Question-driven ensembles of flexible ETAS models

Leila Mizrahi, Shyam Nandan, William H. Savran, Stefan Wiemer, & Yehuda Ben-Zion

Published September 11, 2022, SCEC Contribution #12439, 2022 SCEC Annual Meeting Poster #208

We propose a question-driven ensemble (QDE) modeling approach to combine variants of ETAS models. With this approach, we aim to simultaneously address the goals of producing improved earthquake forecasts, and of gaining new insights into the physical processes involved.

First, we describe flexible ETAS models, which use nonparametric formulations of aftershock productivity and background seismicity. Both productivity and background rates are calibrated with data such that their variability is optimally represented by the model.

We then create a suite of 64 QDE models consisting of flexible ETAS variants, and test them in pseudo-prospective forecasting experiments for Southern California and Italy.

QDE models are obtained by combining the parameters of their ingredient models. The rules to combine parameters are driven by questions about the number of expected future events, and the time and location of background events and aftershocks. A QDE model can be viewed as a model which answers different questions with different ingredient models.

Some model combinations are able to clearly outperform all their ingredient models. We also find striking similarities between the results of the two regions: The same model combinations tend to answer the same questions well in Southern California and Italy.

Key Words
ETAS, flexible ETAS, ensemble modeling, forecasting experiments

Citation
Mizrahi, L., Nandan, S., Savran, W. H., Wiemer, S., & Ben-Zion, Y. (2022, 09). Question-driven ensembles of flexible ETAS models. Poster Presentation at 2022 SCEC Annual Meeting.


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