Super_7_2022_658x90.jpg

Actions

Can artificial intelligence better predict flooding in coastal areas?

ai flooding predictions
Posted at 7:42 PM, Jan 28, 2022
and last updated 2022-01-28 19:42:18-05

NEW ORLEANS, La. — Coastal communities around the world are especially vulnerable to flooding, storms, hurricanes and heavy rainfall. Now, scientists are studying whether artificial intelligence can better predict the impact of the storms.

More information would help areas like New Orleans, Louisiana, which is forced to fix and rebuild after severe flooding.

Clint Dawson, a professor at the University of Texas Austin, is part of a team of investigators working on a project funded by the Department of Energy’s Office of Advanced Scientific Computing Research.

“The only reason that place still exists is because there is fairly extensive levy system that protects it," Dawson said. “More than half of the world’s population lives close to the coast so it’s a huge problem worldwide.”

Essentially, the team is using artificial intelligence and machine learning to predict future flooding.

“The focus here is on hazards such as hurricane storm surge, rainfall, flooding, particularly in coastal regions," Dawson said.

"They need to be trained on physics, on physical models that obey physics. So the research part, if really can we even build in the physics into the AI."

Harmut Kaiser, a research scientist and adjunct professor at LSU, is helping with the research. 

“These newer networks can be trained to answer questions very quickly. So you show them a large amount of data and train them based on some criteria and later on you can show them similar data and they do a prediction," Kaiser said. “We hope to increase the fidelity of the forecasting. In terms of special resolution and in terms of speed of producing results.”

That speed is something Julie Lesko says could make a big difference. She is a Senior Service Hydrologist at the National Weather Service. 

“Anything that could give us better confidence and better forecasting abilities will always be more helpful," Lesko said. “In the moment, it is hard to say that this storm may produce five-times more rain or 5% more rain than a storm like this did last year. So we really can’t do a lot of that calculation in the moment. You have to actually wait until the event is over and see how it played out and compare it to past storms.”

Currently, there is only so much time they have to forecast a weather system coming into a coastal area, and alert the correct teams for evacuations and protections.

“We usually start looking at that about 72 to 48 hours out as confidence increases about rainfall," Lesko said.

This project could change that.

“The hope is that we will be able to produce data faster so that the people who rely on the data to make the decisions like the emergency management community or the energy sector can make decisions earlier," Kaiser said.

“The DOE is investing heavily in understanding our systems in general and so we’re working very closely to try to make their model more applicable to natural hazards and the gulf coast," Dawson.

Dawson points out they have already seen some success.

“We have seen in some simple examples if you have good data you can use AI to make predictions very quickly," Dawson said.