Google has created the largest flash flood dataset by utilizing its Gemini AI to analyze two decades of global news reports, addressing a critical gap in flash flood forecasting. This innovative dataset comprises 2.6 million historical flood events across over 150 countries, making it publicly accessible for further research. The newly developed AI model, trained on this dataset, predicts urban flash floods up to 24 hours in advance, supplementing Google’s Flood Hub, which already informs around 2 billion individuals about river flooding globally.
The challenge was significant, as traditional sensors for urban flooding were lacking. Google transformed historical news articles into geolocated data that trains the AI model using LSTM neural networks to assess flood risks based on various local factors. Although the system has limitations, initial results have shown its effectiveness in aiding humanitarian responses during flash flood alerts, thereby enhancing community resilience against natural disasters.
Source link
