Dongwei Lyu,Rie Nakata,Pu Ren,Michael W. Mahoney,Arben Pitarka,Nori Nakata,N. Benjamin Erichson
Abstract:Large earthquakes can be destructive and quickly wreak havoc on a landscape. To mitigate immediate threats, early warning systems have been developed to alert residents, emergency responders, and critical infrastructure operators seconds to a minute before seismic waves arrive. These warnings provide time to take precautions and prevent damage. The success of these systems relies on fast, accurate predictions of ground motion intensities, which is challenging due to the complex physics of earthquakes, wave propagation, and their intricate spatial and temporal interactions. To improve early warning, we propose a novel AI-enabled framework, WaveCastNet, for forecasting ground motions from large earthquakes. WaveCastNet integrates a novel convolutional Long Expressive Memory (ConvLEM) model into a sequence to sequence (seq2seq) forecasting framework to model long-term dependencies and multi-scale patterns in both space and time. WaveCastNet, which shares weights across spatial and temporal dimensions, requires fewer parameters compared to more resource-intensive models like transformers and thus, in turn, reduces inference times. Importantly, WaveCastNet also generalizes better than transformer-based models to different seismic scenarios, including to more rare and critical situations with higher magnitude earthquakes. Our results using simulated data from the San Francisco Bay Area demonstrate the capability to rapidly predict the intensity and timing of destructive ground motions. Importantly, our proposed approach does not require estimating earthquake magnitudes and epicenters, which are prone to errors using conventional approaches; nor does it require empirical ground motion models, which fail to capture strongly heterogeneous wave propagation effects.
What problem does this paper attempt to address?
This paper proposes an AI-driven seismic wavefield prediction framework called WaveCastNet, aimed at improving earthquake early warning systems. Existing warning systems rely on fast and accurate prediction of ground motion intensity, but this task is challenging due to the complexity of earthquakes, wave propagation, and spatio-temporal interactions. WaveCastNet adopts a sequence-to-sequence (seq2seq) architecture, combined with a novel Convolutional Long Expression Memory (ConvLEM) model, to capture long-term dependencies and multi-scale patterns in space and time. Compared to the resource-intensive Transformer model, WaveCastNet has fewer parameters, shorter inference time, and better generalization ability in different seismic scenarios.
WaveCastNet works through the encoder and decoder components, where the encoder compresses the input seismic wavefield sequence into a single state, and the decoder generates predictions of the target wavefield sequence based on this state. The ConvLEM unit enhances the modeling capability of multi-scale structures by integrating convolutional operations. The paper demonstrates the prediction results of WaveCastNet on simulated data, proving its effectiveness in predicting seismic wave propagation, especially in handling high-intensity seismic wavefields with good prediction accuracy and generalization performance. In addition, by using the ensemble of WaveCastNet, uncertainty estimation can be achieved, which is crucial for validating and demonstrating the reliability of the framework.
In the experimental section, the paper first showcases the performance of WaveCastNet in point source small-scale seismic scenarios, accurately predicting the velocity and arrival time of ground motion regardless of whether the input data is densely sampled or sparsely sampled. Next, the performance of WaveCastNet in simulating large-scale earthquakes with finite faults is studied, and it is found that the prediction accuracy decreases as the earthquake scale increases, possibly due to the insufficient input time window to capture the entire energy release process. By extending the input time window, the prediction performance is improved.
In summary, this paper proposes an efficient AI framework called WaveCastNet for earthquake early warning, which can predict the propagation of seismic waves and reduce dependence on earthquake parameter estimation, thereby improving the prediction accuracy and response speed of the warning system.