APPLES-to-APPLES: Optimizing and benchmarking a ground-motion algorithm for ShakeAlert

Elizabeth S. Cochran, Jessie K. Saunders, Julian Bunn, Colin O'Rourke, Sarah E. Minson, Annemarie S. Baltay, Clara E. Yoon, & Tim Clements

Submitted August 30, 2026, SCEC Contribution #15134, 2026 SCEC Annual Meeting Poster #TBD

Earthquake early warning (EEW) relies on rapid, accurate ground motion predictions to provide warning of potentially damaging shaking. The U.S. ShakeAlert system currently implements algorithms that characterize an earthquake’s source (location, magnitude), but direct ground-motion forecasting approaches like the Propagation of Local Undamped Motion (PLUM), adapted for the U.S. West Coast as APPLES, offer a robust complementary strategy. For example, previous work has shown APPLES to provide faster warning for near-source regions and to perform well during complex ruptures and aftershock sequences.

During 2025 and 2026, we streamlined the APPLES algorithm to directly address ShakeAlert’s priorities for detection of moderate and larger (M5+) earthquakes and to minimize bias in ground motion predictions. Key modifications to APPLES achieved near-zero prediction bias across the West Coast dataset and reduced alerts for small-magnitude earthquakes by requiring ground-motion confirmations across multiple grid cells. Additionally, code optimizations increased prediction speeds by ~10%, reduced the volume of alert updates, and aligned output metrics with current ShakeAlert production standards. Alongside these structural improvements, we provide an update on the real-time performance of the algorithm that is operating on the ShakeAlert development servers, including ground shaking predictions during recent earthquake detections.

Current efforts are focused on benchmarking APPLES against the production ShakeAlert algorithms using the standard System Testing and Performance (STP) approach. STP replays a suite of historic moderate to large earthquakes, synthetic events, and anomalous signals to stress-test the EEW algorithms and evaluate whether proposed changes improve system warning times and accuracy. We present the initial results of these STP matching assessments, evaluating APPLES using established ShakeAlert metrics to measure overall event detection reliability, alert latency, and the accuracy of ground-shaking forecasts.

Key Words
earthquake early warning

Citation
Cochran, E. S., Saunders, J. K., Bunn, J., O'Rourke, C., Minson, S. E., Baltay, A. S., Yoon, C. E., & Clements, T. (2026, 08). APPLES-to-APPLES: Optimizing and benchmarking a ground-motion algorithm for ShakeAlert. Poster Presentation at 2026 SCEC Annual Meeting.

Related Projects & Working Groups
Seismology