1 INTRODUCTION

This work continues a series of studies on the recognition of strong-earthquake-prone areas by pattern recognition methods. This direction has been developed at the Institute of Earthquake Prediction Theory and Mathematical Geophysics of the Russian Academy of Sciences (IEPT RAS) since 1990 and earlier, since the 1970s, at Schmidt Institute of Physics of the Earth of the Russian Academy of Sciences (IPE RAS). The fundamentals of the method were formulated by I.M. Gelfand, V.I. Keilis-Borok, and E.Ya. Rantzman who hypothesized that the epicenters of the strong earthquakes (with magnitude M ≥ M0 where M0 is a given threshold) are confined to the intersections of morphostructural lineaments which bound crustal blocks and are determined by the special technique of morphostructural zoning (MSZ) described in (Alekseevskaya et al., 1977a; 1977b; Ranzman, 1979; Gorshkov et al., 2003). The hypothesis of epicenters’ confinement was validated by the statistical analysis of relative positions of the epicenters and intersections of morphostructural lineaments conducted by A.D. Gvishiani and A.A. Soloviev (1981) and by direct testing the results of recognition by this approach against the data on the earthquakes that occurred in a study region after publication of the recognition results (Gorshkov et al., 2001; Gorshkov, 2010; Soloviev et al., 2015). The results of the last testing for all the 26 previously studied regions (Gorshkov and Novikova, 2018) have shown that about 90% of the earthquakes of considered magnitudes occurred at the lineament intersections and 86% of such events are associated with the intersections that were recognized as highly seismic. The violations of the hypothesis revealed by testing are probably associated with errors in the determination of the location and magnitude of the earthquakes and with the errors in the location of intersections.

In this paper, we consider one seismically most active regions in the Mediterranean which includes the southern part of the Balkan Peninsula belonging to Greece. The high seismicity of the study area is associated with the complex geodynamic situation resulting from the interaction between the different-size Eurasian, African, Adriatic, Aegean, and Anatolian continental plates. According to (Papazachos et al., 2000; Makropoulos and Burton, 1981), 19 events with M ≥ 7.0 occurred there since 1444. The maximum magnitudes were observed during the earthquake of 1867 (Мw = 7.4) on the island of Kefalonia in the Ionian Sea and during the Peloponnesian earthquake of 1886 (Mw = 7.3) in the southwestern Greece (Papazachos et al., 2000).

The studies on seismic hazard assessment in Greece, typically based on probabilistic approaches, have been actively developed since recently for the entire territory and for its individual segments that have been most frequently affected by seismic events (Ambraseys and Finkel, 1993; Ambraseys and Jackson, 1997; Burton et al., 2003; EAK, 2003; Papadopoulos et al., 2002; Tselentis and Danciu, 2010). The same goal is also pursued by extensive studying and mapping of active faults in the region for exploring their connection with the specific seismic events (Bernard et al., 2006; Boccaletti et al., 1997; Briole et al., 2000; Ganas et al., 2004; Chousianitis et al., 2013; Jackson et al., 1982; Kokkalas et al., 2007; Pavlides et al., 2002; 2007; Tsimi et al., 2007; http://eqgeogr.weebly. com/database-of-active-faults.html).

In this work, we aim to determine the areas prone to the earthquakes with М ≥ 7.0 in the Hellenides. The approach we use here was previously applied to the region under study for identifying the D-intersections relative to the magnitude threshold М0 = 6.5 based on the 1 : 2 500 000 morphostructural zoning (MSZ) scheme covering a vast region from the Balkan Peninsula in the west to the Armenian Plateau in Transcaucasia in the east (Gelfand et al., 1974). In contrast to the cited work, we now solve the problem of the recognition of D-intersections for a higher magnitude threshold М0 = 7 using a more detailed 1 : 1 000 000 MSZ scheme .

2 RECOGNITION OBJECTS: DEFINITION AND PARAMETERS

As an object of analysis in this work, we use the intersections of morphostructural lineaments whose position in the Hellenides is identified based on the scale 1 : 1 000 000 MSZ (Alekseevskaya et al., 1977; Rantzman, 1979) (Fig. 1).

Fig. 1.
figure 1

Morphostructural zoning Scheme of Hellenides. Lines show lineaments. Line thickness corresponds to rank of lineaments: thickest lines correspond to first rank; lines with medium thickness correspond to second rank; thin lines correspond to third rank. Solid lines show longitudinal lineaments, dotted lines indicate the transverse ones. Black circles show epicenters of earthquakes with М ≥ 7.0 after 1900; light circles show epicenters with Мw ≥ 7.0 before 1900. Dark and light squares mark earthquake epicenters with 6.0 ≤ Мw < 6.99 after and before 1900, respectively. I to V are numbers of megablocks; 1 to 150 are numbers of lineament intersections.

Full size image

2.1 Morphostructural Zoning of the Hellenides

MSZ is based on the concept of a hierarchical block structure of the Earth’s crust with interaction between the blocks thought of as the cause of seismicity (Sadovskii and Pisarenko, 1991; Keilis-Borok, 1990). Three hierarchical levels of the blocks are distinguished in the MSZ schemes. The lowest level (third rank) blocks are characterized by the close values ​​of the informative topographic traits (height elevation level and orientation of linear landforms). The interblock boundaries are drawn along the lines where the value of at least one characteristic trait changes sharply and significantly. The blocks are combined into megablocks representing the second level of the hierarchy. If the values ​​of the informative traits monotonously vary from one block to another, the violation of this monotony demarcates the boundaries between the megablocks. The largest unit of the highest (first) rank zoning is a mountain country—the territory with a common appearance of the relief and a common type of the orogenic process. The lineament is assigned the highest rank of the morphostructures it divides.

The present structure and configuration of the Hellenic mountain belt was formed by the complex interaction of the African Plate with the Eurasian Plate and Adriatic, Aegean, and Anatolian microplates located between these plates (McKenzie, 1970). The mountain edifice of the Hellenides extends steeply northwesterly along the coastline of the Ionian Sea. The Hellenides are subdivided into the northern segment composed of the Pindos Mountains and the southern segment occupying the Peloponnese peninsula (Fig. 1). These segments are separated by the Gulf of Corinth filling the zone of the young Quaternary rift (Armijo et al., 1996). The mountain ranges of northern Hellenides (External Hellenides, externides) sharply lower down eastwards being replaced by the scattered relatively low ridges which alternate with the intramontane and coastal troughs. Structurally, these low-hill areas belong to the Internal Hellenides (internides) extending eastwards to the Aegean Sea (Kilias et al., 2002).

The first-rank lineaments separate the mountain country of Hellenides from the adjacent large-scale geostructures of the first rank. In the west, east, and south, the lineaments of the first rank separate the mountain edifice of Hellenides from deep basins of the Ionian and Aegean seas and the Sea of Crete, respectively (Fig. 1). The zones of these lineaments are traced along the continental slope and include large tectonic faults (Kilias et al., 2002).

The second-rank lineaments are the boundary zones separating the territorial units of the second rank—the megablocks. The division of the mountain country into the megablocks is determined by the described peculiarities of the relief of the Hellenides. Five megablocks (I–V) which differ in the height level and in the strike of their constituting topographic units are distinguished. Megablocks I (the Pindos Mountains) and IV (the Peloponnese ridges) are identified within the External Zone of the Hellenides. Megablock I is dominated by linearly elongated north-northwest striking ridges, whereas in megablock IV, the strike of the Peloponnese ridges becomes submeridional. Megablocks II, III, and V, characterized by the lower relief, are established in the Internal Hellenides zone. Megablock III includes a vast intermontane Larisa basin. Megablock V (Peloponnese) is separated from the others by a transverse lineament of the second rank which is traced along the tectonically most active southern edge of the Gulf of Corinth (Armijo et al., 1996). The second-rank longitudinal lineaments 10–53 and 51–96 (Fig. 1) separate the megablocks of the External Hellenides (I and IV) from the megablocks of the Internal Hellenides (II, III, and IV).

Lineaments of the third rank are the block boundaries. They mark the regions where the large relief units sharply change their height and strike. A fairly dense network of the transverse third-rank lineaments striking northeast and sublatitudinally and cutting the predominant north-northwestern strike of the Hellenides mountain ranges is revealed (Fig. 1). This is consistent with the geological data on the widespread presence of the “anti-Hellenidic” faults in the region (Goldsworthy et al., 2002).

Overall, 150 lineament intersections were identified by MSZ (Fig. 1). Below, 139 of them are considered as the recognition objects because the crustal Mw ≥ 7.0 earthquakes are confined to the intersections of lineaments (Fig. 1).

2.2 Parameters of the Recognition Objects

The experience of the recognition of the strong-earthquake-prone areas in different seismic regions has shown that using the adequate set of the geological and geophysical parameters describing the recognition objects is one of the most important conditions for success in solving this problem (Gorshkov, 2010). This description is typically based on the geological and geophysical data that indirectly reflect the intensity of the tectonic movements and fragmentation of the Earth’s crust. These data include morphometric indicators of surface relief, geometrical features of the lineament network, and the data on gravity anomalies. Recently, these studies began to consider magnetic anomalies (Soloviev et al., 2016). Here, we used the parameters (Table 1) containing the information on the contrast and intensity of neotectonic movements in the form of topographic elevations, combinations of the terrain types, the area of Quaternary deposits within the vicinity of the intersections, as well as the data on the geometry of the lineament-block structure characterizing the fragmentation of the medium in the vicinity of intersections based on the information about the number and rank of the lineaments forming the intersection, and the distances to the neighboring intersections and block boundaries.

Table 1.  Lineament intersection parameters and discretization thresholds
Full size table

The values ​​of the parameters were determined from the topographic and geological maps and the MSZ scheme. The parameter values ​​were measured in a circle with a radius of 30 km centered at the intersection of lineaments.

In this work, the recognition algorithm Cora-3 (Bongard, 1967; Gvishiani et al., 1988; Gorshkov et al., 2003; Gorshkov, 2010) is applied to the objects in the form of binary vectors. The initial vectors of parameter values ​​corresponding to the recognition objects were converted into binary vectors with the use of the discretization and coding procedures described in detail in (Gelfand et al., 1976a; 1976b; Gvishiani et al., 1988; Gorshkov et al., 2003; Gorshkov, 2010). As a result of the discretization, parameter values ​​are converted into the components of binary vectors indicating the discretization segment the parameter value belongs to.

3 RECOGNITION OF HIGHLY SEISMIC (М ≥ 7.0) LINEAMENT INTERSECTIONS

The earthquakes with М ≥ 7.0 are detected in the vicinity of a relatively small part of intersections of morphostructural lineaments identified in the region (Fig. 1). The purpose is to determine the criteria for discriminating highly seismic (dangerous) seismic intersections (class D) from low seismic (not-dangerous) intersections (class N) relative to the magnitude threshold М0 = 7.0 using the geological and geophysical information about these intersections. The problem is solved by the pattern recognition methods (Bongard, 1967) with the use of the Cora-3 algorithm with learning (Gelfand et al., 1972a; 1972b). The recognition objects are intersections of lineaments each of which is described by the vector of geological and geophysical parameters (Table 1). As a result of recognition, all the intersections of the studied region are divided into those whose vicinity is prone to the earthquakes with M ≥ 7.0 and those whose vicinity is only prone to the events of lower magnitudes. This separation is carried out by the algorithm based on the decision rule determined at the training stage.

The training material for the Cora-3 algorithm is formed based on the information about the seismicity of the studied region. We considered the crustal earthquakes with M ≥ 7.0 including the historical ones. The epicenters of these events are shown in Fig. 1 according to catalogs (Papazachos et al., 2000; 2010) and the database “Active faults of the broader Aegean region in the Greek database of seismogenic sources” (http://eqgeogr.weebly. com/database-of-active-faults.html). The earthquake catalogs used in our study contain information about the earthquakes from 550 B.C. to present. Using the macroseismic magnitude data for the earthquakes in the catalogs, we determined the moment magnitudes Мw. Based on these catalogs, we selected 19 events with Мw ≥ 7.0. The epicenters and years of occurrence of these events are shown in Fig. 1 and Fig. 2.

Fig. 2.
figure 2

Morphostructural zoning scheme of Hellenides and lineament intersections attributed to class 0 for Мw ≥ 7.0. Circles mark highly seismic (dangerous, D) intersections for Мw ≥ 7.0. Other notations are indicated in Fig. 1.

Full size image

One hundred and thirty nine lineament intersections composed the set of the recognition objects. The training set for class D comprised 19 intersections whose 30-km vicinity contained earthquake epicenters with Мw ≥ 7.0. Fifty six intersections close to earthquake epicenters with 6.0 ≤ Мw ≤ 6.99 were attributed to subset X whose objects are correlated to the events with magnitudes close to the considered magnitude threshold. These objects were not used by the Cora-3 algorithm for determining the decision rule at the training stage; however, they were divided into classes D and N at the classification stage. The remaining 64 intersections composed the training set for class N.

Recognition results. The decision rule (Table 2) consists of two sets of the characteristic traits: the characteristic traits of class D (or D-traits) and the characteristic traits of class N (or N-traits) as well as the voting threshold Δ. The recognition object is attributed to class D if the difference between the number of D-traits and the number of N-traits possessed by this object does not exceed Δ. The established traits at voting threshold Δ = 1 are presented in Table 2 and the recognition result is shown in Fig. 2.

Table 2.  Characteristic traits of D- and N-intersections (decisive rule)
Full size table

The characteristic traits comprise the following parameters: the maximum elevation of the relief in the vicinity of the intersection (Hmax), the minimum elevation of the relief in the vicinity of the intersection (Hmin), ΔH = Hmax – Hmin, the distance L between the points of Hmax and Hmin, the percentage of the area covered with loose Quaternary rocks within the vicinity of the intersection (Q), the distance from the intersection to the nearest lineament of the second rank (P2) and similar distance for the first rank lineament (P1), as well as the distance to the nearest intersection (Pint). Yet another characteristic (Mor) reflects the combination of the relief types in the vicinity of the intersection. The stability of the obtained classification of the intersections into classes D and N was verified by a series of the control experiments described in (Gvishiani et al., 1988). In these experiments, at most 10% of the objects of the main result shown in Fig. 2 changed their membership in classes D or N. According to the empirical rule substantiated in (Gvishiani et al., 1988), this indicates the stability of the obtained classification.

4 DISCUSSION

Local researchers associate the strong earthquakes in mainland Greece with the individual active faults (Briole et al., 2000; Ganas et al., 2004; Chousianitis et al., 2013; Jackson et al., 1982; Kokkalas et al., 2007; Pavlides et al., 2002; 2007; Tsimi et al., 2007). In the western part of the region, the strongest earthquakes are associated with the marine Kefalonia transform fault (the first-rank lineament 62–104 in Fig. 1). The approach used in this work has shown that the strong earthquakes in the region are confined to the intersections of lineaments. There is no fundamental contradiction here because most lineament zones established by MSZ (Fig. 1) fully or partially include the tectonic faults, inter alia, the active ones known in the region (Goldsworthy et al., 2002; http://eqgeogr.weebly.com/database-of-active-faults.html). However, the maps of active faults are always incomplete because these faults are typically identified with the consideration of seismicity for individual regions where an earthquake occurred with surface rupture (earthquake rupture reaching the Earth’s surface). The reliable recording of seismicity dates back a few dozen years. According to (Pavlides et al., 2010), reliable recording in Greece began in the 1970s. During MSZ, the region of interest is studied by the unified rules for the entire area and without the consideration of seismicity. Therefore, the MSZ schemes more comprehensively reflect the potential seismogenic structures of the region.

As can be seen from Fig. 2, all the intersections correlated to the earthquakes with M ≥ 7.0 are attributed by the recognition algorithm to the highly seismic (D) class. Besides, eleven intersections from the training set of the low-seismic (N) class and 13 intersections from the testing set Х which contains intersections whose vicinity hosts the epicenters with 6.0 ≤ М ≤ 6.99 are also referred to class D.

The recognized D-intersections outline several highly seismic (D) zones in most of which the earthquakes with М ≥ 7.0 have already occurred. In particular, this is the case with the clusters of D-intersections recognized on the continental slope of the Ionian Sea and southern coast of the Peloponnese Peninsula. At the same time, the zones formed by D-intersections are also revealed in the areas where the earthquakes with М ≥ 7.0 have never been observed before. These D-intersections are recognized on the lineaments bordering the rift zone of the Gulf of Corinth. This testifies to the high seismic potential of this region whose increased tectonic activity was noted by many researchers (Briole et al., 2000; Kokkalas et al., 2007; Tsimi et al., 2007). Another group of D-intersections (nos. 44, 46 and 47) was recognized in the region of Euboea Island where the western continuation of the North Anatolian Fault approaches the Greek coast.

Since a significant part of the lineaments corresponds to the active faults, the presence of the recognized D-intersections on these lineaments allows some conclusions to be made about the potential of the corresponding active fault. In particular, the recognized D-intersections nos. 90 and 97 (Fig. 2) are located on the lineament corresponding to the Psathopyrgos fault whose seismic potential is actively debated (Sokos et al., 2012). Our results suggest the high potential of this fault, supporting the similar conclusions made for this fault in (Vassilakis et al., 2011).

Most of the recognized D-intersections are spatially correlated to the seismogenic zones of the Greek database “Active faults of the broader Aegean region in the Greek database of seismogenic sources” (Pavlides et al., 2007; http://eqgeogr.weebly.com/database-of-active-faults.html). In this database, the seismogenic zones are not graded in terms of the seismic potential; it is only noted that this potential is fairly high (Pavlides et al., 2007). The recognition results provide the information for assessing the seismic potential of these zones relative to М ≥ 7.0. In the cases when these zones contain D-intersections, the magnitude of the earthquakes can reach and exceed 7. Correspondingly, the seismogenic zones which contain lineament intersections but the latter have not been recognized as dangerous for M ≥ 7.0 can be attributed to the areas of lower seismic potential. This refers to several seismogenic zones which are marked by the intersections in Fig. 2:

—3, 4, and 26,

—73 and 74,

—76,

—101 and 107.

Thus, the results of this work provide information for dividing the seismogenic zones of the Greek database according to the value of their seismic potential.

The characteristic traits (Table 2) show that D-intersections are marked by not “small” values ​​of the maximum height (Hmax > 1143 m), “small” values ​​of the minimum height (Hmin ≤ –106 m), “large” values ​​of the height difference (ΔH > 2183 m) and not “large” values ​​of the height gradient (ΔH/L ≤ 74) combined with “large” values ​​of the area occupied by ​​the Quaternary deposits (Q > 20). The “large” height difference indicates that the relief in the vicinity of the D‑intersections is fairly contrasting. The intervals of the listed parameters selected by the algorithm suggest a lowered position of the vicinity of D-intersections in the setting of a contrasting topography and steady subsidence as follows from the “large” Q values. These ranges of the parameters can be explained by the conditions of extension dominating both in the Aegean basin and the mainland Greece (Armijo et al., 1996).