AI Predicts Natural Disasters: Google Launches "Groundsource" Framework, Using Gemini to Transform Global News into 2.6 Million Life-Saving Data Points

Google officially announced a new innovative technology framework called “Groundsource” on March 12. This framework cleverly leverages the powerful understanding capabilities of the Gemini AI model to automatically convert complex global news reports into structured historical disaster data. The initial open-source database includes 2.6 million records of flash floods worldwide, laying a critical foundation for future climate research and early warning systems.
(Background: Google’s Gemini 3 Deep Think upgrade: reasoning ability surpasses Opus 4.6, GPT-5.2, aiming to create the “most research-capable AI”)
(Additional background: Wikipedia 25th anniversary announcement: licensing content to AI giants like Microsoft, Google, Amazon for “training authorization”)

Table of Contents

Toggle

• Overcoming the global data scarcity challenge
• First open-source release: 1.5 billion records of flash floods across 150 countries
• Future expansion to more disaster predictions

As global climate change becomes increasingly extreme, artificial intelligence (AI) demonstrates enormous potential to save lives. Google Research announced on March 12, Taipei time, the launch of “Groundsource,” a new scalable data extraction framework powered by Google’s flagship large language model Gemini.

Overcoming the global data scarcity challenge

Natural disasters pose ongoing threats to populations and economies worldwide, causing millions of victims and billions of dollars in direct economic losses each year. Google scientists stated in a blog that advancing climate research, building accurate hydrological models, and providing timely disaster warnings all depend on having “robust historical baseline data.”

However, in reality, comprehensive historical data on various natural disasters is often extremely scarce and scattered globally. To address this issue, the Groundsource framework was developed. It uses Gemini’s advanced natural language processing capabilities to extract verified “ground truth” data from vast amounts of unstructured global news reports and online information, accurately reconstructing the historical footprints of disasters with unprecedented precision.

First open-source release: 1.5 billion records of flash floods across 150 countries

To demonstrate Groundsource’s powerful capabilities, Google released its first practical result: an open-source database focused on “urban flash floods.”

This database is remarkably extensive, containing records of over 2.6 million historical flood events across more than 150 countries worldwide. Google states that the database is now fully open-source, aiming to provide a reliable, high-quality data source for scientists and relevant organizations globally, helping governments make more accurate decisions in urban planning, insurance assessment, and emergency response.

Future expansion to more disaster predictions

This groundbreaking progress is not only a key step in Google’s Crisis Resilience strategy but also sets a new benchmark for non-generative applications of generative AI. The Google team emphasized that the Groundsource methodology is highly scalable and can be extended to build historical data sets for other natural disasters such as earthquakes and wildfires, further accelerating the global effort to enhance climate resilience.