main concepts scientists sections
SimplePopularAdvanced

How to Predict Floods on Rivers with No Gauges ✛

Original: "Hourly U.S.-wide flood simulation beyond the limits of traditional and data-driven models"
· Wencong Yang, Leo Lonzarich, Yalan Song, Haoyu Ji, Ming Pan, Kathryn Lawson, Chaopeng Shen
arXiv:2609.06794v1 · 2026-09-06 · CC BY 4.0 · 1 min
Geophysics
Scientists have taught a computer to track water in every stream across America, even where no one has ever measured water levels.

In a control room where every stream has its own console, lights used to glow only where sensors were installed. The rest of the rivers were dark. Now, you can switch on all the lights at once.

The model works as a hybrid. Neural networks tune the parameters for physical laws rather than guessing water flow directly. It is like letting a computer pick the pipe diameter and wall roughness, and then everything is calculated using the rules of hydraulics. First, a long daily component accumulates snow and soil moisture; then, a fast hourly component tracks each rainstorm. Water flows down a digital map of rivers, and the model outputs discharge for each of the 800,000 reaches every hour.

Previously, it was unclear whether such a model could be trained on data from all of America. It turns out it can: it catches 33% more rare floods than the current system and 159% more than a purely neural-network counterpart.

Now, all the lights are on in the control room. Even on a small side channel where no gauge has ever existed, you can see the water rising in advance. That means warnings will arrive in time, giving people enough time to get out of harm's way.

Training the entire model for the continental United States took about five days on a single A100 GPU, and simulating 10 years of hourly streamflow for more than 800,000 river reaches took only about four hours on the same GPU. This is orders of magnitude faster than traditional calibration of distributed models, which could take years.

In Kim Stanley Robinson's novel 'The Ministry for the Future', a global system for predicting climate disasters warns of floods hours before they strike. The δHBV2.0MTS-MC model is a step toward such a system: it covers the entire river network of a country and provides forecasts every hour, enabling warnings about flash floods in ungauged places.