Floods can destroy more than property, taking away the feeling of security that comes with a home.
Researchers from Pennsylvania State University and other institutions may have cracked a big part of the puzzle in ensuring future safety from such disasters. Their model predicts floods with more accuracy and in far less time than older tools. “With our new approach, we can create simulations using the same process, regardless of the region we are trying to simulate,” civil and environmental engineering professor Chaopeng Shen said.
NOAA’s National Water Model is the standard hydrologists rely on. But while it’s trusted, it’s also slow. Traditional calibration involves feeding it decades of river data for every site. One site at a time. Shen described it as “time-consuming, expensive, and tedious.” The team’s new method speeds things up using artificial intelligence systems that can spot patterns in mountains of data.
Instead of having to start over with every river basin, AI can generalize information using past readings. “Rather than approaching each site individually, the neural network applies general principles it interprets from past data to make predictions,” co-author Yalan Song said.
The model still follows physics-based rules about how water behaves, but it quickly adapts to new areas. And while AI is great when it comes to deciphering patterns, rare storms throw a monkey wrench in the works. Song said their model keeps the water physics in place, then allows the network to learn from the messy parts.
This method predicts extreme rainfall much better than older tools. Researchers used 15 years of river data and asked the system to replay 40 years of streamflow. Against real records, its projections landed about 30% closer across 4,000 sites.
“With a trained neural network, we can generate parameters for the entire U.S. within minutes,” Shen said. Work that once took weeks on many supercomputers now is finished in hours on one machine.
Similar methods have been used to design safer solid-state batteries. They can also map city vegetation for cooling plans. Labs even test AI in nuclear fusion research.
MIT News reported that training models uses large amounts of electricity and water. A study by Hugging Face and Carnegie Mellon University found some systems can consume as much electricity as a small country. The industry, however, is moving toward renewable energy, and this could buy families time to save more than just their possessions.
Join our free newsletter for weekly updates on the latest innovations improving our lives and shaping our future, and don’t miss this cool list of easy ways to help yourself while helping the planet.