ground-motion prediction equations of intermediate...

Ground-Motion Prediction Equations of Intermediate-Depth Earthquakes

in the Hellenic Arc, Southern Aegean Subduction Area

by A. A. Skarlatoudis, C. B. Papazachos, B. N. Margaris, C. Ventouzi,I. Kalogeras, and the EGELADOS Group

Abstract A response-spectra database is compiled of hundreds of seismic recordsfrom intermediate-depth earthquakes (earthquakes whose foci are located between45 to 300 km from the earth’s surface) with moment magnitudes of M 4.5–6.7 thatoccurred in the South Aegean subduction zone. The database consists of high-qualitydata from both acceleration-sensor and broadband velocity-sensor instruments. Thedatabase is much larger than previous databases used in the development of past em-pirical regressions enabling the determination of various parameters of ground-motionattenuation not previously examined. New variables accounting for the highly com-plex propagation of seismic waves in the Greek subduction zone are introduced basedon the hypocentral depth and the location of the event, as these factors control theeffects of the back-arc low-velocity/low-Q mantle wedge on the seismic-wave propa-gation. The derived results show a strong dependence of the recorded ground motionson both hypocentral depth and distance, which leads to the classification of the datasetinto three depth-hypocentral distance categories. Ground motions from in-slab earth-quakes, especially with hypocentral depths �h� > 100 km, are amplified for along-arcstations, an expected effect of channeled waves through the high-velocity slab. Theground motions are also strongly attenuated in the back-arc region, due to the low-Qmantle wedge, which are almost independent of the recording hypocentral distance. Incontrast, for shallower in-slab events (60 km < h < 100 km), the corresponding dif-ferentiation of seismic motion for along-arc and back-arc stations is observed beyonda specific critical distance range. Moreover, for longer periods, both along-arc ampli-fication and back-arc anelastic-attenuation factors strongly diminish, suggesting thatthe longer wavelengths of seismic waves are not affected by the complex geophysicalstructure, resulting in more similar ground motions for both back-arc and along-arcstations. Finally, results for interface events (h < 45 km) occurring along the outerHellenic arc suggest their wave propagation is not affected by the presence of thelow-velocity/low-QS mantle wedge, but is mainly controlled by the differences of theanelastic attenuation between the Mediterranean and Aegean lithospheres.

Introduction

Ground-motion prediction equations (GMPEs) for earth-quakes that occur in subduction zones are often an importantinput for seismic-hazard analysis. Significant hazard canoriginate from earthquakes both along the subduction inter-face as well as from large events within the subducting slab.The southern Aegean area is located along an active plateboundary environment (Hellenic arc) and has a complicatedgeological and seismotectonic setting (Fig. 1). Thrust-faultinterface earthquakes are found at shallow depths (typically30–60 km) while oblique-thrust intermediate-depth (in-slab)earthquakes occur along a well-defined Wadati–Benioff zoneat depths ranging from 60–170 km (Papazachos and

Comninakis, 1969; LePichon and Angelier, 1979). The fociof the intermediate-depth earthquakes form two segments ofthe southern Aegean–Benioff zone with different dipping an-gles. The first shallower segment (focal depths betweenroughly 30 and 90 km) has a lower dipping angle and cor-responds to the external (outer-arc) section of the Benioffzone (see Fig. 1), which extends below the outer sedimentaryHellenic arc. The second deeper segment (depths ∼90–160 km) corresponds to the internal (inner-arc) Benioff zonesection (see also Fig. 1), dipping steeply below the southernAegean volcanic arc (e.g., Papazachos, 1990; Papazachoset al., 2000).

1952

Bulletin of the Seismological Society of America, Vol. 103, No. 3, pp. 1952–1968, June 2013, doi: 10.1785/0120120265

The back-arc area, schematically shown in Figure 1, ex-hibits very low levels of ground motions for intermediate-depth events as is evident in instrumental recordings and, toa large extent, the damage pattern of large intermediate-depth earthquakes (Papazachos and Comninakis, 1971). Thesouthern border of this area coincides with the volcanic arc,in agreement with the suggestion that this strong attenuationis related to the presence of the volcanic arc and the associ-ated mantle wedge (Papazachos et al., 2005; Boore et al.,2009; Skarlatoudis et al., 2009).

Because of the highly complicated structure of theGreek subduction zone, seismic-wave propagation pathsfrom the earthquake source to the ground surface vary sig-nificantly, depending on the earthquake type and recordingsite. Ground motions generated by different types of earth-quakes (in-slab, interface, and crustal events) exhibit promi-nent differences at the various recording sites even for eventsthat have identical magnitudes and hypocentral distances.For example, intermediate-depth earthquakes in the Hellenicsubduction zone are characterized by very large differencesof the recorded amplitudes at along-arc and back-arc stations,with high-frequency recordings strongly attenuated (mainlythe S-wave phase) at back-arc stations (e.g., Papazachos andComninakis, 1971). This finding was confirmed by laterstudies, such as the work of Konstantinou and Melis (2008)who identified high-attenuation areas in the central/NorthernAegean (back-arc area) and low-attenuation zones in the sub-ducting slab area (along-arc region) by analyzing the shear-wave propagation properties from intermediate-depth eventsalong the Hellenic subduction zone. Moreover, Boore et al.(2009) have also found significant differences in the ex-pected ground-motion levels when comparing 5% pseudo-

spectral acceleration (PSA) values for various periods fromrecordings of the strong 8 January 2006 Kythera M 6.7 in-termediate-depth earthquake and recordings of two shallowearthquakes of comparable magnitudes at similar hypocen-tral distances.

In the present study, a response-spectra database ofground-motion recordings of interface and in-slab events,with magnitudes M 4.5–6.7, which occurred in the southernAegean subduction zone has been compiled. The hypo-central parameters as well as the magnitudes of these earth-quakes are compared against the values reported by severalinternational seismological centers and research institutes inorder to compile a unified earthquake catalog. The size of thedatabase is much larger than that used in previous regressionsfor Greek subduction-zone earthquakes because it includesdata from permanent accelerometric and seismological net-works, as well as temporary seismological networks thatoperated for a limited time period in the southern Aegeanarea. The database compiled for earthquakes that occurredbetween 1994 and 2008 contains 743 horizontal-componentresponse spectra (5% of critical damping), which are used toexplore various aspects of the ground-motion scaling withmagnitude and distance.

Database Used for the Regression

The database that was used for the regression analysisconsists of existing data from previous work (Boore et al.,2009; Skarlatoudis et al., 2009) as well as additional datathat have become available over the time period that fol-lowed. In Table 1, the earthquakes (depths > 45 km) thatwere used in the present study are listed. The spatial distri-bution of these earthquakes is presented in Figure 2 togetherwith the corresponding fault-plane solutions. As a first step,the earthquakes were separated into in-slab and interfaceevents. This separation was based on experience from similarwork for global data (Atkinson and Boore, 2003), as well ason the source location relative to the subducting slab as con-firmed by previous studies (Papazachos and Comninakis,1971; Konstantinou and Melis, 2008; Boore et al., 2009;Skarlatoudis et al., 2009).

The intermediate-depth event categorization was mainlybased on their spatial location in the Hellenic arc (see Fig. 2),as well as their hypocentral depth (typically ∼50–170 km forin-slab events and ∼30–60 km for interface ones; e.g., Papa-zachos, 1990). Moreover, we also considered the existingknowledge on the active tectonics of the southern Aegeansubduction area. As can be seen in Figure 1, interface eventsoccur on the outer Hellenic arc with mostly thrust faults (e.g.,Papazachos and Delibasis, 1969; McKenzie, 1972), whileintermediate-depth in-slab events occur mainly in the innerHellenic arc, typically with oblique thrust mechanisms thatshow a characteristic down-dip extension and arc-parallelcompression (Taymaz et al., 1990; Kiratzi and Papazachos,1995; Benetatos et al., 2004). Following this categorization,epicenters for earthquakes characterized as in-slab are shown

Figure 1. A schematic geotectonic setting of the Hellenic sub-duction and its Benioff zone. The Aegean microplate is overridingthe African plate at a convergence rate of ∼35–40 mm=yr in anortheast–southwest direction (thick black arrow). Typical fault-plate solutions (Papazachos et al., 2000; black and gray focalmechanism plots) are in very good agreement with the regional sub-duction tectonics. The shaded area roughly depicts the high-attenuation area (back-arc) as identified by previous studies, clearlyassociated with the volcanic arc. The color version of this figure isavailable only in the electronic edition.

GMPEs of Intermediate-Depth Earthquakes in the Hellenic Arc 1953

by the white triangles (black and white focal mechanismplots), while black triangles (gray and white focal mecha-nism plots) denote epicenters for earthquakes characterizedas interface events in Figure 2. As can be seen from this fig-ure, the fault-plane solutions of the examined events are inquite good agreement (especially the larger ones) with thetypical fault-plane solutions presented in Figure 1 for interfaceand in-slab events. The final categorization is listed in columnCR of Table 1 (0 for in-slab events and 1 for interface, respec-tively). The international seismological centers and institutesthat reported the hypocentral parameters that were finallyadopted for these earthquakes are also given in the last columnof Table 1.

Because originally reported moment magnitudes wereavailable only for a few earthquakes, additional checks wereperformed before adopting the final moment magnitudeslisted in Table 1. For this reason mb and ML magnitude es-timates reported from the International Seismological Center(ISC), the U.S. Geological Survey (USGS), the NationalEarthquake Information Center (NEIC), and the GeophysicalSurvey of Russian Academy of Sciences (MOS) have beenused to estimate an equivalent moment magnitude,M�, usingappropriate conversion relations proposed by Scordilis(2006). The equivalent moment magnitude, M�, determinedthis way has been compared with the originally reported one,Mw, whenever such an original estimate was available (e.g.,

from Global Centroid Moment Tensor [Global CMT] solu-tions or the catalogs of the Geodynamic Institute of theNational Observatory of Athens and the Seismological Sta-tion of Aristotle University of Thessaloniki). The compari-son showed a very small scatter between the originallyreported and equivalentmomentmagnitudes for the examinedearthquakes, with a mean difference M� −Mw of 0.04 and astandard deviation of 0.16. This excellent agreement justifiesthe approach to use M� as equivalent moment magnitudewhenever original Mw values are not available. The final-moment magnitude (original or otherwise equivalent) willbe denoted as M.

Among the earthquakes used for compiling the presentstudy dataset, the 8 January 2006 Kythera earthquake(M 6.7, h � 67 km) has contributed a large portion of thedata used in the regression analysis as it was recorded byboth temporary and permanent velocity and acceleration-sensor networks. Velocity and acceleration data were alsoavailable for almost all the intermediate-depth earthquakesthat comprise the present study dataset.

In Figure 2, the spatial distribution of the acceleration-sensor and broadband velocity-sensor stations that are usedin the present study is shown. The majority of these stationsbelong to the same networks and/or institutes that provideddata for the Boore et al. (2009) and Skarlatoudis et al. (2009)studies. However, additional stations from CYCNET (Bohnhoff

Table 1Earthquakes Used in the Present Study

Id Origin Time (yyyy/mm/dd hh:mm:ss.ss) Latitude (°) Longitude (°) Depth (km) Mw CR* Source†

1 1994/05/23 06:46:12.00 35.5409 24.6968 68 6.1 0 ISC2 2003/04/29 01:51:20.20 36.9395 21.7314 66 5.1 1 ISC3 2003/09/13 13:46:21.68 36.6910 26.8488 134 5.2 0 CYG4 2004/03/28 14:54:38.26 35.5700 22.9900 55 4.7 1 HRVD5 2004/11/04 06/22/37.56 35.9633 23.1454 70 5.2 0 ISC6 2005/08/01 13:34:58.92 36.6092 26.6775 127 4.8 0 EGE7 2005/11/20 21:20:56.50 35.0332 27.2676 50 4.6 1 ISC8 2006/01/08 11:34:54.64 36.1853 23.4037 67 6.7 0 THE9 2006/05/11 01:47:47.61 36.1256 23.3697 72 4.5 0 ISC

10 2006/05/15 04:22:39.87 35.7490 25.9830 68 4.7 0 EGE11 2006/07/09 03:12:54.22 36.4597 27.2451 118 4.6 0 ISC12 2006/12/02 10:26:54.60 34.7687 26.8962 52 4.7 1 ISC13 2007/02/03 13:43:22.10 35.8092 22.6367 47 5.4 1 EGE14 2008/01/06 05:14:20.18 37.2569 22.7037 84 6.2 0 ISC-NEIC15 2008/03/28 00:16:19.90 34.7922 25.3423 49 5.6 1 ISC16 2008/06/18 01:58:42.90 37.6700 22.7800 83 5.1 0 ISC17 2008/07/15 03:26:34.70 35.8500 27.9200 56 6.4 1 ISC18 2008/09/16 02:58:39.80 36.6900 24.0300 137 4.5 0 ISC19 2008/11/04 12:05:43.50 36.1900 23.3500 68 4.5 0 ISC20 2010/07/16 08:11:05.30 36.776 27.008 163 5.2 0 THE21 2011/02/25 21:33:29.90 36.645 27.011 118 4.4 0 THE

*CR: 0 for in-slab events, 1 for interface events.†Data source definitions: THE, Seismological Station of Aristotle University of Thessaloniki; EGE, EGELADOS

temporary seismological network deployed in the Southern Aegean area, coordinated by the Ruhr–University ofBochum (Germany) and operated by a large working group involving University of Thessaloniki, NationalObservatory of Athens, Technical University of Chania (Greece), Istanbul Technical University (Turkey), Universityof Hamburg and GeoForschungszentrum Potsdam (Germany); ISC, International Seismological Centre; HRVD,Global Centroid Moment Tensor database; NEIC, National Earthquake Information Center.

1954 A. A. Skarlatoudis, C. B. Papazachos, B. N. Margaris, C. Ventouzi, I. Kalogeras, and the EGELADOS Group

et al., 2004, 2006) and SIMBAAD networks (Paul et al., 2008)are also used in the present study. Additional information for theinstrumentation and data availability can be found in Data andResources.

The distribution of the data is shown in Figure 3 in termsof magnitude, hypocentral distance, and depth. Despite thesmall number of available earthquakes, there is a large num-ber of data, which can adequately cover a wide hypocentral-distance range, studied in terms of hypocentral depth. Therelative lack of data for deeper earthquakes (h > 100 km)and larger magnitudes (M > 5:5) is mostly due to the lowerseismicity of the deeper branch of the Wadati–Benioff seis-mic zone where smaller magnitude earthquakes occur incomparison to the shallower branch of the Wadati–Benioffzone (Papazachos, 1990; Papazachos et al., 2005).

Data Analysis

In order to construct an appropriate ground-motion pre-diction model for the two types of intermediate-depth earth-quakes (in-slab and interface) and to determine the form ofthe predicting relation to be used in the final regressionanalysis, we performed a detailed analysis of the main data-set properties. For this reason, a simple functional form,

logY � c1 � c2�M − 5:5� � c3 logR� c4R� c51S

� c52SS� ε; (1)

was initially used in order to perform separate regressions forin-slab and interface earthquakes for the study of the dom-inant distance scaling and the identification of the possibleprominent bias between back-arc and along-arc recordings.In equation (1) Y corresponds to peak ground acceleration(PGA), peak ground velocity (PGV), and 5% damped PSAfor the period range 0.01–4 s, M is the moment magnitude,R �

��D2 � h2

pis the hypocentral distance with D being the

epicentral distance (in km) and h the hypocentral depth, S isequal to 1 for the National Earthquake Hazards ReductionProgram (NEHRP) C soil conditions, and 0 otherwise, andSS is equal to 1 for NEHRP D soil conditions, and 0 otherwise(NEHRP, 1994; UBC, 1997). The Rot50 measure (Boore,2010) of the horizontal components of ground motion is usedin regressions for PGA and PGV and the RotD50 for the5% damped PSA. The specific measures of horizontal-component ground motions were selected because they arequite independent of the in situ orientations of the recordedground motions and represent them in a consistent way with-out computing geometric means. Moreover, the data used inthe regressions were subjected to maximum and minimumusable period limitations, defined by the sampling rate ofeach recording and the cutoff frequency of the low-pass filter(fclow) used in data processing, respectively. More specifi-cally, the minimum usable period of each recording was setto Tmin � 1:25 × TNyquist, while the maximum usable periodto Tmax � 0:67 × Tclow.

Classification of In-Slab Data in Depth Bins

The study of the 2006 Kythera intermediate-depth earth-quake by Boore et al. (2009) and Skarlatoudis et al. (2009)revealed that the expected levels of ground-motion attenua-tion for back-arc stations in the Hellenic arc are strongly de-pendent on the hypocentral distance of the recording station.This dependence was attributed to the presence of the low-velocity/low-QS mantle-wedge layer (mantle LVL) abovethe subducting slab and specifically on the distance traveledby seismic waves within this region. What could not be eval-uated in those studies was the attenuation dependence on thedistance traveled by seismic waves in this LVL as a functionof the hypocentral depth. On the basis of the knowledge ofthe seismotectonic setting (Hatzfeld et al., 1988; Papazachoset al., 2000) and regional travel-time and surface-wavetomography results (Spakman, 1988; Spakman et al., 1993;Papazachos et al., 1995; Papazachos and Nolet, 1997; Kar-agianni et al., 2005) for the geometry and spatial extensionof the mantle-wedge LVL, it is reasonable to expect that fordeeper earthquakes, the distance traveled by seismic waveswithin the mantle wedge is longer, hence the correspondinghigh-anelastic attenuation should be more evident especiallyin the S-wave phase.

In order to investigate the dependence of ground motionon the hypocentral depth for in-slab events, the correspondingdata were classified into three depth bins (60 ≤ h < 80 km,80 ≤ h < 100 km, and h ≥ 100 km, respectively). Moreover,

Figure 2. Acceleration- and velocity-sensor recording stations(see legend for symbols) and spatial distribution of the analyzedearthquakes, including their corresponding fault-plane solutions.White stars (black and white focal mechanism plots) correspondto epicenters of in-slab earthquakes, while black stars (gray andwhite focal mechanism plots) depict epicenters of earthquakes clas-sified as interface. The white square and diamond denote the loca-tions from two additional earthquakes used for result evaluation (seecorresponding text). The color version of this figure is availableonly in the electronic edition.


the recording stations were also separated in back-arc andalong-arc, based on their location, using the back-arc areashown in Figure 1. Although the extent of the high-attenuationback-arc area is not accurately known, the border depicted inFigure 1 was based on the preliminary analysis by Skarlatou-dis et al. (2009) and Boore et al. (2009) as well as regionalhigher-resolution P and S tomographic results (Papazachoset al., 1995; Papazachos and Nolet, 1997), which clearlydelineate the spatial extent of the low-velocity (low-Q) mantlewedge at depths 50–80 km beneath the southern Aegean vol-canic arc. After performing an initial regression using the sim-ple functional form of equation (1), the corresponding data arepresented in Figure 4 against the hypocentral distance for PGAand four selected periods 0.025 s, 0.2 s, 1 s, and 4 s after beingreduced to magnitude M 5.5 and rock-site conditions usingthe preliminary coefficients c2, c51, and c52. In this figure, wepresent running averages rather than actual data in order tofacilitate the visualization of the main differences amongback-arc and along-arc data. An impressive difference of thelevel of ground motion for hypocentral depths h ≥ 100 km,for back-arc and along-arc data is observed, which reaches

roughly one order of magnitude for PGA and short period/higher frequency PSA (0:025 s=40 Hz). This differencebecomes smaller for the shorter periods (1 s and 4 s), sug-gesting a frequency-dependent attenuation mechanism, suchas anelastic attenuation.

Although the back-arc/along-arc bias is extremely largefor all hypocentral distances for deep events (h > 100 km),this is not the case for the other two categories studied.For hypocentral depths 60 ≤ h < 80 km, even though theground-motion levels from back-arc and along-arc dataare almost identical for small hypocentral distances, a differ-ence between back-arc and along-arc observations is gradu-ally built within a critical distance range, which for this caseis roughly defined from 205 to 355 km (vertical solid lines).At larger hypocentral distances (Rhyp > 355 km), the ob-served back-arc/along-arc difference obtains its maximumvalue, which remains practically constant for larger distances,similar to the constant bias observed for deep (h > 100 km)events for the whole distance range. Similarly, for data fromevents with hypocentral depths 80 ≤ h < 100 km, the criticaldistance for which differences in the levels of ground motion

Figure 3. The distribution of the earthquake-moment magnitude and hypocentral depths, as well as the corresponding hypocentral dis-tances for the database used in the present study. In-slab earthquakes are denoted by the open circles, while interface earthquakes are depictedwith the open triangles.


start to gradually build up between back-arc and along-arcrecordings is Rhyp ∼ 140 km. The back-arc/along-arc biasobtains its maximum rather constant value, for distancesgreater than Rhyp ∼ 240 km (the distance range is denotedwith vertical dashed lines in Fig. 4).

The results presented in Figure 4 exhibit several interest-ing characteristics. As earlier noted, the fact that the differencebetween back-arc and along-arc recordings is more prominentfor higher frequencies suggests that intrinsic (anelastic) at-tenuation plays an important role. Moreover, this attenuationis stronger for deeper events, which is expected if the source ofattenuation is the low-velocity mantle wedge above thesubducting slab. On the other hand, the appearance of the

back-arc/along-arc bias at progressively larger distances forshallower events suggests a specific wave-propagation patternthat allows themantlewedge to affect thewaveformonly after acertain epicentral/hypocentral distance. Furthermore, theintroduced attenuation from this low-Q area has a maximumeffect that does not increase for larger epicentral distances.This observation is compatible with a specific, limited-sizeattenuation mantle wedge “pocket,” hence attenuation doesnot further increase for more distant recordings.

A second important observation from Figure 4 is that theobserved distance-decay pattern, which cannot be explainedonly on the basis of a mantle-wedge high attenuation. This isclearly seen by the fact that, for example, the along-arc datafor h ≥ 100 km have much higher PGAs than the corre-sponding values for shallower focal depths. Furthermore,after the initial critical distance at which PGA and PSA startto deviate for h < 100 km curves, the along-arc distance de-cay exhibits a much smaller slope than for shorter distances.These observations are compatible with an amplification ef-fect for along-arc recordings. Such an effect could be a resultof high-amplitude channeled waves traveling through thesubducting slab (e.g., Okal and Talandier, 1997) or a subduc-tion channel, as already shown is possible for the southernAegean subduction zone (Essen et al., 2009). Similar to ane-lastic attenuation, this effect is also expected to be reducedfor longer periods, as longer wavelengths do not “see” theslab and its effect, explaining the convergence of distance-decay patterns for longer periods seen in Figure 4.

In order to adequately capture the previously describedcharacteristics of ground motion seen in Figure 4, appropri-ate variables accounting for the “additional” attenuation oramplification of the data were incorporated in the final re-gression model. The data that were recorded at short hypo-central distances (shorter than the estimated critical distancewhere along-arc/back-arc bias starts to appear) are assumedto be the “reference” data that control the “average” level ofground motions relative to which the “additional” attenuationor amplification is defined. Clearly, data from the depthrange h > 100 km do not belong to this group. Becauseof the small number of data in the 80 ≤ h < 100 km bin, theobserved differences with respect to the 60 ≤ h < 80 km bincould not be incorporated as separate variables in the finalregression model, hence common variables were used to takeinto account for the “additional” attenuation and/or amplifi-cation introduced to the corresponding back-arc and along-arc data after the critical distance.

The previous data patterns and conceptual description ofthe main attenuation/amplification mechanisms are summa-rized in Figure 5, which schematically presents the mainpatterns of wave propagation along a profile parallel to thesubduction direction. Along the subducting slab, both inter-face (outer-arc, shallower depth, mainly thrust-faults), as wellas in-slab events (inner-arc, larger depth, oblique thrust) aregenerated. Seismic waves from shallower in-slab eventspropagate along the Aegean lithosphere and are recordedat short distances at both back-arc (behind the volcanic

Figure 4. Running averages for peak ground acceleration (PGA)and 5% damped pseudospectral acceleration (PSA) data for the peri-ods of 0.025, 0.2, 1, and 4 s, reduced to M 5.5 and rock-site con-ditions, and plotted against Rhyp for the preliminary regression. Dataare classified into three hypocentral-depth bins, h ≥ 100 km (thickblack and gray curves), 80 ≤ h < 100 km (dashed black and graycurves), and 60 ≤ h < 80 km (thin black and gray curves), respec-tively. Black lines correspond to the back-arc data, while gray linescorrespond to along-arc data. The vertical dashed and solid linesdenote the two critical hypocentral-distance ranges, 140–240 kmfor the depth bin 80 ≤ h < 100 km and 205–355 km for the depthbin 60 ≤ h < 80 km, for which the back-arc/along-arc bias gradu-ally develops (see text for explanation).


front) and along-arc stations (outer-arc) without any dif-ferentiation. Thesewaves (denoted as “normal” in Fig. 5) con-stitute the “reference” propagation data for hypocentraldepths < 100 km. At larger distances the waves are eitherchanneled, hence amplified, through the fast subducting slab(towards the south, outer-arc) or they cross the slow, low-Qmantle wedge, hence are attenuated, resulting in the observedPGA/PSA bifurcation at a certain critical distance. This back-arc/along-arc bias increases at larger hypocentral distancesuntil waves cross the entire mantle-wedge “pocket,” hence at-tenuation does not further increase at more distant stations.For deeper events, this pattern occurs at shorter distances dueto the position of the hypocenter relative to the mantle wedge.Eventually, for deep events (h > 100 km) that occur in theslab section after its slope change, as verified by regionaltomography and the available Benioff-zone data (PapazachosandNolet, 1997; Papazachos et al., 2000), the back-arc/along-arc bias occurs for all records, as waves are either channeledthrough the high-Q slab or have to travel the low-Q mantlewedge. As the low-velocity/low-Qmantle wedge has specificlimited dimensions beneath the volcanic arc (as verified byregional tomography; e.g., Papazachos et al., 1995) all wavestravel the same distance through the high-attenuation area,hence the back-arc/along-arc difference remains practicallyconstant for events with h > 100 km for the entire examineddistance range, as seen in Figure 4. Additional support for theconceptual model of Figure 5 is provided by the final regres-sion results, later presented.

In addition to the previous data classification accordingto the event focal depth, new criteria for the characterizationof back-arc and along-arc stations were adopted in order

to better describe the observed ground-motion properties ofeach record for the different earthquake locations. The origi-nal classification we employed in Figure 4 used a rough def-inition of the high-attenuation back-arc area, on the basismainly of the proposed boundary of the mantle-wedge LVLfrom tomography and the corresponding earthquake and therelative recording station location. This definition is adequatewhen studying the properties of a single earthquake but mayfail to describe ground-motion properties when studying theproperties of several earthquakes.

The modified approach is presented in Figure 6a. For adeeper event (event number 2) that lies in the back-arc area(beneath the volcanic arc) and a large hypocentral depth(>100 km, see Benioff zone in Figs. 1 and 5), the back-arc/along-arc definition is not changed as ground motions inback-arc stations (station A) are strongly attenuated, whileground motions at outer-arc stations (B and C) are amplifiedas they travel along the subducting slab. For shallower eventscloser to the outer-arc (in general h < 100 km) such as eventnumber 1 of Figure 6, stations A and B are also classified asback-arc and along-arc, respectively. However, waves trav-eling to station C are characterized as back-arc, even if thestation is located in the outer-arc, because seismic waveshave to travel through the back-arc, high-attenuation area.This simple geometrical modification was employed for re-cordings of shallower in-slab events (h < 100 km), in orderto roughly account for 3D attenuation effects that otherwisewould normally require a full 3D wave-propagation approach.

In Figure 6b, we present the actual effect of the back-arcattenuation and the outer-arc high-frequency channeling am-plification on real data. More specifically, for a deep event

Figure 5. A schematic presentation of the main patterns of wave propagation for intermediate-depth events in the Hellenic arc, along aprofile parallel to the subduction direction. The main geophysical features affecting the wave propagation (high-Q subducted slab, low-Qmantle wedge, and so forth) are also depicted. The color version of this figure is available only in the electronic edition.


(h � 134 km, event number 3 in Table 1), we present thespatial distribution of PGA/PGV ratios only for rock sites(class B). Although this ratio is usually employed for siteeffects assessment (Seed et al., 1976), it has been success-fully used to depict the frequency content of intermediate-depth earthquakes such as the Kythera event (Skarlatoudiset al., 2009). In our case, the whole back-arc area exhibitsvery low PGA/PGV values as a result of the anelastic-attenu-ation effect, which mainly damps higher frequencies, henceaffecting mainly PGA. Similarly, the outer-arc exhibits veryhigh PGA/PGV values, indicating higher frequency amplifi-cation, compatible with the idea of high-frequency channeledwaves through the subducting slab. Notice that the spatialextent of the back-arc area defined on the basis of the avail-able geophysical information coincides very well with thelow PGA/PGV region, with the values of PGA=PGV ∼12roughly corresponding to the back-arc/along-arc boundary.

Following the previous discussion, recording stationsare classified as back-arc and along-arc on the basis of the

relative position of the recording station and the earthquakehypocenter, also taking into account the distance criteria pre-viously defined for the depth bins. Hence, for events withhypocentral depth h ≥ 100 km, recording stations are classi-fied only as back-arc or along-arc, depending on the locationof the recording station, independent of their hypocentral dis-tance. For shallower in-slab events, with hypocentral depths60 ≤ h < 80 km or 80 ≤ h < 100 km, recording stations areclassified as: (a) back-arc or along-arc, if their hypocentraldistance is longer than the upper limit of the distance rangeof the corresponding depth bin where data bifurcation(along-arc/back-arc bias) is found; (b) reference data, if theyare recorded at distances shorter than the lower limit of thedistance range specified for the corresponding depth bin; and(c) intermediate back-arc or along-arc, if they are distributedwithin the distance range defined for the corresponding depthbin where data bifurcation gradually develops.

Study of the Distance-Decay Rates

The distance-decay rates, for both in-slab and interfaceevents, were studied for the various PSA periods before pro-ceeding in the final regression. For in-slab events, the hypo-central distance range for which data are available does notallow the reliable estimation of the geometrical spreading co-efficient as an independent variable due to the large trade-offwith the anelastic-attenuation coefficient and the lack of dataat short hypocentral distances (due to the large hypocentraldepths). In order to estimate a representative geometricalspreading coefficient, the method applied by Atkinson andBoore (2003) was adopted and the simple functional formof equation (1) was used to perform a regression to the dataconsidering a fixed, a priorivalue for the anelastic-attenuationcoefficient, c4 � exp�0:001R�. The regression was per-formed for two typical PSA periods, namely 0.5 s and 1 s, andfor two magnitude bins 4:5 ≤ M < 5:5 and 5:5 ≤ M ≤ 6:7 inorder to also explore possible magnitude dependence. Theestimated values for the geometrical spreading coefficient,c3, ranged from −1:60 to −1:77, with an average value of−1:70. These results also showed that the magnitude depend-ence of the coefficient is negligible and in very good agree-ment with the findings of Atkinson and Boore (2003) forin-slab earthquakes that also estimated a similar value(∼ − 1:75) for the twomagnitude bins studiedwith very smallmagnitude dependence for in-slab events. Based on these re-sults, the finally adopted value for the geometrical spreadingcoefficient for in-slab events was adopted as −1:7 and washeld fixed for all regressions.

For interface events, a similar procedure was also ap-plied, however the very small number of data available atshort hypocentral distances did not allow the estimation ofa robust value for the geometrical spreading coefficient, asthe regression analysis resulted in unrealistically high values.Thus, considering the similar decay rate of the interfaceevents with the available data for in-slab events, we adoptedthe same geometrical spreading coefficient.

Figure 6. (a) A revised definition of back-arc/along-arc charac-terization for in-slab events: For deeper events (h > 100 km, event 2)stations are characterized by their position in the back-arc/along-arcarea, while for shallower in-slab events (h < 100 km, event 1) seis-mic energy has to travel through the higher-attenuation area to reachcertain outer-arc stations (e.g., station C), hence the correspondingpath is redefined as “back-arc”. (b) Spatial variation of the PGA/PGVratio for a deep event (h � 134 km, event 3 in Table 1). Notice thatthe low PGA/PGV values in the back-arc area and the very highPGA/PGV values in the outer-arc, in agreement with the proposedback-arc anelastic attenuation and the along-arc channeled wavepropagation, both mainly affect higher frequencies. The colorversion of this figure is available only in the electronic edition.


Final Regression Analysis

In-Slab Events

The final regression of the dataset was performed usingthe mixed-effects model, as implemented with the algorithmintroduced by Abrahamson and Youngs (1992). For themixed-effects model, the error term can be expressed as twoseparate terms, namely the interevent and intraevent error.Thus, the regression model has the following form:

logYij � f�Mi; Rij; hi; ξ� � ηi � εij; (2)

where ηi represents the error (bias) term for event i and εijrepresents the intraevent residual for recording j of event i.Event terms and the intraevent error are assumed to be nor-mally distributed with zero mean and standard deviation. Thefunctional form that was chosen for the regressions based onthe classification previously described was

logY � c1 � c2�M− 5:5� � c31 logR� c32�R− Rref�� c41�1−ARC��H�h− h0�� c42�1−ARC��H�h0 − h�f�h;R�� c51ARC�H�h− h0� � c52ARC�H�h0 − h�f�h;R�� c61S� c62SS� ε; (3)

and where

f�h; R� �

8>>>>>>>>>>>>>><>>>>>>>>>>>>>>:

if 60 km ≤ h < 80 km0 if R < 205 km�205 − R�=150 if 205 km ≤ R < 355 km1 if R > 355 km;

orif 80 km ≤ h < 100 km

0 if R < 140 km�140 − R�=100 if 140 km < R ≤ 240 km1 if R > 240 km

;

where H is the Heaviside function [H�ξ� � 0 for ξ < 0,H�ξ� � 1 for ξ ≥ 0] and h0 � 100 km. All logarithms arebase 10, M is the moment magnitude, R is the hypocentraldistance, h is the hypocentral depth, and ARC � 0; 1 forback-arc and along-arc stations (on the basis of the updateddefinition previously described in Fig. 6a). Finally, S � 1

and SS � 1 for soil (class C) and soft-soil (class D) sites,respectively (and 0 otherwise). The geometrical spreadingcoefficient c31 was fixed to −1:7, in accordance with theearlier distance-decay analysis and the reference distanceRref (Boore et al., 2009) was set to 1 km.

The functional form of equation (3) contains typicalterms (c1, constant; c2, magnitude term; c31, geometricalspreading term; c61, c62, site-effect term). Three types ofadditional terms are also introduced: (a) the c32 is a typicalanelastic-attenuation term however it mainly reflects the

anelastic attenuation of reference data; that is, data corre-sponding to “normal waves” (see Fig. 5) that are recorded atdistances where the back-arc attenuation/along-arc amplifi-cation do not appear; (b) the c41 and c42 terms that concernonly back-arc data (ARC � 0) and correspond to two differ-ent constant anelastic-attenuation terms (in agreement withprevious discussion), which apply for depths h > 100 km(c41) and h < 100 km (c42); and (c) the similar c51 and c52terms that concern only along-arc data (ARC � 1), and cor-respond to two different constant amplification terms, whichapply for depths h ≥ 100 km (c51) and h < 100 km (c52).These three families of terms attempt to account for the spe-cific attenuation/amplification features identified in Figure 4and schematically described in Figure 5.

A special handling was required for the previously de-scribed intermediate data, namely data for events with depthsh < 100 km, which belong to the distance transition zonefor the two depth ranges examined (60 km ≤ h < 80 km and80 km ≤ h < 100 km), where the back-arc/along-arc biasgradually develops. For this reason, a simple linear weightfunction, f�h; R�, was used to model these “transition”zones, that is, the critical distance range where the back-arc/along-arc difference gradually builds up, as these have beenalready previously defined for each depth bin. Therefore,corresponding intermediate back-arc and along-arc data areassigned a weighted contribution of the total back-arc ane-lastic attenuation or along-arc amplification, on the basis oftheir hypocentral distance.

Evaluation of the Results

The final coefficients from the regression of equation (3)are given in Table 2. The event terms as a function of recordnumber, moment magnitude, and hypocentral depth areshown in Figure 7, and no apparent trend can be identified.The results presented in this figure and especially the distri-bution of event terms with hypocentral depth show that theproposed classification of data in three depth bins was ap-propriate and produced relatively unbiased results. Thisassumption is also supported by the results presented inFigure 8, which show the distribution of interevent residualsas a function of hypocentral distance (the symbols and color-ing are the same as in Fig. 7). Both classes of data do notexhibit any trend with hypocentral distance for any of thethree periods shown.

In Figure 9, the variation of total misfit is plotted for theperiod range 0.01–4 s. The total misfit of the regressions ex-hibits the highest values (∼0:5) for the periods of 0.1 and0.2 s, while for longer periods the values are getting smaller,with the smallest value estimated for the period of 4 s (∼0:3).This general increase with frequency has been also identifiedin the single-event results of Boore et al. (2009; crossedcircles) and Cauzzi and Faccioli (2008), which is opposite ofthe trend found in empirical regression analysis of responsespectra (e.g., Abrahamson et al., 2008).


The variation of the final regression coefficients for theperiod of in-slab events is shown in Figure 10 by the thincurves and gives a general overview of the properties ofthe determined anelastic attenuation for back-arc and ampli-fication for along-arc data. Coefficients c41 and c42 account-ing for the additional attenuation of the back-arc data exhibitthe largest differences for shorter periods, while for longerperiods they both tend to zero, suggesting that the anelastic-attenuation source (mantlewedge) is not significant (“visible”)for longer wavelengths. Coefficient c51, accounting for theadditional amplification of the data from deep in-slab events,h ≥ 100 km, is practically stable for shorter periods andshows diminishing values for longer periods. A similar patternis observed for shallower in-slab data (60 ≤ h < 100 km),

where the amplifications (coefficient c52) are smaller thanfor deeper events for shorter periods, although for periodslonger than 1 s the amplifications are quite similar. Despitethe small number of data recorded on soil and soft soil con-ditions, the corresponding coefficients follow the trend ob-served in the corresponding global data results of Booreet al. (2009), showing a rather stable, frequency independentpattern, with slightly lower values in longer periods andrelatively higher values for the period range 0.1–1 s.

In order to evaluate the performance of the estimatedGMPEs, we plotted them together with the observed data fromearthquakes with 60 ≤ h < 100 km reduced to M 5.5 androck-site conditions (class B) for PGA and five periods(0.025, 0.2, 1, 2, and 4 s) in Figure 11. The definition of

Table 2Regression Coefficients for In-Slab Events

Period (s) c1 c2 c32 c41 c42 c51 c52 c61 c62 σ τ ε

PGA 4.229 0.877 −0.00206 −0.481 −0.152 0.425 0.303 0.267 0.491 0.352 0.112 0.369PGV 2.965 1.069 −0.00178 −0.264 0.018 0.390 0.333 0.408 0.599 0.315 0.144 0.3460.01 4.235 0.876 −0.00206 −0.482 −0.153 0.425 0.304 0.265 0.488 0.353 0.111 0.3700.025 4.119 0.877 −0.00202 −0.490 −0.140 0.415 0.326 0.301 0.511 0.352 0.103 0.3670.05 4.320 0.863 −0.00212 −0.483 −0.178 0.410 0.286 0.245 0.475 0.376 0.095 0.3880.1 4.565 0.867 −0.00244 −0.515 −0.185 0.452 0.371 0.234 0.442 0.404 −0.066 0.4100.2 4.613 0.842 −0.00199 −0.596 −0.221 0.396 0.291 0.289 0.469 0.379 0.154 0.4090.4 4.463 0.926 −0.00190 −0.427 −0.110 0.459 0.295 0.298 0.516 0.322 0.141 0.3511.0 3.952 1.102 −0.00178 −0.199 0.112 0.316 0.442 0.371 0.512 0.305 0.201 0.3652.0 3.281 1.260 −0.00106 −0.136 0.055 0.196 0.352 0.408 0.578 0.277 0.203 0.3434.0 2.588 1.384 −0.00039 −0.179 −0.046 0.113 0.189 0.264 0.475 0.278 0.176 0.329

The equation used is logY � c1 � c2�M − 5:5� � c31 logR� c32�R − Rref� � c41�1 − ARC��H�h0 − h� � c42�1 − ARC��H�h0 − h�f�h; R� � c5ARC�H�h0 − h� � c52ARC�H�h0 − h�f�h; R� � c61S� c62SS� ε. See equation (3) for the variabledefinition. Coefficients c31 and Rref have fixed values of −1:70 and 1 km, respectively.

Figure 7. The distribution of event terms as a function of (a) record numbers, (b) moment magnitude, and (c) hypocentral depth, for PGAand PSA for the periods 1 and 4 s. Gray circles correspond to data from deeper in-slab events (h ≥ 100 km). Black circles correspond to datafrom in-slab events (60 ≤ h < 100 km). Open squares correspond to interface earthquakes.


back-arc, along-arc, intermediate back-arc, intermediatealong-arc, and reference data for each depth bin was adoptedas earlier described in the article. For all periods studied, thepredicted results describe the observations with sufficient ac-curacy. Moreover, the good quality of the fit for the distancetransition zones should be noted, with an exception for theperiods of 1 and 2 s for which the small positive values ofcoefficient c42 combined with the high-positive values of co-efficient c52 result in quite flat (distance-independent) transi-tion curves for along-arc data. In Figure 12, the predictedcurves for the deeper in-slab earthquakes (h ≥ 100 km) fit theobservations quite adequately for the presented periods. How-ever, the most interesting feature is the continuously decreas-ing difference of the predicted ground-motion levels betweenback-arc and along-arc data for longer periods, implying thatthe mantle-wedge attenuation and slab amplification do notcontribute as significantly for lower frequencies. This behav-ior is not as significant in Figure 11, which is an additionalindication that shallower in-slab events have different wave-propagation characteristics than the deeper ones, in accor-dance with the description presented in Figure 5.

Present study results have been compared with similarGMPEs derived from worldwide or regional intermediate-

Figure 8. The distribution of intraevent residuals as a functionof hypocentral distance, Rhyp, for PGA and PSA for the periodsof 1 and 4 s.

Figure 9. Distribution of the total misfit of the regression for in-slab (black triangles) and interface (black circles) events. Crossedcircles depict the single-event misfit (Boore et al., 2009) of the 2006intermediate-depth Kythera earthquake (number 9 in Table 1) pre-sented for comparison.

Figure 10. Avariation of the PSA regression coefficients for in-slab (thin curves) and interface events (thick curves) for the periods0.01–4 s.


depth data, such as Atkinson and Boore (2003; herein re-ferred to as AB03), Kanno et al. (2006; herein referred to asKea06), and Zhao et al. (2006; herein referred to as Zea06),as well as with results from Boore et al. (2009; herein re-ferred to as Bea09) derived for a single event in the Aegeanarea (Kythera mainshock, event number 7 in Table 1).Figure 13 presents the predictions for M 6.7, rock-site con-ditions for a shallow in-slab event (h � 66 km) for PGA andPGV, as well as for 1 and 4 s PSA. In Figure 14, a similar com-parison is presented for a deeper in-slab event (h � 120 km).

The previous figures show that comparisons for theselected magnitude M 6.7 are more consistent among thevarious GMPEs presented for h � 66 km. The largest dis-crepancies are observed for distances up to ∼200 km (refer-ence curve), with present study results predicting a strongerdistance decay of ground motion with respect to Bea09results (determined for the same area for a single event).For larger distances, the results are in a good agreement withBea09, especially for PGA, while for PGV the along-arccurve predicts lower ground motions and a steeper slope fordistances larger than 350 km. Results for deeper events, such

as h � 120 km, show that the predictions for back-arc dataare lower compared to the other GMPEs and larger in generalfor along-arc ones, clearly as a result of the different defini-tion that we employ for back-arc and along-arc data in com-parison to other global and regional relations.

Figure 15 plots the predicted PSA amplitudes of anM 6.7 in-slab earthquake for two typical hypocentral depths(h � 66 and h � 120) at the distance of 66, 120, and 250 kmand for NEHRP class C (soil). In the top left figure(h � 66 km, Rhyp � 66 km), comparisons are performedwith AB03, Bea09, and Zea06 for the distance and frequencyranges that the corresponding predictions are valid. The re-sults verify the conclusions from Figures 13 and 14, where astrong variability of the compared GMPEs is identified.Nevertheless, there is a good agreement of the Zea06 predic-tion with this study’s results in most of the cases presented,while Bea09-predicted PSA levels are generally lower com-pared with this study’s results.

Figure 11. Ground-motion prediction equations for PGA andPSA for five periods (0.025, 0.2, 1, 2, and 4 s) plotted together withobserved in-slab data for depths 60 ≤ h < 100 km, reduced toM 5.5 and rock-site conditions (see text for explanation on the vari-ous data classes). The color version of this figure is available only inthe electronic edition.

Figure 12. Ground-motion prediction equations for PGA andPSA for five periods (0.025, 0.2, 1, 2, and 4 s) plotted together withobserved in-slab data for depths h ≥ 100 km, after reduction toM 5.5 and rock-site conditions. Black and gray colors denoteback-arc and along-arc data, respectively. Data used for predictionvalidation are also shown with squares (event number 20 in Table 1)and diamonds (event number 21).


Interface Events

Deep interface events, limited to hypocentral depthsranging between 45 km and 60 km, were considered as asingle-depth bin following the same approach as previouslydescribed for in-slab events. As a first step, we adopted thesame original back-arc and along-arc categorization on thebasis of the position of the recording station relative to theepicenter. In Figure 16, data-running averages from interfaceevents are plotted against the hypocentral distance for PGAand 0.025 s, 0.2 s, and 1 s periods (regressions are limitedto 1 s due to the lack of data from acceleration-sensor instru-ments). The observed pattern of back-arc/along-arc biasvaries with period, with higher frequencies showing a gradualdevelopment of this bias, especially after the hypocentral dis-tance of 300–350 km, while 1 s data exhibits a rather constantdifference for the complete common back-arc/along-arc dis-tance overlap (from Rhyp ∼ 200 km). In order to overcomethese discrepancies and after several trials, we employed thefunctional form that was used for the regressions by Skarla-toudis et al. (2009), and written specifically as

logY � c1 � c2�M − 5:5� � c3 logR

� c41�1 − ARC��R − Rref� � c42ARC�R − Rref�� c51S� c52SS� ε; (4)

where symbols and notations are the same as in equation (3).Coefficients c41 and c42 are used to describe the additional

Figure 13. Ground-motion prediction equations for PGA,PGV, and 1 and 4 s PSA, reduced to M 6.7 and rock-site conditions,for h � 66 km. Thick solid black (reference), dotted black (back-arc), and dotted gray (along-arc) curves show present study results.Additional predictions are also shown for comparison fromAtkinson and Boore (2003; AB03, gray open-squared curves),Boore et al. (2009; Bea09, thin black and gray solid curves), Zhaoet al. (2006; black dashed curves), and Kanno et al. (2006; graydashed curves).

Figure 14. Ground-motion prediction equations for PGA, PGV,and 1 and 4 s PSA, reduced to M 6.7 and rock-site conditions forh � 120 km. Present study results are shown with the thick black(back-arc) and gray (along-arc) curves. Additional prediction rela-tions by Atkinson and Boore (2003; AB03-dashed gray curves),Zhao et al. (2006; black squared curves) and Kanno et al.(2006; black dotted curves) are also presented.

Figure 15. Predicted spectra forM � 6:7 with (a) h1 � 66 kmand (b) h2 � 120 km for NEHRP C (soil) site conditions and in-slabevents. The spectra are calculated for two hypocentral distances,one at zero epicentral distance (above each hypocenter, left panels)and one at Rhyp � 250 km (right panel). The corresponding predic-tions for the 8 January Kythera intermediate-depth earthquake(Boore et al., 2009; Bea09, Atkinson and Boore, 2003; AB03,and Zhao et al., 2006; Zea06) are also shown for comparison.


back-arc and along-arc anelastic attenuation, while forcoefficient c3 the fixed value of −1:7 was adopted as earlierdescribed. Regressions were performed using the Abraham-son and Youngs (1992) algorithm and the correspondingcoefficients are presented in Table 3.

The distribution of event terms with hypocentral depthand magnitude, shown in Figure 7 for PGA and PSA 1 s, doesnot exhibit any significant trends, although the total number ofthe events is rather small to perform a complete residual analy-sis. Similar conclusions can be drawn for the intraevent-termsdistribution with hypocentral distance, presented in Figure 8,where no apparent trend can be observed. The overall trend ofthe total misfits for the interface events, presented in Figure 9,is following the corresponding one for in-slab events, exhib-iting lower values for longer periods.

The most interesting feature derived from the regressioncoefficients for interface events is the comparison of theanelastic-attenuation factors with respect to the generalanelastic-attenuation coefficient (c32) of the in-slab ones. Ascan be seen in Figure 10, coefficients c41 and c42 are practi-cally independent of period exhibiting, a constant differencebetween back-arc and along-arc data and quite similar valuesfor the coefficient c32 of in-slab events for the same periodrange. These results show a possible different wave-propaga-tion mechanism for interface events in comparison to in-slabevents, for which seismic waves are not significantly affectedby the presence of the mantle-wedge LVL but are mostly con-trolled by the attenuation properties of the different crustalformations between the Mediterranean and the Aegean crust.

Discussion

One of the main challenges of this study was to integratethe conceptual geotectonic and wave-propagation model pre-sented in Figure 5, which is based on existing geophysicalstudies, with a functional form that could yield reasonableestimates in the regression analysis. Therefore, it is interest-ing to consider the behavior of the additional coefficients,c32, c4, and c5, introduced in the regression, with respectto the implications of the proposed geophysical model. Thereference anelastic-attenuation coefficient, c32, which de-scribes the general anelastic attenuation of seismic waves inthe Aegean lithosphere, exhibits a typical behavior, with rel-atively small (absolute) values decreasing rapidly with period(increasing with frequency) up to ∼1 s and showing asmaller decrease for lower periods (0.01–1 s). On the otherhand, coefficients c41 and c42, which describe the additionalanelastic attenuation imposed on seismic waves crossing themantle-wedge LVL (also low-Q area) beneath the southernAegean volcanic arc, show a very compatible pattern withthe model of Figure 5. In particular, coefficient c41 for deeperevents (h ≥ 100 km) has larger absolute values than c42 (forshallower events, h < 100 km), as expected from Figure 5,because seismic waves from deep (h ≥ 100 km) in-slabearthquakes travel larger distances within the LVL (low-Q)

Figure 16. Data-running averages for PGA and PSA data for theperiods 0.025, 0.2, and 1 s, reduced to M 5.5 and rock-site condi-tions, plotted against Rhyp for interface events. Back-arc and along-arc data are denoted with the black and gray curves, respectively.

Table 3Regression Coefficients for Interface Events

Period (s) c1 c2 c41 c42 c51 c52 σ τ ε

PGA 3.945 0.974 −0.00172 −0.00099 0.189 0.707 0.330 0.257 0.418PGV 2.783 1.186 −0.00122 −0.00064 0.232 0.428 0.261 0.095 0.2770.01 3.950 0.972 −0.00172 −0.00099 0.187 0.708 0.331 0.261 0.4210.025 3.842 0.951 −0.00169 −0.00096 0.193 0.792 0.326 0.261 0.4180.05 4.005 0.938 −0.00167 −0.00100 0.167 0.694 0.347 0.288 0.4510.1 4.112 0.910 −0.00163 −0.00091 0.163 0.731 0.377 0.364 0.5240.2 4.296 0.907 −0.00174 −0.00099 0.182 0.725 0.354 0.299 0.4630.4 4.244 0.985 −0.00177 −0.00089 0.251 0.736 0.338 0.149 0.3691.0 3.900 1.171 −0.00162 −0.00094 0.329 0.521 0.259 0.110 0.282

The equation is logY � c1 � c2�M − 5:5� � c3 logR� c41�1 − ARC��R − Rref��c42ARC�R − Rref� � c51S� c52SS� ε. See equation (4) for the variable definitions. Coefficients c3 andRref have fixed values of −1:70 and 1 km, respectively.


wedge than the shallower (60 ≤ h < 100 km) ones. The factthat we employed a constant maximum anelastic-attenuationterm is in agreement with the fixed size of the mantle wedgelow-Q “pocket.” Moreover, the estimated attenuation has arather constant value for higher frequencies (5–100 Hz) andsignificantly diminishes for longer periods (0.2–4 s), as ex-pected for anelastic-attenuation effects.

An additional feature is the derived slab-amplificationeffect, showing a pattern expected from the geophysicalmodel of Figure 5, with coefficient c51 (deeper eventsh ≥ 100 km) exhibiting in general higher values than c52(shallower events, h < 100 km). Because coefficient c5 cor-responds to the “additional” amplification of the seismicwaves that travel within the colder, high-Q, subducting slab,it is quite reasonable to assume that seismic waves fromdeeper in-slab events are more prone to channeling/focusingeffects compared to shallower in-slab events. The results inFigure 10 show that this amplification effect starts to dimin-ish at the frequency of 1–2 Hz, becoming very small at theperiod of 4 s, as longer wavelengths do not “see” the sub-ducting slab, and are therefore gradually less affected by itspresence. Considering typical mantle velocities, this patternchange corresponds to wavelengths of 2–5 km, suggestingthat this slab channeling occurs along a relatively narrowwaveguide, possible a subduction channel (Essen et al.,2009) and/or the subducted ocean-type crust of the easternMediterranean lithosphere.

The applicability of the proposed prediction relationscan be confirmed by the use of observations from two eventsthat occurred more recently in the Hellenic subduction zoneand were not used in the regressions (events numbers 20 and21 in Table 1 and Fig. 2). The data from these events werealso reduced to magnitude M 5.5 and rock-site conditions(Fig. 12; event number 20, squares; event number 21, dia-monds). The observed values are in good agreement with theprediction equations, for both back-arc and along-arc dataand within the limits of the data dispersion of the originalregressions.

The south Aegean subduction zone exhibits several par-ticular characteristics in comparison to other subductionareas worldwide. An example of this particularity is the verylarge difference observed in the predicted response-spectralevels between back-arc and along-arc data. This differenceof ground motions in along-arc areas by a factor of almost∼10 with respect to back-arc areas, may be large and notusual for other subduction zones worldwide, however it isfully justified by past damage observations for large inter-mediate-depth earthquakes in the Hellenic arc.

Conclusions

The large number of high-quality records employed inthis work allowed the derivation of an updated attenuationmodel incorporating variables based on event hypocentraldepth, as well as on the influence of the low-velocity/low-Q mantle wedge and the high-Q slab on seismic waveforms

for the southern Aegean subduction zone. The need to definean updated model emerged from the observation that fordeeper earthquakes the influence of the mantle LVL/low-Q wedge and subducting slab on seismic waves is moreprominent, suggesting that appropriate variables accountingfor the “additional” attenuation or amplification needed tobe incorporated in the ground-motion prediction model.After data classification into three hypocentral-depth bins(60 ≤ h < 80 km, 80 ≤ h < 100 km, and h ≥ 100 km) forin-slab events, the dependence on hypocentral distanceshowed that a critical hypocentral-distance range could beidentified for the two shallower depth bins (Rhyp ∼ 205–355 km and 140–240 km, respectively) within which aback-arc/along-arc difference develops and beyond whichobtains a maximum, constant value. For deeper earthquakes(h > 100 km), the back-arc/along-arc difference was ob-served for the whole distance range studied hence no criticaldistance range was introduced. Moreover, a new categoriza-tion of back-arc and along-arc stations was introduced, basedon the relative position of the recording station with respectto the in-slab earthquake location, its hypocentral depth, andthe critical distance range identified for the two shallowerdepth bins (i.e., only data for hypocentral distances longerthan the critical distance are characterized as back- or along-arc, while shorter hypocentral distances data are character-ized as “reference”). Comparisons with other GMPEs derivedfrom worldwide or regional data are in good agreement,given the particular characteristics and properties of theSouth Aegean subduction zone previously discussed.

Results presented in this work also allowed constraint ofthe different properties of deeper intermediate-depth (in-slab)and shallower interface events that occur in the SouthAegean subduction zone. As shown in Figure 17, the largerobserved differences occur at short distances for PGA, whilefor longer distances there is a general agreement for bothback-arc and along-arc PGA data. On the other hand, PGVpredictions for interface events are higher compared withthe ones from intermediate-depth events, both for back-arcand along-arc data, mainly at larger distances. Consideringalso the comparisons of recorded ground motions of inter-mediate-depth with shallow-crustal events (already discussedby Boore et al., 2009), the individuality of intermediate-depthevents is verified and the need for considering the contributionto earthquake hazard from the different earthquake types andareas (back-arc/along-arc) for the southern Aegean subduc-tion system becomes evident.

Data and Resources

Velocity-sensor data used in this study were collectedfrom permanent Greek seismological networks, operated bythe National Observatory of Athens (NOA) and the AristotleUniversity of Thessaloniki and are available to the publicupon request. The main body of broadband velocity-sensorrecordings for this study came from the EGELADOS,CYGNET, and SIMBAAD temporary networks, which are


not yet publicly released. Additional broadband velocity-sensor recordings were used from GEOFON and are avail-able from the corresponding online database http://www.webdc.eu/arclink/query?sesskey=f60cb2a3 (last accessedAugust 2012). The website for broadband velocity sensorand accelerometer in Patras, Greece, is http://seis12.karlov.mff.cuni.cz/greece/ (last accessed August 2012). Accelera-tion-sensor data used in this study were collected using the

acceleration-sensor networks operated by the Institute ofEngineering Seismology and Earthquake Engineering(ITSAK), the NOA, the Public Power Corporation, and theAstronomical Observatory of Larissa. Data from ITSAK andNOA networks are available upon request, while data fromthe other two acceleration-sensor networks are not publiclyreleased.

Acknowledgments

We would like to thank David M. Boore for his fruitful comments andvaluable suggestions on the original manuscript. We also thank AssociateEditor I. G. Wong and two anonymous reviewers for their thorough reviewsand important comments, which significantly improved the manuscript. Weare also grateful to GEOFON, the Greek Public Power Corporation, and theAstronomical Observatory of Larissa for their data contribution. This workhas been partly supported by the 3D-SEGMENTS project number 1337 ofthe ARISTEIA-I call funded by EC European Social Fund and the GreekSecretariat of Research and Technology.

References

Abrahamson, N. A., and R. R. Youngs (1992). A stable algorithm for re-gression analyses using the random effects model, Bull. Seismol.Soc. Am. 82, 505–510.

Abrahamson, N., G. Atkinson, D. Boore, Y. Bozorgnia, K. Campbell, B.Chiou, I. M. Idriss, W. Silva, and R. Youngs (2008). Comparisonsof the NGA ground-motion relations, Earthq. Spectra. 24, 45–66.

Atkinson, G. M., and D. M. Boore (2003). Empirical ground-motion rela-tions for subduction zone earthquakes and their application to Cascadiaand other regions, Bull. Seismol. Soc. Am. 93, 1703–1729.

Benetatos, C., A. A. Kiratzi, C. B. Papazachos, and G. F. Karakaisis (2004).Focal mechanisms of shallow- and intermediate-depth earthquakesalong the Hellenic Arc, J. Geodyn. 37, 253–296.

Bohnhoff, M., M. Rische, T. Meier, D. Becker, G. Stavrakakis, and H.-P.Harjes (2006). Microseismic activity in the Hellenic volcanic arc,Greece, with emphasis on the seismotectonic setting of the Santor-ini–Amorgos zone, Tectonophysics 423, 17–33.

Bohnhoff, M., M. Rische, T. Meier, B. Endrun, H.-P. Harjes, and G.Stavrakakis (2004). A temporary seismic network on the Cyclades(Aegean Sea, Greece), Seismol. Res. Lett. 75, no. 3, 352–357.

Boore, D. M. (2010). Orientation-independent, nongeometric-mean mea-sures of seismic intensity from two horizontal components of motion,Bull. Seismol. Soc. Am. 100, 1830–1835.

Boore, D. M., A. A. Skarlatoudis, B. N. Margaris, C. B. Papazachos, and Ch.Vendouzi (2009). Along-arc and back-arc attenuation and source spec-trum for the intermediate-depth 18 January, 2006, M 6.7 Kythera,Greece, Earthquake, Bull. Seismol. Soc. Am. 99, 2410–2434.

Cauzzi, C., and E. Faccioli (2008). Broadband (0.05 to 20 s) prediction ofdisplacement response spectra based on worldwide digital records, J.Seismol. 12, 453–475, doi: 10.1007/s10950-008-9098-y.

Essen, K., M. Braatz, L. Ceranna, W. Friederich, and T. Meier (2009).Numerical modeling of seismic-wave propagation along the platecontact of the Hellenic Subduction Zone—the influence of a deepsubduction channel, Geophys. J. Int. 179, 1737–1756, doi: 10.1111/j.1365-246X.2009.04369.x.

Hatzfeld, D., G. Pedotti, P. Hatzidimitriou, D. Panagiotopoulos, M. Scordilis,J. Drakopoulos, K. Makropoulos, N. Delibasis, J. Latoussakis, J.Baskoutas, and M. Frogneux (1988). The Hellenic subduction beneaththe Peloponesse: First results of a microearthquake study, Earth Planet.Sci. Lett. 93, 283–291.

Kanno, T., A. Narita, N. Morikawa, H. Fujiwara, and Y. Fukushima (2006).A new attenuation relation for strong ground motion in Japan based onrecorded data, Bull. Seismol. Soc. Am. 96, 879–897.

Figure 17. Ground-motion prediction equations for PGA andPGV, reduced toM 6.7 and rock-site conditions, for interface eventsand for h � 60 km. Present study predictions are presented with thedotted black and gray curves for back-arc and along-arc data, re-spectively. The thick dark gray (reference), thin black (back-arc),and thin gray (along-arc) curves depict predictions for in-slab events(60 ≤ h < 80 km), while the dashed gray curve corresponds to thepredictions by Atkinson and Boore (2003; AB03). The Skarlatoudiset al. (2003; 2007; Sea03) relations for shallow crustal earthquakesin the Aegean area are also presented for comparison.


http://www.webdc.eu/arclink/query?sesskey=f60cb2a3



http://seis12.karlov.mff.cuni.cz/greece/

http://seis12.karlov.mff.cuni.cz/greece/

http://dx.doi.org/10.1007/s10950-008-9098-y

http://dx.doi.org/10.1111/j.1365-246X.2009.04369.x

http://dx.doi.org/10.1111/j.1365-246X.2009.04369.x

Karagianni, E. E., C. B. Papazachos, D. G. Panagiotopoulos, P. Suhadolc, A.Vuan, and G. F. Panza (2005). Shear velocity structure in the Aegeanarea obtained by inversion of Rayleigh waves, Geophys. J. Int. 160,127–143.

Kiratzi, A. A., and C. B. Papazachos (1995). Active deformation of the shal-low part of the subducting lithospheric slab in the southern Aegean, J.Geodyn. 19, 65–78.

Konstantinou, K. I., and N. S. Melis (2008). High-frequency shear-wavepropagation across the Hellenic subduction zone, Bull. Seismol.Soc. Am. 98, 797–803.

LePichon, X., and J. Angelier (1979). The Hellenic arc and trench system:A key to the neotectonic evolution of the eastern Mediterranean area,Tectonophysics 60, 1–42.

McKenzie, D. (1972). Active tectonics of the Mediterranean region,Geophys. J. Roy. Astron. Soc. 30, 109–185.

NEHRP (1994). Recommended provisions for seismic regulations for newbuildings and other structures, Part 1: Provisions, FEMA 222ABuilding Seismic Safety Council, Washington D.C., 290 pp.

Okal, E., and J. Talandier (1997). Twaves from the great 1994 Bolivian deepearthquake in relation to channeling of S-wave energy up the slab, J.Geophys. Res. 102, 27,421–27,437.

Papazachos, B. C. (1990). Seismicity of the Aegean and surrounding area,Tectonophysics 178, 287–308.

Papazachos, B. C., and P. E. Comninakis (1969). Geophysical features ofthe Greek island arc and eastern Mediterranean ridge, Seance de laConference reunie a Madrid, 1–12 September 1969, Comptes Rendus16, 74–75.

Papazachos, B. C., and P. E. Comninakis (1971). Geophysical and tectonicfeatures of the Aegean arc, J. Geophys. Res. 76, 8517–8533.

Papazachos, B. C., and N. D. Delibasis (1969). Tectonic stress field andseismic faulting in the area of Greece, Tectonophysics 7, 231–255.

Papazachos, C. B., and G. Nolet (1997). P and S deep velocity structure ofthe Hellenic area obtained by robust non-linear inversion of traveltimes, J. Geophys. Res. 102, 8349–8367.

Papazachos, B. C., S. T. Dimitriadis, D. G. Panagiotopoulos, C. B. Papazachos,and E. E. Papadimitriou (2005). Deep structure and active tectonics of theSouthern Aegean volcanic arc, in Developments in Volcanology: TheSouth Aegean Volcanic Arc, Elsevier, 47–64.

Papazachos, C. B., P. M. Hatzidimitriou, D. Panagiotopoulos, and G. N.Tsokas (1995). Tomography of the crust and upper mantle in SEEurope, J. Geophys. Res. 12, 405–422.

Papazachos, B. C., B. G. Karakostas, C. B. Papazachos, and E. M. Scordilis(2000). The geometry of the Wadati–Benioff zone and lithospherickinematics in the Hellenic arc, Tectonophysics 319, 275–300.

Paul, A., D. Hatzfeld, H. Karabulut, P. Hatzidimitriou, D. M. Childs, S.Nikolova, C. Pequegnat, F. Hubans, A. Schmid, M. Aktar, A. K.Mutlu, T. Afacan, Y. Ozakin, D. Samut, C. Papazachos, I. Karagianni,D. Kementzetzidou, E. Karagianni, Z. Roumelioti, D. Vamvakaris, M.Scordilis, and H. Lyon-Caen (2008). The Simbaad experiment inW-Turkey and Greece: A dense seismic network to study the crustaland mantle structures, AGU Fall Meeting, San Francisco, 15–19 De-cember 2008.

Scordilis, E. M. (2006). Empirical global relations convertingMS andMb tomoment magnitude, J. Seismol. 10, 225–236.

Seed, B., R. Murarka, J. Lysmer, and I. M. Idriss (1976). Relationships ofmaximum acceleration, maximum velocity, distance from source, andlocal site conditions for moderately strong earthquakes, Bull. Seismol.Soc. Am. 66, 1323–1342.

Skarlatoudis, A. A., C. B. Papazachos, B. N. Margaris, Ch. Papaioannou,Ch. Ventouzi, D. Vamvakaris, A. Bruestle, T. Meier, W. Friederich,G. Stavrakakis, T. Taymaz, R. Kind, A. Vafidis, T. Dahm, and theEGELADOS group (2009). Combination of strong-and weak-motionrecordings for attenuation studies: The case of the January 8, 2006Kythera intermediate-depth earthquake, Bull. Seismol. Soc. Am. 99,694–704.

Skarlatoudis, A. A., C. B. Papazachos, B. N. Margaris, N. Theodoulidis, C.H. Papaioannou, I. Kalogeras, E. M. Scordilis, and V. G. Karakostas(2003). Empirical peak ground motion predictive relations for shallowearthquakes in Greece, Bull. Seismol. Soc. Am. 93, 2591–2603.

Skarlatoudis, A. A., C. B. Papazachos, B. N. Margaris, N. Theodoulidis,C. H. Papaioannou, I. Kalogeras, E. M. Scordilis, and V. G. Karakostas(2007). Erratum to Empirical peak groundmotion predictive relations forshallow earthquakes in Greece, Bull. Seismol. Soc. Am. 97, 2219–2221.

Spakman, W. (1988). Upper-mantle delay time tomography with an appli-cation to the collision zone of the Eurasian, African, and Arabianplates, Ph.D. Thesis, Univ. of Utrecht, The Netherlands, Vol. 53,200 pp.

Spakman, W., S. Van der Lee, and R. D. Van der Hilst (1993). Travel-timetomography of the European–Mediterranean mantle down to 1400 Km,Phys. Earth Planet. Int. 79, 3–74.

Taymaz, T., J. Jackson, and R. Westaway (1990). Earthquake mechanisms inthe Hellenic trench near Crete, Geophys. J. Int. 102, 695–731.

Uniform Building Code (UBC) (1997). Itern. Conf. Building Officials,USA, Vol. II, 489 pp.

Zhao, J. X., J. Zhang, A. Asano, Y. Ohno, T. Oouchi, T. Takahashi, H.Ogawa, K. Irikura, H. K. Thio, P. G. Somerville, Y. Fukushima,and Y. Fukushima (2006). Attenuation relations of strong groundmotion in Japan using site classification based on predominant period,Bull. Seismol. Soc. Am. 96, 898–913.

Geophysical LaboratoryAristotle University of ThessalonikiPO Box 352-1, GR-54124Thessaloniki, [email protected]@geo.auth.gr

(A.A.S., C.B.P.)

Institute of Engineering Seismology and Earthquake Engineering (ITSAK)P.O. Box 53 GR 551 02 FinikasThessaloniki, [email protected]

(B.N.M.)

Geodynamic Institute of the National Observatory of Athens (GEIN-NOA)P.O. Box 20048118 10 Athens, [email protected]@gein.noa.gr

(C.V., I.K., .)

Manuscript received 24 August 2012


ground-motion prediction equations of intermediate...

Documents