Google’s Groundsource AI Predicts Natural Disasters with Gemini

Google has launched Groundsource, a latest AI-powered methodology leveraging the Gemini model to transform unstructured public news reports into actionable historical data for predicting natural disasters. Initially focused on urban flash floods, the system aims to provide early warnings and improve crisis resilience globally.

Addressing the Data Gap in Disaster Prediction

Historically, a significant challenge in predicting rapid-onset natural disasters like flash floods has been the lack of high-fidelity historical data. Without comprehensive records of past events, training effective AI models has been tricky. Groundsource directly addresses this data gap by extracting verified information from news coverage.

How Groundsource Works

Developed by Google software engineers Oleg Zlydenko and Rotem Mayo and research scientist Deborah Cohen, Groundsource functions as a framework for extracting “ground truth” from unstructured text. The process involves several key steps:

• Data Collection: Groundsource analyzes news reports where a disaster is the primary subject, using the Google Read Aloud user-agent to isolate article text.
• Multilingual Support: News coverage in 80 languages is translated into English using the Cloud Translation API.
• AI-Powered Extraction: The Gemini large language model classifies, dates, and locates flood events reported from 2000 onward, adhering to a structured schema.
• Filtering and Validation: Gemini filters out reports focused on forecasts, policy discussions, or general risk, focusing solely on accounts of actual or ongoing floods. It also resolves relative time phrases and maps locations to standardized geographic polygons using Google Maps Platform.

The Urban Flash Flood Dataset

The first output of the Groundsource methodology is an open-source urban flash flood dataset containing 2.6 million records from over 150 countries. This dataset is available via the Zenodo repository as open access. The methodology and validation results are detailed in a dedicated research paper.

Expanding Beyond Flash Floods

Google highlights the potential for applying the Groundsource approach to other natural disasters, including landslides and heat waves. By converting verified public reports into datasets, the system can contribute to improved global resilience. As stated by Google, the same AI-driven approach can turn verified reports from around the world into datasets that enable improved global resilience [1].

Availability and Integration with Flood Hub

The Urban Flash Flood model and dataset are now integrated with Google’s Flood Hub, alongside existing riverine flood forecasts. The system can forecast floods up to 24 hours in advance, providing crucial time for communities to prepare. Google CEO Sundar Pichai announced the launch on X, emphasizing the importance of addressing the flash flood data gap .

Key Takeaways

• Groundsource uses AI to extract historical disaster data from news reports.
• The initial focus is on urban flash floods, with a dataset of 2.6 million events.
• The system can predict flash floods up to 24 hours in advance.
• The methodology is open-source and has the potential to be applied to other natural disasters.