Fixed recurrence and slip models better predict earthquake behavior than the time‐ and slip‐predictable models: 1. Repeating earthquakes

Abstract

[1] The behavior of individual events in repeating earthquake sequences in California, Taiwan and Japan is better predicted by a model with fixed inter-event time or fixed slip than it is by the time-and slip-predictable models for earthquake occurrence. Given that repeating earthquakes are highly regular in both inter-event time and seismic moment, the time-and slip-predictable models seem ideally suited to explain their behavior. Taken together with evidence from the companion manuscript that shows similar results for laboratory experiments we conclude that the short-term predictions of the time-and slip-predictable models should be rejected in favor of earthquake models that assume either fixed slip or fixed recurrence interval. This implies that the elastic rebound model underlying the time-and slip-predictable models offers no additional value in describing earthquake behavior in an event-to-event sense, but its value in a long-term sense cannot be determined. These models likely fail because they rely on assumptions that oversimplify the earthquake cycle. We note that the time and slip of these events is predicted quite well by fixed slip and fixed recurrence models, so in some sense they are time-and slip-predictable. While fixed recurrence and slip models better predict repeating earthquake behavior than the time-and slip-predictable models, we observe a correlation between slip and the preceding recurrence time for many repeating earthquake sequences in Parkfield, California. This correlation is not found in other regions, and the sequences with the correlative slip-predictable behavior are not distinguishable from nearby earthquake sequences that do not exhibit this behavior.Citation: Rubinstein, J. L., W. L. Ellsworth, K. H. Chen, and N. Uchida (2012), Fixed recurrence and slip models better predict earthquake behavior than the time-and slip-predictable models: 1.

Fixed recurrence and slip models better predict earthquake behavior than the time‐ and slip‐predictable models: 2. Laboratory earthquakes

“…This test relies upon the assumption that a sequence that is perfectly described by the time(slip)‐predictable model will not appear to be fit by these same models if reshuffled. Synthetic tests using perfectly time‐ or slip‐predictable data for sequences of 10 events and longer (all the experiments we study are longer than 10), show that no more than 13 percent of the time there is a false positive [ Rubinstein et al , 2012]. Thus, we can be confident that this test accurately demonstrates whether our observations of the relationship between inter‐event time and slip come from the data distribution or a real connection between these two quantities.…”

“…We also note in the companion manuscript [ Rubinstein et al , 2012] that the requirement of the time‐ and slip‐predictable models for a constant loading rate was too strict and is another reason they do not offer additional value in predicting earthquake behavior. While we have good reason to believe that there is a variable loading rate for natural repeating earthquakes [e.g., Chen and Lapusta , 2009], we impose a constant loading rate in the laboratory.…”

“…We test both models against an IID model with fixed mean recurrence interval, fixed mean earthquake size and zero variability in both measures. More thorough descriptions of all the models tested here can be found in the companion manuscript [ Rubinstein et al , 2012]. …”

“…As a complement to this study, we have also studied the utility of the time‐ and slip‐predictable models in predicting the behavior of natural repeating earthquakes in California, Taiwan, and Japan in a companion manuscript [ Rubinstein et al , 2012]. While repeating earthquakes are not as ideal for testing of the time‐ and slip‐predictable models as laboratory earthquakes, they offer many advantages over most natural earthquakes for this testing.…”

Spatiotemporal variations of characteristic repeating earthquake sequences along the Middle America Trench in Mexico

“…Figure shows a histogram of covariance coefficients for both sets. Group A histogram peaks at COV = 0.5, suggesting a set of stable sequences that repeat according to a deterministic time‐predictable model [ Rubinstein et al , ]. Sequences with short or highly variable interevent times suggest a high interaction between sequences and larger ruptures.…”