Hydroperoxides were measured as the sum of organic hydroperoxides and hydrogen peroxide and were determined using a wet chemical system named the HYdrogen Peroxide and Higher Organic Peroxides monitor (HYPHOP; Klippel et al., 2011; Bozem et al., 2017; Hottmann et al., 2020; Hamryszczak et al., 2022) based on a previous design by Lazrus et al. (1985, 1986). The ambient air was sampled from the top of the aircraft fuselage via a trace gas inlet (TGI) with a 1/2 in. perfluoroalkoxy (PFA) liner that left the cabin again through a second exhaust line. From this bypass, a 1/4 in. PFA sampling line was connected to a Teflon-coated membrane pump (type MD 1C; Vacuubrand, Wertheim, Germany) and a pressure control unit, regulating the pump speed to a line pressure of 1000 hPa (constant pressure inlet, CPI). The CPI provides a constant inlet pressure covering external pressure variations between 1000 and 150 hPa. Following the CPI inlet, the ambient air passed through a stripping coil with a buffered sampling solution (potassium hydrogen phthalate / NaOH; pH 6; stripping efficiency of 1 for hydrogen peroxide and between 0.6 and 1 for organic peroxides; Lee et al., 2000). The hydroperoxide solution was sampled in two individual channels in response to addition of p-hydroxyphenyl acetic acid (POPHA) and horseradish peroxidase (HRP). The formed chemiluminescent 6,6′- dihydroxy-3,3′-biphenyldiacetic acid was detected via fluorescence spectroscopy using a Cd pen-ray lamp at 326 nm. The hydroperoxide-specific fluorescence (Guilbault et al., 1968) was detected at 400–420 nm using photomultiplier tubes for each channel separately.

Hydrogen peroxide mixing ratios are then calculated from the difference between the entirety of the measured peroxides (channel A) and the sum of organic ROOH hydroperoxides (channel B), where H2O2 is selectively destroyed by the addition of catalase. Further information on the mixing ratios of individual organic peroxides cannot be provided by the monitor, due to the characteristics of the measurement technique. Please note that, especially within the boundary layer and due to biomass burning, a variety of organic peroxides might contribute to the total measured signal of organic peroxide (Fels and Junkermann, 1994; Slemr and Tremmel, 1994; Valverde-Canossa et al., 2005; Hua et al., 2008; Dienhart et al., 2023).

Prior to ambient measurements, both channels are simultaneously calibrated using a liquid standard (0.98 µmol L-1) produced from serial dilution of a H2O2 stock solution. The H2O2 destruction efficiency in channel B corresponding to the added catalase was determined to be 0.95–0.98 based on liquid calibrations. In-flight background measurements were performed using purified zero air, generated by a cartridge with silica gel (type IAC-502; Infiltec, Speyer, Germany) and Hopcalite (type IAC-330; Infiltec, Speyer, Germany). Using a gas-phase calibration source (low-density polyethylene (LDPE) permeation devices), the H2O2 transmission efficiency through the inlet (±1σ) was determined regularly by measuring the difference between the addition of the standard before and after the CPI and was found to be 0.61 ± 0.06. Due to a positive ozone interference, the H2O2 data were further corrected by subtraction of 0.056 ppbv H2O2 / 100 ppbv O3 based on a scatter plot of hydrogen peroxide vs. ozone mixing ratios in the lower stratosphere, assuming that ambient H2O2 above the tropopause is essentially zero. The total measurement uncertainty (TMU) of the monitor was estimated as 1TMU=P2+US2+UOI2+UTE2 by considering the instrument's precision (P), uncertainty of the standard (US), uncertainty of the H2O2 transmission efficiency (UTE), and the uncertainty of the ozone interference (UOI). The determined precision with 1σ confidence interval was determined from the reproducibility of the liquid calibrations performed during the campaign to be 1.3 % at 5.46 ppbv for hydrogen peroxide and 0.8 % at 5.64 ppbv for the organic hydroperoxides. The uncertainty of the standard was included in the instrument precision calculation. The uncertainty of the transmission efficiency was calculated to be 6 %. The total measurement uncertainty was determined to be 9 % for hydrogen peroxide and 41 % for the sum of organic hydroperoxides. The total measurement uncertainty of organic hydroperoxides is increased by 40 % due to the varying solubility of individual organic hydroperoxides in aqueous solution, which ranges between 60 % (e.g., methyl hydroperoxide) and 100 % (e.g., peroxyacetic acid). The instrumental time resolution was determined to be 122 s based on the calibration signal rise and fall time from 10 % to 90 % and 90 % to 10 %, respectively. The detection limit with a 2σ confidence was derived from the reproducibility of the in-flight background measurements as 15 pptv for hydrogen peroxide and 6 pptv for the sum of organic hydroperoxides, respectively. For the purposes of this study, hydrogen peroxide data were filtered for stratospheric influences by removing all data points with ozone mixing ratios higher than 100 ppbv.

GPS altitude and coordinates, temperature, pressure, and wind speed were obtained using the BAsic HALO Measurement And Sensor System, BAHAMAS. Water vapor mixing ratios and the corresponding air humidity were measured with the Sophisticated Hygrometer for Atmospheric ResearCh (SHARC) based on a tunable diode laser (TDL) setup (Krautstrunk and Giez, 2012). HOx radicals were measured by laser-induced fluorescence with the HydrOxyl Radical measurement Unit based on fluorescence Spectroscopy (HORUS; Marno et al., 2020). Spectrally resolved upward and downward actinic flux density was obtained with two spectroradiometers (Bohn and Lohse, 2017). A brief overview of the campaign instrumentation, measurement methods, their TMU values, and the corresponding technical references are listed in Table 1.

For the purposes of this study, the in situ observations are compared to numerical simulations from the global chemistry and climate 3-D model EMAC (ECHAM/MESSy Atmospheric Chemistry; Jöckel et al., 2010). The model numerically simulates the chemistry and dynamics of the troposphere and stratosphere using a large variety of submodels describing chemical and meteorological processes and the influences arising from anthropogenic and natural emissions from continental and marine environments (Jöckel et al., 2006). The basis of the atmospheric model is the fifth generation of the European Centre HAMburg general circulation model (ECHAM5; Roeckner et al., 2003, 2006). The communication between the various submodels is achieved by the Modular Earth Submodel System (MESSy; Jöckel et al., 2005, 2010, 2016). Atmospheric chemistry is simulated by the Module for Efficiently Calculating the Chemistry of the Atmosphere (MECCA) submodel (Sander et al., 2005, 2011, 2019), using the Mainz Organic Mechanism (MOM) and photolysis rate calculations from a radiation transfer model (Sander et al., 2014, 2019). Primary emissions and dry deposition as well as aqueous-phase chemistry in clouds and cloud scavenging are simulated by the ONLEM, OFFLEM, TNUDGE, and DRYDEP submodels (Kerkweg et al., 2006a, b), as well as the scavenging of tracers submodel (SCAV; Tost et al., 2006). Anthropogenic emissions are based on the EDGARv4.3.2 inventory (European Joint Center, JRC; Crippa et al., 2018) and are distributed vertically according to Pozzer et al. (2009). Biomass burning emissions were simulated based on the Global Fire Assimilation System (GFAS; Kaiser et al., 2012). The model has a vertical resolution of 47 vertical levels up to 0.01 hPa, a horizontal resolution of T63 (i.e., approximately 1.8∘ × 1.8∘), and a time resolution of 6 min. The model was further weakly nudged towards the ECMWF ERA-Interim data (Tadic et al., 2021). This model setup has been extensively evaluated for different trace gases and aerosols (Pozzer et al., 2022). For comparison to observations, the simulation results were interpolated on the GPS flight tracks using the S4D submodel (Jöckel et al., 2010).

The hydrogen peroxide mixing ratios in the upper troposphere under the assumption of photostationary steady-state conditions were calculated based on measured precursors and photochemical loss processes of hydrogen peroxide (HO2, OH, j(H2O2)) and the rate coefficient data from Atkinson et al. (2004).

In the upper troposphere, the production rate P(H2O2) of hydrogen peroxide due to the self-reaction of HO2 can be calculated from Eq. (2). The photochemical loss rate of hydrogen peroxide, L(H2O2), can be derived from H2O2 photolysis and the reaction with OH as shown in Eq. (3). The equations are derived from the reaction pathways presented in Sect. 1 (Reactions R4–R6). 2PH2O2=kHO2+HO2⋅HO22,3LH2O2=kH2O2+OH⋅OH+jH2O2⋅H2O2. The calculations of the rate coefficients were based on the measured parameters along the flight tracks according to Atkinson et al., 2004 (Eqs. 4–5). Due to the water dependence of the hydrogen peroxide production rate coefficient, both expressions in Eq. (4) were further extended by the factor 1 + 1.4 × 10-21 cm3 [H2O] exp⁡(2200/T).

4kHO2+HO2=2.2×10-13⋅exp⁡600Tcm3s-1+1.9×10-33[N2]⋅exp⁡980Tcm6s-1,5kH2O2+OH=2.9×10-12⋅exp⁡-160Tcm3s-1. Neglecting deposition and transport processes impacting the hydrogen peroxide budget, the mixing ratio of H2O2 was calculated via Eq. (6). 6H2O2PSS=HO22⋅kHO2+HO2OH⋅kH2O2+OH+jH2O2

For the purpose of the present study, we used measured H2O2, OH, HO2, water vapor, j(H2O2), temperature, and pressure and compared these with the concurrent spatially interpolated EMAC simulations. To synchronize the time resolution of the simulated data with the measurement output, we calculated a mean of the measurement data with a matching temporal resolution of 6 min (equivalent to model output).

Vertical profiles of all species under investigation were calculated as 1000 m bins (means and medians) over the entire sampled atmospheric column. Profile information is restricted to 30 takeoffs and landings at Sal, while other areas are not considered due to a lack of statistically significant data.

The spatially resolved data based on measurements, photostationary steady-state (PSS) model calculations, and EMAC simulations were binned into 1∘ × 1∘ subsets over the full extension of the flight tracks in the upper troposphere (≥ 8 km).

The latitudinal distribution of the species was examined using 2.5∘ bins over the upper tropospheric region. Please note that due to a reduced amount of data in the lower troposphere, the analysis of spatial and latitudinal distributions was restricted to measurements performed in the upper troposphere.

The observed mixing ratios of hydrogen peroxide during the CAFE-Africa campaign are presented as a latitude vs. longitude plot with mean mixing ratio values binned into a subset of 1∘ × 1∘ bins for the entirety of the upper troposphere (≥ 8 km) along the flight tracks (Fig. 2). The color scale represents the measured mixing ratio of H2O2.

The mean (±1σ) and median mixing ratios based on all measured H2O2 mixing ratios during the campaign were 0.18 (±0.13) and 0.15 ppbv, respectively, with maximum hydrogen peroxide mixing ratios reaching 1.03 ppbv. Slightly higher H2O2 levels were observed in the ITCZ (approx. 5–20∘ N), where locally mixing ratios up to 0.67 ppbv over a 1∘ × 1∘ bin of merged data were observed. These maxima are most likely due to atmospheric transport of H2O2 into the upper troposphere. This is in accordance with previous reports about increased hydrogen peroxide mixing ratios due to biomass burning and convective activity, elevating the H2O2 mixing ratios in the upper troposphere (Lee et al., 1998; O'Sullivan et al., 1999; Allen et al., 2022). Locally enhanced H2O2 was further observed during the measurement flight in close proximity to the tropical storm Florence on 2 September 2018 (approx. 18∘ N, 38.5∘ W). Here, the mean mixing ratios were between 0.10 and 0.43 ppbv with a maximum of up to 0.94 ppbv. Based on the high convective activity during the tropical storm, as reported by Nussbaumer et al. (2021), the H2O2 mixing ratios were expected to rise due to the rapid transport of air masses from the marine boundary layer (MBL) into the UT. Table 2 gives an overview of the estimated mean and median hydrogen peroxide mixing ratios measured during CAFE-Africa in relation to previous airborne measurements performed at a comparable latitudinal range.

The mean and median values during CAFE-Africa are comparable to previously reported mixing ratios during TRACE A and OMO campaigns, which covered a comparable latitude and altitude range (Table 2). Enhanced mixing ratios for CAFE-Africa relative to observations during the SONEX campaign are most likely due to differences in the examined regional range, since the latter campaign focused on the north Atlantic. Mean and median values in the Northern Hemisphere (20–40∘ N; Table S1) during CAFE-Africa (0.14 ± 0.11 and 0.12 ppbv, respectively) are comparable to observations during SONEX. During the ATom campaigns, slightly higher mean values were observed, although median values are comparable. This could be due to differences in the sampled altitudes, since ATom measurements were generally restricted to altitudes below 12 km.

Based on the comparison with previous studies, the observed mixing ratios of hydrogen peroxide during CAFE-Africa fit well into the general range of recent studies over the equatorial and subtropical Atlantic. The observed H2O2 distribution confirms further that mixing ratios of hydrogen peroxide in the upper troposphere seem to be far less dependent on latitude than those at lower altitudes. The latitudinal distribution of H2O2 during CAFE-Africa displays a rather small symmetrical latitudinal decrease between the inner tropics and the subtropics.

In order to investigate the impact of deep convection in the ITCZ on the H2O2 budget in the upper troposphere, a comparison of the in situ data with the output of photostationary steady-state (achieved) calculations and EMAC simulations was performed. The complementary spatial distributions of the hydrogen peroxide levels were expressed as latitude versus longitude plots of mean mixing ratios aggregated over a spatial grid of 1∘ × 1∘ in the upper troposphere (≥ 8 km) (Fig. S4 in the Supplement).

The calculated PSS-H2O2 levels range between approximately 0.01 and 0.88 ppbv with mean (±1σ) and median mixing ratios of 0.14 (±0.16) and 0.07 ppbv, respectively, which is a factor of 1.3 lower than the observations. PSS-based H2O2 mixing ratios tend to be higher at the southernmost coastal area (2.5∘ S, 10.5∘ W), where the levels range between 0.40 and 0.88 ppbv, and in proximity to the tropical storm Florence at up to 0.40 ppbv (approx. 18∘ N, 38.5∘ W). Hydrogen peroxide mixing ratios simulated by EMAC vary between 0.10 and 0.75 ppbv. The mean (±1σ) and median simulated mixing ratios are 0.30 (±0.19) and 0.29 ppbv, respectively, with maximum mixing ratios up to 1.04 ppbv (3.5∘ S, 9.5∘ W; Fig. S4b), which is slightly higher than the observations. The spatial distributions of the point-by-point ratio between PSS calculations and EMAC simulations versus the observations in the upper troposphere above 8 km are presented in Fig. 3, which gives an overview of the local differences relative to the observations varying from low ratios (yellow) to high (deep blue) values. Please note that for resolution purposes the color scaling is restricted to ratios up to 4.5.

Generally, the H2O2 (EMAC) / H2O2 (measurement) ratios indicate better agreement between the simulations and the measurements in the Northern Hemisphere (≥ 20∘ N; Fig. 3b). With decreasing latitude, the model tends to significantly overestimate hydrogen peroxide; H2O2 (EMAC) / H2O2 (measurement) ratios are increasing from approximately 2 to 4 with decreasing latitude. Locally, most likely due to underestimated cloud scavenging as will be further discussed in this work, EMAC simulates highly elevated hydrogen peroxide with a factor of up to 14 higher than the measurements (4.5∘ N, 9.5∘ W).

Good agreement between the observation-based PSS calculations and the measurements was found in proximity to the tropical storm Florence. Here, the H2O2 (PSS) / H2O2 (measurement) ratios agree between 0.83 and 1.04. Beyond the measurements taken here, the agreement with respect to the measured hydrogen peroxide levels is generally less satisfactory. Towards the southern subtropics as well as locally in the coastal area and at the base of operations in Sal (Cabo Verde), the ratios increase to 2.8, indicating an overestimation of the hydroperoxide levels relative to the observations, similar to the EMAC simulations. On the other hand, the PSS calculations tend to strongly underestimate hydrogen peroxide concentrations in the ITCZ (5–20∘ N) and in the northern part of the investigated region (≥ 20∘ N) by factors of up to 10 and 12, respectively. This indicates that, since the local photostationary steady-state conditions based on observed radical levels do not account for additional sources and sinks of the H2O2, the observed discrepancy between the observations and local PSS are most likely related to transport and cloud scavenging. The calculated difference between production (Eq. 2) and loss (Eq. 3) of hydrogen peroxide, (P–L)H2O2 based on the observations is expressed as a latitude vs. longitude plot with mean mixing ratios binned into a subset of 1∘ × 1∘ bins for the entirety of the upper troposphere (≥ 8 km) along the flight tracks (Fig. 4).

Generally, the majority of the sampled region is dominated by loss, especially in the ITCZ (approx. 5–20∘ N) and towards the north, where an H2O2 deficit of up to approximately -0.01 ppbv h-1 was determined. H2O2 production-dominated regions of up to 0.03 ppbv h-1 are observed towards the south and in the coastal area. The difference between the photochemical production and loss paths is directly linked to the deviations from the photostationary steady state and thus to the additional sources and sinks of hydrogen peroxide on a local scale. These are associated with the local meteorological conditions and transport processes. Based on ERA5 reanalysis results, especially towards the south and in the coastal area, an enhanced presence of clouds at altitudes above 6 km (gray shading in Fig. 4) and convective precipitation were observed during the measurement period (Hersbach et al., 2019). At a mean (±1σ) horizontal wind speed of 14.3 (±7.3) m s-1 measured along the flight tracks and a species lifetime of 3–4 d, transport towards the subtropics impacts H2O2 levels and further justifies the difference between the local PSS calculations and the observations.

Based on the coincidence with the latitudinal range of the ITCZ, the enhanced loss in the H2O2 budget relative to the PSS in the upper troposphere is most likely due to convective injection of H2O2 from lower layers into the upper troposphere and the subsequent redistribution of the species towards the north and south. Below, we show that, based on the comparison with EMAC simulation output, convective transport is important for the budget of hydrogen peroxide in the upper troposphere not only in the ITCZ but also in the subtropics.

As discussed above, the comparison between observations and both PSS calculations and EMAC simulations indicate large deviations at the most southern latitudes that were visited by HALO. This is clearly demonstrated in Fig. 5, which shows observations, PSS calculations, and EMAC simulations of H2O2 as functions of latitude in the upper troposphere (above 8 km). The mean values of each dataset with 6 min time resolution are binned into subsets of 2.5∘ of latitude for the investigated region from 6∘ S to 40∘ N. The lines and the complementary shading represent mean values and the supplementary standard deviations. The area of the ITCZ between 5 and 20∘ N is highlighted by gray shading.

Contrary to the calculations based on local photostationary steady-state conditions and the simulations by EMAC, the observations show very little latitudinal variation with mean values of approximately 0.1–0.2 ppbv. A tendency towards slightly enhanced values is observed in the ITCZ, where the mixing ratios increase up to 0.22 ppbv. However, considering the standard deviation range (up to 0.19 ppbv), the rather flat distribution in the upper troposphere indicates nearly constant hydrogen peroxide levels throughout the whole investigated tropospheric region (see also Table S1).

Both the EMAC-simulated and PSS-calculated latitudinal hydrogen peroxide distributions display decreasing trends with increasing latitudes towards the north. The highest mean values of 0.49 (±0.29) ppbv for PSS and 0.583 (±0.40) ppbv for EMAC are found in the southernmost part of the sampled region. The elevated levels of H2O2 in the EMAC simulations starts already in the ITCZ, while the PSS calculations only increase south of 5∘ N.

Overall H2O2 mixing ratios from the PSS calculations show a decreasing tendency from the Equator towards the subtropics. At approximately 5∘ N, the PSS-based mixing ratios decrease from 0.09 to 0.25 ppbv and tend to be lower than the measured levels at the northern part of the domain by a factor of up to 5. In contrast, the mixing ratios of hydrogen peroxide simulated by EMAC remain elevated from 6∘ S to 15∘ N (0.30 to 0.35 ppbv), yielding H2O2 (EMAC) / H2O2 (measurements) ratios of 2.2 to 2.5. Further north of 15∘ N, the H2O2 levels in EMAC decrease by almost half to 0.15 ppbv, resulting in H2O2 (EMAC) / H2O2 (measurements) ratios between 1.1 and 1.2 and a rather good agreement between simulations and observations. An overview of numerical values for measured means (±1σ) as well as medians and PSS-calculated and EMAC-simulated hydrogen peroxide levels subdivided into three hemispheric regions – the Northern Hemisphere (20–40∘ N), ITCZ (5 < 20∘ N), and Southern Hemisphere (10∘ S < 5∘ N) – is given in Table S1 in the Supplement.

Please note that the steady-state calculations only account for photochemical production and loss of hydrogen peroxide. EMAC simulations additionally account for vertical and horizontal transport processes, as well as local losses due to cloud scavenging (Hamryszczak et al., 2022). Thus, deviations between local photostationary steady-state budget calculations and EMAC simulations can indicate the impact of convective processes in the ITCZ in the upper troposphere. This requires that the EMAC model correctly simulates precursors (HO2) and sinks (OH, H2O2 photolysis) of photochemical H2O2 formation.

A comparison between observations and EMAC simulations for the basic species reveals that while HO2 is adequately reproduced by EMAC the simulations tend to underestimate OH and the photolysis frequencies in particular, south of 15∘ N (Fig. S5b). At the same time, model results partly tend to overestimate HO2, indicating issues with HO2 / OH partitioning. Since the production of OH in the UT depends to an extent on the reaction of HO2 with NO, these might be associated with the underestimation of NO by EMAC in the southern part of the domain, as shown by Tadic et al. (2021). Additionally, the measured H2O2 photolysis frequencies show minor discrepancies with those from the model (Fig. S5c). Due to the overestimated HO2 mixing ratios, EMAC simulates higher levels of H2O2. At the same time, the lower OH mixing ratios and the underestimated H2O2 photolysis frequencies might cause decreased loss rates, thus leading to underestimation of the H2O2 loss pathways. Therefore, overestimation of the photochemical source and underestimation of the photochemical H2O2 sinks by EMAC explain the differences between PSS calculations and simulations at southern latitudes; however, these observations do not explain the differences relative to the observations south of the ITCZ. Also, it is highly unlikely that a measurement error is responsible for the discrepancies, as this would have to be restricted to the most southerly latitudes.

Potential causes leading to the discrepancy between the measurement and the simulations might be an underestimation of cloud scavenging and removal of hydrogen peroxide by wet deposition processes within the ITCZ and also further south. A number of flights south of 15∘ N were performed in close proximity to the western shores of Africa and also even over land (Fig. 2), close to enhanced convective precipitation (Fig. S3). Based on the ERA5 reanalysis results on cloud coverage during the measurement period (Hersbach et al., 2019; Fig. S2), we hypothesize that the air masses sampled in this area were affected by cloud processing especially in the UT, causing the model discrepancies.

Since H2O2 in the upper atmosphere is influenced by convective transport from the lower troposphere as well as by losses to clouds and rainout, it is important that EMAC simulations and PSS calculations reproduce the levels of hydrogen peroxide in the middle troposphere and the boundary layer. In Fig. 6, averages of hydrogen peroxide mixing ratios based on in situ observations, PSS calculations, and EMAC simulations and the corresponding H2O2 (EMAC) / H2O2 (measurements) and H2O2 (PSS) / H2O2 (measurements) ratios are shown. The data are binned into subsets of 1 km of altitude with respect to the center of the bin width based on takeoff and landings in proximity to the base of operations in Sal, Cabo Verde. The lines and the shadings represent mean values and the 1σ standard deviations. Dashed lines represent median values.

In general, the observed, PSS-calculated, and EMAC-simulated vertical profiles of hydrogen peroxide follow the expected trend throughout the troposphere (Fig. 6a–b). The lowest hydrogen peroxide mixing ratios of 0.141 ppbv were measured in the upper troposphere, where the availability of the H2O2 precursor HO2 is limited due to low water vapor concentrations. The highest mean values (±1σ) of 2.44 (±0.78) ppbv were measured directly above the boundary layer (2–5 km) in the free troposphere. Below 2 km, the levels of observed H2O2 decrease to 1.7 (±1.1) ppbv, reflecting the impact of deposition processes on H2O2 in the boundary layer in proximity to Cabo Verde. Based on the good agreement of the observed vertical distribution with the expected trend as well as logbook reports, the presence of clouds and their subsequent scavenging is assumed to have a minor impact on the local budget of the species.

A good agreement between the measured and EMAC-simulated datasets with a H2O2 (EMAC) / H2O2 (measurement) ratio of approximately 1 was found in the lower troposphere (2–6 km; Fig. 6c). Model results tend to overestimate hydrogen peroxide in the boundary layer, which might be due to the model resolution (1.8∘ × 1.8∘) and the corresponding restrictions in resolving small-scale variations in hydrogen peroxide deposition processes in proximity to the base of operations on the island. The vertical profiles of the observations and the model show that the differences arise mainly in the upper troposphere, with increased ratios of simulated vs. measured H2O2 of up to a factor 5.6 at 12 km altitude, which might indicate convective outflow in the model. Vertical profiles of observed and simulated HO2, OH, and H2O2 photolysis rates are in excellent agreement (Fig. S6), indicating that the model accurately simulates photochemical processes throughout the troposphere, so the remaining differences for H2O2 are most likely caused by physical processes (e.g., deposition and transport).

The comparison of measured and PSS-calculated H2O2 vertical profiles indicates a missing source or an overestimated sink below altitudes of 5 km (Fig. 6b). Here, the PSS-calculated hydrogen peroxide levels fall short by about 1 ppbv at altitudes below 4 km, yielding H2O2 (PSS) / H2O2 (measured) ratios smaller than unity at these altitudes (Fig. 6d). The absolute difference between the measured and calculated mixing ratios seems to be very prominent in and directly above the boundary layer (< 5 km) and can be associated with air masses affected by Saharan dust, which was often sampled during takeoff and landings at Sal. Heterogeneous loss of HO2 on desert dust particles, while modest, is expected to lower the production of H2O2 (de Reus et al., 2005). Thus, local mixing ratios of HO2 could be affected by heterogeneous loss, while PSS is not yet achieved (Fig. S7).

In order to investigate the extent of the potential hydrogen peroxide injection into the upper troposphere, we calculate excess hydrogen peroxide mixing ratios as the difference between the observed and the corresponding H2O2 based on PSS. Analogously, potential excess of H2O2 using model-simulated data was determined. The spatial distribution of the calculated excess H2O2 mixing ratios in the upper troposphere is presented in Fig. 7 as latitude vs. longitude plots of mean hydrogen peroxide levels aggregated over a spatial 1∘ × 1∘ grid at altitudes above 8 km. The color scale represents the average excess mixing ratios determined for the species (in pptv).

The H2O2 observations exceed the calculated values based on PSS mostly in the range of 70 to 110 pptv with the exception of the values derived in the Southern Hemisphere. Highest deviations were derived in the ITCZ and reach up to 310 pptv at their maximum. The excess mixing ratios show a clear trend with the most impactful injection events in the ITCZ region (5–15∘ N), where convective transport is expected (Waliser and Gautier, 1993; Fontaine et al., 2011). From there, a subsequent redistribution of hydrogen peroxide towards the Northern Hemisphere and Southern Hemispheres occurs, which agrees well with the decreasing gradient towards the north and south. EMAC simulates exceeding H2O2 mixing ratios mostly in the range of 20 to 150 pptv, with maximum excess up to 240 pptv in the ITCZ (12.5∘ N, 25.5∘ W; Fig. 7b). EMAC reproduces the transport rates to a lower extent in the ITCZ, but local convective events such as those occurring within tropical storm Florence and above the African coast are not simulated by the model. In contrast, no significant excess of H2O2 was determined at 1.5∘ N, 4.5∘ W.

The calculated absolute difference between the measured and the PSS-calculated H2O2 in the upper troposphere displays an average excess of 44 (±120) pptv hydrogen peroxide over the entire region relative to the PSS-modeled conditions. In comparison, injections based on EMAC simulations show about 60 % lower values of 18 (±120) pptv. The mixing ratios of the H2O2 measurements below 4 km at Sal were in the range of 1.7–2.4 ppbv, which would contribute with up to 1.8 %–2.6 % of H2O2 in the outflow, assuming potential inflow below 4 km. A similar calculation based solely on EMAC data indicates a contribution of 0.6 %–0.8 % within the model (based on EMAC mixing ratios of 2.1–3.0 ppbv). Thus, although EMAC reproduces potential transport processes from the lower troposphere, the discrepancy in the transmission efficiencies indicates a smaller contribution to the simulated hydrogen peroxide levels. Consequently, the enhanced hydrogen peroxide mixing ratios cannot be justified exclusively by photochemical reactions within the upper troposphere. Additional injections from the lower troposphere via convective transport and the subsequent redistribution towards the subtropics have to be considered.