Google Introduces Groundsource AI Methodology To Predict Urban Flash Floods: Everything To Know

Hyderabad: Google has announced the launch of Groundsource, a new AI-powered methodology that aims to help predict flash floods before they happen. According to the World Meteorological Organisation (WMO), flash floods are responsible for about 85 per cent of flood-related fatalities worldwide.

Striking within just six hours of intense rainfall, they can transform city streets into raging rivers and claim more than 5,000 lives each year, making them among the deadliest natural disasters globally, Google said, highlighting the importance of early warning systems (EWS) for keeping communities safe and informed. However, the lack of high-fidelity data for certain disasters like flash floods simply did not exist, preventing the development of predictive AI models.

Regions in the world covered by Groundsource AI model. (Google Research Blog)

This is where Groundsource comes into the picture, as it transforms public information into a high-quality record of historical disaster data. Using this technology, Google has developed Urban Flash Flood forecasts, which are now available on Flood Hub along with the tech giant's existing riverine flood forecasts. The Urban Flash Floods model and dataset also join the Google Earth AI family of geospatial models and datasets.

Posting on X, Google CEO Sundar Pichai said, "We trained a new flood forecasting model designed to predict flash floods in urban areas up to 24 hours in advance. To help address a flash flood data gap, we created Groundsource: a new AI methodology using Gemini to identify 2.6M+ historical events across 150+ countries."

This chart illustrates the exponential growth of digitised news and the corresponding increase in flood events captured by the Groundsource pipeline, showing a significant density of data in recent years. (Google Research Blog)

The team at Google Research said that these predictions build on years of research and mark a significant breakthrough in their flood forecasting capabilities, and are basically an expansion of their flood coverage.

Groundsource used Gemini to analyse decades of public reports, identifying more than 2.6 million historical flood events across 150+ countries since the year 2000. Using Google Maps, it mapped the precise geographic boundaries of each event to build a dataset focused specifically on flash floods. With this dataset, the team trained a new model that demonstrates significant progress in predicting urban flash floods up to 24 hours in advance.

Top row: Precision (a) and recall (b) by country. Bottom row: Event count in Groundsource (c) and GDACS (d) by country. These graphs exclude countries with less than 10 ground truth events, as their metrics are very noisy. For a clear visual display, Google limits the precision and recall color scale to 50%. (Google Research Blog)

While communities around the world can access the tool to better prepare for disasters, Groundsource's open-source benchmark is available for scientists to scale their impact. Google says that the same AI-driven approach of Groundsource has the potential to be applied to other natural disasters, such as landslides and heat waves, by turning verified public reports into actionable datasets. "We’re open-sourcing this dataset to advance global research," Google CEO Pichai said.