NG24B-01 INVITED 17:10h
Can variations in precursory seismicity be used to forecast (predict) earthquakes?
There are basically two approaches to the initiation of a major earthquake rupture. The first is to consider the fault on which the earthquake will occur. A focus is on subcritical crack growth. Although rate-and-state friction predicts a precursory creep phase, there is no observational evidence for its occurrence. Also, if the fault dominates, one would expect a systematic increase in the background seismicity during the earthquake cycle. This is not observed. The second approach is to consider that the earth's crust as a thermodynamic system. The near constant background seismicity that satisfies Gutenberg-Richter scaling is a thermal noise. This background noise generates fluctuations in stress level that are responsible for the initiation of rupture on major faults. This approach explains the relatively large variations in wait times between great earthquakes. But a more important question is whether there are observational statistical patterns precursory to major ruptures. Analogies to phase changes indicate that systematic increases in fluctuation levels may be an observable precursory effect. There is also observational support for this in the success of forecasting algorithms such as M8 and pattern informatics and the observations of accelerated moment release before large earthquakes.
NG24B-02 INVITED 17:25h
Equilibrium, Non-Equilibrium and Critical Points; Universality in Models of Earthquake Faults
Due to the presence of scaling laws that describe earthquake fault system data (Gutenburg-Richter, Omori) it has long been thought that earthquakes have something to do with critical phenomena. However, spatial and temporal clustering, runup to failure as seen in, for example, the increase in Benioff strain and the success, albeit limited, of forecasting algorithms indicates that critical phenomena is not the entire story. Models that produce scaling are quite common however models that reproduce the wide range of behavior seen in fault systems are, at present, not available. We have pursued a long term study of various forms of models and have obtained some understanding of their physics. The purpose of this talk is to describe the complex behavior of a large class of models and to indicate how their behavior on several scales can aid in forecasting.
NG24B-03 INVITED 17:40h
Foreshock Sequences and Short-Term Earthquake Predictability on East Pacific Rise Transform Faults
A predominant view of continental seismicity postulates that all earthquakes initiate in a similar manner regardless of their eventual size and that earthquake triggering can be described by an Endemic Type Aftershock Sequence (ETAS) model [e.g. Ogata, 1988, Helmstetter and Sorenette 2002]. These null hypotheses cannot be rejected as an explanation for the relative abundances of foreshocks and aftershocks to large earthquakes in California [Helmstetter et al., 2003]. An alternative location for testing this hypothesis is mid-ocean ridge transform faults (RTFs), which have many properties that are distinct from continental transform faults: most plate motion is accommodated aseismically, many large earthquakes are slow events enriched in low-frequency radiation, and the seismicity shows depleted aftershock sequences and high foreshock activity. Here we use the 1996-2001 NOAA-PMEL hydroacoustic seismicity catalog for equatorial East Pacific Rise transform faults to show that the foreshock/aftershock ratio is two orders of magnitude greater than the ETAS prediction based on global RTF aftershock abundances. We can thus reject the null hypothesis that there is no fundamental distinction between foreshocks, mainshocks, and aftershocks on RTFs. We further demonstrate (retrospectively) that foreshock sequences on East Pacific Rise transform faults can be used to achieve statistically significant short-term prediction of large earthquakes (magnitude $\ge$ 5.4) with good spatial (15-km) and temporal (1-hr) resolution using the NOAA-PMEL catalogs. Our very simplistic approach produces a large number of false alarms, but it successfully predicts the majority (70%) of M$\ge$5.4 earthquakes while covering only a tiny fraction (0.15%) of the total potential space-time volume with alarms. Therefore, it achieves a large probability gain (about a factor of 500) over random guessing, despite not using any near field data. The predictability of large EPR transform earthquakes suggests that a fault preparation process with a characteristic times scale of hours exists and is observable through foreshocks. The predictability of EPR transform earthquakes, their abundant foreshock activity, and their predominance of aseismic fault slip are all consistent with a model in which slow slip transients trigger ordinary earthquakes, enrich their low-frequency radiation, and accommodate much of the subseismic plate motion.