The hydroperoxide data (H2O2obs and total organic hydroperoxides, ROOHobs) during OMO were obtained using a modified commercial instrument (AEROLASER, model AL2021, Garmisch-Partenkirchen, Germany) called HYPHOP (hydrogen peroxide and higher organic peroxide monitor). The HYPHOP instrument was installed in a 19′′ rack together with the infrared laser absorption instrument TRISTAR (tracer in situ TDLAS for atmospheric research) mounted in the back of HALO. Air was sampled from the top of the aircraft fuselage through a forward-facing trace gas inlet (TGI) designed as a bypass, consisting of a 1/2′′ PFA (perfluoroalkoxy alkane) tube inside the aircraft with an exit through a second TGI. From this bypass, air was sampled at a flow rate of 2 slpm (standard liters per minute) through a 1/4′′ PFA tube and directed to HYPHOP. To obtain constant pressure at the HYPHOP inlet, a constant pressure inlet (CPI) consisting of a dual-stage membrane pump (Vacuubrand MD1C VARIO SP, Wertheim, Germany) was used (Klippel et al., 2011).

HYPHOP relies on a dual-enzyme detection method after transfer of gaseous hydroperoxides into a buffered solution (potassium hydrogen phthalate and NaOH, pH 6) in a glass stripping coil (Lazrus et al., 1985, 1986). This stripper also contains EDTA (ethylenediaminetetraacetic acid) to prevent the reaction of transition metal ions with the hydroperoxides. Additionally, formaldehyde (HCHO) is added to prevent the oxidation of dissolved SO2 (in alkaline solutions HSO3-) by the hydroperoxides. Instead, HCHO and HSO3- form hydroxymethyl sulfonate (HOCH2SO3-). After the stripping coil the hydroperoxide-containing solution is divided into two channels. Catalase is added to one channel in order to selectively destroy H2O2. This first channel thus measures only ROOH, while the second channel (without catalase) measures the sum of ROOH and H2O2. Since hydroperoxides cannot be detected by fluorescence directly, a second enzyme (horseradish peroxidase) and p-hydroxyphenylacetic acid (POPHA) are added to both channels. In a quantitative and selective reaction, the enzyme catalyzes the oxidation of POPHA by hydroperoxides, forming the fluorescent dye 6,6′-dihydroxy-3,3′-biphenyldiacetid acid. After excitation at 326 nm with a Cd lamp, the fluorescence at 400–420 nm is detected. To enlarge the fluorescence intensity, sodium hydroxide is added.

In order to perform zero measurements, the sampled air is directed through a cylinder filled with Hopcalite (MnO2 and CuO) to eliminate H2O2, ROOH, and ozone. Since the efficiency of Hopcalite decreases with increased humidity, the air is dried beforehand with the help of orange gel (SiO2 beads plus indicator).

To convert the detected signal into a concentration, a four-point calibration was performed before and after every flight. In the first two steps a liquid standard of H2O2 (1 µmol L-1; freshly diluted from stock solution) followed by zero air is measured in both channels without catalase. Afterwards this is repeated with catalase in the ROOH channel for the last two steps. The sensitivity for both channels and the catalase efficiency are determined via this procedure. The concentration of the liquid standard is based on titration of the stock solution (10 mmol L-1) with potassium permanganate.

To determine the stripping efficiency for H2O2, a gas-phase standard based on a permeation source (Teflon tube filled with 30 % H2O2 in a temperature-controlled glass flask) is used at a constant flow rate of approximately 40 sccm, diluted with synthetic air and measured with the instrument. The permeation rate of the source is quantified by collecting the output of the source into cooled water. The addition of hydrochloric titanium tetrachloride yields the formation of the yellow η2-peroxo complex [Ti(η2-O2)Cl4]2- (Pilz and Johann, 1974), whose concentration is determined via a UV photometer. The stripping efficiency of MHP was assumed to be 60% and that of H2O2 to be 100 % (AEROLASER, 2006; Lee et al., 2000).

The inlet efficiency was determined with the help of the permeation source, which was measured with and without the CPI. In laboratory studies the inlet efficiency was determined to be 87 % ± 3 %, decreasing during the campaign to 62.7 % ± 0.8 %, which is mainly due to the higher humidity.

The limit of detection (LOD) and precisions for H2O2 and MHP (assuming total ROOHobs to be only MHP), respectively, have been calculated for each flight from the reproducibility (1σ standard deviation) of in-flight zero (650 values) and liquid calibration (100 values) measurements, taking into account the sensitivity, stripping, and catalase efficiency. LOD values are in the range of 8–53 pptv for H2O2obs (median 23 pptv) and 9–52 pptv for ROOHobs (median 23 pptv), respectively, assuming that ROOHobs is composed of MHP only. Please note that due to the fact that the exact composition of ROOHobs is unknown, and the solubility of different ROOH species can be quite variable, a detection limit for ROOHobs cannot be given. Instead, we calculate an upper limit of the detection limit, assuming that all ROOHobs consists of MHP, the species with the smallest solubility. Precision values were determined from the reproducibility of standard measurements and are in the range of 0.2 % at 5.2 ppbv and 1.3 % at 5.9 ppbv for H2O2 and 0.3 % at 5.0 ppbv and 2.1 % at 6.0 ppbv for MHP. The time resolution (signal increase from 10 % to 90 %) of the instrument is 120 s. An ozone interference of 53 pptv H2O2 per 100 ppbv O3, which was determined by H2O2 measurements in the stratosphere during the OMO-EU test campaign, was taken into account and corrected.

The total uncertainty calculated from statistical errors and uncertainties of liquid standard, inlet and stripping efficiency, and ozone interference is 25 % for H2O2 and 40 % for MHP.

For this study CO, CH4, OH, HO2, O3, acetone, NO, NOy, JH2O2, and JMHP data measured by other instruments have been used for data interpretation, steady-state calculations, and interference corrections (see Sect. 3.1). A complete list of all measured compounds can be found in Lelieveld et al. (2018). CO and CH4 have been measured by the IR quantum cascade laser absorption spectrometer TRISTAR (Schiller et al., 2008; Tadic et al., 2017). The measurements comprised an ambient-air mode and in-flight calibrations. The latter were realized with secondary standards from pressurized bottles (6 L bottle, Auer GmbH, Germany), which were calibrated against certified reference gases (Tomsche et al., 2019). With the help of the in-flight calibrations, the in situ data are drift-corrected by interpolation between two calibrations (Tadic et al., 2017). The observed CO and CH4 mixing ratios have a total uncertainty of 5.1 % and 0.275 %, respectively. The relatively high CO uncertainty reflects problems with the stability of the CO quantum cascade laser during the second half of OMO.

Laser-induced fluorescence was the method utilized for HOx measurements (instrument name: HORUS; Faloona et al., 2004; Martinez et al., 2010). The accuracies of the measurements are 17.1 % for OH and 17.6 % for HO2. The limit of detection of the instrument does vary depending on altitude as this system has a sensitivity that depends on pressure. As altitude increases the LOD decreases from 0.1 pptv to 0.02 pptv for OH and 0.361 to 0.175 pptv for HO2.

FAIRO (fast airborne ozone instrument) is a lightweight (14.5 kg) and accurate two-sensor device for measuring O3. It combines two techniques, i.e., (a) a UV photometer that measures the light absorption by O3 at a wavelength of λ = 250–260 nm emitted by a UV-LED and (b) a chemiluminescence detector that monitors the chemiluminescence generated by O3 on the surface of an organic dye adsorbed on dry silica gel. These techniques are simultaneously applied in order to combine the high measurement accuracy of the UV photometry with the high measurement frequency of the chemiluminescence detection. The UV photometer has a 1σ precision of 0.08 ppbv at a measurement frequency of 0.25 Hz (and a pressure of 1 bar) and an accuracy of 1.5 % (determined by the uncertainty of the O3 cross section). The chemiluminescence detector has a precision of 0.05 ppbv at a measurement frequency of 12.5 Hz (Zahn et al., 2012). In postprocessing the chemiluminescence detector data are calibrated using the UV photometer data.

Nitrogen oxide (NO) and total reactive nitrogen (NOy) were measured using the AENEAS atmospheric nitrogen oxide measuring system. The measurements were performed by a dual-channel NO chemiluminescence detector (CLD-SR 790, Eco Physics, Switzerland) in combination with a converter technique for the detection of total reactive nitrogen as NO. NOy comprises, among others, NO, NO2, HNO3, NO3, N2O5, HNO2, HO2NO2, PAN, and organic nitrates. The individual NOy species were detected after conversion to NO using a gold tube maintained at about 300 ∘C, with H2 as a reducing agent (Ziereis et al., 2000). Ambient air was sampled using a standard HALO trace gas inlet equipped with a heated (∼40 ∘C) PFA inlet line. The time resolution of the measurements was about 1 s. The overall uncertainty of the NO and NOy measurements depends on its ambient concentrations and is about 8 % (6.5 %) for volume mixing ratios of 0.5 nmol mol-1 (1 nmol mol-1), respectively (Stratmann et al., 2016).

VOCs (e.g., acetone) were measured with a homebuilt lightweight (∼55 kg without rack) proton-transfer-reaction mass spectrometer, which uses a commercial quadrupole mass analyzer (Pfeiffer, QMA 410, Germany). A modular V25 microcomputer system (MPI-C, Mainz, Germany) is applied for instrument control and data acquisition. A custom-built inlet system comprises a platinum–quartz wool scrubber (Shimadzu, High Sensitivity Catalyst) held at 300 ∘C and components for flow and pressure control. The instrument was calibrated between flights with a dynamically diluted gas standard containing approximately 500 ppbv of VOCs (Apel-Riemer Environmental Inc., USA). The accuracy for acetone is typically ±10 %, and the detection limit is ∼60 pptv.

Photolysis frequencies were calculated from spectral actinic flux density spectra (280–650 nm) obtained from charge-coupled device (CCD) spectroradiometer measurements on the top and bottom fuselage of the aircraft, covering the upper and the lower hemisphere, respectively (Bohn and Lohse, 2017). Recent recommendations of absorption cross sections and quantum yields were used in the calculations as well as their temperature and pressure dependencies (if available) by taking into account measured static air temperatures and pressures. Radiometric uncertainties range around 5 %–6 % under typical flight conditions. Additional uncertainties related to the molecular parameters are process-specific. For H2O2 in particular, recommended absorption cross sections and their temperature dependencies were applied, and unity quantum yields were assumed (Burkholder et al., 2015). However, the recommended H2O2 absorption cross sections are confined to a wavelength range below 350 nm, which is insufficient to capture atmospheric photolysis completely. Because measured cross sections decay exponentially over 2 orders of magnitude in the range of 280–350 nm, this dependence was further extrapolated up to 370 nm, where values drop well below 10-22 cm2. Dependent on conditions, this extrapolation increases atmospheric H2O2 photolysis frequencies by 10 %–20 %. For MHP the temperature dependence of the absorption cross sections is unknown. Therefore the recommended room temperature data were used under all conditions as well as unity quantum yields (Burkholder et al., 2015). Combined total uncertainties of 15 % and 25 % are estimated for H2O2 and MHP photolysis frequencies, respectively.

Latitude, longitude, and altitude data as well as temperature and pressure were collected with the BAHAMAS (basic HALO measurement and sensor system) instrument. More detailed information about the installation of scientific instruments and mission flights can be found at http://www.halo.dlr.de/science/missions/omo/omo.html (last access: 27 February 2019).

Since only the sum of organic hydroperoxides was measured, we estimated the contribution of MHP using a photostationary-state (PSS) approximation relying on in situ measurements of HO2, OH, CO, CH4, NO, JMHP, and JH2O2 (see Sect. 3.2) and rate coefficient data from Atkinson et al. (2004, 2006).

In the free troposphere the production rate P of H2O2 and MHP is due to the self-reaction of HO2 and reaction of CH3O2 with HO2, respectively, and can be calculated from Eqs. (1) and (2). 1PH2O2=kHO2+HO2⋅HO222PMHP=kCH3O2+HO2⋅CH3O2⋅HO2 Photochemical loss rates L of H2O2 and MHP are due to photolysis and reaction with OH according to Eqs. (3) and (4). 3LH2O2=kH2O2+OH⋅OH+JH2O2⋅H2O24LMHP=kMHP+OH⋅OH+JMHP⋅MHP For steady-state conditions the production and loss reactions are at equilibrium, and the MHPPSS-to-H2O2obs ratio can be calculated from Eq. (5). 5MHPPSSH2O2obs=kCH3O2+HO2⋅CH3O2⋅HO2kHO2+HO2⋅HO22⋅kH2O2+OH⋅OH+JH2O2kMHP+OH⋅OH+JMHP Because individual peroxy radicals were not measured, the CH3O2-to-HO2 ratio must be estimated from their production and loss terms. This ratio can be deduced as written in Eq. (6). 6CH3O2HO2=LHO2⋅PCH3O2PHO2⋅LCH3O2 Dominant loss processes for HO2 and CH3O2 are reactions with NO and the production of H2O2 and MHP, respectively, neglecting the production of peroxy nitrates due to low NO2 concentrations in the UT (Eqs. 7 and 8). 7LHO2=kCH3O2+HO2⋅CH3O2⋅HO2+kHO2+NO⋅HO2⋅NO+kHO2+HO2⋅HO228LCH3O2=kCH3O2+HO2⋅CH3O2⋅HO2+kCH3O2+NO⋅CH3O2⋅NO The first terms on the right side of both equations are identical. The second terms are dominated by the rate coefficients of the reactions with NO and the NO concentration. For the calculations of the rate coefficients, the mean temperature of 259.18 K, the mean altitude of 10 992.8 m, and the mean pressure of 22 932.9 Pa were used. The resulting values are shown in Eqs. (9)–(11). As the relative humidity is very low in the upper troposphere, the water dependence in Eq. (11) was neglected. 9kHO2+NO=3.45×10-12⋅exp⁡270T=9.78×10-12cm3(molec.s)-110kCH3O2+NO=2.3×10-12⋅exp⁡360T=9.22×10-12cm3(molec.s)-111kHO2+HO2=2.2×10-13⋅exp⁡600T+1.9×10-33⋅N2⋅exp⁡980T=2.64×10-12cm3(molec.s)-1 This indicates that the reaction of HO2 with NO is more than a factor of 3 faster than the self-reaction. The measured NO concentration is an order of magnitude larger than measured HO2 so that reaction with NO is the dominant process for both HO2 and CH3O2, resulting in similar loss rates for both radicals in the UT. Thus, the ratio of CH3O2 to HO2 is dominated by their production rates (Eq. 12). 12CH3O2HO2=PCH3O2PHO2=kCH4+OH⋅CH4⋅OHkCO+OH⋅CO⋅OH The combination of Eqs. (5) and (12) results in Eq. (13), which was used to calculate the MHPPSS concentrations based on the observed mixing ratios during OMO. 13MHPPSS=kCH3O2+HO2kHO2+HO2⋅kH2O2+OH⋅OHobs+JH2O2obskMHP+OH⋅⋅OHobs+JMHPobs⋅kCH4+OH⋅CH4obskCO+OH⋅COobs⋅H2O2obs Please note that other sources of HO2 and CH3O2, in particular the photolysis of formaldehyde (HCHO) and acetaldehyde, respectively have been neglected. This is justified by the generally low mixing ratios of these species at high altitudes. Measurements of HCHO with the TRISTAR instrument yielded values below the detection limit of 30 pptv (parts per trillion by volume), and although acetaldehyde was not measured, we assume that its mixing ratio is within a factor of 2 of those for HCHO.

The total uncertainty of MHPPSS from the calculation according to Eq. (13) can be deduced from error propagation, taking into account uncertainties in OHobs (17.1 %), JH2O2obs (15 %), JMHPobs (25 %), CH4obs (0.275 %), COobs (5.1 %), H2O2obs (25 %), and rate constants, to be of the order of 45 % (1σ).

To estimate the contribution of MHP to the total organic hydroperoxides, the calculated concentration of MHPPSS was subtracted from the measured sum of all organic hydroperoxides ROOHobs. This leads to a concentration of unidentified organic hydroperoxides (UHPPSS; Eq. 14). Please note that MHPPSS only accounts for local CH4 oxidation production and not for transport phenomena. 14UHPPSS=ROOHobs-[MHP]PSS

In order to classify the measured H2O2obs, HO2obs, and OHobs data, we calculated H2O2PSS from measured HOx and JH2O2obs. For the calculation, Eq. (15) was used, which is based on Eqs. (1) and (3). 15H2O2PSS=HO2obs2⋅kHO2+HO2OHobs⋅kH2O2+OH+JH2O2obs A total uncertainty of 45 % (1σ) due to uncertainties in OHobs (17.1 %), HO2obs (17.6 %), JH2O2obs (15 %), and reaction rate constants was calculated.

The EMAC (ECHAM–MESSy atmospheric chemistry) model comprises the fifth generation of the European Center Hamburg (ECHAM5; Roeckner et al., 2006; version 5.3.01) general circulation model and the Modular Earth Submodel System (MESSy; Jöckel et al., 2016; version 2.52, http://www.messy-interface.org/, last access: 25 August 2019). For this study EMAC simulations (T42L90, 2.8∘ × 2.8∘ horizontal resolution, 90 vertical levels to 0.01 hPa, time resolution 12 min) were sampled along the OMO flights tracks. Detailed specifications and results have been published previously (Lelieveld et al., 2018; Tomsche et al., 2019).

Back trajectories of 10 d were calculated along the flight path using FLEXPART to identify the air mass origin (Tomsche et al., 2019). Convective transport can be simulated in FLEXPART with the convection parameterization by Emanuel and Zivkovic-Rothman (1999). To represent moist convection realistically in models, the parametrization includes cloud microphysical processes, the physics of entrainment and mixing, and large-scale control of ensemble convective activity. It builds on temperature and humidity fields to provide mass flux information (Stohl et al., 2005). The back trajectories in the present paper are calculated with the convective parametrization. Further, the Lagrangian particle dispersion model FLEXPART produces so-called centroid trajectories based on the analysis of a cluster of trajectories. These trajectories are comparable to traditional trajectories but include convection via the centroid of all particles per time step. As indicated by Tomsche et al. (2019), prominent source regions of AMA air masses are identified to be the Indo-Gangetic Plain, northeastern India, Bangladesh, and the Bay of Bengal. Additionally, Tomsche et al. used observations of methane to differentiate between air masses influenced by the AMA and background air. A comparison of vertical profiles indicated that the air inside the AMA showed significantly higher CH4 concentrations than outside. Thus a threshold of CH4≥1879.8 ppbv was used to distinguish between air masses influenced by the AMA (CH4≥1879.8 ppbv), the SH background (CH4<1820 ppbv), and the NH background (1820 ppbv≤CH4<1879.8 ppbv; Tomsche et al., 2019).

Data were collected from a merged data set given as 60 s means (calculated from the original data set obtained at higher resolutions) in order to get the same time resolution for all compounds. The given time is the middle of the block mean.

For the histograms the concentrations of all species shown were binned into samples with a width of 10 pptv, starting with the plots with the lowest bin. To compare the simulations from EMAC with measured and PSS-calculated data, the corresponding values (out of the 60 s means) were used at the given times from EMAC.

In a case study analyzing Flight 17 from 10 August 2015, the method used to determine the origin of the measured air masses and a quantification and comparison of measured and simulated mixing ratios of H2O2obs, MHPPSS, and UHPPSS is presented. During this flight we encountered the SH and NH as well as the AMA in the UT. The flight track is shown in Fig. 2 (dotted line). Takeoff was in Gan (Maldives) and landing in Bahrain, on the Arabian Peninsula. Besides takeoff and landing, the entire flight took place in the UT (<230 hPa). The calculated back trajectories show the origin of the air masses. At the beginning of the flight the measured air masses had their origin over the Indian Ocean and Indonesia. During the remaining flight the measured air stemmed from India. Tomsche et al. (2019) showed that the measured air in the AMA was affected by deep convection over India, resulting in methane mixing ratios above the threshold. Figure 3 shows the time series for H2O2obs during the flight at the time steps given from the frequency of EMAC output (orange circles). The colored bar on top shows the origin of air masses, i.e., red for AMA, green for NH, and blue for SH. The H2O2obs mixing ratios vary between 128 and 366 pptv. The modeled H2O2EMAC data are in the range of 110–799 pptv (gray triangles). For the beginning of the flight, model simulations agree rather well with the measurement data. At around 06:00 UTC the model-calculated mixing ratios increase to more than 500 pptv, while the measured data decrease to 200 pptv and lower. In this period a maximum difference between model and in situ data of 386 pptv was found. After 1 h the EMAC model data decrease to 416 pptv, and the in situ data increase to 214 pptv. During the following hour until around 08:00 UTC and thus at higher altitude, both mixing ratios increase, with the modeled data showing a much stronger increase up to approximately 800 pptv, while the in situ data increase only to 230 pptv. During the last period of the flight, simulated and measured data are again in good agreement. Here the mixing ratios from EMAC are in the range of 110–157 pptv, while the measured data are in the range of 128–203 pptv. This steep drop of H2O2EMAC might arise due to a change in the flight altitude since between 08:01 and 08:06 UTC the aircraft changed from a flight level at 11 700 m to one aloft at 13 900 m.

In addition to EMAC simulations, Fig. 3 also shows the calculated H2O2PSS obtained from Eq. (15). Observed and PSS values for H2O2 mixing ratios agree very well, with a median deviation of 42 pptv, well within the combined uncertainties of measured data (25 %) and PSS simulations (45 %).

In Fig. 4 the time series of H2O2obs, MHPPSS, and UHPPSS mixing ratios are shown (5 min means). In the beginning of the flight, MHPPSS is the dominant organic hydroperoxide. The mixing ratios are in the range of 140–341 pptv, similar to those of H2O2obs (143–337 pptv). UHPPSS mixing ratios are in the range of 24–162 pptv, with a mean of 89 pptv. Later during the flight, UHPPSS are the dominant hydroperoxides in air masses inside the AMA originating from India. Here we found UHPPSS mixing ratios up to 275 pptv. The H2O2obs mixing ratios show a similar temporal pattern and mixing ratio levels to those of UHPPSS over the Arabian Sea and the Arabian Peninsula, with values in the range of 140–243 pptv. MHPPSS mixing ratios are much lower (62–130 pptv, median of 72 pptv) in this area. During this part of the flight, the similarity in the time series of acetone and UHPPSS (Fig. 5) indicates either similar source regions for both species or the role of acetone as a precursor for the UHPPSS in AMA-influenced air masses. From 06:20 UTC onwards an increase in both compounds is observed until 07:45 UTC, followed by a steep drop, with a minimum at 08:17 UTC. The UHPPSS and acetone mixing ratios in this part of the flight are strongly correlated (Fig. 6), with a slope of 0.19±0.02 (ppbv ppbv-1) and an offset of (-0.003±0.02) ppbv. The regression coefficient R2 is very high (0.82). For H2O2obs, MHPPSS, and ROOHobs the correlation is not that strong, with slopes of -0.02±0.02 (ppbv ppbv-1), -0.07±0.01 (ppbv ppbv-1), and 0.13±0.03 (ppbv ppbv-1), respectively, and offsets of (0.21±0.02) ppbv, (0.12±0.01) ppbv, and (0.11±0.03) ppbv (Fig. 6). The relation between ROOH mixing ratios and an air mass age tracer based on the ratio between [NO] and [NOy] shows higher values of ROOH at smaller ratios, representing older or more processed air masses (Fig. 7) since the highest ROOH mixing ratios (>200 pptv) are found at the lowest [NO] / [NOy] ratios (all <0.19). Thus, most of the observed ROOH were measured in aged air masses transported within the anticyclone. The correlation with UHPPSS shows that this effect is mainly due to UHPPSS. For H2O2obs there are also some higher mixing ratios for high [NO]-to-[NOy] mixing ratios and thus fresher air (Fig. 7).

To extend the analysis to the entire campaign, Fig. 8 shows all flight tracks in the UT during OMO. The color code represents observed mixing ratios of H2O2obs, MHPPSS, and UHPPSS varying from low (purple) to high values (red). Histograms for the whole campaign of H2O2obs mixing ratios as well as inferred MHPPSS and UHPPSS mixing ratios are presented in Fig. 9. Here only data from the UT (<300 hPa, which corresponds to altitudes >9 km) were included in the analysis. Mixing ratios for all species were further differentiated by methane levels such that data in air masses with CH4 mixing ratios above the threshold of 1879.8 ppbv were classified as AMA-influenced, while air masses with CH4 mixing ratios between 1820 ppbv and 1879.8 ppbv were classified as NH background and those with CH4<1820 ppbv as SH following Tomsche et al. (2019). Figure 9a indicates that H2O2obs mixing ratios are most abundant at values of 70–90 pptv in NH background air masses (green); 130 to 270 pptv with three notable peaks at 180–190, 210–220, and 250–270 pptv in SH air masses (blue); and 150–170 pptv in AMA-influenced air masses (red). The medians are 115 pptv for the NH background, 211 pptv for the SH, and 167 pptv for the AMA, indicating an excess of 52 pptv in AMA-influenced air masses compared to the NH background, while in the SH the H2O2obs mixing ratio is twice as high as the NH background. For MHPPSS (Fig. 9b) the frequency distribution in the NH background shows a maximum at 30–40 pptv (green). For AMA-influenced air a sharp maximum at 50–70 pptv (red) is found. Air masses from the SH exhibit a rather flat distribution, with a maximum at values of 40–50 pptv and a median of 152 pptv (blue). With median mixing ratios of 70 and 64 pptv, we found only slightly higher mixing ratios for AMA-influenced air masses in comparison to the NH background. For UHPPSS (Fig. 9c) we again found a flat distribution of mixing ratios in the SH (blue), with a maximum at values of 140–150 pptv and a median of 129 pptv. The maximum in the frequency distribution for NH background conditions is found at 70–90 pptv (green), while in AMA-influenced air masses it was significantly higher, with 210–230 pptv (red). Thus, mixing ratios of UHPPSS are approximately 2–3 times higher in the AMA outflow than in the NH background. This is also represented in the medians of 210 pptv in the AMA and 89 pptv in the NH background.

In the analysis of Flight 17 we found a strong correlation between UHPPSS and acetone (Fig. 6) and an increase in UHPPSS at the highest air mass ages, represented by low [NO] / [NOy] (Fig. 7). Extension of this analysis to all observations in the upper troposphere obtained during OMO yields similar results for the relation between UHPPSS, ROOHobs, and H2O2obs and acetone (Fig. 10). Enhanced mixing ratios of hydroperoxides are typically associated with enhanced acetone mixing ratios, especially for UHPPSS. A simple calculation of the production of MHP out of the photolysis of acetone and the reaction of acetaldehyde (from EMAC) with OH shows that per day approximately 40 pptv MHP can be formed within the AMA. The lifetime of MHP was calculated to be around 1.5 d. Thus not all of the UHPPSS in the AMA (median 210 pptv) can be accounted for MHP that was chemically produced from VOCs in the AMA. The scatterplots of the hydroperoxides vs. [NO] / [NOy] for the whole data set show no clear correlation with a large spread of hydroperoxide mixing ratios at the lowest [NO] / [NOy] ratios, representing the oldest, i.e., chemically most processed, air masses (Fig. 11).

To our knowledge we present the first observations of H2O2 and ROOH mixing ratios in the Asian monsoon anticyclone. Previous studies have been mainly focused on the northern hemispheric upper troposphere. Several aircraft campaigns including peroxide measurements were performed over North America. They are summarized in Snow et al., 2007): the SONEX campaign took place in fall 1997 in the UT and yielded mean values of 120 pptv for H2O2 and 50 pptv for MHP (medians: 80 and 30 pptv, respectively). The TOPSE campaign in winter/spring 2000 probed the middle troposphere, yielding median H2O2 and MHP mixing ratios of 150 pptv for both species. During the INTEX-NA campaign in summer 2004, observed median mixing ratios at altitudes of 6–10 km were about 400 pptv for H2O2 and 200 pptv for MHP. A comparison with our results (Table 2) shows that we found similar mixing ratios as in SONEX in the northern hemispheric background of 115 and 64 pptv for H2O2 and MHP, respectively. Mixing ratios for both species reported for TOPSE and INTEX-NA are slightly higher than ours, which may be related to the lower altitude range of 6–10 km (in comparison to >9 km for OMO) in these studies. Previous observations have shown that H2O2 and MHP show the highest mixing ratios at altitudes between 2 and 5 km followed by a sharp decrease towards higher altitudes (see e.g., Daum et al., 1990; Heikes, 1992; Weinstein-Lloyd et al., 1998; Snow, 2003; Snow et al., 2007; Klippel et al., 2011).

Heikes et al. (1996) associated enhanced H2O2 mixing ratios above 5 km in the North Pacific of the Asian coast (30∘ N) with outflow from Typhoon Mireille (Heikes et al., 1996). These observations were made close to the source region for the AMA-influenced air masses described here (see back trajectories in the case study of Flight 17; Fig. 2 or Tomsche et al., 2019). For MHP Heikes et al. (1996) found mixing ratios of 250–500 pptv in the southern longitudinal section above 5 km, similar to median mixing ratios of 152 pptv for MHP in SH air masses in the UT found in this study.

Although the mixing ratios observed during this study are similar to previous observations in the upper troposphere, one striking result is that a state-of-the-art global circulation model (EMAC) and a local steady-state calculation constrained by measured radical levels significantly underestimate H2O2 mixing ratios in particular in the AMA. The general tendency is that the steady-state model produces the lowest values, with EMAC falling in between steady state and observations (e.g., Fig. 16). A comparison of the EMAC simulations for the two radicals that affect H2O2 most strongly (OH and HO2) yields a rather good agreement. A scatterplot between modeled and observed HO2 yields a slope of 0.72±0.01 (pptv pptv-1) and an offset of 4.30±0.09 pptv, with a regression coefficient R2 of 0.58 (Fig. 18 left). The OH data show more scatter, with a tendency for EMAC to overestimate the mixing ratios (slope: 1.7±0.2 (pptv pptv-1); offset: -0.1±0.1 pptv; regression coefficient R2: 0.09; Fig. 18 right). Although there is rather good agreement between EMAC simulations and observations for all the species that affect the local photochemical budget of H2O2, EMAC significantly exceeds PSS calculation for H2O2. This is an indication that an additional H2O2 source is accounted for in the global model and that the local photostationary-state assumption is not fulfilled. The additional source is attributed to transport associated with deep convection over India, yielding an upwind source of H2O2 that is significant throughout the western part of the AMA. In the AMA, clouds are absent so that gas-phase photochemical processes may determine the lifetime of H2O2. Based on observed OHobs levels and photolysis frequencies during OMO, the H2O2 lifetime in the upper troposphere is of the order of several days, sufficiently long for the excess H2O2 to reach the western parts of the AMA, producing the observed longitudinal H2O2 gradient observed in both observations and EMAC simulations (Fig. 16). The total amount of H2O2 injected into the UT by convective outflow depends on the scavenging efficiency (Mari et al., 2000; Barth et al., 2016; Bozem et al., 2017). Differences between H2O2obs and H2O2EMAC are most likely due to an overestimation of scavenging in the model as also pointed out by Klippel et al., 2011).

To investigate this assumption we performed a sensitivity study with the wet scavenging for all soluble species being switched off globally. The result is shown in Fig. 19. The H2O2EMAC mixing ratios significantly increase with longitude by a factor of 3–4 and thus to the level of H2O2obs. Please note that significant enhancements in MHPEMAC and ROOHEMAC were not found in the sensitivity study with switched-off scavenging, indicating that the strong underestimation by the model of these species is not due to an overestimation of wet removal in convective clouds. Instead, we found that EMAC underestimates ROOH in all air masses and not only in the AMA. The reasons for this underestimation are unknown. In a previous comparison of MHP observations and EMAC simulations over Europe for July 2007, Klippel et al. (2011) also reported a difference of a factor of 10 in the upper troposphere, while a comparison during the fall season (October 2006) yielded a rather good agreement (within a factor of 2).

There is a rather large uncertainty regarding the scavenging efficiency of MHP in deep convection (Barth et al., 2016). For the Trace A campaign Mari et al. (2000) found observed (modeled) enhancement ratios of postconvective mixing ratios in comparison to preconvective mixing ratios of 11 (9.5) for MHP and 1.9 (1.2) for H2O2. Such efficient transport in the Indian summer monsoon would yield a strong source of upper-tropospheric MHP, explaining the large enhancement of ROOHobs in the AMA described here. Please note that large enhancements of MHPEMAC and ROOHEMAC were not found in the sensitivity study with switched-off scavenging, indicating that the strong underestimation by the model of those species is not due to an overestimation of wet removal in convective clouds. It seems that a large part of the UHPPSS is actually MHP advected throughout the AMA after deep convective transport over India. In the EMAC simulations the transport of MHP is less efficient, and thus MHPEMAC is lower than MHPPSS and UHPPSS. Please note that EMAC has a general tendency to overestimate CO in the UT, especially for the NH background, while it tends to underestimate CH4 (Tomsche et al., 2019). The deviations in general are not significant, with the exception of CH4 in the AMA (Table 1 in Tomsche et al., 2019). As discussed in Tomsche et al. (2019), the CH4 mixing ratio in the AMA depends of the colocation of convection and underlying methane sources. The model resolution of 2.8∘ × 2.8∘ is not sufficient to resolve small-scale variations in both convection and CH4 source distribution. The H2O2 mixing ratio over the Indian subcontinent is not expected to show large spatial variations since latitudinal gradients are generally small (see, e.g., Klippel et al., 2011). Therefore, we do not expect that the model resolution will have a strong influence on the deviation between H2O2obs and H2O2EMAC. Another uncertainty arises from missing information on the absolute mixing ratios of H2O2, MHP, and higher organic hydroperoxides in the inflow region of deep convection over India since observations of these species in the boundary layer over India are not available. Note that the amount of hydrogen peroxide and number of organic hydroperoxides transported to the upper troposphere depends on the scavenging efficiency and on the mixing ratios of the individual species in the inflow region (Barth et al., 2016; Bozem et al., 2017). Thus, an underestimation of hydroperoxides in the upper troposphere after convective injection can be due either to an underestimation of the scavenging efficiency for individual species, an underestimation of their mixing ratio in the inflow region, or a combination of both and might differ for individual hydroperoxides. Due to a lack of observations in the inflow and outflow region of convection over India, this question cannot be resolved in this study.