Abstract
Large earthquakes can be destructive and quickly wreak havoc on a landscape.To mitigate immediate threats, early warning systems have been developed toalert residents, emergency responders, and critical infrastructure operatorsseconds to a minute before seismic waves arrive. These warnings provide time totake precautions and prevent damage. The success of these systems relies onfast, accurate predictions of ground motion intensities, which is challengingdue to the complex physics of earthquakes, wave propagation, and theirintricate spatial and temporal interactions. To improve early warning, wepropose a novel AI-enabled framework, WaveCastNet, for forecasting groundmotions from large earthquakes. WaveCastNet integrates a novel convolutionalLong Expressive Memory (ConvLEM) model into a sequence to sequence (seq2seq)forecasting framework to model long-term dependencies and multi-scale patternsin both space and time. WaveCastNet, which shares weights across spatial andtemporal dimensions, requires fewer parameters compared to moreresource-intensive models like transformers and thus, in turn, reducesinference times. Importantly, WaveCastNet also generalizes better thantransformer-based models to different seismic scenarios, including to more rareand critical situations with higher magnitude earthquakes. Our results usingsimulated data from the San Francisco Bay Area demonstrate the capability torapidly predict the intensity and timing of destructive ground motions.Importantly, our proposed approach does not require estimating earthquakemagnitudes and epicenters, which are prone to errors using conventionalapproaches; nor does it require empirical ground motion models, which fail tocapture strongly heterogeneous wave propagation effects.