Computing Advances, AI Make Waves for Flood Software at DHS

Funding boosts in the modeling software that has been a cornerstone for federal agencies preparing and responding to natural disasters, like Hurricane Katrina, is enabling the Department of Homeland Security to better study coastal hazards in the U.S.

The software, called Advanced Circulation Storm Surge Model (ADCIRC), predominantly evaluates flooding hazards along the U.S. coasts. That data forms the basis for the Federal Emergency Management Agency’s flood maps in coastal areas and helps determine flood insurance rates in those areas.

As a principal developer of the software, Rick Luettich leads the Coastal Resilience Center (CRC), a DHS Science and Technology Directorate Center of Excellence that is working to innovate the accuracy and timelines of storm surge models. The center is led by the University of North Carolina at Chapel Hill, made possible by a five-year, $20 million grant from the agency.

In recent years, the center has expanded ADCIRC’s capabilities “substantially,” according to Luettich. This includes increasing capabilities in flooding predictions, storm forecasting, and planning around response and recovery.

“A colleague and I started [ADCIRC] prior to DHS’ existence with support from the Army Corps of Engineers back in the early 90s,” Luettich told GovernmentCIO Media & Research in an interview. “It was used to support the dredging program. As the early 2000s unfolded, we did more and more storm surge work with it. We were applying it in the greater New Orleans area prior to Hurricane Katrina. Once Katrina hit the area, it was a critical tool in the assessment of what went wrong. … DHS said we need a center that works with coastal natural hazards. Obviously this is a big center of concern.”

The software has also been proven to have cost benefits. According to a 2020 DHS report, ADCIRC “has by far the highest net present value (NPV), which was primarily due to the reduction of unnecessary evacuations during the hurricane season.”

The report notes total federal investment in ADCIRC clocked in around $36.9 million (adjusted for inflation) in 2017. ADCIRC’s projected return on investment over a 10-year period ranges from $72.1 million to almost $533 million.

Even before the DHS commission, the flagship software helped preserve lives and property along the nation’s coasts.

“We’ve worked a lot with the Coast Guard along the East Coast in particular, when Hurricane Irene came up the East Coast (in 2011), turns out their offices were flooded out, and we predicted that was going to happen several days in advance and they moved all their offices to St. Louis and were able to keep continuity of operations,” Luettich said. “One can only make that call using good predictive tools. We also worked heavily with them for hurricanes Irma and Maria in Miami.”

The CRC also works consistently with Louisiana due to a “constant need for storm search and flooding” along the state’s coast. Helping coastal communities build resilience to these types of natural hazards is a key mission driver for the center.

“We do a lot of work in the planning area and do a lot with the American Planning Association,” Luettich said, adding that the center worked with the association to develop the Plan Integration for Resilience Scorecard (PIRS), which helps communities integrate a variety of plans to make overarching decisions. The scorecard has since been made part of their standard practice materials.

ADCIRC’s data helps coastal community planners make better decisions to bolster resilience to flooding. The interactive web mapper, called Coastal Emergency Risks Assessment (CERA) and an extension to ADCIRC, provides a web portal for users to see a variety of storm scenarios.

But running ADCIRC comes with its own challenges. ADCIRC’s data is highly accurate and reliable, but due to the time-sensitive, risk-sensitive nature of the data, the CRC employs high-performance computing (HPC) clusters across the country to run the software.

“[ADCIRC is] very computationally demanding,” Luettich said, noting that computing capabilities were not as advanced in the 90s at the center’s founding. “Computing has dramatically changed over the 30 years we’ve been working on this, so we’ve had to rewrite ADCIRC probably three times to take advantage of computing architectures.”

Running ADCIRC via HPC clusters also allows the center to run various storm scenarios to ensure local and federal emergency response teams have all the data they need to refine response plans. The center employs multiple clusters across the country, including one at UNC-Chapel Hill and the Texas Advanced Computing Center (TACC) in Austin.

“One of the cool things about the ADCIRC prediction system is, again, we spent a lot of time rewriting it so it will work as efficiently as possible on these machines,” Luettich said. “We’re able to shift quickly and put our software and run them at HPC centers and coordinate the whole process in real time. We may have it running on six computers the whole time and cross coordinating. We can run wherever the storm is impacting and marshal HPC resources across the country to try to allow us to run faster and run multiple scenarios. If we watch hurricanes, they tend to shift. What they predict five days ago, and what happens when it gets to landfall, is likely to be different. What we try to do is take the prediction and use that in our model.”

This is where artificial intelligence comes into play via surrogate modeling.

“What surrogate modeling does is it says ok, if you do three different ADCIRC runs, now you have these ADCIRC outputs — what if the storm goes in between, do you have to run ADCIRC again? Or is the data rich enough that we can figure out what another storm is going to do with the data you have in hand?” Luettich said. “FEMA is very interested in this because they’ve already had their contractors do a number of ADCIRC runs in support of the national flood insurance program, but not every possible scenario has been run through ADCIRC.”

Due to the forward-looking nature of DHS S&T COEs, Luettich is already thinking about what storms and hurricanes will look like in 50 years and how they’ll impact coastal communities.

“The 2020 hurricane season definitely kept us busy — every storm that comes by, we’ve learned something about how we can get better,” Luettich said.