It's hard to know when a tornado has formed or why. A new dataset could hold answers. It contains radar returns from thousands of tornadoes that have hit the United States in the past 10 years.
Some storms spawned tornadoes, while others did not. Researchers at MIT Lincoln Laboratory have released TorNet, a dataset of radar returns from thousands of tornadoes in the United States. They hope to make breakthroughs in detecting tornadoes.
Mark Veillette, the project's co-principal investigator, said:
A lot of progress is driven by easily available, benchmark datasets. We hope TorNet will lay a foundation for machine learning algorithms to detect and predict tornadoes.
Researcher work in the Air Traffic Control Systems Group.
The team is releasing models trained on the dataset. The models show that machine learning can spot a twister. Building on this work could help forecasters provide more accurate warnings.
About 1,200 tornadoes occur in the United States every year, causing up to billions of dollars in damage and claiming lives. Last year, one unusually long-lasting tornado killed 17 people and injured at least 165 others in Mississippi.
Forecasters must decide when to issue a tornado warning. They often err on the side of caution, which leads to more than 70% of false alarms. This makes people less likely to take warnings seriously over time.
Scientists have recently turned to machine learning to improve tornado detection and prediction. However, they have had trouble accessing the data and models, which has slowed progress. TorNet aims to help. It's dataset includes over 200,000 radar images. 13,587 show tornadoes, while the rest show severe storms or false alarms.
Building an operational algorithm is difficult, especially in safety-critical situations. The next steps will be taken by researchers around the world who are inspired by the dataset and want to build their own algorithms. Those algorithms will be tested by forecasters to start the process of transitioning into operations.