Current flood forecasts in coastal areas often lack the precision people need. It’s like being told, “it will rain this afternoon,” without knowing whether it’ll be a light drizzle or a heavy downpour. But scientists are working to change that. By leveraging powerful graphics processing unit (GPU) technology, researchers are making coastal flood predictions faster, more accurate and more detailed — more like knowing there’s a 70% chance of rain and whether it will be light, moderate or heavy.
“The better we can predict the impacts of extreme weather events, such as hurricanes, the more lives we can save,” said Jose Maria Gonzalez Ondina, Ph.D., a research assistant scientist at the UF Center for Coastal Solutions (CCS) who co-leads a public-private collaboration that is focused on enhancing a forecasting tool called the Semi-implicit Cross-scale Hydroscience Integrated Systems Model (SCHISM).
This advanced model simulates how water moves across environments from small creeks to the open ocean. By supercharging SCHISM to run hundreds and even thousands of simulations during storm events, scientists aim to provide communities with better tools to anticipate and prepare for coastal flooding during storms, among other applications.
Current circulation models used for forecasting have significant limitations. Scientists rely on meteorological predictions to run simulations, but time constraints often allow for only a single projected outcome — without showing how likely that outcome is.
“Ensemble forecasting, however, is changing the game,” said Gonzalez Ondina. “This method runs multiple simulations and incorporates slight variations in factors like wind speed, to predict a range of potential outcomes and their likelihoods. This approach provides a much clearer and accurate picture of potential storm impacts.”
“When we complete this project, we will have drastically accelerated our three-dimensional earth system simulations. The results from this work will mark a major stride forward in our field.”
– Y. Joseph Zhang, Ph.D., professor, Virginia Institute of Marine Science
The leap in SCHISM’s ability to support ensemble forecasting is being driven by a shift from standard central processing units (CPUs) to GPUs. GPUs are known for their high computational speed, allowing them to handle complex, multi-scenario simulations more efficiently.
“GPUs are fueling AI innovations, but we’re applying them in a different way,” said Gonzalez Ondina. “We’re optimizing traditional numerical models to run on GPUs, increasing their speed and efficiency. This is a hybrid approach that blends conventional models with emerging AI techniques.”
Gonzalez Ondina is putting the finishing touches on an update of a portion of SCHISM, in preparation for testing its speed and performance. Over the next year, experts from the Virginia Institute of Marine Science (VIMS), University of Florida Center for Coastal Solutions (CCS), National Oceanic and Atmospheric Administration (NOAA) and NVIDIA, plan to expand SCHISM’s capabilities, moving from two-dimensional to three-dimensional simulations. This advancement will improve forecasting accuracy and help coastal communities better prepare for extreme weather events before they happen.
—
By Megan Sam