The next chapter in flood resilience: Open sourcing Google’s hydrology framework

2026-06-03 · Source: The latest research from Google · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Environmental Science & Earth Systems, Software Development & Engineering · Depth: Intermediate, medium

Summary

Google Research open-sourced its hydrology modeling framework on GitHub on June 3, 2026, to empower National Meteorological and Hydrological Services (NMHSs) with advanced AI-based flood forecasting. This Python package, built on PyTorch, enables researchers and forecasters to train models using architectures and data similar to Google's Flood Hub. It processes geographical and meteorological inputs like rainfall and temperature to predict daily river flow rates. The framework includes LSTM-based models and a training pipeline utilizing the open-source Caravan dataset, allowing local data integration. Two versions are available: an original from a 2024 benchmarking study and an upgraded v2 model. The v2 model extends the reliable predictive horizon by six days in gauged basins and one day in ungauged basins, using an ME-LSTM architecture to unify diverse weather inputs such as Graphcast, IFS, IMERG, and CPC. This initiative addresses the World Meteorological Organization's 2025 report on integrating local data, with the Czech Hydrometeorological Institute (CHMI) validating the model and integrating it into the Delft-FEWS platform.

Key takeaway

For hydrological scientists and operational forecasters seeking to enhance flood preparedness, Google's open-sourced framework offers a direct path to integrate advanced AI models. You can refine forecasts using your specialized local data, achieving accuracy comparable to traditional models without their complexity. This scalable tool allows you to incorporate advanced predictions into existing workflows, like Delft-FEWS, significantly improving early warning systems and local water resource management. Explore the GitHub repository to begin implementation.

Key insights

Open-sourcing Google's AI hydrology framework democratizes advanced flood forecasting, enabling local adaptation and global resilience.

Principles

• Open science accelerates innovation.
• Local data improves forecast accuracy.
• AI models can simplify hydrology.

Method

The framework uses PyTorch and LSTM networks, training on geographical, meteorological, and historical river data from the Caravan dataset to predict daily river flow.

In practice

• Train AI models with local data.
• Integrate into Delft-FEWS platform.
• Utilize the interactive tutorial notebook.

Topics

• Flood Forecasting
• Hydrology Modeling
• AI Models
• Open-Source Software
• PyTorch
• Early Warning Systems

Code references

• google-research/flood-forecasting

Best for: AI Scientist, Research Scientist, Domain Expert

Related on AIssential

• Protecting cities with AI-driven flash flood forecasting — AI-driven flash flood forecasting using news data significantly enhances urban climate resilience globally.
• Catalyzing scientific impact through global partnerships and open resources — Open science, driven by global partnerships and shared resources, accelerates high-impact discoveries across diverse scientific disciplines.
• NVIDIA Launches Earth-2 Family of Open Models — the World’s First Fully Open, Accelerated Set of Models and Tools for AI Weather — NVIDIA's Earth-2 provides an open, accelerated AI software stack for comprehensive weather and climate prediction.
• HaorFloodAlert: Deseasonalized ML Ensemble for 72-Hour Flood Prediction in Bangladesh Haor Wetlands — Deseasonalized ML ensembles and SAR proxies accurately predict flash floods in complex, flat wetland environments.
• Evaluating Transformer and LSTM Frameworks for Prediction in Ungauged Basins — LSTMs demonstrate superior performance over encoder-only Transformers for upstream streamflow prediction, especially when augmented with downstream hydrologic context.
• Learning the coupled dynamics of global climate modes — UniCM unifies global climate mode forecasting by learning coupled ocean-atmosphere dynamics and inter-mode interactions, revealing emergent predictability.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The latest research from Google.