S43D-01
Stress interaction and the 2004/2005 Sumatran earthquakes
Following the great 26 December Sumatra - Andaman Islands earthquake, we computed the Coulomb stress changes on other faults in the region and found that the stress had increased by up to 8 bars on both the Sunda Trench, immediately south of the 2004 rupture plane, and the Sumatra fault which runs down the center of the Island. In a paper published on 17 March, 2005, we suggested that the seismic risk on both structures had increased as a result of this stress change. On 28 March, 2005, the M=8.7 Simeulue-Nias earthquake ruptured a portion of the Sunda Trench immediately to the south of the 26 December rupture plane. Although slip in this earthquake was contiguous with the December event, the hypocentre was approximately 100 km to the south. Coulomb stresses at the earthquake hypocentre were very low, ranging between 0.07 - 0.17 bars depending upon choice of slip model. The Simeulue-Nias event, in turn, increased the stress farther south along the Sumstra Faults and on the Sunda megathrust south along the subduction zone. Paleogeodetic studies in the Mentawai island chain have shown that beneath the island of Siberut the megathrust has not ruptured since 1797 and has therefore accumulated about 10m of slip. Farther south, the megathrust has not ruptured since 1833 and an event triggered beneath Siberut could potentially propagate southward into this region resulting in an earthquake that would, again, have a large magnitude, possibly M8.0-9.0. A good model for this earthquake would be the M8.5 1833 event which generated a large and destructive tsunami.
S43D-02
Spatial and Temporal Length Scales Characterizing the Evolution of Seismicity Rates.
Numerous studies have documented systematic changes in seismicity rates preceding large magnitude events. Many works suggest that these changes can be used to conduct time-dependent earthquake forecasting. We use two approaches to examine the spatial and temporal scales characterizing the seismicity rate changes, with the goal of exploring the underlying physical process. The first set of analyses follow the methodology outlined in Tiampo et al. [2002], for determining the eigenfunctions describing spatial and temporal correlation in regional seismicity. We extend the method by incorporating a temporal lag in construction of the covariance matrix. Decomposing the matrix into its eigenmodes then highlights correlated activity separated in time by the specified lag. Here, we present the results obtained for southern California seismicity from 1932 to 2004, using a range of temporal lags. Our second approach considers changes in yearly seismicity rates as a function of distance from the rupture plane of major historical events. To quantify the significance of trends in the seismicity rates, we auto-correlate the data, using a range of spatial and temporal lags. Here, we focus on the results for the 1987 Superstition Hills, 1992 Landers, and 1994 Northridge, California, earthquakes. We also briefly address the results for the 1971 San Fernando, 1983 Coalinga, 1986 Chalfant Valley, 1989 Loma Prieta, 1999 Hector Mine events and the 2002 Denali, AK, earthquake.
S43D-03
Exaggerated Claims About Success Rate of Earthquake Predictions: "Amazing Success" or "Remarkably Unremarkable"?
On October 1, 2004, NASA announced on its web site, "Earthquake Forecast Program Has Amazing Success Rate." This announcement claimed that the Rundle-Tiampo earthquake forecast method has accurately predicted the locations of 15 of California's 16 largest earthquakes this decade. Since words like "amazing" carry a lot of meaning to consumers of scientific information, claims of "amazing success" should be limited only to cases where the success is truly amazing. We evaluated the statistical likelihood of the reported success rate of the Rundle-Tiampo prediction method by applying a cellular seismology approach to investigate whether proximity to past earthquakes is a sufficient hypothesis to yield the same level of success as the Rundle-Tiampo method. To delineate where to expect future earthquakes, we used the epicenters of the ANSS earthquake catalog for California from 1932 through 1999 with magnitude ≥ 4.0 ("before" earthquakes). We then tested how many of the 15 events that are shown on the NASA web page ("after" earthquakes) occurred near the "before" earthquake epicenters. We found that with only a 4 km radius around each "before" earthquake epicenter, we successfully forecast the locations of 13/15 (87%) of the "after" earthquakes, and with a 7 km radius we successfully forecast 14/15 (93%) of the earthquakes. The zones created by filling in a 7 km radius around the "before" epicenters cover 18% of the study area. The scorecard maps on the JPL "QuakeSim" web site show an 11 km margin of error for the epicenters of the forecast earthquakes. With an 11 km radius around the past epicenters (covering 31% of the map area), we catch 14/15 of the "after" earthquakes. We conclude that the success rate referred to in the NASA announcement is perhaps better characterized as "remarkably unremarkable", rather than "amazing." The 14/15 success rate for the earthquakes listed on the NASA scorecard is not a rigorous test of the Rundle-Tiampo method, since it appears that any method that uses past seismicity maps as a basis for earthquake forecasting should have a comparable rate of success.
S43D-04
Earthquake forecasting using the pattern informatics (PI) index
Recent large earthquakes include the M ~ 7.4 event that struck Izmit, Turkey in August of 1999, the M ~ 7.6 Taiwan earthquake which occurred in September of 1999, the M ~ 7.1 Hector Mine, California earthquake of October 1999, and the M ~ 9 Indonesian earthquake of December 2005. Many similar examples have been documented over the course of time, yet, until recently, no reliable precursors have been detected with any repeatability. The most successful recent geophysical research associated with earthquakes forecasting has centered on investigating the spatial and temporal patterns in seismicity data. In the past we have employed a pattern informatics analysis technique, formulated based on the physical and theoretical understanding of complex, nonlinear fault systems, to isolate emergent regions of coherent, correlated seismicity prior to their occurrence in southern California (Tiampo et al., 2002). This new technique, the PI index, identifies the characteristic patterns associated with the shifting of small earthquakes from one location to another through time prior to the occurrence of large earthquakes. These identify regions of increased probability of a future large earthquake, on an intermediate length time scale. Examples of the application of this technique to other regions, such as Turkey, Spain, Japan, and the Caribbean, will be discussed, and the current extent of its ability to forecast the magnitude of the upcoming event.
S43D-05
Triggering of Great Earthquakes: Seasonal Variation.
It is now well established that even great earthquakes are triggered by relatively small stress changes provided that the stress change increases the Coulomb failure. Stress changes of a few tenths to a few bars have both advanced and inhibited earthquakes. Four centuries of earthquake occurrence data in Northeast Japan allowed a ~36 year lag time of subduction events after on-land earthquakes to be determined (Rydelek and Sacks, 1990). The elastic-viscoelastic model of the crust-lithosphere derived from geodetic observations over about 50 years can explain the time lag. The stress diffusion from the 1940's Nankai trough earthquakes, M ~8, slowly unloaded the normal stress clamping the Nojima fault over a 50 year period, resulting in the 1995 Kobe earthquake, m=6.9 (Pollitz and Sacks, 1997). From the earthquake record spanning about 12 centuries, the 1940's Nankai trough earthquakes were themselves advanced in time, the interval since the previous event of 1854 being clearly the shortest on record. Strain diffusion from the on-land great Nobi earthquake of 1891 explains not only the advance, but also the two year delay between the eastern (Tonankai) and western (Nankaido) events. The failure mechanism was modeled by Rydelek and Sacks, 2003. In the above examples, the strain diffusion has created stress changes of a fraction to a few bars at the fault so as to increase the Coulomb failure and modify the occurrence time by a significant amount, i.e. many years on a fault with recurrence interval of more than a century. Here we describe a situation in which strain changes, even though less than a tenth of a bar have governed great earthquake occurrence for more than a thousand years. Continuous GPS observations enable insight into much lower stress triggering of great events. It has long been recognized that most Nankai trough events (since 684), with recurrence intervals of one to two centuries, occur in the winter. A seasonal shortening of the continental plate (Heki, 2004) overlying the subducting Philippine sea plate, causes a reduction of stress on the thrust fault. Even though this stress change is less than 0.1 bar, and the yearly stress loading of the fault may be 0.5 to 1 bar, it seems to be sufficient to influence the failure time.
S43D-06
Shallow Seismicity in Stable Continental Regions (SCRs): Implications for Earthquake Hazards
A world-wide compilation of strong ($M_w$ 4.5-8.0) and well constrained earthquakes in stable continental regions (SCRs) reveals a bimodal depth distribution with a very shallow upper crustal component. SCR-earthquake ruptures are confined within the upper or the lower thirds of the crust (0-10 km or 20-35 km). Thus, while the mid crust accounts for much of the moment released in active continental regions (ACRs; excluding regions where tectonics has changed crustal thickness), the same depth range in SCRs shows a minimum of seismicity. This remarkable difference has been partly hidden by a tendency to overestimate hypocentral depths in SCRs, where instrumental coverage tends to be sparse. The upper 5 km of ACR-crust are generally weak and seismically opaque, thus releasing relatively little seismic moment and attenuating seismic energy coming from below. In contrast, the upper SCR-crust is generally strong and many large SCR-earthquakes nucleate close to the surface, with severe implications for hazard. On the other hand, the tendency of SCR-earthquakes to occur in sequences and to rupture downward offers an opportunity for improving hazard estimates; after an earthquake occurs the probability of another earthquake increases dramatically in a source area with no prior seismicity. Such a probability can be quantified from available data. The aforementioned issues become more important, when human activities are more likely to cause significant stress changes in the depth range for nucleation of SCR-earthquakes than for ACR-earthquakes. A world-wide compilation of human triggered earthquakes due to \textit{geomechanical pollution} confirms the argument of shallow earthquake nucleation in SCRs.
S43D-07
Why Weibull?
The statistical distribution of recurrence times of characteristic earthquakes plays an important role in hazard assessment. Assumed distributions include the exponential (random), Weibull (stretched exponential), log-Normal, and Brownian passage time (inverse Gaussian). In this paper we argue that the Weibull distribution provides the proper scaling. This distribution has found wide applicability in statistical physics. In this paper we present the results of numerical simulations using a hybrid model that combines the forest-fire model with the site-percolation model in order to better understand the earthquake cycle. We consider a square array of sites. At each time step, a "tree" is dropped on a randomly chosen site and is planted if the site is unoccupied. When a cluster of "trees" spans the site (a percolating cluster), all the trees in the cluster are removed ("burned") in a "fire". The removal of the cluster is analogous to a characteristic earthquake and planting "trees" is analogous to increasing the regional stress. We find that the statistical distribution of recurrence times (number of time steps between model earthquakes) is in much better agreement with the Weibull distribution than either the log-Normal or Brownian passage time distributions. The coefficient of variation (aperiodicity) of the distribution is 0.394. We also show that the synthetic distribution of recurrence times obtained using the "Virtual California" model is in excellent agreement with the Weibull distribution. For the Parkfield section the simulated earthquakes have a coefficient of variation with a value 0.354. The actual earthquakes on the Parfield section are in good agreement with the Weibull distribution with a coefficient of variation with a value 0.378.
S43D-08
Examination of the Predictive Waiting Time Method Using Global Earthquake Catalogs
The Gutenberg-Richter magnitude frequency relationship offers a single process triggering model to build statistical models of seismicity. And by using these statistical models, one can potentially predict the likelihood of aftershocks within a predetermined level of uncertainty. While this simple relationship is shown to be important as a statistical measure of the number of aftershocks, it is not effective in describing the occurrence time (to within some margin of error) of a given aftershock, particularly large aftershocks that could be important for seismic hazard mitigation. In this paper we examine the effectiveness of an alternative large-aftershock predictive method, the Waiting Time Method (WTM, first proposed by Li and Gu, 1979), using the NEIC catalog. The relationship can be written as log(dT)=Alog(T)+B, where A and B are constants, T is the time of the large aftershocks and dT is the elapsed time of the aftershock in discussion from the previous one. This relationship explicitly explores a simple linear relationship (in log-log domain) between the occurrence of an event and its association with the previous event; A and B can be determined numerically in real time. We examined all of the magnitude 7 -8+ (M) events in the NEIC catalog. We select different aftershock zone radii and aftershock cutoff periods after the main event based on the aftershock magnitude. Our results show WTM explains the occurrences of large aftershocks for nearly 90% of the events above 7.6. The averaged slopes are 0.7-0.9 with uncertainties less than 0.1. WTM explains ~70% of the large aftershocks of mainshocks with magnitudes less than 7.6. Considering the lower aftershock magnitude threshold (M5 or less) for M7 mainshocks, the slightly worse (but still statistically significant) performance may result from magnitude uncertainties. Our results strongly suggest that WTM is a robust empirical relationship for describing large aftershocks and may have important applications in earthquake prediction and classification. Among the parameters that could be improved for further considerations are the lower threshold values for the definition of aftershocks. The lowest magnitude values depend on the type of the fault and the mainshock location. We find that empirical values of 2-3 yield optimal results. Efforts are being made to develop an automated classification algorithm based on WTM and to quantify the regional variation of A and B values.