Google open-sources the hydrology framework behind the Flood Hub flood-forecasting service

Google Research announced on 3 June 2026 that it is open-sourcing the hydrology framework that powers Flood Hub, its global flood-forecasting service. The code is available on GitHub under the Apache 2.0 license and is written in Python with PyTorch. With this, Google enables researchers, meteorological services and civil protection around the world to use, verify and build upon the technology behind an operational early-warning system.

What did Google open-source?

The released framework is Python/PyTorch code that powers Flood Hub, a global flood-forecasting service. The Apache 2.0 license permits free use, modification and commercial application, which is important for institutions that want to integrate the model into their own systems. By open-sourcing the code, Google moves from providing a finished service to sharing the technology itself.

This move means that national hydrometeorological services can run and adapt the models to local conditions, instead of relying solely on Google’s infrastructure.

How does the framework forecast floods?

The framework uses the LSTM and the newer ME-LSTM architectures (types of neural networks specialized for processing time series) over the open-source Caravan dataset. Caravan is a publicly available collection of hydrological data, so the combination of open code and open data enables full reproduction and extension of the models.

The upgraded v2 model brings a concrete improvement: it extends the reliable forecast horizon by 6 days in gauged basins, that is, in basins that have gauging stations, and by 1 day in ungauged basins. A longer lead time directly increases communities’ ability to prepare for an incoming flood.

Who confirmed the operational value?

The framework’s operational feasibility was validated by the Czech Hydrometeorological Institute (CHMI). CHMI also built an adapter for the Delft-FEWS platform, a widely used data-management system in flood forecasting. This shows that the framework is not just a research prototype but a solution ready to be integrated into existing operational systems.

Validation by a national meteorological service gives the project the credibility required by early-warning systems, where an error can have severe consequences.

Why does the open-science approach matter?

The World Meteorological Organization (WMO) publicly endorsed the open-science approach of this project. The WMO is a United Nations body responsible for meteorology and hydrology, so its support underscores the global significance of open-sourcing technology like this.

By open-sourcing the code and data, Google enables countries with limited resources to access advanced flood-forecasting models without high development costs. In the context of increasingly frequent extreme weather events, such an approach can help protect vulnerable communities around the world.