There's Hype Around Machine Learning And AI For Flood Mitigation — What's Really Involved If A City Or County Wants To Dive In?

By Jeff Albee

Southern California’s recent severe floods are flashing a warning for local and state governments to reexamine their flood mitigation preparedness plans. The uncertainty and flooding risks from a more extreme weather future are driving some governmental bodies to deploy new artificial intelligence (AI)-powered technologies to analyze and prepare for similar events. 

Understanding flood patterns is a key concern of state and local officials across the country, who depend on these projections for both emergency services planning and structural engineering purposes. During a major flood event like California’s, governments depend on their own preparations and predictions to lay the groundwork for their emergency response and hazard mitigation. 

Currently, those governmental units depend on data provided by FEMA that projects what 100- and 500-year floods would look like. While that’s industry standard practice, the FEMA datasets have significant limitations: First, they’re based on historical data that may lack relevance for a volatile climate future, and second, their scope is limited to riverine flooding, just one part of the flood threat that faces communities hit by extreme weather events. 

Flooding like Southern California just faced wouldn’t have gotten much illumination from the FEMA projections. In the direct path of an atmospheric river, California faced a 1,000-year flood that not only overtook rivers but streets as well. Flash flooding occurred, dams were threatened, and very little of the danger and damage was projectible based on FEMA datasets.

With these kinds of extreme weather events appearing to occur with greater frequency, state and local governments are scrambling to prepare for the possibilities. Properly scaled preparation can save lives — and getting a sense of that scale is where AI is helping. 

AI-enhanced flood projections are offering governments improved scenario planning, which in turn enables the development of better response alternatives. As with other AI applications throughout the economy, properly applied AI capabilities leverage existing knowledge, and in this case build on FEMA’s work to project nearly every type of flood eventuality. 

The strength of AI systems for this kind of planning is that they’re built on statistical foundations but aren’t limited to the existing data. Rather than only looking backward, AI flood mitigation technology pulls in additional factors that can intensify flood scenarios. The output allows governments to understand where their biggest risks lie, and what the choices and costs of mitigation will be.

While FEMA data helps those in traditional flood zones — like communities along riverbanks — AI data helps look at risks to previously unthreatened areas. Bringing in elevation models and land use maps can then paint a more descriptive and realistic picture for state officials. The downside possibilities become real, and the resulting planning is more focused and urgent.

Leveraging the AI platforms, government engineers and emergency services workers can pose simple questions based on real-world scenarios. What happens if they get 20 inches of rainfall? What happens if, like California, they get stuck under an atmospheric river that stays for days and days? What roads are under threat, and who is likely to be at the greatest risk? 

The answers help allocate resources and direct capital construction projects. Governments move from the micro — where culverts need to be built or drainage systems improved — to the macro: What level of rainfall would cause flash floods sufficient to force evacuations or the closure of highways? 

These are vital questions that can help save lives; a large proportion of deaths during flooding events are caused by individuals trying to escape their stalled cars on flooded roads. 

As extreme weather events become more common, engineers can also use the data to better understand the vulnerabilities of their infrastructure. Bridges are especially vulnerable; many were not built to withstand the velocity and volume of water that can be created by unprecedented rainfall. 

Do regulations in the areas need to change to ensure long term protection, or should certain bridges be prioritized to address their structural weaknesses? Having instant access to data and projections that can be updated with real-life variables will be key for decision makers in an increasingly volatile time. 

Of course, no technology is perfect. AI tools still lever off existing data, and even with a wider set of variables, a perfect understanding of the future is impossible. It’s unwise for any engineering team to rely solely on these projections when determining total resource allocation.

Deploying these kinds of AI projection tools does require resources; most tools will cost at least a million dollars to implement and fine-tune statewide. It’s not cheap, but it’s far cheaper than some of the open-ended, impossible-to-integrate technologies that have a habit of busting budgets without offering concrete results.

But the potential return on investment — reducing loss-of-life and economic impacts — is significant. Having the ability to keep pace with the rate of change is essential to ensuring safety during extreme weather events.

Floods are changing. The technology we use to defend against them should change as well.

Jeff Albee is the VP/Director of Digital Solutions at Stantec.