With the goal to deepen the understanding of cloud physics, a collaboration of the Swiss Federal Institute of Technology (ETH) Zurich and Leibniz Institute for Tropospheric Research (TROPOS) conducted targeted seeding experiments in Eriswil, Switzerland (47.071° N, 7.874° E, 920 m above sea level, a.s.l.). The setup of CLOUDLAB is described in detail in . The CLOUDLAB experiments focus on studying cloud processes and precipitation formation, utilizing a comprehensive range of in-situ and remote-sensing observations and numerical modeling techniques. During the winter 2023/2024, the CLOUDLAB project involved three additional collaborations, two of which were part of the German Science Foundation Priority Program PROM Polarimetric Radar Observations Meet Modelling,. The project PolarCAP (Polarimetric observations of Clouds and Precipitation) involved a deployment of the entire mobile platform LACROS Leipzig Aerosol and Clouds Remote Observations System, to the CLOUDLAB field site. The second collaboration, the CORSIPP project (Characterization of Orography-Influenced Riming and Secondary Ice Production and Their Effects on Precipitation Rates Using Radar Polarimetry and Doppler Spectra), is led by the Leipzig Institute of Meteorology (LIM) of University of Leipzig and utilized a 94 GHz cloud radar and the Video in Situ Snowfall Sensor (VISSS). The objective of this project is to investigate the microphysical processes that lead to the formation of distinct hydrometeor populations. Finally, the third collaboration was conducted with École Polytechnique Fédérale de Lausanne (EPFL), which employed an X-band radar and MASC (Multiple Angle Snowflake Camera) for in-situ measurements of ice crystal properties. As a result, the CLOUDLAB campaign 2023/2024 became one of the largest joint deployments of multi-wavelength radar and lidar systems ever. Further details on the instrumentation are provided in the following sections.

The primary subset of instruments from the CLOUDLAB winter campaign 2023/2024 that was utilized in the framework of this study is listed in Table . This study involves three cloud radars: one modified and one standard MIRA-35 operating at Ka-band, and one cloud radar operating at W-band. The full set of instrumentation during the 2023/2024 CLOUDLAB campaign was depicted by .

Firstly, MIRA-35 MBR5 (MicroBlaze Radar with serial number 5), is a scanning 35 GHz Doppler cloud radar, a modified version of the conventional LDR mode cloud radar, operated in SLDR mode. This device was part of the CLOUDLAB equipment of ETH Zurich. Unlike the standard LDR mode, SLDR measurements are less influenced by variations in hydrometeor orientation, even at low elevation angles . In this configuration, it transmits slant linear polarized pulses and receives in both co- and cross-polarized channels, enabling the detection of particle shape and orientation, particularly useful for distinguishing between ice crystal habits and liquid droplets. SLDR measurements from the 35 GHz cloud radar MIRA-35 MBR5, operating in SLDR mode , serve as the foundation for the VDPS method applied in this study . The co-cross correlation coefficient (ρcx) is also collected by MIRA-35 MBR5. It can be used as an additional constraint in the determination of the particle orientation . As part of this study, MIRA-35 MBR5 conducted range height indicator (RHI) scans, sweeping from 90° (zenith) to 150° elevation (equivalent to 30° elevation angle), at an angular velocity of 0.5° s-1. These RHI scans enable one to capture the polarimetric signatures of hydrometeors within the observation volume. The specified elevation angle notation is used consistently throughout the article. A detailed description of the standard LDR mode MIRA-35 was provided in .

Another 35 GHz cloud radar of the LACROS suite operated in Simultaneous Transmit Simultaneous Receive (STSR) scanning mode (MIRA-35 MBR7) and was co-located with MIRA-35 MBR5. This cloud radar is used in this study to provide zenith-pointing LDR overviews, which are comparable to SLDR measurements in the zenith direction.

The third from the row of cloud radars utilized for this study is the zenith-pointing, dual-polarization (DP) radar RPG-FMCW-94-DP (hereafter RPG94), which is operating at 94 GHz and was manufactured by Radiometer Physics GmbH (RPG). It employs Frequency-Modulated Continuous Wave (FMCW) signals to provide vertically resolved profiles of the Doppler spectra of SNR in the co- and cross-channels, radar reflectivity factor Ze, and LDR. RPG94 serves two main purposes in this study. First, it is used together with data from MIRA-35 in the computation of the Dual-Wavelength Ratio (DWRKa–W) and to provide continuous vertical-stare measurements of Ze and LDR when scanning was performed by both MIRA-35 MBR5 and MBR7 cloud radars (see Table ).

The VISSS was co-located beside the cloud radars during the CLOUDLAB campaign and characterizes particle shape and size of snowfall hydrometeors 1.5 m above the ground. The VISSS system is composed of two cameras equipped with LED backlights and telecentric lenses, enabling precise particle sizing. Its design combines a large observation volume with high pixel resolution while minimizing wind-induced disturbances. VISSS data products include a range of particle characteristics such as maximum dimension, cross-sectional area, perimeter, shape complexity, and fall velocity. demonstrates that VISSS delivers reliable statistics, capturing up to 10 000 distinct particle observations per minute. This instrument is used as an in situ measurement reference in this study to compare the hydrometeors observed at the surface with the particle shapes retrieved by the VDPS method.

Finally, the temperature information used in this study is obtained from the European Centre for Medium-Range Weather Forecasts (ECMWF) Integrated Forecasting System (IFS). Indeed, IFS temperature profiles represent grid-box averages and rely on model physics and data assimilation, which may limit their ability to capture local-scale and small-scale temperature variability, particularly in the boundary layer and under stable conditions. However, report generally good agreement between IFS and ground-based observations, while also highlighting differences on the order of several tenths of a Kelvin, which is sufficient for the purposes of our analysis. In order to provide a guideline for the robustness of the IFS temperature information used in the discussion of the case studies presented here, Fig.  has been added to the Appendix. The Payerne radiosonde station (46.813° N, 6.944° E, 491 m a.s.l) was selected for comparison with the IFS data from Eriswil as it is the nearest operational meteorological sounding station, providing the most representative in-situ vertical profile observations for the study area. Within Fig. , one representative example of the profile of atmospheric temperatures from the IFS at Eriswil station is compared with the corresponding radiosonde observation from Payerne for each case study. Given that IFS data at Eriswil and radiosonde profiles from Payerne show good agreement, we assume that the IFS can be reliably used to infer temperature in our study. Nevertheless, the deviations shown in Fig.  are used in the discussion of the case studies as an indicator for the presence of possible spatial inhomogeneities in the temperature field. Indeed, for the 22 February 2024, Fig. d shows a strong spatial variability, especially visible below 2.5 km height. Given this strong variability, the time-resolved temperature contours have been removed from Fig. , while they are shown in the overview plots of all other presented case studies.

The VDPS method is detailed in , where it is validated through three case studies representing the three main particle shape categories: prolate, isometric, and oblate. The VDPS method is based on a simplified Rayleigh scattering-based model with spheroidal geometry approximation and used a SLDR mode scanning cloud radar (MIRA-35 MBR5 in this study) and is designed to characterize the shape of cloud particles using. Indeed, the SLDR approach is particularly advantageous due to its slanted geometry, which enhances sensitivity to particle shape while reducing the impact of particle oscillations while falling. The employed spheroidal scattering model relates the elevation-angle dependency of SLDR with the polarizability ratio (ξ) and the degree of orientation (κ). ξ and κ describe the apparent particle shape through a density-weighted axis ratio and their preferred orientation, respectively . A polarizability ratio ξ≈1 (when ξ ranges from 0.8 to 1.2, depending on the radar calibration) corresponds to isometric particles, representing spherical particles or those with low density which appear as isometric ones to the radar. In contrast, a polarizability ratio ξ<0.8 or ξ>1.2 describes oblate and prolate particles, respectively. The utilization of the VDPS method in combination with auxiliary measurement techniques was demonstrated to capture the occurrence of riming and aggregation processes . In the current version of the method, SLDR is calculated based on the main peak of the Doppler spectrum of the SNR measured in the co-polarized channel (SNRco), which represents the dominant hydrometeor population, potentially composed of multiple hydrometeor types sharing the same terminal velocity, with the highest backscattering cross-section. An alternative approach proposed in this paper calculates SLDR using the main peak of the Doppler spectrum of SNR measured in the cross-polarized channel (SNRcross) highlighting hydrometeor populations that induce significant depolarization, such as columnar ice crystals. By comparing the polarizability ratio ξ derived by the VDPS method using the main peak signatures from both SNRco and SNRcross, respectively, it is possible to identify microphysical processes that generate distinct hydrometeor populations, such as secondary ice production (SIP), as will be demonstrated in Sect. .

The difference between the measured reflectivity within a cloud at 35 and 94 GHz can be used to detect large ice particles. Indeed, high values of DWRKa–W suggest the departure of the W-band radar from the Rayleigh scattering regime, indicating the presence of large aggregates or dense graupel, whereas values of DWRKa–W≈0 correspond to small hydrometeors. In order to compute DWRKa–W the following steps are performed. First, the observed reflectivity Ze from MIRA-35 MBR7 and RPG94 is mapped onto a common time-height grid using nearest-neighbour interpolation. The attenuation due to atmospheric gases, including water vapour, is estimated using the Passive and Active Microwave TRAnsfer (PAMTRA) tool and based on the temperature and humidity profiles from the European Centre for Medium-Range Weather Forecasts Integrated Forecasts System (ECMWF-IFS). Since DWRKa–W is only used qualitatively in this study, no correction for liquid water attenuation is applied. MIRA-35 MBR7 reflectivity Ze(Ka) is offset-corrected to align with RPG94 reflectivity Ze(W) in the Rayleigh regime, following the approach outlined by . This correction is performed using 30 min time intervals, selecting cloud regions that are at least 1 km above the 0 °C isotherm and 4 km above the surface, where Ze(Ka) ranges from −30 to −10dBZ. In these regions, both cloud radars are expected to operate within the Rayleigh regime. The reflectivity difference, Ze(W)-Ze(Ka), represents the offset, which is then applied to Ze(Ka) for the corresponding 30 min period. To ensure the quality of the offset correction, the correlation between Ze(Ka) and Ze(W) must exceed 0.9 for the selected cloudy pixels.

The interpretation of hydrometeor shape with VISSS is detailed in and follows a multi-step procedure. First, VISSS detects and analyses hydrometeors in video frames by first identifying motion between intermittent frames, reducing data volume. Particles are identified by background subtraction and the individual particle shapes are determined through dilation, filling and erosion. The final mask is used to compute particle properties like maximum size, area, perimeter, aspect ratio, canting angle, and complexity. Some quality criteria are applied: only particles with the maximal diameter Dmax≥2 pixels are retained. The VISSS combines particles detected by two separate cameras into a single 3D coordinate system, discarding those only visible in one camera. This is obtained by comparing the vertical position and extent of particles, assuming normally distributed measurement errors. The tracking of matched particles over time with VISSS enables the reconstruction of their 3D trajectories, allowing estimation of sedimentation velocity and interaction with turbulence. Natural tumbling provides multiple perspectives, improving estimates of particle properties such as the diameter, area, perimeter, and aspect ratio of particles.

The focus of this first case study is the characterization of liquid precipitation (rain) that was observed at Eriswil on 22 February 2024 from 11:00–13:00 UTC. A general overview of the precipitating cloud system is shown in Fig. . The black rectangle in Fig. a marks the time period of the RHI scan performed by MIRA-35 MBR5. The radar reflectivity factor Ze and LDR from RPG94 are shown in Fig. a and c, respectively. In Fig. b, the DWRKa–W calculated with difference between Ze(W) from RPG94 and Ze(Ka) from MIRA-35 MBR7, describes the size of particles.

The melting layer is clearly indicated by a sudden increase in Ze and LDR at around 1 km height, as shown in Fig. a and c, respectively. The location of the melting layer is confirmed by means of the RHI scan of SLDR by MIRA-35 MBR5 shown in Fig. a, where SLDR ≈-10dB. The vertical distribution of the polarizability ratio derived by the VDPS method is presented in Fig. b. Between 1.1 and 0.9 km above the ground, the VDPS method detects a transition from prolate (ξ>1) to oblate (ξ<1) shapes which is associated to the melting layer detected previously. Below the melting layer, low values of SLDR of around -30dB are measured with both MIRA-35 MBR7, and RPG94 as shown in Figs. a and c, respectively. The VISSS measurements, presented in Fig. c for the time period from 11:14–12:31 UTC provide in situ information on the shape and size of hydrometeors. In Fig. c, liquid droplets are detected as spherical particles with the VISSS and described as isometric particles by the VDPS method which derives a polarizability ratio ξ≈1 from the melting layer to the ground level, as shown in Fig. b. The comparison between particle shapes derived from the VDPS method and in situ VISSS measurements demonstrates the ability of the VDPS method to identify liquid droplets as isometric particles, as indicated by a polarizability ratio of approximately ξ≈1 from the melting layer down to the ground level.

The focus in this case study is put on a low-level stratus cloud layer that was observed at Eriswil on 8 January 2024 from 09:00–11:00 UTC. An overview on the observation is presented in Fig. . During the whole period, a low-level stratus cloud was located below 1.2 km height, capped by an inversion layer where temperatures are approximately at around -11°C (Fig. ). The cloud top exhibits low values of Ze and LDR, as shown in Fig. a and c, respectively, suggesting the presence of a supercooled liquid layer. Simultaneously, rimed dendritic or stellar plates are visually observed by eye at the site between 10:00 and 11:00 UTC, which is corroborated by the VISSS observations (Fig. c). These observations suggest the presence of a dendritic growth layer within the cloud, which would be the prerequisite for any subsequent riming to occur. Two case studies are presented in this section: the case outlined by the blue rectangle at around 10:08 UTC shown in Fig. a, which illustrates an early stage of the riming process, resulting in the formation of rimed dendrites, whereas the case highlighted by the red rectangle at around 10:38 UTC represents a more advanced stage of riming, leading to the development of graupel. The color codes defining the boxes are consistently used in Figs.  and .

As a starting point, the low-LDR layer that is present in both case studies depicted previously in Fig. c is examined. Regarding the Doppler spectrogram depicted in Fig. a and d, a layer with low values of SNRco associated with a fall velocity v≈0ms-1 is observed from 1.2–1 km height. The absence of detectable signals in SNRcross and SLDR at this height, as shown in Fig. b and c, respectively, indicates that these particles are likely supercooled liquid droplets at the top of the cloud. Between 1 and 0.8 km height, low values of SLDR are measured by MIRA-35 MBR5 as shown in Fig. a and d at all elevation angles. This corresponds to the presence of isometric particles as demonstrated by means of the VDPS retrievals shown in Fig. b and e, where ξ≈1. In parallel, an increase of SNRco is observable in Fig. a and low values of SNRcross and SLDR are shown in Fig. b and c, respectively, associated with a Doppler velocity v≈0ms-1. These features can be assigned to isometrically shaped particles, such as drizzle droplets. Indeed, the presence of an inversion layer can contribute to the sustainment of supercooled liquid water in clouds at temperatures as low as -10°C . Below 0.8 km height, another hydrometeor population is present which is characterized by Doppler velocities v≈-1ms-1, as shown in Fig. a, while low values of SNRcross and SLDR are presented in Fig. b and c, respectively. At this altitude, the VDPS method derives oblate particles according to Fig. b and e, which are characterized by low values of the polarizability ratio ξ=0.5. At the same altitude, a temperature inversion occurs with temperatures around -11°C, allowing the formation of plate-like crystals. Finally, supercooled liquid droplets are detected between 0.7 and 0.6 km height as indicated by the black rectangles in Fig. a and d, where no detectable signals are observed in SNRcross and SLDR, as shown by the black boxes in Fig. b, c and e and f.

Next, the blue-framed case from 10:08:16 UTC (Fig. ), which represents an early stage of a riming process, is discussed. Between 0.6 km height and the ground, the polarizability ratio increases and stabilizes at around ξ=0.8, as shown in Fig. b. This indicates the presence of slightly oblate-shaped particles falling at a velocity not exceeding v=-1ms-1, as shown in Fig. a–c. This phenomenon is attributed to the transformation of ice crystals upon contact with the supercooled liquid droplets that were detected down to 0.6 km height (Fig. a and d), leading to the transformation of dendritic crystals toward rimed dendrites, particles which appear oblate but more isometric than dendrites, and that fall more slowly than graupel. Regarding the particle shapes presented in Fig. c, rimed dendrites are detected using VISSS from 10:15–10:17 UTC. This in-situ observation confirms the hypothesis that the VDPS-based retrieval of weakly oblate particles, as shown in Fig. b below 0.4 km height, is associated to the presence of rimed dendrites.

Finally, the second red-framed case from 10:38:14 UTC representing an advanced stage of the riming process is discussed. The polarizability ratio is found to show a decrease at heights below 0.6 km height and stabilizes at ξ≈1 below 0.4 km height as shown in Fig. e. Unlike the previous case, more compact and spherical particles are detected with VISSS from 10:45:00 and 10:46:30 UTC, according to Fig. f. In this case, the advanced stage of riming allows the formation of spherical particles such as graupel, which are described by a polarizability ratio ξ≈1 using the VDPS method. Below 0.4 km in height, these particles fall faster, reaching velocities of v≈-2ms-1, as shown in Fig. d–f, due to their higher density.

In conclusion, the VDPS method identifies the formation of dendrites with a polarizability ratio ξ=0.5 from 0.9–0.7 km height where an inversion layer is occurring at temperatures around -11°C, conditions which are favourable for their development. Close to the surface, in turn, the presence of rimed dendrites is derived, which is associated to higher polarizability ratios ξ≈0.8. The riming process alters both the structure and shape of ice crystals, causing dendrites to appear more isometric to the radar. Finally, an advanced stage of riming produces spherical graupel particles, which are characterized by a polarizability ratio ξ≈1.

The following analysis is divided into two separate case studies. In Fig.  the observation of a convective cloud system is presented that was observed on 6 January 2024 from 16:00–23:00 UTC. The first case study, described in Sect. , is highlighted by a blue rectangle in Fig. a and was observed at around 17:40 UTC when aggregates formed. The second case study, described in Sect. , is highlighted by the red rectangle and was observed at around 22:08 UTC when two ice particle types were detected.

The first case study focuses on the period from 17:00–19:00 UTC on 9 January 2024, during which aggregates formed in the columnar crystal growth regime were observed. An overview about the scenario is provided in Fig. . In the course of the observation period, the cloud-top temperature of the precipitation system was constantly below -20°C. As the discussion of the case concentrates on the lower parts of the cloud system, the visualization is restricted to the height range from 0–4 km above the ground. The time period of interest ranges from 17:30–18:30 UTC, when a fallstreak of high-reflectivity (Fig. a), elevated DWR (Fig. b) and pronounced LDR signatures especially at W-band (Fig. c) and partially at Ka-band (Fig. d) was observed at heights between 2.8 km and the surface. According to VISSS measurements shown in Fig. e, large particles are detected at the surface between 17:50 and 18:30 UTC, where the diameter of hydrometeors reaches up to 15 mm. Such values are no longer compatible with the Rayleigh scattering regime for both Ka- and W-band radars. Also the high values of DWRKa–W during this time period indicate the presence of large particles causing the W-band to depart from the Rayleigh scattering regime. The formation of the large particles begins at around 2.8 km height where DWRKa–W ranges from 6 to 8 dB. At this altitude, the temperature is between -15 to -10°C (Fig. ) where aggregation is known to be rather efficient . This zone represented by the white rectangle in Fig. b, is correlated with high values of LDR from RPG94 (Fig. c). A small decrease in DWRKa–W is visible around 18:08 UTC, coinciding with an increase in LDR values from MIRA-35 MBR7 (Fig. d) and the presence of particles exceeding 15 mm (Fig. e). While the PSD is assumed approximately stable over the fall streak, slight changes in particle shape or orientation could influence DWRKa–W. Therefore, the observed decrease in DWRKa–W likely reflects the combined effects of both bands entering the non-Rayleigh regime and small variations in the PSD, rather than a large Rayleigh-to-non-Rayleigh transition. In this case, DWRKa–W stabilizes around 5 dB along the fallstreak.

According to the RHI scans of SLDR and ρcx presented in Fig. a and b, respectively, higher values of SLDR (at around -23dB) and low values of ρcx (at around zero) at all elevation angles are measured from 1 km height to the ground. As shown in Fig. c, the VDPS method derives slightly prolate particles, exhibiting a polarizability ratio ξ>1. Low values of ρcx are defined as fully chaotic orientation which can be considered as a special case of reflection symmetry , and are not compatible with columnar crystals, characterized as prolate particles which are mostly horizontally oriented. However they can be associated with a polarizability ratio ξ≈1, characterizing isometric particles.

Examining the Doppler spectrogram presented in Fig. c, low SLDR values at around -30dB are observed in regions where the particle fall velocity is v≈-1ms-1. As the fall velocity increases to v=-2ms-1, the SLDR rises to about -20dB, as highlighted by the black rectangle. This increase in SLDR is not correlated with any secondary ice population in SNRco, as shown in Fig. a. This indicates that only one hydrometeor population, which may be composed of multiple types of particles, is present at Doppler velocities between v=-1ms-1 and v=-2ms-1. However, the high values of SLDR in this region are correlated with low values of SNR highlighting that this hydrometeor population has a low particle concentration, consistent with the low concentration attributed to the large particles detected with VISSS (Diameter>10mm) shown in Fig. e. Large aggregated ice particles that are associated with higher fall velocities, lead to the non-Rayleigh scattering regime for both RPG94 and MIRA-35 as SLDR increases in Fig. c and d, respectively. In Fig. , values of SLDR are reaching only -23dB in the RHI scan while SLDR maximum values are higher according to the Doppler spectrogram presented in Fig. c. This occurs because SLDR is calculated using the main peak of SNRco defining the prominent hydrometeor population contained in the cloud, reducing the margin of error. Indeed, the particles which are falling slowly (v≈-1ms-1) exhibit lower SLDR values at around -27dB describing isometric particles. Finally, on the right side of the Doppler spectrogram where ice particles are falling with v>-1ms-1, high values of SLDR (SLDR>-15dB) are in general associated with prolate particles such as columnar crystals. During the same time period, VISSS observations (Fig. ) reveals a high number of aggregates and a low concentration of columnar crystals. A range of aggregate sizes is shown in Fig.  corresponding to the spectrum of fall velocities observed in the Doppler spectrograms shown in Fig. .

In the case of large aggregates, columns do indeed cluster together in all possible orientations (see Fig. ). The internal microphysics and the low-density structure containing small needles can amplify the crystal depolarization and increase the SLDR at all elevation angles. The inclusion of ρcx to the VDPS method under conditions of large ice crystals exhibiting high SLDR values may hinder the detection of large aggregates primarily composed of columnar crystals.

In conclusion, the increase in LDR in the W-band radar observations is intrinsically correlated with the occurrence of higher values of DWRKa–W (highlighted in the black box in Fig. ), indicating that the departure from the Rayleigh scattering regime occurs slightly earlier (in time as aggregation is ongoing) due to the smaller wavelength and influences the LDR. In contrast, LDR values from the Ka-band radar increase only when DWRKa–W decreases, typically when the ice particle diameter reaches 10–15 mm, allowing a broader range of particle sizes to form without significantly affecting the LDR.

Consequently, the VDPS method is using at this stage the MIRA-35 cloud radars, operating at Ka-band, and the impact of the non-Rayleigh scattering will occur for very large particles, only in this regard, the most of aggregates will be detectable as isometric particles. In the case of actually isometric aggregates larger then 15 mm, the VDPS method will derive slightly prolate shaped particles with a polarizability ratio ranging from 1.2 to 1.3 which is already considered as quasi isometric particles (Fig. c). Nevertheless for any potential future version of the VDPS method using a W-band radar, the non-Rayleigh scattering produced by particles larger than 5 mm has to be taken into account riming and aggregation processes.