|
|
Probabilistic Seismic Hazard Assessment for
the North Anatolian Fault Zone Based on Areal Sources
by
Ali Osman Oncel
Istanbul University, Department of Geophysical Engineering, 34850 Avcilar-Istanbul/Turkey
U.S. Bureau of Reclamation, Denver, Colorado, USA; email roland@seismo.usbr.gov
Geological Survey of Japan, Tsukuba 305, Japan; oncel@gsj.go.jp
Abstract: The North Anatolian Fault Zone (NAFZ) is one of the most active fault zones in the world. It is a dextral strike-slip fault more than 1000 km long which forms the northern boundary of the westward moving Anatolian plate. NAFZ is splayed into three branches at about 31 E and continues to the Gulf of Corinth in the northern Aegean Sea. This study consists of a description of the development of a historic seismicity catalog for the NAFZ, and the subsequent use of this catalog in a probabilistic seismic hazard analysis (PSHA).
The analysis is based on the subdivision of the NAFZ into western, central, and eastern zones, with areas of about 200,000, 250,000, and 100,000 square kms, respectively. Instead of modeling activity on the fault directly, seismicity is modeled as occurring randomly within three rectangular zones centered on the fault zone. The results are area-based, in other words, it is assumed that the hazard within a given zone (or within a few tens of kilometers from the zone boundaries) is the same. A Cornell (1968)-type analysis was conducted, with complete enumeration of uncertainties in seismicity rate and ground motion value as a function of magnitude and distance. Results consist of hazard curves for peak horizontal acceleration and equal hazard acceleration response spectra.
The PSHA results in this study consist of hazard curves for peak horizontal acceleration, and uniform hazard acceleration response spectra for return periods of 100, 500, 1000, and 5,000 years. Earthquake recurrence calculations show that the central section fits the log-linear Gutenberg-Richter relation quite well. The western section exhibits anomalously larger numbers of earthquakes in the magnitude 6.5-7.0 range, implying "characteristic" behavior. Recurrence for the eastern section is poorly constrained due to the small number of events.
Probabilistic Seismic Hazard Analysis (PSHA) is currently the most accepted method of portraying ground motion hazards from earthquakes. The method, originally described by Cornell (1968) that was applied to determine the preliminary seismic risk in Turkey (Bath, 1979), integrates all relevant seismic sources and their activity rates, along with their estimated uncertainties, and through an appropriate ground motion attenuation function, produces annual frequency of exceedance values for a continuum of ground motion levels. While Peak Horizontal Acceleration (PHA) has been the most commonly applied ground motion measure in the engineering community, the response spectrum, defined as the maximum response of a damped, single-degree-of-freedom oscillator to an input ground motion at various frequencies, is also used. In this study results consist of hazard curves for PHA (annual frequency of exceedance vs. PHA), and equal-hazard response spectra, defined as response spectra where each spectral ordinate has the same annual probability of being exceeded.
Earthquakes can be considered to emanate from two sources: identifiable active faults and hidden, randomly distributed sources. Activity rates on faults can be estimated from historic seismicity (if earthquakes can unequivocally be associated with the fault), slip rates, moment rates, and/or regional strain rates (e.g., Youngs and Coppersmith, 1985; WGCEP, 1995). Rates of randomly occurring regional seismicity, however, must be estimated from the historic record. Random earthquake magnitudes are generally less than 7, while earthquakes of engineering significance that occur on faults have magnitudes of ~5 and above.
For this study the historic seismicity record in the vicinity of the North Anatolian Fault Zone (NAFZ) is used as the sole estimate of activity rate as input to the PSHA. Although simplistic, this approach has some advantages. A fault-based modelling approach requires information regarding the location, dip, downdip extent, and sense of motion of each fault, in addition to estimates of maximum magnitude, activity and/or slip rates, and distribution of magnitudes. For many areas of the NAFZ this information is not available. Also, the problematic task of associating individual historic events, most of which have large location uncertainties, to individual faults is circumvented.
The NAFZ is the northern boundary of the Anatolian block, which is moving to the west as a result of the continuing northward movement of the Arabian plate towards Eurasia
(Figure 1) . The NAFZ is well defined morphologically from about 31°E to 41°E (Allen,1969). Dextral strike-slip faulting along the NAFZ appears to continue eastward, beyond the triple junction (41°E) with the East Anatolian Fault Zone (EAFZ), but is not as continuous as it is along the NAFZ (Jackson, 1992). There it breaks into three strands from about 31° up to the northern Aegean Sea region (Barka, 1992). Estimates of earthquake hazards from the NAFZ have been made by separating the fault zone into three subregions based on its geometric character and temporal and spatial patterns of seismicity (Öncel et al., 1995; Öncel and Alptekin, 1996a,b).
This study consists of a description of the development of a historic seismicity catalog for the NAFZ, and the subsequent use of this catalog in a PSHA. The analysis is based on the subdivision of the NAFZ into western, central, and eastern zones. The results are area-based, in other words, it is assumed that the hazard within a given zone (or within a few tens of kilometers from the zone boundaries) is the same. The PSHA results in this paper consist of hazard curves PHA, and uniform hazard acceleration response spectra for return periods of 100, 500, 1000, and 5,000 years.
The instrumental part of the catalogue was compiled from the International Seismological Center (ISC) data file for the time period between 1916 and 1992 and Ambraseys and Finkels (1987) catalogue for the time period between 1899 and 1915. The historical part of the catalogue for the period from 1800 to 1899 was compiled from Soysal et al., (1980) catalogue and the intensities converted to M
magnitude (Oncel,
The Gardner-Knopoff method window method was used for producing the mainshocks catalog (Gardner and Knopoff, 1974) since it is proposed to be the best technique for removing aftershocks when the earthquake catalog has variable quality station coverage in different regions and time periods (Savage and Depolo, 1993).
Earthquake recurrence statistics were calculated for each of the three zones, based on the maximum likelihood methodology of Weichert (1980). This method takes into account unequal completeness periods for different magnitude ranges. Parameter uncertainties were computed after Bollinger and others (1989). The event counts and completeness periods for each zone are listed in Table 1 . The same completeness periods were used for each zone.
Incremental recurrence curves for each of the three zones are shown in Figure 2 . Incremental curves, as opposed to more commonly seen cumulative curves, are displayed because the rates for a particular magnitude interval, and the associated uncertainty distribution for that interval, are input directly into the PSHA. A magnitude interval of 0.5 units was used in all cases. 95% data and model bounds are shown in the curves. The data error bounds refer to each observation as an independent Poisson variable, while the model bounds reflect how well each data point fits the maximum likelihood solution.
The recurrence curve for the Western Zone was originally found to exhibit distinctly non-linear behavior in the magnitude 6 to 7 range (e.g., Bath and Duda, 1979; Karnik and Klima, 1993). This has been noted by other workers, and has been cited as evidence for "characteristic" earthquake behavior, where larger events occur more frequently than the Gutenberg-Richter relation for smaller events would suggest (e.g., Youngs and Coppersmith, 1985). To account for this, recurrence calculations for the Western Zone were calculated independently for three magnitude ranges: 4.0 - 5.75, 5.75 - 7.25, and 7.25 - 8.0. While the physical basis and validity of the "characteristic" recurrence model is a subject of ongoing debate (e.g., Wesnousky, 1994), in this case a segmented model appears to fit the data well.
In contrast, the Central Zone appears to be well modeled by a linear recurrence relation. Although a suggestion of higher than "normal" rates in the 6.5 - 7.5 range is evident, they lie within the 95% confidence bounds of the maximum likelihood solution. The curve for the Eastern Zone is poorly constrained for magnitudes of 6 and above. This is most probably due to the smaller number of events, and more uncertain detection capabilities in this region. Because earthquakes in the magnitude 4.0 - 4.5 range fell significantly below the computed curve (probably due to incomplete detection), they were deleted from the data set. The poor fit to the empirical data lessens our confidence in the PSHA results for the Eastern Zone. The a and b Gutenberg-Richter recurrence parameters (in the formula LogN = a - bM) are shown in Table 2 , along with their errors at one standart deviations. The a values have been normalized to km
A probabilistic seismic hazard analysis was conducted for the site, using the general methodology of Cornell (1968). Because of the areal nature of the seismic source characterization, the results presented here can be considered valid for locations more than about 50 km from the zone boundaries. The hazard for locations near the boundaries of adjoining zones can be calculated; however this is beyond the scope of the present paper. The hazard at locations near (about 50 km or less from) the outer boundaries will be influenced by seismicity rates outside the zones considered here.
The hazard was calculated for a site at the center of each zone. Earthquakes were distributed uniformly throughout each zone at 5 km spaced grid points, with the seismicity rates scaled appropriately. Earthquake depths were assumed to have a triangular probability distribution with a modal peak at 10 km, and an upper limit of 20 km. In order to estimate the minimum hypocentral depth of a non-surface rupturing earthquake, depths were assumed to be equivalent to the radius of a circular crack with a 100 bar stress drop (Kanamori and Anderson, 1976), occurring on an "average" 60o dipping fault. Magnitudes of 6.75 and above were assumed to break the surface, and thus have zero depth, as used in the source to site distance calculations.
A single attenuation relation, that of Sadigh and others (1997), was used. Two others, Spudich and others (1997), and Abrahamson and Silva (1997) were used in the Central Zone peak acceleration calculations, and were found to give results within a few percent of those resulting from the Sadigh and others (1997) relation. All relations are based largely on California and western United States earthquakes. A relation developed specifically for the NAFZ may give different results. The tectonic character of the two regions consists of a major plate boundary with predominantly strike-slip faulting, however, and thus are similar in gross tectonic features. Maximum magnitudes in each zone were taken from Oncel and Alptekin (1998); the Western Zone limit was set at 7.75, the Central Zone at 8.0, and the Eastern Zone at 7.75. The lower magnitude limit was set to 5.0, and "rock" site conditions were assumed.
Uncertainties in ground motion attenuation and seismicity rates were treated by complete enumeration of the respective distributions. This was accomplished by discretizing the lognormal distributions for ground motion attenuation into 50 equal-probability values, and for seismicity rates into 20 equal-probability values. Discretization intervals greater than these were found to have an insignificant impact on the results. The distributions were bounded at 2.5 standard deviations above and below the median value. The random source model was then executed for each parameter combination and each of five depth values, resulting in 50 x 20 x 5 = 5,000 annual frequencies for each ground motion exceedance level. The arrays of annual frequency values for each exceedance level were then ordered and integrated into probability density functions. By this procedure, 50th percentile values are mean values.
The results presented here consist of hazard curves for peak horizontal acceleration (PHA), and uniform hazard acceleration response spectra at 5% damping (UHS) for return periods of 100, 500, 1,000, and 5,000 years (corresponding to annual probabilities of exceedance of .001, .002, .0001, and .0002). These were chosen to represent annual probabilities and risk levels appropriate for a variety of engineered structures. Other PSHA products, such as hazard curves for different response periods, UHS for different return periods, various fractile levels for the hazard curves and UHS (only means are shown for the UHS here), and deaggregation results for use in selecting acceleration time histories for dynamic analysis (e.g., McGuire, 1995), could also be derived from the basic modeling procedure.
Figure 3 show mean hazard cur ves for PHA for the three zones, along with 16th and 84th fractile curves.
Figure 4 displays the entire distribution surface of PHA exceedances for the Western Zone. In thi s plot, the log
Several probabilistic seismic hazard results that have been suggested for Turkey in general (e.g., Bath, 1979, Burton et al., 1982, Erdik et al., 1985; Gulkan et al., 1993) and the NAFZ in particular (e.g., Erdik and Oner, 1982; Alptekin et al., 1992). The study of Erdik et al., (1985) and Gulkan et al., (1993) that are the most important examples computing the seismic hazard map for Turkey separate the region into subregions considering the seismotectonic nature and computes the peak ground accelerations based on the Cornell-type model for the different return periods similar to this study, but use the raw earthquake data for the instrumental period without attempting to homogenize catalogue by declustering difference than the present study. Also, the parameter a of the Gutenberg-Richter equation that changes with length of time and size of the area are used in their study without normalizing to the area (km
This study represents an alternative to a purely fault-based hazard characterization (Reiter, 1991). It is simplistic in that it assumes nothing about the location, slip rates, and magnitude distributions of specific faults. Instead, it only assumes that seismicity rates will continue to be those seen in the historic record, and that earthquakes can occur with equal spatial probability in the "vicinity" of the major North Anatolian fault zones. A more accurate assessment would model individual faults with estimates of slip rates, magnitude distributions, and maximum magnitudes on each segment, and an accounting for "random", smaller magnitude seismicity not associated with mapped faults. It would be interesting to compare such an assessment to that presented in this paper.
Comparing results from the three zones, it is evident that the results are quite similar, with values for the more active Western Zone slightly higher than the other two. The effect of seismicity rate uncertainty is illustrated in figures 3 and 4 . The fractile PHA hazard curves for the Western Zone (Figure 3) are closer together than those for the other two zones, reflecting both a better fit to the historical seismicity, and also a larger data set. The broader uncertainty bounds are evident in the results for the Eastern Zone. Greater rate uncertainty also tends to drive up the overall level of the curves. Figure 4 also shows that increasing return periods result in broader uncertainty bounds in ground motion estimates. This reflects the lesser confidence in predicting infrequent events.
As stated in Section 5, these results should be considered valid at distances greater than about 50 km from the zone boundaries. Results for locations near adjoining zones will be some weighted average of the results of each zone, and locations near the outer zone boundaries will be affected by seismic activity beyond the zones considered here.
Acknowledgement
We are grateful to Dr. Arabasz and Dr.Ian Main for their constructive criticism of an earlier draft of this paper.
References
Turkish), MSc.thesis,pp. 93, Istanbul University.
Eastern Sierra Nevada, Bull. Seism. Soc. Am., 83, 1910-1938.
939-965.
