The airborne measurement-based CAFE-Africa project took place in August and September 2018 in the tropical troposphere over the central Atlantic Ocean and western Africa. Starting from and returning to the international airport on Sal, Cabo Verde (16.75∘ N, 22.95∘ W), a total of 14 scientific measurement flights (MFs) was carried out with the German High Altitude and Long-range research aircraft (HALO). For the analysis of the MFs, we consecutively numerate each MF, starting with MF03 on 7 August 2018 for the ferry flight from Oberpfaffenhofen (Germany, Deutsches Zentrum für Luft- und Raumfahrt) to Sal (Cabo Verde islands) on and ending with MF16 on 7 September 2018 for the back ferry flight from Sal to Oberpfaffenhofen. The test flights MF01 and MF02 conducted over Germany are not included in this study. MF03–MF16 covered a latitudinal range from 8∘ S to 48.2∘ N and a longitudinal range from 47.9∘ W to 12.5∘ E and reached maximum flight altitudes of about 15 km. Before landing at the home base airport in Sal, a fixed-altitude leg of 30 min duration at FL150 (∼4600 m altitude) was flown for calibration purposes. Take off (T/O) time of the flights was typically 09:00 or 10:00 UTC, except for MF08 with T/O at 04:00 UTC and landing around 13:00 UTC and MF11 with T/O at 16:00 UTC and landing around 01:00 UTC the next day.

The location of the campaign home base on Sal provided the unique possibility to analyze the impact of the Intertropical Convergence Zone (ITCZ) on physical and chemical processes in the airspace above the Atlantic Ocean and western Africa. The ITCZ is a low-pressure region evolving near the Equator, which is characterized by deep convection, strong precipitation and frequent lightning (Collier and Hughes, 2011), producing nitrogen oxides, mostly as NO through the Zeldovich reactions from atmospheric N2 and O2. The campaign was performed in late summer (August and September) 2018 when the ITCZ had reached its northernmost position at around 5–15∘ N (Collier and Hughes, 2011) and was henceforth located only a few degrees in latitude to the south of the campaign base at 16.75∘ N. The flight tracks of the 14 MFs performed during the campaign are shown in Fig. 1. An overview of the corresponding flight dates and objectives of each particular MF is given in Table S1 in the Supplement.

In situ measurements of NOx on ground-based and mobile platforms are challenging in terms of the demand for high sensitivity and high precision (Tadic et al., 2020). During CAFE-Africa, we deployed a modified commercially available chemiluminescent detector CLD 790 SR (ECO Physics Inc., Dürnten, Switzerland) aboard HALO. It is the same instrument that has been used during previous shipborne (Hosaynali Beygi et al., 2011; Tadic et al., 2020) and airborne campaigns (Bozem et al., 2017). The measurement method is based on the gas-phase reaction of NO with O3, which will partly produce excited-state NO2∗ followed by the spontaneous emission (chemiluminescence) of a photon (Ridley and Howlett, 1974; Ryerson et al., 2002). R13NO+O3→NO2∗+O2R14NO2∗→NO2+hv(λ>600nm) Photons generated through the emissions from excited-state NO2∗, which are directly proportional to the NO concentration in the sample flow (Ridley and Howlett, 1967), are detected by a photomultiplier tube and converted to an electric pulse. Carrying out the oxidation of NO by O3 at low pressure (7–8 mbar) and in a temperature-stabilized (25 ∘C) main reaction chamber minimizes quenching (non-radiative de-excitation of NO2∗ via collisions) (Reed et al., 2016; Tadic et al., 2020). Detector dark noise and artifacts due to the reaction of O3 with species other than NO (such as alkenes and sulfides) are corrected for by using a pre-chamber setup, as first described by Ridley and Howlett (1967). A residual instrumental background (due to memory effects within the instrument) is corrected for by regularly sampling synthetic zero air (Tadic et al., 2020). During the MFs, we sampled zero air from a tank (17 L composite tank, AVOX) with a Purafil-activated carbon adsorbent installed downstream of the zero-air tank to ensure NO-free zero-air measurements. The residual instrumental background of the NO measurement was calculated at 5 pptv from measurements obtained at nighttime during MF11. As chemiluminescent detection of NO is an indirect measurement method, regular calibrations against a known standard are needed. During the MFs, we diluted the secondary NO standard (1.187±0.036 ppmv NO in N2) at a mass flow of 8.6 sccm in a zero-air flow of 3.44 slpm (standard liter per minute), resulting in NO calibration gas mixing ratios of ∼3 ppbv. NO calibration measurements were performed six to eight times during a MF of 9–10 h duration by manually initiating calibration slots consisting of 2 min zero-air measurement, 2 min NO calibration and 2 min zero-air measurement, similar to previous deployments of the instrument (see Tadic et al., 2020).

The limit of detection (LOD) of the NO data was calculated at 9 pptv from the FWHM (full width at half maximum) of a Gauss fit applied to the distribution of 1 s NO data obtained at nighttime during MF11 (see Fig. S1 in the Supplement). Analogously we estimate the LOD of the NO data at 1 min time resolution to be 5 pptv from the FWHM of a Gauss fit applied to the distribution of 1 min NO data obtained at nighttime during MF11. The precision of the NO data was calculated from the average reproducibility of all in-flight calibrations to be 5 % at 1σ. The uncertainty in the used secondary standard mixing ratio was 3 %. The total measurement uncertainty (TMU) of the NO data was estimated at 6 % as the quadratic sum of the precision and the uncertainty of the secondary standard (Tadic et al., 2020). 1TMUNO=5%2+3%2≈6%

O3 was quantified with a chemiluminescence detector calibrated by a UV photometer (Fast AIRborne Ozone instrument; Zahn et al., 2012). CO and CH4 were measured by mid-infrared quantum cascade laser absorption spectroscopy (QCLAS) with TRISTAR, a multichannel spectrometer (Schiller et al., 2008; Tadic et al., 2017). OH and HO2 radicals were measured by laser-induced fluorescence with the custom-built HORUS instrument (Marno et al., 2020). Note that both OH and HO2 data are preliminary. We conservatively estimate the total relative measurement uncertainty of the OH and HO2 data at 50 %. Spectrally resolved actinic flux density measurements were obtained with upward- and downward-looking spectroradiometers installed on the top and bottom of the aircraft fuselage, respectively. The particular photolysis frequencies were calculated from the actinic flux density spectra between 280 and 650 nm (Bohn and Lohse, 2017). The uncertainty in the used j values was estimated to be 13 %. Water vapor mixing ratio and further derived humidity parameters were measured by SHARC (Sophisticated Hygrometer for Atmospheric ResearCh) based on dual-path direct absorption measurement by a tunable diode laser (TDL) system (Krautstrunk and Giez, 2012). The measurement range of SHARC covers the whole troposphere and lower stratosphere (5–40 000 ppmv) with an absolute accuracy of 5 % (+1 ppmv). The BAHAMAS (BAsic HALO Measurement And Sensor System) provided measurements of temperature and pressure (Krautstrunk and Giez, 2012). All instruments deployed on the aircraft have been developed to meet the high standards of airborne measurements in terms of operability, accuracy and sensitivity. Table 1 lists the used instrumentation with the associated total measurement uncertainties. A reference is given regarding the use of each method during previous measurements.

EMAC is a 3-D global general circulation, atmospheric chemistry–climate model, which has been used and described in a number of previous studies (Roeckner et al., 2006; Jöckel et al., 2010; Sander et al., 2019; Tadic et al., 2020). Briefly, EMAC comprises the fifth generation of the European Center Hamburg (ECHAM5) circulation model (Roeckner et al., 2006) and the Modular Earth Submodel System (MESSy) in version 2.52 (Jöckel et al., 2010). Here we use the model in the T63L47 resolution with a spatial resolution of roughly 1.8∘×1.8∘, with 47 vertical levels and one data point every 6 min. The model has been weakly nudged towards the ECMWF ERA-Interim data (Jeucken et al., 1996; Berrisford et al., 2009). The chemical mechanism (the Mainz Organic Mechanism, MOM) and the photolysis rate calculations used in this work have been presented in Sander et al. (2019) and in Sander et al. (2014), respectively. The Emissions Database for Global Atmospheric Research (EDGARv4.3.2; Crippa et al., 2018) data were used for anthropogenic emissions, while biomass burning emissions were from the GFAS (Global Fire Assimilation System) database with a daily temporal resolution (Kaiser et al., 2012). Important for this work, the NOx emissions from lightning activity have been estimated using the submodel LNOX (Tost et al., 2007), where the parameterization by Grewe et al. (2001) was applied. The global NOx emissions from lightning were scaled to 6.3 Tg(N)yr-1, following Miyazaki et al. (2014). Tracer and aerosol wet and dry deposition were estimated following Tost et al. (2006) and Kerkweg et al. (2006), respectively. The NOx soil biogenic emission flux is calculated based on a semiempirical emission algorithm implementation by Yienger and Levy (1995) and Kerkweg et al. (2006). For the current study, we use EMAC simulations of NO, O3, OH, HO2, CH3O2, specific humidity, j(O1D), temperature and pressure spatially interpolated to the flight tracks (latitude, longitude and altitude). Based on the simulations, we perform calculations of α and NOPRs along the flight tracks (see Sect. 2.5). To synchronize the time stamp of the model data (6 min) with the measurement data (1 min), we have calculated a running mean of the measurement data within ±2 min around the simulated data point along the measurement flight tracks such that every sixth measurement data point (if available) was neglected.

Calculation of NOPRs utilizes the chemical reactions related to ozone formation described in the introduction. EMAC model calculations show that during CAFE-Africa CH3O2 represents on average 80 % of the sum of all organic peroxy radicals with respect to ozone formation at typical flight altitudes of 200 hPa (and even up to 90 % at lower altitudes). Model calculations further show that the sum of HO2 and CH3O2 represents on average 95 % of the sum of HO2 and all organic peroxy radicals (RO2), yielding that the ratio (HO2+CH3O2)/(HO2+RO2) is practically one. In analogy to Tadic et al. (2017), we estimated RO2 as the sum of all organic peroxy radicals with less than four carbon atoms. See Table S2 for an overview of the used organic peroxy radicals. Therewith we calculate photochemical gross production of O3 by the rate-limiting reaction of NO with HO2 and CH3O2 (Thornton et al., 2002; Bozem et al., 2017). 2. The IUPAC Task Group on Atmospheric Chemical Kinetic Data Evaluation (Atkinson et al., 2004, 2006) provides rate coefficients used in this study. Note that other studies use P(O3)gross as an acronym for P(O3) in Eq. (2). The photochemical lifetimes of both HO2 and CH3O2 are similar with respect to self-reactions yielding hydrogen peroxide and methyl hydroperoxide and reactions with NO and HOx (Bozem et al., 2017). We further assume photostationary steady state (PSS) for the probed air masses. As the typical time to acquire PSS during CAFE-Africa varied between 40 s at 2 km altitude and about 70–80 s at 15 km altitude (Mannschreck et al., 2004; Tadic et al., 2020), we can calculate the concentration of CH3O2 by the equation derived by Bozem et al. (2017). 3CH3O2=kOH+CH4[CH4]kOH+CO[CO]⋅[HO2] Note that the reaction of CO with OH represents the dominant term in HO2 production during CAFE-Africa. Assuming mixing ratios of 500 ppbv and 100 pptv for H2 and HCHO, respectively, we find that HO2 production rate from the reaction of OH with CO is on average 5 times greater than the sum of the HO2 production rates from photolysis of HCHO and the reactions of HCHO and H2 with OH during CAFE-Africa. Note that the assumed mixing ratio of 100 pptv represents a rather conservative upper estimate for HCHO in the upper troposphere. As mentioned above, ozone loss due to photolysis (and subsequent reaction of O(1D) with H2O) will only partly lead to a net loss effect as most O(1D) will deactivate via collisions with air molecules, mostly N2 and O2, to O(3P) and reform O3 in the subsequent reaction with O2. The share of O3 photolysis that will eventually lead to a net loss in O3 can be calculated using Eq. (4) (Bozem et al., 2017; Tadic et al., 2020). 4α=kO(1D)+H2O[H2O]kO(1D)+H2O[H2O]+kO(1D)+N2[N2]+kO(1D)+O2[O2] In the troposphere, α ranges from about 15 % in the PBL to 1 % in the upper troposphere, where absolute humidity is very low (Bozem et al., 2017). Further loss processes of O3 (reactions with alkenes, sulfides and halogen radicals) are considered small and are therefore neglected in this study. Equation (5) then gives gross loss of ozone.5 j(O1D) expresses the photolysis frequency of ozone to O(1D). The NOPR is given as the difference of the gross ozone production rate (Eq. 2) and the gross ozone loss rate (Eq. 5) (Lin et al., 1988; Cantrell et al., 2003). 6NOPR=PO3-LO3=NO⋅kNO+HO2HO2+kNO+CH3O2CH3O2-O3⋅α⋅j(O1D)+kOH+O3OH+kHO2+O3HO2 Note that other studies use P(O3)net as an acronym for NOPR in Eq. (6).

In the following, we will investigate averages of the vertical profiles, which are calculated based on an altitude bin width of 1 km. The profiles are calculated with respect to the center of the particular bin, e.g., the average at 3.5 km includes all data points obtained at or above 3 km altitude and below 4 km altitude. Data are filtered for stratospheric influence by removing all data points for which concurrent O3 concentration is larger than 100 ppbv; this is a conservative criterion which has been discussed by Prather et al. (2011). Figure 2 shows the vertical NO and O3 profiles obtained during CAFE-Africa. The orange and blue lines represent vertical average profiles of experimental and model-simulated data, respectively. The blue and orange shading in the respective colors represent the ±1 standard deviation of the vertical averages.

The vertical profile of measured NO data shows lowest NO mixing ratios of less than 20 pptv observed between 2 and 8 km altitude, which reflect the absence of emission and transport sources at these altitudes. Highest NO mixing ratios of 0.15–0.2 ppbv are observed above 12 km altitude and reflect the increasing amount of lightning-produced NOx, and to a lesser extent influence of relatively NOx-rich stratospheric air. Below 2 km altitude, the vertical profile shows a weak increase of NO, which reproduces the low amount of anthropogenic sources in the investigated MBL and PBL in proximity to the Cabo Verde islands. This suggests that the contribution of local convective uplift of PBL air to the increased NOx above 12 km altitude is negligible.

The vertical average profile of simulated NO data is in good agreement with the vertical profile of measured NO data, which is also indicated by the median NO(model)/NO(measurement) ratio throughout the whole campaign at 0.97. Although the vertical profiles are in overall agreement, the variability when comparing single measurement and simulation data points is substantial, as indicated by the large variability of simulated NO data above 10 km altitude. The 25th percentile, 75th percentile and average of the NO(model)/NO(measurement) ratio throughout the whole campaign are 0.25, 2.2 and 2.27, respectively, which illustrate the significant spread among measurement and model data in a comparison of individual data points. The minimum and maximum mixing ratios of modeled NO are zero and 2.13 ppbv, respectively. The minimum and maximum mixing ratios of observed NO are zero and 0.95 ppbv, respectively.

The average vertical profile of measured O3 shows lowest mixing ratios of 30–40 ppbv below 3 km altitude and a steady increase in O3 at about 5–10 ppbv per kilometer altitude to mixing ratios of 60–65 ppbv at 6 km. Above this altitude, O3 is relatively constant (60–65 ppbv) until it further increases above 12 km altitude, simultaneous with the increase in NO.

Although simulated O3 slightly underestimates the measurement data throughout the troposphere, we find that the vertical profiles of simulated and measured O3 data are in again in good agreement. EMAC O3 data match lowest mixing ratios of 30–40 ppbv observed below altitudes of 3 km as well as the vertical gradient in O3 mixing ratios between 3 and 6 km with mixing ratios of 60 ppbv at 6 km altitude. Above 6 km altitude, both the vertically constant O3 mixing ratio as well as the further increase in O3 above 12 km altitude deduced from the measurements are well reproduced by the model. Except for lowest altitudes (<1 km), it seems that simulated O3 mixing ratios slightly underestimate the measurement data. This is confirmed by the median and average O3(model)/O3(measurement) ratio throughout the campaign at 0.97 and 0.98, respectively, confirming the general agreement between measurements and simulations, as well as the slight underestimation by the latter. The 25th percentile and 75th percentile of the O3(model)/O3(measurement) ratio are 0.85 and 1.11, respectively, and indicate that the spread among single data points when comparing measurement and model for O3 is less than for NO.

O3 profiles observed in this study are in good agreement with results from the ATom mission (Bourgeois et al., 2020). For the June–August season, Bourgeois et al. (2020) show that in the tropical troposphere O3 increased with altitude to 50 ppbv at 5–6 km, whereas above 9 km O3 varied from 40 to 80 ppbv, supporting the results presented here (see Figs. 9 and 10 in Bourgeois et al., 2020).

As most of the measurement time (>60 %) of the CAFE-Africa campaign was dedicated to upper tropospheric measurements above 12 km altitude and as both NO and O3 show highest mean mixing ratios above 12 km altitude, we characterize the spatial distribution of NO and O3 above that altitude in the following. Data above 12 km altitude have been aggregated and averaged over a spatial 2∘×2∘ grid. We again remove stratospheric measurement data by only considering those for which O3 was below 100 ppbv. Note that this does not necessarily exclude influence of mixing with air of stratospheric origin. Figure 3 shows the color-coded spatial NO distributions based on the measured data (left plot) and simulated data (middle plot). The right plot shows the tropospheric average spatial distribution of the point-by-point NO(model)/NO(measurement) ratio above 12 km. Note that the color scales presented in the following emphasize the most relevant features of the spatial distribution. Thus a few single data points might exceed the given color range, such as in the case of simulated NO (Fig. 3b) with single maximum NO mixing ratios of >1 ppbv.

The spatial distribution of NO from the measurement data shows that NO in the upper tropical troposphere above western Africa and the Atlantic Ocean is generally 0.1–0.2 ppbv. The range in mixing ratios includes lowest NO of less than 20 pptv observed between -30 to -40∘ E and 16 to 20∘ N close to a deep convective system without lightning to mixing ratios of 0.1–0.2 ppbv over wide areas over the Atlantic Ocean to (more than) 0.4 ppbv over western Africa. Although deep convective systems over oceanic regions rarely evolve lightning (Zipser, 1994), we encountered large amounts of NO close to a marine cumulonimbus cloud system with potential lightning activity resulting in more than 0.3 ppbv NO at -28 to -32∘ E and at 12 to 16∘ N. The coincidence of the ITCZ (5 to 15∘ N during August and September) and enhancements in upper tropospheric NO above western Africa underlines the substantial influence of the seasonal migration of the ITCZ and its impact on lightning and nitrogen oxides in the upper tropical troposphere (Zipser, 1994; Xu and Zipser, 2012). The data also suggest a longitudinal increase in NO from about 0.1 ppbv observed at -40∘ E westbound to 0.4 ppbv above western Africa. Note that we observe a slight decrease in upper tropospheric NO over the Côte d'Ivoire and partly also over Guinea compared to upper tropospheric NO over Senegal, although the lightning flash rate over the Côte d'Ivoire and Guinea is reported to be a factor of about 3 larger than that over Senegal (Collier and Hughes, 2011). A linear fit applied to the longitudinal average profile of all NO data weighted by the standard deviation and collected in the troposphere above 12 km altitude between -42 and -8∘ E reveals an increase in average NO of 4–5 pptv per degree longitude (see Fig. S2).

Although the comparison of the vertical profiles of measured and simulated NO suggests generally good correspondence between measurements and model simulations, the agreement with respect to the spatial NO distributions in the upper troposphere is much less satisfactory. The EMAC model does not reproduce the large NO enhancements in the area of the ITCZ, as shown in the latitudinal profile of measured and simulated tropospheric NO data above 12 km in Fig. S3. On contrary the model tends to underestimate the observations south of 10∘ N. Interestingly, this holds also for observations above western Africa (except for the airspace over Senegal around 12 to 14∘ N) where simulated NO mixing ratios are highest. This general underestimation of the measurements by the model over large parts of western Africa extends to large parts of the Atlantic Ocean between 5 to 15∘ N. On the other hand, further north the model tends to overestimate the measurements across large areas north of 16∘ N and west of -20∘ E with spatially averaged mixing ratios that exceed 1 ppbv. However, reasonable agreement between measurements and numerical results is observed towards and over southern Europe.

Figure 4 shows color-coded spatial O3 distributions based on observations (a) and simulated data (b). Figure 4c shows the average tropospheric distribution of the point-by-point O3(model)/O3(measurement) ratio. In analogy to Fig. S3, Fig. S4 shows latitudinal profiles of measured and simulated O3 mixing ratios obtained above 12 km in the troposphere.

Measured O3 shows a rather uniform distribution in the upper troposphere above the Atlantic Ocean and western Africa. The observed mixing ratios range from less than 40 ppbv between -30 to -40∘ E and 16 to 20∘ N, areas where NO is likewise decreased compared to other regions with more than 80 ppbv towards and over southern Europe, which partly reflects the increasing stratospheric impact above 12 km altitude. Over western Africa, we observed average O3 at 50–70 ppbv, which is in approximate agreement with previous studies (Galanter et al., 2000). Note that measured O3 mixing ratios over the African continent are not significantly different from O3 mixing ratios over adjacent oceanic areas.

In general, the simulated O3 reproduces the observed absolute O3 mixing ratios in the upper tropical troposphere, as well as regional tendencies. It is of note that, although the model underestimates NO over the tropical continental area of Africa, O3 is reproduced remarkably well. Nevertheless, the model is not able to reproduce local O3 variations such as at -20∘ E and 10∘ N or at -10∘ E and 0∘ N. Moreover, simulated O3 seems to be rather uniformly distributed throughout the whole ITCZ region. The right panel of Fig. 4 further illustrates that the majority of the simulated data points deviate by less than 10 %–15 % from the observational data and that larger deviations between model simulations and measurement are mainly restricted to situations when the measurements show either the lowest or highest mixing ratios not reproduced by the simulations. The overall spatial agreement between O3 observation and O3 simulation is also demonstrated in the latitudinal profile given in Fig. S4. Furthermore, Figs. S5 and S6 show 2-D latitudinal or altitudinal distributions of measured, tropospheric NO and O3, respectively.

In the following, NOPRs are calculated based on Eq. (6) and analyzed both vertically and spatially. The left graph of Fig. 5 shows the vertical profile of NOPRs derived from measured and simulated data in orange and blue (±1 standard deviation of the corresponding vertical average), respectively. The middle and right graph show the vertical average profiles of the components of gross ozone loss and gross ozone production derived from experimental in situ data and simulated data, respectively. We provide a vertical profile of α calculated based on Eq. (4), for which we obtain good agreement between measurements and simulations, for which we refer to the left graph of Fig. S7. Figure S7 also provides a comparison of vertical profiles of measured and simulated H2O mixing ratios. The vertical profiles are calculated based on an altitudinal bin width of 1 km and are filtered for stratospheric influence by removing data points for which O3 is higher than 100 ppbv. Figure S8 presents latitudinal profiles of NOPRs above 12 km altitude in the troposphere derived from measured and simulated data. A spatial distribution of OH and HO2 (derived from both measured and simulated data) is given in Fig. S9.

During CAFE-Africa NOPRs derived from in situ measurements varied between -1 to about 0.6 ppbvh-1 within ±1 standard deviation of the vertical average. We found net ozone destruction for all altitudes below 7–8 km with a minimum of (-0.6±0.2) ppbvh-1 between 2 and 3 km. A general increase of NOPRs with altitude results in net ozone production of 0.2–0.4 ppbvh-1 above 9 km altitude with a maximum of (0.4±0.3) ppbvh-1 between 13 and 14 km altitude. The large standard deviation associated with the vertical profile at 13–14 km altitude reflects the large variation in NOPRs along the flight tracks. The vertical NOPR profile derived from in situ data further shows a rather smooth transition from net ozone destruction to net ozone production between 7–8 km altitude, which is in good agreement with the value estimated by Bozem et al. (2017) for the tropical troposphere over the South American rainforest at latitudes of 5 to 10∘ N. The negative net ozone tendencies observed between 3 and 5 km altitude for the tropical troposphere stand in opposition to positive net ozone tendencies of about 0.1 ppbvh-1 (Zanis et al., 2000a) and balance net ozone tendencies (NOPR=0) (HOOVER campaign over Europe; Bozem et al., 2017) deduced from previous measurements at similar altitudes at mid-latitudes.

In general, the vertical tendencies in NOPRs derived from the observations are well reproduced by the NOPR calculation based on simulated data. However, the model calculations indicate a minimum in net ozone destruction at (-0.3±0.2) ppbvh-1 between 1 to 3 km, which represents about half of that derived from the in situ measurements for these altitudes. This underestimation of the measurement by the model is directly related to an underestimation of simulated humidity and j(O1D), which are both underestimated by EMAC by about 15 %–20 % below 4 km altitude (see the right graph of Fig. S7 for a comparison of the vertical profiles of measured and simulated j(O1D)). The model reports net ozone production of 0.2 ppbvh-1 above 8 km (except between 9 and 10 km altitude) and also suggests that the transition from net destruction to net production occurs between 6 and 8 km altitude, which again agrees with the measurement-based calculation. Nevertheless, the atmospheric variability of the (simulation-based) average NOPR profile reveals that transition from net ozone destruction to net ozone production occurs within a wider altitudinal range of 4 to 10 km altitude.

The reaction of NO with HO2 dominates gross ozone production (Fig. 5b) for NOPRs derived from measured in situ data. Whilst the reaction of NO with CH3O2 contributes about 0.03 ppbvh-1 to gross ozone production throughout the whole troposphere, the vertical average of the ozone production rate from the reaction of NO with HO2 yields 0.1 ppbvh-1 at the lowest altitudes with a linear increase (r2≈0.6) to about 0.3 ppbvh-1 at 14–15 km altitude. From our observations, it follows that the ozone production rate due to the reaction of NO with HO2 is from a factor of 2–3 (below 3 km altitude) to a factor of 10 (above 12 km altitude) stronger than gross ozone production due to the reaction of NO with CH3O2. For the measurement-based estimate, photolysis of ozone dominates gross ozone loss below 6 km altitude. Between 1 and 2 km, it is largest in absolute values at -0.8 ppbvh-1, where it contributes to about 80 % of total gross ozone loss. With increasing altitude, the gross ozone loss rate due to photolysis sharply decreases in absolute value to less than -0.05 ppbvh-1 above 8 km altitude. Between 6 and 10 km altitude, (total) gross ozone loss is on the order of -0.15 ppbvh-1, mainly due to photolysis of O3 while reaction of HO2 with O3 (-0.05 ppbvh-1) and the reaction of OH with O3 (-0.03 ppbvh-1) are significantly smaller in absolute values. Above 10 km, gross ozone loss rate decreases to -0.03 to -0.05 ppbvh-1. This is mainly due to a diminishing ozone loss via photolysis and reaction with H2O at low humidity, leaving ozone loss by the reaction of O3 with OH and HO2 at -0.01 to 0.02 ppbvh-1 as major loss processes. Ozone loss rates observed above 10 km during CAFE-Africa have only little impact on NOPRs as they balance only about 10 %–20 % of the absolute value of concurrent gross ozone production rates at these altitudes.

The model generally reproduces the NOPR tendencies in gross production and loss as shown above, yielding net ozone destruction at a rate of -0.1 to -0.3 ppbvh-1 below 4 km altitude, which is significantly lower in absolute values than the measurement-based calculation. This is due to a combination of a weaker loss term due to photolysis and of a larger production term due to the reaction of NO with HO2 and CH3O2 represented in the model. The contribution of the reaction of NO with HO2 represents a vertically constant value of about 0.2–0.3 ppbvh-1 with a slightly larger production rate from the reaction of NO with CH3O2 at 0.05 to 0.1 ppbvh-1 than inferred from the measurements. Above 10 km altitude, EMAC reproduces the relative and absolute tendencies of the particular gross ozone loss rates remarkably well.

Our results are comparable to a previous study on NOPRs derived from in situ airborne observations at similar latitudes over the rainforest in South America (Bozem et al., 2017). Bozem et al. (2017) report net ozone destruction of -0.2 to -0.6 ppbvh-1 between 2 to 4 km and net ozone production between 7 and 9 km altitude with the transition from net ozone destruction to net ozone production occurring at 7 km, similar to our results. Below 6 km altitude gross ozone loss is dominated by photolysis in both studies (Bozem et al., 2017). Bozem et al. (2017) found net ozone production in the continental PBL layer. In this study O3 destruction prevails, most likely due to the absence of large emission sources in the proximity of the Cabo Verde islands. In the marine boundary layer both studies tend towards net ozone destruction (Bozem et al., 2017).

Our results add to the understanding of photochemical net ozone production in the upper troposphere of the region. Using a photochemical trajectory model initiated by in situ measurements, Real et al. (2010) derived photochemical net ozone production rates of 2.6 ppbvd-1 over a period of 10 d downwind of western Africa. Our study supports the findings by Real et al. (2010) by underlining that photochemical ozone production in the upper troposphere over the tropics is positive at about 0.2–0.4 ppbvh-1, which supports the concept of significant photochemical ozone production in the upper troposphere of the region. Note that during CAFE-Africa, measurements at low altitudes were generally performed over the Atlantic Ocean. Hence, we cannot compare to previous results from Saunois et al. (2009) reporting ozone production ranging from 0.25 to 0.75 ppbvh-1 in the continental boundary layer over western Africa.

In the following, we investigate the spatial distribution of NOPRs derived from measured and simulated data. Figure 6 shows the color-coded spatial, tropospheric distribution of upper tropospheric (>12 km altitude) NOPRs calculated from observations (a) and model simulated data (b). The right plot shows the spatial, average tropospheric distribution of the point-by-point NOPR(model)/NOPR(measurement) ratio. Note that a few single data points exceed the given color scales. Also note that NOPR calculations based on observational data are restricted to periods of simultaneous availability of a number measured species and parameters so that data gaps will be more likely than for spatial distributions of in situ NO or O3.

The spatial distribution of NOPRs calculated based on measured data shows the already discussed, generally positive net ozone production tendencies in the upper tropical troposphere, but with distinct, characteristic regional features. While NOPRs are generally on the order of (0.2±0.1) ppbvh-1 north of 16∘ N and west of -20∘ E; spatially averaged NOPRs in the area of the ITCZ are ∼0.4 ppbvh-1 at several locations. Largest spatially averaged NOPRs based on the observations (>0.8 ppbvh-1) are found over tropical western Africa, mirroring strong NO enhancements (see Fig. 3). Nevertheless, the highest NOPR values are inferred over the Côte d'Ivoire although NO is lower than over Guinea or over Senegal, where the NOPR calculation yields comparable, but slightly smaller values. Over the Côte d'Ivoire ozone formation is mainly driven by large HO2 mixing ratios of up to 15–20 pptv (see Fig. S9 for spatial distributions of in situ measured and simulated OH and HO2 data). Similar as for NO, the spatial distribution further suggest a longitudinal increase of NOPRs towards the western African coast reflecting the general absence of LNOx over oceanic areas and increased lightning flash rates over the tropical parts of western and central Africa (Williams and Satori, 2004; Collier and Hughes, 2011).

The spatial distribution of NOPRs calculated based on simulated data largely follows the spatial distribution of simulated NO. Although the model indicates lowest NOPRs of less than 0.2 ppbvh-1 over most oceanic area, NOPRs derived from simulated data exhibit values of about 0.4 ppbvh-1 at several locations between -20 and -30∘ E, which correlate with enhancements in NO retrieved from EMAC. Over Africa, EMAC yields significant enhancements in NOPRs only between 10 and 14∘ N over Senegal, where NO is also enhanced.

The strong dependence of ozone formation on ambient NO concentrations for both measurement and model raises the question to which extent ozone formation was NOx limited. Figure 7 shows NOPRs calculated based on measured and simulated data in orange and blue, respectively, aggregated to a bin width of 0.025 ppbv of NO on the x axis. There is only limited data coverage for measured NO above 0.325 ppbv and for simulated NO above 0.15 ppbv. See Table S3 for the number of data points in each NO mixing ratio bin.

NOPRs derived from both observations and model simulations exhibit similar dependencies on ambient NO mixing ratios. Both measurements and model simulations show net ozone destruction below 30 and 20 pptv, respectively, and a nearly linear increase in NOPRs with increasing ambient NO mixing ratios above this threshold value, with an NOPR increase of 0.1 to 0.15 ppbvh-1 per 100 pptv increase in NO. Especially the NO compensation mixing ratio (for which ozone production equals ozone loss) reproduces results from previous studies remarkably well. Cantrell et al. (2003) report NO compensation mixing ratios between 10 and 30 pptv over the Pacific, depending on whether modeled or measured HO2 and RO2 is used. A study conducted by Zanis et al. (2000b) for the Swiss Alps also reports balance in ozone production for similar NO compensation mixing ratios. Due to low data coverage above 0.4 ppbv, we cannot resolve with much certainty whether beyond this value the increase in NOPR will continue. Note that one possible limitation of this figure arises from the fact that the data aggregated in the respective NO mixing ratio bins stem from different atmospheric layers and origins, which causes the spiky signature of the profile for both measurement and model. However, both model simulations and observations indicate that O3 in the upper troposphere in the tropics is NOx limited.