J Seismol (2016) 20:151–165 DOI 10.1007/s10950-015-9517-9

ORIGINAL ARTICLE

Modelling macroseismic observations for historical earthquakes: the cases of the M = 7.0, 1954 Sofades and M = 6.8, 1957 Velestino events (central )

Giannis Papazachos · Costas Papazachos · Andreas Skarlatoudis · Harris Kkallas · Efthimios Lekkas

Received: 4 February 2015 / Accepted: 7 August 2015 / Published online: 26 August 2015 © Springer Science+Business Media Dordrecht 2015

Abstract We attempt to model the spatial distribution intensity, allowing the generation of synthetic of the strong ground motion for the large M =7.0, (stochastic) isoseismals. Also, different site amplifi- 1954 Sofades and M = 6.8, 1957 Velestino events cation factors proposed for the broader Aegean area, (southern basin, central Greece), using the according to local geology, are tested. Finally, we also macroseismic intensities (IMM up to 9+) observed perform a sensitivity analysis of the fault location, within the broader Thessaly area. For this reason, taking into account the available neotectonic data we employ a modified stochastic method realised for the broader southern Thessaly fault zone. The by the EXSIM algorithm for extended sources, in finally determined fault locations are different than order to reproduce the damage distribution of these previously proposed, in agreement with the available earthquakes, in an attempt to combine existing earth- neotectonic information. The observed macroseismic quake information and appropriate scaling relations intensities are in good agreement with the ones derived with surface geology and to investigate the efficiency from the synthetic waveforms, verifying both the use- of the available macroseismic data. For site-effects fulness of the approach, as well as of the macroseismic assessment, we use a new digital geological map data used. Finally, site-effects show clear correlation of the broader Thessaly basin, where geological with the geological classification employed, with formations are grouped by age and mapped on constant amplification factors for each soil class gen- appropriate NEHRP soil classes. Using the previous erally providing better results than generic transfer approach, we estimate synthetic time series for differ- functions. ent rupture scenarios and employ various calibrating relations between PGA/PGV and macroseismic Keywords Macroseismic observations · Stochastic simulation · EXSIM · Thessaly basin · Southern Thessaly fault zone · G. Papazachos () · E. Lekkas Historical earthquakes · Site-effects Faculty of Geology & Geoenvironment, National & Kapodistrian University of Athens, Panepistimioupoli Zografou, GR15784, Athens, Greece e-mail: [email protected] 1 Introduction

· · C. Papazachos A. Skarlatoudis H. Kkallas Thessaly is located in the back-arc Aegean micro- Geophysical Laboratory, Aristotle University of Thessaloniki, PO Box 352-1, GR54124, plate area and is one of the most seismically active Thessaloniki, Greece regions of Greece. The area has been shown to expand 152 J Seismol (2016) 20:151–165 in a more-or-less N-S direction (see Fig. 1), with a of earthquakes along the southern Thessaly rupture velocity of ∼1 cm/year, due to the active stress field. zone, which also involved the 1955 and 1957 events The result of this extension is the creation of nor- (Papastamatiou and Mouyaris 1986). The available mal faults mainly along its southern and northern descriptions for the M = 6.8, 1957 Velestino earth- borders, with dominant east-west strikes, which dip quake (March 8, 39.38◦ N, 22.63◦ E, see Fig. 3), given towards the north (along the southern Thessaly fault by Papazachos and Papazachou (2002), report seri- zone) or the south (mostly in the northern Thessaly ous damages in the prefectures of , , area, Caputo and Pavlides 1993; Mountrakis et al. and . A total of 32701 buildings were 1993; Papazachos and Papazachou 2002). This dom- damaged, 6934 of which collapsed, 10847 were seri- inant E-W faulting pattern is also found along the ously damaged and 14920 were lightly damaged. Two western Thessaly border, where it co-exists with the people were killed while 71 were injured. A large N-S normal faults of the Hellenides mountain belt foreshock (M = 6.5) occurred a few minutes before (e.g. Lekkas 1987). the mainshock and its consequences cannot be dis- The Thessaly basin, the largest basin in central tinguished from those of the mainshock. The largest Greece, has a well-known history of large earth- aftershock (M = 6.0) occurred several hours after the quakes, with mainshocks having typical moment mag- mainshock. nitudes between 6.0 and 7.0 (we have employed In the present work, we attempt to simulate the moment magnitudes for all processing and figures damage distribution for the 1954, M = 7.0 Sofades throughout this study). The seismicity (both instru- earthquake, as well as the M=6.8, 1957 Velestino mental and historic) follows the two discrete rup- earthquake, using simulated seismic motions. These ture zones of the Thessaly basin (e.g. Papazachos motions are appropriately converted to macroseismic et al. 1993), also presented in Fig. 1. The north- intensity, after incorporating the site-effect through the ern rupture zone, along Peneus River, is formed by local geology. For this reason, we have applied the relatively small faults (typically <25 km), associ- stochastic simulation method, which has been effi- ated mainly with historical earthquakes with mag- ciently used for strong-motion simulation applications nitudes up to ∼6.5. In the southern Thessaly zone, during the last three decades. along the boundaries of the Thessaly plain, the The stochastic simulation method was initially pro- faults are relatively large (up to ∼50 km), associ- posed by Boore (1983) as a point-source method and ated with earthquakes exhibiting maximum magni- then applied by many researchers, in order to simulate tudes up to M∼7.0. During the twentieth century, the ground motion from seismic sources (e.g. Boore eight major seismic sequences with mainshock mag- and Atkinson 1987; Toro and MacGuire 1987;Ou nitudes equal or larger than 6.0 occurred in this and Hermann 1990; Atkinson and Boore 1995; Zonno area (1905, 1911, 1930, 1941, 1954, 1955, 1957, et al. 2010). EXSIM is a modified stochastic simu- 1980), with the M = 7.0, 1954 Sofades earth- lation algorithm, using finite-fault modelling with a quake (April 30, 39.28◦ N, 22.29◦ E, see Fig. 2) dynamic corner frequency approach (Motazedian and being the largest event. The available descriptions for Atkinson 2005). In the present work, we have used this event (Papastamatiou and Mouyaris 1986; EXSIM as adapted by Boore (2009), in order to gener- Papazachos and Papazachou 2002) report damages in ate synthetic time series for selected earthquake fault the prefectures of Karditsa, Larisa, Trikala, Phthiotida, rupture scenarios and estimate peak ground acceler- Magnesia and Evritania. Overall, 6599 buildings were ation (PGA) and peak ground velocity (PGV) from destroyed, 9154 were heavily damaged, 12920 were the synthetic time series. The main advantage of the slightly damaged, while the town of Sofades, in the proposed approach is that it can reproduce realistic Karditsa prefecture, was severely damaged. In total, strong motion intensity measures (e.g. PGA, PSA, 25 people were killed and 157 were injured. Also, etc.), although the modelled ruptures are specified by a ground fissures, liquefaction phenomena and hydro- few simple metrics, such as earthquake magnitude and logical changes were observed in several places. The distance, with options to include more detailed infor- largest foreshock (M =4.6)occurredonApril25 mation on fault geometry and slip. In EXSIM, the fault and the largest aftershock (M =5.7)onMay4. is divided into equal sized subfaults, considered as The 1954 event marked the beginning of a series point sources. The ground motions are calculated for J Seismol (2016) 20:151–165 153

22˚ 23˚

Seismological data 0 km 20 Neotectonic data

Average stress field

Larisa

39.5˚ Trikala

Karditsa Thessaly basin

39˚ 39˚ 22˚ 23˚

Fig. 1 N-S extensional stress field in the Thessaly basin, and neotectonic and earthquake observations, while large arrows the associated major active E-W striking faults (adopted from depict the average stress field of the study area (modified from Papazachos et al. 2001). Small arrows denote T-axes from Panagiotopoulos and Papazachos 2008) each subfault using the original point-source method 2 Simulation of the 1954 Sofades earthquake and then summed at the observation point. In most cases, a random rupture scenario can be considered 2.1 Generating synthetic macroseismic maps in the simulations, though it is possible to intro- with stochastic simulation duce specific rupture scenarios (e.g. bi-directional, etc.) Macroseismic data were collected from the published To assess the effect of the local site conditions on database of macroseismic information for the Aegean seismic motions, it is clear that detailed geophysical area (Papazachos et al. 1997). Observed intensities, or geotechnical information could not be employed, ranging from IMM = 5toIMM = 9+,were due to the scale of the study area. For this reason, we available for 75 settlements. The locations of these set- relied on the available geological information, which tlements were used as target sites for the simulations. can provide an initial base for site-effect characterisa- Overall, 19 of the simulation target sites fall into geo- tion. Geological formations were hand-digitized using logical class 1, 4 fall into class 2 and 52 fall into class the available Greek Institute of Geology and Mineral 4. Exploration (IGME) maps (scale 1:50000) and then The fault location was initially constrained follow- clustered according to their age, hence their expected ing the focal parameters (hypocentre and fault loca- dynamic amplification behaviour, into the following tion, fault strike and dip, faulting type, etc.) provided four classes (also see Fig. 2): by Papazachos et al. (2001). The fault’s dimensions (length, L, and width, w) were estimated using the 1. Bedrock/Basement rocks (Mesozoic-Paleogene following Eqs. 1 and 2: age) 2. Molassic type sediments (Paleogene-Neogene log10L = 0.50M − 1.86 (1) age) 3. Neogene sediments 4. Quaternary-Plio/Pleistocene sediments log10w = 0.28M − 0.70 (2) 154 J Seismol (2016) 20:151–165

Fig. 2 Observed macroseismic intensities (Modified Mercalli Sofades fault edge (L1–L3) considered in the present study scale), catalogue epicentre of the M = 7.0, 1954 Sofades earth- for the 1954 event, as well as the main active faults in south- quake, and main geological formation classes within the broader ern Thessaly, proposed by Mountrakis et al. (1993), are also Thessaly basin area. The three candidate locations for the upper presented

which have been proposed for the estimation of earth- For the simulation, the stress parameter was set to quake fault dimensions from moment magnitude for 70 bars (Boore and Joyner 1997). The value 56 bars continental dip-slip faults (Papazachos et al. 2004; which has been proposed for normal and strike-slip Papazachos et al. 2006). The dimensions (length, L faults in Greece by Margaris and Boore (1998)and and width, w) of each subfault were constrained is based on a relatively small number of data, was from Eq. 3, proposed by Beresnev and Atkinson 1999: also tested, but did not have significant impact on the resulting ground motion levels, hence its effect could not be resolved, at least using the available log10l =−2 + 0.4M (3) macroseismic information. The crust’s average shear wave velocity (β) and density (ρ) were derived from and were slightly modified in order to match Papazachos et al. (1966). The rupture propagation the calculated fault dimensions. The available field velocity, Vrup, was set at a typical value of 0.8. Larger observations (Papastamatiou and Mouyaris 1986; values (Vrup = 0.9 and 0.95) were also tested, Mountrakis et al. 1993) showed a large number of resulting in slightly lower (underestimated) synthetic surface fissures and very small surface rupture seg- macroseismic values for the southern Thessaly obser- ments, without revealing a clear surface manifestation vation sites for the higher Vrup value. Different values of the seismic fault along its length. For this reason, were used for the high-frequency attenuation param- the depth of the upper edge of the fault was set to eter, κ, according to the geology/soil class of each 1 km, close to the surface. The sensitivity analysis per- simulation site, as shown in Table 1. For the bedrock formed for this parameter showed that larger depths sites, we used a smaller value than previously sug- resulted in systematic underestimation of macroseis- gested for rock formations in Greece (Margaris and mic intensities, suggesting that the adopted depth Boore 1998) using judgement based on our experi- choice was appropriate. ence for κ estimates from bedrock sites in the broader J Seismol (2016) 20:151–165 155

Fig. 3 Observed macroseismic intensities (Modified Mercalli Thessaly basin area. The three candidate locations for the upper scale), catalogue epicentre of the M = 6.8, 1957 Velestino earth- fault edge (L1–L3) examined in the present study for the quake, and main geological formation classes within the broader 1957 event are also presented

Aegean area. Table 2 summarises all the parameters of (2003, 2007). Moreover, an additional simulation was the model adopted in the simulations. performed, assuming the generic spectral amplifica- The expected synthetic seismograms were initially tions factors proposed for Greece by Klimis et al. simulated without considering the local site-effects (1999), for site classes C and D. (all formations were considered as basement rocks, The final step for generating the synthetic macro- class 1, with no site amplifications). Following this seismic maps involves the estimation of the expected step, the calculated PGA and PGV values were cor- rected for site-effects using the constant amplification factors of the GMPE proposed for the Greek area by Table 2 Parameters of the preferred model used for the strong- Skarlatoudis et al. (2003, 2007). For this estimation motion simulations of the M = 7.0, 1954 Sofades earthquake it was assumed that Bedrock, Molassic/Neogene, and Strike 271◦ Quaternary/Plio-Pleistocene sediments corresponded Dip 47◦ to NEHRP soil categories A/B, C and D, respectively, L44km as these were also considered by Skarlatoudis et al. w18km Top-fault depth 1 km Table 1 High-frequency attenuation parameter κ and NEHRP Sub-faults 7x3 classification, adopted for each geological soil class in the L 6.286 km simulations w6km M7.0 Geological class NEHRP class κ Reference Stress parameter 70 bars 1 A/B 0.015 Adopted after tests Vrup 0.8 2 C 0.044 Klimis et al. (1999) β 3.4 km/s 3 C 0.044 Klimis et al. (1999) ρ 2.7 gr/cm3 4 D 0.066 Klimis et al. (1999) κ (as in Table 1) 156 J Seismol (2016) 20:151–165

Bedrock Molassic Quaternary 10 10 (a) (b) (c) 9 9

8 8

I 7 7

6 6 Bias=0.16 Bias=-0.36 Bias=-0.49 5 RMS=0.82 RMS=0.98 RMS=1.07 5

567891056789105678910 Ibso Ibso Ibso

Fig. 4 Comparison of modelled, I, against observed, Iobs, amplification factors of Skarlatoudis et al. (2003, 2007). c macroseismic intensities for Sofades fault location 1 scenario With the generic spectral site-effect amplifications proposed by (see Fig. 2). a No site-effects. b With the constant site-effect Klimis et al. (1999) macroseismic intensities from the synthetic stochas- as they are closer to the average of all previous rela- tic seismograms. We have adopted a simple approach, tions published for the area of Greece (Theodulidis where the PGA and PGV values from the synthetic and Papazachos 1992; Koliopoulos et al. 1998). waveforms were converted to macroseismic intensi- ties (Modified Mercalli scale, IMM, as adapted in 2.2 Fault location optimisation and site-effect Greece). To perform the conversion, we employed the assessment most recently developed approach by Tselentis and Danciu (2008) for the Greek area and the correspond- Initially, we adopted the fault location proposed ing conversion (4)and(5): by Papazachos et al. (2001), depicted as L1 in Fig. 2. The results of this initial simulation, pre- IMM =−0.946 + 3.563 ∗ log10PGA (4) sented in Fig. 4, showed significant deviations between observed (Iobs) and synthetic (I) intensities. These differences were observed both without site IMM = 3.3 + 3.358 ∗ log10PGV (5) amplifications, as well as when using the constant

Bedrock Molassic Quaternary 10 10 (a) (b) (c) 9 9

8 8

I 7 7

6 6 Bias=0.31 Bias=-0.21 Bias=-0.34 5 RMS=0.84 RMS=0.84 RMS=0.91 5

567891056789105678910 Ibso Ibso Ibso

Fig. 5 Same as Fig. 4, for the second fault location simulation (L2 in Fig. 2) J Seismol (2016) 20:151–165 157

Bedrock Molassic Quaternary 10 10 (a) (b) (c) 9 9

8 8

I

I 7 7

6 6 Bias=0.39 Bias=-0.13 Bias=-0.26 5 RMS=0.89 RMS=0.80 RMS=0.85 5

567891056789105678910 Ibso Ibso Ibso

Fig. 6 Same as Fig. 4, for the third fault location examined (L3 in Fig. 2) site amplification factors proposed by Skarlatoudis deviations between observed and synthetic macroseis- et al. (2003, 2007), and the generic spectral amplifica- mic intensities. The third fault location examined (L3 tions suggested by Klimis et al. (1999). In general, the in Fig. 2), which was constrained by the observed results show that the modelled intensities for bedrock surface ruptures and neotectonic faulting of the south- sites (diamonds in Fig. 4) are lower than the observed ern Thessaly basin (Papastamatiou and Mouyaris intensities. The opposite is observed for soft soil sites, 1986; Caputo and Pavlides 1993; Mountrakis et al. with synthetic macroseismic intensities (no site-effect 1993), also shown in Fig. 2) led to the best results, amplification, Fig. 4a) showing much larger values as is presented in Fig. 6. Figure 6a shows that syn- than the observed ones. This pattern completely con- thetic and macroseismic intensities for bedrock corre- tradicts the usual pattern observed for soft soil sites, late well when no site amplifications are considered. which are expected to exhibit significantly higher site This correlation verifies our assumption, which is amplifications. Due to this discrepancy, the bias and in agreement with the available neotectonic infor- RMS of the fit between synthetic and observed macro- mation for the optimal fault location. However, the seismic data became larger when also using site-effect amplifications (−0.49/1.07 intensity bias/RMS for the Klimis et al. (1999) site effects instead of 0.16/0.82 6 for the case of no site-effects, see also Fig. 4b, c). 5 In order to explain this discrepancy, it is clear that any simulation scenario would require a reduc- 4 tion of the synthetic macroseismic intensities for the soft soil (essentially Quaternary/Plio-Pleistocene) 3 sites, and an increase for the bedrock sites. This pat- tern can be partly fulfilled if the fault’s upper edge Amplification 2 would be located to the south of the initial loca- tion examined, being in better agreement with the 1 mapped active faults of the southern Thessaly fault 0 zone (Mountrakis et al. 1993), as these are presented in 0.1 0.2 0.51 2 510 20 50 100 Fig. 2. For this reason, two additional possible fault Frequency (Hz) locations were considered, depicted as L2 and L3 in Fig. 7 Comparison of the spectral amplifications proposed by Fig. 2. Klimis et al. (1999) for soil classes C and D against the typical As Fig. 5 indicates, simulations using the second HVSR variation with frequency as measured within the urban fault location (L2 in Fig. 2) resulted in slightly smaller Karditsa area 158 J Seismol (2016) 20:151–165 soil sites (mostly Quaternary/Plio-Pleistocene sites, It should be noticed that while the overall intensity class D) show significantly higher observed intensi- data modelling improvement is small, it is quite sig- ties when no site amplifications are considered for nificant. For example for the selected (optimal) fault the stochastic simulation results, as expected due to position (L3), the introduction of site-effects through the presence of significant site amplification effects the local geology index results in smaller intensity bias for class D formations. The optimal results have been (−0.13 instead of 0.39 intensity units) and RMS (0.8 obtained for the third fault location (L3 in Fig. 2), instead of 0.89), corresponding to ∼20 % variance using the constant amplification factors for PGA and reduction between synthetic and observed intensities. PGV proposed by Skarlatoudis et al. (2003, 2007)for While this reduction is not very large, it is clearly sta- the broader Aegean area. Macroseismic intensities for tistically significant, as it arises from the introduction both bedrock and Quaternary formation sites show a of a single site-effect factor. Moreover, recent results very good correlation between predicted and observed for the area of Greece (Stewart et al. 2014) justify intensities (see Fig. 6b). the use of geological proxies as site-effects indica-

Fig. 8 Estimated macroseismic intensities distribution for the figure) and with the use of the constant PGA/PGV site-effect M = 7.0, 1954 Sofades event, considering the optimal fault loca- amplification factors of Skarlatoudis et al. (2003, 2007)(lower tion 3 (see Fig. 2), without the use of site amplifications (upper figure) J Seismol (2016) 20:151–165 159 tors, showing excellent correlation with traditional Table 3 Parameters of the preferred model used for strong- motion simulations of the M = 6.8, 1957 Velestino earthquake site-effect controlling quantities such as Vs30. An interesting feature is that the simulations which Strike 269◦ were performed using the spectral amplification fac- Dip 47◦ tors proposed by Klimis et al. (1999)haveledto L40km a systematic overestimation of macroseismic inten- w16km sities for these mainly Quaternary formation sites Top-fault depth 1 km (Fig. 6c), especially for the central Thessaly basin part. Sub-faults 8x3 Selected HVSR measurements which were performed L5km in the southern Thessaly basin showed that the recov- w5.33km ered resonance frequencies for the central section M6.9 of the basin (Karditsa area) exhibit very low values Stress parameter 70 bars (∼0.4 Hz, as shown in Fig. 7), probably due to the Vrup 0.8 significant thickness of the Quaternary deposits. This β 3.4 km/s low-frequency amplification is very different than the ρ 2.7 gr/cm3 corresponding typical (generic) amplification factors κ (as in Table 1) proposed by Klimis et al. (1999) for NEHRP class D formations. This discrepancy can be considered as the main reason for the observed bias and resulting macroseismic intensity overestimation. In order to a posteriori assess the effect of the Velestino event seismic motions. The depth of the performed modelling for the broader Thessaly area, upper edge of the fault was also set to 1 km, relatively we have adopted the final modelling parameters close to the surface, as existing information sug- (Table 2, L3, constant amplification factors for soil gested the presence of sparse surface fissures/breaks formations, etc.) and performed computations for a for this event. More specifically, a post-event grid of target points with 4 km spacing, in order to field survey (Papastamatiou 1957) showed mainly the estimate the expected damage distribution of the 1954 presence of limited surface fissures (∼500–1000 m), event, in terms of macroseismic intensities, throughout while additional surface deformation effects have not the whole study area. The final damage distribution been confirmed (Ambraseys and Jackson 1990). Simi- maps (with and without using site amplification) are lar to the 1954 Sofades event, slightly larger depths of presented in Fig. 8. The results include the effect of 2 and 3 km were also tested for the upper fault edge the fault geometry, as well as local geology on the but no significant differences were observed. expected (and observed) seismic motions. For exam- It is evident that the consequences of the foreshock ple, significant modifications of the damage pattern and the mainshock (M =6.5andM = 6.8, respec- are observed due to the presence of local site-effects, tively) cannot be practically distinguished. Although while the hanging-wall area is exhibiting higher seis- some efforts have been made to discriminate mixed mic motion levels, which are further enhanced for the macroseismic effects, especially in the case of after- largest part of the southwestern Quaternary Thessaly shocks following a damaging earthquake (e.g. Ferrari basin, due to local site effects. et al. 1995; Vannucci et al. 1999), we essentially have a quite comparable foreshock occurring shortly before the mainshock, hence the adopted simulation method- 3 Simulation of the 1957 Velestino earthquake ology should not be appropriate for such a case. However, reckoning that this is a typical case with The collected macroseismic data for the M =6.8, macroseismic and limited seismological data avail- 1957 Velestino event, included observed intensities ability, we considered useful to attempt to simulate + + (IMM), ranging from 5 to 9 , for 79 settlements (32 the 1957 event seismic motions, as if it was a histori- bedrock sites, 3 Neogene sediments sites and 44 Qua- cal earthquake without any information on the detailed ternary sediments sites). Table 3 summarises all the features of its foreshock or aftershock sequence (i.e. parameters adopted for the simulations of the 1957 presence of a strong preshock). 160 J Seismol (2016) 20:151–165

Bedrock Neogene Quaternary 10 10 (a) (b) (c) 9 Bias=0.77 Bias=0.35 Bias=0.23 9 RMS=1.24 RMS=1.07 RMS=1.06 8 8

I 7 7

6 6

5 5

4 4 5678910 5678910 567891010 Ibso Ibso Ibso

Fig. 9 Comparison of modelled, I, against observed, Iobs, amplification factors of Skarlatoudis et al. (2003, 2007). c macroseismic intensities for Farsala fault location 1 scenario With the generic spectral site-effect amplifications proposed by (see Fig. 3), with a uni-directional rupture and a larger Klimis et al. (1999) fault length. a No site-effects. b With the constant site-effect

As in the 1954 Sofades earthquake simulation, examined simulations. However, in all cases, simula- we initially adopted the original fault location pro- tions using the examined fault locations resulted in posed by Papazachos et al. (2001). This loca- synthetic intensities significantly different from the tion was revised through a trial-and-error approach, observed ones, throughout the whole Thessaly area. and two additional fault locations, to the south of In general, the calculated (synthetic) intensities where the original one, were also tested (L2 and L3 in systematically smaller than the observed ones, possi- Fig. 3), compatible with the active faulting of the bly a result of the combined effect of the mainshock southern Thessaly presented in Fig. 2. A random rup- and its strong foreshock on the observed damage ture scenario was also considered in this case for all pattern.

Fig. 10 Sites for which significant bias (I≥2) is found between observed (Iobs) and synthetic (I) macroseismic intensties for the 1957 Velestino earthquake J Seismol (2016) 20:151–165 161

In order to handle this problem, additional con- approximating the observed intensity distribution cor- figurations were examined, where the rupture length responded to the original Farsala fault location (L1 in was slightly increased, in order to produce larger Fig. 3), however with a larger length (increased by synthetic intensities. Moreover, scenarios involving 5 km towards the east) and an eastwards rupture direc- directive rupture were also considered. It should be tivity with Vrup = 0.8. This length corresponds to a noted that the observed macroseismic data (Fig. 3) slightly larger moment magnitude event (M =6.9), exhibited much larger intensities along the eastern which was expected since additional effects of the side of the fault (IMM ∼8+ to 9+), in compari- large (M = 6.5) foreshock could not be taken into son to its western section (IMM ∼7+), indicating account in the simulations in a systematic manner. the presence of strong directional rupture phenom- The results of the preferred simulation scenario ena (West-to-East rupture propagation). After these are presented in Fig. 9, where a good correla- modifications, the only scenario that was sufficiently tion between observed and synthetic macroseismic

Fig. 11 Estimated macroseismic intensities distribution for the without the use of site amplifications (upper figure) and with the M = 6.8, 1957 Velestino event, considering the optimal fault use of the constant PGA/PGV site-effect amplification factors location 1 (see Fig. 3) with a larger fault length of 40 km, of Skarlatoudis et al. (2003, 2007)(lower figure) 162 J Seismol (2016) 20:151–165 intensities is noticed for most sites. However, there are case of typical overestimation of the assigned inten- six sites (highlighted by a grey triangle in the compar- sity value, considering that all other intensity values isons presented in Fig. 9b, c) for which simulations in the vicinity of this site do not exhibit similar bias. significantly underestimate the observed macroseis- An additional reason for the local underestimation of mic intensities, with differences larger than two inten- macroseismic intensities is that some of the examined sity units. As can be seen in Fig. 10, almost all of sites are located upon flysch formations (classified as these sites are clustered in an area located at the west- Bedrock, class 1), which, in this case, may locally ern part of Thessaly, at a distance of approximately behave as more loose, susceptible to amplification for- 60–80 km from the fault, along its strike. This pat- mations, due to the local geological conditions (e.g. tern suggests that the increased observed intensities presence of a significant weathered siltstone layer, may be due to specific wave propagation phenom- etc.). It should be noted that the bias and RMS are ena, such as complex source radiation combined with significantly improved if we exclude these divergent critical reflections from the Moho (or other inter- values from the computations. More specifically, the crustal discontinuities) toward this part of the Thessaly original bias and RMS (Fig. 9b) are reduced to 0.18 basin, or simply the fact that we are modelling a com- and 0.87, respectively, while the ones presented in plex event with a simple uni-directional rupture. Such Fig. 9c are also reduced to 0.06 and 0.84. specific source-rupture/wave-propagation phenomena The final intensity distribution maps (with and cannot be simulated using the available information without using site amplification) are presented in or the adopted stochastic simulation modelling. This Fig. 11. While the constant site-effect amplification suggests that the proposed methodology cannot fully factors of Skarlatoudis et al. (2003, 2007)wereused replace detailed wave propagation models. No obvious also for this case, similar to the 1954 Sofades event explanation exists for the underestimation of macro- simulation, the generic spectral amplification fac- seismic intensity regarding the village of Kokkotoi tors of Klimis et al. (1999) show a similar (if not (eastern Thessaly, see Fig. 10). This is probably a slightly better) performance. This pattern change is

Fig. 12 Comparison of the Synthetic values Bedrock Molassic Quaternary synthetic PGA (top figures) Skarlatoudis et al. (2003, 2007) Chiou & Youngs (2014) bottom figures and PGV ( ) 0 20406080020406080 values obtained for the final 1954 Sofades simulations, Class A/B Class B (hanging wall) against the empirical Class C Class B (foot wall) predictive relations of Class D Class C (hanging wall) Class D (hanging wall) Skarlatoudis et al. (2003, 2007) for Greece (left

figures) and of Chiou and PGA Youngs (2014) based on the 100 100 100 NGA-West2 PEER world database (right figures)

0 20406080020406080

PGV 10 10 10

0 20406080020406080 Epicentral distance (km) Rrup (km) J Seismol (2016) 20:151–165 163

Fig. 13 Comparison of the Synthetic values Bedrock Neogene Quaternary synthetic PGA (top figures) Skarlatoudis et al. (2003, 2007) Chiou & Youngs (2014) and PGV (bottom figures) 0 204060801000 20406080 values obtained for the final 1957 Velestino simulations, Class A/B Class B (hanging wall) against the empirical Class C Class B (foot wall) Class D Class C (hanging wall) predictive relations of Class D (hanging wall) Skarlatoudis et al. (2003, 2007) for Greece (left figures) and of Chiou and 100PGA 100 100 Youngs (2014) based on the NGA-West2 PEER world database (right figures)

0 204060801000 20406080

PGV 10 10 10

0 204060801000 20406080 Epicentral distance (km) Rrup (km) compatible with the general geological setting, since both examined events) with the empirical predictive the southeastern part of the Thessaly basin exhibits relations of Skarlatoudis et al. (2003, 2007) pro- smaller thicknesses for Quaternary formations, hence posed for the broader Aegean area, as presented in the discrepancy observed in Fig. 7 is expected to be Figs. 12 and 13. A good fit is also observed in the less pronounced or even not present. same figures for the Chiou and Youngs (2014) rela- tion, which is based on the recent NGA-West2 world strong-motion database from similarly active seismo- 4 Conclusions tectonic environments like the Aegean area. It should be noticed that we employ the Rrup (instead of epi- The results obtained in the present study show that central) distance for the later comparison, presenting the simulation of the rupture of historical earthquakes equations for the hanging-wall (where most synthetic can be performed by employing macroseismic data PGA/PGV values refer), while the foot-wall relation and appropriate simulation tools, such as stochastic is also presented for bedrock sites for completeness. simulation algorithms (e.g. EXSIM), and by apply- Similar fits were also obtained for all other NGA2 ing appropriate relations converting PGA and PGV PEER relations (Abrahamson et al. 2014; Boore et al. values (from synthetic waveforms) to macroseismic 2014; Campbell and Bozorgnia 2014) intensities, IMM (e.g. Tselentis and Danciu 2008). From the performed simulations, it can be sug- The simulations also suggest that the use of cal- gested that the originally proposed locations for the ibrating equations for fault dimensions assessment Sofades and Farsala faults, as constrained by the (e.g. Papazachos et al. 2004; Papazachos et al. 2006), already published information, can be updated on can lead to realistic results, even for historical earth- the basis of the available macroseismic information. quakes. This agreement is also confirmed by the Using the simulated macroseismic intensities, espe- comparison of the synthetic PGA and PGV val- cially for sites on bedrock formations which are not ues (obtained from the final simulation scenarios of affected by site-effects, it was possible to relocate 164 J Seismol (2016) 20:151–165 the Sofades fault towards the southern edge of the manuscript. The contribution of Costas Papazachos and Harris southwest Thessaly basin. The proposed location (L3 Kkallas was partly realised within the 3D-SEGMENTS ARIS- in Fig. 2) is in very good agreement with the avail- TEIA project, funded by the EC Social Fund and the Greek Secretariat of Research and Technology. able neotectonic data for the area (e.g. Mountrakis et al. 1993). Moreover, the obtained results confirm the significant effect of local geology on strong seis- References mic motion, showing important differentiation in the ground response of bedrock formations against Qua- Abrahamson NA, Silva WJ, Kamai R (2014) Summary of the ternary / Plio-Pleistocene sediment sites. ASK14 ground motion relation for active crustal regions. The results obtained in this study have showed Earthq Spectra 30:1025–1055 that the generic transfer functions proposed by Klimis Ambraseys NN, Jackson JA (1990) Seismicity and associated et al. (1999) for the area of Greece could not real- strain of central Greece between 1890 and 1988. Geophys J Int 101:663–708 istically describe the observed site-effects, resulting Atkinson GM, Boore DM (1995) Ground-motion relations for in very high amplifications for the Quaternary/Plio- eastern North America. Bull Seism Soc Am 85:17–30 Pleistocene sediments. This discrepancy is most prob- Beresnev IA, Atkinson GM (1999) Generic finite-fault model ably due to the high Quaternary sedimentary thick- for ground-motion prediction in eastern North America. Bull Seism Soc Am 89:608–625 ness of the southern Thessaly basin (several hundred Boore DM (1983) Stochastic simulation of high-frequency meters), leading to low-frequency amplifications that ground motions based on seismological models of the are not adequately reflected in the generic trans- radiated spectra. Bull Seism Soc Am 73:1865–1894 fer functions, in agreement with the available H/V Boore DM (2009) Comparing stochastic point-source and information. Contrariwise, the constant amplification finite-source ground-motion simulations: SMSIM and EXSIM,Bull Seism Soc Am 99(6):3202–3216 factors proposed by Skarlatoudis et al. (2003, 2007) Boore DM, Atkinson GM (1987) Stochastic prediction of for PGA and PGV resulted in satisfactory simulation ground motion and spectral response parameters at hard- results and realistic site amplification estimates. rock sites in eastern north America. Bull Seism Soc Am The simulation for the M = 6.8, 1957 Velestino 77:440–467 Boore DM, Joyner WB (1997) Site amplifications for generic event was less effective, probably due to the fact that rock sites. Bull Seism Soc Am 87:327–341 the observed damage distribution was also affected by Boore DM, Stewart JP, Seyhan E, Atkinson GM (2014) NGA- its strong foreshock. Moreover, the employed spec- West2 equations for predicting PGA, PGV, and 5 % tral and constant site-amplification factors produced damped PSA for shallow crustal earthquakes. Earthq Spec- identical synthetic damage patterns, with equivalent tra 30:1057–1085 Campbell KW, Bozorgnia Y (2014) NGA-West2 ground motion fit to the original damage distribution. Despite this model for the average horizontal components of PGA, being a complex event, we were still able to model it PGV, and 5 % damped linear acceleration response spectra. using a single event damage distribution, by involv- Earthq Spectra 30:1087–1115 ing a fault with larger dimensions. A uni-directional Caputo R, Pavlides S (1993) Late cainozoic geodynamic evolu- tion of thessaly and surroundings (central-northern Greece). rupture distribution has been employed in order to Tectonophysics 223(3):339–362 explain the observed spatial asymmetry of the damage Chiou BS, Youngs RR (2014) Update of the Chiou and Youngs distribution. This observation suggests that it is still NGA model for the average horizontal component of possible to model efficiently even complex historical peak ground motion and response spectra. Earthq Spectra 30:1117–1153 events using their macroseismic data, without intro- Ferrari G, Gasperini P, Guidoboni E (1995) Macroseismic ducing significant bias in the final model estimates intensity evaluation with the fuzzy sets logic. Annali di (e.g. moment magnitude, etc.). Geofisica 38(5-6):811–826 Klimis NS, Margaris BN, Koliopoulos PK (1999) Site- Acknowledgments The new digitized geological map of the dependent amplification functions and response spectra in broader Thessaly area used in this study was produced from Greece. J Earthq Eng 3(02):237–270 the digitization of the corresponding 1:50000 geological maps Koliopoulos PK, Margaris BN, Klimis NS (1998) Duration and of the Greek Institute of Geology and Mineral Exploration energy characteristics of Greek strong motion records. J (IGME). We would like to thank George Vargemezis for provid- Earthq Eng 2(3):391–417 ing help with the geological maps used. HVSR results for the Lekkas E (1987) Geological structure and geodynamic evolu- broader Karditsa area were obtained from HVSR measurements tion of the Koziakas range (western Thessaly) (in Greek), performed within this study. All remaining data used in this PhD thesis, Department of Geology, National and Kapode- work were collected from published sources, referenced in the strian University of Athens J Seismol (2016) 20:151–165 165

Margaris BN, Boore DM (1998) Determination of σ and κ0 seismic fault parameters and earthquake moment. Bull from response spectra of large earthquakes in Greece. Bull Seism Soc Am 36:1482–1489 Seism Soc Am 88:170182 Papazachos BC, Karakaisis GF, Papazachos CB, Scordilis Motazedian D, Atkinson GM (2005) Stochastic finite-fault EM (2006) Perspectives for earthquake prediction in the modelling based on a dynamic corner frequency. Bull Seism Mediterranean and contribution of geological observations. Soc Am 95:991010 In: Robertson AHF, Mountrakis D (eds) Tectonic Develop- Mountrakis D, Kilias A, Pavlides S, Zouros N, Spyropoulos N, ment of the Eastern Mediterranean, vol 260. Publ Geol Soc, Tranos M, Soulakelis N (1993) Field study of the south- London, p 689707 ern Thessaly highly active fault zone. In: Proceedings of Skarlatoudis AA, Papazachos CB, Margaris BN, Theodulidis the 2nd Congress of the Hellenic Geophysical Union, vol 2, N, Papaioannou Ch, Kalogeras I, Scordilis EM, Karakostas p 603614 V (2003) Empirical peak ground motion predictive rela- Ou GB, Hermann RB (1990) Estimation theory for peak ground tions for shallow earthquakes in Greece. Bull Seism Soc motion. Seismol Res Lett 61(2):99–107 Am 93:2591–2603 Panagiotopoulos DG, Papazachos CB (2008) Application of Skarlatoudis AA, Papazachos CB, Margaris BN, Theodulidis N, modern methods for assessing the seismicity of Karditsa. Papaioannou Ch, Kalogeras I, Scordilis EM, Karakostas V In: Proceedings of the 1st development conference of (2007) Erratum to empirical peak ground motion predictive Karditsa relations for shallow earthquakes in Greece. Bull Seism Soc Papastamatiou I (1957) The earthquake of Velestino of 8 March Am 97:2219–2221 1957, Report Inst. Geologias & Erevnon Ypedaphus, p 11, Stewart JP, Klimis N, Savvaidis A, Theodoulidis N, Zargli Athens E, Athanasopoulos G, Pelekis P, Mylonakis G, Margaris Papastamatiou D, Mouyaris N (1986) The earthquake of April B (2014) Compilation of a local Vs profile database 30, 1954, in Sophades (central Greece). Geophys J R astr and its application for inference of Vs30 from geologic- Soc 87:885–895 and terrain-based proxies. Bull Seism Soc Am 104:2287– Papazachos BC, Papazachou C (2002) The earthquakes of 2841 Greece. Ziti Publ., Thessaloniki Theodulidis NP, Papazachos BC (1992) Dependence of strong Papazachos BC, Comninakis PE, Drakopoulos JC (1966) Pre- ground motion on magnitude-distance, site geology and liminary results of an investigation of crustal structure in macroseismic intensity for shallow earthquakes in Greece: southeastern Europe. Bull Seism Soc Am 56(6):1241–1268 I, peak horizontal acceleration, velocity and displacement. Papazachos BC, Hatzidimitriou PM, Karakaisis GF, Papaza- Soil Dyn Earthq Eng 11(7):387–402 chos CB, Tsokas GN (1993) Rupture zones and active Toro GR, McGuire RK (1987) An investigation into earthquake crustal deformation in southern Thessalia, central Greece. ground motion characteristics in eastern north America. Boll Geof Teor Appl 35:363–374 Bull Seism Soc Am 77(2):468–489 Papazachos BC, Papaioannou Ch, Papazachos CB, Savvaidis Tselentis GA, Danciu L (2008) Empirical relationships between AS (1997) Atlas of isoseismal maps for strong shallow modified Mercalli intensity and engineering ground-motion earthquakes in Greece and surrounding area (426bc-1995). parameters in Greece. Bull Seism Soc Am 98(4):1863– Ziti Publ., Thessaloniki 1875 Papazachos BC, Mountrakis DM, Papazachos CB, Tranos MD, Vannucci G, Gasperini P, Ferrari G, Guidoboni E (1999) Encod- Karakaisis GF, Savvaidis AS (2001) The faults which have ing and computer analysis of macroseismic effects. Phys caused the known major earthquakes in Greece and sur- and Chem of the Earth 24:505–510 rounding region between the 5th century bc and today. Zonno G, Oliveira CS, Ferreira MA, Musacchio G, Meroni F, In: Proceedings of 2nd National Conference Anti-Seismic Mota-de-Sa F, Neves F (2010) Assessing seismic damage Engineering and Technical Seismology, p 1726 through stochastic simulation of ground shaking: the case of Papazachos BC, Scordilis EM, Panagiotopoulos DG, Papaza- the 1998 Faial earthquake (Azores Islands). Surv Geophys chos CB, Karakaisis GF (2004) Global relations between 31:361–381