How 12,000 Houston Residents Got 30 Extra Minutes During the February Floods

Houston has flooded before. Hurricane Harvey in 2017 dropped over 60 inches of rain in some areas, caused $125 billion in damage, and killed 68 people in Harris County alone. The city sits on a flat coastal plain with clay soils that resist absorption, crossed by bayous that function as the primary drainage system. When rainfall exceeds the bayous' capacity, water backs up into residential neighborhoods — often with little warning.

The Harris County Flood Control District operates one of the most advanced flood monitoring networks in the country: 150+ stream gauge stations, 2,500+ rain gauges, and a real-time flood warning system that publishes gauge readings every 5 minutes. It is, by any standard, an exemplary local system.

The gap is the same gap that existed in LA: the monitoring network tells you what the water is doing right now. It does not tell individual households what the water will do to them in 30 minutes.

The February 2026 Event

On February 14, 2026, a slow-moving Gulf moisture surge delivered 8 to 12 inches of rain across Harris County in a 6-hour window. The National Weather Service issued a Flash Flood Warning at 2:15 PM for the county. Wireless Emergency Alerts went out at 2:22 PM with the standard message: "Flash Flood Warning in effect for your area. Avoid flooded areas."

ZoneCastAI's flood model, running on NOAA's HRRR precipitation forecast and Harris County's real-time gauge data, had identified the high-risk drainage basins 47 minutes earlier. At 1:35 PM, 12,400 ZoneCastAI users in the Brays Bayou, White Oak Bayou, and Greens Bayou watersheds received personalized notifications that included three specific data points the WEA did not: estimated time to street-level flooding at their address, the nearest elevated shelter accepting residents, and which of their pre-computed evacuation routes was still passable.

What Made the Difference

Hydrological modeling, not just weather. Rain gauges tell you how much water has fallen. Flood models tell you where that water is going. ZoneCastAI's flood engine ingests real-time precipitation data, terrain elevation (from USGS 3DEP at 1-meter resolution), soil saturation estimates, and bayou gauge levels to project which streets will flood and when. The model runs every 5 minutes and compares projections against the most recent gauge readings to continuously correct its estimates.

Community reports as ground truth. By 1:45 PM, ZoneCastAI users in the Meyerland area were submitting photo reports of water on the street — 30 minutes before the nearest bayou gauge showed overflow. These reports were geotagged, timestamped, and fed into the model as real-time validation points, improving downstream predictions for users further along the drainage path.

Route intelligence. By the time the WEA was delivered at 2:22 PM, three of the major evacuation routes from the Brays Bayou area were already impassable. ZoneCastAI users who left at 1:40 PM had access to all three. Users who waited for the WEA had access to one. This is the 30-minute gap in practice — not a theoretical improvement, but a measurable difference in available options.

Limitations and Transparency

We are not claiming to have saved lives. We are claiming that 12,400 people received more specific, earlier, and more actionable information than they would have from the WEA system alone. Whether that information changed outcomes depends on whether individuals acted on it — and we don't have that data. What we do have is user engagement metrics showing that 74% of users who received the 1:35 PM alert opened the notification within 3 minutes, compared to a 31% open rate for the subsequent WEA.