A RELM Earthquake Forecast Based on Pattern Informatics

James R. Holliday, Chien-chih Chen, Kristy F. Tiampo, John B. Rundle, Donald L. Turcotte, & Andrea Donnellan

Published 2007, SCEC Contribution #941

There have been a wide variety of approaches applied to forecasting earthquakes (Turcotte 1991; Kanamori 2003). These approaches can be divided into two general classes. The first is based on empirical observations of precursory changes. Examples include precursory seismic activity, precursory ground motions, and many others. The second approach is based on statistical patterns of seismicity. Neither approach has been able to provide reliable short-term forecasts (days to months) on a consistent basis.

Although short-term predictions are not available, longer-term seismic-hazard assessments can be made. A large fraction of all earthquakes occur in the vicinity of plate boundaries, although some do occur in plate interiors. It is also possible to assess the long-term probability of having an earthquake of a specified magnitude in a specified region. These assessments are primarily based on the hypothesis that future earthquakes will occur in regions where past, typically large, earthquakes have occurred (Kossobokov et al. 2000). As we will discuss, a more promising approach is to begin with the hypothesis that the rate of occurrence of small earthquakes in a region can be analyzed to assess the probability of occurrence of much larger earthquakes.

The Regional Earthquake Likelihood Models (RELM) forecast described in this paper is primarily based on the pattern informatics (PI) method (Rundle et al. 2002, 2003; Tiampo et al. 2002a, 2000c). This method identifies regions of strongly correlated fluctuations in seismic activity. These regions are the locations where subsequent large earthquakes have been shown to occur, therefore indicating a strong association with the high stress preceding the main shock. The fluctuations in seismicity rate revealed in a PI map may be related to the preparation process for large earthquakes. Seismic quiescence and seismic activation (Bowman et al. 1998; Wyss and Habermann 1988), which are revealed by the PI map, are examples of such preparation processes. The PI method identifies the existence of correlated regions of seismicity in observational data that precede the main shock by months and years. The fact that this correlated region locates the aftershocks as well as main shocks leads us to identify this region of correlated seismicity with the region of correlated high stress (Tiampo et al. 2002a, 2002b, 2002c).

The PI method does not predict earthquakes; rather it forecasts the regions (hot spots) where earthquakes are most likely to occur in the relatively near future (typically five to 10 years). The objective is to pinpoint more narrowly the areas of earthquake risk relative to those given by long-term hazard assessments. The result is a map of areas in a seismogenic region (hot spots) where earthquakes are likely to occur during a specified period in the future. In this paper a PI map is combined with historic seismicity data to produce a map of probabilities for future large events. These probabilities can be further converted, using Gutenberg-Richter scaling laws, to forecast rates of occurrence of future earthquakes in specific magnitude ranges. This forecast can be evaluated using the RELM likelihood test. In the following sections we present details of the PI method and the procedure for producing a composite forecast map. A discussion on binary forecasts and forecast verification techniques is provided in appendixes A and B.

Key Words
United States, California, patterns, geological hazards, earthquake prediction, seismic risk, intensity, risk assessment, algorithms, Regional Earthquake Likelihood Models, earthquakes

Holliday, J. R., Chen, C., Tiampo, K. F., Rundle, J. B., Turcotte, D. L., & Donnellan, A. (2007). A RELM Earthquake Forecast Based on Pattern Informatics. Seismological Research Letters, 78(1), 87-93. doi: 10.1785/gssrl.78.1.87.