Novel AI-Based Multiview Detailed Classification, AI-Enabled 3D Shape Reconstruction, Multicamera Instrumentation, and Comprehensive Analysis & Profiling of Snowflakes and Snowfall

About This Grant

This research addresses an important topic related to snowflakes and snowfall using state-of-the-art and yet-to-be-developed methodologies, tools, and instrumentation and has significant potential to advance knowledge about the natural diversity of snowflake properties. The principal objective is to enhance scientific understanding of snow and ice particles with applications in improving weather forecasting and precipitation estimation, the importance of which to economy, safety, and everyday life can hardly be overstated. In line with the saying that “no two snowflakes are alike,” there indeed is a huge natural variability of shapes, sizes, internal compositions, densities, and habits of snow and ice particles, which are both truly fascinating and extremely challenging to observe, measure, analyze, understand, and predict. This project develops and applies a synergistic systematic approach to analysis, characterization, and quantification of a large variety of snowflakes and snowfall advancing and integrating microphysics and computational methodologies, emerging artificial intelligence (AI) and deep machine learning (ML) techniques, sophisticated image and data processing and computer vision methods, and cutting-edge optical and computer technology and equipment. The project performs new comprehensive microphysical analysis, profiling, and parametrization of precipitation particles, whose outcomes will be systematized and organized into the “Snowflake World Bank” database. Education and outreach plan includes advising/training Ph.D. students, course modules, undergraduate capstone projects, and vertically integrated “AI and Snowflakes” outreach program. This research develops new general methods for classification, characterization, and quantification of snowflakes based on multicamera measurements and AI/ML and image-processing techniques. Main outcomes are a novel AI/ML-enabled methodology for automatic classification of precipitation, namely, deciding to which of the predefined classes of winter hydrometeors the observed particles belong, based on multiview images by multicamera instrumentation measurements, as well as a novel method for realistic reconstruction of 3D shapes of winter hydrometeors using full images from multiview cameras and AI/ML. To aid the observations, a novel multicamera instrument for snowflakes in freefall, namely, the Multicamera Snowflake Profiler (MSP), will be developed. The MSP will incorporate and apply novel classification and shape reconstruction methods, on-site, in real time, which will, conversely, be used to advance the methods. The research measures, calculates, and estimates a wide range of properties and parameters of ice particles including geometric categories; degrees of riming; melt/dry state; fall speed; 3D reconstructed shape; volume; mass; effective density; and effective dielectric constant among others. Generally, and not exclusively, it pursues the concept of classification first and then quantification by assigning these properties and parameters to various categories of previously classified particles. The accuracy, reliability, and versatility of the data and information are enabled by the new general methods for classification, characterization, and quantification of snowflakes. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.