About This Project

Atmospheric researchers have collected millions of hologram images while flying through clouds. This project will identify those images that contain snowflakes, allowing scientists to pinpoint the conditions that create similar "flavors" of snowflakes.

Ask the Scientists

What is the context of this research?

Scientists have been collecting hologram images from clouds at a rate much faster than they can be processed. Most of the holograms contain useless noise, making it extremely difficult to find the images that provide scientific value. The aim of this project is to use modern "machine learning" algorithms to speed up this process, allowing researchers to extract the full value from the data that has been collected.

The algorithms we develop will extract the holograms that are of most important to atmospheric scientists, allowing them to better understand how different atmospheric conditions influence ice formation. Essentially, we are removing the "hay" from the haystack so that scientists can focus their time and resources to to understand the "needles".

What is the significance of this project?

It's is hard to believe, but scientists still don't know the exact conditions that produce various shapes and sizes of snowflakes. The shape and size of snowflakes has a significant impact on various atmospheric properties that are important for weather forecasting as well as climate models. This project will use instrumentation data (holograms and well as surrounding temperature, humidity, and other information) to deepen our understanding of the science ice crystal formation in the atmosphere.

What are the goals of the project?

This is a pilot project to use state-of-the-art machine learning algorithms to solve a challenging atmospheric physics problem. Our goal is to validate this approach helping us to secure a larger grant to fund a full-time researcher to continue the project.

Our target is to build an algorithm that is 10-100 times faster than current rule-based methods for sifting through atmospheric hologram images while at the same time improving accuracy over current methods (in other words, finding more "interesting" images while also doing a better job at ignoring the noise).

We anticipate at least one peer-reviewed paper to be published from the results of this pilot project.