I'll be honest - when I first saw this paper's title, "A Tale of Two Coasts," I figured it was going to be a straightforward climate doom scroll. Two coastlines, some flood maps, maybe a scary chart. What actually confused me was the machine learning angle. Why would you need three different AI models to tell you that New Orleans floods? Turns out, knowing that a city floods and knowing why, how badly, and who gets hurt worst are completely different questions - and the answers are wilder than I expected.

Three Robots Walk Into a Floodplain

Researchers Hemal Dey and Wanyun Shao at the University of Alabama didn't just eyeball a map and circle the wet spots. They fed historical FEMA flood damage data from 2012 to 2017 into three machine learning models - a support vector machine (SVM), a random forest, and a multilayer perceptron (MLP) - alongside 16 different risk factors covering hazard, exposure, and vulnerability (Dey & Shao, 2026). Think of it like giving three very different students the same exam: one's a geometry nerd, one's a statistics whiz, and one's a neural network that learned everything from flashcards. When all three independently agree that your city is in trouble, that's not a fluke. That's a consensus.

The result? Eight US coastal cities flagged as high-risk, stretching from Houston to New York. More than 17 million Americans live in the highest-risk zones along the Gulf and Atlantic coasts. That's roughly the population of the Netherlands, except the Netherlands actually planned for this.

New York Has the Numbers, New Orleans Has the Percentages

Here's where the data gets genuinely unsettling. New York City tops the list with 4.75 million people at risk under general flood damage scenarios and 4.4 million under extreme conditions. Those are staggering absolute numbers. But New Orleans? Nearly 99% of its population is exposed under both scenarios. Ninety-nine percent. That's not a risk factor - that's a city-wide condition.

The remaining six cities rounding out the list - Norfolk, Charleston, Jacksonville, Miami, Mobile, and Houston - each bring their own cocktail of problems. Houston's impervious surfaces channel water like a bowling alley gutter. Miami sits on porous limestone that laughs at seawalls. Norfolk is sinking while the ocean rises to meet it, like the world's worst handshake (Tulane University subsidence study, 2025).

It's Not Just Water - It's Who the Water Hits

The part of this study that separates it from your standard flood map is the vulnerability dimension. Dey and Shao didn't stop at geography and infrastructure. They factored in poverty rates, minority populations, residents without high school diplomas, children under five, and the elderly. And the pattern is exactly what you'd expect if you've been paying attention: the people least equipped to recover from flooding are the ones most likely to get flooded.

This tracks with broader research showing that machine learning flood models consistently outperform traditional physical models at capturing these nonlinear relationships between environmental and social factors (Springer, 2025). A river doesn't care about your income bracket, but your recovery absolutely does.

The Framework That Actually Scales

What makes this more than an alarming data dump is the scalability argument. The researchers built their framework to be portable - plug in local FEMA data, feed it the same 16 factors, and you can run this analysis for any flood-prone region. It's a diagnostic tool, not just a report card. And the dual-scenario approach, separating general flood damage from extreme events, gives policymakers something they rarely get: nuance.

The proposed fixes aren't exactly shocking - levees, wetland preservation, permeable surfaces like grass-tile parking lots - but having ML-backed evidence pointing at specific cities with specific vulnerabilities makes the "we should probably do something" conversation a lot harder to dodge. As co-author Wanyun Shao put it, the goal was turning an "abstract level of awareness into tangible numbers" (Scientific American, 2026).

So What Now?

Jeremy Porter at CUNY raised the uncomfortable truth: people don't leave waterfront property just because a study says it floods. They adapt. They raise houses, buy pumps, file insurance claims, and stay. Which means the real value of frameworks like this isn't convincing people to move - it's giving cities the data to build smarter before the next storm, not after.

Seventeen million people on two coastlines. Three algorithms that agree. Eight cities that should probably start returning FEMA's calls.

References

1. Dey, H., & Shao, W. (2026). A tale of two coasts: Unveiling US Gulf and Atlantic coastal cities at high flood risk. Science Advances. DOI: 10.1126/sciadv.aec2079. PMID: 42018627

2. Springer Nature. (2025). Application of machine learning for coastal flooding. Discover Cities. https://link.springer.com/article/10.1007/s44327-025-00125-8

3. Tulane University. (2025). Study reveals uneven land sinking across New Orleans, raising flood-risk concerns. Phys.org. https://phys.org/news/2025-06-reveals-uneven-orleans.html

4. Scientific American. (2026). New York City, New Orleans at greatest risk of extreme damage from floods, new analysis reveals. https://www.scientificamerican.com/article/new-york-city-new-orleans-at-greatest-risk-of-extreme-damage-from-floods-new-analysis-reveals/

5. Houston Public Media. (2026). New study finds 'alarming' high flood risk for 17 million Americans on Atlantic and Gulf coasts. https://www.houstonpublicmedia.org/articles/news/flooding/2026/04/24/550002/houston-flood-risk-study-gulf-coast-atlantic-cities/

Disclaimer: This blog post is a simplified summary of published research for educational purposes. The accompanying illustration is artistic and does not depict actual model architectures, data, or experimental results. Always refer to the original paper for technical details.