An overview of the observed microphysical cloud structure is shown in Fig. . The radar reflectivity shows that the precipitation began at 15:10 UTC and was convective in nature (Fig. a). At around 17:30 UTC, the reflectivity increased at all altitudes, and the highest precipitation rates were observed at the surface (Fig. e). The period of high reflectivity (>10 dBZ) lasted for about 1 h. After this period, the cloud top lowered from 5000 to 2800 m, and the precipitation ended shortly after 18:40 UTC. The bulk of the precipitation originated at cloud top, as can be seen from the fallstreak pattern of enhanced radar reflectivity (>10 dBZ; Fig. a). The contour frequency by altitude diagram (CFAD; Fig. ) of the radar reflectivity (Fig. a) indicates a rapid increase in the radar reflectivity near cloud top, suggesting that the ice crystals were formed in the layer between 5000 and 4000 m. The ice crystals rapidly grew to large sizes between 4000 and 3000 m before they partly sublimated in the layer between 3000 and 2000 m, as indicated by the decreasing radar reflectivity (Figs. a and a) below 3000 m (assuming horizontal homogeneity). The majority of upward motion was observed above 3500 m (Figs. b and b). It is important to note that the measured vertical Doppler velocity is the sum of the particle fall speed and the air motion. Thus, as the ice particles grow to larger sizes while falling towards the ground, their fall speed increases and therefore masks the updrafts. The Doppler velocity CFAD shows large variations between -4 to 2 m s -1 near cloud top (Fig. b), indicative of turbulent motions. Indeed, the strong variability in the Doppler velocity was collocated with the enhanced shear layer from the wind profiler (Fig. b). Furthermore, the spectral width was also enhanced locally near cloud top (Fig. c), which can be attributed to the presence of turbulence (see Fig. b) near cloud top.

The occurrence of (1) high radar reflectivity fallstreaks (Fig. a), (2) positive Doppler velocities (Fig. b) and (3) increased spectral width (Fig. c) near cloud top suggests the presence of cloud top generating cells. Cloud top generating cells can enhance ice nucleation and growth and as such have important implications for precipitation formation (e.g., ; ; ; ; ; ; ; ; ; ), as will be further discussed in Sect. .

Ice particles that formed within the seeder region interact with other cloud particles while falling through the cloud and thus influence the microphysics of the feeder region below. The low-level cloud structure was observed with the tethered balloon system HoloBalloon (see Fig. ). The balloon-borne measurements indicate the presence of a low-level liquid layer that was confined to the lowest 300 m of the cloud (see Fig. ).

The cloud droplet number concentration (CDNC) increased from 100 to 350 cm-3 between 16:00 and 17:45 UTC (Fig. a) before the CDNC decreased after 18:00 UTC. The mean cloud droplet diameter ranged between 8 and 12 µm, as shown by the size distribution in Fig. a.

The ICNC was in the range of 1 to 4 L-1 between 16:00 and 18:00 UTC (Fig. b). ICNCs were higher when fallstreaks of enhanced radar reflectivity reached the surface. During the main precipitation event, after 18:00 UTC, the ICNC increased up to 14 L-1. During the same time period, the ratio between the ice water content (IWC) and total water content (TWC), which is often used to characterize the cloud phase (e.g., ; ), increased from 0.05 to 0.3 (liquid to mixed phase) to 0.9 (ice phase). Thus, a transition from a mixed-phase low-level cloud (before 18:00 UTC) to an ice-dominated low-level cloud (after 18:00 UTC) was observed during the passage of the cloud system (Fig. c). The cloud radar and microwave radiometer observations suggest that the entire cloud layer glaciated as an increase in the radar reflectivity (Fig. a) and a decrease in the LWP (Fig. d) was observed after 18:00 UTC. In the absence of sufficiently large updraft velocities for additional cloud droplet activation, the presence of large ice particles or high ICNC can lead to rapid glaciation of the cloud by the WBF process .

Even though downward motions were present on the lee side of the mountain barrier (see increased fraction of negative Doppler velocities in Fig. b) which contributed to hydrometeor evaporation/sublimation (see decreased reflectivity in Fig. a), a persistent low-level liquid layer was observed at Wolfgang. We suggest that this shallow low-level feeder cloud formed due to orographic lifting as the low-level flow in the lee of the mountain barrier was decoupled from the large-scale flow (Fig. a) and was forced to rise from Klosters (1200 m) to Wolfgang (1630 m) over the local topography. Similarly, in another case study of the RACLETS campaign, we found that the interaction between local flow effects and topography can induce the formation of updrafts and low-level feeder clouds . It is assumed that this shallow cloud could not generate significant precipitation by itself due to the limited time available for the collision and coalescence of cloud droplets to produce precipitation-sized particles and due to the high temperatures (>-3 ∘C) which were limiting the amount of INPs and thus ice formed through primary ice nucleation. However, the hydrometeors that formed in the generating cells can “feed” on the low-level liquid layer and thus enhance precipitation by riming and depositional growth. Additionally, it can provide an environment favorable for the production of secondary ice particles, as will be discussed in Sect. .

Observations from the cloud radar, microwave radiometer, HoloBalloon platform and ground-based aerosol measurements were combined to study the microphysics within cloud top generating cells. Since no in situ observations within generating cells or near cloud top were available during the RACLETS campaign, the analysis of the microphysics was limited to observations from remote sensing instrumentation and balloon-borne in situ measurements near cloud base. In the first part of this section, the overall dynamical and microphysical structure of generating cells is characterized, whereas in the second part the origin of ice crystals and the microphysical growth processes active within generating cells are investigated from an INP-cloud perspective.

When the strongest generating cells were present, vertical overshooting of up to 500 m was observed at the cloud top (Fig. ; e.g., at 16:00 and 16:45 UTC), indicating the presence of strong updrafts.

This was also supported by observations of the maximum Doppler velocity (Fig. b), which was derived from the Doppler spectra (see Appendix ) and used as a proxy to identify updraft regions. The maximum Doppler velocity suggests that the strongest updrafts were present in the core regions of the cloud top generating cells (>3 m s-1), whereas updrafts were weaker outside of the generating cells and at altitudes below 3000 m (Fig. b). It is likely that liquid water was produced in these updraft cells as a positive correlation was found between the vertically integrated maximum Doppler velocity and the LWP measured by the microwave radiometer (see Fig. b). Moreover, anomalies in the cloud top properties and the LWP were observed during the periods with generating cells (Fig. d). Coinciding peaks in the anomaly signal were labeled as GC1 (16:00 UTC), GC2 (16:45 UTC) and GC3 (17:55 UTC). The Spearman's rank correlation coefficients of the anomalies ranged between 0.46 (for reflectivity and spectral width) and 0.73 (for reflectivity and LWP), significant at the 5 % level. Thus, given the significant correlation between updrafts, LWP and radar reflectivity within generating cells, it is likely that the updrafts acted as a major driver for the formation and maintenance of generating cells by providing a continuous source of liquid water and thereby enhancing ice nucleation and growth through immersion freezing, subsequent vapor deposition and riming.

To further explore the microphysics within cloud top generating cells, the Doppler spectra along the 17:00 UTC fallstreak were investigated (Fig. ). This approach allows us to obtain a continuous picture of the evolution of the particle populations along the fallstreak and to draw conclusions regarding the microphysical processes active. Previous studies used the Doppler spectra information for the classification and characterization of ice particle shape and particle populations (e.g., ; ).

The vertical profile of the Doppler spectra shows a broad particle distribution spanning from -5 to 4 m s-1 between 3300 and 5000 m height, indicative of a turbulent layer. This layer likely marked the extent of the generating cell where ice crystals were produced and initial growth occurred. The Doppler spectra show a spectral bimodality below 3300 m (Fig. ; i.e., presence of multiple particle populations with different fall speeds) which extends down to the surface. When analyzing the Doppler spectra of the full period with the peakTree technique (Sect. ), multi-peaked situations become evident at the leading edges of the fallstreaks (Fig. c). For example, the Doppler spectrum in Fig. b (red line) indicates the presence of two particle populations: a fast falling one (-2 m s-1) and a slow falling one (-0.5 m s-1). The LDR of the slower falling particle population was slightly higher (-25 dB; not shown) compared to the faster-falling population (-28 dB; not shown). These LDR values are characteristic for oblate or plate-like particles . The observed Doppler spectra and the ice habits observed near cloud base (Fig. ) suggest that the faster falling population represents heavily rimed ice particles and/or graupel, whereas the slower falling population was associated with stellar dendrites. This is also consistent with the observed temperature (dendrite regime; ; ) and the presence of supercooled liquid (riming) within the generating cells. It is likely that these two particle populations were already present above but only separated below the turbulent layer due to the weaker updrafts and their difference in fall speed.

In the following, we will further investigate the origin of ice particles that formed within generating cells. Numerous studies have observed enhanced ice formation and growth in these updraft regions (; ; ; ; ; ; ). For example, found that the ICNC was enhanced by a factor of 2 to 3 within the core region of generating cells compared to the region between the cells. While most of the studies agree that radiative cooling is a major driver for the formation and maintenance of cloud top generating cells, less research has focused on the reason for the enhanced ICNCs that were observed within these cells. Here we provide potential reasons from an INP-cloud perspective and propose possible mechanisms by considering the measured INP concentrations and cloud base observations of the ICNC and ice particle size.

INP concentrations were measured at the valley site of Wolfgang (1630 m) and at the mountaintop station of Weissfluhjoch (2700 m) (Fig. ).

The observed INP concentrations at a given temperature spanned over 1 order of magnitude. The INP concentration measured at 07:15 and 09:30 UTC was a factor of 3–10 higher at Wolfgang compared to Weissfluhjoch, which was presumably a consequence of the decoupled low-level flow (see Fig. a) and thus the sampling of different air masses. Based on the INP measurements at Weissfluhjoch, an INP concentration of 0.27 L-1 (0.15 to 0.48 L-1) was extrapolated at cloud top (Fig. ). It is important to note that the cloud top INP concentration was estimated from the Weissfluhjoch measurements in the morning (i.e., representative for pre-cloud INP concentrations) as no INP concentrations were measured at Weissfluhjoch during the passage of the cloud system.

Additionally, cloud measurements were conducted by the HoloBalloon platform near cloud base. Since no in situ observations were available within the generating cells, assumptions regarding the upper-level cloud properties were required. We assumed that the largest ice particles (>400 µm; derived from particle size distribution in Fig. b) formed near cloud top and grew to these large sizes while falling to the surface. This criterion is based on the assumption that the large ice particles did not sublimate completely prior to reaching the surface. The ICNC>400µm observed near cloud base was in the same order of magnitude as the radar-retrieved ICNC at cloud top (red dots in Fig. ). This observation further supports the assumption that ice particles larger than 400 µm originated near cloud top.

The comparison between the observed ICNC>400µm and the estimated INP concentration at cloud top shows a discrepancy between the INP concentration and observed ICNC during certain time periods (Fig. ). This suggests that the observed ICNC cannot be solely explained by primary ice nucleation, but that other mechanisms were active.

Static instability driven by cloud top radiative cooling can produce strong updrafts (Fig. b) and lead to convective overshooting of cloud top generating cells (see red arrows in Fig. a). This convective overshooting can decrease the cloud top temperature and therefore increase the ICNC formed by primary ice nucleation. For example, the cloud top height during GC1 increased by 500 m from 4500 to 5000 m. Considering the observed temperature profile in Fig. b, the cloud top temperature decreased by 3.6 ∘C from -21 ∘C (at the average cloud top height) to -24.6 ∘C (at 5000 m) upon convective overshooting. Consequently, the estimated INP concentration increased by a factor of 3.1 from 0.27 L-1 (0.15 to 0.48 L-1) to 0.85 L-1 (0.42 to 1.7 L-1) (Fig. ) due to the lower cloud top temperature. The ICNC>400µm measured at cloud base lied below or near the extrapolated INP concentration at -24.6 ∘C before 18:00 UTC (Fig. ). This suggests that the observed ICNC>400µm before 18:00 UTC can be solely explained by primary ice nucleation and convective overshooting. After 18:00 UTC, the ICNC>400µm measured near cloud base lied several factors above the convective overshooting line (Fig. ), suggesting that other processes were occurring.

For example, the positive feedback between supercooled liquid water, radiative cooling and turbulence that has been observed near cloud tops (e.g., ) might have contributed to enhanced ice formation. The presence of supercooled liquid can lead to strong longwave radiative cooling (e.g., ). This radiative cooling decreases the stability near cloud top, which causes turbulent motions that in turn can produce further supercooled liquid water. The magnitude of the longwave radiative cooling strongly depends on the cloud phase, the liquid water content (LWC) and particle size distribution, among other factors (e.g., ). Indeed, the LWP, as measured by the microwave radiometer, was enhanced within generating cells (see Fig. d) and thus likely increased the longwave radiative cooling at cloud top. The question is by how much the radiative cooling was enhanced within generating cells due to the increased cloud liquid water compared to their surrounding regions. Previous studies observed longwave radiative cooling rates in the range of 1 to 5 K h-1 near cloud top (e.g., ; ; ; ; ; ; ; ; ). Additionally, computed radiative heating rate (RHR) profiles in the atmosphere as a function of cloud type and LWP by using an observational dataset. According to , an increase in the LWP from 50 to 150 g m-2 (e.g., GC2 in Fig. ) in MPCs can increase the longwave radiative cooling rate from around 1.7 to 2.9 K h-1 (ΔRHR = 1.2 K h-1). This could potentially cool the cloud top temperature by 0.3 K if a lifetime of 15 min is assumed for generating cells (i.e., 1.2 K h-1 × 15 min = 0.3 K) and increase the estimated INP concentration from 0.27 to 0.3 L-1 (a factor of 1.1; see Fig. ). Thus, in the present case study, longwave radiative cooling only plays a minor role in enhancing primary ice nucleation. Nevertheless, longwave radiative cooling is of major importance for the production of radiatively driven turbulence near cloud top and thus for maintaining generating cells.

Other mechanisms must be active to explain the increased ICNCs after 18:00 UTC. For instance, the enhanced updrafts in generating cells allow for all hydrometeors to grow to larger sizes. It is unlikely that the larger cloud droplet size would significantly increase primary ice nucleation by immersion freezing, which is the dominant ice nucleation mechanism in MPCs (e.g., ; ; ). However, it can play an important role for SIP. For example, the freezing of drizzle-sized droplets can release small secondary ice particles (e.g., ; ; ; ). This process is known as droplet shattering and has been observed to be strongly dependent on the cloud droplet size and to be potentially effective over a large temperature range (; ). Previous field studies have observed the presence of drizzle-sized droplets in the size range of 100 to 300 µm in regions of strong vertical updrafts (e.g., ; ; ). Thus, droplet shattering could increase the ICNC in generating cells by several factors if supercooled drizzle drops are present in the updraft regions. However, in situ observations within generating cells would be necessary to further investigate this hypothesis.

In summary, the increased ICNC in generating cells can be the result of different mechanisms or a combination of several mechanisms. Three possible mechanisms have been proposed in this study, and their potential contributions are summarized in Fig.  and further discussed in Appendix .

Firstly, primary ice nucleation in generating cells can be increased due to convective overshooting or radiative cooling. The ICNC observed before 18:00 UTC can likely be explained by these two mechanisms since the estimated INP concentration and the ICNC>400µm measured near cloud base agreed within the same order of magnitude (Fig. ). For the present study, we found that the ice crystal enhancement factor from convective overshooting (factor 2.2 to 5.2) was larger than that of radiative cooling (factor 1 to 1.4). On the other hand, the ICNC>400µm after 18:00 UTC exceeded the estimated cloud top INP concentration by up to a factor of 8, suggesting that SIP processes such as droplet shattering might have been active within generating cells and contributed to the glaciation of the MPC. However, more targeted studies are necessary to understand which mechanisms are responsible for enhanced ice formation and growth within cloud top generating cells. In particular, in situ measurements of the cloud properties within generating cells and their environmental conditions (e.g., temperature, updrafts, INP conditions) are of major importance to address these questions.

Ice crystals that formed in the seeder region can grow by microphysical interactions with other cloud particles while falling through the cloud layer and thus influence the microphysics of the entire cloud. For example, if large ice particles fall through a supercooled liquid layer, they can initiate the glaciation of the cloud layer through the WBF process and/or grow by riming. The total number of peaks in Fig. c shows multi-peaked situations below 3300 m, indicating the presence of multiple particle populations with different fall speeds. This suggests that secondary ice particles might be produced in the feeder region of the cloud. In the following, we investigate the importance of ice growth and SIP in the feeder region by analyzing the phase-resolved cloud properties measured in situ with the HoloBalloon platform. In particular, the analysis of the ice crystal habit and size can provide important information about the formation and growth history of ice particles.

Figure  shows a representative set of ice particle images observed by HOLIMO as a function of height and time. It can be seen that ice crystal habits varied greatly during the passage of the cloud system.

For example, the images indicate the presence of numerous columns between 17:00 and 17:20 UTC at altitudes above 1780 m (yellow boxes) which are known to grow at temperatures between -3 and -10 ∘C (; ). Furthermore, irregular-shaped particles including ice fragments were abundant (green boxes), consistent with previous studies (e.g., ; ). A large fraction of graupel and rimed particles was observed between 17:00 and 17:40 UTC (red boxes). After 18:00 UTC, the ice crystals became more aggregated (blue boxes) and less rimed (see also MASC data in Fig. d), suggesting a decrease in the amount of liquid water available for riming. Furthermore, from 18:00 UTC onwards dendrites and broken branches of dendrites were more abundant. Small pristine ice crystals (plates and columns) were present over the entire period (see Fig. c and purple boxes in Fig. ).

The large variability in ice crystal habit and size suggests that the ice crystals have formed and grown in different cloud regions. As discussed in Sect. , it is likely that the heavily rimed ice particles and large dendrites (Fig. ) were produced within the seeder region of the cloud and gained mass by riming and deposition while falling through the cloud. On the other hand, the small pristine ice crystals were likely formed within the feeder region of the cloud. Previous studies have found that small pristine ice crystals (<100 µm) were spatially correlated with their environment of origin (e.g., ). For example, it is possible that the observed columns originated within the multi-peaked structures (Fig. c) as the temperature below 3000 m was in the temperature regime of columnar growth . Pristine plates likely grew in the lowest part of the cloud where the prevailing temperature was above -3 ∘C. These small ice crystals (<100 µm) could have formed either by primary ice nucleation or by SIP processes within the feeder cloud and rapidly grown by diffusion to larger sizes (e.g., ). The contribution of primary ice nucleation to the observed ICNC can be estimated from the measured INP concentration at Wolfgang (Fig. ; assuming coupling between the surface and the lower part of the cloud), which was below the minimum detectable concentration at a temperature of -3 ∘C. Thus, the minimum detectable concentration of 6.3×10-4stdL-1 (see Sect. ) represents an upper limit for the INP concentration within the feeder region. The ICNC<100µm in diameter observed in the feeder cloud (1 to 2 L-1; Fig. a) exceeded the estimated INP concentration by 3 orders of magnitude, suggesting that primary ice nucleation alone cannot explain the small ice crystals observed.

Secondary ice production processes are necessary to explain the observed ICNC in the feeder cloud. Since the cloud droplets in the low-level feeder cloud were small (< 50 µm in diameter; Fig. a), droplet shattering was likely not the responsible mechanism. However, as the temperature at 1900 m was around -3 ∘C and large rimed particles (Fig. a) and cloud droplets larger than 25 µm in diameter (Fig. b) were observed in the feeder cloud, the Hallett–Mossop process may have been active (; ).

Another mechanism that could have led to the production of secondary ice particles in the low-level feeder cloud is ice particle fragmentation upon ice–ice collisions (e.g., ; ). As the low-level liquid layer contained small pristine and large rimed ice particles (Fig. a), which have different terminal fall velocities and therefore enhanced collision efficiencies, collisional ice fragmentation may have been occurring. Indeed, the ice crystal images in Fig.  indicate the presence of ice fragments (e.g., broken-off arms of dendrites after 18:00 UTC). Based on the temporal evolution of the cloud properties within the feeder cloud (Fig. a, b), which shows an increase in the ICNC and a decrease in the number of large cloud droplets after 18:00 UTC, we suggest that ice particle fragmentation upon collision was the dominant SIP process after 18:00 UTC. In contrast, the presence of large cloud droplets (>25 µm) before 18:00 UTC suggests that both the Hallett–Mossop process and collisional ice fragmentation contributed to the observed ICNC.

Previous studies have observed large discrepancies between the INP concentration and ICNC in the feeder region of clouds (e.g., ; ; ; ). These observations were frequently conducted at mountaintop research stations or near mountain slopes where ICNCs of several hundreds to thousands per liter have been reported (e.g., ; ; ). These large ICNCs were attributed to the influence of surface processes such as blowing snow (, ), hoar frost , riming on snow-covered surfaces or ice crystal enhancement through turbulence and convergence , whereas the contribution of SIP processes has been suggested to be minor or has been difficult to assess (, ). By performing balloon-borne measurements in a mountain valley, we measured ICNCs that were 2 orders of magnitude lower than previous observations at mountaintops (1 to 10 L-1 instead of 100 to 1000 L-1) and thus were able to significantly reduce the impact of surface processes. Based on the estimated INP concentration (Fig. ) and observed ICNC (Fig. a), we suggest that SIP processes contributed up to 1–2 L-1 to the observed ICNC in the presented case study and thus accounted for up to 50 % of the total ICNC before 18:00 UTC. However, the increase in the ICNC from 3 up to 14 L-1 after 18:00 UTC (Fig. a) cannot be solely explained by SIP within the feeder cloud since the observed increase was primarily due to large ice particles (>300 µm; see Fig. b). As a substantial fraction of aggregates after 18:00 UTC are dendrites and broken-off arms of dendrites, the ICNC might be attributed to ice–ice collision breakup within the seeder region. Interestingly, the discrepancy between the INP concentration and ICNC observed in the present study after 18:00 UTC (factor of around 8; see Sect. ) is strikingly similar to the ice multiplication factor of dendrites previously observed at Jungfraujoch in winter clouds even if both studies applied different approaches to determine the enhancement factor and to reduce the influences of surface processes.

If only a small concentration of secondary ice particles is captured by updrafts or turbulence within the feeder region and lifted aloft, they can initiate further ice formation and growth at temperatures well above typical INP activation temperatures and have a significant impact on the development of the cloud (e.g., cloud properties, glaciation, lifetime). While the CDNC and CDNC>25µm decreased above 1850 m, the vertical profiles of the ICNC showed no height dependence over the 200 m height interval (Fig. ).

This suggests that SIP was active over the entire low-level feeder cloud. However, due to the limited vertical extent of the profiles, we cannot make a final statement regarding the impact of SIP within the feeder region on the cloud microphysics aloft. Further observations in “surface-decoupled” environments (i.e., reduced influence of surface processes) with a larger vertical extent are required to assess the role of SIP in feeder clouds. This is important as secondary ice production in the feeder region can potentially enhance orographic precipitation.