Meteorologists in 2019 have a powerful host of technological tools at their disposal to help them predict severe weather early and accurately. Still, as anyone who has crossed paths with an ill-timed thunderstorm this summer knows, it remains an inexact science. There are simply too many variables to forecast every storm, every time. Plus, the finite supply of meteorologists means that less developed and/or smaller storm systems often don’t get coverage, even when it’s justified. This handicap results in injuries, property loss, and even deaths each year when people are caught by extreme weather.
Using AI in Real-Time Weather Analysis
To address this bottleneck, a multi-pronged research effort headed by scientists from AccuWeather, Pennsylvania State University, and Spain’s University of Almeria was launched with the goal of developing an AI model that could predict storms better and sooner. In the group’s paper, published last month in IEEE Transactions on Geoscience and Remote Sensing, they outlined a model that predicts severe weather based on satellite imagery of cloud rotation.
Using machine learning linear classifiers, the tool notes specific rotational movements that might indicate early storm formation that could be missed by humans. Scientists working on the project described this imagery as a valuable data point. “The very best forecasting incorporates as much data as possible,” said Steve Wistar, senior forensic meteorologist at AccuWeather. “There’s so much to take in, as the atmosphere is infinitely complex. By using the models and the data we have [in front of us], we’re taking a snapshot of the most complete look of the atmosphere.”
Beware the Comma-Shaped Cloud
The team built the AI to look for a singular, tell-tale sign of an early-stage storm: comma-shaped clouds. These formations are powerful indicators that a cyclone could soon form, bringing with it attendant pleasantries like gusty winds or hail. To “train” the system, the team analyzed over 50,000 satellite images taken above various atmospheric conditions. Through a combination of computer vision and machine learning, the AI became able to autonomously detect the dreaded comma-shaped cloud.
The system is also able to notify human meteorologists of these danger zones. This could allow them to cut through less important noise and focus their attentions on looming cyclones. Helping people wade through mountains of data to get to the good stuff is among the burgeoning tech’s most valuable functions. The researchers found that the model can pick out comma-shaped clouds at a rate of 99 percent at a speed of just 40 seconds per cloud identified. That could prove pivotal for a field where quickly parsing dynamic and growing datasets is often the hardest task.
The Future of Meteorology?
Crucially, the study also revealed that the AI was able to predict well over half—64%—of severe weather events, an improvement over existing models. It also flashed the ability to pick out danger spots before humans watching in real-time. If that early detection advantage proves durable, the task of predicting dangerous weather may soon be a lot easier. As well, thousands of storm victims each year could have the comma-shaped-cloud detective work of a computer model to thank for their increased preparedness.