Here we use the two-moment bulk microphysics scheme developed by . It distinguishes between the six hydrometeor classes: cloud droplets, rain, ice crystals, snow, hail, and graupel, and employs prognostic equations to estimate the mass densities and number concentrations of these hydrometeor particles. Therefore, it can provide more accurate predictions of cloud formation and precipitation than the simpler single-moment scheme employed in the standard COSMO setup . In the standard setup of the two-moment scheme, the number of activated cloud droplets and ice particles is calculated using prescribed CCN and INP values, respectively, and saturation adjustment is applied. We enhance the model configuration for this study by replacing these prescribed concentrations with calculated activation of cloud droplets and ice particles directly from aerosol mass concentrations simulated by MUSCAT. This dynamic approach allows the model to account for spatial and temporal variability in aerosol properties, which is expected to enable a more realistic representation of CCN and INP with the goal of improving the simulation of aerosol–cloud interactions.

We build upon the work of , who included the effects of dust, soot, and organic carbon on ice particle formation and cloud droplet activation. Cloud droplet activation was computed using the parameterization for multiple aerosol types developed by , which accounts for multiple soluble and insoluble aerosol species, representing a multimodal aerosol size distribution. It uses the hygroscopicity parameter κ, which describes the relationship between particle hygroscopicity, dry diameter, and CCN activation . Experimentally determined values of κ range from greater than 1 for highly hygroscopic particles down to 0 for hydrophobic particles.

In this study, the treatment of ice particle activation from is retained, while the cloud droplet activation scheme is extended to include additional aerosol species, such as ammonium sulfate, ammonium nitrate, sulfate, organic carbon, elemental carbon, and two sea salt size classes. The corresponding κ values used in this study are listed in Table . Those species were chosen, as they are the most important natural and anthropogenic sources of cloud condensation nuclei . For each species, the aerosol number size distribution is calculated from the simulated mass concentrations, using fixed values for radius and standard deviation, as in . These size distributions are then used to calculate the number of activated particles using the parametrization, which links particle size distribution and composition to the number of particles activated at a given supersaturation. The maximum supersaturation is determined by the competition among aerosols for water vapor and depends on their composition, size distribution, and the updraft velocity. To enable in-cloud activations, already activated cloud droplets are treated as an additional aerosol mode with κ≈0 and a diameter equal to the mean droplet size. This allows the activation scheme to distinguish between activated droplets and activatable aerosol at each model time step, enabling secondary nucleation in updrafts throughout the cloud depth.

Two cases with different meteorological characteristics, including flow patterns and boundary layer conditions, were chosen. The cases were systematically selected through evaluation of public reports and the RADKLIM CatRaRe event catalog and whether the cases were captured by COSMO-D2 reanalysis. Case I features a small-scale convective event approaching the region of interest with northerly winds in the evening of 13 July 2019, while case II shows a larger event with westerly flow into the region in the evening of 20 July 2019. Despite these differences, both produced considerable precipitation across the region.

To simulate both cases, we employ a 3-step one-way nesting strategy with domains: D0, D1, D2. The outer domain, D0, covers Europe with a grid resolution of 14 km (Fig. ). There are 40 vertical layers with the highest layer at 22 km altitude. In this domain, COSMO-MUSCAT is run with the single-moment bulk water continuity scheme and without DCEP. The timestep is 80 s and the simulation starts 15 d prior to the actual event. This spin-up time is necessary to allow the model chemistry to develop. To keep the model meteorology close to the real synoptic situation, the meteorological fields are re-initialized for the next simulation cycle every 48 h. The COSMO simulations are initialized and driven by ERA5 meteorological reanalysis data . The first MUSCAT cycle is initialized with an atmospheric chemical composition reanalysis product of CAMS (Copernicus Atmosphere Monitoring Service) , while subsequent cycles use aerosol and trace gas fields from the previous cycle to maintain temporal consistency in chemical composition. Natural primary aerosols, such as dust and sea salt, are computed within the MUSCAT model e.g. using meteorological fields from COSMO, combined with surface property data. Emission of anthropogenic aerosols and trace gases are taken from the CAMS-REG inventory , while the temporal profiles of the main air pollutants and greenhouse gases are based on the CAMS-REG-TEMPO dataset .

The next inner domain D1 covers Germany and is configured with a horizontal grid resolution of 2.3 km (Fig. ). The vertical resolution includes 50 levels with the highest level at 22 km. As meteorological input, the COSMO-D2 reanalysis provided by the German Weather Service (DWD) is used. This high-resolution input is essential for resolving the small convective cells that are the focus of this study. The simulation starts on 12 July 2019 at 00:00 UTC for case I and at 19 July 2019 00:00 UTC for case II, each simulation runs continuously for 48 h without a restart and has a timestep of 20 s. The first 24 h serve as a meteorological spin-up, while the subsequent 24 h comprise the coupled chemistry-meteorology simulation. In this domain, the model is run with the two-moment bulk microphysics scheme. The chemical fields simulated from the coarser European domain are interpolated onto the D1 grid as initial and lateral boundary conditions. Point source, detailed area, and traffic emissions that are used build on the GRETA dataset of the German Federal Environmental Agency (Umweltbundesamt, UBA) .

The innermost domain D2 focuses on the greater metropolitan area of Leipzig, with a horizontal grid resolution of 1 km. The chemistry and meteorology for D2 are driven by the outputs from the D1 simulations, maintaining the same simulation time and vertical layers for both cases. The model in D2 operates with the two-moment bulk microphysics scheme, along with the urban parametrization scheme DCEP activated for the city of Leipzig. The simulation uses a timestep of 10 s, with output saved at 10 min intervals. The domain is characterized by predominantly flat topography, bounded by the Erzgebirge along its southern edge. The highest peak of this mountain range rises to an altitude of about 1.2 km. The city of Leipzig lies at the center of the domain.

To capture the movement of air volumes influenced by urban emissions, we used the Lagrangian analysis tool LAGRANTO . We calculated backward trajectories starting from a region centered on the area of highest precipitation downwind of Leipzig for each ensemble member, using meteorological fields from the D2 output. For case I, starting at 13 July 2019, 23:00 UTC, trajectories were traced backward for six hours. The starting region of trajectories horizontally covers approximately 40×40km2 with about 5 km spacing between points. Vertically, the range extends from 1–4.5 km above ground level and is divided into 20 vertical levels. For case II, starting at 20 July 2019, 22:00 UTC, the start points were similarly traced back for six hours. The horizontal box spans roughly 40×60km2 with 5 km spacing, and vertically, it extends from 1–7.5 km subdivided into 50 levels. These spatial and vertical configurations ensure that the trajectories originate from sufficiently large and well-resolved volumes, capturing the key atmospheric dynamics of each precipitation event. To make sure that we only follow the Leipzig emissions, a filter region was defined as a three-dimensional box in longitude, latitude, and height. By retaining only trajectories that passed through the urban boundary layer over Leipzig, this approach facilitates the attribution of aerosol–cloud interactions (see Fig. S1 in the Supplement).

For analyzing cloud dynamics and microphysics along the trajectories, a moving box analysis was implemented. At each time step, a rectangular box is centered on the mean urban trajectory point of each ensemble member, and vertical profiles are extracted by horizontally averaging all grid point values within the box, including grid points with zero values (unconditional average). Each box is defined in the COSMO rotated geographical coordinate system as a rectangular area with the mean trajectory point at its center, extending symmetrically in both horizontal directions (east–west and north–south) and including the full vertical column of model layers. The grid point values extracted within each box define the analysis volume at that time step. Although the trajectories are calculated backward in time, the moving box analysis follows the trajectory points in forward temporal order, tracing the evolution of the analysis volume from the urban emission region toward the precipitation area. As the trajectory point advances in time, the box moves accordingly through the model domain. This approach differs from conventional methods that either analyze a single convective system or average over the entire model domain: rather than tracking an individual convective system, the moving box isolates the microphysical state of convection within the aerosol-influenced region at each time step. The box may therefore contain portions of one or more convective cells at any given time, and the resulting profiles reflect the aggregate microphysical state of convection along the plume trajectory. We selected different box sizes depending on the spatial scale of the respective precipitation event. For case I, a box size of approximately 15×15km2 was chosen to fully enclose the smaller, localized precipitation structure. For case II, a larger 35×35km2 box was used to capture the broader precipitation area associated with the more extensive event. This selection ensured that the dominant precipitation features were included in the spatial averaging for each case. The associated backward trajectories are shown in Fig. a (case I) and Fig. b (case II). At each mean trajectory point and time step, we averaged the number of activated cloud droplets, all hydrometeor species, and vertical wind components horizontally for all grid points within each box. This produced one vertical profile per ensemble member every 10 min. These profiles were then aggregated across all ensemble members to derive the ensemble mean, minimum, and maximum values.

The same analysis was repeated for the overall mean trajectory (Appendix A). While the analysis box along the filtered mean trajectories captures the effects of urban aerosols on cloud dynamics and microphysics, the analysis box along the overall mean trajectory (Appendix A) represents convective cloud properties from grid cells irrespective of their interaction with urban aerosols. This comparison provides evidence that only air volumes interacting with urban aerosols exhibit systematic changes.

We are aware that five ensemble members are likely insufficient to capture the full range of variability, found that an ensemble of ten members is required to sufficiently represent internal variability in aerosol-convection interaction studies. Increasing the ensemble size beyond five members was, however, not computationally feasible within this project. To partially compensate for the limited ensemble size, the grid cells within the analysis box are considered to represent the variability in cloud processes. So, the bootstrapping significance test is applied as follows. Rather than comparing only the five ensemble-mean values per experiment, all grid point values within the moving analysis box (Sect. ) at a given height level and time step were pooled across all five ensemble members. This pooling approach increases the effective sample size without requiring additional simulations and better represents the internal spatial variability within the analysis box. For each combination of height level and time step, 1000 bootstrap resamples were drawn with replacement from each pooled distribution, and the difference of the resample means was computed. A difference was considered statistically significant at the 90 % confidence level if the resulting bootstrap confidence interval did not include zero. The same approach was applied to the precipitation fields, where grid point values within a 0.1°×0.1° neighborhood centered on each pixel were pooled across all ensemble members prior to bootstrapping.

In the first case, atmospheric conditions were dominated by a high-pressure system shifting from the Norwegian Sea toward the British Isles and a broad low-pressure system over Eastern Europe (Fig. S2a–c). This synoptic setup established a northerly flow across Germany, advecting cold, moist air masses (Fig. S3a) into the region of interest. Several small-scale convective events developed over northern and eastern Germany (Fig. S4a). Northerly winds transported large amounts of sulfate into the model domain, reflecting long-range transport from coal combustion and industrial sources in Eastern Europe, and leading to elevated background concentrations between 2 and 6 km (Fig. S5a). The analysis box captures a single convective cell that dominates the precipitation field throughout the analyzed period. The following analysis therefore reflects the evolution of this single cell. The convective cell formed east of Leipzig and propagated southwestward, passing over the city of Chemnitz. Low-level winds transported urban emissions from Leipzig toward the southeast, mixing into the convective cell at higher altitudes. The total accumulated precipitation associated with this convective event exceeded 9 mm over the course of a 4 h period (Fig. a). The convective cell began to intensify north of Chemnitz around 20:00 UTC and propagated southward at approximately 10–20 kmh-1 (Fig. ). Maximum precipitation occurred between 21:00 and 21:30 UTC, when the precipitation core (approximately 5 km in diameter) reached intensities of 6–7 mm in 30 min, representing a typical heavy rainfall rate for convective summer events. The core was surrounded by moderate precipitation of 2–4 mm(30min)-1. The convective cell then continued to propagate southward, gradually dissipating.

The analysis reveals a complex and spatially variable pattern of precipitation differences between the BASE and the NONURBAN simulations (Fig. ). Some regions within the pattern showed statistically significant differences. Throughout, the precipitation differences remained consistently small at ±0.5-1mm(30min)-1. Between 20:00 and 20:30 UTC (Fig. b), a precipitation enhancement appeared on the western flank, coinciding with the intersection of the convective cell and the urban plume. This represents the first influence of urban emissions on the precipitation pattern. Simultaneously, the eastern flank exhibited reduced precipitation, indicating a localized atmospheric response to the urban aerosol plume. In the following (Fig. d), the differences at both the western and eastern flanks persisted but shifted further south as the cell propagated, demonstrating a continued evolution of the urban emission influence. Around 21:00 UTC (Fig. f), the difference shows a more mixed spatial pattern, suggesting complex interactions between the urban emissions and the convective event's internal dynamics. After 21:30 UTC (Fig. h), the difference pattern shows a shift to more negative values, indicating that the urban emissions increased rainfall intensity at this later period of the convective event. Additionally, the pattern, initially more positive and later more negative, suggests a southward shift of the precipitation core.

The influence of urban emissions on local precipitation rates ranges between 10 % and 20 %, with both increases and decreases depending on location, although not all differences are statistically significant. Significant changes mainly occur at the edges of the cell, where emissions can reach higher cloud layers, making their impact more pronounced in these regions. The precipitation core maintains its structure and intensity in both experiments, with variations only after 21:00 UTC. This spatial distribution highlights the complex and heterogeneous response of convective events to urban emissions.

In the second case, a shortwave trough moved northeastward over Benelux on the forward flank of a broad Atlantic upper-level low (Fig. S2d–f). Meanwhile, a cold front reached western Germany during the day. Flow from the southwest advected warm, moist subtropical air into the lower troposphere (Fig. S3b), while colder air aloft associated with the approaching trough increased lapse rates and conditional instability (Fig. S4b). Combined with enhanced vertical wind shear, these conditions favored the development of organized convective events, including a squall line that propagated northeastward across northern and central Germany. The wind field promoted background concentrations dominated by secondary organic aerosols and ammonium sulfate (Fig. S5b). The analysis box captures a portion of the squall line system, which may contain multiple convective cells at different levels of intensity. The following analysis reflects the aggregate microphysical and dynamical response within the analysis box rather than the evolution of a single cell.

The analyzed convective event exceeded 24 mm accumulated precipitation within a 4 h period, with a total horizontal extent of approximately 100–150 km and moved diagonally through the upper model domain (Fig. c). Peak intensity was reached between 20:00 and 20:30 UTC. The core of the event exhibited a multicellular structure with at least two pronounced local maxima (Fig. e). Rainfall rates within the core reached 15–20 mm(30min)-1, classifying the event as heavy precipitation by German Weather Service (DWD) standards. Toward the edges of the event, rainfall intensities decreased to 2.5–10 mm(30min)-1. Following its peak, the convective event continued to move northeastward and gradually dissipated.

Comparing the BASE and NONURBAN experiment reveals a temporal evolution of the urban influence of the event. Between 19:00 and 20:00 UTC, differences were relatively small below ±1mm(30min)-1 and randomly distributed (Fig. b and d). However, after 20:00 UTC when the event reached its maximum intensity, a negative precipitation difference between experiments is seen in Fig. f, indicating a precipitation increase in the presence of urban emissions. These differences concentrate mainly around the central precipitation core, while other areas of the event are only marginally affected. The differences range from 2–3 mm(30min)-1. Additionally, some heterogeneously distributed areas of positive differences ranging from 1–2 mm(30min)-1 appear at the edges of the convective event. As the convective event propagates, the pattern persists within the same range (Fig. h).

These results demonstrate that urban emissions can increase local precipitation intensity by 10 %–20 %. Mainly the precipitation core after 20:00 UTC is affected, showing a clear increase of precipitation by Leipzig's emissions. This demonstrates that urban modifications to cloud microphysics and precipitation processes require time to establish. Given the convective event's rapid propagation speed (∼50-70kmh-1), its residence time over the Leipzig area remains limited (∼1h). Nevertheless, this exposure proves sufficient for the convective event to obtain measurable urban modifications. These modifications continue to influence precipitation patterns as the convective event moves downstream, with increased precipitation effects observable 50–100 km downwind of the urban source region.

In this case, the mean trajectory originates approximately 30 km northwest of Leipzig and passes through the city diagonally between 17:50 and 18:40 UTC (Fig. a). It continues southeast for approximately 15–20 km following the low-level wind field. At this point, it encounters convective activity and is lifted from surface level to cloud height (2–4.5 km altitude), after which it moves straight south for another 80 km within the convective cell. As the trajectory crosses the urban emission plume, which is characterized by up to 3 % increased aerosol concentrations relative to the NONURBAN experiment (Fig. a), the air volume picks up an urban signature. This influence is limited to the section of the trajectory prior to convective uplift.

Cloud droplets first appear along the trajectory at 18:30 UTC (Fig. a). The number of activated cloud droplets (QNC) increases steadily to a first maximum of 45 cm-3 at 20:00 UTC, then continues rising to 55 cm-3 after 21:00 UTC when strongest updrafts enhance supersaturation conditions. Subsequently, QNC declines to 5 cm-3 as precipitation processes dominate particle removal. The NONURBAN experiment shows a slightly higher QNC peak, but differences remain within ensemble spread. The total amount of cloud condensate (QSUM) reaches a maximum of 0.2 gkg-1 (Fig. b). Detailed analysis reveals that QSUM (see Fig. S6) is dominated by graupel, cloud and rain droplets. Indicating that the dynamics in the cloud lead to riming and rapid droplet growth. No substantial differences exist between experiments for total condensate or individual hydrometeor categories. Precipitation begins after 19:00 UTC and increases until reaching a maximum precipitation rate of 0.6 mm(10min)-1 between 21:00 and 21:30 UTC (Fig. c). Both experiments achieve precipitation maxima at the same time, the BASE experiment shows slightly higher rain rates after 21:30 UTC, though the differences remain within ensemble variability.

Overall, we found no substantial differences in the vertically averaged quantities between the experiments. All analyzed variables show nearly identical temporal evolution, with only minor differences that fell within the natural ensemble variability. Therefore, we conducted a more detailed analysis of vertical variability to better understand the potential urban impact on convective processes that may not be captured by horizontally and vertically averaged fields.

The analysis of the number of activated cloud droplets reveals a distinct temporal evolution of cloud development. Cloud droplet activation initiates at 18:00 UTC and rapidly intensifies reaching concentrations of 150–250 cm-3 with a vertical extent of 1–4 km (Fig. a). Between 20:00–22:00 UTC the convective cell continuous to develop vertically, with QNC reaching its maximum vertical extent of up to 6 km. The hydrometeor distribution (Fig. c) shows a vertical extent of 6 km between 19:30 and 22:00 UTC. Peak QSUM exceeds 0.8 gkg-1 at 3 km height at 21:30 UTC, corresponding with highest rain intensities. After 21:30 UTC the cloud is mainly dominated by ice phase processes, with substantial formation of graupel and snow.

Comparing the BASE and NONURBAN experiment reveals two distinct periods of urban influence on cloud microphysics. During the first period between 18:30 and 20:30 UTC, the negative QNC difference values of 5–10 cm-3 are more prevalent (Fig. b), indicating that the urban aerosols act as additional condensation nuclei and the formation of cloud droplets is increased by 2 %–10 % compared to the background aerosol conditions (see Fig. S8). Alongside, QSUM shows a slight increase of 2 %–4 % during this first period.

In the second period, after 20:30 UTC, the pattern in QNC reverses. Positive differences dominate across all vertical layers, showing a 2 %–7 % reduction in QNC compared to the NONURBAN experiment. However, this period is characterized by statistically significant intensification of QSUM between 4 and 6 km altitude of 3 %–7 % (Fig. d). This is driven by enhanced ice microphysical processes including increased formation of graupel and snow within the cloud, while simultaneously reducing ice crystals and cloud droplets (see Fig. S8).

This two-stage pattern indicates a spatial redistribution of the cloud microphysics caused by the urban emissions. The additional droplets formed in the first phase are lifted to higher altitudes where temperatures fall below the freezing level, which is located at approximately 2.8 km. At these altitudes, the increased droplet population enhances riming and aggregation processes, leading to more efficient formation of graupel and snow particles. The freezing of additional droplets releases latent heat, which warms the surrounding air and strengthens the updrafts, creating a positive feedback loop between microphysics and dynamics.

The vertical redistribution of cloud condensate with increased concentrations above 4 km after 20:00 UTC and decreased concentrations below supports this interpretation. Urban emissions effectively enhance the ice phase development, promoting more efficient precipitation processes through intensified microphysical interactions. They also provide additional condensation nuclei between 18:30 and 20:30 UTC, which subsequently undergo vertical transport and freezing, ultimately promoting more efficient precipitation formation through intensified ice microphysical processes.

To further support this conclusion, we analyzed the domain-averaged vertical profiles of the positive and negative vertical velocities (Fig. ). Here we see an increase of the updraft region in the same height as the increased number of hydrometeors. Supporting the presumption that with urban emissions, the ice phase is enhanced, as more particles reach the freezing level, resulting in more freezing of droplets. Hence, more energy is released, which intensifies the updraft in the cloud, leading to more graupel and snow in the clouds. While Fig. d shows positive values of the downdraft region difference before 20:00 UTC, which temporally coincide with the increase of precipitation by urban emissions (Fig. d).

For case II, the mean trajectory originates 40–60 km southwest of Leipzig and progresses northeastward, passing over the city between 18:00 and 19:00 UTC (Fig. b). North of Leipzig, trajectories follow the urban emissions plume eastward for about 30 min before continuing northeastward. Between 20:00 and 21:00 UTC, convective lifting causes the trajectory to ascend vertically by approximately 2–4.5 km. Prior to convective uplift, the air volume along the trajectory acquires up to 2 % higher urban aerosol concentrations relative to the NONURBAN experiment.

The number of activated cloud droplets (QNC) increased between 19:00 and 20:00 UTC reaching a maximum of 47 cm-3, indicating strong CCN activation (Fig. a). The subsequent loss of cloud droplets at a faster rate suggests coalescence processes becoming dominant over nucleation, consistent with the transition to precipitation development. Both experiments display similar temporal patterns, with minor differences occurring after the peak period (20:00–21:00 UTC). Although the BASE simulation shows slightly higher values, these remain within the ensemble spread. The total cloud condensate QSUM peaked at a maximum of 2.0 gkg-1 at 20:00 UTC (Fig. b). Comparing the experiments, we see no significant difference. Consistent with case I, QSUM is dominated by rain drops and graupel. The onset of precipitation occurred around 19:00 UTC, coinciding with the increase in QNC and QSUM, intensifying to maximum rates of 2.7 mm(10min)-1 by 20:30 UTC (Fig. c). An observed lag of approximately 30 min between the peak in QNC and the maximum rain rate reflects the microphysical timescale necessary for cloud droplets to grow via condensation and collision-coalescence processes until precipitation-sized particles are formed.

Similarly to case I, no pronounced influence of urban emissions was detected. The temporal evolution of all examined variables was largely consistent between experiments, with minor variations well within the ensemble spread. Here as well, we performed a more detailed examination of vertical variability to better assess potential subtle impacts that might not be evident in horizontally and vertically averaged fields.

The BASE experiment shows a typical convective evolution with a vertical extent reaching up to 13 km (Fig. c). The number of activated cloud droplets peaked around 20:00 UTC with maximum values exceeding 150 cm-3 at around 4–5 km height. The total hydrometeor mass mixing ratio reached their maximum of 3.5 gkg-1 at 20:00 UTC (Fig. c). During this peak intensity period, strong graupel production dominates (see Fig. S9).

The comparison between BASE and NONURBAN experiments reveals contrasting effects across different periods of the analysis. Between 19:00–20:00 UTC, a slight increase in hydrometeor mass mixing ratio is observed in the BASE experiment (Fig. d). This increase is primarily attributed to increased cloud droplet numbers above 6 km altitude and enhanced graupel formation shortly before the precipitation maximum. A slight increase in the QNC signal is also detected during this period (Fig. b). At the time of peak precipitation (20:00–20:30 UTC), the BASE experiment shows significantly higher numbers of rain drops, reflecting increased rain processes during the most intense period (Fig. S9). After 20:00 UTC, the BASE experiment displays an increase of approximately 10 % in activated cloud droplets. The difference pattern in QSUM shifts compared to the earlier period and is characterized by increased graupel and snow formation of up to 5 % in NONURBAN (Fig. S9). While the BASE experiment shows 8 %–12 % more total ice content.

With urban emissions, cloud droplet activation is increased during the early period (19:00–19:30 UTC) followed by rapid ice formation at higher altitudes. This efficient freezing process increases latent heat release, which intensifies graupel and snow formation, leading to intensified precipitation formation. The resulting increased precipitation leads to a stronger downdraft and suppresses the updraft essential for sustaining convective activity. Analysis of vertical velocity fields confirms significantly stronger downdraft regions by 5 %–10 % in the BASE experiment (Fig. d), supporting this mechanism. In contrast, the NONURBAN experiment is characterized by continued graupel and snow production in the later period (20:00–21:00 UTC). Weaker downdrafts allow sustained updraft activity, supporting ongoing aggregation and riming, leading to fewer cloud droplets and ice in the NONURBAN experiment. This is supported by increased updraft regions above 10 km and reduced downdraft regions after 20:30 UTC (Fig. ).

The comparison reveals different convective evolution patterns with significant implications for precipitation efficiency. When including the Leipzig emissions, the BASE experiment produces increased early precipitation accompanied by stronger downdrafts that suppress further updraft development. Without urban emissions, the convective activity maintains more balanced updraft-downdraft circulation during the later period (20:00–21:00 UTC), enabling sustained aggregation and riming processes. These differences in updraft-downdraft balance suggest distinct convective evolution pathways between the two experiments, though the bulk precipitation time series in Fig.  does not reflect a clear temporal shift in overall precipitation amounts.