The resulting dataset spans more than 150 countries and focuses on flash floods in urban areas, where reliable historical records have often been missing. By converting written reports into structured data, Google created what it describes as the first large-scale dataset designed specifically for training machine learning models to predict flash flooding.
Using the Groundsource dataset as a baseline, researchers trained a forecasting system built on a Long Short-Term Memory (LSTM) neural network. The model analyzes global weather forecasts and estimates the probability that flash flooding could occur in a specific location.
Google has begun deploying the forecasts through Flood Hub, the company’s public platform for flood risk information. The new urban flash flood forecasts are now available alongside Google’s existing riverine flood predictions, which cover 2 billion people across more than 150 countries.
Emergency response organizations are already testing the system. António José Beleza, an emergency response official with the Southern African Development Community, said the forecasting tool helped his organization respond to floods more quickly during trials.
Despite the progress, the system still has limitations. The model identifies potential flood risk across areas measuring about 20 square kilometers, which makes it less precise than local warning systems such as those used by the U.S. National Weather Service, which rely on radar data to track rainfall in real time.
Google designed the model partly for regions where governments lack the resources to maintain extensive meteorological infrastructure. By using publicly available reports as input data, the system can generate forecasts in places that previously lacked sufficient monitoring systems.
“Because we’re aggregating millions of reports, the Groundsource dataset actually helps rebalance the map,” said Juliet Rothenberg, a program manager on Google’s Resilience team. “It enables us to extrapolate to other regions where there isn’t as much information.”
Researchers say the same approach could also be used to build datasets for predicting other difficult-to-measure disasters, including heat waves and landslides. Marshall Moutenot, chief executive of Upstream Tech, said assembling large datasets remains one of the biggest challenges in applying machine learning to geophysical forecasting.
“Data scarcity is one of the most difficult challenges in geophysics,” Moutenot said. “Simultaneously, there’s too much Earth data, and then when you want to evaluate against truth, there’s not enough. This was a really creative approach to get that data.”
This analysis is based on reporting from Google.
Image courtesy of Google.
This article was generated with AI assistance and reviewed for accuracy and quality.
