Google Uses Gemini and Maps Data to Push Urban Flash Flood Forecasts 24 Hours Ahead

Google is trying to solve the hardest part of flood forecasting: the missing data layer

On March 12, 2026, Google introduced Groundsource, a new Gemini-powered methodology designed to improve prediction of urban flash floods. The company says the system analyzes public reports and combines them with Google Maps to reconstruct the geographic boundaries of past flood events. That new dataset was then used to train a model that can forecast urban flash floods up to 24 hours ahead, with forecasts now available through Flood Hub.

The importance of this announcement is not only in the model itself, but in the way Google is filling a long-standing data gap. Riverine flood forecasting benefits from physical gauges and historical water measurements. Urban flash floods are different. They emerge quickly, can be highly localized, and often lack the structured historical records needed to train reliable machine learning systems at scale. Google’s pitch is that AI can help turn messy public information into a usable geospatial training set.

Groundsource converts public information into machine-learning-ready flood history

According to Google, Groundsource used Gemini to analyze decades of public reports and identify more than 2.6 million historical flood events across more than 150 countries. It then used Google Maps to estimate more precise geographic boundaries for each event, producing a dataset focused on urban flash floods. That is a meaningful change in method. Instead of waiting for traditional observation systems to cover every city and region, Google is trying to synthesize a record of past disasters from publicly available information and map-based context.

That matters because in many applied AI problems, the bottleneck is not the neural network architecture but the lack of reliable ground truth. By transforming public reports into structured data, Google is effectively building the prerequisite layer for prediction. The company is arguing that frontier AI can do more than classify or summarize information after the fact; it can help create the datasets that make new forecasting systems possible.

Flood Hub expands from riverine floods toward urban flash floods

Google said the new urban flash flood forecasts are available in Flood Hub, alongside its existing riverine flood forecasts. The company says those riverine forecasts already cover more than 2 billion people across more than 150 countries for the most significant flood events. Adding urban flash flood forecasts extends that effort into a class of disasters that often unfolds much faster and leaves less time for response.

In practice, that makes the 24 hours lead time significant. Riverine floods often build over longer periods, giving authorities more time to mobilize. Flash floods can disrupt transportation, housing, and local commerce within hours. If early-warning systems can move from reactive monitoring toward a day-ahead forecast, emergency agencies and communities gain a better chance to stage equipment, redirect travel, and reduce risk before the most dangerous period begins.

Why this matters beyond flooding

Google also said the Groundsource approach could be extended to other hazards, including landslides and heat waves. That suggests the larger story is not a single flood model, but a broader method for turning dispersed public evidence into global disaster datasets. The next questions are empirical ones: how accurate the urban flash flood forecasts prove to be in real deployments, how quickly coverage expands, and whether the same data-building approach works for other hard-to-observe climate and disaster events.

Sources: Google · Google Research