Abstract

Introduction

The concentration and size of ice crystals in clouds have a significant influence on precipitation formation (1, 2), the radiative energy budget (3), and cloud lifetime (4). Despite their important role, fundamental knowledge gaps in ice formation and growth processes exist, owing to complex interactions among the hydrometeors and the wide variety of ice crystal shapes and densities found in nature (5, 6). Understanding and quantifying the growth of ice crystals in the atmosphere is crucial for developing physically based parameterizations in weather and climate models (7).

At temperatures above $-{38}^{\circ }$C, a special subset of aerosols known as ice nucleating particles are required for ice crystals to form (8). Once formed, ice crystals grow by vapor deposition into their preferred ice habit at the given ambient temperature and relative humidity. The vapor attaches to the crystal surface in steps centered on the corners, where the supersaturation is highest, and then spreads as layers along the faces (9–11). While temperature determines the main growth direction by controlling the relative rates of surface diffusion on the basal and prism faces (10, 12, 13), supersaturation determines the magnitude of the growth rate and the degree of secondary growth features such as hollows (9, 14–17). Furthermore, ice crystals can grow at the expense of evaporating cloud droplets when the vapor pressure lies between ice ($e_i$) and water saturation ($e_s)$ (i.e. $e_i<e_s$) (18, 19), a process referred to as the Wegener–Bergeron–Findeisen (WBF) process (20–22). The speed of the WBF process strongly depends on whether cloud droplets and ice crystals are uniformly distributed (genuinely mixed) or spatially separated (conditionally mixed) within a cloud volume (23). As ice crystals grow, they can collide and coalesce with other hydrometeors (e.g. riming, aggregation) or splinter and produce new ice crystals through secondary ice production processes (24). However, like microphysics in general, ice processes cannot be resolved in weather and climate models and need to be parameterized. Improper representation of ice processes in models introduces uncertainties in both weather forecasts and climate projections (7, 19, 25, 26).

Microphysical parameterizations are developed based on both theory and observations from laboratory and field studies (27). These parameterizations consist of simplified rate equations that are tailored to represent a specific microphysical process. However, these process rate parameterizations, such as the vapor diffusional growth rate of ice crystals, are difficult to constrain. The only way to quantify individual process rates under controlled conditions is through laboratory experiments. For example, Ryan et al. (28) measured the axial and mass growth rates of freely suspended ice crystals in a cloud chamber at temperatures ranging from $-{21}^{\circ }$C to $-3^{\circ }$C for growth times of up to 3 min. Similarly, Takahashi et al. (29) measured the vapor diffusional growth rate of ice crystals for extended growth periods of up to 30 min in a vertical supercooled cloud tunnel at temperatures ranging between $-{23}^{\circ }$C and $-3^{\circ }$C. Both studies observed the maximum growth rates on the basal face at around $-6^{\circ }$C (i.e. along c-axis, columnar growth) and on the prism face at $-{15}^{\circ }$C (i.e. along a-axis, plate-like growth). While laboratory studies are crucial for studying processes within a well-controlled environment, they have difficulty exploring the full range of atmospheric conditions present in natural clouds due to design limitations (e.g. short residence time) and the constrained setting (e.g. limited variability in temperature and/or supersaturation).

Up to now, it has been challenging to use field observations to constrain microphysical parameterizations, given the simultaneous occurrence of multiple interacting processes in natural clouds and the significant temporal and spatial variability. This necessitates the use of sophisticated sampling strategies that go beyond measuring the current state of the cloud. Possible approaches include cloud measurements along quasi-Lagrangian flight tracks (30–32) or repeated penetrations in quasi-steady-state clouds such as wave clouds (33, 34). These studies were limited to temperatures below $-{10}^{\circ }$C, where clouds contain sufficiently high concentrations of naturally occurring ice nucleating particles (8), to obtain solid statistics. To address this lack of observations, here we use glaciogenic cloud seeding (i.e. injection of artificial ice nucleating particles) to produce ice crystals and initiate subsequent growth processes in supercooled stratus clouds otherwise devoid of ice (see Methods). By adjusting the distance and time between seeding and measurement, we can quantify the ice crystal growth rates across a range of growth times and temperatures. This approach allows growth rate studies to extend to temperatures above $-{10}^{\circ }$C, a temperature range that is particularly important, as it covers crucial precipitation initiation processes (e.g. WBF, seeder–feeder, Hallett-Mossop).

Results

The observations presented here were obtained during the CLOUDLAB measurement campaigns (35) conducted in the Swiss Plateau, where wintertime stratus clouds were seeded by a burn-in-place flare mounted on an uncrewed aerial vehicle (UAV) to trigger ice formation and subsequent ice growth processes. The seeding-induced microphysical changes were observed 4 to 10 min downwind of the seeding location by ground-based remote sensing instrumentation and in situ instrumentation (e.g. holographic imager HOLIMO) mounted on a tethered balloon system (36). One seeding experiment included several consecutive seeding missions (conducted within 30 min of each other) with changing distance (2–3 km) between seeding and measurement location. The environmental conditions were assumed to remain approximately constant during this time period due to the persistent nature of stratus clouds (37). Following this approach, systematic ice growth studies were conducted in this natural quasi-steady-state environment by adjusting the seeding distance (and consequently ice growth times) and measuring the resulting ice crystal sizes. A conceptual overview of the observational setup and experimental approach is shown in Fig. ​Fig.11 and described in more detail in the Methods section.

Open in a separate window

Fig. 1.

Schematic of the experimental setup of CLOUDLAB. Seeding particles (brown plume) are released into a supercooled stratus cloud by a UAV flying tracks perpendicular to the wind. This triggers ice formation and growth via the WBF process, as illustrated by the increase in the ice (white hexagons) and the depletion of the liquid phase (blue circles). The seeding signal is measured downwind by a tethered balloon system and by vertically pointing and scanning cloud radars (blue). The expected radar seeding reflectivities are highlighted by blue–green–yellow colors. The seeding distance (and thus ice growth time) is varied between consecutive seeding missions, while the environmental parameters such as temperature (T) and wind speed (u, from left to right) remain approximately constant, enabling systematic studies on ice growth.

Two seeding experiments conducted at different seeding temperatures are shown in Fig. ​Fig.22 ($\text{T}=-{5.2}^{\circ }$C) and Fig. S1 ($\text{T}=-{7.2}^{\circ }$C). The experimental parameters of the seeding missions are summarized in Table ​Table1.1. For both sets of experiments, the seeding material was introduced 2 km and 3 km upwind of the observational site by the UAV, leading to different growth times. The seeding pattern consisted of four 400 m horizontal legs at constant altitude perpendicular to the wind direction within the flare burning time of 5 min.

Open in a separate window

Fig. 2.

Temporal evolution of the microphysical properties observed during the seeding missions S5-6.6 and S5-9.3 conducted on 2023 January 24 at a seeding temperature of $-{5.2}^{\circ }$C and at seeding distances of 2 and 3 km, respectively. a) Time–height radar reflectivity measured by the vertically pointing W-band cloud radar. The white line shows the height of the tethered balloon, whereas the white stars indicate the height and start time of the seeding missions. The black bars highlight the time periods of expected seeding signal, taking into account the mean wind speed and the flare burning time of 5 min. b) Radar reflectivity measured by the scanning Ka-band cloud radar during a plan position indicator (PPI) scan at 19:31 UTC at an elevation angle of 30° and an azimuth angle ranging between 212° and 302°. c) Timeseries of the cloud droplet number concentration (gray line) and ice crystal number concentration (blue crosses). The background periods (BG-6.6: 18:45:45–18:55:45 UTC, BG-9.3: 19:16:00–19:26:00 UTC) and seeding periods (S5-6.6: 18:57:45–19:01:45 UTC, S5-9.3: 19:28:00–19:33:20 UTC) are highlighted by the horizontal gray bars. d) Violin plots of the liquid water content (LWC, gray) and ice water content (IWC, blue) measured during the background and seeding periods (same periods as depicted in c). Note that no ice crystals were measured during the background periods (i.e. $\text{IWC}=0$). e) Timeseries of the ice crystal size distributions in terms of the major axis length. f) Cloud droplet size distribution (gray) and ice crystal size distribution (blue) measured during BG-6.6 (dashed), S5-6.6 (solid), BG-9.3 (dotted), and S5-9.3 (dot-dashed) in terms of the equivalent area diameter. The data presented in c)–f) were obtained from HOLIMO and averaged over 2 s c,d) and 10 s e), respectively.

Here, we provide a comprehensive description of the seeding-induced microphysical changes for the experiment conducted at a temperature of $-{5.2}^{\circ }$C (Fig. ​(Fig.2),2), but the findings also remain applicable for the experiment conducted at $-{7.2}^{\circ }$C (Fig. S1). The natural background cloud on 2023 January 24 was characterized by a radar reflectivity of $-25$ dBZ, a cloud droplet concentration of 250–350 ${\text{cm}}^{-3}$, a liquid water content of 0.2–0.25 g m⁻³, and a low ice water content of <0.001 g m⁻³ (Fig. ​(Fig.2a,c,d).2a,c,d). During the passage of the seeding plume, the radar observations indicated regions of locally enhanced radar reflectivities of up to 0 dBZ (Fig. ​(Fig.2a,b),2a,b), which clearly stand out from the natural background. Although the seeding material was injected at a constant altitude, seeding signatures were observed across a large vertical extent in the cloud radar due to turbulent mixing, updrafts induced from latent heat release, and sedimentation of ice crystals (38). HOLIMO revealed high concentrations of ice crystals (up to 2,000 L⁻¹) with a simultaneous reduction in cloud droplet number concentration during the passage of the seeding plume (Fig. ​(Fig.2c).2c). During some time periods, the liquid phase was entirely depleted, demonstrating the efficiency of the WBF process in natural clouds. After the passage of the seeding plume, the cloud properties returned to the natural background levels present prior to seeding. Thus, the crucial assumption of constant microphysical and environmental properties in the background between consecutive seeding missions was substantiated by observations. The ice crystals measured during S5-9.3 were consistently larger than the ice crystals measured during S5-6.6, due to the longer growth time (Fig. ​(Fig.2e,f).2e,f). This increase in ice crystal size was also visible in the enhanced radar reflectivities during S5-9.3 (Fig. ​(Fig.2a).2a). Thus, by employing glaciogenic cloud seeding in supercooled stratus clouds, we were able (i) to produce a sufficient number of ice crystals at temperatures above $-{10}^{\circ }$C, ensuring robust statistics and (ii) to systematically adjust the growth time (by changing seeding distance) between consecutive seeding missions, resulting in different ice crystal sizes, allowing for comprehensive growth studies in natural clouds.

Given that the shape of ice crystals resulting from growth by vapor deposition is influenced by the environmental conditions (6), growth rates are typically reported individually for the basal and prism faces (28, 29). Figure ​Figure33 gives an overview of the ice crystal axial dimensions measured during the seeding experiment at $-{5.2}^{\circ }$C for different growth times. Ice crystals with a major axis length of up to 500 $\mu$m and up to 800 $\mu$m were measured during S5-6.6 and S5-9.3, respectively. The major axis length of the ice crystals measured during S5-9.3 was on average 33% larger than the one observed during S5-6.6. A large variability was observed in the major axis length, while the minor axis varied less for both missions (Fig. ​(Fig.3b,c).3b,c). The variability among the higher values of the ice crystal size distribution might even be underestimated due to sedimentation of larger ice crystals (Fig. ​(Fig.2a).2a). The variability in the major axis length at $-{5.2}^{\circ }$C is likely linked to varying levels of supersaturation experienced by the ice crystals during their growth, as discussed in the following paragraphs. The ice crystals measured at $-{5.2}^{\circ }$C were characterized by a mean aspect ratio of $3.1\pm 1.3$ (Fig. ​(Fig.3a)3a) and a columnar shape (Fig. S3A). On the other hand, the ice crystals observed during the seeding missions at $-{7.2}^{\circ }$C exhibited an aspect ratio of $1.6\pm 0.3$ (Fig. S2A). The ice crystal images showed the occurrence of plates and short columns including a large number of hollow columns indicative of high supersaturation (Fig. S3B).

Open in a separate window

Fig. 3.

Overview of the ice crystal axial dimensions measured during the seeding missions S5-6.6 ($N=\mathrm{17,796}$) and S5-9.3 ($N=\mathrm{7,041}$). a) 2D histogram of the major and minor axis length of the detected ice crystals. The number of ice crystals observed with a given major and minor axis length during S5-9.3 are shown with the green colormap, while S5-6.6 is shown by the purple contour lines. The dots represent the corresponding median axial dimensions (including all observed ice crystals) and the black lines show contour lines of aspect ratio (defined by major axis length divided by minor axis length). b, c) Normalized probability density function (PDF) of the major and minor axis length. The triangles represent the corresponding time-weighted medians, whereas the numbers indicate the mean values and standard deviations of the PDFs. In the upper-right corner, a schematic of a column’s major and minor axes is displayed, corresponding to growth along the basal and prism faces.

The diffusional growth of ice crystals can be expressed as the growth of the axial dimensions as a function of time. While Ryan et al. (41) found that the ice crystal growth can be adequately described as a linear function of time (i.e. constant growth rate) at short time scales (<3 min), Takahashi et al. (29) found a power-law function more suitable for longer time periods (3–30 min). In the following, we evaluate whether the growth of the observed ice crystal populations can be represented by a constant growth rate, i.e. independent of time. At $-{5.2}^{\circ }$C, an average linear growth rate of 0.13 ± 0.038 $\mu$m s⁻¹ was observed for the minor axis and 0.50 ± 0.23 $\mu$m s⁻¹ for the major axis (Fig. ​(Fig.4),4), whereas the average linear growth rates at $-{7.2}^{\circ }$C were 0.17 ± 0.050 $\mu$m s⁻¹ for the minor axis and 0.28 ± 0.087 $\mu$m s⁻¹ for the major axis (Fig. S4). While columnar growth dominated the seeding missions at $-{5.2}^{\circ }$C, a simultaneous presence of columnar and plate-like growth patterns (resulting in either short columns or plates) was observed for the missions at $-{7.2}^{\circ }$C, in line with previous studies (6, 29, 41). Interestingly, no significant difference was observed for the average growth rates measured during S5-6.6 and S5-9.3 (Fig. ​(Fig.4),4), as indicated by a Kolmogorov–Smirnov test ($P\ge$ 0.05). Moreover, the growth rates computed from the difference between consecutive seeding missions (i.e. S5-2.7, see Methods section) was in good agreement with the respective seeding missions. These findings suggest that at $-{5.2}^{\circ }$C, the observed ice crystal populations grew approximately linearly over time (i.e. constant growth rate) during the time period of 6.6 to 9.2 min. This also aligns with the laboratory study of Knight (40), who recorded long time series, lasting up to several hours, of individually grown ice crystals at $-5^{\circ }$C and found nearly constant growth rates over time for needles.

Open in a separate window

Fig. 4.

Linear growth rates of ice crystals along the minor and major axes are shown for seeding missions S5-6.6 (purple) and S5-9.3 (green). The left halves of the violin plots illustrate the mean distribution including all data, while the right halves of the violin plots show the mean distribution within stable mixed-phase regions (≤ 50 L⁻¹, light color) and rapid glaciating regions (≥ 500 L⁻¹, dark color), with the ice crystal number concentration averaged over a 1 s interval. The colored numbers represent the number of time steps included in the respective half violin plots. Black markers denote the linear growth rates reported in the specified studies (Ryan et al. (28): $T=-{5.0}^{\circ }$C, $t=2.5$ min; Takahashi et al. (29): $T=-{5.3}^{\circ }$C, $t=5$ min; Castellano et al. (39): $\text{T}=-{6.5}^{\circ }$C, $\text{t}=4.5$ min; Knight (40): $\text{T}=-{5.0}^{\circ }$C, $t=40$ min). Additionally, the linear growth rate computed from the difference between the seeding missions (S5-2.7) is shown as a black square (see Methods section). A Kolmogorov–Smirnov test was conducted at a significance level of 0.05 to evaluate whether the distribution of growth rates measured during S5-6.6 and S5-9.3 were significantly different; a P-value ≤ 0.05 indicates a significant difference. The test was performed on 500 independent subsamples, each with a sample size of 100. The mean P-values across these subsamples are reported, with rejection rates of 7% for the minor axis and 3% for the major axis.

The average linear growth rates observed in Figs. ​Figs.44 and S4 were notably lower compared to the linear growth rates documented in earlier laboratory studies (28, 29, 39) and showed a substantial variability. Nevertheless, the largest measured ice crystals exhibited comparable or even higher growth rates than those reported in previous studies. Considering the high ice crystal number concentrations measured within the seeding plume (Figs. ​(Figs.2c2c and S1C), the variability in the observed linear growth rates and the lower averages compared to earlier studies can partly be attributed to the competition among the ice crystals for the limited available water vapor. Indeed, a depletion of the liquid phase was observed during the seeding missions, particularly in time periods with high ice crystal number concentrations, with certain cloud regions experiencing complete glaciation (Figs. ​(Figs.2c,d2c,d and S1C,D). To account for that, we investigated the growth rates of ice crystal populations for different mixed-phase conditions, specifically exploring the growth rates within stable mixed-phase regions and rapid glaciating regions. Based on the mean ice crystal number concentrations and glaciation times reported in Korolev et al. (42), our data were divided into two regimes using an ice crystal concentration threshold of below 50 L⁻¹ to define “stable mixed-phase regions” (i.e. vapor unlimited regime) and above 500 L⁻¹ to define “rapid glaciating regions” (i.e. vapor-limited regime). The separation into these regions shows that at $-{5.2}^{\circ }$C, the growth rates along the major axis are significantly lower in the rapid glaciating regions (0.38 ± 0.18 $\mu$m s⁻¹) compared to the stable mixed-phase regions (0.53 ± 0.24 $\mu$m s⁻¹), as indicated by the Kolmogorov–Smirnov test ($P\le 0.05$). This suggests that the number concentration of ice crystals and thus the competition for the limited available water vapor can significantly reduce the supersaturation and ice crystal growth rates. Indeed, direct numerical simulations conducted by Chen et al. (43) confirm that fluctuations in the small-scale supersaturation field in the immediate surrounding of ice crystals broaden the ice crystal size distribution, particularly in environments close to ice saturation.

To investigate the impact of the supersaturation on the growth rates of individual ice crystals, we used the aspect ratio at $-{5.2}^{\circ }$C as a proxy for the supersaturation, where a high aspect ratio is assumed to be an indicator for high supersaturation experienced by the ice crystals during growth. While fluctuations in the supersaturation do not change the basic habit of ice crystals (e.g. column vs. plate), it influences how fast the preferred axis grows (15). Indeed, as depicted in Fig. ​Fig.5b,5b, ice crystals with a larger aspect ratio were observed at $-{5.2}^{\circ }$C when lower ice crystal number concentrations and higher cloud droplet number concentrations were present. These ice crystals likely experienced higher supersaturation during growth, consistent with previous laboratory studies (44, 45) that observed faster ice growth in the presence of higher liquid water content. This accelerated growth was attributed to the repeated close proximity of supercooled cloud droplets, which enhanced rapid vapor transport from the cloud droplets to the ice crystals ((44), “vapor flush” effect). As a result, the observed growth rates (1.03 ± 0.16 $\mu$m s⁻¹, Fig. ​Fig.5c)5c) align more closely or even exceed those reported in previous laboratory studies and can be considered representative of conditions in which ice crystal growth is not vapor-limited or in which cloud droplets and ice crystals are uniformly distributed within a cloud volume ((23), genuinely mixed). On the other hand, when considering ice crystals with lower aspect ratio, the ice crystal growth rates were significantly lower (0.24 ± 0.090 $\mu$m s⁻¹) and likely limited by water vapor. This case can be important in regions with high ice crystal number concentrations, such as those with strong secondary ice production (46, 47) or in mixed-phase regions in which cloud droplets and ice crystals are spatially separated ((23), conditionally mixed).

Open in a separate window

Fig. 5.

Linear growth rates of ice crystals along the major axis are shown for seeding missions S5-6.6 (purple) and S5-9.3 (green) as a function of aspect ratio. a) The histograms show the number of counts in the respective aspect ratio bin. b) Mean cloud droplet number concentration (CDNC, solid gray lines) and mean ice crystal number concentration (ICNC, dashed blue lines) measured within each aspect ratio bin. c) The boxplots show the linear growth rate along the major axis. As depicted by the black arrow, the aspect ratio is used as a proxy for supersaturation experienced during ice growth, where a low aspect ratio corresponds to low supersaturation (rapid glaciating region), while a high aspect ratio corresponds to high supersaturation (stable mixed-phase region). It is important to note that other factors (e.g. temperature fluctuations) can also influence the aspect ratio.

The observed variability in ice crystal growth rates in natural clouds, and consequently in ice crystal sizes and fall speeds, can potentially accelerate the onset of precipitation initiation through more efficient aggregation (48) and secondary ice production from ice–ice collisions (49). Quantifying the variability in ice crystal growth rates and identifying the conditions that influence them has significant implications for predicting the amount and spatial distribution of precipitation, particularly in shallow clouds and in regions with complex terrain (50, 51).

Conclusion

In this study, we repurposed weather modification for fundamental cloud research by conducting repeated glaciogenic cloud seeding experiments in persistent supercooled stratus clouds to infer ice crystal growth rates in natural clouds. Seeding material was released from a multirotor UAV and the seeding-induced microphysical changes were measured 4–10 min downwind by ground-based remote sensing and in situ instrumentation on a tethered balloon system. Seeding signatures were detected by increased radar reflectivities, along with elevated ice crystal number concentrations and a simultaneous reduction in the cloud droplet number concentration up to complete glaciation in single cloud patches. This emphasizes the efficiency of the WBF process in producing precipitation-sized particles ($\ge 100\mu$m) within a few minutes, highlighting its importance for precipitation initiation.

The seeding distance and consequently the ice growth time were systematically adjusted between consecutive seeding missions, while the environmental conditions remained approximately constant, enabling comprehensive ice growth studies within a quasi-steady-state natural environment. At $-{5.2}^{\circ }$C, the seeding experiments revealed columnar growth with an average growth rate of 0.50 ± 0.23 $\mu$m s⁻¹ along the major axis, while a transition to short columns and plates was observed at $-{7.2}^{\circ }$C, characterized by an average growth rate of 0.28 ± 0.087 $\mu$m s⁻¹. Our findings indicate that the ice crystal populations in natural clouds grew nearly linearly over time at $-{5.2}^{\circ }$C within the observed time scales (6–10 min). Further investigations over a broader range of time and ambient conditions are required to thoroughly characterize the temporal (in)dependence of growth rates. In-depth knowledge of ice growth processes and the ways they are represented in numerical weather prediction models directly impacts the precipitation forecast skill (e.g. extreme precipitation).

Our observations show large variability in ice crystal growth rates within natural clouds, underscoring their sensitivity to both ice crystal number concentrations and supersaturation. Specifically, in rapid glaciating regions characterized by high ice crystal number concentrations, the linear growth rates at $-{5.2}^{\circ }$C (0.38 ± 0.18 $\mu$m s⁻¹) were significantly lower compared to those in water vapor unlimited regions (0.53 ± 0.24 $\mu$m s⁻¹), as the ice crystals in rapid glaciating regions likely experienced lower supersaturation in their immediate vicinity during growth. This variability in ice crystal growth rates potentially accelerates precipitation initiation, with significant implications for both precipitation forecast and climate simulations. Thus, we recommend that future research focuses on investigating ice crystal growth rates under conditions of fluctuating supersaturation and/or temperature, using well-designed field experiments or innovative laboratory approaches.

This study provides unambiguous evidence that the targeted use of weather modification techniques can be employed to study microphysical processes in natural clouds, specifically enabling the investigation of ice crystal growth rates at temperatures above $-{10}^{\circ }$C—a pursuit previously limited in field studies due to the scarcity of naturally occurring ice nucleating particles at these high temperatures. Future work aims to extend this analysis to different environmental and microphysical conditions, leveraging these observations to constrain and validate model simulations. In particular, these comprehensive observations can help to develop more realistic parameterizations for weather and climate models, extending beyond what is achievable in a laboratory setting. Overall, this study shows the versatile use of weather modification, not only for precipitation enhancement and climate intervention but also for advancing our understanding of cloud processes (e.g. ice growth processes, aerosol-cloud interactions), validating remote sensing retrievals, and refining model parameterizations.

Materials and methods

Supplementary Material

Contributor Information

Fabiola Ramelli, Department of Environmental Systems Science, Institute for Atmospheric and Climate Science, ETH Zurich, Zurich 8092, Switzerland.

Jan Henneberger, Department of Environmental Systems Science, Institute for Atmospheric and Climate Science, ETH Zurich, Zurich 8092, Switzerland.

Christopher Fuchs, Department of Environmental Systems Science, Institute for Atmospheric and Climate Science, ETH Zurich, Zurich 8092, Switzerland.

Anna J Miller, Department of Environmental Systems Science, Institute for Atmospheric and Climate Science, ETH Zurich, Zurich 8092, Switzerland.

Nadja Omanovic, Department of Environmental Systems Science, Institute for Atmospheric and Climate Science, ETH Zurich, Zurich 8092, Switzerland.

Robert Spirig, Department of Environmental Systems Science, Institute for Atmospheric and Climate Science, ETH Zurich, Zurich 8092, Switzerland.

Huiying Zhang, Department of Environmental Systems Science, Institute for Atmospheric and Climate Science, ETH Zurich, Zurich 8092, Switzerland.

Robert O David, Department of Geosciences, Section for Meteorology and Oceanography, University of Oslo, Oslo 0316, Norway.

Kevin Ohneiser, Department Remote Sensing of Atmospheric Processes, Leibniz Institute for Tropospheric Research (TROPOS), Leipzig 04318, Germany.

Patric Seifert, Department Remote Sensing of Atmospheric Processes, Leibniz Institute for Tropospheric Research (TROPOS), Leipzig 04318, Germany.

Ulrike Lohmann, Department of Environmental Systems Science, Institute for Atmospheric and Climate Science, ETH Zurich, Zurich 8092, Switzerland.

References