ORIGINAL ARTICLE

Attenuation tomography of the Southern Apennines (Italy)

F. Zolezzi & P. Morasca & K. Mayeda &

W. S. Phillips & C. Eva

Received: 18 April 2007 /Accepted: 26 November 2007# Springer Science + Business Media B.V. 2007

Abstract The aim of this study is to improve ourknowledge of the attenuation structure in the SouthernApennines using a new amplitude ratio tomographymethod (Phillips et al., Geophys Res Lett 32(21):L21301, 2005) applied on both direct and codaenvelope measurements derived from 150 eventsrecorded by 47 stations of the Istituto Nazionale diGeofisica e Vulcanologia National Seismic Network(Rete Sismica Nazionale Centralizzata). The two-dimensional (2-D) analysis allows us to take intoaccount lateral crustal variations and heterogeneitiesof this region. Using the same event and stationdistribution, we also applied a simple 1-D methodol-ogy, and the performance of the 1-D and 2-D pathassumptions is tested by comparing the averageinterstation variance for the path-corrected amplitudesusing coda and direct waves. In general, codameasurement results are more stable than using directwaves when the same methodology is applied. Using

the 2-D approach, we observe more stable results forboth waves. However, the improvement is quitesmall, probably because the crustal heterogeneity isweak. This means that, for this region, the 1-D pathassumption is a good approximation of the attenuationcharacteristics of the region. A comparison between Qtomography images obtained using direct and codaamplitudes shows similar results, consistent with thegeology of the region. In fact, we observe low Qalong the Apennine chain toward the Tyrrhenian Seaand higher values to the east, in correspondence withthe Gargano zone that is related to the ApuliaCarbonate Platform. Finally, we compared our resultswith the coda Q values proposed by Bianco et al.(Geophys J Int 150:10–22, 2002) for the same region.The good agreement validates our results as theauthors used a completely independent methodology.

Keywords Coda waves . Attenuation Apennines .

Q tomography . Southern Apennines

1 Introduction

On October 31st, 2002, an earthquake with magnitudeMw=5.7 struck the town of San Giuliano di Puglia(southern Italy), resulting in the collapse of a schooland the death of many children. The event occurredalong the westward prolongation of the Mattinatafault, interpreted as reactivation of pre-existing E–Wdiscontinuities situated within the Apulia foreland in

J SeismolDOI 10.1007/s10950-007-9079-6

F. Zolezzi : P. Morasca (*) :C. EvaUniversity of Genova (Dip.Te.Ris.),Genoa, Italye-mail: alpocc@dipteris.unige.it

K. MayedaWeston Geophysical Corporation,Lexington, MA, USA

W. S. PhillipsLos Alamos National Laboratory,Los Alamos, NM, USA

response to the NW–SE Africa–Europe convergence(Di Bucci and Mazzoli 2003; Valensise et al. 2004).

After this tragic and unexpected event, the neces-sity of re-examining the characteristics of this regionbecame a priority because, previously, the area wasclassified as only moderately hazardous according tothe seismic code (Decanini et al. 2004). In order toaccurately predict ground motion for future events inthis zone, a study of attenuation features and theirspatial variations in the Southern Apennines region isnecessary.

One-dimensional (1-D) attenuation parameterswere studied in this region by Bianco et al. (2002)who obtained separate estimates of the intrinsic andscattering Q that were compared with coda Q, (Qc).They found a predominant intrinsic attenuation inSouthern Apennines indicating a weakly heteroge-neous crust for this area. However, this zone isbordered by heavy heterogeneous volcanic areas suchas Vesuvio and Campi Flegrei that may have aninfluence on the average attenuation values. To takeinto account the possible effects due to heterogene-ities and lateral variations, well-established tomo-graphic methods have been used for many years (e.g.,Young and Ward 1980; Singh and Herrmann 1983;Mitchell et al. 1997). More recently, Phillips et al.(2005) have developed a tomography approach toinvert for Q along the path in central Asia using thecoda envelope amplitude that they idealized as if itwere a direct wave, which is valid for the shortsegment of the early coda that they used (Mayedaet al. 2005). This method was successfully appliedto northern California by Mayeda et al. (2005) whotested the performance of this approach comparing2-D distance-corrected coda and direct waveamplitudes with results obtained using more simple1-D path corrections (Mayeda et al. 2003) for thesame stations and data. The improvement observedfor this region leads us to test the calibration of the2-D path corrections for the Southern Apennines,where a detailed knowledge of attenuation mecha-nisms is important for seismic risk analysis. More-over, because attenuation varies with the physicalproperties of the media through which the wavestravel (presence of water, tectonic age, heat flow,etc.; Mitchell et al. 1997), differences in wavepropagation can be linked with the geologic andtectonic settings of the areas and play a key role inhazard studies.

Analyzing a data set of 150 earthquakes recordedin the Southern Apennines, we apply the newtomographic approach proposed by Phillips et al.(2005) to estimate 2-D attenuation for regional phases(S-waves) and coda waves. The attenuation tomogra-phy methodology of Phillips et al. (2005) is anextension of well-known amplitude ratio techniques(Chun et al. 1987; Shih et al. 1994; Fan and Lay2003) that allow for removing source terms from theformulation to obtain the spatial variation of Q for thestudy area. The method assumes that the codaamplitudes could be modeled in the same way as thedirect waves and that the source radiates isotropically.This assumption is certainly valid for coda waves dueto its averaging nature. In fact, coda waves are thescattered waves that follow the direct arrivals, thismeans that “coda” is homogeneously distributed inthe crust (e.g., Mayeda et al. 2007). For this reason,they are nearly insensitive to the radiation pattern anddirectivity of the earthquake source. As for directwaves, the source directivity effect is negligible for ourdataset because for each event we have many recordingstations, and this allows a sufficient sampling of thefocal sphere to average over the radiation pattern.

We are interested in understanding which struc-tures mainly contribute to the average attenuationobserved in this region and in quantifying theirinfluence in order to obtain a more detailed propagationcharacterization, which is fundamental for an accurateseismic hazard analysis. The approach we use permitsthis, and moreover the 2-D inversions of early codawaves, as well as the direct S-waves, have the addedvalue of permitting stable estimates of the source spectraafter the application of the 2-D path corrections.

2 Tectonic framework and data

The Southern Apennines is an east-verging Neogeneaccretionary wedge resulting from the westward sub-duction of the Apulian–Ionian lithosphere (Doglioniet al. 1996). The accretionary wedge consists of a pileof nappes (east-verging Meso-Cenozoic limestones andbasinal units) over the flexured south–western marginof the Apulia foreland (Patacca and Scandone 1989).The present configuration of the Southern Apenninesthrust belt (see Fig. 1) is the result of alternations inspace and time of carbonate platforms and basins thattogether with the discontinuous nature of the paleo-

J Seismol

geographic domains make the modeling of geologicalprofiles of this area a difficult task (Menardi Nogueraand Rea 2000).

Recent geophysical studies using active seismicdata show strong lateral velocity variations at 3–4-kmdepth. The low velocity in the upper 4 km of the crustare thought to be the result of Cenozoic flyshoidcovers and basinal successions that overlap the Meso-Cenozoic carbonate multilayer of the Apulia platform(Improta et al. 2000). During the middle Pleistocene,the Southern Apenninic wedge has been uplifted byextensional tectonics along NE–SW direction cross-cutting the contractional structures. This stress regimeis responsible for the historical and present-dayseismic activity (Anderson and Jackson 1987). Theseismicity distribution shows the existence of shallowearthquakes, with depths less than 30 km andlocations that approximately follow the axis of thechain, with extensional or transtensive focal planesolutions. This normal faulting stress regime observedalong the Apenninic chain seems to be replaced by astrike–slip regime in the foreland (Frepoli and Amato2000). In the southern edge of the studied area, adeeper seismicity is observed that is related to thewest-dipping subduction of the Apulian plate litho-

sphere (Menardi Noguera and Rea 2000). An expres-sion of this subduction of the lithospheric slab isrepresented by the positive gravity anomaly of theMurge bulge in contrast to the negative anomaly ofthe Bradano foredeep (Amato and Selvaggi 1991;Royden 1993). Magnetic data show high anomalies ofthe magnetic basement in the axial part of the Apenninicchain (Mostardini and Merlini 1986). Studies on thedeep structure of the Southern Apennines define aMoho that runs parallel to the bottom of the Apuliaplatform and reaches a depth of about 30–35 km closeto the Campanian coast (Adriatic Moho) whileupwards to 20–25-km depth in the Tyrrhenian offshore(Menardi Noguera and Rea 2000).

We based the current study on regional, deconvolved,and vertical component seismograms recorded between1985 and 2003 at 47 stations of the National SeismicNetwork (Rete Sismica Nazionale Centralizzata—Istituto Nazionale di Geofisica e Vulcanologia) that aremostly equipped with 1-Hz S13 Geotech sensors. Weused roughly 150 earthquakes that occurred within50-km depth in the investigated area, which after acareful manual picking of each seismogram werelocated. Data were selected by requiring goodsignal-to-noise ratio and magnitudes larger than 3.5.

Fig. 1 Simplified geologi-cal sketch of the ItalianApennines by Fracassi andValensise (2007)

J Seismol

Figure 2 shows the distribution of events andstations used in this study.

3 Materials and methods

In order to study the attenuation effects in theSouthern Apennines, we performed 2-D inversionsfor both coda and direct S-waves and the results arecompared to the ones obtained with the application ofthe 1-D method outlined by Mayeda et al. (2003).

In order to estimate 2-D path Q in the region, weuse a tomography technique developed by Phillipset al. (2005). Here, we give a brief description of theprocedure and direct the reader to Phillips et al.(2005) for more details. We want to invert separatelyfor coda and direct amplitudes that are measured fornarrow frequency bands in a range between 0.7 and10 Hz.

Direct amplitudes are defined as the maximumamplitude following the onset of the S-wave arrival.Coda envelope measurements are determined using

the procedure proposed by Mayeda et al. (2003) thatprovides synthetic envelopes after a careful calibra-tion of S-wave velocity and coda-shape parametersfor the study region. The coda measurement windowlength is determined by manual inspection because itdepends on the frequency band and event magnitude.As explained by Mayeda et al. (2003), for eachfrequency, it is possible to define a minimum windowlength beyond which the interstation standard deviationstabilizes. Choosing shorter windows would imply alarge scatter comparable to the direct waves. Thesynthetic coda envelope is scaled vertically to matchthe observed envelope and the amount of the shift isdefined as the nondimensional coda amplitude that weassume to have a distance dependence with the sameform as the direct waves (i.e., geometrical spreadingand Q). We want to highlight that for this study weused early coda measurements so that the path Q isclose to the direct ray path. In fact, it is well knownthat the late coda consists of multiple scattered energythat certainly sees different parts of the propagationmedium than the early coda and the direct waves.

Fig. 2 Map showing theearthquakes (black circles)and seismic stations (whitesquares) used in this study

J Seismol

Thanks to this choice, a ray tomographic approach isappropriate for amplitudes measured from coda waves.

All direct and coda amplitudes are then correctedfor the assumed 1/r geometrical spreading (Yang2002) before inverting for the laterally varying (2-D)Q. Considering an isotropic source radiation, weapply the following amplitude ratio formulation toboth kinds of waves, separately:

Aij� < Aij >j¼ Si� < Sj >j

þ Pkdxijkαk� < Pkdxijkαk >j

� �log10 eð Þ

ð1Þ

where A represents the log10 amplitude, i, j, k are site,source, and path indices respectively, S representslog10 site terms, dx are path lengths throughout adiscretized area, α is the discretized attenuationcoefficient (αk=ω /2Qkc) and the Pk is the ray pathsum. The < >j represent averages taken over all raypath or sites associated with event j. Equation 1 islinear and can be solved easily using sparse matrixmethod that allow us to estimate the average (2-D) Qvalue for every frequency band for both S and codawaves. In our inversion, we assumed cell dimensionof 0.5° by 0.5°. Smoothing constraints are applied tothe attenuation model to avoid artifacts due to thenoise, and the site term sum is assumed to be zero.

This approach allows elimination of source termsfrom the inversion because the problem is formulatedin terms of amplitude ratios (or difference in logspace) between at least two stations that recorded thesame earthquake. This means that we need to use dataonly from multiple recorded events as a single stationrecording will result in a null equation.

In addition to obtaining 2-D path corrections, wealso estimate the average 1-D attenuation for theregion following Mayeda et al. (2003), for both directand coda amplitudes. This allows us to make somecomparisons and verify the performance of the twomethodologies.

For coda measurements, 1-D path corrections aredefined empirically by Eq. 2 that simply looks at thedistance dependence (r) for each considered frequencyband (f):

P r; fð Þ ¼ 1þ r

P2

� �p1� ��1

ð2Þ

P is the source-normalized coda amplitude, r is theepicentral distance and p1 and p2 are the pathparameters.

Mayeda et al. (2003) observed that Eq. 2 is roughlyconstant up to a certain critical distance (p2) beyondwhich the coda is no longer homogeneously distrib-uted in the space and then falls off with increasingdistance proportionally to p1 that controls how rapidlythe coda envelope level decreases with the distance. Agrid-search technique is then applied to find the bestp1 and p2 parameters for each frequency band. Thesetwo parameters are those who minimize the amplitudeinterstation scatter for common events recorded atmultiple pairs of stations.

No direct correlation exists between Eq. 2 and codaQ because, while appearing physical, Eq. 2 is onlyused for empirical fitting and fit parameters shouldnot be overinterpreted.

Although Eq. 2 allows to define path parametersfor each possible station pair, as observed in otherstudies where this methodology was applied (Morascaet al. 2005; Eken et al. 2004), a single set of pa-rameters p1 and p2 can be used for the whole regionusing an average over all possible station pairs.

For 1-D direct wave amplitude corrections, we alsoapplied a grid-search technique, which minimized theinterstation scatter in order to solve for the attenuationand geometrical spreading using the formulation ofStreet et al. (1975):

r < rc; A r; fð Þ ¼ A0 � S fð Þ � 1r� e�

pfbQb

h i

ð3Þ

r � rc; A r; fð Þ ¼ A0 � S fð Þ � 1rc

� rcr

� �g�e�

pfbQb

h i

ð4Þ

where A0 is the source term, S is the site effect, rc isthe critical distance (fixed at 100 km), γ is thespreading coefficient (set to 0.5), β is the S-wavevelocity (set to 3.5 km/s), and Qβ is the S-wavequality factor. Through a grid-search technique thebest Qβ is defined as the value that minimized thatscatter between all station pairs.

Although the two methodologies (1-D and 2-D)are independent of each other, Eqs. 3 and 4 used forthe direct waves are the same as for the 2-D case

J Seismol

(i.e., Eq. 1), with the difference that in Eq. 1 there isthe addition of a sum over discretized path terms inplace of 1/Q. On the other way, no correlation existsbetween Eq. 2, used for coda waves, and Eqs. 3 and4 since Eq. 2 is an empirical formulation that simplylook at the distance dependence.

The step following the 1-D and 2-D path calibra-tion is to apply the corrections to direct and codaamplitudes and analyze the four cases: (a) 1-Dcorrected coda waves; (b) 1-D corrected direct waves;(c) 2-D corrected coda waves; and (d) 2-D correcteddirect waves.

For each case, amplitudes are only corrected forpath effects and still carry the S-to-coda transferfunction and the site effects that we need to removeto obtain a source spectrum for each event in each ofthe four cases. In this way, we obtain four sourcespectra for each event to be compared with eachother in order to see the differences due to the dif-ferent path corrections applied and to verify the bestcase.

Source spectra are computed applying the method-ology of Mayeda et al. (2003). The approach consistsin building a nondimensional, distance-correctedspectrum for each event and to transform this into amoment–rate spectrum by correcting for frequency-dependent S-to-coda transfer function and site effect.The nondimensional spectrum can be visualized as aplot of the previously obtained path-corrected ampli-tudes as a function of the frequency. It is called“nondimensional” because amplitudes are computedas a shift respect to starting reference amplitudes. Thenecessary corrections are obtained analyzing thenondimensional spectra of a set of events that we call“calibration events”. Their characteristic is that wehave independent information on their moments fromlong-period waveform modeling. Assuming that theS-wave source spectrum is flat below the cornerfrequency, we can flatten their spectra proportionallyto the given moments. In this way, we determine amoment–rate constant to add to all amplitudes foreach frequency. Because these corrections are inde-pendent of distance they can be uniformly applied toall distance-corrected amplitudes for the entire dataset. For high frequencies, we derive the moment–rateconstants by analyzing small events. In fact, eventhough such events are too small to be waveformmodeled, we can still estimate moments usingfrequencies less than a few Hertz, and, assuming that

they have flat source spectra to high frequency, we areable to derive the moment–rate constants for higherfrequencies.

3.1 2-D and 1-D attenuation results

In Fig. 3, we show the results obtained after applyingthe attenuation tomography results to both direct andcoda waves for two frequency bands (3.0–4.0 Hz and8.0–10.0 Hz). Although the event–station path distri-bution shows a slightly prevalent NW–SE trend dueto the narrow shape of Italy, this has an influenceespecially outside the resolved area indicated by theblue rectangle in Fig. 3. In fact, the station distribu-tion is quite good inside this area and allows for agood azimuthal coverage.

The resolution area (blue rectangle of Fig. 3) wasdefined through a synthetic test. We used a checker-board cell of 0.30° in latitude and longitude inter-spaced with a low and high input Q of 50 and 500,respectively. The area in which the checkerboard testis well reproduced is the region that we can considerwell resolved by the tomography inversion. Thesynthetic data values were computed for the sametravel paths used in the inversion of the real data;moreover, the same damping parameters were as-sumed in the inversion of both synthetic and realdata.

For both S-wave and coda wave analysis, we usedthe same event and station distribution. We foundgood agreement between the average Q derived in thetwo cases: we observed slightly smaller values for thedirect waves, more consistent at higher frequency,while the variation from the mean is quite similar.

Looking at the frequency dependence, the aver-age Q values increase from about 100 at 0.7–1.0 Hzto about 625 at 8.0–10.0 Hz. However, we did notnote any significant regional variations of Q, eventhough the methodology is capable of resolving suchvariation (Mayeda et al. 2005). This is probably dueto the weak small-scale heterogeneity of the crust ofthe investigated area, as pointed out from the deepseismic sounding profile data collected in this region(Menardi Noguera and Rea 2000). Another valida-tion of this thesis is the predominance of intrinsicdissipation over scattering attenuation observed byBianco et al. (2002). The attenuation results are inagreement with the geological setting of the region,exhibiting a difference between the Tyrrhenian and

J Seismol

the Adriatic side. The tomography maps for both Sand coda waves show a trend of the average Q(yellow in Fig. 3) that roughly follows the Apenninicthrust front (see Fig. 1) with a progressive eastward

migration of its limit at higher frequencies (Fig. 3).We observed low Q along the Apenninic chaintoward the Tyrrhenian sea (with a variation of up to20% with respect to the average Q) and higher

Table 1 Numerical values referred to in Fig. 3 (a, b, a’, b’)

Wave Frequency (Hz) Average Q (<Q>) <Q> −20% <Q> +50%

S-waves 3.0–4.0 283 226.4 424.5Coda waves 3.0–4.0 329 263.2 493.5S-waves 8.0–10.0 435 348 652.5Coda waves 8.0–10.0 625 500 937.5

The minimum Q values obtained from the inversions are 20% less than the average Q (<Q> −20%) and are representative of the redareas in Fig. 3; while the maximum Q values are a factor 0.5 higher than the average Q (<Q> +50%) and are observed in the Garganozone, the blue areas of Fig. 3

Fig. 3 Maps of Q from S-wave (a, a’) and coda wave (b, b’)inversions for the frequency band 3.0–4.0 and 8.0–10.0 Hz.The results are represented as difference in percentage from theaverage Q (yellow) for each frequency. Figures c, c’ show the

ray path of the events used in the tomographic inversion. Theresolved area is extending roughly from 13.5°E to 16.0°E andfrom 40.5°N to 42.0°N (blue box)

J Seismol

values to the east, in correspondence with theGargano zone (with a variation of up to 50% withrespect to the average Q). This low-attenuation areamay be correlated to the high-velocity anomaly ofthe Apulia carbonate platform (Di Stefano et al.1999; Chiarabba and Amato 1996; Mele et al. 1997;Castro et al. 2004), while the high heat flow of theTyrrhenian side may be the cause of the observed highattenuation along the Southern Apennines. In Table 1,we report the numerical values of the average Q andits minimum and maximum variations referred to theexamples of Fig. 3. In general, the Q variationsacross the region rarely vary by more than a factor of0.5. We compared our average coda-derived Q withthe coda Q measured in the same area by Bianco etal. (2002). As shown in Fig. 4, we obtained similarresults for the comparable frequency range. This is agood validation of our values because the method-ologies are completely independent of each other.

Finally, we used the well-established methodologyof Mayeda et al. (2003) to define 1-D path correctionsfor coda and direct waves. Results are reported inTable 2 that shows the best p1 and p2 values obtainedempirically using a grid-search technique on 27station pairs. These values are used in Eq. 2 thatrepresents the path correction to be applied to the rawcoda amplitudes. Likewise, for the direct waves, agrid-search on the same station pairs provided theS-wave Q values (Fig. 4) to be coupled with thegeometrical spreading defined in Eqs. 3 and 4.

3.2 Source spectra results

For the four examined cases, 1-D direct waves, 2-Ddirect waves, 1-D coda waves, and 2-D coda waves,the path-corrected amplitudes were used to computethe source spectra and to calibrate the moment magni-tude of the 150 events used in this work. Figure 5 showsan example of direct- and coda-derived source spectrafor the 1997, March 19th Mw=4.5 earthquake. Tocompute the spectra we applied a frequency-dependentcorrection to get a flat, low-frequency spectra belowthe estimated corner frequency as outlined in Mayedaet al. (2003). Relative site terms are introduced basedon common event measurements throughout the

Table 2 Path correction parameters used in the hyperbola Eq. 2for all stations for coda measurement

Frequency (Hz) p1 p2

0.7–1.0 1,700 20,0001.0–1.5 2,000 30,0001.5–2.0 1,900 30,0002.0–3.0 2,000 30,0003.0–4.0 2,000 35,0004.0–6.0 2,200 30,0006.0–8.0 2,700 50,0008.0–10.0 2,700 45,000

p2 is in kilometer because it is a critical distance as explained inthe text, while p1 is a constant that controls the amplitude decaybeyond the critical distance

Fig. 4 Comparison ofdifferent Q estimates for theSouthern Apennines. Blackcurves are the 2-D-calibratedcoda Q derived in this study,and the observed coda Qobtained for the same regionby Bianco et al. (2002). Greycurves are Q values obtainedin this study for directS-waves using the 1-Dmethod by Mayeda et al.(2003) and the averagevalues obtained using the2-D approach of Phillipset al. (2005)

J Seismol

network, which avoids interstation discrepancies thatresult from the lack of spatial and temporal distributionof calibration events. As shown in Fig. 5, the coda-derived source spectra show less scatter than thosecomputed from direct waves for both 1-D and 2-Dpath-correction cases.

Finally, we compare the performance between codaand S-wave source spectra estimates using both 1-Dand 2-D methodology. As observed in Fig. 6, asignificant improvement is obtained using codawaves, independent of the attenuation methodologyadopted. Note that due to the stability of coda waves(Sato and Felher 1998), the simple 1-D path correc-tion applied on codas result in lower standarddeviations than the 2-D direct wave results.

However, comparing the two attenuation method-ologies for the same waves, the improvement is quitesmall. One of the reasons for this could be due toweak heterogeneity of the crust in this sector of theSouthern Apennines (Menardi Noguera and Rea2000; Bianco et al. 2002). This means that the 1-Dmethod is good enough to describe the attenuationcharacteristics of the crustal volume of the SouthernApennines that we investigated. However, for codawaves, we note that the lack of significant improve-

ment between 1-D and 2-D corrections could be aresult of short measurement window lengths whichmight contribute to increased amplitude scatter.

4 Conclusion

This study provides detailed information on theattenuation of the Southern Apennines giving an

Fig. 6 Plot of the average interstation data standard deviation(y-axis) versus frequency. The top two curves represent theresults for S-waves, while the two bottom curves show resultsfor coda wave inversion. Note that coda waves significantlyreduce the SD values

Fig. 5 Example of S- andcoda-wave-derived sourcespectra for the 1997, March19th Mw=4.5 earthquake.Grey and black curves arestation and average spectrarespectively

J Seismol

important contribution to the hazard analysis of thearea since understanding the weak-motion attenuationin the region can help in better understanding strongground motion from future, damaging earthquakes(e.g., Malagnini et al. 2007).

We applied a new amplitude ratio tomographymethod (Phillips et al. 2005) to both coda and directwaves in the Southern Apennines to estimate thevarying quality factor (Q) for the region. The method-ology removes the source parameters from the inver-sion by using amplitudes ratios between two or morestations that recorded the same event.

For both direct and coda waves, the Q tomographyimages show similar results, consistent with the majortectonic features of the studied area. Relative to theaverage Q (yellow in Fig. 3) estimated for theresolved area (blue box in Fig. 3), we observed highattenuation (low Q) up to 20% of variation in theTyrrhenian side of the Southern Apennines associatedwith the high heat flow of this volcanic region, whileefficient propagation, with Q variations up to 50% ofthe average Q, is found in the Gargano zone incorrespondence of the high-velocity anomaly (DiStefano et al. 1999; Chiarabba and Amato 1996;Mele et al. 1997; Castro et al. 2004) of the Apuliacarbonate platform for both coda and direct wavesanalysis. Table 1 shows the numerical values of Qreferred to the examples reported in Fig. 3. In general,the Q variations across the region rarely vary by morethan a factor of 0.5.

The comparison with the coda Q measured for thesame region by Bianco et al. (2002) through theMultiple Lapse Time Window method, gives a goodvalidation of our study because the two methodolo-gies provided similar results. As expected, the bestagreement is found between their Qc and our 2-Dcoda Q because coda waves are less sensitive to localpath effects than direct waves. Slightly strongerfrequency dependence is observed for both 1-D and2-D direct wave results when compared to coda-derived values. The difference between the S-waveand the coda decay with frequency could be explainedaccording to the energy flux model by Frankel andWennerberg (1987) that suggests that scatteringattenuation at low frequencies becomes important.

Besides the 2-D attenuation tomography, we alsoapplied the 1-D path correction formulation proposedby Mayeda et al. (2003) on both direct and codameasurements to obtain four sets of path-corrected

amplitudes (1-D S-waves, 2-D S-waves, 1-D codawaves, and 2-D coda waves). For all four cases, wecomputed the source spectra and calibrated themoment magnitude for the 150 earthquakes used inthis study following the methodology described byMayeda et al. (2003).

We evaluated the performance of each approach bycomparing the data scatter of distance-correctedamplitudes using the same events and station pairs(27 station pairs). For the same waves (direct orcoda), the 2-D path correction slightly reduces theaverage interstation data standard deviation. Thisresult could be due to the weak heterogeneity of thecrust in the study area, as suggested by MenardiNoguera and Rea (2000) and Bianco et al. (2002) whopointed out how in this area Qc is close to intrinsic Q.On the other hand, a noticeable improvement isobserved for coda waves when compared to directwave results. Because coda waves average over anysource radiation pattern as well as any lateral crustalheterogeneity, even the simple 1-D approach appliedto coda amplitudes allow for more stable results thanthe 2-D inversion using direct waves. The bestperformance is obtained for coda amplitudes usingthe 2-D path correction, even though the 1-Dapproach seems to be adequate to describe theattenuation in the region.

Acknowledgements K. Mayeda was supported under WestonGeophysical subcontract No. GC19762NGD and AFRL con-tract No. FA8718-07-C-0010.

References

Amato A, Selvaggi G (1991) Terremoti crostali e sub-crostalinell’Appennino Settentrionale. Studi Geol Camerti VolSpeciale 1991(1):75–82

Anderson HJ, Jackson J (1987) Active tectonics of the Adriaticregion. J R Astron Soc 91:937–987

Bianco F, Del Pezzo E, Castellano M, Ibanez J, Di Luccio F(2002) Separation of intrinsic and scattering seismicattenuation in the Southern Apennine zone, Italy. GeophysJ Int 150:10–22

Castro RR, Gallipoli MR, Mucciarelli M (2004) An attenuationstudy in southern Italy using local and regional earth-quakes recorded by seismic network of Basilicata. AnnGeophys 47(5):1597–1608

Chiarabba C, Amato A (1996) Crustal velocity structure of theApennines (Italy) from P-waves travel time tomography.Ann Geophys 39(6):1133–1148

Chun K-Y, West GF, Kokoski RJ, Samson C (1987) A noveltechnique for measuring Lg attenuation: results from

J Seismol

eastern Canada between 1 to 10 Hz. Bull Seismol Soc Am77:398–419

Decanini L, Di Pasquale G, Galli P, Mollaioli F, Sanò T (2004)Seismic hazard and seismic zonation of the region affectedby the 2002 Molise, Italy earthquake. Earthquake Spectra20(S1):S131–S165

Di Bucci D, Mazzoli S (2003) The October–Novembre 2002Molise seismic sequenze (southern Italy): an expression ofAdria intraplate deformation. J Geol Soc London160:503–506

Di Stefano R, Chiarabba C, Lucente F, Amato A (1999) Crustaland uppermost mantle structure in Italy from the inversionof P-wave arrival times: geodynamic implication. GeophysJ Int 139:483–498

Doglioni C, Harabaglia P, Martinelli G, Monelli F, Zito G(1996) A geodynamic model of the Southern Apenninesaccretionary prism. TerraNova 8:540–547

Eken T, Mayeda K, Hofstetter A, Gök R, Orgülü G, Turkelli N(2004) An application of the coda methodology formoment–rate spectra using broadband stations in Turkey.Geophys Res Lett 31:L11609 DOI 10.1029/2004GL019627

Fan G, Lay T (2003) Strong Lg wave attenuation in thenorthern and eastern Tibetan plateau measured by a two-station/two-event stacking method. Geophys Res Lett 30(10):1530 DOI 10.1029/2002GL016211

Fracassi U, Valensise G (2007) Unveiling the sources of thecatastrophyc 1456 multiple earthquake: Hints to ununexplored tectonic mechanism in Southern Italy. BullSeis Soc Am 97(n.3):725–748

Frankel A, Wennerberg L (1987) Energy-flux model of seismiccoda: separation of scattering and intrinsic attenuation.Bull Seismol Soc Am 77:1223–1251

Frepoli A, Amato A (2000) Spatial variation in stresses inpeninsular Italy and Sicily from background seismicity.Tectonoph 317:109–124

Improta L, Ianaccone G, Capuano P, Zollo A, Scandone P(2000) Inferences on the upper crustal structure ofSouthern Apennines (Italy) from seismic refraction inves-tigations and subsurface data. Tectonoph 317:273–297

Malagnini L, Mayeda K, Uhrhammer R, Akinci A, HerrmannRB (2007) A regional ground motion excitation/attenua-tion model for the San Francisco region. Bull Seismol SocAm 97:843–862 DOI 10.1785/0120060101

Mayeda K, Hoffstetter A, O’Boyle J, Walter WR (2003) Stable andtransportable regional magnitudes based on coda-derivedmoment–rate spectra. Bull Seismol Soc Am 93:224–239

Mayeda K, Malagnini L, Phillips WS, Walter WR, Dreger D(2005) 2-D or not 2-D, that is the question: a northernCalifornia test. Geophys Res Lett 32:L12301 DOI10.1029/2005GL022882

Mayeda K, Malagnini L, Walter WR (2007) A new spectralratio method using narrow band coda envelopes: evidence fornon-self-similarity in the Hector Mine sequence. GeophysRes Lett 34:L11303 DOI 10.1029/2007GL030041

Mele G, Rovelli A, Seber D, Barazanghi M (1997) Shear waveattenuation in the lithosphere beneath Italy and surround-ing regions: tectonic implication. J Geophys Res 102:11863–11875

Menardi Noguera A, Rea G (2000) Deep structure of theCampanian–Lucanian Arc (Southern Apennines, Italy).Tectonoph 324:239–265

Mitchell BJ, Pan Y, Xu J, Cong L (1997) Lg coda Q variationacross Eurasia and its relation to crustal evolution. J GeophysRes 102:11767–22779

Morasca P, Mayeda K, Malagnini L, Walter WR (2005) Coda-derived source spectra, moment magnitudes and energy–moment scaling in the western Alps. Geophys J Int160:263–275

Mostardini F, Merlini S (1986) Appennino centro-meridionale.Sezioni geologiche e proposta di modello strutturale. MemSoc Geol It 35:177–202

Patacca E, Scandone P (1989) Post-Tortonian mountainbuilding in the Apennines. The role of the passive sinkingof a relic lithospheric slab. In: Boriani A, Bonafede M,Piccardo GB and Vai GB (eds), The lithosphere in Italy.Acc Naz Lincei 80:157–176

Phillips WS, Harts H, Rutledge JT (2005) Amplitude ratiotomography for regional phase Q. Geophys Res Lett 32(21):L21301 DOI 10.1029/2005GL023870

Royden LH (1993) The tectonic expression of slab pull atcontinental convergent boundaries. Tectonics 12(2):303–325

Sato H, Fehler M (1998) Seismic wave propagation andscattering in the heterogeneous earth. Modern acousticand signal processing. AIP Press, Melville

Shih XR, Chun K-Y, Zhu T (1994) Attenuation of 1–6 s Lgwaves in Eurasia. Geophys Res 99(B12):23859–23874

Singh SK, Herrmann RB (1983) Regionalization of crustal codaQ in the continental United States. J Geophys Res 88:527–538

Street RL, Hermann R, Nuttli O (1975) Spectra characteristicsof the Lg wave generated by central United Statesearthquakes. Geophys J R Astr Soc 41:51–63

Valensise G, Pantosti D, Basili R (2004) Seismology andtectonic setting of the 2002 Molise, Italy, earthquake.Earthquake Spectra 20:S23–S37

Yang X (2002) A numerical investigation of Lg geometricalspreading. Bull Seismol Soc Am 92:3067–3079

Young CY, Ward RW (1980) Three-dimensional Q-l model ofthe Coso Hot Springs known geothermal resource area.J Geophys Res 85:2459–2470

J Seismol