S23A-0282 1340h
GEMS: the Opportunity for Stress-Forecasting all Damaging Earthquakes Worldwide
Worldwide observations of shear-wave splitting indicate that rocks in the crust are pervaded by stress-aligned fluid-saturated grain-boundary cracks. These cracks are so closely spaced that they verge on fracture-criticality and failure in earthquakes. The evolution of such cracked rocks before fracture-criticality is reached can be modeled by anisotropic-poro-elasticity (APE). APE-modeling matches a huge range of phenomena pertaining to cracks, stress, and shear-wave splitting. This has allowed the accumulation of stress to be monitored, with hindsight, before about 12 earthquakes worldwide with magnitudes ranging from ML1.7 to Ms7.7, using swarms of small earthquake as shear-wave source. The time and magnitude of one M5 earthquake was successfully stress-forecast. Shear-waves carry no information about location, but precursory data predicted the fault-break. Suitably-persistent swarms of earthquakes are rare, and routine stress-forecasting requires controlled-source measurements in crosshole observations between adjacent boreholes in stress-monitoring-sites (SMSs). The first preliminary SMS in Iceland, using the highly-repeatable Downhole Orbital-Vibrator (DOV) source, recorded spectacularly sensitive anomalies in: crosswell Vp, Vs1, Vs2, and Vs2-Vs1 (anisotropy); there were also well-level changes; and changes in NS and EW GPS measurements, all correlating with low-level seismicity (energy equivalent to one M$\sim$3.5 earthquake) at 70km-distance. These confirm the science and technology of SMSs for monitoring stress accumulation before earthquakes. The sensitivity means that SMS networks on 400km-grids, say, would be able to recognize the accumulation of stress before all earthquakes M$>$5 within the network. Consequently, GEMS, a global network of SMSs on a 400km grid, would: 1) stress-forecast the times and magnitudes of all damaging earthquakes (M greater or equal to 5, say) worldwide; 2) provide data for stress-forecasting service analogous to weather-forecasting; 3) provide control for reducing the potential for large earthquakes threatening vulnerable locations by monitoring the effects of massive hydraulic operations nearby; 4) provide a powerful new tool for investigating the evolution of the interior of the Earth and provide an intellectual stimulus for understanding the Earth in the 21st century.
http://www.glg.ed.ac.uk/~scrampin/opinion/
S23A-0283 1340h
Earthquake Probabilities and Magnitude Distribution in 100a along the Haiyuan Fault, northwestern China
The Haiyuan fault is a major seismogenic fault in north-central China. One of the most devastating great earthquake in 20th century occurred near Haiyuan in northwestern China in December 16, 1920. More than 220000 people were killed and thousands of towns and villages were destroyed during this devastating earthquake. A 230 km long left-lateral surface rupture zone formed along the Haiyuan fault during the earthquake with maximum left lateral displacement of 10 m. In recent years, some researchers have studied the paleoseismology along the Haiyuan fault and revealed a lot of paleoearthquake events. All available information allows more reliable analysis of earthquake recurrence interval and earthquake rupture patterns along the Haiyuan fault. Based on fault geometry, segmentation pattern, and paleoearthquake events along the Haiyuan fault we can identify three scales of earthquake rupture: rupture of one segment, cascade rupture of two segments, and cascade rupture of entire fault (three segments), and obtain the earthquake recurrence intervals of these scales of earthquake rupture. The earthquake probability and magnitude distribution in 100-year along the Haiyuan fault can be obtained through weighted computation, by applying these paleoseismological information mentioned above, using Possion and Brownian passage time model and considering different rupture patterns. The result shows that the earthquakes probability of is about 0.035 in 100-year along the Haiyuan fault.
S23A-0284 1340h
An Inhomogeneous Distribution Model of Strong Earthquakes along Strike-slip Active Fault Segments on the Chinese Continent and Its Implication in Engineering Seismology
Through the statistical analysis of earthquake distribution along 51 strike-slip active fault segments on the Chinese continent, We found that strong earthquake distribution along the seismogenic fault segments is inhomogeneous and the distribution probability density can be stated as K=1.1206exp-3.947 in which K=S/(L/2), S refers to the distance from earthquake epicenter to the center of a fault segment , L is the length of the fault segment. The above model can be utilized to modify the probability density of earthquake occurrence of the maximum magnitude interval in a potential earthquake source. Nevertheless, it is only suitable for those potential earthquake sources delineated along a single seismogenic fault. This inhomogeneous model has certain effect on seismic risk analysis, especially for those potential earthquake sources with higher earthquake reoccurrence rates of the maximum magnitude interval. In general, higher reoccurrence rate of the maximum magnitude interval and lower exceeding probability level may bring larger difference of the results in seismic risk analysis by adopting the inhomogeneous model, the PGA values increase inner the potential earthquake source, but reduce near the vicinity and out of the potential earthquake source. Taking Tangyin potential earthquake source as an example, with exceeding probability of 10% and 2% in 50 years, the difference of the PGA values between inhomogeneous model and homogenous models can reach to 12%.
S23A-0285 1340h
Probability Map of the Next M$>=$5.5 Earthquakes in Italy
We provide a probability map of the next M$>=$5.5 earthquakes in Italy. For this purpose, we use a nonparametric multivariate hazard model to characterize the spatio-temporal distribution of earthquakes. The method is able to account for seismological/geological parameters that can potentially influence the spatio-temporal variability, and tests their relative importance. Moreover, it allows straightforward testing of a variety of hypotheses, such as Seismic Gap, Cluster, and Poisson. The method has been applied to the Italian seismicity (M$>=$5.5) of the last four centuries. We developed a regionalization of Italy in order to account for heterogeneities both in the tectonic domains and in the spatial distribution of earthquakes. Italy has been divided in 61 irregular zones that are homogeneous with regard to seismic behavior and to kinematics and orientation of the stress field resulting from active stress data (borehole breakouts data, earthquake fault plane solutions, seismogenic fault data). The results indicate a time clustering of the M$>=$5.5 earthquakes for a few years after an event, then the distribution becomes similar to a memoryless Poisson process, leading to a time dependent probability map. We have found that the most likely regions where the next moderate to large earthquakes may occur in the next 10 years are the Friuli in northeastern Italy, the Umbria-Marche in central Italy, and part of the Southern Apennines and the Calabrian Arc.
S23A-0286 1340h
A synthetic fault system in a spherical and viscoelastic Earth model
In order to improve our knowledge about fault interaction at long distances, we developed a synthetic fault system based on a spherical, viscoelastic, selfgravitating Earth model. Though seismicity in our simulation is controlled mainly by tectonic loading, our results highlight the significant role played by stress transfer in assessing the spatio-temporal pattern of the earthquakes also on global scale. In particular we observed that the postseismic stress diffusion could modify the seismicity rate of a local structure up to 70% and it affects structures thousands kilometers far away from the source.
S23A-0287 1340h
Comparison of Short-term and Long-term Earthquake Forecast Models for Southern California
Many earthquakes are triggered in part by preceding events. Aftershocks are the most obvious examples, but many large earthquakes are preceded by smaller ones. The large fluctuations of seismicity rate due to earthquake interactions thus provide a way to improve earthquake forecasting significantly. We have developed a model to estimate daily earthquake probabilities in Southern California, using the Epidemic Type Earthquake Sequence model [Kagan and Knopoff, 1987; Ogata, 1988]. The forecasted seismicity rate is the sum of a constant external loading and of the aftershocks of all past earthquakes. The background rate is estimated by smoothing past seismicity. Each earthquake triggers aftershocks with a rate that increases exponentially with its magnitude and which decreases with time following Omori's law. We use an isotropic kernel to model the spatial distribution of aftershocks for small ($M\leq5.5$) mainshocks, and a smoothing of the location of early aftershocks for larger mainshocks. The model also assumes that all earthquake magnitudes follow the Gutenberg-Richter law with a unifom $b$-value. We use a maximum likelihood method to estimate the model parameters and tests the short-term and long-term forecasts. A retrospective test using a daily update of the forecasts between 1985/1/1 and 2004/3/10 shows that the short-term model decreases the uncertainty of an earthquake occurrence by a factor of about 10.
http://moho.ess.ucla.edu/~helmstet/forecast.html
S23A-0288 1340h
Coulomb Stress Distribution Along the Fairweather and Queen Charlotte Transform Fault System
Tectonic loading and Coulomb stress transfer are modeled along the right-lateral Fairweather and Queen Charlotte transform fault system using a three-dimensional boundary element program. The loading model includes slip below 12 km along the transform as well as motion of the Pacific plate and is consistent with most available GPS displacement rate data. Coulomb stress transfer from adjacent fault segments is shown to be a weak contributing factor to the failure of the southeastern (Sitka) segment of the Fairweather fault (M 7.6, 1972), advancing the clock by only about 8 months. Failure resulted from a combination of loading from below (99 percent) by slip of nearly 5 cm/yr since before1900, and of stress transfer (1 percent) from major earthquakes on adjoining segments of the Queen Charlotte fault to the southeast (M 8.1 in 1949) and the Fairweather fault to the northwest (M 7.8 on Lituya segment in 1958). Combined Coulomb stress increases exceeded 4 MPa at a depth of 8 km prior to the Sitka earthquake. Coulomb stress transferred from the nearby M 9.2 Alaska earthquake of 1964 also may have advanced the clock for the 1972 event, but only by a month or two. Minimum recurrence times, based on average co-seismic displacements estimated from seismic moments and fault dimensions, range from about 80 years for the 1958 and 1972 events to 160 years for the 1949 earthquake. This implies stresses of 5 to 10 MPa at 8 km depth at failure, assuming total stress drops. Continued tectonic loading over the last half century and stress transfer from the M 7.6 Sitka event has resulted in re-stressing the adjacent segments by about 3 MPa at 8 km depth, as evidenced by the occurrence of a M 6.8 earthquake on the northwestern part of the Queen Charlotte fault on June 28, 2004, the largest since 1949. The segment of the Queen Charlotte fault immediately southeast of the 1949 rupture has accumulated about 6 MPa at 8 km through loading since 1900 and stress transfer in 1949. A continued rise in earthquake hazard is indicated for the Alaska panhandle and Queen Charlotte Islands region in the decades ahead as the potential for damaging earthquakes increases.
S23A-0289 1340h
The Pattern of Fractal Dimension Change before Large Earthquakes in Taiwan
We investigate the fractal dimension D, of the seven largest earthquakes (M$_{L}$ $>=$ 6.4) in the ten year period of 1994 $\sim$ 2003 located in the Taiwan area, including the 1999 M$_{L}$ = 7.3 Chi-Chi earthquake. Most showed significant increase in D before mainshocks. The D value of earthquake events without large foreshock events (M$_{L}$ $>=$ 5.2) had an obvious increase in the month preceding the mainshocks, and then fell abruptly due to clustered aftershock sequences. The pattern of D value change is slightly different for an earthquake event preceded by a large foreshock (M$_{L}$ $>=$ 5.2); here D increased earlier, about 3 or 4 months before the mainshock. The D value then declined slowly until the month with big foreshocks. Under the influence of foreshock sequences, the D value had a sharp drop before the mainshock event. This fractal analysis proved that a similar change of D value preceding large earthquakes is not due to the network processing procedure. The rise or fall of D value is not controlled by differences in efficiency of hypocenter relocation before and after large earthquake events. Even foreshocks with relatively smaller magnitudes produce an increased D value. This fact suggests that before large earthquakes, the spatial distribution of background seismicity becomes sparser then before, causing the increase of D value. Any period of steady rise in D value in the Taiwan area implies a higher potential for large earthquakes in the next few months.
S23A-0290 1340h
Aftershock Number for Forecasting Short-Term Earthquake Probabilities
Data from earthquakes worldwide with depths shallower than 70 km were combined from the International Seismological Centre, the US National Earthquake Information Center, Blacknest, and Harvard. An extensive magnitude and catalogue completeness study defined a `best' magnitude using the Harvard moment as a reference. The catalogue covers the period 1964 to 1995 and is effectively complete for earthquakes of magnitude 5.0 and above. The data were divided into six tectonic settings, and searched for related events using a simple window in space and time. An objective method was developed to define an elliptical aftershock area. The database of aftershock sequences has about 28,000 mainshocks of which about 2,400 have a magnitude $M \ge 6.0$, and these were followed by a total of about 7,000 aftershocks. The database was analyzed in space, time, magnitude, and in the number of aftershocks in a sequence, hereafter called abundance. The aftershock decay in time and the magnitude-frequency distribution follow well- established empirical laws, Omori's law and the Gutenberg and Richter relationship. These relationships were analyzed by stacking data from various sequences within the same tectonic setting. The $p$-value for the aftershock decay in time was found to be 1.0 for subduction and collision zones, and for regions of mixed tectonic character like New Zealand. For mid-ocean ridges the $p$-value of the present dataset is $1.19 \pm 0.08$ and for intracontinental zones $0.86 \pm 0.14$. The $b$-value of the magnitude-frequency relation is 1.0 for aftershock sequences in all settings. No variation of the $b$-value with time was observed. The abundance varies greatly from sequence to sequence. It can be modeled by a geometric distribution, where the mean abundance $N$ grows exponentially with mainshock magnitude, $M$ i.e. log $N$ is proportional to $M$. The distribution parameters for time, magnitude and abundance can be combined to probabilistically predict the number of aftershocks in a given time and magnitude interval for a given mainshock. This study was funded by the New Zealand Earthquake Commission.
S23A-0291 1340h
A Time\-Dependent Probabilistic Seismic Hazard Model For The Central Apennines (Italy)
Earthquake hazard in the Central Apennines, Italy has been investigated using time-independent probabilistic (simple Poissonian) and time-dependent probabilistic (renewal) models. We developed a hazard model that defines the sources for potential earthquakes and earthquake recurrence relations. Both characteristic and floating earthquake hypothesis\/model is used for the Central Apennines faults (M$>$5.9). The models for each fault segment are developed based on recent geological and geophysical studies, as well as historical earthquakes. Historical seismicity, active faulting framework and inferred seismogenic behavior (expressed in terms of slip rates, recurrence intervals, elapsed times) constitute the main quantitative information used in the model assignment. We calculate the background hazard from Mw 4.6-5.9 earthquakes using the historical catalogs of CPTI04 (Working Group, 2004) and obtain a-value distribution over the study area. This is because the earthquakes occur in areas where they cannot be assigned to a particular fault. Therefore, their recurrence is considered by the historic occurrence of earthquakes, calculating the magnitude-frequency distributions. We found good agreement between expected earthquake rates from historical earthquake catalog and earthquake source model. The probabilities are obtained from time-dependent models characterized by a Brownian Passage Time function on recurrence interval with aperiodicity of 0.5. Earthquake hazard is quantified in terms of peak ground acceleration and spectral accelerations for natural periods of 0.2 and 1.0 seconds. The ground motions are determined for rock conditions. We have used the attenuation relationships obtained for the Apennines by Malagnini et al. (2000) together with the relationships predicted from Sabetta and Pugliese (1996) and Ambraseys et al. (1996) for the Italian and European regions, respectively. Generally, time dependent hazard is increased and the peaks appear to shift to the ESE of the central Apennines with respect to the results of the Possionian source model. In order to present the most likely earthquake magnitude or the most likely source-site distance and determine predominant sources of seismic hazard we deaggreagted the seismic hazard for 1.0 Hz PSA and PGA in some cities across Central Apennines.
S23A-0292 1340h
Correlation Length as an Indicator of Critical Point Behavior Prior to a Large Earthquake
A large earthquake preparation is often manifested in correlation of seismicity in an area whose characteristic dimension greatly exceeds a dimension of source of main shock. Zoller et al. (2001) show the growth of correlation length of earthquakes prior to nine large earthquakes in California according to a power low. We argue that the algorithm of correlation length estimation proposed by Zöller et al.(2001) can result in a decrease of correlation length preceding its precursory growth before large earthquakes if the area in which earthquake activity is correlated grows with time during a main shock preparation. The correlation length analysis of acoustic emission events recorded in laboratory experiments on destruction of rocks and correlation length analysis of intermediate magnitude earthquakes in the area of large earthquakes preparation on Kamchatka and in Italy confirms the theoretical argument. This effect can be considered as an additional premonitory pattern of large earthquake preparation.
S23A-0293 1340h
Breaks in Fractal Scaling of Real and Synthetic Earthquake Catalogues
Earthquake generation within the crust is the result of a series of complicated spacio-temporal interactions between different tectonic blocks and units. The end product of the process is a function of both long term deterministic-chaotic processes in a regional scale and short-term Self-Organized Critical (SOC) processes of a local nature [e.g. McCloskey and Bean,1994; ]. In the past three to four decades many models of seismicity have been developed [e.g. Burridge and Knopoff, 1967; Huang and Turcotte, 1990; McCloskey, 1993; Ben-Zion, 1996] trying to model the observed patterns of earthquake generation and seismicity. Some of these studies have shown that it is possible to reproduce the main features of the real earthquake populations. In this study the fractal dimension of SCSN, JMA and ISC seismicity catalogues have been studied. the aim was to see whether all the different sizes of earthquakes within a catalogue (i.e. a single spacio-temporal window) belong to the same population and whether any breaks in fractal scaling exists within the catalogue concerned. Furthermore, a selection of synthetic earthquake models were analyzed with the same approach to determine whether they are able to reproduce the same results as empirical ones. Subsequent analysis of the data have revealed several distinct breaks in fractal scaling of earthquakes of different magnitudes. In other words, it emerged that small and large earthquakes in each catalogue are obeying different fractal dimensions hence belonging to different earthquake populations. It is possible to associate one of the breaks, observed in the SCSN catalogue to the average thickness of the seismogenic crust of California ( $\sim$ 15 km as calculated by Nazareth and Hauksson, 2004). With the same technique used for the empirical catalogues, three different synthetic catalogues [McCloskey, 1993; Ben-Zion, 1996; Khademi and McCloskey, this study ] were analyzed. Results have shown that all the models are able to predict the fractal dimension of the empirical catalogues to some degree and further, two latter models are able to simulate the breaks in the fractal scaling. The fractal dimensions from models F, U, M and A of Ben-Zion [1996] are generally in good agreement with observed dimensions of empirical catalogues, though the observed breaks in the empirical catalogues cannot be seen in these synthetic models. However, a gradual decrease in the fractal dimension with increasing treshold magnitude can be observed. The McCloskey [1993] Chaotic-SOC hierarchical model, with reasonable accuracy, predicts both the fractal dimensions and dimension breaks, observed in the empirical catalogues. The model's success in the prediction of behaviour of empirical data is particularly due to the combination of low-dimensional chaotic behaviour of bigger blocks (i.e. larger events) and high-dimensional SOC behaviour of smaller blocks (i.e. small events resulting from activity of smaller portions of the main fault or adjacent minor discontinuities). And finally the Khademi-McCloskey (KMC) model is able to reproduce both the dimension and one of the breaks of scaling. But, the model is unable to produce more than one break in scaling (i.e. to distinguish more than two earthquake populations within the same dataset). It is concluded that hierarchical earthquake models (though with some modifications) can be used to extend the temporally limited empirical catalogues to much longer time spans and to overcome the temporal limitations of the existing empirical catalogues.
S23A-0294 1340h
Time-dependent evolution of stress over the past 200 years in Southern California
Seismic hazard estimates can be refined by examining the current state of stress change in regions prone to earthquakes. Active faults which have not recently failed and lie in regions experiencing high rates of stress increase represent the best candidates for near future earthquakes. Evolution of stress in the crust is influenced by three main factors: interseismic loading, static stress changes from large earthquakes; and postseismic relaxation of a viscous lower crust or upper mantle. Southern California has had more than eighteen large $(M_{w} \geq 6.5)$ earthquakes in the past two centuries. Previous research has calculated the evolution of stress in southern California over this period of time due to coseismic slip associated with these events and interseismic loading. Here we extend this analysis to include the effects of viscous relaxation that follows each quake and can significantly alter the stress field. Our approach is to numerically model the evolution of Coulomb stress for southern California due to all the major $(M_{w} \geq 6.5)$ earthquakes over the past 200 years using the superposition of a viscoelastic model and the regional stress rate. The regional stress rate is due to the relative motion of the Pacific and North American plates as constrained by the SCEC velocity field map. We plan to present results that discuss where Coulomb stress is currently building the fastest and how these stress changes resolve themselves on major strike-slip faults in the region and thrust faults within the LA Basin. We are particularly interested in how the stress shadows (region of Coulomb stress decrease) cast by the great Fort Tejon earthquake of 1857 and other large historic events have been influenced by the competing role of viscoelastic relaxation, which would tend to prolong the longevity of shadows associated with strike-slip earthquakes, and the build up of regional stresses which work to erase them.
S23A-0295 1340h
OpenAMR - A JAVA-Based Computational Facility for Earthquake Rupture Forecasts Using Accelerating Moment Release
In conjunction with the RELM OpenSHA project, we are developing a regional earthquake probability estimator based on the Stress Accumulation (SA) model for accelerating moment release (AMR). The SA model suggests that precursory AMR is controlled by the evolution of the stress field preceding large events. Bowman and King [GRL, 2001] demonstrate that the pre-mainshock stress field, as indicated by a simple backslip model of the event, can be used to define the critical region that optimizes the precursory AMR signal. Recent work has shown that the Stress Accumulation AMR model can also be applied prospectively for forecasting future large events by calculating the pre-event stress field for a hypothetical earthquake scenario with a size and slip distribution consistent with geologic and geodetic evidence of the fault behavior. This stress field is then used to search for an AMR signal that would indicate the temporal proximity of the scenario event. We outline here the methodology being developed for a regional earthquake probability estimation package using the SA model for AMR. The approach combines a fault-based algorithm that utilizes geologically and historically supported rupture scenarios drawn from the SCEC Earthquake Rupture Forecast database, with a grid-based algorithm that searches for AMR preceding physically realistic scenario events occurring off of the fault network. The output of the analysis is a regional map of time-dependent earthquake probabilities. The package utilizes the object-oriented capabilities of JAVA programming both to interface with other components of OpenSHA, and to facilitate the direct comparison of the Stress Accumulation model forecasts with forecasts based on alternative search criteria. An alternative we are currently exploring for the grid-based algorithm uses circles rather than stress contours to search for precursory AMR. This will provide an assessment of the merit of the SA model for earthquake forecasting, independent of the merit of AMR.
S23A-0296 1340h
Comparison of Seismicity Models for Hazard Assessment in Low Seismicity Regions
Statistical models such as ETAS (Epistemic Type Aftershock Sequence) propose to describe the occurrence of earthquakes based on two well-established laws, the Gutenberg-Richter and the Omori laws. Such a model is thus able to reproduce clustering of earthquakes, as observed in reality. If integrated in probabilistic seismic hazard assessment, this model can be an alternative to the non-realistic Poisson model usually assumed. This study first aims at evaluating the ability of ETAS model to describe the behaviour of seismicity in low seismicity regions. The parameters required by the model are estimated on French and German instrumental data. To estimate the uncertainty on these parameters, due to the use of short time periods, synthetic catalogues are simulated with Monte Carlo method. In the second step, we compare the hazard estimates computed assuming that seismicity is fully described by the ETAS model, with classical probabilistic results of PSHA studies relying on the Poisson hypothesis.
S23A-0297 1340h
A Comprehensive Search for Tidal Triggering of Southern California Earthquakes
Tidal forces cause crustal stress changes that might be expected to trigger earthquakes but previous analyses searching for tidal periodicities in earthquake catalogs have yielded generally negative or inconclusive results. This study examines the Caltech/USGS southern California earthquake catalog from 1932 to 2003, which includes 429,886 events. We examine the data for any correlation between event occurrence and lunar phase by plotting histograms of event count versus phase as well as simple plots of event phase versus time. We analyze both the complete catalog and specified subsets of the data, including grouping by magnitude and depth. We also refine our data spatially by binning events in individual 0.5 by 0.5 degree latitude/longitude bins. Finally, we examine over 900 individual similar event clusters containing 50 or more events as defined by cross-correlation analysis. These clusters typically are less than 3 km across and include earthquakes that likely to have similar focal mechanisms and sensitivity to tidal phase. However, none of our results find a significant correlation between tidal forces and seismicity, suggesting that tidal forces do not contribute significantly to earthquake occurrence times anywhere in southern California.
S23A-0298 1340h
A Time-dependent Seismic Hazard Model for Mexico
The fast subducting of Cocos-Rivera plate beneath the western margin of the North American plate leads to many large and great earthquakes. The recurrence times of large earthquakes in this region is relatively short, and thus plenty historical earthquake data are available for time-dependent seismic hazard analysis. However, large uncertainties on magnitudes, locations and rupture processes of historical data make it difficult to estimate recurrence times of characteristic earthquakes which are important for regional earthquake potential modeling. We formulate an earthquake potential model for Mexico based on stochastic simulations of the interactions between subduction zone segments and time-dependent rupture scenarios to capture the uncertainties in the historical data and the variations in return periods of characteristic earthquakes. The subduction zone in Mexico is divided into 14 segments based on the work by Nishenko [1991] and Ordaz and Reyes [1999]. We translate all subduction zone related historic earthquake data in conjunction with the plate convergence rates into coupling coefficients for segments. The coupling coefficients that are obtained based on different pieces of information are evaluated and applied in seismic rate calculation. We create a stochastic model for the subduction segments interactions to simulate both single and multi-segment rupture scenarios. Time dependent rates of large interface earthquakes are calculated based on the renewal model with lognormal distribution. We considered uncertainties in the mean recurrence and distribution sigma for the analysis. The uncertainties from the cascading model are incorporated into the time dependent model. Based on these concepts, a regional time-dependent seismicity model for subduction zone is created. For other regions in Mexico, we apply smoothed seismicity method and characteristic model for modeling background seismicity and fault-based earthquakes, respectively. We construct a seismic hazard model for Mexico and evaluate the result by comparing it with a hazard model purely based on Poissonian earthquake rate.
S23A-0299 1340h
Crustal stress evolution of last 700 years in North China and earthquake occurrences
We simulate the evolution process of cumulative Coulomb failure stress change ($\Delta$CFS) in North China since 1303, manifested by secular tectonic stress loading and occurrence of large earthquakes. Secular tectonic stress loading is averaged from crustal strain rates derived from GPS. Fault rupture parameters of historical earthquakes are estimated as follows: the earthquake rupture length and the amount of slip are derived based on their statistical relationships with the earthquake intensity distribution and magnitude, calibrated using parameters of instrumental measured contemporary earthquakes. The earthquake rake angle is derived based on geologically determined fault setting parameters and seismically estimated orientation of regional tectonic stresses. Assuming a layered visco-elastic medium, we calculate stress evolution resulted from secular tectonic loading and coseismic and postseismic deformation. 49 M 6.5 earthquakes occurred in North China since 1303. Statistics shows that 39 out of the 48 subsequent events were triggered by positive $\Delta$CFS, yielding a triggering rate of 81.3%. The triggering rate for M 5 earthquakes after the 1976 Tangshan earthquake is 82.1%. The triggering rate is up to 90% if corrections are made for some aftershocks which were wrongly identified as occurred in stress shadow zones because of errors in parameter estimates of historical earthquakes. Our study shows very high correlation between positive $\Delta$CFS and earthquake occurrences. Relatively high $\Delta$CFS in North China at present time is concentrated around the Bohai Sea, the west segment of the Northern Qinling fault, western end of the Zhangjiakou-Bohai seismic zone, and the Taiyuan basin in Shanxi rift zone, suggesting relatively higher earthquake potential in these areas.
S23A-0300 1340h
Analyzing earthquake clustering features by stochastic reconstruction -- foreshocks
The ETAS model has been proved to be able to be used the first approximation for describe the earthquake clustering features (Zhuang et al 2004, JGR, 109, No. B5, B05301, doi:10.1029/2003JB002879). This model classifies the seismicity into two components, the background and the cluster, where each earthquake event, no matter if it is from the background component (usually assumed to be a space-time Poisson process, stationary or non-stationary, homogeneous or non-homogeneous) or generated by another event, produces (triggers) its own offspring (aftershocks) according to some branching rules. Its conditional intensity function at the space-time-magnitude location $(t, x, y, M)$, has the form of $$\lambda(t,x,y,M|\hit)=J(M)\left[\mu(x,y)+\sum_{i:t_i<t} \xi(t-t_i, x-x_i, y-y_i;M_i)\right],$$ where ${\cal H}_t$ is the observation history of the past events up to time $t$ but not including $t$. We define the probability of the $j$th event being directly produced by a previous $i$th event as the relative contribution of the event $i$ to the total conditional intensity at the occurrence time and location of $j$, i.e., $$\rho_{ij}=\left\{ \begin{array}{cc} {\frac {J(M_j)\xi(t_j-t_i,x_j-x_i,y_j-y_i;M_i)}{\lambda(t_j, x_j, y_j|{\cal H}_{t_j}) }}, &~~~~~\mbox{when }j>i, 0, &~~~\mbox{otherwise}, \end{array} \right ., $$ and its probability of being produced from background to be the relative contribution of the background rate, $$\varphi_j= \frac {\mu(x_j,y_j)J(M_j)}{\lambda(t_j,x_j,y_j,M_j| {\cal H}_{t_j})}.$$ These probabilities can be used to realize the family trees of earthquakes stochastically. Thus, a foreshock is naturally defined as a background event that has at least one offspring of larger magnitude, direct or indirect, and a mainshock is defined as the biggest event in the family tree. With this new definition we studied the characteristics of foreshock features and their relationship to mainshocks.
S23A-0301 1340h
The False-Alarm Rate of Accelerating Moment Release, and its Relationship to Probabilistic Earthquake Rupture Forecasts.
It has been suggested that large earthquakes are preceded by a systematic increase in the rate of background seismicity in a broad region around the impending event. This rate change, known as "accelerating moment release" (AMR), has been proposed as a precursory signal that could be used to forecast large earthquakes. In this approach, the observation of accelerating seismicity would represent a period of increased likelihood of a large earthquake, termed an "alarm". However, as with any pattern-recognition scheme there is a finite probability that the observed rate changes could be due to random variations in background seismicity rate. An observation of accelerating moment release that does not culminate in a large earthquake is called a "false-alarm". To test the probability of accelerating moment release arising from random fluctuations in the seismicity rate, synthetic earthquake catalogs are generated and searched for AMR. The frequency of accelerating moment release in these randomly generated catalogs represents the false-alarm rate of the AMR forecasting methodology. The importance of aftershocks and the spatial distribution of seismicity on discrete faults is also explored by introducing random clustering in the synthetic catalogs. The false-alarm rate is shown to depend strongly on the degree of acceleration in the background seismicity. Assuming that the false-alarm rate is the probability that the observed acceleration is occurring randomly, then the complement of the probability represents the likelihood that the observation is a precursor to an earthquake. This allows time-dependent probabilistic rupture forecasts to be made based on observations of accelerating moment release.
S23A-0302 1340h
Tossing the Earth: How to Reliably Test Earthquake Prediction Methods
One of the most consequential issues of the earthquake prediction problem is reliable testing of hypothetical prediction methods. The danger of self-deception by data overfitting here is especially high due to both the scarceness of large earthquakes and the absence of a conventional wide-reaching theoretical framework. This talk gives an overview of the methods currently employed to test prediction algorithms and bridges the commonly accepted approaches to the problem. The main focus is on the two most widely used approaches to assessing prediction methods. Both of them evaluate the amount of new information revealed by the prediction method about the impending earthquake activity. The first one starts by estimating the expected spatio-temporal distribution of seismicity, and uses the classical likelihood paradigm to evaluate the prediction power. Accordingly, it uses the nomenclature of statistical estimation. The second one applies results of G. Molchan [Pure Appl. Geophys., 149: 233-247, 1997] that can be considered as a time-dependent analog of the Neyman-Pearson lemma to make a decision whether or not to expect an earthquake within a given spatio-temporal region. Accordingly, it uses the nomenclature of hypothesis testing. Importantly, this approach does not require the explicit knowledge of the earthquake hazard rate; in other words, the correct decision can be made with the realistically imprecise data. We discuss how the choice of the assessment method depends on a specific prediction situation using the outcomes of real-time prediction experiments. The best choice happened to depend crucially on the specifics of the prediction problem: set of target earthquakes; prediction time-span, resolution, etc.