Analysis of Ice Crystal Morphologies in Deep Convective Clouds Using In-Flight Measurements during HAIC–HIWC Cayenne

Abstract Deep convective clouds (DCCs) are characterized by distinct microphysical processes across their cirriform, stratiform, and convective regions. This study is analyzing hydrometeor microphysical and associated morphological properties of hydrometeors in DCCs from the high-altitude ice crystals–high ice water content (HAIC–HIWC) Cayenne campaign, thereby distinguishing the abovementioned cloud regions. Two airborne optical array probes [two-dimensional stereo (2D-S) and precipitation imaging probe (PIP)] collected 2D images of hydrometeors (with 300 ≤ D max ≤ 6000 μ m), processed by a convolutional neural network to automatically retrieve their habit. Distinct morphological crystal properties were observed from PIP images (2–6 mm) across the regions of DCCs. Cirriform regions are dominated by fragile aggregates (FAs) and columns (Cos), originating from convective turrets and undergoing riming and aggregation during their ascent. In stratiform regions, aggregation dominates, with FA dominating in the upper layers and compact particles (CPs) and rimed aggregates (RAs) prevailing in the lower layers due to sedimentation and precipitation. In convective regions, CP fractions, exceeding 50%, are fostered by updrafts and turbulence leading to intense riming. The 2D-S images of submillimetric hydrometeors (300–1200 μ m) reveal that over 50% of the particles are found irregular, classified as CP and to a smaller amount as FA. In general, low pristine number fractions of hydrometeors in DCCs can be attributed to rapid transformation processes via riming, aggregation, and possibly fragmentation leading to secondary ice production (SIP). This study provides first quantitative results on morphological number fractions of ice crystals in DCCs, emphasizing the need to incorporate these findings into models to refine parameterizations not only for ice crystal growth and sedimentation but also for their radiative properties. Significance Statement This study introduces a convolutional neural network (CNN)-based identification of particle shapes closely related to ice crystal growth regimes. Performing the quantitative analysis of morphological shapes in addition to state-of-the-art microphysical analysis (particle size distributions) considerably improves not only knowledge and model parameterization of hydrometeor fall speeds, thus cloud lifetimes, but also parameterization of optical properties of clouds.