Building AI-based wavefield-forecasting framework for earthquake early warning: collaborative research with international computer science institute and university of california berkeleyproject…

The International Computer Science Institute (ICSI) received a $100,000 Project Grant award from the U.S. Geological Survey under the Earthquake Hazards Program Assistance (CFDA 15.807) for the period September 1, 2025 through August 31, 2026. The award supports collaborative research with the University of California Berkeley to develop WaveCastNet-V2, an artificial intelligence-enabled framework designed to forecast earthquake ground motions in real-time earthquake early warning applications. The project builds upon ICSI's existing WaveCastNet model and adapts it to real earthquake observations in the San Francisco Bay Area.

Under this award, ICSI will deliver four primary products: (1) a robust AI-based framework (WaveCastNet-V2) capable of forecasting earthquake ground motion across a wide range of earthquake magnitudes; (2) a pre-trained AI model calibrated for San Francisco Bay Area earthquakes; (3) a curated database of real waveforms from actual earthquakes; and (4) a synthetic waveform database for model training and validation. The project activities include collecting real observations and simulating ground motions using the San Francisco Bay Region 3D seismic velocity model, developing the novel WaveCastNet-V2 framework, and conducting extensive testing and validation with both synthetic and real waveforms. The framework aims to reduce false warnings and missing alerts inherent in traditional earthquake early warning methods by improving accuracy in rapid magnitude and location estimation.

Activities

(1) collecting real observations in sfba and simulate ground motions using the SAN francisco BAY region 3D seismic velocity model V21.1 (2) developing a novel AI-enabled framework wavecastnet-V2 (3) extensive testing and validation with synthetic and real waveforms for small and moderate earthquakes.deliverables and expected outcomes: the recipient plans to provide (i) robust AI-based framework, wavecastnet-V2 for earthquake ground motion forecasting that are applicable for a wide range of earthquake magnitudes (II) a pre-trained AI model for sfba earthquakes (III) a curated real waveform database for earthquakes (IV) a synthetic waveform database used for training and validation. The recipient provides practical theory and novel AI models that will be applicable to real data.

Intended beneficiaries

This research will be fundamental for building an AI-based earthquake early warning system for the real world where only irregular sparse measurements are available. The recipient expects that wavecastnet-V2 will be capable of forecasting wavefield from real observations of small to moderate magnitude earthquakes in sfba. Once the model is fine-tuned, the recipient expects that the proposed deep learning approach is significantly fast at the order of a few seconds.