The formation of secondary ice in clouds, that is, ice particles that are created at temperatures above the limit for homogeneous freezing without the direct involvement of a heterogeneous ice nucleus, is one of the longest-standing puzzles in cloud physics. Here, we present comprehensive laboratory investigations on the formation of small ice particles upon the freezing of drizzle-sized cloud droplets levitated in an electrodynamic balance. Four different categories of secondary ice formation (bubble bursting, jetting, cracking, and breakup) could be detected, and their respective frequencies of occurrence as a function of temperature and droplet size are given. We find that bubble bursting occurs more often than droplet splitting. While we do not observe the shattering of droplets into many large fragments, we find that the average number of small secondary ice particles released during freezing is strongly dependent on droplet size and may well exceed unity for droplets larger than 300 μm in diameter. This leaves droplet fragmentation as an important secondary ice process effective at temperatures around −10°C in clouds where large drizzle droplets are present.

Abstract. An accurate prediction of the ice crystal number concentration in clouds is important to determine the radiation budget, the lifetime, and the precipitation formation of clouds. Secondary-ice production is thought to be responsible for the observed discrepancies between the ice crystal number concentration and the ice-nucleating particle concentration in clouds. The Hallett–Mossop process is active between −3 and −8 ∘C and has been implemented into several models, while all other secondary-ice processes are poorly constrained and lack a well-founded quantification. During 2 h of measurements taken on a mountain slope just above the melting layer at temperatures warmer than −3 ∘C, a continuously high concentration of small plates identified as secondary ice was observed. The presence of drizzle drops suggests droplet fragmentation upon freezing as the responsible secondary-ice mechanism. The constant supply of drizzle drops can be explained by a recirculation theory, suggesting that melted snowflakes, which sedimented through the melting layer, were reintroduced into the cloud as drizzle drops by orographically forced updrafts. Here we introduce a parametrization of droplet fragmentation at slightly sub-zero temperatures, where primary-ice nucleation is basically absent, and the first ice is initiated by the collision of drizzle drops with aged ice crystals sedimenting from higher altitudes. Based on previous measurements, we estimate that a droplet of 200 µm in diameter produces 18 secondary-ice crystals when it fragments upon freezing. The application of the parametrization to our measurements suggests that the actual number of splinters produced by a fragmenting droplet may be up to an order of magnitude higher.

Abstract. The seeder–feeder mechanism has been observed to enhance orographic precipitation in previous studies. However, the microphysical processes active in the seeder and feeder region are still being understood. In this paper, we investigate the seeder and feeder region of a mixed-phase cloud passing over the Swiss Alps, focusing on (1) fallstreaks of enhanced radar reflectivity originating from cloud top generating cells (seeder region) and (2) a persistent low-level feeder cloud produced by the boundary layer circulation (feeder region). Observations were obtained from a multi-dimensional set of instruments including ground-based remote sensing instrumentation (Ka-band polarimetric cloud radar, microwave radiometer, wind profiler), in situ instrumentation on a tethered balloon system, and ground-based aerosol and precipitation measurements. The cloud radar observations suggest that ice formation and growth were enhanced within cloud top generating cells, which is consistent with previous observational studies. However, uncertainties exist regarding the dominant ice formation mechanism within these cells. Here we propose different mechanisms that potentially enhance ice nucleation and growth in cloud top generating cells (convective overshooting, radiative cooling, droplet shattering) and attempt to estimate their potential contribution from an ice nucleating particle perspective. Once ice formation and growth within the seeder region exceeded a threshold value, the mixed-phase cloud became fully glaciated. Local flow effects on the lee side of the mountain barrier induced the formation of a persistent low-level feeder cloud over a small-scale topographic feature in the inner-Alpine valley. In situ measurements within the low-level feeder cloud observed the production of secondary ice particles likely due to the Hallett–Mossop process and ice particle fragmentation upon ice–ice collisions. Therefore, secondary ice production may have been partly responsible for the elevated ice crystal number concentrations that have been previously observed in feeder clouds at mountaintop observatories. Secondary ice production in feeder clouds can potentially enhance orographic precipitation.

Abstract. During typical field campaigns, millions of cloud particle images are captured with imaging probes. Our interest lies in classifying these particles in order to compute the statistics needed for understanding clouds. Given the large volume of collected data, this raises the need for an automated classification approach. Traditional classification methods that require extracting features manually (e.g., decision trees and support vector machines) show reasonable performance when trained and tested on data coming from a unique dataset. However, they often have difficulties in generalizing to test sets coming from other datasets where the distribution of the features might be significantly different. In practice, we found that for holographic imagers each new dataset requires labeling a huge amount of data by hand using those methods. Convolutional neural networks have the potential to overcome this problem due to their ability to learn complex nonlinear models directly from the images instead of pre-engineered features, as well as by relying on powerful regularization techniques. We show empirically that a convolutional neural network trained on cloud particles from holographic imagers generalizes well to unseen datasets. Moreover, fine tuning the same network with a small number (256) of training images improves the classification accuracy. Thus, the automated classification with a convolutional neural network not only reduces the hand-labeling effort for new datasets but is also no longer the main error source for the classification of small particles.

Abstract. The discrepancy between the observed concentration of ice nucleating particles (INPs) and the ice crystal number concentration (ICNC) remains unresolved and limits our understanding of ice formation and, hence, precipitation amount, location and intensity. Enhanced ice formation through secondary ice production (SIP) could account for this discrepancy. Here, in a region over the eastern Swiss Alps, we perform sensitivity studies of additional simulated SIP processes on precipitation formation and surface precipitation intensity. The SIP processes considered include rime splintering, droplet shattering during freezing and breakup through ice–graupel collisions. We simulated the passage of a cold front at Gotschnagrat, a peak at 2281 m a.s.l. (above sea level), on 7 March 2019 with the Consortium for Small-scale Modeling (COSMO), at a 1 km horizontal grid spacing, as part of the RACLETS (Role of Aerosols and CLouds Enhanced by Topography and Snow) field campaign in the Davos region in Switzerland. The largest simulated difference in the ICNC at the surface originated from the breakup simulations. Indeed, breakup caused a 1 to 3 orders of magnitude increase in the ICNC compared to SIP from rime splintering or without SIP processes in the control simulation. The ICNCs from the collisional breakup simulations at Gotschnagrat were in best agreement with the ICNCs measured on a gondola near the surface. However, these simulations were not able to reproduce the ice crystal habits near the surface. Enhanced ICNCs from collisional breakup reduced localized regions of higher precipitation and, thereby, improved the model performance in terms of surface precipitation over the domain.