Generative AI Enables Real-time Probabilistic Tsunami Forecasting.

Traditional tsunami forecasting often provides deterministic predictions of onshore inundation, which can be misleading due to high uncertainties, especially for near-field tsunamis. This can lead to a false sense of security outside predicted boundaries, prompting current warnings to focus on coastal height rather than specific inundation. A new study introduces a probabilistic ensemble model built on a conditional diffusion model, a type of generative AI. This model aims to balance prediction accuracy with the crucial aspect of uncertainty quantification, moving beyond deterministic outputs. The framework was validated using data from the 2011 Tohoku-oki earthquake. Results show that the model effectively tracks the reduction of uncertainty over time following an earthquake, while also accurately forecasting both the depth and extent of tsunami inundation. This represents a significant step towards next-generation early warning systems.