Large-scale, high-resolution flood maps can take decades to build. Deep learning completes them at a fraction of the time and cost, giving the public access to crucial flood information that would otherwise remain unavailable.
The problem
Large-scale, high-resolution flood maps can take decades and substantial resources to produce. FEMA has detailed studies for only about one-third of U.S. rivers, leaving many communities without equivalent information.
The solution
The model learns from existing FEMA flood patterns and terrain, then applies those relationships where detailed studies do not exist. This produces high-resolution flood information at national scale, at a fraction of the time and cost.
Existing FEMA flood patterns provide the training signal. Terrain provides the physical context. The model learns the relationship between the two and estimates flood extent where detailed maps are missing.
National scale. High resolution. Low cost.
Read the method in the paper →The light layer shows FEMA’s published 100-year flood extent. Blue shows model-generated coverage created by applying learned spatial patterns across the country.
Applied across the contiguous United States, the method reveals 11.04 million more people and 4.12 million more buildings within modeled 100-year flood extents beyond FEMA’s published coverage.
Miami and New York have the largest additional exposed populations; McAllen and Houma have almost no FEMA-mapped exposure.
Ranked by the share of exposed residents outside FEMA’s extent, among metropolitan areas with at least 10,000 exposed people.
These city views show how the same national model produces locally detailed spatial insight.