparent that large plume-like features in NO x coincide with similar features in SO 2 . We now separate the data set into two regimes in which the NO z and NO y measurements indicate either relatively "fresh" emissions (NO z / NO y ratio < 0.4) or relatively "aged" emissions (NO z / NO y ratio > 0.8). In Fig. 5a we show that, for fresh emissions, SO 2 and NO x are highly correlated (Pearson's R = 0.84) with a slope of 4 ± 0.1 ppbv NO x per ppbv SO 2 and an intercept (at zero SO 2 ) of −1.9 ± 0.3 ppbv. This strongly suggests that fresh NO x emissions are generally accompanied by SO 2 and thus indicates that either ships or power plants, e.g. in coastal locations, are the likely sources of a large fraction of the NO x . The slope is similar to that derived by Celik et al. (2020) (2.7 ± 0.8), who examined single ship plumes in a more detailed analysis and with literature values that range from 6.8 ± 6.3 near the coast of Texas (Williams et al., 2009) to 11.2 ± 10.9 (Diesch et al., 2013) at the Elbe river near Hamburg/Germany. In comparison to Celik et al. (2020), however, the two other literature studies only sampled very fresh and unprocessed ship plumes, from a distance of less than ca. 5 km to the emission source. Fig. 5a shows that NO z and SO 2 are not correlated (Pearson's R = 0.38) in air masses impacted by fresh emissions.
In more aged air masses (Fig. 5b) the slope of NO x per SO 2 is, as expected, much smaller (0.16 ± 0.01 ppbv NO 2 per ppbv SO 2 ) which reflects the significantly longer lifetime of SO 2 (∼ 10 d) compared to NO x . After a few days of transport an air mass containing co-emitted NO x and SO 2 will still contain SO 2 but the initially emitted NO x will, to a large extent, have been converted to NO z . The intercept (NO x = 0.049 ± 0.005 ppbv at zero SO 2 ) is consistent with the re-generation of NO x from NO z (see above) but is also in the area of the detection limit of the NO x measurement.
The plot of NO z versus SO 2 for aged emissions indicates a significant intercept (at zero SO 2 ) of 0.4 ppbv NO z . As the lifetime of SO 2 (∼ 10 d) is longer than of NO z (∼ half a day) (Dickerson et al., 1999;Romer et al., 2016), the residual NO z at zero SO 2 cannot stem directly from ship emissions (or combustion sources that generate both NO x and SO 2 ) but represents the background level of NO z in the Mediterranean Sea in aged air masses and is consistent with an average HNO 3 mixing ratio of 0.48 ppbv observed during the MINOS campaign at Finokalia on Crete (Metzger et al., 2006).
The analysis above, when combined with back-trajectory information, provides clear evidence that shipping emissions are responsible for a large fraction of NO x in the Mediterranean Sea. The impact of shipping emissions on the atmospheric sulfur budget has been assessed in numerous stud- ies which identify coastal areas and international shipping lanes as important hotspots for SO 2 emissions (Capaldo et al., 1999;Dalsøren et al., 2009;Eyring et al., 2010), with emissions of SO 2 severely impacting air quality in port regions (Isakson et al., 2001;Cooper, 2003;Saxe and Larsen, 2004;Marmer and Langmann, 2005;Ledoux et al., 2018). A detailed analysis of SO 2 data with regard to ship emissions during AQABA is provided by Celik et al. (2020), who analysed emission factors from individual ship plumes during the AQABA campaign.

Red Sea
Measurements over the Red Sea (from the Suez Canal and the Strait of Bab al-Mandab) were made from 2-16 July 2017 on the first leg and 17-25 August 2017 on the second leg. On the first leg, the Kommandor Iona reversed direction in the northern Red Sea three times (twice for 9 h and once for 6 h), in order to sail into the wind and avoid contamination by the ship's own stack. Additionally, there was a 3 d layover in Jeddah (10 to 13 July 2017). Temperatures on the first leg were usually above 27 • C, with maxima of 37-38 • C in the Suez Canal, in Jeddah, and on the approach to Bab al-Mandab. The relative humidity was usually between ca. 60 % and 80 % but dropped below 30 % in the Suez Canal and in Jeddah. Winds came predominantly from northerly directions with speeds generally between 2 and 10 m s −1 . On the second leg, temperatures were constantly above 30 • C in the southern Red Sea; relative humidities were similar to the first leg. The wind was consistently from the north, with wind speeds between 5 and 12 m s −1 until the ship reached the Suez region. During the first leg, the air masses intercepted above the northern Red Sea were impacted by emissions from Cairo and the Nile valley. Two-day back-trajectories for the southern Red Sea start in the centre of the Red Sea and do not indicate transport from the Suez region. Extended back-trajectories for the southern Red Sea showed that 3 to 4 d prior to sampling, the air parcel passed over southern Egypt, and 5 to 6 d before it was located over the Cairo area. Similar back-trajectories were obtained for the second leg. Air masses in the northern Red Sea were influenced by the Suez region, north-eastern Egypt, and Israel.

NO
x NO x mixing ratios in the Red Sea (excluding the 3 d layover in the port of Jeddah) as measured by the TD-CRDS and the CLD instruments are displayed in Fig. 6a. NO x mixing ratios were highly variable and there were only short periods free of NO x plumes > 10 ppbv (e.g. during the second leg on 19 and 20 August 2017). The mean NO x mixing ratios (2.8 ppbv measured by the TD-CRDS and 3.2 ppbv measured by the CLD) were therefore significantly higher than the median values of 1.0 ppbv. Figure 6b indicates that the NO x mixing ratios are broadly distributed around the median of 1.0 ppbv with 21 % of all data points > 3 ppbv. The highest NO x levels during AQABA were found in narrow shipping corridors of the Suez region and the Strait of Bab al-Mandab. When excluding the Suez and Bab al-Mandab regions, a median NO x mixing ratio of 0.7 ppbv can be derived for the maritime central part of the Red Sea.
To the best of our knowledge, in situ measurements in the Red Sea area are not available for comparison with our NO x data. Satellite-based modelling studies show that high NO 2 column densities above the Red Sea are associated with shipping emissions (Richter et al., 2004;Alahmadi et al., 2019), which is consistent with our observation of a strong correlation between NO x and SO 2 (see below). Johansson et al. (2017) have estimated a NO x emission rate of 0.70 t km −2 yr −1 for the Red Sea (including the Suez region).

NO z
The mean mixing ratio of NO z over the Red Sea was 1.0 ppbv, with a maximum value of 8.0 ppbv measured in the Gulf of Suez on the first leg. NO z mixing ratios are narrowly distributed (see Fig. 6d) around a median value of 0.7 ppbv, with 53 % of the measurements between 0.4 and 1.0 ppbv and 41 % between 1.0 and 4.0 ppbv.
The NO z / NO y ratios along the ship's track are plotted in Fig. 7: values > 0.6 were mostly observed over the northern Red Sea on the first leg, after leaving the Gulf of Suez. On the second leg, the NO z / NO y ratio was higher in the southern Red Sea. NO z data coverage was limited in the Red Sea on both legs and the NO z / NO y ratio was more variable than values found in the Mediterranean Sea and the Arabian Gulf. The high variability in the NO z / NO y ratios is caused by the route of the Kommandor Iona along the main shipping lane connecting the Suez Canal and the Gulf of Aden and the frequent sampling of plumes from nearby ships. The observed NO z / NO y ratios of < 0.6 in the Red Sea highlight the impact of NO x emissions from shipping on the reactive nitrogen budget and the air quality in the Red Sea region (as discussed in Sect. 3.2.1).
For the Red Sea, we have defined four subregions in which we calculate the contributions of NO x and various NO z species to NO y : these are RS1 on the first leg and RS2, RS3, and RS4 on the second leg. Note that RS1 and RS4 are both located in the northern Red Sea, but the measurements (∼ 5 weeks apart) revealed different chemical characteristics; hence the separate treatment.
Due to poor data coverage, mainly of organic nitrates, we were not able to perform this calculation in further subregions on the first leg. In all four regions, NO x was the largest component of NO y which results from continuous NO x input from onshore and shipping emissions.
In RS1 we observed the lowest contribution (36.4 %) of NO x to NO y and the largest contribution of ONs (23.8 %) to NO y , over the Red Sea. The latter value is the highest found during the entire AQABA campaign and is comparable to the contribution of HNO 3 (30.0 %). In roughly colocated RS4, but 5 weeks later, the NO x contribution was much larger (69.5 %). The divergent median NO x / NO y and NO z / NO y for subregions RS1 and RS4 can be understood . The pie charts indicate the components of NO y at various segments along the ship's track (ONs = organic nitrates, pNit = particulate nitrate). HNO 3 was calculated via HNO 3 = NO z -(ONs + pNit + ClNO 2 + HONO). The colours of the pie chart segments are assigned as follows (clockwise): pNit in yellow, NO x in red, HNO 3 in green, ONs in blue, ClNO 2 in grey, and HONO in magenta (© Google Maps). when one examines the air mass back-trajectories for the two legs. On the second leg, strong northerly winds transported NO x from the highly polluted southern end of the Gulf of Suez to RS4, whereas during the first leg the back-trajectory for RS1 passed (with lower wind speeds) mainly over eastern Egyptian deserts, with emissions from Cairo requiring 36 h to reach RS1 during which a significant fraction of NO x was converted to NO z . We expect that the large contribution of ONs in RS1 is a result of the unique chemical environment at the southern end of the Gulf of Suez and in the northern Red Sea. A large coherent oil field is located south of the Gulf of Suez and the coast of eastern Egypt (Alsharhan, 2003), and the numerous facilities for oil extraction result in abundant emissions of VOCs, while the proximity to the Gulf of Suez and the narrowing shipping corridor on the ap-proach to Suez provides the NO x required for formation of organic nitrates (ONs). Meteorological conditions additionally favoured a build-up of ONs during our passage through RS1: elevated wind speeds of up to 11 m s −1 coincided with temperatures below 30 • C, which slowed down the thermal decomposition of PAN (peroxyacetyl nitrate) compared to the ca. 35 • C regime in the Arabian Gulf. Average PAN mixing ratios, as measured by chemical ionisation mass spectroscopy (CIMS), were 190 pptv in this area, which constitutes ca. 20 % of the total ONs signal. On the second leg in RS4, the fractional contribution of ONs was overshadowed by the stronger impact of NO x pollution from the Suez region (see above).
In RS1 and RS4 the contributions of HONO and ClNO 2 to NO y were minor (≤ 3 %). RS2 and RS3 are both located in the southern half of the Red Sea. For RS3 we observed the highest contribution (15 %) of AMS-measured particulate nitrate to NO y , and RS3 was characterised in large parts by coarse-mode OPC fractions > 85 % (i.e. (PM 10 -PM 1 ) / PM 10 ; see 18 and 19 August 2017 in Fig. S1). It is reasonable to assume that the coarse-mode particle mass concentrations in this area was due to sea salt, which reacts heterogeneously with HNO 3 to form particle-phase nitrates (Mamane and Gottlieb, 1990). Refractory sea salt aerosol particles in the PM 1 size range are, however, not expected to be detectable via AMS  or with only very low efficiency (ca. 1 %) (Zorn et al., 2008).
Region RS2 shows a intermediate behaviour, as NO z / NO y increases after leaving Bab al-Mandab and transported air only came from the surrounding southern Red Sea without being influenced by shore-side anthropogenic activities. Here, NO x and HNO 3 contribute 57 % and 27 %, respectively. The relatively high NO x contribution, considering the remote area, can be explained by sampling ship plumes on the departure from Bab el-Mandab, which led to several NO x peaks above 10 ppbv (see Fig. 6a). Consequently, background NO x levels also did not fall below ca. 1.5 ppbv on the night from 17 to 18 August 2017. Overall, the fractional contributions of NO x were positively biased by short-term spikes in NO x mixing ratios caused by ship plumes in all Red Sea subregions. The use of mean values to assess the fractional contributions of NO y species in certain subregions is thus a caveat of this analysis, as NO z signals exhibit less variability during pollution events (see Fig. 6c). Employing the median values, however, would not allow the relative contributions to NO y to be assessed.

Lifetime and sources of NO x
Analogous to Sect. 3.1.3, we now investigate the day-and night-time chemical losses of NO 2 in the Red Sea (see Fig. 8). As described previously, we used an interpolated OH data set based on a scaling factor between the available OH data and J O 1 D . As OH was not measured over the Red Sea on the first leg, our analysis is restricted to the second leg only. Daytime NO 2 lifetimes with respect to loss by reaction with OH were usually in a range between 2 and 4 h, with a minimum of 1.7 h on 21 August 2017, where the noon-time OH concentration peaked at 1.1 × 10 7 molec. cm −3 . Night-time NO 2 lifetimes (determined by O 3 levels) exhibited a larger variability but were mostly between 5 and 10 h. The average (day and night) NO 2 lifetime in the Red Sea was 5.0 h.
Over the entire period of measurements in the Red Sea (8 d and 8 nights) we calculate that a cumulative total of 62 ppbv of NO 2 were lost (Fig. 8c). Despite the shorter lifetime of NO 2 at noon, the greater integrated loss of NO 2 occurred during night-time (5.7 ppbv per night on average) when continually high O 3 levels (median 54 ppbv) were available. At midday, NO 2 mixing ratios are reduced due to the shift in the NO 2 / NO ratio caused by the rapid photolysis of NO 2 and also because the OH levels are highest then. On average, daytime loss rates were 2.0 ppbv per day.
In order to assess the contribution of shipping on NO x emissions, we correlated NO x and SO 2 mixing ratios for freshly emitted (NO z / NO y < 0.4) and chemically more aged (NO z / NO y > 0.8) air masses. The results are illus- trated in Fig. 9 and summarised in Table 2, which reveal a positive correlation (slope of 3.7 ± 0.1 and a regression coefficient R of 0.61) between NO x and SO 2 in air masses containing freshly emitted pollutants. Six data points far above 20 ppbv (range 43-128 ppbv SO 2 ) were excluded, as they would bias the linear regression result. Including these data points lowers the slope to 1.26 ± 0.04 and the correlation coefficient R to 0.40. The NO x / SO 2 ratio is thus highly variable throughout the Red Sea, potentially reflecting variable NO x / SO 2 emission ratios of different vessels, using various fuels, as well as the impact (on NO x ) of offshore oil-drilling rigs and shore-side oil refineries. The latter are most important in the northern Red Sea, whereas shipping emissions dominate in the narrow shipping lanes of the Suez Canal.
For chemically aged air masses, the NO x / SO 2 ratio is 0.20 ± 0.01 with R = 0.61 and the reduction in slope reflecting the shorter lifetime of NO x compared to SO 2 . We find, however, that in chemically aged air masses, NO z and SO 2 are highly correlated (Fig. 9b) with a slope NO z / SO 2 of 1.25 ± 0.04 and R = 0.85. The intercept (see Fig. 9b) at an SO 2 mixing ratio of zero is 0.40 ± 0.03 ppbv, which can be taken to be the regional NO z background mixing ratio (i.e. NO z formed from NO x which was not emitted from SO 2containing fuels).

Arabian Gulf
Data over the Arabian Gulf (see Fig. 10) were obtained from 24 to 31 July 2017 (first leg) and 31 July to 3 August 2017 (second leg). During the 4 d layover in the harbour of Kuwait, the TD-CRDS was not operational. The highest temperatures during the AQABA campaign were found in the Arabian Gulf with daytime temperatures up to 46 • C at Kuwait harbour and 38-39 • C offshore. Night-time temperatures were constantly above 30 • C on both legs. Offshore relative humidities were between 60 % and 90 % during both legs; wind speeds were generally below 6 m s −1 and frequently 1-2 m s −1 . The Arabian Gulf crossing was divided into four subregions: A1 and A2 on the first leg and A3 and A4 on the second leg (see Fig. 11). Air mass back-trajectories indicated that air sampled in the Gulf of Oman originated in Oman; the south-eastern Arabian Gulf was influenced by transport from the central Arabian Gulf and Saudi Arabia. Inside A1, samples were affected by the eastern coast of Saudi Arabia. When approaching Kuwait (area A2), back-trajectories originated from Iraq. During the second leg, the northern Arabian Gulf region was dominated by stagnating air masses, mainly containing emissions from local sources and from the direction of Iran. Air from this area was also transported to the central Arabian Gulf, which is covered by subregion A3. Local sources from inside the shipping lane were dominant when passing the Strait of Hormuz (A4). The Gulf of Oman experienced influx from the remote Arabian Sea in contrast to the first leg.

NO x
Elevated NO x mixing ratios were detected by both TD-CRDS and CLD throughout the Arabian Gulf (see Fig. 10a). The TD-CRDS measured mean and median NO x mixing ratios of 3.3 and 1.6 ppbv, respectively. By comparison, the CLD measured an average of 4.1 ppbv and a median of 1.8 ppbv. The large difference between median and mean reflects the numerous plumes of high NO x detected by both instruments (Fig. 10a). The deviation of the TD-CRDS and the CLD data is caused by different data coverage as the CLD continued measuring in the most polluted areas close to Fujairah and Kuwait, while the TD-CRDS was switched to zeroing mode, in order to avoid contamination of the inlet lines. When limiting the comparison to periods where both instruments were operating, very similar median values are obtained, with 1.6 ppbv from the TD-CRDS and 1.5 ppbv from the CLD. A histogram of the NO x measurements (CLD data only) made in the Arabian Gulf (Fig. 10b) shows a broad distribution, reflecting high variability in the region, with 77 % of the data points falling into a range between 0.4 and 10 ppbv and a broad maximum at 1-3 ppbv. The highest NO x daily maxima were observed near Fujairah (up to 34 ppbv on the first and 153 ppbv on the second leg), in the Strait of Hormuz (26 and 30 ppbv), and when approaching and departing Kuwait (43 and 90 ppbv). The locations of these maxima close to the shore or in narrow shipping corridors and the plume-dominated time series suggest the influence of mostly local pollution sources of NO x , i.e. from ship traffic or from industrial activities in the shore-side areas of the neighbouring cities. NO x mixing ratios < 0.5 ppbv were found exclusively in the central part of the Arabian Gulf, which is the widest part (least influence from onshore activity) with the largest spread of the shipping lanes.
The generally very high levels of NO x in the Arabian Gulf are consistent with results from satellite measurements which have identified high NO 2 tropospheric vertical column den- sities over the Gulf of Oman, the Strait of Hormuz, and the south-eastern Arabian Gulf (Beirle et al., 2004). Model studies estimate a NO x emission rate of 1.13 t km −2 yr −1 for the Arabian Gulf (Johansson et al., 2017). With a NO x lifetime of 4.0 h (see Sect. 3.3.3) and a boundary layer height of 1 km (Wu et al., 2008), this emission rate translates to a NO x mixing ratio of 0.3 ppbv. The lower mixing ratio, compared to the median NO x observed on AQABA (see above), is likely caused by the averaging of the model over the entire Arabian Gulf water surface area, whereas the Kommandor Iona followed common shipping routes with larger NO x emissions. To the best of our knowledge, there are no in situ measurements of NO x over the Arabian Gulf with which to compare our data.

NO z
The Arabian Gulf featured the highest NO z levels during the AQABA campaign (see Fig. 10c), with mixing ratios from < 0.1 ppbv up to 6.9 ppbv (mean 2.0 ± 1.5 ppbv (standard deviation) and median of 1.5 ± 0.7 ppbv (median absolute deviation)). The histogram of NO z mixing ratios (Fig. 10d) shows a maximum in the frequency distribution at 1-3 ppbv, with 73 % of all data above 1 ppbv and 15 % above 4 ppbv. Our results thus indicate that the Arabian Gulf is a hotspot for NO z formation, which is a result of high levels of the NO x and VOCs precursors and also O 3 . The spatial distribution of the NO z / NO y ratio for both legs is presented in Fig. 11. On both legs, NO z / NO y ratios above 0.8 were found in the central part of the Arabian Gulf, which results from the processing of NO x emissions during transport from the shore to the centre of the Arabian Gulf.
We now examine the partitioning of NO y into its various components in the four subregions (A1-A4) defined above for the Arabian Gulf (Fig. 11). On the approach to Kuwait (A2), winds from the north transported fresh NO x emissions from cities in Kuwait and Iraq to the ship and NO x accounted for 81 % of NO y . More aged air masses were found in other regions (A1, A3, and A4) with a roughly equal split between NO x and HNO 3 (both 45 %-50 %) observed in A3 and A4. The major component of NO z was HNO 3 in all regions, with significant but very variable contribution from organic nitrates, especially in A1 (13 ± 16 %) where the air masses originated from the eastern coast of Saudi Arabia, which ac- Figure 11. The NO z / NO y ratio over the Arabian Gulf during the (a) first and (b) second legs. Coloured lines are 2 d back-trajectories (HYSPLIT). The pie charts indicate the components of NO y at various segments along the ship's track (ONs = organic nitrates, pNit = particulate nitrate). HNO 3 was calculated via HNO 3 = NO z -(ONs + pNit + ClNO 2 + HONO). The colours of the pie chart segments are assigned as follows (clockwise): pNit in yellow, NO x in red, HNO 3 in green, ONs in blue, ClNO 2 in grey, and HONO in magenta.
commodates numerous facilities for oil and gas extraction and processing, resulting in high levels of organic trace gases including alkanes, alkenes, and aromatics . Particulate nitrate contributed only minor amounts to NO z in the Arabian Gulf, which reflects the high temperatures and resultant partitioning of nitrate into the gas phase. Other NO z species contributed only weakly to the NO z as indicated in Fig. 11.

Lifetime and sources of NO x
Analogously to Sect. 3.1.3, we also determined NO 2 lifetimes and the cumulative loss of NO 2 in the Arabian Gulf. The results are presented in Fig. 12. Limited by the availability of OH data, these calculations include only the time period after 29 August 2017 on the first leg. In the same way as in Sect. 3.1.3, we used an interpolated OH data set in the following calculations.
In the Arabian Gulf, daytime NO 2 lifetimes (considering loss by OH) were generally between 2 and 4 h. Night-time lifetimes were in a similar range but also occasionally exceeded 10 h, e.g. when leaving the Arabian Gulf towards the Gulf of Oman and the Arabian Sea on the second leg, where O 3 mixing ratios fell below 20 ppbv. The average NO 2 lifetime was calculated to be 4.0 h. Figure 12c shows that 50 ppbv of NO 2 was lost cumulatively throughout the period of measurements over the Arabian Gulf, with night-time losses (black data points) being more important than daytime losses (red data points). On average 6.0 ppbv NO 2 were lost per night and only 2.1 ppbv was lost per day. Large night-time compared to day-time losses are related to moderate OH levels in large parts of the Arabian Gulf (see Fig. 12b). The daytime average OH concentration was 2.4 × 10 6 molec. cm −3 , while on average 73 ppbv O 3 was present. The measured OH concentrations were generally low, given the NO x and O 3 levels in the Ara-bian Gulf, which may have resulted from its reactions with VOCs. With a loss rate constant of 11.6 s −1 , the Arabian Gulf was the AQABA region with the largest median OH reactivity (Pfannerstill et al., 2019), with 61 % of the total OH reactivity attributed to various measured VOCs. The daytime losses of NO 2 are therefore indirectly limited by the availability of VOCs from the oil and gas production (see above).
Via analysis of correlation between SO 2 and NO 2 ( Fig. 13 and Table 2), we can assess the influence of shipping emissions on NO x mixing ratios in the Arabian Gulf. In air masses recently influenced by NO x emissions (NO z / NO y < 0.4), NO x and SO 2 are only weakly correlated (slope = 4.1 ± 0.2, R = 0.41), indicating that many different NO x sources (i.e. not only shipping emissions) contribute. These might include vehicular traffic and industrial activity (e.g. production of nitrogen-based fertilisers Khan et al., 2016) in Kuwait City, the Iraqi city of Basra, and in Iranian harbours and offshore oil and gas terminals. Considering the limited NO x lifetime, the land-based emission sources of NO x from urban/industrialised areas gain in importance over plumes from nearby ships, when approaching the coast. In aged air masses, the slope of the NO x versus SO 2 correlation is 0.11 ± 0.01 with a large correlation coefficient (R = 0.72). This indicates that in aged air masses, the NO x levels are linked to SO 2 emissions, which is consistent with the photolysis of HONO being a major source of NO x in the region. From the intercept (SO 2 mixing ratio = zero = 0.0 ± 0.3 ppbv), we would expect negligible background levels of NO z . Overall, shipping was an important source of NO x in the Arabian Gulf, both through direct emissions and via photolysis of ship-related HONO.
3.4 NO x and NO y and the role of ship-emission-related HONO formation during AQABA In this section, we perform a steady-state analysis, assessing to what extent chemical source strengths can explain the background mixing ratios of NO x observed during AQABA. Background conditions refer to NO x mixing ratios found during periods when ship plumes were rarely encountered. "Background" NO 2 varied from region to region and was, for example, 50-150 pptv in the Mediterranean Sea. The required NO x source strength (P , in molec. cm −3 s −1 ) to maintain the observed NO x levels is derived from the measured mixing ratios [NO x ] and the NO 2 reactivity (k NO 2 ; see Sect. 3.1.3), whereby P = P chem + E is a combination of chemical production (P chem ) and direct emission (E). Notably, we neglect direct emissions under background conditions (i.e. E = 0) and assume that NO x is only lost via the reaction of NO 2 with OH (i.e. k NO x = k NO 2 ).
Chemical processes that result in the formation of NO x include the degradation of two gas-phase NO z components, HONO and HNO 3 , and the photolysis of particulate nitrate.
In a first step, we examine whether the HONO levels observed on AQABA can be explained by the photolysis of pNit in the PM 1 size range. This calculation is based on the assumption of a steady state for HONO established at noon through its photolytic loss and its production through the photolysis of pNit. Using average noon-time Mediterranean Sea concentrations for HONO (2.44 × 10 9 molec. cm −3 ) and pNit (2.93 × 10 9 molec. cm −3 ) and a photolysis rate J HONO (1.45 × 10 −3 s −1 ), we calculate that a value for J pNit of ∼ 1.21 × 10 −3 s −1 would be required in order to maintain the observed HONO concentrations. This is a factor of ∼ 5-6 higher than a reported value of J pNit ≈ 2 × 10 −4 s −1 , based on observations over the western North Atlantic Ocean (Ye et al., 2016). It is, however, unclear whether the type and age of particles examined by Ye et al. (2016) are comparable to those in AQABA. In addition, photolysable nitrate associated with particles that are > 1 µm diameter remain undetected by the AMS and could also contribute to the discrepancy between required and literature J pNit .
Laboratory studies have demonstrated the conversion of NO 2 to HONO on BC particles, with a clear enhancement under UV irradiation (Acker et al., 2006;Elshorbany et al., 2009;Monge et al., 2010;Ma et al., 2013). Monge et al. (2010) postulated the transport of HONO and NO to remote low-NO x areas, enabled via this heterogeneous mechanism. Besides the effect of irradiation, heterogeneous BCassisted HONO and NO generation also shows a remarkable humidity dependence (Lammel and Perner, 1988;Kalberer et al., 1999;Kleffmann et al., 1999). Further information on the particulate-phase chemistry of HONO can be found in comprehensive reviews by Ma et al. (2013) and George et al. (2015). Sources of HONO during the AQABA campaign will be discussed in more detail in a separate publication.
Using Eq.
(2), we now calculate what values of P chem are required to maintain the background levels of NO x observed and assess the individual contributions from Reactions (R12)-(R14) (results presented in Table 3). The analysis was restricted to data points where NO z / NO y was greater than 0.6 and to the 4 h time frame around local noon, in order to focus on aged air mass conditions during maximum photochemical activity. NO 2 reacting with O 3 was not considered a NO x loss mechanism, due to the rapid reformation of NO x by the photolysis of NO 3 . For the pNit photolysis rate constant to form NO x , we used 0.33 · J pNit (from Ye et al., 2016), which accounts for the HONO / NO x production ratio of 2 : 1. Additionally, we scaled J pNit with J HONO (normed to the average daytime maximum of J HONO ) to introduce diurnal variability. Due to limited data availability and rare occurrence of NO z / NO y > 0.6 (i.e. sampling of aged air) in the other regions, we performed this calculation for the Mediterranean Sea only. The results indicate that the measured HONO concentrations should result in a factor ca. 4.7 times larger NO x production term than calculated via Eq. (2). Possible explanations for this include a positively biased HONO measurement or the underestimation of NO x losses, e.g. due to undetected OH (despite the upper limit chosen in the interpolation). Our measurements and calculations, nonetheless, allow for the qualitative identification of HONO photolysis as a major source of daytime background NO x levels during AQABA. The production rate from pNit photolysis can also account for ca. 64 % of the chemical NO x generation, whereas the reaction of OH and HNO 3 forms an order of magnitude less NO x .
Throughout AQABA, shipping emissions were responsible for fresh input of pollutant NO x into the atmosphere. Our observations that levels of NO x (with a lifetime of a few hours) were correlated with SO 2 (with lifetimes of more than a week) levels even in aged air masses and that HONO photolysis was an important source of NO may be reconciled by Table 3. Average required production rates to maintain the observed NO x mixing ratios in aged air masses during the Mediterranean Sea transit and contributions from processes (R12)-(R14).

Mediterranean Sea
P chem ± SD (10 5 molec. cm −3 s −1 ) 2.8 ± 2.2 P (HONO + hν) ± SD (10 5 molec. cm −3 s −1 ) 13.1 ± 9.1 P (pNit + hν) ± SD (10 5 molec. cm −3 s −1 ) 1.8 ± 0.4 P (OH + HNO 3 ) ± SD (10 5 molec. cm −3 s −1 ) 0.14 ± 0.06 Number of data points (5 min averages) 90 considering that HONO (and thus NO x ) production is driven by heterogeneous photochemistry on nitrate-containing particulate matter, the formation of which is associated with emissions of NO x and SO 2 as well as black carbon. The latter has a lifetime in the boundary layer (defined by its deposition) of about a week or longer in the absence of precipitation and is thus comparable to that of SO 2 . The slow, photochemically induced conversion of nitrate to HONO thus provides a long-lived source of NO x and a link with SO 2 , together with an explanation for the detection of short-lived HONO even in processed air masses in the eastern Mediterranean Sea. We emphasise that the analysis presented here focussed on the daytime chemistry of HONO. At night-time, a pseudo stationary state, independent of fresh NO x input, has been observed by Wojtal et al. (2011) and explained with a reversible deposition of HONO on marine surfaces. This will, however, be insignificant during the day. Ship-derived HONO also has a substantial effect on the rates of photochemical O 3 formation in the remote marine boundary layer, largely as a result of higher RO x production rates (Dai and Wang, 2021).
3.5 Inter-regional ozone production efficiency (OPE) The OPE can be calculated from the relationship between O x and NO z where O x = O 3 + NO 2 , and the O 3 mixing ratios are augmented by those of NO 2 , 95 % of which potentially photolyses to O 3 (Wood et al., 2009). Note that in any air mass where HNO 3 is a major component of NO z , the derived OPE may represent an upper limit if HNO 3 is lost during transport from the NO x source region to the measurement location. The NO y / CO ratio has been used to estimate the impact of NO z losses on the values of OPE obtained in this type of analysis  the rationale being that CO (like O 3 ) is a product of photochemical activity and relatively long lived, at least compared to NO z . The high variability in the NO y / CO ratio during AQABA is, however, indicative of local (non-photochemical) sources of CO (e.g. via combustion) and precludes use of this corrective procedure so that the values of OPE we present should be regarded as upper limits. An introduction into the OPE metric and on typical literature values is given in Sect. 1. In Fig. 14 we plot O x versus NO z for which the NO 2 photolysis rate constant was > 1 × 10 −3 s −1 , which restricts  Table 4 were derived from linear fits of these data points. A clear regional variability can be observed for O x and NO z mixing ratios. Elevated O x and NO z levels were measured in the Arabian Gulf and the Red Sea. the analysis to hours of the day with active photochemistry. Regional OPE values are 10.5 ± 0.9 for the Red Sea, 19.1 ± 1.1 for the Arabian Gulf, and 15.4 ± 2.4 for the eastern Mediterranean Sea. The heterogeneity of NO z and O 3 mixing ratios, i.e. the chemical conditions frequently varying between aged and plume situations (see Sect. 3.1), resulted in a low correlation coefficient in the western Mediterranean Sea (R 2 = 0.19), which precluded derivation of an OPE for this region and led us to restrict the Mediterranean Sea OPE analysis to the more homogenous eastern part (encompassing subregions M1-M5).
The range of OPE values measured during AQABA (10.5-19.1) is comparable to the value of 10, derived in the marine boundary layer (MBL) at Oki Islands, Japan, a site which is influenced by pollution arriving from the Korean peninsula and the Japanese mainland (Jaffe et al., 1996), but much lower than the value of 87 which was derived from observations off the coast of Newfoundland (Wang et al., 1996), where the median NO x mixing ratio was < 100 pptv. As alluded to above, high values in remote locations may in part be a result of reactive nitrogen loss via deposition. By comparison, during AQABA the median NO x mixing ratio was > 600 pptv, which together with the relatively low OPE indicates that the vast majority of the AQABA ship track cannot be considered representative of remote MBL conditions. Figure 14 and Table 4 indicate that the Arabian Gulf, for which the highest O 3 levels in the entire campaign were found (up to 150 ppbv), also has the largest OPE, despite high median NO x mixing ratios. The high OPE value, however, is consistent with the analysis of Pfannerstill et al. (2019), who used VOC and OH reactivity measurements to derive the fraction of OH that reacts with VOCs (fuelling the forma- Figure 15. NO z , O 3 , and HCHO mixing ratios, together with NO 2 photolysis rates, during the transitions between the Arabian Sea and Arabian Gulf, as well as in the Arabian Gulf. tion of RO 2 , conversion of NO to NO 2 , and thus O 3 formation) versus the fraction that reacted with NO x (resulting in NO z formation) to identify regions where O 3 formation was NO x -limited, VOC-limited, or (as was generally the case) in a transition regime. Pfannerstill et al. (2019) indicated that formation of O 3 was favoured around the Arabian Peninsula where VOCs from petroleum extraction and processing industries were important sinks of OH. The highest net ozone production rates (NOPRs) during AQABA were also found in the Arabian Gulf where calculations of the rate of RO 2 induced oxidation of NO to NO 2 resulted in a median (over the diel cycle) value of NOPR = 32 ppbv per day which was driven by high noon-time mixing ratios of RO 2 (73 pptv in the Arabian Gulf) . In contrast to the OPE, NOPR accounts for the total amount of O 3 produced in 1 d, considering production (governed by the formation of NO 2 via reactions of NO with HO 2 and RO 2 ) and loss (via photolysis and reaction with OH or HO 2 ). The OPE, on the other hand, focusses on the product side and assesses the competition between O 3 formation and sequestering into NO z from a given initial level of NO x . By approximating the O 3 production rate via the NO 2 formation from NO reactions with HO 2 and RO 2 , the NOPR thus neglects the alternative branch leading to NO z . In the other two regions, the correlation coefficients are notably smaller, due to the lower span in O 3 and NO z , resulting in increased relative errors for the derived OPE values. 0.5-1.0 0.5-2.1 0.9-4.9 NO y / CO (%) b,c 1.4-7.0 1.9-14.6 OH max (10 6 molec. cm −3 ) (d) 9.1 5.7 11.8 a Median. b 10-90 percentiles. c No CO data after 16 August 2017. d Average of daily OH peak concentrations; no data before 18 July 2017.
In Fig. 15 we plot a time series of NO z mixing ratios during the transition from the Arabian Sea to the Arabian Gulf along with NO 2 photolysis rates, O 3 , and formaldehyde (HCHO) which is formed during the photochemical processing of many VOCs (Fischer et al., 2003;Klippel et al., 2011;Wolfe et al., 2016;Wolfe et al., 2019) and which can therefore be used as a tracer for photochemical activity (Dodge, 1990;Altshuller, 1993;Garcia et al., 2006;Duncan et al., 2010;Parrish et al., 2012). The transition from low NO z levels in the Arabian Sea to values up to ∼ 7 ppbv in the Strait of Hormuz (SH) is accompanied by increases in both O 3 (up to 160 ppbv) and HCHO (up to 12.5 ppbv). Based on the analysis by Duncan et al. (2010), Tadic et al. (2020) calculated a median HCHO / NO 2 ratio of 9.3 for the Arabian Gulf, indicating that O 3 production in this region is NO x limited. The high levels of NO z , O 3 , and HCHO in the Arabian Gulf result from the combination of intense solar radiation with high levels of reactive VOCs Pfannerstill et al., 2019) and NO x and are accompanied by the highest levels of gas-phase organic nitrates observed during AQABA, with absolute mixing ratios up to 2.5 ppbv on the approach to Kuwait. In conclusion, our NO x / NO y measurements and the OPE values derived from them confirm the exceptional photochemical activity in the Arabian Gulf.

Conclusions
During the AQABA campaign in the summer of 2017, we collected a unique NO x and NO y data set that covers the Mediterranean Sea, the Red Sea, and the Arabian Gulf, which are regions with only few previously published observational data sets. The highest median NO x and NO z mixing ratios were observed in the Arabian Gulf (NO x : 1.6 ppbv; NO z : 1.5 ppbv), followed by the Red Sea (NO x : 1.0 ppbv; NO z : 0.7 ppbv) and the Mediterranean Sea (NO x : 0.3 ppbv; NO z : 0.7 ppbv). Night-time losses of NO 2 exceeded daytime losses by factors of 2.8 and 2.9 in the Arabian Gulf and the Red Sea, respectively, whereas daytime losses were 2.5 times higher in the Mediterranean Sea, which is a result of consistently high daytime OH concentrations.
The derivation of NO x lifetimes enabled us to calculate the NO x source strength required to reproduce the observed mixing ratios and indicated that HONO photolysis was a significant source of NO x in the Mediterranean Sea. The strong correlation between NO x and SO 2 in air masses that were impacted by fresh emissions of NO x indicated that ships are the dominant source of NO x throughout the AQABA campaign. HONO may have been generated on particulate nitrate, possibly associated with black carbon that has been processed (to contain sulfate, organics, and nitrate) as the ship plumes evolve chemically.
The fractional contributions to NO x of NO y and the various components of NO z were highly variable in the three regions. The lowest regional mean contribution of NO x to NO y (i.e. most aged air masses) was found in the Mediterranean Sea with 41 % compared to 47 % in the Red Sea and 46 % in the Arabian Gulf. Of the NO z trace gases, HNO 3 represented the most important contribution to NO y with 39 % in the Arabian Gulf, 25 % in the Red Sea, and 35 % in the Mediterranean Sea. A clear regional variability was observed for the contribution of organic nitrates, with the highest value (16 % in the Red Sea) related to the concurrent availability of precursor NO x and VOCs from the oil and gas industry. Comparable figures were derived for the Arabian Gulf and the Mediterranean Sea, with 10 % and 11 %, respectively. pNit (particle diameter < 1 µm) contributed only a few percent, with the largest value (10 %) found in the Mediterranean Sea. HONO and ClNO 2 were generally only minor components (< 3 %) of NO z . Future studies on the reactive nitrogen budget in the AQABA region might benefit from longer stationary measurements (e.g. to identify diurnal patterns), together with the detection of more speciated NO z compounds (especially HNO 3 and ONs).
Data availability. The NO x and NO z data sets are available at https://doi.org/10.5281/zenodo.4746367 (Friedrich and Crowley, 2021). Other data sets (e.g. HONO, OH, SO 2 ) can be obtained via the person responsible on request.
Author contributions. NF analysed the NO x and NO y data sets and wrote the article. NF and JNC operated the TD-CRDS. PE and JNC provided CIMS measurements of SO 2 and ClNO 2 . JSh, NS, and JNC performed and evaluated ONs measurements. JSc set up and processed data from the spectral radiometer. DD, BH, IT, and HF contributed NO, NO 2 , HCHO, and CO measurements. MM, RR, ST, and HH provided OH concentrations. EYP, NW, and JW were responsible for the OH reactivity measurements. JB and FD performed measurements with the AMS and OPC instruments. HS, GL, and YC contributed the HONO data set. JL designed the AQABA campaign. All authors contributed to the writing of the article.
Competing interests. The authors declare that they have no conflict of interest.