The announcement, published on March 12, 2026, was authored by Yossi Matias, Google's Vice President of Engineering and Research and head of the company's Crisis Response initiative.
What is Groundsource and how does it work?
Groundsource is an AI-powered methodology developed by Google Research. It uses Google's Gemini model to comb through decades of public reports and extract verified historical records of disaster events. For its initial application, the system identified more than 2.6 million historical flood incidents spanning more than 150 countries. Google Maps was subsequently used to assign precise geographic boundaries to each recorded event, refining the data into a focused flash flood dataset.That dataset was then used to train a new AI model designed to forecast flash floods specifically in urban environments, a category of disaster that has long been difficult to predict because of the lack of organized historical data.
Why urban flash floods present a unique challenge
Unlike riverine floods, which follow predictable drainage patterns and have been studied at scale, urban flash floods are rapid, localised, and highly variable. Google noted that the absence of high-fidelity historical data for such events had previously blocked efforts to build predictive models. Groundsource was designed explicitly to address that gap.Where forecasts are available
Urban flash flood predictions generated by the new model are now live on Google's Flood Hub, the company's existing public-facing platform for flood risk information. The Flood Hub already delivers riverine flood forecasts that serve approximately 2 billion people across more than 150 countries. The addition of urban flash flood forecasting represents a significant expansion of the platform's scope.The Urban Flash Floods model and its dataset have been formally integrated into Google's Earth AI portfolio of geospatial models and data resources. Both have been released as open-source tools, making them available to scientists, emergency management agencies, and humanitarian organisations, particularly those working in urban areas that have historically lacked reliable flash flood records.
Broader applications beyond flooding
Google indicated that the underlying logic of Groundsource is not restricted to floods. The company said the same methodology, using AI to convert verified public reports into structured disaster datasets, could be applied to other types of natural hazards, including landslides and heat waves. The ambition, as stated by the company, is to contribute to a goal in which no community is caught off guard by a natural disaster.Part of Google's wider Crisis Resilience effort
Groundsource sits within Google's broader Crisis Resilience programme, which has been working to deploy AI in support of disaster preparedness and early warning systems. The methodology represents a shift from relying solely on purpose-built sensor networks or government monitoring systems, instead treating the accumulated mass of public reporting as a source of scientific data in its own right.For emergency planners, local governments, and disaster response organisations, the release of an open-source urban flash flood benchmark dataset addresses a long-standing analytical gap, and offers a foundation on which new forecasting tools can be built.
I truly appreciate you spending your valuable time here. To help make this blog the best it can be, I would love your feedback on this post. Let me know in the comments: How could this article be better? Was it clear? Did it have the right amount of detail? Did you notice any errors?
If you found any of the articles helpful, please consider sharing it.