Google recently announced a fascinating new way to predict flash floods using decades-old news articles and cutting-edge language models. Juliet Rothenberg, a program manager on Google’s new Resilience team, is the force behind this innovative program. It’s the first time the tech giant has deployed such generative AI models in real-time for flood forecasting. Like all projects that advance resilience strategies, this one improves immediate flood response. Beyond that, it sets a precedent for forecasting other major weather disasters going forward.
Flash floods are the most deadly weather event worldwide, killing more than 5,000 people annually. Understanding this urgent need for reliable forecasting, Google has produced a model predicting risk on a 20-square-kilometer level. This is a model that António José Beleza, an emergency response official at the Southern African Development Community, has tested. He added that it dramatically shortens the time it takes to respond to flooding emergencies.
This project utilizes data from the Groundsource Opus dataset. It combines millions of these reports to provide a less alarmist and more comprehensive picture of flood threats. Rothenberg further commented on the importance of this dataset, saying,
“Because we’re aggregating millions of reports, the Groundsource data set actually helps rebalance the map. It enables us to extrapolate to other regions where there isn’t as much information.”
There is no shortage of challenges in geophysics, but data scarcity is perhaps the most crucial. Marshall Moutenot, CEO of Upstream Tech, speaks to this pressing challenge. His company uses deep learning models to predict river flows for clients including hydropower companies. Moutenot applauded Google’s smart approach to filling in gaps in their data.
“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,” he remarked.
This collaborative initiative has resulted in important research and data sets. We have released them publicly to demonstrate our ongoing commitment to an open and collaborative approach to improving weather forecasting. Rothenberg said that the team is indeed excited to use similar methodologies. Similar partnerships build datasets for other significant but temporary weather occurrences, such as heat waves and mudslides.
This pioneering effort reflects a broader trend within the tech industry to harness deep learning-based models for improved weather prediction. As organizations like Google strive to assemble comprehensive datasets for effective forecasting, the implications for emergency response and disaster management could be profound.