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

Within ice-containing clouds, ice crystals can have many shapes, or habits. The radiative properties of ice crystals, as well as the rate of microphysical processes such as sublimation, sedimentation, and accretion, vary depending on habit, which is consequential for parameterizing ice-containing clouds within weather and climate models. Although atmospheric parameters, most notably temperature and ice supersaturation, are useful for determining crystal growth mode, the advection of ice crystals from one part of a cloud to another often causes habits to vary from expectations given a set of atmospheric conditions.

The best way to determine habit is to analyze aircraft in-situ ice crystal images. Images were collected from multiple cloud imaging probes during the Investigation of Microphysics and Precipitation for Atlantic Coast-Threatening Snowstorms (IMPACTS) aircraft campaign; these probes include the Two-Dimensional Stereo Probe (2DS), the Cloud Particle Imager (CPI), and the Particle Habit Imaging and Polar Scattering probe (PHIPS). One of seven possible habits is assigned to each crystal image: column, sphere, dendrite, plate, aggregate, graupel, and irregular. Two methods of ice crystal classification are used: a simple manually tuned decision tree, that utilizes morphological properties of ice crystals, including crystal area (A), perimeter (p) and maximum dimension (D), and a neural network that is trained using a set of manually pre-classified crystal images.

A unique aspect of this work is that habit classifications from a similar algorithm are compared among multiple probes. Initial classification results from one flight (normalized over two-minute blocks of time) show 69% agreement between classifications (D > 100 um) for the 2DS and PHIPS probes, although normalized identification rates of columns continue to show discrepancies, averaging 14% for the PHIPS and 5% for the 2DS. The fractional habit classifications also suggest that a large proportion of crystals during the IMPACTS campaign were irregularly shaped (averaging 51% for the 2DS and 36% for the PHIPS); this in addition to manual image analysis revealing many crystals to be ambiguous in habit presents a challenge for incorporating habit-related parameters into models.