Articles | Volume 18, issue 12
https://doi.org/10.5194/amt-18-2781-2025
© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
https://doi.org/10.5194/amt-18-2781-2025
© Author(s) 2025. This work is distributed under
the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Research article
| 
27 Jun 2025
Research article |
| 27 Jun 2025
| 27 Jun 2025
Exploring the effect of training set size and number of categories on ice crystal classification through a contrastive semi-supervised learning algorithm
Yunpei Chu
Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland
Department of Geosciences and Remote Sensing, Faculty of Civil Engineering and Geosciences, Delft University of Technology, Delft, the Netherlands
Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland
Xia Li
Institute for Machine Learning, ETH Zurich, Zurich, Switzerland
Jan Henneberger
CORRESPONDING AUTHOR
Institute for Atmospheric and Climate Science, ETH Zurich, Zurich, Switzerland
Related authors
No articles found.
Christopher Fuchs, Fabiola Ramelli, David Schweizer, Ulrike Lohmann, and Jan Henneberger
Atmos. Meas. Tech., 18, 2969–2986, https://doi.org/10.5194/amt-18-2969-2025, https://doi.org/10.5194/amt-18-2969-2025, 2025
Short summary
Short summary
We present a new instrument based on digital in-line holography (SmHOLIMO) for in situ cloud measurements. SmHOLIMO is designed to specifically measure small cloud droplets with diameters > 3.7 μm. This way we retrieve accurate cloud droplet size distributions, which are crucial to understand the evolution and governing microphysical processes of a cloud. Results of a field study are compared to co-located measurements of a second holographic imager, microwave radiometer, and cloud radar.
Kevin Ohneiser, Patric Seifert, Willi Schimmel, Fabian Senf, Tom Gaudek, Martin Radenz, Audrey Teisseire, Veronika Ettrichrätz, Teresa Vogl, Nina Maherndl, Nils Pfeifer, Jan Henneberger, Anna J. Miller, Nadja Omanovic, Christopher Fuchs, Huiying Zhang, Fabiola Ramelli, Robert Spirig, Anton Kötsche, Heike Kalesse-Los, Maximilian Maahn, Heather Corden, Alexis Berne, Majid Hajipour, Hannes Griesche, Julian Hofer, Ronny Engelmann, Annett Skupin, Albert Ansmann, and Holger Baars
EGUsphere, https://doi.org/10.5194/egusphere-2025-2482, https://doi.org/10.5194/egusphere-2025-2482, 2025
Short summary
Short summary
This study focuses on a seeder-feeder cloud system on 8 Jan 2024 in Eriswil, Switzerland. It is shown how the interaction of these cloud systems changes the cloud microphysical properties and the precipitation patterns. A big set of advanced remote-sensing techniques and retrieval algorithms are applied, so that a detailed view on the seeder-feeder cloud system is available. The gained knowledge can be used to improve weather models and weather forecasts.
Anna J. Miller, Christopher Fuchs, Fabiola Ramelli, Huiying Zhang, Nadja Omanovic, Robert Spirig, Claudia Marcolli, Zamin A. Kanji, Ulrike Lohmann, and Jan Henneberger
Atmos. Chem. Phys., 25, 5387–5407, https://doi.org/10.5194/acp-25-5387-2025, https://doi.org/10.5194/acp-25-5387-2025, 2025
Short summary
Short summary
We analyzed the ability of silver iodide particles (a commonly used cloud-seeding agent) to form ice crystals in naturally occurring liquid clouds at −5 to −8 °C and found that only ≈ 0.1 %−1 % of particles nucleate ice, with a negative dependence on temperature. By contextualizing our results with previous laboratory studies, we help to bridge the gap between laboratory and field experiments, which also helps to inform future cloud-seeding projects.
Christopher Fuchs, Fabiola Ramelli, Anna J. Miller, Nadja Omanovic, Robert Spirig, Huiying Zhang, Patric Seifert, Kevin Ohneiser, Ulrike Lohmann, and Jan Henneberger
EGUsphere, https://doi.org/10.5194/egusphere-2025-688, https://doi.org/10.5194/egusphere-2025-688, 2025
Short summary
Short summary
We quantify diffusional ice crystal growth in natural clouds using cloud seeding experiments. We report growth rates for 14 experiments between -5.1°C and -8.3°C and observe strong variations depending on the cloud characteristics. Comparing our growth rates to laboratory data, we found similar temperature-dependent trends, but the laboratory rates are higher. This data fills the gap in quantitative in situ observation of ice crystal growth, helping to validate models and laboratory experiments.
Huiying Zhang, Xia Li, Fabiola Ramelli, Robert O. David, Julie Pasquier, and Jan Henneberger
Atmos. Meas. Tech., 17, 7109–7128, https://doi.org/10.5194/amt-17-7109-2024, https://doi.org/10.5194/amt-17-7109-2024, 2024
Short summary
Short summary
Our innovative IceDetectNet algorithm classifies each part of aggregated ice crystals, considering both their basic shape and physical processes. Trained on ice crystal images from the Arctic taken by a holographic camera, it correctly classifies over 92 % of the ice crystals. These more detailed insights into the components of aggregated ice crystals have the potential to improve our estimates of microphysical properties such as riming rate, aggregation rate, and ice water content.
Nadja Omanovic, Sylvaine Ferrachat, Christopher Fuchs, Jan Henneberger, Anna J. Miller, Kevin Ohneiser, Fabiola Ramelli, Patric Seifert, Robert Spirig, Huiying Zhang, and Ulrike Lohmann
Atmos. Chem. Phys., 24, 6825–6844, https://doi.org/10.5194/acp-24-6825-2024, https://doi.org/10.5194/acp-24-6825-2024, 2024
Short summary
Short summary
We present simulations with a high-resolution numerical weather prediction model to study the growth of ice crystals in low clouds following glaciogenic seeding. We show that the simulated ice crystals grow slower than observed and do not consume as many cloud droplets as measured in the field. This may have implications for forecasting precipitation, as the ice phase is crucial for precipitation at middle and high latitudes.
Anna J. Miller, Fabiola Ramelli, Christopher Fuchs, Nadja Omanovic, Robert Spirig, Huiying Zhang, Ulrike Lohmann, Zamin A. Kanji, and Jan Henneberger
Atmos. Meas. Tech., 17, 601–625, https://doi.org/10.5194/amt-17-601-2024, https://doi.org/10.5194/amt-17-601-2024, 2024
Short summary
Short summary
We present a method for aerosol and cloud research using two uncrewed aerial vehicles (UAVs). The UAVs have a propeller heating mechanism that allows flights in icing conditions, which has so far been a limitation for cloud research with UAVs. One UAV burns seeding flares, producing a plume of particles that causes ice formation in supercooled clouds. The second UAV measures aerosol size distributions and is used for measuring the seeding plume or for characterizing the boundary layer.
Ghislain Motos, Gabriel Freitas, Paraskevi Georgakaki, Jörg Wieder, Guangyu Li, Wenche Aas, Chris Lunder, Radovan Krejci, Julie Thérèse Pasquier, Jan Henneberger, Robert Oscar David, Christoph Ritter, Claudia Mohr, Paul Zieger, and Athanasios Nenes
Atmos. Chem. Phys., 23, 13941–13956, https://doi.org/10.5194/acp-23-13941-2023, https://doi.org/10.5194/acp-23-13941-2023, 2023
Short summary
Short summary
Low-altitude clouds play a key role in regulating the climate of the Arctic, a region that suffers from climate change more than any other on the planet. We gathered meteorological and aerosol physical and chemical data over a year and utilized them for a parameterization that help us unravel the factors driving and limiting the efficiency of cloud droplet formation. We then linked this information to the sources of aerosol found during each season and to processes of cloud glaciation.
Guangyu Li, Elise K. Wilbourn, Zezhen Cheng, Jörg Wieder, Allison Fagerson, Jan Henneberger, Ghislain Motos, Rita Traversi, Sarah D. Brooks, Mauro Mazzola, Swarup China, Athanasios Nenes, Ulrike Lohmann, Naruki Hiranuma, and Zamin A. Kanji
Atmos. Chem. Phys., 23, 10489–10516, https://doi.org/10.5194/acp-23-10489-2023, https://doi.org/10.5194/acp-23-10489-2023, 2023
Short summary
Short summary
In this work, we present results from an Arctic field campaign (NASCENT) in Ny-Ålesund, Svalbard, on the abundance, variability, physicochemical properties, and potential sources of ice-nucleating particles (INPs) relevant for mixed-phase cloud formation. This work improves the data coverage of Arctic INPs and aerosol properties, allowing for the validation of models predicting cloud microphysical and radiative properties of mixed-phase clouds in the rapidly warming Arctic.
Julie Thérèse Pasquier, Jan Henneberger, Fabiola Ramelli, Annika Lauber, Robert Oscar David, Jörg Wieder, Tim Carlsen, Rosa Gierens, Marion Maturilli, and Ulrike Lohmann
Atmos. Chem. Phys., 22, 15579–15601, https://doi.org/10.5194/acp-22-15579-2022, https://doi.org/10.5194/acp-22-15579-2022, 2022
Short summary
Short summary
It is important to understand how ice crystals and cloud droplets form in clouds, as their concentrations and sizes determine the exact radiative properties of the clouds. Normally, ice crystals form from aerosols, but we found evidence for the formation of additional ice crystals from the original ones over a large temperature range within Arctic clouds. In particular, additional ice crystals were formed during collisions of several ice crystals or during the freezing of large cloud droplets.
Guangyu Li, Jörg Wieder, Julie T. Pasquier, Jan Henneberger, and Zamin A. Kanji
Atmos. Chem. Phys., 22, 14441–14454, https://doi.org/10.5194/acp-22-14441-2022, https://doi.org/10.5194/acp-22-14441-2022, 2022
Short summary
Short summary
The concentration of ice-nucleating particles (INPs) is atmospherically relevant for primary ice formation in clouds. In this work, from 12 weeks of field measurement data in the Arctic, we developed a new parameterization to predict INP concentrations applicable for pristine background conditions based only on temperature. The INP parameterization could improve the cloud microphysical representation in climate models, aiding in Arctic climate predictions.
Jörg Wieder, Nikola Ihn, Claudia Mignani, Moritz Haarig, Johannes Bühl, Patric Seifert, Ronny Engelmann, Fabiola Ramelli, Zamin A. Kanji, Ulrike Lohmann, and Jan Henneberger
Atmos. Chem. Phys., 22, 9767–9797, https://doi.org/10.5194/acp-22-9767-2022, https://doi.org/10.5194/acp-22-9767-2022, 2022
Short summary
Short summary
Ice formation and its evolution in mixed-phase clouds are still uncertain. We evaluate the lidar retrieval of ice-nucleating particle concentration in dust-dominated and continental air masses over the Swiss Alps with in situ observations. A calibration factor to improve the retrieval from continental air masses is proposed. Ice multiplication factors are obtained with a new method utilizing remote sensing. Our results indicate that secondary ice production occurs at temperatures down to −30 °C.
Jörg Wieder, Claudia Mignani, Mario Schär, Lucie Roth, Michael Sprenger, Jan Henneberger, Ulrike Lohmann, Cyril Brunner, and Zamin A. Kanji
Atmos. Chem. Phys., 22, 3111–3130, https://doi.org/10.5194/acp-22-3111-2022, https://doi.org/10.5194/acp-22-3111-2022, 2022
Short summary
Short summary
We investigate the variation in ice-nucleating particles (INPs) relevant for primary ice formation in mixed-phased clouds over the Alps based on simultaneous in situ observations at a mountaintop and a nearby high valley (1060 m height difference). In most cases, advection from the surrounding lower regions was responsible for changes in INP concentration, causing a diurnal cycle at the mountaintop. Our study underlines the importance of the planetary boundary layer as an INP reserve.
Paraskevi Georgakaki, Aikaterini Bougiatioti, Jörg Wieder, Claudia Mignani, Fabiola Ramelli, Zamin A. Kanji, Jan Henneberger, Maxime Hervo, Alexis Berne, Ulrike Lohmann, and Athanasios Nenes
Atmos. Chem. Phys., 21, 10993–11012, https://doi.org/10.5194/acp-21-10993-2021, https://doi.org/10.5194/acp-21-10993-2021, 2021
Short summary
Short summary
Aerosol and cloud observations coupled with a droplet activation parameterization was used to investigate the aerosol–cloud droplet link in alpine mixed-phase clouds. Predicted droplet number, Nd, agrees with observations and never exceeds a characteristic “limiting droplet number”, Ndlim, which depends solely on σw. Nd becomes velocity limited when it is within 50 % of Ndlim. Identifying when dynamical changes control Nd variability is central for understanding aerosol–cloud interactions.
Fabiola Ramelli, Jan Henneberger, Robert O. David, Johannes Bühl, Martin Radenz, Patric Seifert, Jörg Wieder, Annika Lauber, Julie T. Pasquier, Ronny Engelmann, Claudia Mignani, Maxime Hervo, and Ulrike Lohmann
Atmos. Chem. Phys., 21, 6681–6706, https://doi.org/10.5194/acp-21-6681-2021, https://doi.org/10.5194/acp-21-6681-2021, 2021
Short summary
Short summary
Orographic mixed-phase clouds are an important source of precipitation, but the ice formation processes within them remain uncertain. Here we investigate the origin of ice crystals in a mixed-phase cloud in the Swiss Alps using aerosol and cloud data from in situ and remote sensing observations. We found that ice formation primarily occurs in cloud top generating cells. Our results indicate that secondary ice processes are active in the feeder region, which can enhance orographic precipitation.
Fabiola Ramelli, Jan Henneberger, Robert O. David, Annika Lauber, Julie T. Pasquier, Jörg Wieder, Johannes Bühl, Patric Seifert, Ronny Engelmann, Maxime Hervo, and Ulrike Lohmann
Atmos. Chem. Phys., 21, 5151–5172, https://doi.org/10.5194/acp-21-5151-2021, https://doi.org/10.5194/acp-21-5151-2021, 2021
Short summary
Short summary
Interactions between dynamics, microphysics and orography can enhance precipitation. Yet the exact role of these interactions is still uncertain. Here we investigate the role of low-level blocking and turbulence for precipitation by combining remote sensing and in situ observations. The observations show that blocked flow can induce the formation of feeder clouds and that turbulence can enhance hydrometeor growth, demonstrating the importance of local flow effects for orographic precipitation.
Annika Lauber, Jan Henneberger, Claudia Mignani, Fabiola Ramelli, Julie T. Pasquier, Jörg Wieder, Maxime Hervo, and Ulrike Lohmann
Atmos. Chem. Phys., 21, 3855–3870, https://doi.org/10.5194/acp-21-3855-2021, https://doi.org/10.5194/acp-21-3855-2021, 2021
Short summary
Short summary
An accurate prediction of the ice crystal number concentration (ICNC) is important to determine the radiation budget, lifetime, and precipitation formation of clouds. Even though secondary-ice processes can increase the ICNC by several orders of magnitude, they are poorly constrained and lack a well-founded quantification. During measurements on a mountain slope, a high ICNC of small ice crystals was observed just below 0 °C, attributed to a secondary-ice process and parametrized in this study.
Claudia Mignani, Jörg Wieder, Michael A. Sprenger, Zamin A. Kanji, Jan Henneberger, Christine Alewell, and Franz Conen
Atmos. Chem. Phys., 21, 657–664, https://doi.org/10.5194/acp-21-657-2021, https://doi.org/10.5194/acp-21-657-2021, 2021
Short summary
Short summary
Most precipitation above land starts with ice in clouds. It is promoted by extremely rare particles. Some ice-nucleating particles (INPs) cause cloud droplets to already freeze above −15°C, a temperature at which many clouds begin to snow. We found that the abundance of such INPs among other particles of similar size is highest in precipitating air masses and lowest when air carries desert dust. This brings us closer to understanding the interactions between land, clouds, and precipitation.
Georgios Touloupas, Annika Lauber, Jan Henneberger, Alexander Beck, and Aurélien Lucchi
Atmos. Meas. Tech., 13, 2219–2239, https://doi.org/10.5194/amt-13-2219-2020, https://doi.org/10.5194/amt-13-2219-2020, 2020
Short summary
Short summary
Images of cloud particles give important information for improving our understanding of microphysical cloud processes. For phase-resolved measurements, a large number of water droplets and ice crystals need to be classified by an automated approach. In this study, a convolutional neural network was designed, which exceeds the classification ability of traditional methods and therefore shortens the analysis procedure of cloud particle images.
Fabiola Ramelli, Alexander Beck, Jan Henneberger, and Ulrike Lohmann
Atmos. Meas. Tech., 13, 925–939, https://doi.org/10.5194/amt-13-925-2020, https://doi.org/10.5194/amt-13-925-2020, 2020
Short summary
Short summary
Boundary layer clouds are influenced by many physical and dynamical processes, making accurate forecasting difficult. Here we present a new measurement platform on a tethered balloon to measure cloud microphysical and meteorological profiles. The unique combination of holography and balloon-borne observations allows high-resolution measurements in a well-defined volume. Field measurements in stratus clouds over the Swiss Plateau revealed unique microphysical signatures in the cloud structure.
Alexander Beck, Jan Henneberger, Jacob P. Fugal, Robert O. David, Larissa Lacher, and Ulrike Lohmann
Atmos. Chem. Phys., 18, 8909–8927, https://doi.org/10.5194/acp-18-8909-2018, https://doi.org/10.5194/acp-18-8909-2018, 2018
Short summary
Short summary
This study assesses the impact of surface processes (e.g. blowing snow) on in situ cloud observations at Sonnblick Observatory. Vertical profiles of ice crystal number concentrations (ICNCs) above a snow-covered surface were observed up to a height of 10 m. The ICNC near the ground is at least a factor of 2 larger than at 10 m. Therefore, in situ measurements of ICNCs at mountain-top research stations close to the surface are strongly influenced by surface processes and overestimate the ICNC.
Alexander Beck, Jan Henneberger, Sarah Schöpfer, Jacob Fugal, and Ulrike Lohmann
Atmos. Meas. Tech., 10, 459–476, https://doi.org/10.5194/amt-10-459-2017, https://doi.org/10.5194/amt-10-459-2017, 2017
Short summary
Short summary
In situ observations of cloud properties in complex alpine terrain are commonly conducted at mountain-top research stations and limited to single-point measurements. The HoloGondel platform overcomes this limitation by using a cable car to obtain vertical profiles of the microphysical and meteorological cloud parameters. In this work example measurements of the vertical profiles observed in a liquid cloud and a mixed-phase cloud at the Eggishorn in the Swiss Alps are presented.
J. Grazioli, G. Lloyd, L. Panziera, C. R. Hoyle, P. J. Connolly, J. Henneberger, and A. Berne
Atmos. Chem. Phys., 15, 13787–13802, https://doi.org/10.5194/acp-15-13787-2015, https://doi.org/10.5194/acp-15-13787-2015, 2015
Short summary
Short summary
This study investigates the microphysics of winter alpine snowfall occurring in mixed-phase clouds in an inner-Alpine valley during CLACE2014. From polarimetric radar and in situ observations, riming is shown to be an important process leading to more intense snowfall. Riming is usually associated with more intense turbulence providing supercooled liquid water. Distinct features are identified in the vertical structure of polarimetric radar variables.
G. Lloyd, T. W. Choularton, K. N. Bower, M. W. Gallagher, P. J. Connolly, M. Flynn, R. Farrington, J. Crosier, O. Schlenczek, J. Fugal, and J. Henneberger
Atmos. Chem. Phys., 15, 12953–12969, https://doi.org/10.5194/acp-15-12953-2015, https://doi.org/10.5194/acp-15-12953-2015, 2015
Short summary
Short summary
The paper explores the microphysical structure of clouds at the high-alpine measurement site Jungfraujoch, Switzerland. High concentrations of ice crystals were measured by a range of instruments. The presence of these high concentrations could not be explained through conventional understanding of ice formation processes in clouds and the possibility that the surface provides a significant source of ice crystals is investigated.
J. Henneberger, J. P. Fugal, O. Stetzer, and U. Lohmann
Atmos. Meas. Tech., 6, 2975–2987, https://doi.org/10.5194/amt-6-2975-2013, https://doi.org/10.5194/amt-6-2975-2013, 2013
Related subject area
Subject: Clouds | Technique: In Situ Measurement | Topic: Data Processing and Information Retrieval
Convolutional neural networks for specific and merged data sets of optical array probe images: compatibility of retrieved morphology-dependent size distributions
An analysis of cloud microphysical features over United Arab Emirates using multiple data sources
IceDetectNet: a rotated object detection algorithm for classifying components of aggregated ice crystals with a multi-label classification scheme
Distribution characteristics of the summer precipitation raindrop spectrum on the Qinghai–Tibet Plateau
In situ observations of supercooled liquid water clouds over Dome C, Antarctica, by balloon-borne sondes
Partition between supercooled liquid droplets and ice crystals in mixed-phase clouds based on airborne in situ observations
Innovative cloud quantification: deep learning classification and finite-sector clustering for ground-based all-sky imaging
Revealing halos concealed by cirrus clouds
Quantifying riming from airborne data during the HALO-(AC)3 campaign
Estimation of 24 h continuous cloud cover using a ground-based imager with a convolutional neural network
Neural network processing of holographic images
Ice crystal images from optical array probes: classification with convolutional neural networks
Detection and analysis of cloud boundary in Xi'an, China, employing 35 GHz cloud radar aided by 1064 nm lidar
The University of Washington Ice–Liquid Discriminator (UWILD) improves single-particle phase classifications of hydrometeors within Southern Ocean clouds using machine learning
Twenty-four-hour cloud cover calculation using a ground-based imager with machine learning
Application of cloud particle sensor sondes for estimating the number concentration of cloud water droplets and liquid water content: case studies in the Arctic region
Clouds over Hyytiälä, Finland: an algorithm to classify clouds based on solar radiation and cloud base height measurements
A convolutional neural network for classifying cloud particles recorded by imaging probes
Spatiotemporal variability of solar radiation introduced by clouds over Arctic sea ice
Analysis algorithm for sky type and ice halo recognition in all-sky images
Study of the diffraction pattern of cloud particles and the respective responses of optical array probes
A method for computing the three-dimensional radial distribution function of cloud particles from holographic images
A new method for calculating number concentrations of cloud condensation nuclei based on measurements of a three-wavelength humidified nephelometer system
Cloud radiative effect, cloud fraction and cloud type at two stations in Switzerland using hemispherical sky cameras
Evaluation of radar reflectivity factor simulations of ice crystal populations from in situ observations for the retrieval of condensed water content in tropical mesoscale convective systems
Cloud detection in all-sky images via multi-scale neighborhood features and multiple supervised learning techniques
From pixels to patches: a cloud classification method based on a bag of micro-structures
Evaluation of cloud base height measurements from Ceilometer CL31 and MODIS satellite over Ahmedabad, India
Software to analyze the relationship between aerosol, clouds, and precipitation: SAMAC
Block-based cloud classification with statistical features and distribution of local texture features
Assessment of the performance of the inter-arrival time algorithm to identify ice shattering artifacts in cloud particle probe measurements
Assessment of cloud supersaturation by size-resolved aerosol particle and cloud condensation nuclei (CCN) measurements
Inversion of droplet aerosol analyzer data for long-term aerosol–cloud interaction measurements
Response of the Nevzorov hot wire probe in clouds dominated by droplet conditions in the drizzle size range
Louis Jaffeux, Jan Breiner, Pierre Coutris, and Alfons Schwarzenböck
Atmos. Meas. Tech., 18, 2311–2331, https://doi.org/10.5194/amt-18-2311-2025, https://doi.org/10.5194/amt-18-2311-2025, 2025
Short summary
Short summary
Airborne cloud observation relies on high-frequency black-and-white-image information. The study presents automatic shape recognition tools developed with machine learning techniques and adapted for this type of image. Applied on a recent field campaign, these tools produce morphology-specific size distributions that can be compared across four instruments covering different size ranges. The analysis show that the tools are performing well and are consistent across the different instruments.
Zhenhai Zhang, Vesta Afzali Gorooh, Duncan Axisa, Chandrasekar Radhakrishnan, Eun Yeol Kim, Venkatachalam Chandrasekar, and Luca Delle Monache
Atmos. Meas. Tech., 18, 1981–2003, https://doi.org/10.5194/amt-18-1981-2025, https://doi.org/10.5194/amt-18-1981-2025, 2025
Short summary
Short summary
Water is a precious resource, and it is essential to monitor and predict the current and future occurrence of precipitation-producing clouds. We investigate the cloud characteristics related to precipitation using several cloud cases in the United Arab Emirates with data from aircraft measurements, satellite observations, and weather radar observations. This study provides scientific support for the development of an applicable framework to examine cloud precipitation processes.
Huiying Zhang, Xia Li, Fabiola Ramelli, Robert O. David, Julie Pasquier, and Jan Henneberger
Atmos. Meas. Tech., 17, 7109–7128, https://doi.org/10.5194/amt-17-7109-2024, https://doi.org/10.5194/amt-17-7109-2024, 2024
Short summary
Short summary
Our innovative IceDetectNet algorithm classifies each part of aggregated ice crystals, considering both their basic shape and physical processes. Trained on ice crystal images from the Arctic taken by a holographic camera, it correctly classifies over 92 % of the ice crystals. These more detailed insights into the components of aggregated ice crystals have the potential to improve our estimates of microphysical properties such as riming rate, aggregation rate, and ice water content.
Fuzeng Wang, Yuanyu Duan, Yao Huo, Yaxi Cao, Qiusong Wang, Tong Zhang, Junqing Liu, and Guangmin Cao
Atmos. Meas. Tech., 17, 6933–6944, https://doi.org/10.5194/amt-17-6933-2024, https://doi.org/10.5194/amt-17-6933-2024, 2024
Short summary
Short summary
This study analyzes precipitation features over the Qinghai–Tibet Plateau, focusing on drop size distribution (DSD), terminal velocity, number density, radar reflectivity (Z), and rainfall intensity (I) at four sites. Terminal velocity is fastest at Nyalam and slowest at Lhasa. Five microphysical characteristics exhibit different correlations with altitude. Z–I relationships differ by site, and both exponential and gamma distributions fit DSD, with gamma parameters varying with altitude.
Philippe Ricaud, Pierre Durand, Paolo Grigioni, Massimo Del Guasta, Giuseppe Camporeale, Axel Roy, Jean-Luc Attié, and John Bognar
Atmos. Meas. Tech., 17, 5071–5089, https://doi.org/10.5194/amt-17-5071-2024, https://doi.org/10.5194/amt-17-5071-2024, 2024
Short summary
Short summary
Clouds in Antarctica are key elements affecting climate evolution. Some clouds are composed of supercooled liquid water (SLW; water held in liquid form below 0 °C) and are difficult to forecast by models. We performed in situ observations of SLW clouds at Concordia Station using SLW sondes attached to meteorological balloons in summer 2021–2022. The SLW clouds were observed in a saturated layer at the top of the planetary boundary layer in agreement with ground-based lidar observations.
Flor Vanessa Maciel, Minghui Diao, and Ching An Yang
Atmos. Meas. Tech., 17, 4843–4861, https://doi.org/10.5194/amt-17-4843-2024, https://doi.org/10.5194/amt-17-4843-2024, 2024
Short summary
Short summary
The partition between supercooled liquid water and ice crystals in mixed-phase clouds is investigated using aircraft-based in situ observations over the Southern Ocean. A novel method is developed to define four phases of mixed-phase clouds. Relationships between cloud macrophysical and microphysical properties are quantified. Effects of aerosols and thermodynamic and dynamical conditions on ice nucleation and phase partitioning are examined.
Jingxuan Luo, Yubing Pan, Debin Su, Jinhua Zhong, Lingxiao Wu, Wei Zhao, Xiaoru Hu, Zhengchao Qi, Daren Lu, and Yinan Wang
Atmos. Meas. Tech., 17, 3765–3781, https://doi.org/10.5194/amt-17-3765-2024, https://doi.org/10.5194/amt-17-3765-2024, 2024
Short summary
Short summary
Accurate cloud quantification is critical for climate research. We developed a novel computer vision framework using deep neural networks and clustering algorithms for cloud classification and segmentation from ground-based all-sky images. After a full year of observational training, our model achieves over 95 % accuracy on four cloud types. The framework enhances quantitative analysis to support climate research by providing reliable cloud data.
Yuji Ayatsuka
Atmos. Meas. Tech., 17, 3739–3750, https://doi.org/10.5194/amt-17-3739-2024, https://doi.org/10.5194/amt-17-3739-2024, 2024
Short summary
Short summary
Many types of halos appear in the sky. Each type of halo reflects the state of the atmosphere; therefore observing them from the ground greatly helps in understanding the state of the atmosphere. However, halos are easily obscured by the contrast of the cloud itself, making it difficult to observe them. This study describes the construction of a sky-color model for halos and a new effective algorithm to reveal halos in images.
Nina Maherndl, Manuel Moser, Johannes Lucke, Mario Mech, Nils Risse, Imke Schirmacher, and Maximilian Maahn
Atmos. Meas. Tech., 17, 1475–1495, https://doi.org/10.5194/amt-17-1475-2024, https://doi.org/10.5194/amt-17-1475-2024, 2024
Short summary
Short summary
In some clouds, liquid water droplets can freeze onto ice crystals (riming). Riming leads to the formation of snowflakes. We show two ways to quantify riming using aircraft data collected in the Arctic. One aircraft had a cloud radar, while the other aircraft was measuring directly in cloud. The first method compares radar and direct observations. The second looks at snowflake shape. Both methods agree, except when there were gaps in the cloud. This improves our ability to understand riming.
Bu-Yo Kim, Joo Wan Cha, and Yong Hee Lee
Atmos. Meas. Tech., 16, 5403–5413, https://doi.org/10.5194/amt-16-5403-2023, https://doi.org/10.5194/amt-16-5403-2023, 2023
Short summary
Short summary
A camera-based imager and convolutional neural network (CNN) were used to estimate ground cloud cover. Image data from 2019 were used for training and validation, and those from 2020 were used for testing. The CNN model exhibited high performance, with an accuracy of 0.92, RMSE of 1.40 tenths, and 93% agreement with observed cloud cover within ±2 tenths' difference. It also outperformed satellites and ceilometers and proved to be the most suitable for ground-based cloud cover estimation.
John S. Schreck, Gabrielle Gantos, Matthew Hayman, Aaron Bansemer, and David John Gagne
Atmos. Meas. Tech., 15, 5793–5819, https://doi.org/10.5194/amt-15-5793-2022, https://doi.org/10.5194/amt-15-5793-2022, 2022
Short summary
Short summary
We show promising results for a new machine-learning based paradigm for processing field-acquired cloud droplet holograms. The approach is fast, scalable, and leverages GPUs and other heterogeneous computing platforms. It combines applications of transfer and active learning by using synthetic data for training and a small set of hand-labeled data for refinement and validation. Artificial noise applied during synthetic training enables optimized models for real-world situations.
Louis Jaffeux, Alfons Schwarzenböck, Pierre Coutris, and Christophe Duroure
Atmos. Meas. Tech., 15, 5141–5157, https://doi.org/10.5194/amt-15-5141-2022, https://doi.org/10.5194/amt-15-5141-2022, 2022
Short summary
Short summary
Optical array probes are instruments used aboard research aircraft to capture 2D images of ice or water particles in clouds. This study presents a new tool using innovative machine learning, called convolutional neural networks, designed to identify the shape of imaged ice particles for two of these imagers, namely 2DS and PIP. Such a tool will be a very strong asset for understanding cloud microphysics. Beyond traditional evaluation metrics, human inspections were performed of unknown data.
Yun Yuan, Huige Di, Yuanyuan Liu, Tao Yang, Qimeng Li, Qing Yan, Wenhui Xin, Shichun Li, and Dengxin Hua
Atmos. Meas. Tech., 15, 4989–5006, https://doi.org/10.5194/amt-15-4989-2022, https://doi.org/10.5194/amt-15-4989-2022, 2022
Short summary
Short summary
We put forward a new algorithm for joint observation of the cloud boundary by lidar and Ka-band millimetre-wave cloud radar. Cloud cover and boundary distribution characteristics are analysed from December 2020 to November 2021 in Xi'an. More than 34 % of clouds appear as a single layer every month. The maximum and minimum normalized cloud cover occurs in summer and winter, respectively. The study can provide more information on aerosol–cloud interactions and forecasting numerical models.
Rachel Atlas, Johannes Mohrmann, Joseph Finlon, Jeremy Lu, Ian Hsiao, Robert Wood, and Minghui Diao
Atmos. Meas. Tech., 14, 7079–7101, https://doi.org/10.5194/amt-14-7079-2021, https://doi.org/10.5194/amt-14-7079-2021, 2021
Short summary
Short summary
Many clouds with temperatures between 0 °C and −40 °C contain both liquid and ice particles, and the ratio of liquid to ice particles influences how the clouds interact with radiation and moderate Earth's climate. We use a machine learning method called random forest to classify images of individual cloud particles as either liquid or ice. We apply our algorithm to images captured by aircraft within clouds overlying the Southern Ocean, and we find that it outperforms two existing algorithms.
Bu-Yo Kim, Joo Wan Cha, and Ki-Ho Chang
Atmos. Meas. Tech., 14, 6695–6710, https://doi.org/10.5194/amt-14-6695-2021, https://doi.org/10.5194/amt-14-6695-2021, 2021
Short summary
Short summary
This study investigates a method for 24 h cloud cover calculation using a camera-based imager and supervised machine learning methods. The cloud cover is calculated by learning the statistical characteristics of the ratio, difference, and luminance using RGB channels of the image with a machine learning model. The proposed approach is suitable for nowcasting because it has higher learning and prediction speed than the method in which the many pixels of a 2D image are learned.
Jun Inoue, Yutaka Tobo, Kazutoshi Sato, Fumikazu Taketani, and Marion Maturilli
Atmos. Meas. Tech., 14, 4971–4987, https://doi.org/10.5194/amt-14-4971-2021, https://doi.org/10.5194/amt-14-4971-2021, 2021
Short summary
Short summary
A cloud particle sensor (CPS) sonde is an observing system to obtain the signals of the phase, size, and the number of cloud particles. Based on the field experiments in the Arctic regions and numerical experiments, we proposed a method to correct the CPS sonde data and found that the CPS sonde system can appropriately observe the liquid cloud if our correction method is applied.
Ilona Ylivinkka, Santeri Kaupinmäki, Meri Virman, Maija Peltola, Ditte Taipale, Tuukka Petäjä, Veli-Matti Kerminen, Markku Kulmala, and Ekaterina Ezhova
Atmos. Meas. Tech., 13, 5595–5619, https://doi.org/10.5194/amt-13-5595-2020, https://doi.org/10.5194/amt-13-5595-2020, 2020
Short summary
Short summary
In this study, we developed a new algorithm for cloud classification using solar radiation and cloud base height measurements. Our objective was to develop a simple and inexpensive but effective algorithm for the needs of studies related to ecosystem and atmosphere interactions. In the present study, we used the algorithm for obtaining cloud statistics at a measurement station in southern Finland, and we discuss the advantages and shortcomings of the algorithm.
Georgios Touloupas, Annika Lauber, Jan Henneberger, Alexander Beck, and Aurélien Lucchi
Atmos. Meas. Tech., 13, 2219–2239, https://doi.org/10.5194/amt-13-2219-2020, https://doi.org/10.5194/amt-13-2219-2020, 2020
Short summary
Short summary
Images of cloud particles give important information for improving our understanding of microphysical cloud processes. For phase-resolved measurements, a large number of water droplets and ice crystals need to be classified by an automated approach. In this study, a convolutional neural network was designed, which exceeds the classification ability of traditional methods and therefore shortens the analysis procedure of cloud particle images.
Carola Barrientos Velasco, Hartwig Deneke, Hannes Griesche, Patric Seifert, Ronny Engelmann, and Andreas Macke
Atmos. Meas. Tech., 13, 1757–1775, https://doi.org/10.5194/amt-13-1757-2020, https://doi.org/10.5194/amt-13-1757-2020, 2020
Short summary
Short summary
In the changing Arctic, quantifying the resulting variability of incoming solar radiation is important to better elucidate the net radiative effect of clouds. As part of a multidisciplinary expedition in the central Arctic held in early summer 2017, a novel network of pyranometers was deployed over an ice floe to investigate the spatiotemporal variability of solar radiation under different sky conditions. This study presents the collected data and an analysis of the spatiotemporal variability.
Sylke Boyd, Stephen Sorenson, Shelby Richard, Michelle King, and Morton Greenslit
Atmos. Meas. Tech., 12, 4241–4259, https://doi.org/10.5194/amt-12-4241-2019, https://doi.org/10.5194/amt-12-4241-2019, 2019
Short summary
Short summary
How cirroform clouds affect the radiation balance of the atmosphere depends on their properties, including ice particle types such as crystals, pellets, and fragments. Ice halos form if ice particles in these clouds are in a smooth hexagonal crystalline form. This paper introduces a method to search long-term records of sky images for ice halos, as gathered by total sky imagers (TSIs). Such an analysis will allow one to explore geographical and seasonal variations in cirrus cloud particle types.
Thibault Vaillant de Guélis, Alfons Schwarzenböck, Valery Shcherbakov, Christophe Gourbeyre, Bastien Laurent, Régis Dupuy, Pierre Coutris, and Christophe Duroure
Atmos. Meas. Tech., 12, 2513–2529, https://doi.org/10.5194/amt-12-2513-2019, https://doi.org/10.5194/amt-12-2513-2019, 2019
Michael L. Larsen and Raymond A. Shaw
Atmos. Meas. Tech., 11, 4261–4272, https://doi.org/10.5194/amt-11-4261-2018, https://doi.org/10.5194/amt-11-4261-2018, 2018
Short summary
Short summary
A statistical tool frequently utilized to measure scale-dependent departures from perfect randomness is the radial distribution function. This tool has many strengths, but it is not easy to calculate for particle detections within a three-dimensional sample volume. In this manuscript, we introduce and test a new method to estimate the three-dimensional radial distribution function in realistic measurement volumes.
Jiangchuan Tao, Chunsheng Zhao, Ye Kuang, Gang Zhao, Chuanyang Shen, Yingli Yu, Yuxuan Bian, and Wanyun Xu
Atmos. Meas. Tech., 11, 895–906, https://doi.org/10.5194/amt-11-895-2018, https://doi.org/10.5194/amt-11-895-2018, 2018
Short summary
Short summary
Existing chamber technologies for direct measurements of number concentration of cloud condensation nuclei (NCCN) are sophisticated and expensive. In this paper, a new method is proposed to calculate NCCN based only on measurements of a humidified nephelometer system which have accounted for influences of both aerosol size and aerosol hygroscopicity on NCCN calculation. This new method makes NCCN measurements more convenient and is capable of obtaining NCCN at lower supersaturations.
Christine Aebi, Julian Gröbner, Niklaus Kämpfer, and Laurent Vuilleumier
Atmos. Meas. Tech., 10, 4587–4600, https://doi.org/10.5194/amt-10-4587-2017, https://doi.org/10.5194/amt-10-4587-2017, 2017
Short summary
Short summary
The current study analyses the cloud radiative effect during the daytime depending on cloud fraction and cloud type at two stations in Switzerland over a time period of 3–5 years. Information about fractional cloud coverage and cloud type is retrieved from images taken by visible all-sky cameras. Cloud cover, cloud type and other atmospheric parameters have an influence on the magnitude of the longwave cloud effect as well as on the shortwave.
Emmanuel Fontaine, Delphine Leroy, Alfons Schwarzenboeck, Julien Delanoë, Alain Protat, Fabien Dezitter, Alice Grandin, John Walter Strapp, and Lyle Edward Lilie
Atmos. Meas. Tech., 10, 2239–2252, https://doi.org/10.5194/amt-10-2239-2017, https://doi.org/10.5194/amt-10-2239-2017, 2017
Short summary
Short summary
In this study we evaluate a method to estimate cloud water content (CWC) knowing cloud reflectivity. Ice hydrometeors are replace by ice oblate spheroids to simulate their reflectivity. There is no assumption on the relation between mass and their size. Then, a broad range of CWCs are compared with direct measurements of CWC. The accuracy of the method is ~ ±32 %. This study is performed in areas of convective clouds where reflectivity and CWC are especially high, what makes it unique.
Hsu-Yung Cheng and Chih-Lung Lin
Atmos. Meas. Tech., 10, 199–208, https://doi.org/10.5194/amt-10-199-2017, https://doi.org/10.5194/amt-10-199-2017, 2017
Short summary
Short summary
A cloud detection method for all-sky images is proposed. Obtaining improved cloud detection results is helpful for cloud classification, tracking and solar irradiance prediction. The features are extracted from local image patches with different sizes to include local structure and multi-resolution information. The cloud models are learned through the training process. We have shown that taking advantages of multiple classifiers and various patch sizes is able to increase the detection accuracy.
Qingyong Li, Zhen Zhang, Weitao Lu, Jun Yang, Ying Ma, and Wen Yao
Atmos. Meas. Tech., 9, 753–764, https://doi.org/10.5194/amt-9-753-2016, https://doi.org/10.5194/amt-9-753-2016, 2016
Short summary
Short summary
This paper proposes a new cloud classification method, named bag of micro-structures (BoMS), for whole-sky imagers. BoMS treats an all-sky image as a collection of micro-structures mapped from image patches, rather than a collection of pixels. BoMS identifies five different sky conditions: cirriform, cumuliform, stratiform, clear sky, and mixed cloudiness (often appearing in all-sky images but seldom addressed in the literature). The performance of BoMS overperforms those of traditional methods.
Som Sharma, Rajesh Vaishnav, Munn V. Shukla, Prashant Kumar, Prateek Kumar, Pradeep K. Thapliyal, Shyam Lal, and Yashwant B. Acharya
Atmos. Meas. Tech., 9, 711–719, https://doi.org/10.5194/amt-9-711-2016, https://doi.org/10.5194/amt-9-711-2016, 2016
Short summary
Short summary
Cloud base height observations from Ceilometer CL31 were extensively studied during May 2013 to January 2015 over Ahmedabad (23.03°N, 72.54°E), India. Results indicate that the ceilometer is an excellent instrument to precisely detect low- and mid-level clouds, and that the MODIS satellite provides accurate retrieval of high-level clouds over this region.
S. Gagné, L. P. MacDonald, W. R. Leaitch, and J. R. Pierce
Atmos. Meas. Tech., 9, 619–630, https://doi.org/10.5194/amt-9-619-2016, https://doi.org/10.5194/amt-9-619-2016, 2016
Short summary
Short summary
Measurements of clouds with an aircraft are essential to understand how clouds form and how they affect the Earth's climate. These measurements are used in climate models to help predict how our climate might develop in the next century. Aircraft measurements are, however, difficult for modellers to interpret because the way they were acquired and analyzed varies from one team of scientists to the next. We present a software platform for scientists to share and compare their analysis tools.
H.-Y. Cheng and C.-C. Yu
Atmos. Meas. Tech., 8, 1173–1182, https://doi.org/10.5194/amt-8-1173-2015, https://doi.org/10.5194/amt-8-1173-2015, 2015
Short summary
Short summary
This work performs cloud classification on all-sky images. To deal with mixed cloud types, we propose performing block-based classification. The proposed method combines local texture features with classical statistical texture features. The experimental results have shown that applying the combined feature results in higher classification accuracy. It is also validated that using block-based classification outperforms classification on the entire images.
A. Korolev and P. R. Field
Atmos. Meas. Tech., 8, 761–777, https://doi.org/10.5194/amt-8-761-2015, https://doi.org/10.5194/amt-8-761-2015, 2015
M. L. Krüger, S. Mertes, T. Klimach, Y. F. Cheng, H. Su, J. Schneider, M. O. Andreae, U. Pöschl, and D. Rose
Atmos. Meas. Tech., 7, 2615–2629, https://doi.org/10.5194/amt-7-2615-2014, https://doi.org/10.5194/amt-7-2615-2014, 2014
M. I. A. Berghof, G. P. Frank, S. Sjogren, and B. G. Martinsson
Atmos. Meas. Tech., 7, 877–886, https://doi.org/10.5194/amt-7-877-2014, https://doi.org/10.5194/amt-7-877-2014, 2014
A. Schwarzenboeck, G. Mioche, A. Armetta, A. Herber, and J.-F. Gayet
Atmos. Meas. Tech., 2, 779–788, https://doi.org/10.5194/amt-2-779-2009, https://doi.org/10.5194/amt-2-779-2009, 2009
Cited articles
Bengio, Y., Simard, P., and Frasconi, P.: Learning long-term dependencies with gradient descent is difficult, IEEE T. Neural Networ., 5, 157–166, 1994. a
Chen, T., Kornblith, S., Norouzi, M., and Hinton, G.: A simple framework for contrastive learning of visual representations, in: International conference on machine learning, virtual, 13–18 July 2020, 1597–1607, PMLR, https://proceedings.mlr.press/v119/chen20j.html (last access: 25 June 2025), 2020. a, b
Chen, X. and He, K.: Exploring simple siamese representation learning, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, virtual, 19–25 June 2021, 15750–15758, https://openaccess.thecvf.com/content/CVPR2021/html/Chen_Exploring_Simple_Siamese_Representation_Learning_CVPR_2021_paper.html (last access: 25 June 2025), 2021. a
Chu, Y., Zhang, H., Li, X., and Henneberger, J.: Data for publication “Exploring the effect of training set size and number of categories on ice crystal classification through a contrastive semi- supervised learning algorithm”, Zenodo [data set], https://doi.org/10.5281/zenodo.14696359, 2025a. a
Chu, Y., Zhang, H., Li, X., and Henneberger, J.: Models for publication “Exploring the effect of training set size and number of categories on ice crystal classification through a contrastive semi- supervised learning algorithm”, Zenodo [code], https://doi.org/10.5281/zenodo.14793334, 2025b. a
Chu, Y., Zhang, H., Li, X., and Henneberger, J.: Code for publication “Exploring the effect of training set size and number of categories on ice crystal classification through a contrastive semi- supervised learning algorithm”, Zenodo [code], https://doi.org/10.5281/zenodo.14795274, 2025c. a
Geiss, A., Christensen, M. W., Varble, A. C., Yuan, T., and Song, H.: Self-Supervised Cloud Classification, Artificial Intelligence for the Earth Systems, 3, e230036, https://doi.org/10.1175/AIES-D-23-0036.1, 2024. a
Grill, J.-B., Strub, F., Altché, F., Tallec, C., Richemond, P., Buchatskaya, E., Doersch, C., Avila Pires, B., Guo, Z., Gheshlaghi Azar, M., Piot, B., kavukcuoglu, k., Munos, R., and Valko, M.: Bootstrap Your Own Latent – A New Approach to Self-Supervised Learning, in: Advances in Neural Information Processing Systems, edited by: Larochelle, H., Ranzato, M., Hadsell, R., Balcan, M., and Lin, H., 33, 21271–21284, Curran Associates, Inc., https://proceedings.neurips.cc/paper_files/paper/2020/file/f3ada80d5c4ee70142b17b8192b2958e-Paper.pdf (last access: 25 June 2025), 2020. a, b
Gupta, A. K., Deshmukh, A., Waman, D., Patade, S., Jadav, A., Phillips, V. T., Bansemer, A., Martins, J. A., and Gonçalves, F. L.: The microphysics of the warm-rain and ice crystal processes of precipitation in simulated continental convective storms, Commun. Earth Environ., 4, 226, https://doi.org/10.1038/s43247-023-00884-5, 2023. a
He, K., Zhang, X., Ren, S., and Sun, J.: Deep residual learning for image recognition, in: Proceedings of the IEEE conference on computer vision and pattern recognition, Las Vegas, 26 June–1 July 2016, 770–778, https://openaccess.thecvf.com/content_cvpr_2016/html/He_Deep_Residual_Learning_CVPR_2016_paper.html (last access: 25 June 2025), 2016. a, b
He, K., Fan, H., Wu, Y., Xie, S., and Girshick, R.: Momentum contrast for unsupervised visual representation learning, in: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, virtual, 14–19 June 2020, 9729–9738, https://openaccess.thecvf.com/content_CVPR_2020/html/He_Momentum_Contrast_for_Unsupervised_Visual_Representation_Learning_CVPR_2020_paper.html (last access: 25 June 2025), 2020. a, b, c, d
Henneberger, J., Ramelli, F., Spirig, R., Omanovic, N., Miller, A. J., Fuchs, C., Zhang, H., Bühl, J., Hervo, M., Kanji, Z. A., Ohneiser, K., Radenz, M., Rösch, M., Seifert, P., and Lohmann, U.: Seeding of supercooled low stratus clouds with a UAV to study microphysical ice processes: An introduction to the CLOUDLAB project, B. Am. Meteor. Soc., 104, E1962–E1979, 2023. a, b, c
Jaffeux, L., Schwarzenböck, A., Coutris, P., and Duroure, C.: Ice crystal images from optical array probes: classification with convolutional neural networks, Atmos. Meas. Tech., 15, 5141–5157, https://doi.org/10.5194/amt-15-5141-2022, 2022. a, b, c
Järvinen, E., Jourdan, O., Neubauer, D., Yao, B., Liu, C., Andreae, M. O., Lohmann, U., Wendisch, M., McFarquhar, G. M., Leisner, T., and Schnaiter, M.: Additional global climate cooling by clouds due to ice crystal complexity, Atmos. Chem. Phys., 18, 15767–15781, https://doi.org/10.5194/acp-18-15767-2018, 2018. a
Jiang, J., Shu, Y., Wang, J., and Long, M.: Transferability in deep learning: A survey, arXiv [preprint], https://doi.org/10.48550/arXiv.2201.05867, 15 January 2022. a
Keppas, S., Crosier, J., Choularton, T., and Bower, K.: Ice lollies: An ice particle generated in supercooled conveyor belts, Geo-5 physical Research Letters, 44, 5222–5230, 2017. a
Lamb, D. and Verlinde, J.: Physics and chemistry of clouds, Cambridge University Press, https://doi.org/10.1017/CBO9780511976377, 2011. a
Li, Z., Liu, F., Yang, W., Peng, S., and Zhou, J.: A survey of convolutional neural networks: analysis, applications, and prospects, IEEE T. Neur. Net. Lear., 33, 6999–7019, 2021. a
Lindqvist, H., Muinonen, K., Nousiainen, T., Um, J., McFarquhar, G., Haapanala, P., Makkonen, R., and Hakkarainen, H.: Ice-cloud particle habit classification using principal components, J. Geophys. Res.-Atmos., 117, D16206, https://doi.org/10.1029/2012JD017573, 2012. a, b
Liou, K., Gu, Y., Yue, Q., and McFarguhar, G.: On the correlation between ice water content and ice crystal size and its application to radiative transfer and general circulation models, Geophys. Res. Lett., 35, L13805, https://doi.org/10.1029/2008GL033918, 2008. a
Lohmann, U., Lüönd, F., and Mahrt, F.: An introduction to clouds: From the microscale to climate, Cambridge University Press, https://doi.org/10.1017/CBO9781139087513, 2016. a, b
Lv, Q., Li, Q., Chen, K., Lu, Y., and Wang, L.: Classification of ground-based cloud images by contrastive self-supervised learning, Remote Sens., 14, 5821, https://doi.org/10.3390/rs14225821, 2022. a
Maahn, M., Moisseev, D., Steinke, I., Maherndl, N., and Shupe, M. D.: Introducing the Video In Situ Snowfall Sensor (VISSS), Atmos. Meas. Tech., 17, 899–919, https://doi.org/10.5194/amt-17-899-2024, 2024. a, b
Mumuni, A. and Mumuni, F.: Data augmentation: A comprehensive survey of modern approaches, Array, 16, 100258, https://doi.org/10.1016/j.array.2022.100258, 2022. a
Oord, A. v. d., Li, Y., and Vinyals, O.: Representation learning with contrastive predictive coding, arXiv [preprint], https://doi.org/10.48550/arXiv.1807.03748, 10 July 2018. a
Pasquier, J. T., David, R. O., Freitas, G., Gierens, R., Gramlich, Y., Haslett, S., Li, G., Schäfer, B., Siegel, K., Wieder, J., Adachi, K., Belosi, F., Carlsen, T., Decesari, S., Ebell, K., Gilardoni, S., Gysel-Beer, M., Henneberger, J., Inoue, J., Kanji, Z. A., Koike, M., Kondo, Y., Krejci, R., Lohmann, U., Maturilli, M., Mazzolla, M., Modini, R., Mohr, C., Motos, G., Nenes, A., Nicosia, A., Ohata, S., Paglione, M., Park, S., Pileci, R. E., Ramelli, F., Rinaldi, M., Ritter, C., Sato, K., Storelvmo, T., Tobo, Y., Traversi, R., Viola, A., and Zieger, P.: The Ny-Ålesund aerosol cloud experiment (nascent): Overview and first results, B. Am. Meteor. Soc., 103, E2533–E2558, 2022a. a, b, c
Pasquier, J. T., Henneberger, J., Ramelli, F., Lauber, A., David, R. O., Wieder, J., Carlsen, T., Gierens, R., Maturilli, M., and Lohmann, U.: Conditions favorable for secondary ice production in Arctic mixed-phase clouds, Atmos. Chem. Phys., 22, 15579–15601, https://doi.org/10.5194/acp-22-15579-2022, 2022b. a, b, c, d
Pasquier, J. T., Henneberger, J., Korolev, A., Ramelli, F., Wieder, J., Lauber, A., Li, G., David, R. O., Carlsen, T., Gierens, R., Maturilli, M., and Lohmann, U.: Understanding the history of two complex ice crystal habits deduced from a holographic imager, Geophys. Res. Lett., 50, e2022GL100247, https://doi.org/10.1029/2022GL100247, 2023. a, b, c
Ramelli, F., Beck, A., Henneberger, J., and Lohmann, U.: Using a holographic imager on a tethered balloon system for microphysical observations of boundary layer clouds, Atmos. Meas. Tech., 13, 925–939, https://doi.org/10.5194/amt-13-925-2020, 2020. a
Santachiara, G., Belosi, F., and Prodi, F.: Ice crystal precipitation at Dome C site (East Antarctica), Atmos. Res., 167, 108–117, 2016. a
Schlenczek, O., Fugal, J. P., Lloyd, G., Bower, K. N., Choularton, T. W., Flynn, M., Crosier, J., and Borrmann, S.: Microphysical properties of ice crystal precipitation and surface-generated ice crystals in a High Alpine environment in Switzerland, J. Appl. Meteorol. Clim., 56, 433–453, 2017. a
Schmitt, C. G., Järvinen, E., Schnaiter, M., Vas, D., Hartl, L., Wong, T., and Stuefer, M.: Classification of ice particle shapes using machine learning on forward light scattering images, Artificial Intelligence for the Earth Systems, 3, 230091, https://doi.org/10.1175/AIES-D-23-0091.1, 2024. a, b, c
Shorten, C. and Khoshgoftaar, T. M.: A survey on image data augmentation for deep learning, Journal of Big Data, 6, 1–48, 2019. a
Song, Q., Cui, Z., and Liu, P.: An Efficient Solution for Semantic Segmentation of Three Ground-based Cloud Datasets, Earth Space Sci., 7, e2019EA001040, https://doi.org/10.1029/2019EA001040, 2020. a
Thériault, J. M., Stewart, R. E., and Henson, W.: Impacts of terminal velocity on the trajectory of winter precipitation types, Atmos. Res., 116, 116–129, 2012. a
Touloupas, G., Lauber, A., Henneberger, J., Beck, A., and Lucchi, A.: A convolutional neural network for classifying cloud particles recorded by imaging probes, Atmos. Meas. Tech., 13, 2219–2239, https://doi.org/10.5194/amt-13-2219-2020, 2020. a
van Diedenhoven, B., Fridlind, A. M., Cairns, B., and Ackerman, A. S.: Variation of ice crystal size, shape, and asymmetry parameter in tops of tropical deep convective clouds, J. Geophys. Res.-Atmos., 119, 11–809, 2014. a
Wendisch, M., Yang, P., and Pilewskie, P.: Effects of ice crystal habit on thermal infrared radiative properties and forcing of cirrus, J. Geophys. Res.-Atmos., 112, D08201, https://doi.org/10.1029/2006JD007899, 2007. a
Wendling, P., Wendling, R., and Weickmann, H. K.: Scattering of solar radiation by hexagonal ice crystals, Appl. Optics, 18, 2663–2671, 1979. a
Wu, Z., Xiong, Y., Yu, S. X., and Lin, D.: Unsupervised feature learning via non-parametric instance discrimination, in: Proceedings of the IEEE conference on computer vision and pattern recognition, Salt Lake City, 18–22 June 2018, 3733–3742, https://openaccess.thecvf.com/content_cvpr_2018/html/Wu_Unsupervised_Feature_Learning_CVPR_2018_paper.html (last access: 25 June 2025), 2018. a
Yang, P., Liou, K., Wyser, K., and Mitchell, D.: Parameterization of the scattering and absorption properties of individual ice crystals, J. Geophys. Res.-Atmos., 105, 4699–4718, 2000. a
Ye, L., Cao, Z., and Xiao, Y.: DeepCloud: Ground-based cloud image categorization using deep convolutional features, IEEE T. Geosci. Remote, 55, 5729–5740, 2017. a
You, Y., Gitman, I., and Ginsburg, B.: Large batch training of convolutional networks, arXiv [preprint], https://doi.org/10.48550/arXiv.1708.03888, 13 August 2017. a
Zeiler, M. D. and Fergus, R.: Visualizing and understanding convolutional networks, in: Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, 6–12 September 2014, Proceedings, Part I 13, 818–833, Springer, https://doi.org/10.1007/978-3-319-10590-1_53, 2014. a
Zhang, H.: Development of a classification algorithm for ice crystal habit by using deep learning, Master's thesis, University of Oslo, http://urn.nb.no/URN:NBN:no-91875 (last access: 5 November 2024), 2021. a
Zhang, H., Binder, A., Pasquier, J., Krummenacher, B., Ramelli, F., Storelvmo, T., David, R. O., and Henneberger, J.: Deep-learning based classification of ice crystals: habits and microphysical processes, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-8098, https://doi.org/10.5194/egusphere-egu22-8098, 2022. a, b
Zhang, H., Li, X., Ramelli, F., David, R. O., Pasquier, J., and Henneberger, J.: IceDetectNet: a rotated object detection algorithm for classifying components of aggregated ice crystals with a multi-label classification scheme, Atmos. Meas. Tech., 17, 7109–7128, https://doi.org/10.5194/amt-17-7109-2024, 2024. a, b, c, d, e, f, g, h, i, j, k, l
Zhang, R., Isola, P., and Efros, A. A.: Split-brain autoencoders: Unsupervised learning by cross-channel prediction, in: Proceedings of the IEEE conference on computer vision and pattern recognition, Honolulu, 21–26 July 2017, 1058–1067, https://openaccess.thecvf.com/content_cvpr_2017/html/Zhang_Split-Brain_Autoencoders_Unsupervised_CVPR_2017_paper.html (last access: 25 June 2025), 2017. a
Short summary
Our study improves ice crystal shape classification, key for understanding weather and climate. By adding unsupervised pre-training before supervised classification, our algorithm reduces manual labeling effort while maintaining high accuracy. It outperforms fully supervised models across datasets of varying sizes and categories, showing strong generalization ability. This method improves ice crystal classification techniques, making it adaptable to different environmental datasets.
Our study improves ice crystal shape classification, key for understanding weather and climate....
