Short-term Bayesian ETAS spatiotemporal forecasting of the Ölfus 2008 earthquake sequence in Iceland

• Published in: Tectonophysics
• Main Authors: Darzi, Atefe, Halldorsson, Benedikt, Hrafnkelsson, Birgir, Vogfjorð, Kristín
• Format: Article in Journal/Newspaper
• Language: English
• Published: Zenodo 2022
• Subjects: Operational earthquake forecasting; Spatiotemporal ETAS; Aftershock sequence; Bayesian estimation; Iceland
• Online Access: https://doi.org/10.1016/j.tecto.2022.229522

trong earthquakes occurring frequently in south Iceland are always characterized by intense aftershock activity. The aftershock sequences cause progressive damages to structures and considerable psychological strain on inhabitants as the threat of upcoming large earthquakes looms. In this study, we apply a Bayesian epidemiological spatiotemporal aftershock sequence (ETAS) clustering model to forecast the aftershock sequence that commenced with the damaging 29 May 2008 M w 6.3 Ölfus earthquake. The forecast is carried out manually but in a framework of a provisional operational earthquake forecasting (OEF) system and based on an informative set of ETAS parameters derived from the strong June 2000 sequence in the region. We show that the short-term forecasts are unreliable at the early stages of the ongoing seismic sequence due to incomplete aftershocks collected to feed the Bayesian updating procedure of the ETAS model parameters, mainly during the first hours after the Ölfus mainshock. In contrast, issuing 24-h long forecasts, starting 6–8h after the mainshock gives the best forecasting results. We allow the daily forecasts to adapt to the ongoing 2008 sequence using the adaptive prior estimation method, i.e., by assigning an informative prior distribution to ETAS parameters for the subsequent forecasting interval based on the preceding ETAS parameter posterior distributions. The parametric posteriors do not show narrow distributions, thus indicating inconsistent forecasting performance, both in stark contrast to the June 2000 sequence. Regardless, our results provide a quantitative estimate of the reliability of the forecasts which is a prerequisite for evaluating their usefulness to civil protection.