NASA Funds CEE Professor's Development of Targeted Flood Forecasts
In New Jersey alone, climate change has triggered a summer of damaging heavy rains and extreme flash flooding. Emergency managers rely on a single projected storm track based on a total rainfall prediction to issue evacuations, stage rescues, and close roads. If the forecast is off by even a few miles, the results can be catastrophic.
Efthymios Nikolopoulos, an associate professor in the Department of Civil and Environmental Engineering (CEE) has been awarded a $633,934 grant from NASA to support emergency decision-makers by creating a framework for improved flood risk assessments—and avoid catastrophes. The ultimate goal is, according to Nikolopoulos, to put more precise information in the hands of those who must make these calls under real time pressure.
"New Jersey is where I live and work, and we'll be testing storms in flooded towns I know," he says. "We're doing this work in our own community and the people who will benefit, if we get it right, are my neighbors. I'd like the plac I live to be better prepared for the next storms."
Giving Emergency Managers the Information they Need
The project, Nikolopoulos reports, hopes to accomplish this by "using Prithvi-WxC—the weather and climate foundation model developed jointly by NASA and IBM—to build a probabilistic precipitation forecasting system, and then using that system to advance flood warning procedures by expressing forecasts as probabilities rather than single values."
He adds, "Instead of one forecast, the system generates 50 physically plausible versions of the same storm, rather than issuing one best guess that's either right or wrong. Having access to a range of possible outcomes is an operational advantage. A single forecast that misses by five miles is a failure. An ensemble that puts meaningful probability on that location gives the emergency manager something to act on in advance."
Real World Applications
Partnering with the New Jersey Office of Emergency Management and the Iowa Flood Center and city of Spencer, Iowa, the system will be tested in two settings: densely populated New Jersey which is prone to fast, localized flash flooding, and Iowa with more rural areas that experience flooding from large thunderstorm systems.
Nikolopoulos is excited about the project's "real world applications with a short path to use. Every improvement we make potentially translates into fewer people suffering during flood events. If the approach works in both test settings, it should transfer broadly," he predicts.
While PI Nikolopoulos and CEE colleague and co-PI Professor Jie Gong began work on the two-year NASA project in September, they are also collaborating on a related University of Iowa led—and funded—international Urban Flash Flood Information System—a project aiming to develop early warnings and flood forecasting in vulnerable Caribbean island nations and Comoros, an island in the Indian Ocean.
A Valuable Research Experience
They will be assisted by two PhD students. "One will focus on assembling and harmonizing the decade of historical forecast and radar data the system trains on and on validating the flood predictions that come out the other end. The other will focus on the machine learning side, running the experiments that will tell the team which design choices actually improve forecasts," Nikolopoulos says.
Most important is the hands-on research, which he believes is "exactly the kind of training the next generation of Earth science AI practitioners need: students who understand both the physics and machine learning, and who have worked directly with the emergency managers who use the output."
