← Back to Sessions

A Prospective Evaluation of a Publicly Released ETAS Aftershock Forecast During the 2020 SW Puerto Rico Sequence

Session: Constructing and Testing Regional and Global Earthquake Forecasts I

Type: Oral

Date: 4/20/2021

Presentation Time: 10:30 AM Pacific

Description: 

The magnitude 6.4 earthquake that struck the SW coast of Puerto Rico on January 7, 2020, was accompanied by a robust fore- and aftershock sequence, with 12 magnitude 5 and larger aftershocks recorded in the first year of the sequence. The USGS issued an automated Reasenberg and Jones (RJ) aftershock forecast [Reasenberg and Jones, 1989] following the first magnitude 5.0 foreshock on 29 December 2019, and switched to a Epidemic-Type Aftershock Sequence (ETAS) [Ogata, 1988] forecast on 11 January 2020 as it became clear that this was an unusually productive sequence with many large aftershocks. The ETAS forecasts were generated using the Aftershock Forecaster software, developed in OpenSHA. The USGS additionally released an open-file report [van der Elst et al., 2020] on the potential duration of the aftershock sequence, based on an extrapolation of the sequence as of 17 January 2020. These products were disseminated along with press releases and infographics via the USGS website and social media. Over the last year, the USGS public forecasts have been updated on a daily to biweekly basis, allowing for a public, prospective test of the forecast methodology. Here we evaluate the performance of the regularly updated daily, weekly, and monthly ETAS forecasts, published on the USGS website, with respect to the sequence as observed. We also compare the evolution of the forecasts to the expectations published in the Open-File report. We supplement the analysis with a retrospective comparison between the ETAS and RJ forecast models. While the ETAS-based forecast performed reasonably well, the exercise highlights issues related to non-stationarity of model parameters, which are possibly linked to network and catalog changes made in response to the earthquake and are likely to confound even the most sophisticated statistical seismicity models.

Presenting Author: Nicholas van der Elst

Student Presenter: No

Authors