Google open-sources its PyTorch-based hydrology framework, allowing global researchers to utilise Flood Hub’s AI models to extend critical river flood prediction windows by up to six days.
On 3 June 2026, Google open-sourced the hydrology framework that powers the river forecasting models driving its global Google Flood Hub platform. The hydrology model is provided as a Python package built using the open-source PyTorch machine learning library.
Flooding is one of the most destructive natural disasters globally, disproportionately affecting areas without advanced meteorological infrastructure. The hydrology framework processes geographical data (climate, soil types, land cover, and topography) alongside meteorological data (forecasted temperature, rainfall, etc.) to predict the daily flow rate of rivers globally.
The code repository provides access to two distinct iterations of Google’s AI model. The first version was the baseline model featured in Google’s 2024 benchmarking studies, and the second version is the current framework used by Flood Hub for real-time global flood forecasting.
By open-sourcing the architecture, Google allows local scientists and meteorologists to plug their specific regional datasets directly into the framework. This newer, upgraded architecture introduces multi-source meteorological processing. Benchmarking demonstrates it extends the reliable predictive horizon by six days in gauged basins and by one day in ungauged basins compared to the original model.
The model’s training pipeline is built around Caravan, an open-source hydrological dataset containing historical river tracking information. External researchers, scientists, and local flood forecasting agencies can import their own local watershed datasets into the repository to fine-tune or custom-train the AI architecture for specific regional terrains.
Google has packaged the code with interactive Python tutorial notebooks and companion video walk-throughs to assist global agencies with deploying the framework.
