Evaluation of 10 Years of UCERF3-ETAS Next-day Forecasts

Francesco Serafini, Maximilian J. Werner, Fabio Silva, Philip J. Maechling, Kevin R. Milner, & Edward H. Field

Submitted September 7, 2025, SCEC Contribution #14961, 2025 SCEC Annual Meeting Poster #TBD

Earthquake forecasting models are routinely used for probabilistic seismic hazard analysis (PSHA) and operational earthquake forecasting (OEF) operations. Consequently, rigorously evaluating the consistency between forecasts and observations is crucial to understand a model's performance and provide guidance on potential improvements. Here, we evaluate next-day retrospective seismicity forecasts for California produced with UCERF3-ETAS developed by USGS. UCERF3-ETAS incorporates hypotheses of fault rupture dynamics and elastic rebounding with an Epidemic-Type Aftershock Sequence (ETAS) model for short-term seismicity, and is used both for long-term and short-term tasks. The forecasts cover a 10-year period going from August 1, 2008, to August 31, 2018, and each forecast comprises 100,000 synthetic catalogues of Mw ≥ 2.5 earthquakes, between 00:00:00 and 23:59:59 UTC of each day, and is conditioned on all earthquakes that occurred prior to midnight. During the 2008-2018 period 15,227 earthquakes occurred with Mw ≥ 2.5, with the most relevant sequence being the 2010 Mw 7.2 El-Mayor Cucapah sequence. Evaluating this extensive set of forecasts not only provides insights into the model's ability to accurately forecast daily seismicity, but also to understand which components drive the model in active and quiet periods, and the interplay between them. Analysing the daily number of events distribution, we found that UCERF3-ETAS is mostly consistent with the observations except during the 2014 Sheldon swarm in northwest Nevada, suggesting more adaptive model parameterizations may improve model performance. We assess consistency of the magnitude distribution differentiating between on-fault and off-fault seismicity using a novel magnitude test. We compare UCERF3-ETAS with models in the Next-day CSEP gridded-forecasts database which constitutes a benchmark for this problem using CSEP comparisons tests. UCERF3-ETAS outperforms all models in the database in most periods highlighting the improvements made in model development. These findings demonstrate how rigorous retrospective evaluation can guide targeted refinements to OEF models, ultimately enhancing their reliability for real-time decision-making.

Key Words
CSEP, forecast evaluation, next-day modelling

Citation
Serafini, F., Werner, M. J., Silva, F., Maechling, P. J., Milner, K. R., & Field, E. H. (2025, 09). Evaluation of 10 Years of UCERF3-ETAS Next-day Forecasts. Poster Presentation at 2025 SCEC Annual Meeting.

Related Projects & Working Groups
Earthquake Forecasting and Predictability (EFP)