The HOOVER project included a total of two measurement campaigns in October 2006 and July 2007, composed of four measurement flights per campaign. From the home airport of Hohn (Germany; 54.2∘ N, 9.3∘ E) regular research flights were performed southbound with a stopover at Bastia in Corsica (France; 42.2∘ N, 9.29∘ E) and northbound with a stopover at Kiruna airport (Sweden 67.5∘ N, 20.2∘ E). The majority of the flights were performed in the UT, while regular profiles were flown in and out of the home and stopover airports, as well as halfway towards the respective destinations over either southern Germany or northern Scandinavia. Additional flights in summer 2007 were directed to the Arctic (Svalbard, Norway; 78.1∘ N, 15.3∘ E) and two flights over central Germany to study the influence of deep convection. On 19 July 2007, an eastward moving mesoscale convective system developed over the southern part of Germany. During a research flight (HOOVER II flight no. 7) out of Baden airport (48.4∘ N, 8.4∘ E) the in- and outflow of a strong convective cell were probed close to Dresden, the capital of the Free State of Saxony in Germany. Further details about the campaigns can be found in several previous publications (Klippel et al., 2011; Regelin et al., 2013; Bozem et al., 2017).

During HOOVER, a Learjet 35A from GFD GmbH (Hohn, Germany) was used. The jet aircraft has a range of about 4070 km and a maximum flight altitude of approximately 14 km. In the present configuration, both the range and height maximum were reduced due to the use of two wing pods that housed additional instruments. The instrumentation consisted of a chemiluminescence detector (CLD 790 SR, ECO Physics, Switzerland) for NO, NO2 and O3 measurements (Hosaynali Beygi et al., 2011); a set of up- and downward looking 2π-steradian filter radiometers for j(NO2) measurements (Meteorologie Consult GmbH, Germany); a quantum cascade laser IR-absorption spectrometer for CO, CH4 and HCHO measurements (Schiller et al., 2008); a dual enzyme fluorescence monitor (model AL2001 CA peroxide monitor, Aero-Laser GmbH, Germany) to measure H2O2 and organic hydroperoxides (Klippel et al., 2011); a laser induced fluorescence (LIF) instrument for simultaneous measurements of OH and HO2 (Martinez et al., 2010; Regelin et al., 2013); a non-dispersive IR-absorption instrument (model LI-6262, LI-COR Inc., USA) for CO2 and H2O measurements (Gurk et al., 2008); a proton transfer reaction mass spectrometer (PTR-MS, Ionicon, Austria) for partially-oxidized VOC measurements and a series of canisters for post-flight analysis of non-methane hydrocarbons (Colomb et al., 2006). Details about the instrument performance with respect to time resolution, precision, detection limit and total uncertainty can be found in Klippel et al. (2011), Regelin et al. (2013) and Bozem et al. (2017). For the two species, which are central in the present study, some more details will be given next.

The hydrogen peroxide measurements have a time resolution of 30 s (time for a calibration signal to rise from 10 to 90 % of total reading), a detection limit of 24 pptv (deduced from the 1σ reproducibility of in-flight zero air measurements) and a precision of ±8.3 % at 260 pptv (deduced from the standard deviation (1σ) reproducibility of in-flight calibrations with a liquid standard) resulting in a total uncertainty of ±13.9 % at 260 pptv (Klippel et al., 2011).

The formaldehyde measurements have a time resolution of 30 s (averages over ∼ 75 HCHO spectra at a duty cycle of 60 %, while the remainder of the cycle is dedicated to CO and CH4 measurements), a detection limit of 32 pptv (deduced from the 1σ reproducibility of in-flight zero air measurements) and a total uncertainty of ±9 % (Klippel et al, 2011; Schiller et al., 2008). All data used in this study have been averaged over a time interval of 30 s.

In order to estimate photochemical destruction and secondary production of HCHO and H2O2 in the UT after convective injection, the box model MECCA (Module Efficiently Calculating the Chemistry of the Atmosphere) has been used. MECCA uses an extensive chemistry mechanism for gas-phase and liquid-phase chemistry (Sander et al., 2005). It uses the “Kinetic PreProcessor” (KPP), a flexible package for the numerical integration to translate the chemical mechanism into a set of ordinary differential equations (Sandu and Sander, 2006). For atmospheric studies, MECCA is coupled to either the atmospheric chemistry-general circulation model EMAC (ECHAM5/MESSy2 Atmospheric Chemistry, http://www.messy-interface.org) or the box model CAABA (Chemistry As A Boxmodel Application). The coupling is realized by the MESSy (Modular Earth Submodel System) interface (Jöckel et al., 2006).

For this study, CAABA/MECCA was set up with basic methane, VOC and isoprene chemistry following the Mainz isoprene mechanism (MIM) (Pöschl et al., 2000), a condensed version of the detailed Master Chemical Mechanism (Jenkin et al., 1997). This resulted in a mechanism for tropospheric gas-phase chemistry including 76 species and 153 reactions. Halogen chemistry and liquid phase or heterogeneous chemistry was not included since model calculations were restricted to cloud-free air. Additionally, deposition processes, either wet or dry, were not considered for the same reasons. Equations were solved by a Rosenbrock scheme (Sandu and Sander, 2006). Details of the model set-up and initialization will be discussed in Sect. 3.3.

The meteorological conditions in Europe on 19 July 2007 are shown in Fig. 2. A pool of cold air, associated with a long trough extending from the North Atlantic over the British Isles to the Azores, led to a low pressure system between Spain and Ireland (Fig. 2a). Towards the east, a high pressure ridge extending from North Africa over the eastern Mediterranean, eastern Europe and into Russia started to weaken. Between these systems a strong south-westerly flow was established in the middle and upper troposphere, bringing moist and warm air from Spain to north-western Europe, with a surface frontal zone separating moist, hot air in the south-east from dry, cold air in the north-west (Fig. 2b). Over Germany these conditions led to the development of thunderstorms, favoured by a potentially unstable troposphere with high convective available potential energy (CAPE) values over south-western Germany.

During the night from 18 to 19 July, convective cells developed along a weakening cold front over France, which rapidly developed into a mesoscale convective system (MCS) moving into south-western Germany in the early-morning hours of 19 July. This MCS subsequently travelled in a north-easterly direction accompanied by heavy rain (Fig. 3). Around noon the south-eastern edge of the MCS reached the Nürnberg/Bamberg area. The explosive storm development came with three strong, isolated cells which expanded throughout the whole tropospheric column up to 10 km altitude within 2 hours. The convection in the three cells was accompanied by strong lightning activity. Figure 4 shows the position (Fig. 4a) and intensity (Fig. 4b) of detected lightning flashes on 19 July 2007 between 00:00 and 22:00 UTC. Based on the temporal evolution of the lightning activity (colour code in Fig. 4a), the movement of the MCS over Europe can be traced up to the rapid thunderstorm development over south-eastern Germany (green and yellow points in Fig. 4a). During this phase the number of lightning flashes increased to 50 min-1 (Fig. 4b).

Figure 5a–c shows the RADOLAN (RADar-OnLine-ANeichung) data of the precipitation radar of the German weather service for the time period of rapid development over the Nürnberg/Bamberg area. During the initial storm development (Fig. 5a), from 11:30 to 12:30 UTC, precipitation is observed associated with the MCS. Later on (12:30 to 13:30 UTC; Fig. 5b), the isolated thunderstorm cells, being investigated here, emerged. Superimposed on the figure is the flight track of the Learjet after takeoff from Baaden airport to the UT ahead of the MCS. The precipitation signal from three not fully separated thunderstorm cells in the time interval 13:30–14:30 UTC together with the flight track of the Learjet is shown in Fig. 5c. The strongest cell at the most northerly position with a maximum precipitation intensity of 30 mm h-1 was intensively probed by the research aircraft, performing measurements around the cell at various altitudes in, above and below the outflow during the final period (14:30 to 15:30 UTC; Fig. 5d). The outflow, which occurred at an altitude of 10.5 km, was identified in-flight by the strong enhancements of a number of trace gases (see Sect. 3.2) and sampled for about 5 min. Due to the close proximity of the three cells, the outflow cannot be assigned unambiguously to an individual cell but on-board wind measurements strongly support our interpretation that the outflow originated from the northernmost cell.

The research flight was continued by measurements ahead of the weakening MCS. Around Dresden, descents into the continental boundary layer were flown, followed by ascents to obtain trace gas profiles and to characterize the potential inflow region, spending 15 min in the boundary layer. Although the inflow region cannot be identified unambiguously the storm track and local winds support our interpretation.

Figure 6 shows a time series of trace gas measurements (CH3OH, CH3COCH3, HO2, OH, NO, O3, H2O2, HCHO, CH4, CO and flight altitude) from HOOVER flight 7 on 19 July 2007. The first part of the flight was conducted in the wake of the eastward-moving MCS and up-wind of the developing thunderstorm cells. Signatures of convection can be identified during various parts of the flight. The first significant enhancements in CO and CH4 between 13:00 and 13:10 UTC at an altitude between 7.5 and 8 km were influenced by activity of the MCS that moved over the area earlier during the day. Signatures of recent convection can also be identified from enhancements of O3 and CH3COCH3, while mixing ratios of shorter-lived species e.g. NO, HCHO, H2O2 and HOx-radicals are only slightly influenced. The lack of enhancements of these species and the enhancement of O3 point towards ozone build-up, and indicate that we most likely probed an aged air mass from an up-wind convective event that had occurred recently (DeCaria et al., 2005).

Signatures of rather fresh convective injection into the UT are found between 13:45 and 13.58 UTC, with enhancements above background concentrations observed for CH3OH, CH3COCH3, NO, H2O2, CH4 and CO and a slight decline of O3 mixing ratios, indicating recent transport from lower layers of the troposphere. Formaldehyde does not show clearly enhanced mixing ratios. The strong increase in NO indicates production by lightning, which was directly observed in the area (see Fig. 4).

The major convective event associated with the northernmost thunderstorm cell was probed between 14:27 and 14:42 UTC (marked by the red box in Fig. 6). Before entering the outflow, but in close proximity to the convective cell, all trace gases with the exception of O3 showed significant decreases in mixing ratios. Between 14:10 and 14:27 UTC, CO and CH4 dropped to mixing ratios around 65 and 1770 ppbv, respectively, while O3 increased to 130 ppbv, indicating downward transport from the stratosphere. Maps of potential vorticity indicate that the local tropopause was about 1–2 km above the aircraft flight track. It has been postulated by Poulida et al. (1996) and later demonstrated by Pan et al. (2014) that convective cells in the mid-latitudes are often surrounded by O3 rich air masses associated with transport from the stratosphere, and this phenomenon was also observed during this flight.

As mentioned above, the outflow of the northernmost cell was probed between 14:27 and 14.42 UTC in cloud-free air. Here all trace gases and radicals under investigation show significant changes in mixing ratios. Table 1 lists the changes in mixing ratios in the outflow, the surrounding UT and the enhancement ratio (outflow : surrounding air). The mixing ratios for the surrounding air masses were determined from a part of the flight around 14:15 UTC up-wind of the thunderstorm and before the period affected by the stratospheric influence described above. Values greater than unity for enhancement ratios are reported for all species except O3, with a ratio less than 1. These enhancements vary between 4 % for CH4 and 700 % for NO, while O3 in the outflow was 20 % lower than in the surrounding air masses. Even higher enhancement ratios are observed for HCHO, though also related to the uncertainty in the mixing ratio for the surrounding air mass that was below the detection limit of the instrument. Besides HCHO, the strongest enhancement is observed for NO, most probably due to lightning.

After probing the outflow, the aircraft performed a descent into the boundary layer ahead of the thunderstorm over Dresden (marked by the blue box in Fig. 6). Vertical profiles from this part of the flight (between 14:20 and 15:25 UTC) are shown in Fig. 7. Visual inspection of Fig. 7 indicates that mixing ratios for the longer-lived, insoluble trace gases CO, CH4, CH3COCH3 and CH3OH in the outflow are of the same order of magnitude as in the inflow area, i.e. the boundary layer towards the north-east. The mixing ratios of NO, O3 and OH in the outflow are higher than in the inflow, while for HO2, HCHO and H2O2 we find the reverse. The influence of the high NO on HOx partitioning has been addressed by Regelin et al. (2013) and will not be discussed here. From the behaviour of the longer-lived, insoluble tracers, and assuming that the measurements in the boundary layer are representative of the in-flow of the thunderstorm cell, we infer that the convection transported nearly undiluted boundary-layer air into the UT. Table 2 shows that the median (mean ±1σ) ratios of outflow to inflow mixing ratios for CO, CH4, methanol and acetone were 0.93 (0.94 ± 0.04), 0.99 (0.99 ± 0.01), 0.93 (0.93 ± 0.10) and 1.22 (1.23 ± 0.12) respectively, and thus not significantly different from unity considering their 2σ variability. This indicates that the contribution of entrainment is insignificant for the outflow region, in line with previous observations. Hauf et al. (1995) concluded from a case study of a thunderstorm over Basel (Switzerland) that the cloud contained “protective cores”, in which air from the boundary layer was transported almost undiluted to the anvil. Similar observations were reported by Poulida et al. (1996) and Ström et al. (1999). Thus, the lower than unity values for outflow : inflow ratios for HCHO and H2O2 of 0.54 (0.55 ± 0.09) and 0.59 (0.61 ± 0.08), respectively, are most likely due to partial rain-out of these soluble species. Other studies (Bertram et al., 2007; Fried et al., 2016) derived much lower ratios between outflow and boundary layer inflow, indicating significant entrainment. To estimate the potential role of entrainment for our measurements we apply a two-box model (Cohan et al., 1999) to calculate outflow (OF) mixing ratios from the inflow (IN) and the entrainment (EN) according to OF=xEN+(1-x)IN.

Using values for OF, IN and EN from Tables 1 and 2 and Fig. 7 we derive the following entrainment rates: 24 % (CO), 26 % (CH4), 30 % (Acetone) and 19 % (methanol) indicating that roughly 75 % of the air in the outflow stems from the boundary layer. Thus assuming an average value of 25 % for the entrainment rate we calculate maximum mixing ratios for HCHO and H2O2 at the storm core of 2.05 and 1.82 ppbv, respectively.

As illustrated in the schematic in Fig. 8, the 5 min measurements in the outflow were made in clear, cloud-free air at a distance between 50 and 150 km from the anvil. At a wind speed of 30 m s-1 this would correspond to a transport time of 30 to 90 min, which is sufficient for secondary photochemistry (production or destruction) to become a significant contributor to the budgets of shorter-lived species like HCHO and H2O2. For example, Bozem et al. (2017) calculated the potential net ozone production rate from in situ observation in this convective event and derived a rate of 1.9 ± 0.28 ppbv h-1, indicating secondary ozone formation of 1–3 ppbv in the outflow, explaining the slight difference in O3 between the inflow (80.8 ppbv; 81.2 ± 2.7 ppbv) and outflow region (83.8 ppbv; 83.5 ± 2.6 ppbv) (Table 2).

As shown by Fried et al. (2008), the temporal evolution of the HCHO mixing ratio depends on the concentration of HCHO at the cloud top, the concentrations of HCHO precursors and radicals, and the processing time. Within the first few hours it is unlikely that the HCHO concentration will reach steady state. Here we will use a box model to simulate the temporal evolution of HCHO and H2O2 mixing ratios in the outflow region after exiting the cloud. Please note that this model study neglects mixing with background air. This is justified by the ratios between inflow and outflow given in Table 2 for the longer-lived species, which are close to unity, indicating insignificant mixing with surrounding air masses.

To calculate the temporal evolution of HCHO and H2O2 with the MECCA model, we constrained it with measured median OH, HO2, CO, CH4, CH3OH, CH3COCH3, O3 and NO mixing ratios, and photolysis rates derived from the radiation transfer model TUV (https://www2.acom.ucar.edu/modeling/tropospheric-ultraviolet-and-visible-tuv-radiation-model; Madronich and Flocke, 1999), based on observed j(NO2) frequencies. Calculations of chemical production and loss were performed for 2 hours with a time step of 15 min. In general, the model runs indicate a strong dependency of the mixing ratio evolution on processing time and the amount of HCHO and H2O2 leaving the convective cloud (initial values). At high HCHO and H2O2 initial concentrations, photochemical destruction due to photolysis and reaction with OH prevails (the loss terms are proportional to the initial concentration), while at low initial concentrations secondary production from precursors dominate. Therefore, a number of sensitivity studies were performed that take into account the uncertainties in initial values and processing time. To account for variations in the input parameters and uncertainties of the rate constants of the model we performed additional sensitivity calculations based on the Monte Carlo method, by varying initial concentrations within the 1σ uncertainties and rate constants by up to ±80 %.

Figure 9 shows the temporal evolution of HCHO mixing ratios in the outflow as simulated with MECCA. The blue area (median with standard deviation) indicates the observed HCHO concentration in the outflow (1.45 ± 0.11 ppbv). Sensitivity studies were performed with different initial HCHO concentrations. A first run was performed with an initial HCHO mixing ratio of 0.02 ppbv, corresponding to background conditions (Table 1). This case study represents near-complete removal of HCHO during convective uplifting due to cloud processing, followed by subsequent secondary production from HCHO precursors in the cloud-free outflow. Note that all precursors and radical levels are initialized at observed values in the outflow and do not change during the processing. The red curve in Fig. 9 starts at background HCHO mixing ratios, and exhibits significant production of HCHO during the first 60 min with a rate of approximately 0.01 ppbv min-1 with a maximum HCHO mixing ratio of 0.56 ppbv (range 0.51–0.61 ppbv), corresponding to the minimum and the maximum of the Monte Carlo simulation after around 90 min, and slowly decreasing values afterwards. These results compare well to box model studies of secondary HCHO formation after convective events reported by Stickler et al. (2006) and Fried et al. (2008). Since the mixing ratios in this simulation are at any time smaller than values observed in the outflow, we can assume that a significant portion of HCHO observed in the outflow is due to vertical transport from the inflow region.

A second run (not shown) was performed with an initial HCHO mixing ratio corresponding to the observed value in the outflow (1.45 ± 0.11 ppbv). Photochemical loss of HCHO through photolysis and reaction with OH dominates the temporal evolution of formaldehyde in this run, with loss rates of up to 0.02 ppbv min-1. This case study demonstrates that HCHO initial values have to be larger than observed mixing ratios in the outflow if we take photochemical processing into account. On the other hand, if we fully neglect photochemical processing by setting the processing time to zero, the observed concentration in the outflow is identical to the amount transported upwards in the convective cloud, providing a lower limit for the cloud top outflow. In a third sensitivity study we assumed undiluted transport of HCHO from the boundary layer to the cloud top. Starting with an initial value of 2.70 ± 0.42 ppbv corresponding to the HCHO mixing ratio in the inflow, the green curve in Fig. 9 shows strong photochemical loss of HCHO throughout the simulation. The envelope given by the minimum and maximum of the initial concentrations and the Monte Caro simulation intercepts the observed HCHO mixing ratio range between 20 and 52 min (vertical dashed lines). Due to both the uncertainty of the HCHO initial values and the elapsed processing time it is not possible to assign a single value to the HCHO mixing ratio in the cloud outflow. Instead we used a range corresponding to a minimum value given by the measured HCHO mixing ratio in the cloud-free outflow (1.45 ± 0.11 ppbv) and assuming zero processing, and a maximum value assuming undiluted HCHO transport from the inflow region (2.70 ± 0.42 ppbv) and a processing time of 20 to 52 min. To study the role of entrainment a final sensitivity run (black dashed line) was performed with a mixing ratio for HCHO at the storm core of 2.05 ppbv, calculated based on an entrainment rate of 25 %. This curve is only slightly below the green curve and yields slightly lower processing times.

Figure 10 shows the temporal evolution of H2O2 in the outflow region calculated with MECCA. Independent of the initial H2O2 mixing ratio all simulations indicate a photochemical loss of hydrogen peroxide in the first 2 h after convective injection. Assuming a processing time between 20 and 52 min deduced from the HCHO study (vertical dashed lines), the best fit for the initial H2O2 concentration is obtained for a range between 1.42 and 1.45 ppbv (brown trace in Fig. 10). Lower initial values (red trace for background conditions corresponding to zero transport) or higher values (green trace for complete transport from the inflow region to the outflow and the black dashed line for an assumed entrainment rate of 25 %) are at no time compatible with the observations (blue bar). As in the case of HCHO we will again provide a range of H2O2 initial values that are compatible with the observations in the outflow, with a minimum of 1.25 ± 0.09 ppbv given by the observed mixing ratio and no photochemical processing and a maximum of 1.435 ± 0.015 ppbv provided for the model best fit at processing times between 20 and 52 min. It should be mentioned that the MECCA simulations are based on gas-phase chemistry only. Hydrogen peroxide formation in the liquid phase by HO2 or O3 dissolution is ignored. Based on short transport times from the boundary layer to the UT of the order of 30 min at vertical velocities of 5 m s-1 and a depth of 9 km, the H2O2 production by these processes is negligible (Prather and Jacob, 1997; Jacob, 2000).

Based on the model results presented above, we estimate the HCHO mixing ratio at the cloud top exit to be in the range between 1.45 and 2.70 ppbv. This corresponds to the amount of HCHO transported from the inflow region to the top of the cloud. Assuming no chemical processing in the cloud, the ratio between the modelled cloud top mixing ratio and the mixing ratio in the inflow region yields the transport efficiency for HCHO at a range of 53–100 %. This indicates a strong contribution of transport to the HCHO budget in the outflow, which corresponds to minor cloud scavenging, i.e. at an efficiency between 0 and 47 % (amount of HCHO lost within the cloud). For H2O2 a similar analysis yields a cloud top mixing ratio of 1.25 to 1.45 ppbv, a transport efficiency of 59 to 68 % and a corresponding scavenging efficiency of 32 to 41 %.