AI can quantify hurricane damage, say Ohio State researchers

Researchers at Ohio State University have developed a machine-learning algorithm that predicts the extent of damage to buildings after a hurricane. As reported by Techxplore.

The developers used satellite images taken before and after the hurricane. These images covered 22,686 buildings.

Then, using a convolutional neural network, the researchers reconstructed the outlines of buildings on pre-hurricane satellite images and classified the amount of damage after the storm. The model used four categories:

• undamaged;
• minor damage;
• severe damage;
• destroyed.

The researchers tested their new model on data from Hurricane Michael, which struck the eastern coast of the United States in fall 2018. It turned out that the algorithm correctly assessed the damage in 86.3% of cases in one Florida region. That was 11% better than a model built using a support vector machine (SVM).

“SVM tried to distinguish between minor and severe damage, which could pose a serious problem for hurricane response teams,” said co-author Dashen Liu.

The researcher believes their algorithm could help rescuers in real time to assess the damage caused by the storm.

“In real hurricane situations, the model can be used to assess the degree of building damage to direct emergency services to priority checks,” Liu said.

Earlier in May 2021, researchers developed an AI-powered tool for predicting hail, tornadoes and strong winds during a storm.

In September, DeepMind unveiled the deep-learning tool DGMR for predicting rain 90 minutes ahead.

Subscribe to ForkLog’s Telegram news: ForkLog AI — all the news from the world of AI!