Real-time forecasts of tomorrow's earthquakes in California

Abstract

Despite a lack of reliable deterministic earthquake precursors, seismologists have significant predictive information about earthquake activity from an increasingly accurate understanding of the clustering properties of earthquakes. In the past 15 years, time-dependent earthquake probabilities based on a generic short-term clustering model have been made publicly available in near-real time during major earthquake sequences. These forecasts describe the probability and number of events that are, on average, likely to occur following a mainshock of a given magnitude, but are not tailored to the particular sequence at hand and contain no information about the likely locations of the aftershocks. Our model builds upon the basic principles of this generic forecast model in two ways: it recasts the forecast in terms of the probability of strong ground shaking, and it combines an existing time-independent earthquake occurrence model based on fault data and historical earthquakes with increasingly complex models describing the local time-dependent earthquake clustering. The result is a time-dependent map showing the probability of strong shaking anywhere in California within the next 24 hours. The seismic hazard modelling approach we describe provides a better understanding of time-dependent earthquake hazard, and increases its usefulness for the public, emergency planners and the media.

Aftershock Characteristics as a Means of Discriminating Explosions from Earthquakes

“…The probabilistic aftershock rate model of Reasenberg and Jones (1989;1994) and Gerstenberger et al (2005) provide the perfect tool to make such a discriminant. The model calculates the probability of an aftershock greater than magnitude M at a given time t after the mainshock with magnitude Mm.…”

“…Therefore, in order to simplify the analysis we only look at events greater than Mc to define the aftershock distribution. The aftershock parameters (b, p, c, and a) calculated by Reasenberg and Jones (1989) and used by the real-time hazard analysis of USGS (Gerstenberger et al, 2005) are in Table 1 (SoCal parameters).…”

Adaptive Smoothing of Seismicity in Time, Space, and Magnitude for Time-Dependent Earthquake Forecasts for California

“…Similarly, whereas the GutenbergRichter law with exponent b ≈ 1 approximates the magnitude distribution in sufficiently large space-time volumes well (e.g., Kagan, 1997), b-values have been claimed to vary at smaller scales (e.g., Wiemer et al, 1998;Wiemer and Wyss, 2002). One approach to account for these seismicity fluctuations in models is to continuously recalibrate Omori's p-value and/or Gutenberg-Richter's b-value to the data to improve forecast performance (e.g., Gerstenberger et al, 2005;Wiemer and Schorlemmer 2007). For instance, in the STEP model (Gerstenberger et al, 2005), the parameters p and b are inverted for during aftershock sequences.…”

“…One approach to account for these seismicity fluctuations in models is to continuously recalibrate Omori's p-value and/or Gutenberg-Richter's b-value to the data to improve forecast performance (e.g., Gerstenberger et al, 2005;Wiemer and Schorlemmer 2007). For instance, in the STEP model (Gerstenberger et al, 2005), the parameters p and b are inverted for during aftershock sequences.…”

“…These models usually assume that the seismicity rate is the sum of a background rate (usually heterogeneous in space and stationary in time) and of triggered earthquakes. This class of models includes, among others, the point-process model of Kagan and Knopoff (1987), the epidemic-type aftershock sequence (ETAS) model (Ogata, 1988;Helmstetter et al 2006), the STEP model (Gerstenberger et al, 2005), and the model of Marsan and Lengliné (2008).…”