1 of 1

WaveCastNet: An AI-enabled Wavefield Forecasting �Framework for Earthquake Early Warning

Scientific Achievement

We develop WaveCastNet, an AI-based framework, designed to enhance earthquake early warning systems. WaveCastNet provides rapid predictions that are robust and highly generalizable across different seismic scenarios, including higher magnitude earthquakes, when evaluated on simulated data.

Significance and Impact

Earthquakes pose significant risks to humans and critical infrastructure in seismically active regions. To mitigate these threats, early warning systems have been developed. Our work demonstrates that AI-enabled systems can substantially reduce the computational costs associated with forecasting, thereby increasing the warning time and enhancing the overall effectiveness of these systems in protecting communities.

Technical Approach

WaveCastNet is based on a sequence to sequence approach, which effectively models long-term dependencies, and multi-scale patterns.
We use powerful attention-based embedding and reconstruction layers to handle both dense and sparse measurements.
We use an ensemble to produce uncertainty estimates.

WaveCastNet produces highly accurate forecasts for wavefields, while drastically reducing the inference time, when compared to traditional methods for early warning. Importantly, WaveCastNet is able to generalize to (rare) higher magnitude earthquakes.

PI(s)/Facility Lead(s): Lenny Oliker (LBL)

Collaborating Institutions: ICSI, UC Berkeley, Lawrence Livermore National Laboratory

ASCR Program: SciDAC RAPIDS2

ASCR PM: Kalyan Perumalla (SciDAC RAPIDS2)

LOCAL LAB POC: Ben Erichson

TALKING POINTS:

To mitigate immediate threats posed by large earthquakes, early warning systems issue alerts to provide time for precautions.
The success of these systems depends on the fast and accurate prediction of earthquake ground motion intensities.
We propose WaveCastNet, a novel AI-enabled framework, for forecasting wavefields (which represent ground motions).
Our approach drastically reduces inference times compared to more resource-intensive models like transformers and traditional numerical methods, while improving forecasting accuracy.
We demonstrate the performance on simulated data from the San Francisco Bay Area.
Importantly, we demonstrate that our framework generalizes to different seismic scenarios, including those with higher magnitude earthquakes.

METADATA:

Name of the associated awarded project: SciDac-5 Institutes RAPIDS2

PI name(s): Leonid Oliker

Name of the program manager: Kalyan Perumalla (RAPIDS)

CITATIONS:

Dongwei Lyu, Pu Ren, Michael W. Mahoney, Arben Pitarka, Nori Nakata, Rie Nakata and N. Benjamin Erichson. WaveCastNet: An AI-enabled Wavefield Forecasting Framework for Earthquake Early Warning. (under submission).