A Multi-Probe Automated Classification of Ice Crystal Habits During the IMPACTS Campaign

Virtually everyone has heard of the saying that no two snowflakes (and likewise, ice crystals) are exactly the same. While this is true, two different ice crystals may share many similarities, allowing them to be grouped together according to their shapes, or habits. For a given crystal, rates of microphysical processes, including accretion, sedimentation, and sublimation, as well as light absorption and scattering, are dependent on the habit of the crystal, meaning that an accurate understanding of how habits depend on environmental conditions and cloud formation mechanisms will improve the treatment of processes within mixed and ice phase clouds in weather and climate models. Laboratory studies show the expected habit is a strong function of temperature and ice supersaturation; however, habits frequently do not match those observed in the laboratory as crystals are advected through regions with varying atmospheric conditions, and sometimes undergo secondary growth processes such as riming and aggregation.

While the number of ice crystals within any given cloud is far too large to obtain habit information for every single one, nor is it possible to sample the entire cloud volume at the resolution of a single crystal, a representative sample of ice crystals can be attained using aircraft in-situ cloud imaging probes, including, but not limited to, the Two-Dimensional Stereo Probe (2DS), the High-Volume Precipitation Spectrometer (HVPS), and the Particle Habit Imaging and Polar Scattering probe (PHIPS). Many ice crystal images were collected using these probes during the Investigation of Microphysics and Precipitation for Atlantic Coast-Threatening Snowstorms (IMPACTS), conducted in January and February of 2020, 2022, and 2023 over the Northeast U.S.

This study assesses the successes and limitations of a relatively simple habit classification algorithm using data from the IMPACTS campaign. The algorithm determines habit by using a series of decisions based on crystal morphological properties, including, but not limited to, crystal area (A), perimeter (p) and maximum dimension (D). Seven habit classes are used: column/needle, sphere, dendrite, plate, aggregate, graupel, and irregular. The algorithm is designed to be relatively lenient in classifying crystals as non-irregular, given that nearly all crystals were observed to exhibit some degree of irregularity relative to an idealized habit.

The algorithm is applied to 2DS, PHIPS, and HVPS images, in order to test its adaptability for probes of differing resolutions and size ranges. The algorithm is designed to be nearly identical for each probe, although a few coefficients to adjust threshold values are necessary to handle differences in image quality and resolution. Accuracies are calculated using a set of manually pre-classified images, 100 from each habit class. The high resolution of the PHIPS allowed for higher accuracy (~80%) compared to the 2DS and HVPS (~70%). The application of the same algorithm to different probes is a novel aspect of this study, and moderate correspondence in habits determined using the 2DS and HVPS probes over a similar size range was noted. Given that different field campaigns utilize different probes, it is important to understand how different probes affect habit classifications when comparing habits across multiple field campaigns.

Another unique contribution of this study is the determination of particle habits for crystals with D > 1 mm from the HVPS probe, as there are fewer studies of habit information for larger particle sizes. Using a combination of 2DS and HVPS data, microphysical habits of ice crystals with D > 300 µm are related to mesoscale cloud features using a case study from IMPACTS, in which a heavy band of snowfall associated with a nor’easter was sampled on February 7, 2020. At lower elevations and higher temperatures (~2500 m and -3 to -6°C) significant habit variation was seen between inside and immediately outside the band, including more columns/needles and graupel within the band. At higher elevations and lower temperatures, less variation was seen from inside to outside the band. In addition to allowing for exploration of habits within a mesoscale cloud feature, the case study also provides an opportunity to qualitatively assess the performance of the algorithm for the PHIPS, 2DS, and HVPS.