Natural Hazards and Earth System Sciences Clues to the identification of a seismogenic source from environmental effects : the case of the 1905 Calabria ( Southern Italy ) earthquake

The 8 September 1905 Calabria (Southern Italy) earthquake belongs to a peculiar family of highly destructive (I0=XI) seismic events, occurred at the dawning of the instrumental seismology, for which the location, geometry and size of the causative source are still substantially unconstrained. During the century elapsed since the earthquake, previous Authors identified three different epicenters that are more than 50 km apart and proposed magnitudes ranging from M≤6.2 toM=7.9. Even larger uncertainties were found when the geometry of the earthquake source was estimated. In this study, we constrain the magnitude, location and kinematics of the 1905 earthquake through the analysis of the remarkable environmental effects produced by the event (117 reviewed observations at 73 different localities throughout Calabria). The data used in our analysis include ground effects (landslides, rock falls and lateral spreads) and hydrological changes (streamflow variations, liquefaction, rise of water temperature and turbidity). To better define the magnitude of the event we use a number of empirical relations between seismic source parameters and distribution of ground effects and hydrological changes. In order to provide constraints to the location of the event and to the geometry of the source, we reproduce the coseismic static strain associated with different possible 1905 causative faults and compare its pattern to the documented streamflow changes. From the analysis of the seismically-induced environmental changes we find that: 1) the 1905 earthquake had a minimum magnitudeM=6.7; 2) the event occurred in an offshore area west of the epicenters proposed by the historical seismic Catalogs; 3) it most likely occurred along a 100 ◦ N oriented normal fault with a left-lateral component, consistently with the seismotectonic setting of the area. Correspondence to: A. Tertulliani (tertul@ingv.it)


Introduction
The modern seismological networks currently use powerful technological tools that provide highest quality earthquake locations.On the contrary, we do not know with enough accuracy the epicentral location nor the source of a wide number of strong earthquakes occurred in the early decades of the past century.These earthquakes belong to a kind of 'shadow zone' that encloses all the events for which, for several reasons, we have poor seismological, macroseismic or geological data (seismic records, direct historical accounts, primary surface faulting etc).Therefore, in order to better characterize the source of these events we have to consider other kind of data such as earthquake-induced ground effects and hydrological changes, that have been known and were widely reported for centuries, but were progressively left aside with the advent of the fast-evolving instrumental seismology.Such environmental effects can be properly considered a footmark of the earthquake (e.g.Muir-Wood and King, 1993;Porfido et al., 2002;Pizzino et al., 2004), so they are potentially associated with its seismogenesis.The 8 September 1905 Central Calabria (Southern Italy) seismic event is a perfect candidate for this kind of studies.
In this study we want to support new evidence concerning the location and magnitude of the 1905 event, discriminating between three different epicentral solutions: A offshore westward, B inland and C offshore northward (Fig. 1 and Table 1).We will also provide new clues on the geometry of the event source, checking several individual faults associated with those locations.To these aims we will: 1) describe the remarkable dataset of environmental effects caused by the earthquake; 2) derive reliable magnitudes from the distribution of ground effects and hydrological changes; 3) use these magnitudes in empirical relations to deduce fault area and seismic slip of a number of potential sources at locations A, B, and C, and calculate the coseismic strain field associated Published by Copernicus Publications on behalf of the European Geosciences Union.A offshore westward (Michelini et al., 2006).B inland (Baratta, 1906;Mercalli, 1906;Boschi et al., 1995;CPTI Working Group, 2004).C offshore northward (Rizzo, 1907;Riuscetti and Shick, 1974;Martini and Scarpa, 1983;Mulargia et al., 1984;Postpischl, 1985;Westaway, 1992).CV (Capo Vaticano Fault), VV (Vibo Valentia Fault) and CC (Coccorino Fault) are three fault segments (hachures on the downthrown block) that are potentially associated with the event according to Piatanesi and Tinti (2002) (CV and VV) and Cucci and Tertulliani (2006) (CC).Black lines include: 1) highest level of destruction (intensity 9-10/10 MCS), 2) very heavy damage (intensity 8-9/9 MCS), 3) heavy damage (intensity 7-8/8 MCS).A grey arrow indicates the area in study.The inset shows a sketch of the structural arcs and of the regional stress regime in Italy (modified from Montone et al., 1999Montone et al., , 2004)).with those faults; 4) analyze which of the above cited sources better matches the environmental effects that we collected.
Our main purpose is to verify if it is possible to characterize an earthquake (in terms of magnitude, location, and geometry of the source) by using only the environmental effects produced by the quake itself.

Background on the 1905 earthquake
The earthquake struck in the early hours of 8 September a vast area of the Tyrrhenian side of the Calabria peninsula (Fig. 1), causing the death of 557 people and the heaviest damage (I 0 =XI in CPTI Working Group, 2004) between the towns of Lametia and Nicotera.
The event produced a great number of effects on the environment (Fig. 2).Changes in the flow and in the temperature of rivers and springs were eyewitnessed over the entire Calabria region, as well as diffuse ground cracking, landslides, liquefactions, and sparse light phenomena (e.g.Rizzo, 1907); all these environmental effects were observed soon after the earthquake.Moreover, a moderate tsunami was observed and recorded in the open sea and along the coastlines of Calabria and Sicily, and propagated northward in the Tyrrhenian Sea to large distances from the epicentral region (e.g.Boschi et al., 1995).
Among the highly destructive earthquakes that struck the Italian peninsula in the past century, the 8 September 1905 event is one of the most significant and one of the less understood at the same time.So far no convincing or unambiguous elements exist on the location and the geometry of the earthquake fault, nor on its magnitude.The first hypotheses on the event location were by coeval Authors who investigated the most damaged area in central-western Calabria; among them Baratta (1906) and Mercalli (1906)   inland epicenter close to Vibo Valentia (location B in Fig. 1), while Rizzo (1907) suggested an offshore location in the Gulf of S. Eufemia (location C in Fig. 1), based also on the analysis of few seismic recordings.In the following decades, Riuscetti and Schick (1974) and Martini and Scarpa (1983) provided a similar offshore epicenter, although with different values of magnitude (M s =7.0 and M k =7.3, respectively).Different locations of the event were also reported in the catalogs of Italian historical seismicity, as Postpischl (1985) again located the earthquake offshore, close to location C, while CPTI Working Group ( 2004) moved it inland close to Vibo Valentia (location B).In more recent times Michelini et al. (2006) proposed, by means of probabilistic algorithms, a new hypocentral location about 30 km offshore to the west of Capo Vaticano (location A in Fig. 1).
As for the magnitude, the 1905 event has been considered as a moderate, a strong, and even a major earthquake.Indeed we can pass from Westaway (1992), who claimed a M≤6.2 for the event by comparing the 1 m tsunami height of the event to that (8-12 m) produced by the nearby 1908 earthquake, to M s =6.8 suggested by Abe and Noguchi (1983) in their Catalog of revised magnitudes, to the value of M s =7.9 calculated by Duda (1965).To complete the ref-erence frame of the energy released by the 1905 event we cite Galli (2000), who calculated a minimum M=7.38 on the base of indications of liquefaction features.Further studies about the 1905 event were provided by Fantucci and Sorriso-Valvo (1999), who found earthquake-induced growth anomalies in trees through dendro-geomorphological analysis, and by Tinti et al. (2004), who associated to this event a tsunami intensity 3 (on a scale of 6 degrees) in the New Catalogue of Italian Tsunamis.
Uncertainties on the geometry and kinematics of the source of the 1905 earthquake are also remarkable.According to Riuscetti and Schick (1974) the event probably occurred on a subvertical thrust fault, whilst Mulargia et al. (1984) stressed that the occurrence of a tsunami wave implies large coseismic displacements and suggested a normal faulting mechanism for the event.Piatanesi and Tinti (2002) tried to model the tsunami waves but could not discriminate between the offshore Capo Vaticano Fault and the inland Vibo Valentia Fault (both NE-trending and NW-dipping, Fig. 1).Cucci and Tertulliani (2006) compared the set of geological, topographic and macroseismic data and found that the WNW-trending, SSW-dipping Coccorino normal fault (Fig. 1) is a candidate source of the earthquake.
3 The environmental effects of earthquakes

Ground effects
Moderate to strong earthquakes (M>5.0) are capable to produce peculiar ground effects on the environment.Those coseismic effects can be primary, if permanent features (i.e.surface faulting) are directly produced by the earthquake, or secondary, if they are triggered by the ground motion (e.g.landslides, slope failures, liquefaction, cracks, etc.).The area affected by landslides or other ground failures is strongly dependent on the magnitude and on other critical factors such as the rupture fault distance or the epicentral distance (Keefer, 1984), the lithology and the slope steepness (Keefer, 2000).However, the distribution of ground effects on the territory is conditioned by many triggering factors in addition to the earthquake, as soil conditions, vegetation, rainfall, weathering, slope, water content, drainage (Kojima and Obayashi, 2006).Several statistical investigations found that, for different magnitudes, an upper threshold can be individuated for the distance epicenter/landslide.Strong earthquakes can trigger landslides over hundreds kilometres from the rupture fault or the epicenter (Keefer, 1984).However, most of secondary ground effects are common within 80-90 km from the epicenter (i.e.Harp and Jibson, 1996;Rodriguez et al., 1999), as observed also for several Italian earthquakes (Esposito et al., 2000;Prestininzi and Romeo, 2000;Porfido et al., 2002).

Hydrological changes
Hydrological changes associated with earthquakes have been observed since centuries.The interactions between earthquakes and hydrological processes include variations of streamflow, wells' level, liquefaction, and variations in the chemical characteristics of waters.Most of them are of coseismic nature, but their occurrence has been also observed prior or after a major event.Muir-Wood and King (1993), Esposito et al. (2000), Porfido et al. (2002) tried to infer, for historical earthquakes, some relationship between such disturbances and the earthquake characteristics (fault type, epicentral distance, intensity and magnitude).In particular, Muir-Wood and King (1993) found that hydrological changes would accompany major normal fault, showing increase in streamflow and spring rise; reverse faults, on the contrary, would show negligible or undetectable effects, whilst strike slip and oblique faults would generate a combination of responses.Recently, the spatial pattern of water level changes has been compared with simulated coseismic strain changes (Grecksch et al., 1999;Ge and Stover, 2000;Lee et al., 2002;Montgomery and Manga, 2003;Caporali et al., 2005).The area interested by contractional volumetric strain seems to be in good agreement with the water level rise and water excess in streamflows.Montgomery and Manga (2003) underlined that the response of aquifers is different when considering the We include liquefaction in hydrological effects, considering that it can be one of the mechanisms that contributes to the increase in stream flow by means of the expulsion of water from compaction of unconsolidated deposits.The maximum distance to which liquefaction occurs is consistent with the maximum distance to which increase of streamflow has reported (Manga, 2001;Montgomery et al., 2003).Chiodo et al., 1999;Fantucci and Sorriso-Valvo, 1999).After a careful revision of each quoted effect we stored a dataset of 117 deeply reviewed observations at 73 different localities (Fig. 2 and Table 1).The collection concerned two main groups of observations: ground effects (failures, landslides, cracks) and hydrological changes (flow variations, physical and chemical variations, liquefactions).

Ground effects
We collected useful data for forty-two localities where one or more ground failures (landslides, rock falls, lateral spreads in soil or simply cracks) occurred (Fig. 3 and Table 1).Unfortunately, we were unable to estimate the volume of the generated landslides, although we know from the chronicles that several of them were the main cause of damage and victims in some villages (i.e.Martirano, Ajello, Fitili, Parghelia; Fig. 2 and Table 1).Most of those landslides can be classified as coherent slides, few as disrupted slides and falls (sensu Keefer, 1984); occasionally the documents described reactivated landslides.In addition, chronicles reported many slope-parallel ground cracks, that can often be interpreted as incipient landslides.None of the recognized ground failures has been directly related to primary surface faulting (Mercalli, 1906;Galli and Bosi, 2002;Cucci and Tertulliani, 2006), even if the magnitude of the 1905 earthquake is around 7, so capable of breaking up through the surface.The 90% of cracks and landslides occurred within 60 km from the locations B and C, and within 90 km from location A (Fig. 3 and Table 1).

Hydrological changes
We collected forty-seven reports of localities throughout Calabria (Fig. 4 and Table 1) where one or more observations of hydrological changes occurred following the earthquake (Mercalli, 1906;Rizzo, 1907;Galli, 2000).Most of the data concerned excess flow in streams and springs; less frequently, a flow decrease or spring disappearance was reported.In some cases, we had notices of localities where the coeval accounts reported unspecific flow variations.These variations were sometimes accompanied by changes in the physical characteristics of the waters such as rise of their temperature and turbidity.The six localities where a temperature increase was reported fall into river basins with increase of flow.Liquefaction was observed in several localities, mostly where the same phenomenon was already noticed during past earthquakes.Hydrological changes were observed up to 185, 162 and 148 km from epicentres A, B, and C, respectively (Fig. 4 and Table 1).As expected, hydrological changes occurred at longer distances than other environmental signatures.Indeed 90% of data was distributed within 140, 120, and 100 km from epicentres A, B, and C, respectively.

Magnitude of the 190event from environmental effects
In order to constrain the size of the earthquake on the basis of the above reported characteristics of the environmental effects, we used a number of empirical relations between seismic source parameters and distribution of ground effects and hydrological changes.In particular, in this section we evaluated the total area affected by seismically-induced landslides, or put in relation the maximum distance of each kind of earthquake-induced effect from the three alternative locations A, B, and C.

Magnitude derived from ground effects
We compared our data to the database of the pioneer work of Keefer (1984), who elaborated relations between magnitude and distribution of earthquake-induced landslides.The total area affected by landslides following the 1905 earthquake is ∼6000 km 2 : using the empirical relations we obtained a magnitude M=6.7 (Fig. 5).This is a particularly important value because the area affected by landslides usually shows a strong correlation with magnitude (Keefer, 1984).However, we want to stress that the M=6.7 estimate is a minimum value because of the possibility of an offshore epicenter, of the narrow shape of the Calabria peninsula, and because it is obtained on the upper-bound fit of the Keefer's data (Fig. 5).
Using the relation between earthquake magnitude and maximum epicentral distance of landslides, relative to the three inferred locations A, B, and C, we found mean values of magnitude (weighted means as a function of the number of observations) M=6.71, M=6.44 and M=6.39 for the three major categories of landslides (i.e.coherent slides, disrupted slides and lateral spreads) described by Keefer (1984) (Fig. 6).

Magnitude derived from hydrological effects
To derive a magnitude value from this environmental effect we referred to Montgomery and Manga (2003), who found that the maximum distance to which seismically-induced hydrological changes have been reported is related to the earthquake magnitude.In particular, these authors showed that the limiting distance to which liquefaction is observed also defines the upper limit to the envelope of reported streamflow responses to earthquakes.Starting from this key-point, we selected from Table 1 the maximum distance of the locality where concurrent increase of streamflow and liquefaction Fig. 5. Plot of the area affected by landslides (km 2 ) as a function of earthquake magnitude (after Keefer, 1984).Landslides following the 1905 event covered a total area of ∼6000 km 2 , which corresponds to a magnitude M=6.7.Fig. 6.Upper bound curves of the relation between earthquake magnitude and maximum distance from the epicenter of the landslide distribution (after Keefer, 1984).Mean values of magnitude M=6.71, M=6.44 and M=6.39 result from the intersection of the distances from the locations A, B, and C with the three major categories of landslide.Fig. 7. Plot of the distance from epicenter versus earthquake magnitude for sites that exhibited seismically-induced streamflow response (after Montgomery and Manga, 2003).The solid line represents the empirical limit to the distance from the epicenter, beyond which liquefaction has been observed.Following the 1905 event, concurrent increase of streamflow and liquefaction was testified up to 104 km, 81 km and 67 km from the locations A, B, and C, respectively.This corresponds to M=6.7, M=6.6 and M=6.5 at the three potential epicenters.
was observed, and obtained M=6.7, M=6.6 and M=6.5 at the three potential epicenters A, B, and C, respectively (Fig. 7).We want to remark that also in this case these values are conservative, as they correspond to the minimum magnitude for an earthquake to produce liquefaction at a given distance, and also because liquefaction is not the exclusive cause of flow increasing.

Summary on the magnitudes derived from environmental effects
Magnitudes obtained both from ground and from hydrological effects are quite comparable.Averaging the values derived from the relations between magnitude, area affected by landslides, maximum distance of landslides and maximum distance of liquefaction, we obtain mean magnitudes M=6.70, M=6.58 and M=6.53 at the three locations A, B, and C, respectively.Such magnitudes are placed halfway between the extreme bounds proposed by Westaway (1992, M≤6.2) and Duda (1965, M s =7.9) and fully comparable to the M s =6.8 suggested by Abe and Noguchi (1983).In particular, we want to focus on the constant M=6.7 value obtained for location A from all the studied environmental effects.Therefore, for the 1905 event we define a minimum magnitude M=6.7 based on the empirical relations, and favour the epicenter A on the base of the consistency of the magnitudes obtained at this location.

Hydrological data and models of coseismic strain
In order to provide further constraints to the location and the geometry of the event source, we calculated the coseismic field of deformation produced by a number of individual sources potentially associated with the three locations investigated and compared it to the experimental data of streamflow changes.Our aim was to identify a preferred source, or alternatively to rule out some of the epicenters proposed by previous Authors.We assumed that the observed streamflow variations are the hydrological response to coseismic strain changes (Muir-Wood and King, 1993).Several studies demonstrate that the polarities of calculated deformation are in agreement with those of the observed hydrological effects so that extensional strain produces a discharge fall and compressive strain a discharge rise, especially in the near field (Grecksch et al., 1999;Ge and Stover, 2000;Lee et al., 2002;Montgomery and Manga, 2003;Caporali et al., 2005).
However, in entering into a fall season of higher rainfall, the signature of stream flow associated with the earthquake can be perceptibly masked.For hydrological observations to be employed for this purpose it must be possible to show that, what is being observed is a consequence of seismic strain changes affecting crustal porosity.So we considered the rainfall in the days and weeks before and after the earthquake and found that the dry season prolonged until 20 September (Annali dell'Ufficio di Metereologia e Geodinamica Italiano, 1908).As the reports about stream flow changes referred to the first days following the event, we are confident that the hydrological signatures reflect coseismic strain.
Therefore, using the empirical relationships between magnitude and fault parameters by Wells and Coppersmith (1994) we deduced, from the mean magnitudes inferred through environmental effects, the fault area and the seismic slip of a number of generic sources at the three epicenters.Finally, we assigned a direction to each generic fault on the base of different evidences such as inversion of instrumental and intensity data (Boschi et al., 1995;Cucci and Tertulliani, 2006;Michelini et al., 2006), seismotectonic indications (Galli and Bosi, 2002;Piatanesi and Tinti, 2002;Neri et al., 2003;Tortorici et al., 2003), batymetric data and offshore profiles (Trincardi et al., 1987;Argnani and Trincardi, 1988;Gamberi and Marani, 2007).
At the end of this process we obtained five potential sources, that we summarize in Table 2 and briefly describe in the following: -Western Offshore Fault (WO): it is associated with epicenter A, proposed as new hypocentral location of the 1905 event by Michelini et al. (2006).
-Macroseismic Fault North (MFN) and Macroseismic Fault South (MFS): they are derived from inversion of intensity data (Boxer code by Gasperini et al., 1999) and are associated with epicentral location B.
-Capo Vaticano Fault (CV): this fault has been used by Piatanesi and Tinti (2002) to model the tsunami waves generated by the earthquake.It is associated with epicenter C.
-Coccorino Fault (CC): this fault was selected it on the base of geological constraints (Tortorici et al., 2003;Cucci and Tertulliani, 2006), so that it is not strictly ssociated with any of the three epicentral locations.
For each source, we calculated the pattern of coseismic static strain expected for different geometries and styles of faulting.Calculations of the strain at the surface were made in an elastic halfspace with uniform isotropic elastic properties following Okada (1992) and using COULOMB 3.0 (Lin and Stein, 2004;Toda et al., 2005).The output of our calculations were plots of volumetric strain at the free surface on 60 individual faults (Fig. 8a-e and Table 3), obtained by crossing the five potential sources with two different dips (60 • and 80 • ), two different depths of the top edge of the fault (0 and 5 km) and three different rakes (231 • normal fault with rightlateral component/270 • pure normal fault/309 • normal fault with left-lateral component).We also tried the same calculations on the 500-m-deep surface, but the strain patterns were almost indistinguishable from the 0-m results.
To find out which of the above cited sources better matches the observed streamflow changes we selected only the solutions that showed: 1) within the distance of two fault-lengths, more than 50% of points with polarities of the observed hydrological effects in agreement with the expected deformation, and 2) percentage of consistent polarities at a distance of one fault-length higher than the percentage at two faultlengths.
We comprised in the calculations only the localities up to a distance of two fault-lengths (in our case up to 46-58 km, see Table 2) from the center of the source, i.e. in the region where the most significant hydrological changes should have occurred and where there is the highest chance to discriminate between models.
The results of this elaboration are depicted in Table 3; seven individual faults out of 60 satisfied both the conditions above described.A comparison between the seven preferred faults and the other solutions provided some interesting insights.Within two-faults distance, none of the solutions generated by the Capo Vaticano Fault (CV, Fig. 8e) showed a deformation pattern consistent with the observed one.The same poor fit was exhibited by the whole family of solutions with right-lateral component, independently from other parameters like the dip of the fault or the depth of its upper side.At one-fault distance, also the solutions associated with the Macroseismic Fault South (MFS, Fig. 8b) showed a low percentage of consistent polarities.Moreover, the good fit faults seemed to be greatly influenced by the rake (six solutions are with left-lateral component), partially by the position of the top edge of the fault (five solutions), and scarcely by the dip.

Discussion
Our dataset consisted of 117 observations at 73 different localities diffused over the Calabria territory.Data concerned a great number of notices of ground effects (landslides, rock falls and lateral spreads) and of hydrological changes (streamflow variations, liquefaction, rise of water temperature and turbidity).The analysis of these data provided i) sound evidence to assess the size of the 1905 earthquake, ii) several observations to reduce the uncertainties on its location, iii) some interesting clues to better define the geometry of the causative fault.i) Evidence on magnitude: the empirical relations that we used to evaluate the size of the 1905 event from ground effects and hydrological changes concurred to individuate a magnitude between 6.5 and 7.0.We emphasize the M=6.7 value, which is calculated on the total area affected by landslides, a parameter that strongly correlates to the size of the event.We underline that this is a minimum value, because of the possibility of an offshore location, of the shape of the Calabria region, and of the limits of the Keefer's curve.Similarly, magnitudes between M=6.5 and M=6.7 (depending on the different location) were calculated by hydrological effects; we consider also these estimates as conservative.In summary, values of magnitude obtained from environmental effects were comparable and homogeneous.Taking into account the above described conservative constraints, we indicate M=6.7 as the most likely minimum magnitude of the 1905 event.
ii) Observations on location: we checked three potential target areas, identified as alternative epicenters by previous studies and indicated as A, B, and C in Fig. 1.From the empirical relations between magnitude and distribution of environmental effects we found that epicenter A is preferred because of the consistency of the magnitudes obtained at this location.As for the hydrological effects, we assumed that the streamflow variations are the hydrological response to coseismic strain changes (Muir-Wood and King, 1993), and compared the polarities of the observations (charge and A. Tertulliani and L. Cucci: Identification of a seismogenic source from the environmental effects Solutions like WOL1 satisfy both the condition 1 (within two-faults distance, more than 50% of points with polarities of the observed hydrological effects in agreement with the expected deformation) and the condition 2 (percentage of consistent polarities at one-fault distance higher than the percentage at two-faults distance).Solutions like MFS1 satisfy only condition 1. Solutions like WO1 are discarded.
discharge) to the modeled coseismic static strain.Also in this case epicenter A showed a good agreement with the observed streamflow changes (Fig. 8d).On the contrary, poor fit was obtained for the epicenter C (Fig. 8e): we believe that this was the less probable location for the 1905 earthquake.Epicenter B showed a good agreement with the hydrological effects polarities, too (Fig. 8a and b); however, this localization displayed a higher uncertainty as it derived from the inversion of intensity data rather poorly distributed along the coastline.In addition, the M=6.58 mean magnitude from environmental effects associated with this location was slightly lower than the M=6.7 proposed magnitude.We conclude that A is the most constrained epicentral location of the 1905 event.
iii) Clues on geometry: the comparison of the observed hydrological changes with the expected coseismic static strain calculated on 60 individual faults allowed us to discard all the solutions with right-lateral component.On the contrary, two groups of faults were consistent with the polarities from the observations: the first belonged to a family of 100 • oriented faults, (CC and WOL, Fig. 8c-d and Table 3), associated with the A offshore epicenter.All these faults displayed a component of left-lateral slip.The second belonged to a group of 260 • oriented faults (one pure normal and two normal with left-lateral component solutions, MFN, Fig. 8a and Table 3) associated with the B inland epicenter.Furthermore, whilst the change of the fault dip did not visibly affect the pattern of expected strain, an important indication was that five preferred solutions out of seven derived from models of fault with depth of the top at 5 km; this is consistent with the generally accepted lack of any primary surface faulting produced by the event.

Conclusions
In this study, we significantly reduced the uncertainties concerning the size, location and geometry of the 1905 earthquake by using the environmental effects produced by the event.Summarizing our main conclusions in the light of the above discussed issues, we associate all the observed environmental changes produced by the 1905 earthquake with a M≥6.7 event, most likely occurred in an offshore area west of Capo Vaticano (hence west of the location proposed by the historical seismic Catalogs) and possibly generated by a 100 • N oriented normal fault with a left-lateral component.We support this characterization of the event source because: i) it shows the same minimum M=6.7, constantly obtained at location A from all the studied environmental effects; ii) it is highly consistent with the general pattern of the environmental effects observed following the earthquake (maximum distance landslides-epicenter, good fit between calculated strain and observed hydrological changes); iii) it is consistent with the lack of primary surface faulting associated with the event; iv) it may reasonably explain the moderate tsunami produced by the earthquake; v) it is consistent with the orientation of the extension acting in this area, which strikes NW-SE (Montone et al., 1999(Montone et al., , 2004; see also inset in Fig. 1).In particular, a 100 • N oriented normal fault with left-lateral component, located offshore nerby Capo Vaticano, could act as a regional transversal tectonic lineament that transfers the extensional deformation from the NE-trending structures in the Tyrrhenian northern offshore to the NE-trending faults that are reported inland, south of Capo Vaticano (Valensise and Pantosti, 2001;Galli and Bosi, 2002).
This paper shows the use of "macroscopic evidences", such as hydrological changes and ground failures, to infer the causative source of an earthquake.We suggest that the approach proposed provides tools to widen the knowledge of historical earthquakes for which instrumental data are contradictory or lacking, and to simulate the analysis of earthquakes occurred in non-urbanized areas.

Fig. 2 .
Fig. 2. General map of the environmental effects produced by the 8 September 1905 earthquake.Black dots indicate the 73 localities where one or more environmental effects were observed.The three alternative locations in study are indicated by triangles.The inset shows a blow-up of the area of Capo Vaticano.

Fig. 3 .
Fig. 3. Sites of ground effects.Triangles indicate the three locations in study A, B, and C.

Fig. 4 .
Fig. 4. Sites of hydrological effects.Triangles indicate the three locations in study A, B, and C.

Fig. 8 .
Fig. 8.Comparison between calculated volumetric strain distributions and observed hydrological effects produced by the 1905 earthquake.Plots of surface static volumetric strain along five potential sources are calculated for different geometries and styles of faulting (the abbreviation of the individual fault is shown in the upper right side of each plot, refer to Table 3 for fault parameters); blue shading indicates areas in compression, red shading areas in dilatation.Units: 10 −6 .A red rectangle indicates the surface projection of the fault plane; a green line is the intersection of the updip projection of the fault with the surface.Streamflow changes are indicated by circles (black/discharge increase; white/discharge decrease; grey/unspecified change; red/other effect).An increase of discharge is expected in compressional areas, a streamflow decrease in dilatational areas.

Fig. 8 .
Fig. 8.Comparison between calculated volumetric strain distributions and observed hydrological effects produced by the 1905 earthquake.Plots of surface static volumetric strain along five potential sources are calculated for different geometries and styles of faulting (the abbreviation of the individual fault is shown in the upper right side of each plot, refer to Table 3 for fault parameters); blue shading indicates areas in compression, red shading areas in dilatation.Units: 10 −6 .A red rectangle indicates the surface projection of the fault plane; a green line is the intersection of the updip projection of the fault with the surface.Streamflow changes are indicated by circles (black/discharge increase; white/discharge decrease; grey/unspecified change; red/other effect).An increase of discharge is expected in compressional areas, a streamflow decrease in dilatational areas.

Table 1 .
List of the environment effects observed following the 1905 earthquake.

Table 2 .
List of the five potential sources.