Tornadoes are one of nature’s most destructive and unpredictable forces, causing billions of dollars in damages and claiming lives every year. Though they are difficult to predict, a new open-source dataset created by researchers from the MIT Lincoln Laboratory, known as TorNet, offers hope in improving our detection and prediction abilities. Composed of radar images from thousands of tornadoes that have hit the U.S. over the past decade, TorNet not only includes instances of tornado-spawning storms, but also severe storms that didn’t produce tornadoes–creating a broad comparison base for machine learning algorithms.
The goal of making the dataset open-source is to stimulate advancements in the use of machine learning for tornado detection and prediction. The researchers believe that building on their work could help weather forecasters provide more accurate warnings, potentially saving lives.
Tornadoes are infamous for being challenging to forecast, primarily because their formation is still not fully understood. Moreover, radar, which is a crucial tool in monitoring atmospheric conditions, often struggles with detecting tornadoes since they lie too low on the ground. This has resulted in a high false-alarm rate for tornado warnings, sometimes exceeding 70%.
To combat this, TorNet provides a rich dataset for machine learning utilization. It contains over 200,000 radar images, 13,587 of which depict tornadoes. The remaining images originated from severe storms that either didn’t produce tornadoes or triggered false alarms.
The team is also releasing models trained on the TorNet dataset. Initial results indicated promising performance, with the deep-learning model accurately classifying 50 percent of weaker EF-1 tornados and over 85 percent of EF-2 or higher tornadoes, the latter of which are responsible for most storm-related devastation and costs. The algorithm surpassed or matched the efficiency of all known tornado-detecting algorithms in academic literature.
Looking forward, the researchers hope their work will spur others in the field to continue their advancements, potentially uncovering new methods of tornado prediction, detection, and understanding. This may lead to the enhancement of early warning systems, enabling individuals and communities to take preventive measures when faced with potential tornado threats.
