For the two campaigns BISTUM23 and BISTUM24, two rural sites in German low-mountain ranges were selected. BISTUM23 took place near the city of Albstadt in the Swabian Alb (48°15^′ N, 8°59^′ E) with the ground station at 886 m above mean sea level (a.s.l.). BISTUM24 was performed near the village of Spielberg in the Vogelsberg area with the ground station at 391 ma.s.l. (50°19^′ N, 9°15^′ E). Both sites were chosen because they are in rural areas, which lowers the risk of contamination from local anthropogenic sources. The ground stations were located at the top of a hill, close to the flank of the mountain range where the air mass inflow usually occurs (Fig. S1 in the Supplement). Therefore, the topographical conditions favor orographic lifting of air masses, which can facilitate deep convective events (Barros and Lettenmaier, 1994; Liu and Kirshbaum, 2025).

Measurements were performed on the same three platforms for each campaign and are listed in Table A1 in the Appendix. Continuous ground-based measurements on-board the Mobile Laboratory (MoLa, Drewnick et al., 2012) include the O/C ratio of the organic fraction of the submicron aerosol particles, measured with an aerosol mass spectrometer, particle size distribution (merged data of a fast mobility particle sizer and an optical particle counter), particle number concentration (PNC), O3 mixing ratio, and wind data as well as temperature, humidity, and pressure. The MoLa inlet was 6 m above ground level (a.g.l.). A ceilometer monitored the aerosol backscatter signal above the site up to 10 kma.g.l. Additionally, during BISTUM24, 3D wind measurements at 5 ma.g.l. provided sensible heat flux (QH) and turbulent kinetic energy (TKE) data at 30 min-averaging intervals, as recommended for fair weather and pre-storm weather (Markowski et al., 2019).

Drone-based measurements were performed with the Flying Laboratory research drone (FLab, Moormann et al., 2025), which provides a wide particle and gas phase dataset including O3, PNC, and meteorological data. This allows the estimation of these variables' gradients in the lowest 500 ma.g.l. as well as the calculation of the bulk Richardson number Ri, the potential temperature θ and the equivalent potential temperature θeq in Eqs. (1)–(3) calculated from the acceleration of gravity g, the flight height above ground level ha.g.l., the virtual potential temperature θv, and the horizontal wind speeds u and v at ground at the respective air level: 1Ri=gθv,air-θv,groundha.g.l.θv,airuair-uground2+vair-vground2 θ and θeq were calculated/approximated with the air temperature T in K, the ambient pressure p, the reference pressure p0=1000 mbar, the dry adiabatic constant a=R/cp from the ideal gas const. R and the heat capacity cp of dry air at a constant pressure. Evaporation was considered in θeq with the latent heat at 20 °C Lv(20°C)=2500 kJkg-1 the specific heat for dry air Cp=1004 Jkg-1 and the total water mixing ratio rH2O (Stull, 1988): 2θ=T×p0pa3θeq≈T+LvCp×rH2O×p0pa Data from radiosondes, launched from the same site (Valero et al., 2025), help to verify the model data analysis of the meteorological conditions on a larger scale and provide the convective available potential energy (CAPE) as an indicator of the vertical uplift forcing before and after the weather events as well as the elevated boundary layer heights (Stull, 1988).

Large-scale synoptical information such as 24 h-backward trajectories, the mixing layer height (MLH) and radar information for the respective measurement site were derived as described in the following from data sources listed in Table A2. The trajectories in Fig. S2 were calculated for 24 h-backwards with the analysis tool to detect source regions outside Germany and were started from 0, 120 and 500 ma.g.l. (Stein et al., 2015). For trajectories, which are based on coarsely resolved 0.25°-GFS analysis data and a low-resolution topography, complex terrain poses a substantial challenge. Therefore, ICON-D2 analysis and hourly forecast data were utilized for the gaps between the analysis time events (3 hourly) to calculate corresponding MLHs and backward trajectories within the German domain and to validate the trajectories (Figs. S3 and S4 and in Sect. 3.3.2). ICON-D2 is the operational weather forecasting model of the German Weather Service (DWD). ICON-D2 is a non-hydrostatic model, with parameterisations for shallow convection and a sophisticated boundary layer and surface exchange scheme. Even though ICON-D2 itself operates on a triangular grid, the data has been re-gridded to a regular longitude–latitude grid with a similar grid width to ICON-D2. A detailed description of the ICON model has been provided by Zängl et al. (2015); Crueger et al. (2018) and further descriptions are available at DWD (2025a).

ICON-D2 uses terrain-following Gal-Chen coordinates, according to Klemp (2011), with 60 vertical levels in total. The lowest kilometer above ground for the Albstadt site is described with 16 vertical levels, whereas for Spielberg 15 levels cover the lowest kilometer with corresponding layer thicknesses between 30 m close to the surface and 100 m in 1 km altitude above ground. Turbulence and surface exchange processes are parameterised with a second-order scheme following Raschendorfer (2001).

To obtain a higher temporal resolution than the 3 hourly analysis data, forecasts for the respective two hours in between the analysis time spots have been merged with the analysis data set, i.e., forecasts with up to 2 h lead time. The forecast data was obtained from the PAMORE data archive (DWD, 2025c) for the duration of the campaign.

The backward trajectories in Figs. S3 and S4 have been calculated using a tailored trajectory program, which regrids the model data to a regular latitude–longitude–altitude grid, including the respective grid elevation of the orography. The trajectories themselves were calculated based on a 1 min-timestep, and were finalised when the trajectory left the domain of ICON-D2 or after 24 h.

The MLH is selected as the lowest altitude where all the following criteria in the ICON-D2 data are fulfilled: Brunt–Väisälä frequency >5×10-5 s-1, vertical gradient of potential virtual temperature >0.3 Kkm-1, and vertical gradient of absolute humidity must be curved (Wang and Wang, 2014) with threshold values empirically determined by the analysis of the simulated profiles in conjunction with the local observations and Hyun et al. (2005).

Additional synoptic scale information, displayed in Figs. S5 and S6 (surface pressure, geopotential in 500 hPa), for the selected events has been obtained from ERA5 reanalysis data (Hersbach et al., 2020) as well as from radar data (WN data set) from the radar network of the DWD (e.g., Kreklow et al., 2019). The radar and trajectory data were obtained from the DWD data server (DWD, 2025b) during the campaigns. The respective satellite images are obtained from EUMETSAT from the MSG SEVIRI instrument (Schmetz et al., 2002). Frontal lines in Figs. S5 and S6 were manually added to highlight areas of strong baroclinicity following the DWD analysis charts and maps of the equivalent potential temperature from ICON simulation data.

The synoptic map in Fig. S5 shows that the measurement site was centrally located in a large high-pressure system that covered substantial parts of Western and Central Europe with an indicated warm front crossing the measurement site. Two rain events with different intensities can be attributed to the warm front: first rain: ∼20 min duration, 0.5 mm accumulated rainfall starting at 11:00 (all times are given in local time, i.e., UTC+2); second rain: ∼10 min duration, 0.1 mm precipitation starting at 15:20. FLab measured three vertical profiles before and five after the major rain event (Fig. 2a), allowing investigation of hour-scale influences of the rain event on the atmosphere in the lowermost 500 ma.g.l. The MLH reflects stratification in 300 ma.g.l. in the morning, so that FLab measures inside and above the mixing layer before and directly after the first rain event. After the second rain event, no profiling flights were possible due to temporary airspace restrictions.

At 500 hPa a stable ridge was located above Central Europe, leading to large-scale subsidence. Radio soundings before and after the rain event show no change in tropopause height or CAPE (Fig. S7), suggesting no significant change due to the front. The high-pressure system appears to be stable, and the backward trajectories show no uplift as would be expected for warm fronts, i.e., the uplift at the warm front is compensated or suppressed by the large-scale subsidence. However, from the ceilometer data a cloud uplift from 3 to 5 km (or a trailing higher midlevel cloud) can be deduced immediately after the first rain (Fig. 2b). A strong upper free tropospheric layer (air mass I) that suppresses a trajectory uplift is shown by the local scale observation of an immediate but short-lived dynamical stabilization above the lowermost layer and its hardly changed conditions below. Despite the missing updraft transport, the rain events can indeed be described as a warm front (see Sect. 3.2 and 3.3.1). A follow-up rain event like the one at 15:20 is typical for warm fronts.

Stratification, such as the NBL that forms near the ground during the night, usually dissipates by midday due to convective forcing. The necessary energy for the buoyancy is provided by irradiance as the driving heat source. Typically, irradiance heats the ground, which conserves energy during cloud cover or darkness, and creates turbulences in the air that releases energy by dissipating eddies (Stull, 1988).

In Fig. 2a, the TKE increases in the morning, without being reduced prior to and during the first rain event, in contrast to the irradiance and sensible heat flux (QH). The strong correlation between irradiance and sensible heat flux throughout the day (Pearson correlation coefficient r=0.87) suggests that sensible heat flux is generated mainly by direct irradiance and is probably not driven by thermal energy stored in the ground. During the 11:00 rain event, irradiance and hence the sensible heat flux reach a minimum, while TKE is available independently – most likely due to advected air masses or turbulence generated by the precipitation and its evaporation itself. The existing TKE and the small amount of sensible heat flux are not sufficient to elevate the lowermost layer prior to rainfall. During the first two hours after rainfall, the thermal energy in the young CBL must still accumulate before mixing with the free tropospheric air can occur between 13:00 and 14:00, as the model predicts. To provide evidence for this suggested development and a more holistic understanding, the air masses in the lowest 500 m need to be characterized with in-situ data.

The air masses are characterized in reference to the 11:00 rain event primarily with FLab measurement data, which provide information on air mass stability and history. To reduce statistical and measurement uncertainty, the presented FLab-related data have been binned in 100 m increments for the altitude ranges from 0 to 500 m (Figs. 3 and 4). Note that strong gradients which are often measured within a few tens of meters above the ground are averaged out in the 0 to 100 m bins. To account for concentration changes over time, the O3 and PNC data measured on-board FLab were corrected for temporal trends using the analogous continuous ground-based data measured by MoLa. The identification of different air masses (I, IIa, and IIb) and the underlying processes discussed in the following sections lead to a schematic description of the lowermost troposphere after rainfall at the end of this section.

During 12 August 2023, four precipitation events were observed at ground level at the Albstadt site, while the ceilometer recorded seven, including three that did not reach the ground due to evaporation of rain droplets at higher altitudes (Fig. 7a and b). Even after the last rain event with a strong thunderstorm at 15:00, clouds were still present and the irradiance did not reach the levels of cloudless days, although the conditions became mild and the temperature rose by up to 10 °C.

For this day, synoptic maps show a convergence line and a cold front crossing the site from 14:40 to 15:20 (Fig. S6). A severe thunderstorm may have been caused by the combination of substantial lability and CAPE in addition to orographic lifting of warm air in conjunction with a convergent flow between 14:30 and 15:00. Radiosondes launched from the site recorded CAPE of 1400 Jkg-1 at 14:00, while it decreased to less than 10 % of this value, i.e., 130 Jkg-1, during mild, sunny conditions after the thunderstorm (radiosonde at 16:40, Fig. S9). 24 h-backward trajectories indicate inflow of air masses, amongst other across the mountain ranges of the Black Forest and the Swabian Alb, during the 100 km long track before they reached the measurement site (Fig. S2).

In this section, we use a similar approach as that presented in Sect. 3 to develop an overview schematic that describes the details of the boundary layer processes around the investigated events. The schematic at the end of this section contains air masses characterized by dynamical and chemical properties, which are tagged with roman letters (I, IIa, IIb, III, and IV) and should not be confused with the air masses mentioned in the previous Sect. 3.

The meteorological situation of the boundary layer on 12 August 2023 can be separated into a dynamically unstable pre-thunderstorm period and a stable post-thunderstorm period. In contrast to the situation described in Sect. 3, no ground-based flux measurement data are available. A consistent Ri=0 at the 50 m mark during all times when the Flab was operating indicates unstable conditions, i.e., small-scale turbulence at the ground, which are not significantly influenced by processes above 100 m (Fig. 8a). Before the first rain event at 08:30, a layer boundary at the 250 m mark was identified by sign changes of the weak gradients for Ri, potential temperature, absolute humidity, O3, and PNC (Fig. S10 or orange dots in Fig. 8). While positive gradients of potential temperature and Ri indicate increasing dynamic stability with altitude, the upper air mass above 200 m is classified as laminar with Ri=0.6 with enhanced moisture and O3 levels (O3 increased by 6 ppbv), and reduced particulate pollution (PNC decreased by >1000 particles cm-3). Here, free tropospheric air (I) overlays a weak NBL remnant above 200 m (IIa), similar to the case in Sect. 3.

After the first rainfall event (0.5 mm in 15 min), a remnant of the NBL still remains up to 200 m in agreement with the MLH predicted by ICON. Associated with the rain event, an increase of Ri and dθeq/dz indicates increased stability of air mass I above the NBL remnant. Evaporating precipitation, i.e., latent heat flux can act as an energy sink for sensible heat, which would otherwise enhance convective turbulence. Before 10:00, it rises to a height of 1100 m, where the ceilometer detects the upper boundary of an aerosol layer (Fig. 7b).

After a light second rainfall of 0.1 mm at 10:00, no RL and uniform gradients are observed over the whole 500 m-altitude range for all FLab-measured variables until 13:00. In air mass IIb mixing has eliminated the gradients. Since the first two rain events at 08:30 and 10:00, potential temperature near ground has increased by only 2–3 K, suggesting that diurnal heating primarily contributes to latent heat rather than driving vertical mixing (Fig. 8c). This observation is in agreement with very slow evaporation from moist soil that might explain the unusually stable equivalent potential temperature at the 50 m mark between 10:00 and 13:00 (Fig. 8b), while the absolute difference to the potential temperature decreases (Fig. 8c). A similar pattern is observed for the last rain event, although here evaporation is accelerated by strong irradiance and turbulence-driven mixing.

At 13:30, a cold front delivered 2.4 mm of rain in 10 min leading to a 2.2 °C decrease in air temperature and 2.1 gkg-1 increase in humidity at ground, while equivalent potential temperature remains constant at ground as measured with MoLa (Figs. 8 and 9). However, at 50 m altitude and above no change in temperature, humidity, and dθeq/dz is observed. The slow evaporation of moisture from previous rainfall forms a localized moist patch, which, on a regional scale, might develop into a cold pool. Cold pools can create a positive feedback loop: they generate broader clouds that are less affected by entrainment, leading to increased precipitation, larger moist patches, and further expansion of cold pools – ultimately fostering larger clouds with enhanced convective mass fluxes (Schlemmer and Hohenegger, 2014). Simulations of cold-pool dynamics indicate that, under continental conditions in the mid latitude, strong latent heat flux can increase convective mass flux (Lochbihler et al., 2021).

In our case, following the rain events at 13:30 and 14:45, we observe a cold pool from ground and FLab measurements, identified based on four indicators: (i) stratification between the lowermost 100 m (cold pool) and aloft is visible for the dynamical markers dθ/dz and Ri (Fig. 8a and c) as well as for the O3 mixing ratio (ii), which is ∼(5.3±3.4) pptv (12 %) lower in the lowermost layer (Fig. 8e, presumably due to depletion in the wet cold pool); (iii) the increased negative gradient and variance of the absolute humidity (Fig. 8d) and (iv) an increase of the equivalent potential temperature by ∼6 K at ground and by 2.1 K for the whole 500 m range (Fig. 8b) show the evaporation and conditional enhanced latent heat flux typical for cold pools (Tompkins, 2001; Engerer et al., 2008).

These highly-resolved vertical observations between 13:30 and 14:45 support the model-based hypothesis by Lochbihler et al. (2021) and Haerter and Schlemmer (2018) that strong latent heat flux and associated evaporation can promote the formation of new convective cells like the upcoming rain front in our case study.

Due to increased latent heat flux, thermal convection is suppressed initially, leading to a homogeneously laminar air mass III with Ri=0.28 above 100 m, while air in the lowest 100 m is dynamically stable due to surface roughness. The downdraft of air mass III is likely enhanced by the cold pool, as penetrative downdrafts from the mid-troposphere commonly occur after rainfall exceeding 2 mmh-1 (Barnes and Garstang, 1982), in agreement with model results under convergence zone conditions (Schlemmer and Hohenegger, 2014).

At 14:30 the last rain event of the day with 5.4 mm precipitation in 40 min was associated with a severe thunderstorm under a convergence line as described in Sect. 4.1. Despite the reduced CAPE after the storm (>90 % decrease), the new air mass IV remains unstable. Unlike the unstable air mass III, which might be unstable due to deep-convective vertical forcing (CAPE), this air mass IV is likely unstable due to small-scale turbulence indicated by a suddenly negative Ri above 100 m altitude and a remaining negative dθeq/dz from 15:30 on (Fig. 8a and b). Additionally, the convection did not take place in a frontal system, thus the existing air mass was not replaced by a colder, more stable air mass. Instead, only a part of the CAPE was consumed by the convective event which was initiated and substantially driven by the convergence. After the rain event, the post-convective subsidence led to cloud free conditions and thus growing instability caused by irradiance and enhanced ground temperatures (Fig. 9). Higher temperatures and thermally driven convection at the ground allow for drying of the moist near-ground layer, and for recovery of the lowermost troposphere from the thunderstorm as thermal mixing removes the vertical gradients of potential temperature, absolute humidity, PNC and O3 after the last rain event (Fig. 8c–f).

Based on different characteristics of the air masses I–IV, we were able to create an overview scheme (Fig. 10) that illustrates the temporal exchange of air masses and the accompanying processes to enhance the understanding of the complex meteorological situation for the following sections.

During the day, four rain events with different amounts of precipitation occurred, leading to different changes in the trace matter distribution. Contrary to the morning precipitation events, when there was no lifting of air masses before they reached the measurement site; for the rainfall events at 13:30 and 14:30 the 24 h-backward trajectories show a lifting up to above 2000 ma.s.l. and a sudden downdraft, 7 and 1 h before arrival of the air masses at the measurement site, respectively (Fig. S4). This downdraft dynamic was also observed by the ceilometer, which measured cloud layers subsiding from 2500 m down to 700 ma.g.l. prior to the thunderstorm (Fig. 7b). The downward motion is likely a sign of an onset of convergence leading to more low level and convective clouds and may be enhanced by the cold pool at the ground. This downdraft results in the injection of clean, O3-rich air with low particle load into the lowermost troposphere and represents the transition to air mass III (Figs. 8e, f and 10).

After each rain event, O3 mixing ratios are increased at altitudes above 200 m (due to the air mass downdraft from higher altitudes, Fig. 8e), but decreased near the ground by 3.0±1.3 ppbv, excluding the 10:00 rainfall, as shown in Fig. 11. As a consequence, O3 gradients are enhanced by 7 ppbvkm-1 for each rain event and subsequently disappear within a few hours due to convective mixing within the PBL. Reduced ground-level O3 concentrations following rainfall may result from deposition or enhanced O3 depletion, whereas outwash seems unlikely due to the poor solubility of O3 in water, except for potential aqueous phase SO2 oxidation (Hoyle et al., 2016). Depletion could be driven by the increased release of primary biogenic volatile organic compounds (BVOCs) from vegetation (Rossabi et al., 2018; Miyama et al., 2020; Machado et al., 2024a), by peroxide-scavenging after rainfall (Bela et al., 2018) or reaction with soil-emitted NO after rain (Andersen et al., 2024). BVOCs have been identified as an O3 sink in natural and anthropogenic conditions (Fitzky et al., 2019; Machado et al., 2024a). However, this emission-driven O3 removal mechanism cannot be verified due to the lack of BVOC and NO data.

Figure 8f highlights that for the same rain events, when O3-rich air was injected into the lowest 500 m, the CPC PNC was consistently reduced by 25 % (except when the NBL was still present, because air mass I is already in the free troposphere). The PNC gradient remained unchanged during and after the regeneration period, and the PM1 levels measured at ground remained constant immediately after rain (Fig. 9), indicating that washout was minimal under these conditions. This is likely due to the size of the measured aerosol particles that are in the Greenfield gap and consequentially inefficiently removed by outwash (Cherrier et al., 2017).

These observations show that rain itself does not necessarily significantly influence the distribution of pollutants such as O3, PM1, and PNC. However, post-rain air masses determine the composition and gradients at higher levels, while rain-induced emissions from the ground may act as a sink for reactive substances as O3.

In Sect. 3 we showed that the determination of the MLH using 1 h forecast data from the ICON model is feasible with high accuracy under stable conditions during a weak warm front (Fig. 5). The MLH estimation criteria in the model include the Brunt–Väisälä frequency, humidity, and potential virtual temperature gradients. Until this point of the study the MLH is used as a synonym for the height of the planetary boundary layer (PBLH), which is considered the most relevant measure separating the free troposphere and the ground-influenced layer with different dynamical characteristics and compositions (Stull, 1988; Tignat-Perrier et al., 2020; Kotthaus et al., 2023). Note that other kinds of stratification occur regularly in the troposphere (even within the planetary boundary layer) and are often not predictable, due to local topography, emissions, and heat reservoirs.

For the convergence line/cold front case analyzed in this section, the modelled MLH, shown in Figs. 7a and 10, increases from 0 m at 08:30 to 200 m at 10:00 and 300 m at 11:00. This observation is consistent with the NBL identified from in-situ FLab data and demonstrates additionally to Fig. 5 that the MLH algorithm predicts the height of the NBL (in this case the PBLH) accurately. However, between 12:00 and 18:00, the modelled MLH drops below 500 m before returning to levels above 600 m at 18:00, where it continues following its diurnal cycle. The average predicted diurnal cycle of the MLH at the measurement site from Sect. 4 shows maxima of the MLH between 800 to 1000 m at 14:00 to 15:00 on clear-sky days, while being <20 m during night (Fig. 7a). The difference between the modelled MLH in the cold front case, compared to the average diurnal cycle of the PBLH, may be explained by reduced irradiance (average of 100 Wm-2 during this period).

The formation of a cold pool at ground leads to an increase of latent heat flux and consequentially to less turbulent mixing. Therefore, the cold pool acts as an energy sink and suppresses energy supply for further turbulent mixing. Although convection has driven the PBLH up to the free troposphere before the rain, the cold pool contributes to layering of the convective-driven RL above a shallow turbulent mixing layer. This stratification disappears as the cold pool dissolves. These dynamic processes drastically alter the vertical gradients, and may have influenced the MLH model output, which relies solely on vertical gradients. The ICON 1 h forecast models that the MLH exceeds 500 m after 11:45 and thus differs strongly from our in-situ observations. After irradiance decreased, MLH criteria were met even below 400 m due to rapidly changing gradients caused by vertical dynamics within the convergence zone. ICON data predict a MLH of 600 m at 18:00 that can be assigned as a realistic PBLH 2.5 h after the last rain, in agreement with the mean life time of a cold pool (Tompkins, 2001). Interpolating the MLH between 12:00 and 18:00 may lead to an inaccurate estimate, because the PBLHs maximum is expected in between. In agreement, radio soundings at 14:10 and 16:40 indicate a boundary layer up to slightly above 800 hPa, i.e., 1400 and 1300 m based on changing lapse rates and temperature gradients (Fig. S9, Seidel et al., 2010). The PBLH derived from the radio soundings is 100 to 200 m higher than the estimated average MLH for the measurement site (Fig. 7a), indicating that the PBLH has lifted due to convective forcing by the thunderstorm. A more robust parameterization of the MLH estimation for the PBLH for cold pool scenarios may consider two MLHs: a lower MLH separating an additional shallow cold pool-driven layer at ground from the previously well-mixed PBL and a second MLH separating the PBL from the free troposphere in the altitude range close to the PBL before rain events. Alternatively, the upper boundary of the PBL is described by a kind of “residual layer” where the cold pool leads to a different PBL structure below.

Due to unsafe flight conditions, vertical profiling during the thunderstorm was not possible; instead, ground-based MoLa data were used to analyze meteorological processes during this time (Fig. 9). Between 14:30 and 15:30, a strong cloud layer largely suppressed irradiance and minimized thermal convection, leading to dynamically deep convective inflow-driven air masses with surface wind speeds peaking at 8 ms-1. As the rainfall starts, strong downward movement of air is implied by a descending cloud layer height, a high wind speed period, and a 40 % reduction in PM1 due to mixing in of unpolluted free tropospheric air (Figs. 9 and 10). Although it is known for cold pool-induced deep convective events that the horizontal wind speed is maximal before rainfall starts (Tompkins, 2001), a strong downdraft from the start of rainfall until the maximum of the precipitation rate was not described, yet, probably because observational studies describe longer-lasting rainfalls only (Young et al., 1995).

As the rain band advanced, heavy rain ceased and wind direction turned by 180°, an effect attributed to a moving convergence zone (Crook and Klemp, 2000), which changed the air track from passing a rural forest (Western) to the town of Albstadt (Eastern, Fig. S1a). This resulted in increased NOx and PNC with lower aerosol oxidation levels, indicated by the reduced particulate organics O/C ratio (Fig. 9). After rainfall, near-surface O3 was reduced (possibly due to depletion by enhanced anthropogenic or natural emissions as NOx or BVOCs, see Sect. 4.3), though it slowly recovered as the wind direction turned back from East to West after 40 min. When irradiance exceeded 400 Wm-2 at 15:35, photochemical activity led to enhanced PM1 and PNC; simultaneously, temperature and absolute humidity increased as the grassland dried.

These rapid shifts in air mass history and wind dynamics behind the convergence line highlight the need of in-situ 3D turbulence observations, e.g., with wind LIDAR (Bélair et al., 2025) to fully capture local upwelling and advection processes within the PBL (Borque et al., 2020); measurements at a less polluted site (independent of wind direction) would be helpful to identify compositional changes due to the thunderstorm itself or thunderstorm-related effects like BVOC emissions from the ground.