ROSE SCHOOL FEASIBILITY EVALUATION OF AN EARLY WARNING SYSTEM IN THE LOMBARDY REGION (NORTHERN ITALY)

Transcription

1 Istituto Universitario di Studi Superiori Università degli Studi di Pavia EUROPEAN SCHOOL FOR ADVANCED STUDIES IN REDUCTION OF SEISMIC RISK ROSE SCHOOL FEASIBILITY EVALUATION OF AN EARLY WARNING SYSTEM IN THE LOMBARDY REGION (NORTHERN ITALY) A Dissertation Submitted in Partial Fulfilment of the Requirements for the Master Degree in EARTHQUAKE ENGINEERING by PAOLA TRAVERSA Supervisor: Prof. Carlo Lai Dr. Claudio Luciano Strobbia August, 2006

2 The dissertation entitled Feasibility evaluation of an early warning system in the Lombardy Region (Northern Italy), by Paola Traversa, has been approved in partial fulfilment of the requirements for the Master Degree in Engineering Seismology. Dr. Carlo G. Lai Dr. Claudio Luciano Strobbia i

3 Abstract ABSTRACT Lombardy (northern Italy), being considered one of the engines of Italian economy, is one of the most important regions in Europe. It is therefore evident the relevance of a matter as security and preservation of existing sensitive structures, such as the large number of hospitals serving the area. Lombardy results to be a low-seismicity area, the security of constructions against earthquakes, therefore, should be more oriented at preserving sensitive equipment and safety of people than dedicated to prevention of structure collapsing. This work illustrates a feasibility study of an Earthquake Early Warning System (EEWS) in this Region. The aim is to issue the hospitals with a warning which have to be transmitted sufficiently ahead before the arrival of damaging ground shaking in order to allow the medical staff to start the shut-down operations of medical equipments and all the necessary security operations. Simulating earthquake scenarios represented by occurred events in northern Italy, the EEWS could be able to issue a warning time of few to tens of seconds in Lombardy, depending on the distance of the urban area from the seismic source. Once the earthquake s size has been determined in real time from P-waves characteristics, the warning time is function of the temporal separation between first arrival of P- and S-waves at the observation point. Most of the major centres are therefore likely to be alarmed sufficiently ahead. Being affected by medium-low magnitudes events, in Lombardy the effectiveness of an EEWS is strongly reduced: towns located at a distance allowing for relevant warning times would not be struck by damaging ground motion due to ground motion attenuation with distance, while struck by strong ground motion due to the proximity to the source, would not be alerted sufficiently ahead. Synthetic velocigrams are compared with real recordings of the event being simulated at seismic stations and attenuation of synthetic ground motion with distance is compared with suitable attenuation laws coming from literature. Nearly linear relationship can be defined between temporal separation of first arrival of P- and S-waves and epicentral distance The possibility of simplifying all the procedure by considering a geometrical model is considered too simplistic to reproduce properly the developing ground motion. Such a simplified model, however, allows for sensitivity analysis of the parameters. i

4 Abstract Keywords: early warning; ground motion simulation; seismic hazard; Lombardy Region; hazard mitigation; temporal separation; attenuation ii

5 Acknowledgements ACKNOWLEDGEMENTS I would like to greatly thank Claudio Strobbia for promoting ideas, assisting and solving jamming with Matlab code, providing information and making entertaining the job. Sincere thanks to Carlo Lai, for guiding all the research work, providing software products and finding information about real recordings of occurred earthquakes. Thanks to Mirko Corigliano for providing last versions of Hisada code and information about it. iii

6 Index TABLE OF CONTENTS Page ABSTRACT... i ACKNOWLEDGEMENTS... iii TABLE OF CONTENTS... iv LIST OF FIGURES... vii LIST OF TABLES... xv LIST OF SYMBOLS... xvi 1. INTRODUCTION Motivation for the study Context of application Research objectives Methodology Dissertation outline EARTHQUAKE EARLY WARNING SYSTEMS Background Literature review on Eathquake Early Warning system worldwide Objectives of hazard mitigation provided by EEWs s Perspectives and future developments of EWS s TECTONIC SETTINGS OF ITALY Overview Geological setting of Italy Northern Italy Appennines Southern Italy (Calabria and Sicily) A seismotectonic (kinematic) model iv

7 Index 3.4 Depth of the seismogenic layer Geophysical data and crust model Earthquake scenarios SEISMIC HAZARD IN NORTHERN ITALY Overview Seismogenic zoning Seismogenic sources Focal mechanisms Effective depth Earthquake catalogue Seismic hazard maps GENERATION OF SYNTHETIC SEISMOGRAMS Overview Characteristics of the simulation procedure Source process: a kinematic model Wave propagation Representation of the ground motion: near field and far field approximation Simulation code Comparisons of synthetic seismograms with real records GROUND SHAKING SCENARIOS Overview Reggio Emilia 1996 earthquake Seismogenic model Source characteristics and model parameters Synthetic seismograms Comparison with recorded data Spatial attenuation of synthetic PGV: comparison with attenuation relationships from the literature Temporal separation of P and S phases: scenario of pre-alarming times Comparison with a simple geometric model Sensitivity to the model parameters Comparison of synthetic and real pre-alarming times Salò (Brescia) 2004 earthquake Seismogenic model Source characteristics and model parameters Synthetic seismograms v

8 Index Comparison with recorded data Spatial attenuation of synthetic PGV: comparison with attenuation relationships from the literature Temporal separation of P and S phases: scenario of pre-alarming times Comparison of synthetic and real pre-alarming times Future earthquake scenario: re-activation of the North Giudicarie line Overview Seismogenic model Source characteristics and model parameters Synthetic seismograms Spatial attenuation of synthetic PGV: comparison with attenuation relationships from the literature Temporal separation of P and S phases: scenario of pre-alarming times CONCLUSIONS Conclusions and remarks Recommendations for future research REFERENCES APPENDIX A...Error! Bookmark not defined. vi

9 Index LIST OF FIGURES Page Figure 1.1. Location of the hospitals in the Lombardy region. Main road network is also showed...2 Figure 2.2. Ground surface motions: the three components of a seismogram are included, Z indicates the up-down vertical motion, N the north-south horizontal motion, E the eastwest horizontal motion, where the arrivals are more evident, the first arrival of P and S waves are indicated with the correspondent letter...7 Figure 2.2. Map of the principal towns of the Lombardia region with placemarks indicating the location of the epicentres of the major seismic events.(from Google Earth)... Figure 3.1. Structural-kinematic sketch of Italy and surrounding areas showing the traces of the slip vectors of Africa versus Europe (according to Livermore and Smith, 1985) and Adria versus Europe (according to the rotation pole proposed by Meletti et al 2000). Figure 3.2. Structural sketch of Italy and surrounding regions. 1 trhust, 2 normal faults, 3 strike slip faults, 4 Pliocene-Quaternary anticlines, 5 Pliocene-Quaternary synclines, 6 backarc basin floored by oceanic crust. Figure 3.3. CMT (Harvard) and RCMT (Pondrelli et al. 2002) solutions for the >4.5 seismicity since The extension along the Appennines belt and the compression around the Adria lithosphere and in the Northern Sicily offshore are evident. [Chiarabba et al. 2005]. Figure 3.4. Tectonic scheme summarizing seismological data (Chiarabba et al 2005) and available kinematic constraint (Devoti et al. 2002; Hollestein et al. 2003; Battaglia et al. 2004). Arrows are simplified directions of GPS sites relatives to a stable European plate. [Chiarabba et al. 2005]. vii

10 Index Figure 3.5. Sketch map of main tectonic features of Italy simplified from Bigi et al. [1990]. CMTs for great earthquakes that occurred between 1976 and 1998 are shown. (a) thrust fault (pre-middle Pliocene); (b) thrust fault (middle Pliocene-Recent); (c) normal fault; (d) strike slip fault; (e) undetermined fault. Figure 3.6. E-W vertical sections of seismicity across the western Alps. Events falling ± 20 km from the section are plotted. The geometry of the European and Ligurian Moho taken from available seismic reflection data are shown. [Chiarabba et al. 2005]. Figure 3.7. Schematic lithospheric section across the northen Appenninic arc [Meletti et al. 1995]. Figure 3.8. Summary of the stress data of the Italian peninsula. The main stress provinces delineated by the data are sketched. The structural arcs with shaded triangles indicated the active oceanic subduction; open triangles indicate the active compressional fronts;solid triangles show active oceanic subduction; open triangles delineate the location of Plio-Pleistocene trust front, presently affected by prevalent extension. [Montone et al. 1999]. Figure 3.9. Vertical sections of seismicity across the northern Appennines arc. Events falling ± 20 km from the section are plotted. The lines indicate the geometry of the Adriatic and European Moho, as suggested by seismic reflection data and earthquake hypocenters. The bold lines show a simplified sketch of the main faults in the crust. [Chiarabba et al. 2005]. Figure 4.1. Seismic zonation ZS9. Each zone is defined by a number; zones identified by a letter are not used for the evaluation of the seismic hazard. Black boundaries between zones are drown only on the base of tectonic or geological and structural information. Blue boundaries separate zones of similar deforming style, but with different seismic characteristics. [INGV Mappa di pericolosità sismica 2004]. Figure 4.2. Seismic zonation ZS9, focus on northern Italy. [INGV Mappa di pericolosità sismica 2004] Figure 4.3 ZS9 Seismic Zonation of northern Italy (black borders) [INGV 2004]. Figure 4.4. Focal mechanisms of ZS9 seismic zonation, from the database pubblished by Vannucci and Gasperini [2003]. Dimension of symbols is proportional to the logarithm, of the overall released seismic moment [INGV 2004, Zonazione sismogenetica ZS9]. Figure 4.5. Main fault mechanism expected for the seismic zones of ZS9 [INGV 2004, Zonazione sismogenetica ZS9]. viii

11 Index Figure 4.6. Frequency distribution of the earthquake depths of seismic zone ZS 911 [INGV 2004, Zonazione sismogenetica ZS9] Figure 4.7 Zoom on the central part of northern Italy of the Italian seismic catalogue. Figure 4.8. Hypocemtral distribution of about selected events. Colour scale, continuously varying, indicates the depth of events (blue colours for the crustal seismicity and red colours for the mantle seismicity). The different size of circles is given by the magnitude scale indicated on the lower right corner. [Chiarabba et al 2005]. Figure 4.9. Map of seismic sources determined from intensity data (geological evidence was not available) [Meletti et al. 2000]. Figure Map of CPTI catalog events with M 5.5 [Meletti et al. 2000]. Figure Seismogenic source in northern Italy [DISS 3.0, 2006]. Figure Map of Italian seismic hazard expressed in terms of maximum acceleration with 10% in 50 years of probability to be exceeded for stiff soil (Vs > 800 m/s; cat.a, punto del 30 D.M ). Reference: Ordinanza PCM del 28 aprile 2006 n.3519, All.1b. [INGV 2006]. Figure Maximum observed intensities from 1000 to 1992 [Camassi et al. 1999]. Figure 5.1. Ground motion evaluation: illustration of the three processes which characterize the generation and the propagation of seismic waves [Stewart et al. 2001]. Figure 5.2. Sketch for the spatial definition of the fault plane and of the slip direction. Figure 5.3. Sketch representing Green s functions (from Faccioli [2005]) Figure 6.1. (a) Simplified seismotectonic setting of northern Italy. Red lines are the thrust fronts of the Po Plain (PTS, Pede-apenninic thrust front; MTF, Monferrato thrust front; EFTF, Emilia±Ferrara±Romagna folded arcs). Small circles are instrumental seismicity of the past 15 yr recorded by RSNC and empty squares are historical seismicity (from Boschi et al. 1995). Thick bars are the P-axis directions of the Parma 1971, Parma 1983 and Reggio Emilia 1996 earthquakes and the Caorso (Ca) and Cesena±Forli (Fo) seismic sequences. Thin lines are traces of sections in Fig. 2. Filled squares are the main towns of the region (Bo, Bologna; Fe, Ferrara; RE, Reggio Emilia; Pr, Parma). From Selvaggi et al. [2001] Figure 6.2. Subsurface geological structure interpreted from active seismic data (modifed from Pieri & Groppi 1981). From Selvaggi et al [2001]. Figure 6.3. Horizontal (N-S) component of the velocigrams at different observation points. Figure 6.4. Horizontal (E-W) component of the velocigrams at different observation points. ix

12 Index Figure 6.5. Vertical (U-D) component of the velocigrams at different observation points. Figure 6.6. Directivity effect on the synthetic velocigrams. Figure 6.7. Map of the area showing the location of the epicentres and of the seismic stations; focal mechanisms for the main shock and the aftershocks are reported. Grey triangles indicate the recording sites of the AGIP network, while the bigger square represents the borehole station. (From Selvaggi et al. 2001). Figure 6.8. Seismic signals on the N-S and the E-W components recorded by the borehole station for the main shock. Amplitude units are m/s. (From Selvaggi et al 2001). Figure 6.9. Synthetic velocigrams on the N-S, the E-W and the U-D components simulated by Hisada [1996] code at the borehole location for the main shoch. Figure Attenuation of the ground motion in terms of acceleration, a) vertical component of synthetic acceleration compared with the relationship of Ambraseys [1996]; b) horizontal components of synthetic acceleration compared with the relationships of Sabetta and Pugliese [1996] and of Ambrasyes [1996]. Figure Comparison between the attenuation of the synthetic velocities with the attenuation laws of Sabetta and Pugliese [1996] and Ambraseys [1996]. Figure Attenuation of synthetic velocity and acceleration with respect to direction (the angles represent azimuth with respect to the north); horizontal N-S component. Figure Attenuation of synthetic velocity and acceleration with respect to direction (the angles represent azimuth with respect to the north); horizontal E-W component. Figure Attenuation of synthetic velocity and acceleration with respect to direction (the angles represent azimuth with respect to the north); vertical up-down component. Figure Picking of first P-wave (in pink) and S-wave (in red) arrival on the vertical updown component of ground velocity. Figure Time separation between P-wave and S-wave arrivals at different locations as function of the distance of the observation point from the epicenter. Figure Relation between the time separation between arrival of P and S waves at a given point and the ground shaking level suffered by the same location. Figure Relation between the time separation between arrival of P and S waves at a given point and the ground shaking level suffered by the same location. Figure Variation of temporal separation between P-wave and S-wave arrivals for different moment magnitude values. x

13 Index Figure Sketch illustrating a ray path travelling from a seismic source to the ground surface in a layered medium. Figure Geometric model: travel times of P- and S- waves from source to ground surface as function of the epicentral distance. Figure Comparison of temporal separation between P- and S-waves first arrivals as function of the distance from the epicenter obtained by rigorous simulation and by simplified geometric model. Figure All Temporal separations between P- and S-waves first arrivals at given observation points deployed on the ground surface that can be obtained varying the main parameters in the geometric model. Figure Set of the whole parameter combinations used in the geometric model for the sensitivity study. In blue all the combinations, in red the ten fastest and in green the ten slowest. Figure Near field: correlation between temporal separation and P-wave velocity in the layered media. Figure Near field: correlation between temporal separation and S-wave velocity in the layered media. Figure Far field: correlation between temporal separation and P-wave velocity in the layered media. Figure Far field: correlation between temporal separation and S-wave velocity in the layered media. Figure Near field: correlation between temporal separation and S- (in blue) and P-wave (in red) velocity in the layered media. Figure Far field: correlation between temporal separation and S- (in blue) and P-wave (in red) velocity in the layered media. Figure Near field, relationship between the temporal separation between P- and S-waves cumulated in the two thickest layers (fourth and fifth) and the temporal separation between P- and S-wave first arrivals at an observation point located in proximity of the epicenter on the ground surface. Figure Far field, relationship between the velocity difference of P- and S-waves in the bedrock and the temporal separation between P- and S-wave first arrivals at an observation point located 142 km away from the epicenter on the ground surface. xi

14 Index Figure Montecarlo simulation: bar distributions of the variable parameters for each layer. Figure Seismotectonic sketch of the Garda Lake area (from Pessina et al. [2006]). Figure Map of the area including instrumental epicenter of the November 2004 Salò earthquake, focal mechanisms and field stations where observations were collected [Esposito et al. 2005]. Figure Synthetic velocigrams (N-S horizontal component) from the epicenter (Salò) to Milan (142 km far from the epicenter). Figure Synthetic velocigrams (E-W horizontal component) from the epicenter (Salò) to Milan (142 km far from the epicenter). Figure Synthetic velocigrams (U-D vertical component) from the epicenter (Salò) to Milan (142 km far from the epicenter). Figure Directivity effect on synthetic ground motion. Figure 6.40 Acceleration data recorded at Vallio Terme, (13.3 km epicentral distance) during the November 2004 Salò earthquake. (From Pessina et al. 2006). Figure Synthetic accelerogram, horizontal N-S component. Figure Recorded data at Salò Scuola station during the aftershocks of the 2004 Salò earthquake. Figure Velocigrams calculated by integration from recorded data at Salò Scuola station during the aftershocks of the November 2004 Salò earthquake. Figure Synthetic velocigrams obtained by simulation at Salò Scuola station due to the aftershocks of the November 2004 Salò earthquake. Figure Attenuation of peak ground acceleration (vertical and horizontal components) as function of the distance from the epicenter, comparison with relationships from literature. Figure Attenuation of peak ground velocity (horizontal components) as function of the distance from the epicenter, comparison with relationships from literature. Figure Attenuation of peak ground velocity and acceleration (respectively left side and right side) for the N-S horizontal component with respect to the distance from the epicenter for different directions (azimuth are referred to the north). Figure Attenuation of peak ground velocity and acceleration (respectively left side and right side) for the E-W horizontal component with respect to the distance from the epicenter for different directions (azimuth are referred to the north). xii

15 Index Figure Attenuation of peak ground velocity and acceleration (respectively left side and right side) for the U-D vertical component with respect to the distance from the epicenter for different directions (azimuth are referring to the north). Figure Picking of P and S waves first arrivals. Figure Temporal separation between P-wave and S-wave first arrivals as function of the distance from the epicenter. Figure P- and S-wave temporal separation as function of PGA at observation points. Figure P- and S-wave temporal separation as function of PGV at observation points. Figure Realtionship between temporal separation of P and S wave first arrivals and PGV as function of different earthquake magnitudes. Lines of the same colour refer to events represented by standard deviations of 1 or 2. Figure Simplified tectonic map of the central southern Alps and of the Austroalpine nappes west of the Tauern Window (after Thöni [1981], Castellarin and Vai [1982], Rossi and Rogledi [1988], Schmid and Haas [1989], and Beserzio and Fornaciari [1994]). Bold lines represent the main tertiary faults of the central Alpine chain. The shear sense of these faults has been reported following Martin et al. [1991] and Schmid and Foitzheim [1993]. From Prosser [1998]. Figure Velocigrams, horizontal N-S component. Figure Velocigrams, horizontal E-W component. Figure Velocigrams, vertical U-D component. Figure Attenuation of PGA as getting further away from the epicenter. Comparison of synthetic results with Sabetta and Pugliese [1996] and Ambraseys [1996] attenuation laws. Figure Attenuation of PGV as getting further away from the epicenter. Comparison of synthetic results with Sabetta and Pugliese [1996] and Ambraseys [1996] attenuation laws. Figure 6.61 Picking of first arrivals of P- and S-waves at given observation points. In pink we pointed out P-waves first arrivals and in red S-waves first arrivals. Figure Relationship between temporal separation between P- and S-waves first arrivals and distance from the epicenter. Figure Relation between the temporal separation between P- and S-wave first arrivals at given locations and the PGA at the same point. xiii

16 Index Figure Relation between the temporal separation between P- and S-wave first arrivals at given locations and the PGV at the same point. Figure Relationship between temporal separation between P- and S-waves first arrivals at given location points and PGV at the same points, for different magnitudes (considering an interval of one standard deviation). xiv

17 Index LIST OF TABLES Page Table 1.1. Seismic depth of the considered zones and number of events of different magnitudes...34 Table 4.1. Seismic depth of the considered zones and number of events of different magnitudes Table 6.1. Parameters of the crust model used for the simulation of the 1996 Reggio Emilia earthquake Table 6.2. Distance, main instants and warning times issued to the major towns in Lombardy after the first recording at the reference station. Table 6.3. Parameters of the crust model used for the simulation Table 6.4. Characteristics of the data recorded at Salò Scuola station during aftershocks Table 6.5. Distance, main instants and warning times issued to the major towns in Lombardy after the first recording at the reference station. Table 6.6. Main characteristics of the Giudicarie fault system (from DISS 3.0 [2006]) Table 6.7. Parameters of the crust model used for the simulation Table 6.8. Distance, main instants and warning times issued to the major towns in Lombardy after the first recording at the reference station. xv

18 Index LIST OF SYMBOLS Vp Vs Mw L W log L R A T V i i D D e = P-wave velocity = S-wave velocity = Moment magnitude = Fault length = Fault width = Logarithm of standard deviation = Rupture length = Rupture area = Temporal separation between P and S wave first arrivals = Travel time velocity in the layer i = Incidence angle in the layer i = Maximum surface displacement = Distance of the point from the epicenter xvi

19 Chapter 1. Introduction 1. INTRODUCTION 1.1 Motivation for the study The security of existent structures is quite a relevant issue in Italy, due to its social and economical importance. In fact, maybe due to the fact that the Italian urban planning has tended mostly to keep unvaried the urban landscape, a process of building renewal has been often prevented. Consequently the level of Italian constructions is frequently of low-quality, problem that can also be blamed on the practice of unauthorized building that affect some places of the Peninsula. The problem of security is particularly relevant also in the case of strategic buildings, which belong to risk category due to their age of construction and/or to their collocation in the urban fabric. The aim of this work is to study the feasibility of an Earthquake Early Warning System (EEWS) in the Lombardy Region, located in north-western Italy, with km 2 area and inhabitants population. Lombardy is one of the engines of the global economy with a GDP calculated by ISTAR at 400 billion. The region is one of the three richest regions in Europe, with a per capita gross domestic product that is 50 percent higher than the rest of Italy. In fact the latest Eurostat figures show that Lombardy in 2003 had the highest GDP for a region in the whole of the EU. Many foreign and national companies in fact, have their headquarters in Milan. The idea of setting up of an Earthquake Early Warning System is focused on the preservation of existent strategic constructions, due to the necessity of managing in an optimal way first aid and interventions in areas struck by seismic events. This could be achieved, on one side by protecting buildings and strategic facilities and, on the other side, by fast transmission of the conditions of the region hit by the seism and of the damages suffered by the urban and infrastructural fabric. As shown in figure 1.1, Lombardy region is served by 209 hospitals. In fact, in particular, the objective of this work is trying to investigate the possibility to issue a reasonable warning to the hospitals in order to start the shut-down operations of medical equipments and all the necessary security operations. 1.2 Context of application As mentioned before, our study is focused on the Lombardy Region, in northern Italy, which is, as we will show in the following chapters, a low seismicity zone, that is with infrequent 1

20 Chapter 1. Introduction and low magnitudes events. This actually represents a drawback in designing an EEWS since centres located at short distances from the seismic source would feel high level damaging shaking, but the short distance from the epicenter would represent a disadvantage in issuing a satisfactory warning time due to, as we will explain in the following chapters, the little temporal separation between compressional and secondary waves. On the other hand, centres located further away from the source could be issued with a good warning, but in these cases, the ground shaking level striking sensitive buildings would not be actually damaging due to the attenuation of seismic waves during the propagation in the top crust. Figure 1.1. Location of the hospitals in the Lombardy region. Main road network is also showed 1.3 Research objectives The main research objectives that have been fulfilled in this study are the followings: - Make an overview of the existing Earthquake Early Warning Systems worldwide. - Describe the tectonic and geological setting of Italy, particularly focusing on the northern part and define a suitable crust model from geophysical data. 2

21 Chapter 1. Introduction - Identify the seismic hazard in northern Italy, defining the main characteristics of local seismic zones in terms of focal mechanisms, maximum magnitudes, intensities, and effective seismic depth. - Obtain information about past events in Lombardy region by reviewing historical and instrumental catalogues. - Select real and representative earthquake scenarios from existing literature which effects struck Lombardy territory. - Generate synthetic seismograms from the selected earthquake scenarios at different observation points, spread out over the Regional Territory, by means of a simulation program and comparison with existing recorded data. - Study of the simulated seismic waves behaviour in terms of spatial attenuation of Peak Ground Velocity (PGV) and comparison with attenuation relationships from the literature - Evaluation of the temporal separation between first arrival of P and S waves at selected location points, estimation of the warning time that could be issued to the main towns and study of the effectiveness of the pre-alert by considering the relationship between ground motion level and first arrival time difference at the same observation points. - Demonstrate why a simplified geometric model is not able to correctly reproduce the times and amplitudes of ground motion generated at observation points, but it represents a good approximation for estimating the temporal separation between P- and S- waves first arrivals at given observation points. - Individuate which are the parameters with the strongest influence on the synthetic ground motion generated at given observation points, study the sensitivity of the model to those parameters. - Evaluate the possibility of re-activation of the Giudicarie lineament (northern Italy) and simulation of ground effects generated by this virtual scenario with the aim of testing the EEWS on a future possible earthquake scenario. 1.4 Methodology The idea of Earthquake Early Warning Systems is quite recent, general definitions and concepts on EEWS were recalled by specific articles on the topic, both analysing the reasonableness of the idea and the tests made to evaluate the reliability of such a system. Information about the tectonic setting of Italy and in particular about the northern part was obtained from literature review. After considering the Italian global tectonic setting, we concentrated our attention to the northern part of the Peninsula. In this case we introduced a simplification necessary for applying the analysis in such a low seismicity region. We therefore reported the study carried out by the INGV (Istituto Nazionale di Geofisica e 3

22 Chapter 1. Introduction Vulcanologia) [2004], leading to the definition of the Italian seismogenic zones (see Zonazione Sismogenetica ZS9, INGV [2004]). This allowed us to have an idea about the potential seismic sources whose generated ground shaking could affect the Lombardy Region. Where detailed literature could be found, those sources were individually considered in terms of main source and expected seism characteristics. We based our analysis on historical and/or instrumental seismicity data, obtaining information from the Italian seismic catalogue CPTI04, available online. This allowed for selecting and defining two representative earthquake scenarios, related to two representative historical well-known events. The choice was highly conditioned by the extremely poor information relating to the Italian seismicity currently available, and in particular to the zone we are referring to. The fact of referring to two occurred events makes more realistic our analysis, allowing for the possibility to compare the result of the simulation with real recordings at seismic stations. On the basis of these two reference scenarios and of a ground model in which seismic waves propagate, ground motion simulation was carried out by using the code GRFLT [Hisada, 1996 and modifications]. This allows us to calculate synthetic seismograms generated by the earthquake scenario at different locations. The objective is to determine, taking into account the Regional seismogenic context, the temporal separation between the first arrival of P and S waves at certain observation points for a given scenario earthquake. In this way we can evaluate the actual applicability of an EEWS to the real case. In order to be effective, in fact, the system must be able to issue the arrival time of the damaging shake (S waves generally) reasonably ahead (the order should be of the ten of seconds). In addition to determining the time difference between arrival of P and S waves, we also studied the influence of the distance and the directivity on the synthetic level of shaking. We expressed the results of the simulation in terms of velocigrams since we are concerned in an intermediate range of frequencies. We are interested in the time difference between the arrival of P and S waves, so we carried out a picking of the two first arrivals on the obtained velocigrams and we got the delta time by difference. We related this value with the distance from the epicenter and with the PGV for all the observation points. In order to determine which parameters are affecting the most all the simulation procedure, we built a geometric model based on the ray path theory, having the same configuration as the one used in the rigorous simulation, and we carried out a sensitivity study of the model by randomly varying its main parameters (i.e. wave velocities and thickness of the different layers). The geometric model alone is not able to fully reproduce the synthetic ground motion, but allowed us to carry out a sensitivity study and a calibration of the model. We tested the EEWS on possible future earthquake scenarios referring to a possible reactivation of a potential seismic source represented by the Giudicarie lineament, in northern Italy. Information about it was obtained by literature review. 1.5 Dissertation outline This thesis can be divided in 7 main chapters, we started by introducing the concepts at the base of the work, the objectives of the research, the methodology we used to carry out the study and the outline of the dissertation. 4

23 Chapter 1. Introduction In the second chapter we described the background of EEWSs and the fundamentals on which it is based on. Then we conducted a literature review on existing EEWSs worldwide, describing several countries experiences. In the third paragraph we detailed the objectives of hazard mitigation provided by EEWSs. Finally we discussed the perspectives and future developments of EEWSs. In the third chapter we shortly detailed the tectonic setting of Italy, particularly focusing our attention on the northern part. This allowed us to figure out a kinematic model of the Peninsula, essential point in order to understand the processes responsible for generating ground motions and corresponding levels of shaking. In addition, geophysical recordings coming from AGIP surveys in the Po Plain and studies of crustal structure, allowed us to define a crust model (this has been introduced as input in the ground motion simulations). In this framework we could therefore define two earthquake scenarios to base the simulations on. In the fourth chapter we described the studies carried out by INGV about Italian seismic zoning, giving an overview about seismic sources, focal mechanisms and seismic effective depth in particular for Lombardy. For the seismic zones we were interested in we gave information about expected level of shaking in terms of maximum magnitude, intensity, and PGA (Peak Ground Acceleration). We gave then an overview of the Italian seismic catalogue focusing on Lombardy and surrounding zones, both for historical and instrumental recordings. In conclusion we showed the seismic hazard maps established by INGV [2004] for the Italian territory. In chapter five we described the simulation procedure and the program we used to obtain synthetic seismograms from a given scenario. In chapter six we applied the code to the two selected earthquake scenarios affecting Lombardy, we detailed the characteristics of the two cases in terms of selected model and source parameters as to input in the computation code, and we described the results we obtained from the simulation. We analysed the influence of distance and directivity on the produced level of shaking. The synthetic seismograms were then compared with real recordings of the two events at seismic stations. Afterwards we studied the attenuation of synthetic PGV with distance and we compared what we obtained with attenuation laws from literature. Finally we found the temporal separation between first arrival of P and S waves and we related the delta time with distance from the epicenter and with PGV at the observation points. In conclusion we showed how a simple geometric model can be considered a good approximation for temporal separation calculations, but it does not provide information about the amplitude of the generated ground motion and it is too simplistic to give reliable result for travelling times of seismic waves. It resulted to be very useful for carrying out a study of the sensitivity of the model to its main parameters and thus a calibration of the model. Conclusions from the feasibility study of an Earthquake Early Warning System in the Lombardy region performed in this work and recommendations for future research are presented in chapter seven. 5

24 Chapter 2. Earthquake Early Warning Systems 2. EARTHQUAKE EARLY WARNING SYSTEMS 2.1 Background An Earthquake Early Warning (or alerting) System provides notification that an earthquake is occurring and that potentially damaging ground motion is approaching. Ideally, a network of field stations equipped with strong motion instruments will detect the initiation of an earthquake and if the earthquake meets or exceeds a given ground motion or magnitude value, a signal is transmitted from field stations to a data processing site where it is processed and, as a result, a public warning is then issued to populations and facilities at risk. Warning times depend on the distance between the earthquake source and the populated area and may vary from no warning at all to more than a minute if the source is quite distant. In the absence of a reliable earthquake prediction system which could theoretically provide hours to years of advance warning of a damaging earthquake, earthquake early warning provides an alternative. The public safety benefit of early warning systems is in response readiness, that is, with appropriate training and preparedness, members of the public will learn to take various protective measures to reduce the risk of injury and minimize damage. On an organizational level, actions could be taken to reduce the risks of injury to employees, customers, and patrons and protect property through very rapid mitigation measures. For example, trains could be alerted to slow down or remain in their stations, elevators could be programmed to stop and open their doors at the next floor and telephone calls could be rerouted around areas of impact [Goltz et al., 2002]. The concept which lie at the base of the idea of EEWS is that the first seismic arrival at a given observation point from an earthquake is the P wave, which is usually relatively lowamplitude and causes little damages (see figure 2.1). It is followed by the S wave, which usually has larger amplitude and include the peak ground motion, causing most of the damage to buildings in an earthquake. The most basic systems of EEW, offering no warning time, issue an alarm when ground shaking at the same location exceeds a given threshold. Observation of peak ground motion allows for estimation of the magnitude of an earthquake. By deploying seismometers between the source and the urban area, a warning time can be issued by use of the P-wave arrival to estimate the magnitude of an earthquake [Allen and Kanamori, 2003]. Once determined the magnitude of the in progress event, electronic transmission will issue the alert to areas that may suffer structural damage in the earthquake. Depending on the temporal separation between arrival of P and S waves (which will vary according to the depth of the event and the 6

25 Chapter 2. Earthquake Early Warning Systems proximity of the closest station), this estimation can be determined before the S-wave arrival at the epicenter. Figure 2.1. Ground surface motions: the three components of a seismogram are included, Z indicates the up-down vertical motion, N the north-south horizontal motion, E the east-west horizontal motion, where the arrivals are more evident, the first arrival of P and S waves are indicated with the correspondent letter. In its current work, Allen shows that the relationship between P wave frequency and the total magnitude of the earthquake holds for major quakes, up to magnitude 8 and higher, as well as for medium and small earthquakes. Based on the correlation, he shows that it is possible to predict the total magnitude of the event down to magnitude 1. Allen's findings conflict with the current model of earthquake rupture. The "cascade" model assumes that earthquake faults are made up of lots of different-sized patches, each under some degree of stress. When one of the patches is stressed enough to slip, the slip propagates to adjacent patches, which rupture in turn like falling dominoes. The rupture stops only when the stress propagating along the fault zone reaches a patch that is too solidly locked to slip. Inherent in this model is the idea that the initiating rupture is the same for big and small quakes. Allen's findings suggest this is wrong. Instead, the rupture is different for large and small quakes from the beginning, and the initial rupture contains information that can be used to predict the final size. He proposes that if the initial rupture generates a large "slip pulse" that travels continuously in all directions across the fault plane, the pulse can supply the necessary energy to propagate through patches that would not otherwise have ruptured. Only when the energy in the pulse drops to a level insufficient to overcome the grip of rock on rock does the rupture stop. In their report, Allen and Olson [2005], say that, if the rupture pulse initiates with a large slip, it is more likely to evolve into a large earthquake. 7

26 Chapter 2. Earthquake Early Warning Systems Allen's demonstration that this observation holds in earthquakes around the world, from California to Taiwan and Japan, provides a solid basis for constructing an early warning system. Once the magnitude of the quake has been estimated, computers can predict areas of serious ground shaking based on an understanding of a particular fault. Within five seconds, warnings could be sent to cities in the areas calculated to expect damaging ground motion. In particular, for this study the hospitals of the Lombardy Region are considered as sensitive objectives to protect in case of an earthquake. The evaluation of feasibility had already started by distribution through the hospitals of a questionnaire aimed at the evaluation of their interest on the implementation of an Early Warning System. This step was necessary since the applicability of such a system depends on the correct interaction with the medical staff, who is supposed to identify the biomedical equipment to protect and the most adequate security operation, taking into account the life of the patient undergoing medical treatment at the time of the seismic event. 2.2 Literature review on Earthquake Early Warning System worldwide Seismologists, especially those in the United States, have become increasingly pessimistic about being able to predict earthquakes. Experiments at the intensively monitored Parkfield, California, site have dampened enthusiasm that earthquake ruptures could be predicted hours or days before they happen. To reduce loss of life and property, earthquake-prone regions generally rely on a combination of advance preparation and post-earthquake assessment and notification between five and 10 minutes after a quake. Allen and Kanamori [2003] reported differences in the frequency of seismic signals emanating from small and medium earthquakes during the first four seconds of the rupture, with the larger quakes showing lower frequency signals than the smaller quakes. EEWSs that estimate the severity of ground shaking and the time till that shaking will commence are in operation in California, Japan, Mexico, and Taiwan. The most basic system, offering no warning time, issues an alarm when ground shaking at the same location exceeds some threshold. When the earthquake source region is some distance from a populated area or city, seismometers can be deployed between the sources and the city to detect any earthquake and transmit a warning electronically, ahead of the more slowly moving ground motion. Mexico City is protected by such a front detection EWS, providing up to 70 s of warning time (see Espinosa Aranda et al., [1995], Anderson et al., [1995], and The Central Weather Bureau of Taiwan also uses a front-detection EWS, which can provide warning for areas greater than around 75 km from the epicenter (see Wu et al. [1998] and Wu and Teng [2002]). All these EWSs use observation of peak ground motion to estimate the magnitude of an earthquake, which is the most commonly applied and most accurate method of local magnitude determination. However, this approach does not provide the most rapid magnitude estimate. In Japan, one of the world's most seismically active areas (the country accounts for about 20 percent of the world's earthquakes of magnitude 6 or greater), this approach was introduced in the 1990s; the UrEDAS system uses the P-wave arrival to estimate both the magnitude and location of an earthquake (see Nakamura and Tucker [1988 and 1988 b]. See also Real-time Earthquake Assessment Disaster System in Yokohama READY 8

27 Chapter 2. Earthquake Early Warning Systems On August first 2006 Japan launched a new early warning system for earthquakes. The warning could precede the shaking by about seconds and, at sufficient distance from the epicenter, would also include the expected intensity of the tremor. The network kicked off with 41 institutions, including railway companies, construction firms, factories and hospitals. The system has been running for a test period of over two years since February 2004 and the agency is aiming to implement the system nationwide by the end of March Rapid response systems are also implemented in California (USGS, Caltech and CDMG TriNet ShakeMaps Caltech-USGS Broadcast of Earthquakes (CUBE) System UC Berkeley Seismological Laboratory and USGS Menlo Park (REDI) here Allen and Kanamori [2003] recently proposed an Earthquake Alarm System (ElarmS) for southern California, where the early warning problem is particularly challenging due to the high number of active faults dissecting even metropolitan areas. This system issues a warning based on information determined from the P wave arrival only, providing the potential to issue a warning before peak ground motion at the epicenter. Erdik et al. [2003] proposed also to install an Earthquake Rapid Response System in Istanbul, where significant earthquake hazard and risk is faced, as illustrated by the earthquake risk scenario for the city (see Currently such systems are either implemented on in construction or planning stage also in Romania (see and Greece. 2.3 Objectives of hazard mitigation provided by EEWs s Current effort to mitigate seismic hazard includes long-term hazard assessment and rapid post-event notification. Long-term hazard mitigation is facilitated by probabilistic groundshaking maps [Frenkel et al., 1996], which estimate the probability of ground motion exceeding some threshold during the next i.e. 50 years. Rapid post-earthquake notification is provided in California for example by a network of many station sites with both high dynamic range broadbande and strong-motion instrumentation (see Kanamori et al. [1997]; Hauksson et al. [2001]). As mentioned before, the aim of this work is to try to estimate whether an enough amount of time is passing between the arrival of the first P wave and the arrival of S waves, in order to alert sensitive structures such as hospitals in the Lombardy Region. If this is occurring, the objective is to install a net of sensors in strategic points which will warn the structures ahead of some to tens of seconds in order to allow the shut down of appliances and devices. Rough estimations of the time separation between the arrival of the two waves show time difference of around 1 second each 8 km. With our analysis we got results that show good agreement with this assertion. We individuated two scenario events (see fig 2.2) whose effects potentially affect Lombardy. 9

28 Chapter 2. Earthquake Early Warning Systems Figure 2.2. Map of the principal towns of the Lombardia region with placemarks indicating the location of the epicentres of the major seismic events. (From Google Earth). In the following chapters we will give accurate details concerning the characterization of the two sources considered in this work and of a third potential one, and the warning that an EEWS could be able to issue. To preview some results, however, we found for each case a nearly linear relationship between the time difference between the two arrivals and the distance of the observation point from the epicenter. According to the selected crust model, the wave velocity in the bedrock is around 5 km/s for the P waves and 2.8 km/s for the S waves. With this information we were able to estimate how ahead the EEWS could issue the alert to the major towns of Lombardy. The transmission of the alert to sensitive structures starts electronically and, with an adequate transmission technology, we found possible to communicate a pre-alert tens of seconds ahead the arrival of damaging waves. According to Allen and Kanamori [2003], the delay in transmitting the warning would be dependent on the technology used, but could be reduced to less than 1 second, the system can therefore provide an effective warning to strategic buildings. 10

29 Chapter 2. Earthquake Early Warning Systems 2.4 Perspectives and future developments of EEWS s In our analysis we found that, depending on the location of the urban area with respect to the initiation of fault rupture, it s possible to provide a few to tens of seconds of warning to areas that may suffer damages in an earthquake. The facilities receiving the early warning information will decide proper actions based on the alarm level. In addition to this immediate use, the development of an early warning system will lead to development of infrastructures that can efficiently and readily use the information. A few seconds may not sound like much, but it is enough time for hospitals to shut down medical equipment and start the emergency operation in order to ensure the security of patients. A part from the objective of this work, a warning of seconds allows also school children to dive under their desks, gas and electric companies to shut down or isolate their systems, emergency providers to reach probable trouble areas and endless more applications, which will allow for saving lives and money. It could be also matter of discussion the possibility of another hazard: of people to panic if an alert is issued. In order to prevent such possibility, together with the installation of an EEWS, it should be set up a system of information at all levels between citizens: in this way everybody would be prepared in case of alert and the secondary hazard would be reduced. A drawback of such a system can be the effectiveness of the warning: we will demonstrate in the following chapters that towns which could be issued with a good warning time will not be struck by actually damaging ground motion due to the attenuation of seismic waves travelling in the top crust from the source to the observation point. This inconvenient must be taken into account particularly in case of low-medium magnitude event, such as those that are likely to strike a low seismicity region as Lombardy. In each case it will thus be important to find a reasonable trade-off between the two aspects of the problem. 11

30 Chapter 3. Tectonic settings of Italy 3. TECTONIC SETTINGS OF ITALY 3.1 Overview In this part we are going to describe the present-day stress field in Italy, giving first a general view and then individuating areas of relatively uniform stress field. In particular these zones can be grouped into Northern Italy, Appennines and Southern Italy. Most of the Appenninic belt is affected by an extensional regime, whilst a compressional (or transpressional) regime characterizes the eastern Alp, the eastern side of northern Appennines and the southern Tyrrenian to northern Sicily zone. An abrupt change in stress direction marks the transition between northern and southern Appennines, suggesting that the two arcs are characterized by a different tectonic setting and recent evolution [Montone et al., 1999]. 3.2 Geological setting of Italy The present-day kinematic and stress field of the Central Mediterran area and the tectonic structure of Italy, are related to the N-S convergence between Africa and Eurasia. This process is particularly complex since different processes are acting at the same time and in close proximity. These processes include the northward indentation of the Adriatic microplate beneath the southern Alps, the flexure of the continental Adriatic lithosphere below the Appennines, and the subduction-sinking of the Ionian lithosphere below the Calabrian arc (see figure 3.1). This complex geology is reflected in a strongly variable stress field. It is not clear, however, which is the relative contribution of the driving forces which are deforming the Italian region. [Montone et al., 1999]. Other geodynamics processes which must be taken into account are mainly related to the passive sinking of the subducting Adria lithosphere. In the Appennines, these processes are responsible for the migration of the thrust belt-foredeep system towards the Padan-Adriatic-Ionian foreland, and for the synchronous opening of the Tyrrenian backarc basin according to slip vectors largely exceeding the values of Africa-Europe convergence [Meletti et al., 2000]. 12

31 Chapter 3. Tectonic settings of Italy Figure 3.1. Structural-kinematic sketch of Italy and surrounding areas showing the traces of the slip vectors of Africa versus Europe (according to Livermore and Smith, 1985) and Adria versus Europe (according to the rotation pole proposed by Meletti et al. [2000]). The present day seismicity in the Central Mediterranean region [Amato et al., 1997; INGV 1995] mostly follows the principal mountains chains, that is the Alps, the Appennines and the Dinarides, though clusters of epicentres indicate a certain fragmentation of the foreland areas, notably in Southern Sicily and the Sicily Channel, in the Gargano-Tremiti region and in the Central Adriatic Sea [Meletti et al., 2000]. The Alps are a well known thrust-and-fold belt comprising a huge pile of basement an cover nappes transported towards the European foreland, detached from the lower (Europe) and upper (Adria) plates. See figures 3.2, 3.3 and

32 Chapter 3. Tectonic settings of Italy Figure 3.2. Structural sketch of Italy and surrounding regions. 1 trhust, 2 normal faults, 3 strike slip faults, 4 Pliocene-Quaternary anticlines, 5 Pliocene-Quaternary synclines, 6 backarc basin floored by oceanic crust. [Meletti et al., 2000]. The Appennines fold and thrust belt has developed on top of the eastward migrating subducting hinge associated with compression at the outer front and contemporaneous extension in the backarc region [Malinverno and Ryan, 1986; Patacca and Scandone, 1989]. According to Frepoli and Amato [1997] and Mariucci et al. [1999], this process is still active today in the northern Appennines. On the contrary, there is geological and seismological evidence that the southern part of peninsula underwent a major stress change in the last 1 Myr [Patacca and Scandone, 1989; Pantosti and Valensise, 1890; Hippolyte et al., 1994; Westaway, 1993; Amato and Montone, 1997]. 14

33 Chapter 3. Tectonic settings of Italy Figure 3.3. CMT (Harvard) and RCMT (Pondrelli et al. 2002) solutions for the >4.5 seismicity since The extension along the Appennines belt and the compression around the Adria lithosphere and in the Northern Sicily offshore are evident. [Chiarabba et al. 2005]. With regard to the recent tectonic evolution of the peninsula, most of studies carried out in the last few years, propose several different hypotheses, which are often conflicting with each other. According to Montone et al. [1999], this might be due to the lack of tight constraints which should allow us to discriminate among the proposed mechanism. Therefore, at the moment there is still several questions unanswered in the understanding of the relationship between the stress in the crust and the active tectonic process. Montone et al. [1999] individuates few large ( km 2 ) regions in Italy within which the stress field is relatively uniform. 15

34 Chapter 3. Tectonic settings of Italy Figure 3.4. Tectonic scheme summarizing seismological data (Chiarabba et al. [2005]) and available kinematic constraint (Devoti et al. 2002; Hollestein et al. 2003; Battaglia et al. 2004). Arrows are simplified directions of GPS sites relatives to a stable European plate. [Chiarabba et al., 2005] Northern Italy The eastern Alps are dominated by active compression, which shows thrust faulting earthquakes (see figure 3.3 and 3.5), with N-S direction of maximum compression. Here, the coexistence of ~E-W trending Alpine structures and NW-SE trending Dynarides structures determines a variability of the fault plane solutions, but the driving mechanism seems to be northward push of the Adriatic plate below Alps, consistent with the motion of Africa towards Europe. This mechanism can also explain the stress rotation observed in the western Alps, where the Adriatic plate is pushed to the west below the Alpine arc. [Montone et al. 1999, see also Grünthal and Stromeyer, 1992; Müller et al., 1992]. Eva et al. [1997] have explained the seismicity in this region as due to backthrust faulting in the lowland regions and normal faulting in the most elevated area. Stress data north of the Alps revelas N-S compression [Müller et al., 1992], but the discontinuous data distribution on the Italian side does not allow us to extrapolate the observed stress directions over the Alpine arc [Montone et al. 1999]. 16

35 Chapter 3. Tectonic settings of Italy As regard the southern Alps, the maximum shortening has been calculated in the eastern part of the system [Castellarin, 1978], where the Alpine structures join those of the Dinarides, and where the strongest historical earthquakes of the Alps have been recorded (e.g. Friuli earthquake, M = 6.5). Figure 3.5. Sketch map of main tectonic features of Italy simplified from Bigi et al. [1990]. CMTs for great earthquakes that occurred between 1976 and 1998 are shown. (a) thrust fault (pre-middle Pliocene); (b) thrust fault (middle Pliocene-Recent); (c) normal fault; (d) strike slip fault; (e) undetermined fault. [Montone et al., 1999]. Regarding the seismicity, Chiarabba et al. [2005] show this is abundant and clustered in the eastern and western Alps, while it is sparse in the central Alps. (See fig. 3.7), probably reflecting the density of seismic stations. In the western Alps, we note two main arc-like strips of events with M d usually smaller than 3.0 following the mountain range, namely the Briançonnais and Piemontais arcs (see figure 3.6). A radial extension characterizes the present day tectonics of the inner belt, suggesting complex mechanisms of slab retreat or break off [Sue et al., 1999; Sue and Tricart, 2002]. Along the main strips, regions with few and sparse events identify seismic gaps, where future earthquakes are more likely to occur (i.e. the Pelvoux massive, the Lepontine dome). 17

36 Chapter 3. Tectonic settings of Italy Figure 3.6. E-W vertical sections of seismicity across the western Alps. Events falling ± 20 km from the section are plotted. The geometry of the European and Ligurian Moho taken from available seismic reflection data are shown. [Chiarabba et al. 2005]. In the eastern Alps, the shallow crustal seismicity is located on several south-verging ramps developed within the Adriatic Mesozoic cover (Adriatic thrust fault system). Subparallel faults broad the active region at the intersection between the Alpine and Dinaric structures. The southernmost active thrust is located at the foothill. Large thrust earthquakes are caused by a N-S trending compression (see Slejko et al. [1989]; Bressan et al. [1998], among many others). The deep crustal seismicity indicates the northeastward flexure of the Adria plate beneath the Dinarides belt, indented in the south-dipping European subducting lithosphere and a possible over thrusting of Adria along a S-SW-verging ramp. To the west, the border of the Adria lithosphere is poorly defined by seismicity define possible seismic gaps (i.e. the Cansiglio and Montello westernmost segments of the Adriatic thrust fault system).to the east, the seismicity clusters on N-NW trending faults in the Dinarides. Strike lip mechanisms, such as the M L 5.6 Bovec earthquake, accommodate a transpression of Adria. Chiarabba et al. [2005] show the diffuse occurrence of deep crustal and subcrustal earthquakes that show the location and geometry of the Adria lithosphere, flexured beneath the Alps and Appennines. The northern limit of Adria is well defined by the deep crustal seismicity occurring beneath the eastern, central and western Alps. In the northern and central Appennines from the Po plain down to the Gargano promontory, the arc-like belt of deep crustal seismicity follows the mountain range and defines almost continuously the flexure of Adria beneath the belt. Hypocentral depths are typically between 12 and 25 km. The largest 18

37 Chapter 3. Tectonic settings of Italy events (Parma M L = 5.7, 1973; Parma M L 5.1, 1983; Reggio Emilia, 1996 M L 5.4 and Forlì, 2000 M L 4.5, Selvaggi et al. [2001]) indicate a mechanism of compression on thrust faults buried benath the Po plain that parallel the curvature of the arc. An E-NE elongation of seismicity on subparallel faults connects the Apulian foreland to the Dinarides, in correspondence with the variation of lithospheric thickness [Panza, 1984] Appennines The boundary between the Alps and the Appennines corresponds to a transform fault zone which linked through Tertiary and Quaternary times two orogenic systems generated by opposite lithosphere subductions: Europe beneath Adria in the Alps and Adria beneath Europe in the Appennies [Meletti et al., 2000]. Subduction of Adria beneath Europe started because of active convergence processes in a neutral arc system, but with time the flexure-hinge retreat of the lower plate largely exceeded the amount of convergence. Consequently, backarc extension took place, accompanied by high rate forward migration of the thrust belt-foredeep system. Referring to the post- Tortonian evolution of the Appennines, the progressive opening of the Tyrrenian sea and the synchronous forward migration of the Appennine thrust-and-fold belt, accompanied by consumption of the Adria foreland, are the most striking results of these processes. At present, the Appenninic chain appears to be divided into two major arcs: the Northern Appenninic arc and the southern Appenninic arc. [Meletti et al., 2000]. The Appennines form the backbone of Italian peninsula and Sicily; active extensional tectonics is observed throughout the Appenninic belt, in which the maximum principal stress is generally vertical and the stress regime is extensional, with NW trending fault planes. The largest normal faulting earthquakes (magnitude between 6 and 7) occur in the most elevated regions of the Appennines, as indicated by the 1997 Umbria-Marche seismic sequence [Amato et al., 1998a; Ekstrom et al., 1998], the 1915 Fucino earthquake, the Val Comino earthquake of 1984, the 1980 Irpinia events and the 1998 earthquake at the Basilicata- Calabria border. Montone et al. [1999] compared breakout, earthquake and fault data, and concluded that the NE-SW extension seems to be continuous throughout the Appenninic belt. In southern Appennines the extension seems to be a general and widespread feature overprinting the previous compressional tectonics. It seems that the normal faulting stress regime observed in the belt changes to a strike-slip regime in the foredeep and to a thrust regime in the Adriatic foreland. It is important to remark the absence of compressional stress perpendicular to the thrust front of the southern Appennines [Montone et al., 1999]. On the contrary, there are several indications of the active compression in the outer front of the northern Appenninic arc, including breakouts and seismicity of this region [Montone et al., 1999]. In the backarc region of northern Appennines we still observe predominantly normal faulting, in most cases parallel to the trend of the adjacent belt [Montone et al., 1999; Montone et al., 1995]. Along the Tyrrenian coast of central Italy we observe radial extension. 19

38 Chapter 3. Tectonic settings of Italy The difference between the southern Appennines and the northern Appenninic arc is evident from the 90 rotation of the minimum horizontal stress orientation at latitude 42 N-43 N (see figure 3.8). Figure 3.7. Schematic lithospheric section across the northen Appenninic arc [Meletti et al., 1995]. In the northern Appenninic arc the slab sinking with a flexure-hinge retreat faster than the Adria divergence may wholly justify the regional seismicity pattern, characterised by: - low/medium-energy compressional earthquakes along the Padan-Adriatic margin of the Appennines related to active frontal and lateral ramps which branch off from the sole-thrust at greater depths (but in any case not exceeding about 20 km) moving from the foreland towards the mountains chain; - medium/high-energy earthquakes, mostly displaying extensional dip-slip focal mechanisms, in correspondence to an axial belt located between the Adria flexure hinge and the Tyrrhenian asthenospheric wedge. The bulk of the focuses is contained in a crustal synform where the opposite geometries of the rising Tyrrenian asthenosphere and the sinking Adria lithosphere are accommodated by NE-dipping low angle master-faults and SW-dipping high-angle antithetical-faults; - low-energy very shallow earthquakes above and behind the mobile asthenospheric wedge. In the southern Appennines, the cessation of subduction while the counterclock-wise rotation of the Adria microplate was still continuing produced a strong modification of the lithosphere-asthenosphere system and the establishment of an extensional regime. A seismic axial belt is present, characterized by medium/high-energy earthquakes whose available focal solutions show dip-slip mechanisms. Meletti et al. [2000] relate the seismicity of the southern Appennnines to very young normal faults generated by the Adria divergence, superimposed on inactive contractional structures. 20

39 Chapter 3. Tectonic settings of Italy According to Chiarabba et al. [2005], the main feature is an arc-like belt of seismicity in the upper crust that follows the mountain range. From north to south, the upper crustal seismicity shows a rotation from NW-trending alignments in the north to NNE-trending in Calabria, pralleleing the rotation of the Appeninic and Calabrian arcs. In the northern and southern Appennines, the NW-striking segments are confined within the upper 6-8 and km of depth, respectively. Figure 3.8. Summary of the stress data of the Italian peninsula. The main stress provinces delineated by the data are sketched. The structural arcs with shaded triangles indicated the active oceanic subduction; solid triangles show active compression and open triangles delineate the location of Plio-Pleistocene trust front, presently affected by prevalent extension. [Montone et al. 1999]. The largest events show normal faulting mechanisms, consistent with the regional NEtrending extension [Westaway, 1992; Montone et al., 1999]. The limit between the flexure of Adria and the extensional belt is sharp, well defined by the depth variations of hypocenters. (see figure 3.9). 21

40 Chapter 3. Tectonic settings of Italy In the southern Appennines, the upper crustal seismic belt is narrow (30-50 km). Large earthquakes originate on ~20-40 km NW-elongated normal faults [Pantosti and Valensise, 1990; Amato et al., 1992; Pantosti et al., 1996; Galadini, 1999; Piccardi et al., 1999 among many others], that cut the entire upper crust (down to km depth). Background seismicity mostly occurs at the borders of the silent fault segments. An active NE-trending extension has been observed by focal mechanisms and borehole breakouts [Montone et al., 1999]. Normal fault earthquakes occur mostly beneath the mountain belt, following a rotation from NW-trending in southern Appennines to NE-trending in Calabria and in the Messina strait [Pino et al., 2000]. Figure 3.9. Vertical sections of seismicity across the northern Appennines arc. Events falling ± 20 km from the section are plotted. The lines indicate the geometry of the Adriatic and European Moho, as suggested by seismic reflection data and earthquake hypocenters. The bold lines show a simplified sketch of the main faults in the crust. [Chiarabba et al., 2005]. 22

41 Chapter 3. Tectonic settings of Italy Southern Italy (Calabria and Sicily) This region, which lies on top of the active subduction of Ionian oceanic lithosphere is one of the most seismically active areas in Italy [Boschi et al., 1995]. Unfortunately, however, no recent earthquake data are available. From geological data it was inferred that the sources of large historical earthquakes are normal faults trending mainly NE-SW [Bigi et al., 1990; Valensise and Pantosti, 1999], roughly parallel to the subduction hinge. The few focal mechanisms available seem to be consistent with this trend, both in northern Calabria and in the Messina straits, although no data are available in between. Stress directions in Sicily show a very consistent pattern. A NE-SW trend of the minimum stress orientation thus parallel to the main thrust front, is evident in southeastern Sicily, which belongs to the Hyblean African foreland domain. A rotation to an almost orthogonal trend is detected in the adjacent foredeep, which could be due to the flexure of the African lithosphere beneath the belt. More to the north, the minimum stress orientation rotates again to an almost E-W to NE-SW trend. This stress direction is consistent with the northward motion of Africa towards Eurasia, although the two plates here do not really interact because of the presence of the Tyrrenian basin north of Sicily. [Montone et al., 1999]. 3.3 A seismotectonic (kinematic) model Concerning the seismotectonic model of the entire national territory and adjacent areas, Meletti et al. [2000] show that the seismicity pattern is controlled within a relatively small space by a very complex geodynamic framework. Within this framework they recognize: - Continent-continent convergence (Alps and Dinarides) with development of compressive-transpressive features along the plate margins; - Plate divergence across margins characterized by passive slab sinking (northern Appennines and Calabrian arc), with development of backarc basins (northern Tyrrenian sea and Southern Tyrrenian sea) flanked by forelandward migrating thrust belt-foredeep systems. - Plate divergence across a margin previously characterized by lithosphere sinking and thereafter discharged from the subducted slab (southern Appennines), with development of quite peculiar rift processes within the inactive thrust belt; - Transpression (northern Sicily) due to the combined effect of plate convergence (Africa-Europe) and high-rate flexure-hinge retreat of an intervening plate (Adria microplate) with high angles between the respective slip vectors; - Intraplate strain partition and fault activity (mainly combined strike-slip and thrust motions), possibly in correspondence to inverted structures. Within the single mobile belts the earthquake space distribution and the maximum source dimensions are obviously controlled by the overall geometry of the system, while the slip rate 23

42 Chapter 3. Tectonic settings of Italy and the focal mechanisms are determined by the kinematics of the mobile lithosphereasthenosphere system. [Meletti et al., 2000]. 3.4 Depth of the seismogenic layer Chiarabba et al. [2005] show that the depth of the seismogenic layer strongly varies and follows the tectonics of the area. In the Appennines, the depth increases from 4-6 km beneath the Tyrrhenian coast to 6-8 and km beneath the normal fault belt in the northern and southern Appennines respectively. Towards the east and in the external area of the Appennines system, the depth of the seismogenic layer increases to more than 20 km, following the flexure of the Adria and Ionian lithosphere. Along the northern Appennines chain axis, the belt of deep crustal seismicity is locally displaced [Chiarabba et al., 2005]. In the Alps, Chiarabba et al. [2005] found that the depth is around 10 km, except a western sector and the eastern most strip, where the depth increases northward (12 km) along the dipping of the main active thrust, and both southward and eastward in the Dinarides (z >16 km), where the Adria lithosphere is flexured. In Sicily the seismogenic layer is deeper, down to 20 km depth, beneath the offshore compressional belt, the TL fault system and beneath Mt. Etna. [Chiarabba et al., 2005]. 3.5 Geophysical data and crust model In order to study the propagation of seismic waves travelling from source to site into the Lombardy ground, we defined a crust model beneath the Lombardy region, referring to existing literature. In particular we took as reference point for an uppermost mantle structure model the study of Di Stefano et al. [1999] and, to detail the upper crust beneath Lombardy, to the study of Pieri and Groppi [1981], who carried out geophysical surveys for Agip society. In their work, Di Stefano et al. [1999], present a 3-D image of the P-wave velocity structure in the crust and uppermost mantle beneath Italy, obtained by inverting P-wave arrival times of crustal earthquakes. In their work they focused on the velocity structure of the upper 40 km. They individuated three layers, two of which are in the crust, at 8 and 22 km depth, and one of which is beneath the Moho, at 38 km depth. They identified in their model 34 km depth for the location of the Moho. In the area we are interested in, they individuated areas of low velocity materials in the eastern margin of both the western Alps and the northern Appennines (in the Padanic foredeep) in the upper layer (due to the thick layer of sediments characterizing the Padana Plain). In the second layer, at 22 km depth, they identified instead high velocities beneath the Padanic area. In the third layer (38 km depth) they found again low velocities surrounding the northern Appenninic arc from the Padanic area to Mt Conero; to the west of the low velocity of the northern Appenninic arc, relatively high-velocity areas are observed beneath the belt. Di Stefano et al. [1999] explained their results as following: the upper crustal structure of the Appennines consists of different units, each unit being about 5-6 km thick, mostly composed of marls and limestones (with average V p values of about 4 and km/s, respectively), thrust over low-velocity flysh units (V p km/s) and overlying crystalline basement. 24

43 Chapter 3. Tectonic settings of Italy Information about the deep structure of the Appennines is sparse and less constrained. In the third layer (22 km depth), low-velocity and low-density materials are present beneath the belt. For the Alps, the authors, observed a similarity between the velocity in the lower crust and gravity anomaly. In the western Alps, the Ivrea-Verbano body represents as a high velocity feature, separated from low-velocity areas imaged beneath the Po plain by a sharp velocity gradient. To the north of the Ivrea-Verbano body, other high-velocity features can be related to the Appenninic thrust units. In the lower crust, a low velocity area is found to the northwest of the Ivrea-Verbano high-velocity. This anomaly may indicate the presence of upper crustal rocks at depth, with velocities lower than the Penninic thrust unit. Modeling gravity data, Bayer et al. [1989] found evidence for an approximately 14 km thick low-density unit, possibly composed of upper crustal rock beneath about 20 km depth. Di Stefano et al. [1999] claimed thus that the upper crustal rocks, overthrust by the Penninic unitand actually present at about 22 km depth beneath the western Alps, have not yet been completely metamorphosed. At 38 km dept, a small continuous area of low-velocity anomalies is present beneath the whole Alpine belt. To deepen our knowledge of the upper part of the crust in the Po plain and so to be able to define a crust model for our simulations, we referred to the work of Pieri and Groppi [1981], relating to AGIP Company surveys in that zone. They investigated the structural pattern of the external part of the Northern Appennines and its relationships with the foreland area. From west to east they identified three main folded arcs: the Monferrato (partly outcropping in the western Po plain and bordered to the south by the South Piedmont Basin), the Emilia arc and the Ferrara-Romagna arc. In the Lombardy plain they essentially identified a series of sediment layers down to around 5 km depth, with successions of Quaternary deposits, folded Miocene and Paleogene clastic sections on Mesozoic carbonates. Given the main configuration of the underground feature beneath Lombardy, we zoomed on the shallowest part of the sediments by referring to the work carried out by the Civil Engineering Department of the University of Trento [2004], in which they synthesized all the available information about subsoil in the Friuli region. The Friuli region is located in the eastern part of the Padan Plain, so we considered reasonable to extend the shallow configuration of this area to our region, located in the central part of the Po plain. We have to keep in mind, however, that sediments in the central part of the plain reach much higher depths, that is what indicated in the work Pieri and Groppi [1981]. Stratigraphic sequences obtained by well drilling in the Friuli plain (see Note informative of the Carta del suolo della pianura friulana, 2004) shows series of sands, marls and clays (with some intercalations of gravel) until around 700 m depth, over a limestone basement. 3.6 Earthquake scenarios In the previous paragraphs we have given an overview over the seismotectonic and geologic configuration of the Italian Peninsula. Particular attention has been dedicated to the northern part of the country, where our study takes place. In this framework we wanted to define earthquake scenarios on which to test an Earthquake Early Warning system by computer simulation. In order the analysis to be reliable we decided to refer to real and representative 25

44 Chapter 3. Tectonic settings of Italy scenarios, we selected therefore two scenarios for which exhaustive literature could be found. These are the following: - The M w 5.4 Reggio Emilia 1996 earthquake, which potential source is outside Lombardy boundaries. Its proximity to the southern boundary, however, makes its ground motion effects highly affecting our region. Precisely, the seismic sequence occurred in October 1996 occurred on the southern edge of the Po plain, which unfortunately is a poorly known region that marks the transition between the wellestablished active extension zone of the Appenninic chain and the buried compressive structures of the Po plain [Selvaggi et al., 2000]. - The M w 5.0 Salò 2004 earthquake, with source located inside Lombardy, in the western side of Garda Lake, in the eastern part of the region. This area is located along the margin of the south-alpine chain, formed in the framework of the Africa-Europe convergence. The Garda Lake is affected by the Giudicarie fault system, related to the NNE-SSW trending thrust and transpressional structures. According to the work of Pessina et al. 2006, the southern Garda lake is affected by significant seismicity (1802, 1222, and 1901 earthquakes), while moderate seismicity affects the northern area (1932, 1882, 1876 earthquake). The whole earthquakes show an alignment of epicentres along the Giudicarie fault system. In the Database of individual seismogenic sources (DISS) the seismogenic sources responsible for the 1802 and 1901 earthquakes are identified through geological and geophysical investigations [DISS Working Group, 2006]. In the following chapters we will detail more the two selected earthquakes scenarios, in order to define the basis for simulating the ground motion produced by the seismic sources at different observation points. 26

45 Chapter 4. Seismic hazard in northern Italy 4. SEISMIC HAZARD IN NORTHERN ITALY 4.1 Overview In the previous chapter we described the seismotectonic context of the Peninsula, which is of essential importance in the understanding of the processes generating ground shaking in the region we are interested in. We showed that this area is characterized by low seismicity. Moreover the related available literature is very scant, therefore it s necessary to make some simplifications and to refer to the seismic hazard indicated by the study of INGV [2004] (see Rapporto conclusivo INGV relativo alla redazione della mappa di pericolosità sismica del territorio nazionale prevista dall Ordinanza PCM del 20 marzo 2003, n. 3247, All. 1) as the real seismic potential of a given zone. In this chapter we will therefore provide an insight into the seismic zonation individuated by the INGV [2004], focusing our attention on the northern part of the Italian Peninsula. 4.2 Seismogenic zoning In the Rapporto conclusivo INGV relativo alla redazione della mappa di pericolosità sismica del territorio nazionale prevista dall Ordinanza PCM del 20 marzo 2003, n. 3247, All. 1, the Institute identifies 36 seismogenic zones (referring to ZS9 [2004]) and provides for all of them information about geometrical aspects and expected behaviours. In particular ZS9 [2004] divides the national territory into zones of analogous kinematic mechanism, all differing in the characteristics of seismicity. In figures 4.1 and 4.2 are represented the seismic zones, at National level and relating to the northern Italy respectively. The Lombardy Region is located in a portion of Italian territory which is partly considered non-seismogenic area and partly divided into four seismogenic zones, which are the following: on the north, at the boundary with Trentino Region (Grigioni - Valtellina) towards the boundary with Veneto (Bergamasco) to the south (Savona Valle Scrivia oppure Tortona Bobbio????) at the boundary with Emilia Romagna Region (Appennino Emiliano Romagnolo) 27

46 Chapter 4. Seismic hazard in northern Italy Each zone is associated to a given source system (see GNDT Galadini et al. [2000]; DISS 3.0 Valensise and Pantosti [2004]), whose characteristics are based on the Plio-Quaternary kinematic evolution [Gruppo di lavoro INGV, 2004]. Figure 4.1. Seismic zonation ZS9. Each zone is defined by a number; zones identified by a letter are not used for the evaluation of the seismic hazard. Black boundaries between zones are drawn only on the base of tectonic or geological and structural information. Blue boundaries separate zones of similar deforming style, but with different seismic characteristics. [INGV Mappa di pericolosità sismica 2004]. 28

47 Chapter 4. Seismic hazard in northern Italy Figure 4.2. Seismic zonation ZS9, focus on northern Italy. [INGV Mappa di pericolosità sismica 2004] Seismogenic sources Figure 4.3 shows a map of the seismic sources individuated by the INGV [2004]. We can see that the source relating to the Lombardia Region are of small extension and capacity: in fact they belong just to the first two classes of magnitude M w of expected ground motion. The zone can therefore be defined as a low-seismicity region. In particular we have: - ZS 903, Grigioni Valtellina: historical source, maximum observed magnitude: M w For preventive purposes, INGV suggests to adopt a maximum expected magnitude value of ZS 907, comprising the lower part of the Bergamo and Brescia provinces: geological source for which the database DISS 3.0, suggests a maximum magnitude of 5.9 (even if the maximum observed value is 5.67). Again INGV suggests to adopt a higher value of This zone is characterized by medium-low energy seismicity, with the only exception of the Soncino earthquake (1802 event), whose magnitude has been indicated of around 5.9 (see Albini et al. [2002]) - ZS 911, Savona Valle Scrivia. This zone comprises the so-called Arco di Pavia and related structures. The source is of historical type, with maximum observed magnitude of M w As for the two previous cases the proposed preventive value is The hypothesis is that the structures here included have a function of kinematic release for the supposed migrating system (see for example Patacca et al. [1990]). Here the seismicity cannot be neglected. - ZS 913, Appennino Emiliano Romagnolo: historical-type source, with maximum observed magnitude M w : 5.85; the preventive value suggested by INGV is again This zone is the result of the decomposition of the strip extending from Parma to the Abruzzo region. Earthquakes observed inside this zone are mostly of compressional 29

48 Chapter 4. Seismic hazard in northern Italy type in the north-western part, and probably extensional in the south-eastern part. The historical seismicity is generally represented by low-value magnitude earthquakes. Hypocentral depths are in average higher than in the surrounding zones (this is testified by events that generated effects in pretty wide areas. Figure 4.3 compares ZS9 (Seismic Zoning [2004]) with the distribution of the seismogenic sources of DISS 3.0 database. Each source is represented by a chromatic scale expressing expected earthquake magnitude Mw for that source. Squared symbols represent earthquakes of the referring catalogue (CPTI2) but non-associated to a specific DISS 3.0 source. Their magnitude is represented by the same chromatic scale used for the sources. The magnitude classes representing earthquakes and sources are the same as those used for the seismic rates calculations. INGV however, in its Rapporto Conclusivo [2004], specifies to assume the hypothesis that events of M wmax = 5.0 can occur in all the zones outside the seismic zoning, coming from the observation that low-medium energy events have occurred in various areas. Figure 4.3. ZS9 Seismic Zoning of northern Italy (black borders) [INGV 2004] Focal mechanisms The seismic hazard map (from INGV [2004]) shows also the main focal mechanisms (the ones having the highest probability of characterizing significant future earthquakes) for all the seismic zones. The assessment of these mechanisms comes from the observation that the convergence between Adria and Europe plates is considered the main process responsible for 30

49 Chapter 4. Seismic hazard in northern Italy the active tectonic in Italy. In the central-eastern part of southern Alps INGV bases its characterization on the distribution of seismicity, considering this sector as a structural continuous system of adjacent thrusts. Unfortunately the lack of information about the possible transferring mechanisms between the eastern and western part of the Alpine sector leave space for other geodynamic mechanisms conditioned by the evolution of the same sector. Figures 4.4 and 4.5 represent the main focal mechanisms for the seismogenic zones. Figure 4.4. Focal mechanisms of ZS9 seismic zonation, from the database pubblished by Vannucci and Gasperini [2003]. Dimension of symbols is proportional to the logarithm, of the overall released seismic moment [INGV 2004, Zonazione sismogenetica ZS9]. In particular, in the zones relating to our study, the expected mechanisms are the following: 31

50 Chapter 4. Seismic hazard in northern Italy - ZS 903: INGV defines the rupture mechanism as undetermined. This is due to the fact that this zone has been identified referring more to seismicity characteristics than to geologic information. In this sense the map shows a dip-slip mechanism with some strike-slip component, with T axis oriented SW-NE. - ZS 907, in this zone the only tectonic activity for which the Giudicarie system is responsible is located, probably due to the movement of the compressional structure of the central southern Alps. The expected mechanism here is mixed: thrust and strikeslip. - ZS 911, this is a transferring zone between the Appenines and the Liguria sea, where expected mechanisms are left strike-slip for crustal structures, and dip-slip for deeper structures. - ZS 914, this zone is located in a mixed transition band, related to the passive subsidence of the Adriatic lithosphere under the northern Appennines belt. Expected mechanisms are thrust and strike-slip with axis oriented SW-NE. (See Scandone e Stucchi, Zonazione Sismogenetica ZS4, [1996]). Figure 4.5. Main fault mechanism expected for the seismic zones of ZS9 [INGV 2004, Zonazione sismogenetica ZS9]. 32

51 Chapter 4. Seismic hazard in northern Italy Effective depth The seismogenic layer is conventionally defined as the depth range in which the highest number of earthquakes occurs (90% of the events of each zone is here generated). Thus it is the range in which next seismic events are likely to occur. The upper and lower boundaries of the seismogenic layer have therefore been identified at depths including a cumulative number of events of 5% and 95% respectively. Inside this layer we define the Effective Depth, at which the greatest number of earthquakes occurs, determining the risk of a given zone. INGV identifies this depth as the main mode of the event frequency distribution [INGV, 2004, zonazione sismogenetica ZS9]. Figure 4.6 shows the case of ZS 911, blue arrow indicates the mode, red arrows indicate 5% and 95% of the total of events. Figure 4.6. Frequency distribution of the earthquake depths of seismic zone ZS 911 [INGV 2004, Zonazione sismogenetica ZS9] Some errors related to the procedure must be taken into account, they are due to: - Uncertainty relating to the estimation of depths - Uncertainties on the choice of mode in multi-modal distributions (in these cases the suggestion is to refer to what indicated in the database DISS 2.0 and to the depths of the main seismic sequences) Each zone of ZS9 is associated to a depth class, corresponding to the value of the effective depth. Table 4.1 shows these values for the considered seismic zones. 33

52 Chapter 4. Seismic hazard in northern Italy Seismic zone Table 4.2. Seismic depth of the considered zones and number of events of different magnitudes Md > 2.0 events Md > 2.5 events Md > 3.0 events Maximum magnitude Class of depth [km] Effective depth [km] Earthquake catalogue The information referring to the Italian seismicity is collected in the Italian seismic catalogue CPTI 04, available online at site Figure 4.7 Zoom on the central part of northern Italy of the Italian seismic catalogue (CPTI). 34

53 Chapter 4. Seismic hazard in northern Italy Figure 4.7 shows all the seismic events recorded from the ancient world to 2002 in northwestern Italy, giving the most homogeneous possible estimation of moment magnitude and surface wave magnitude. For all earthquakes are provided the following parameters: identification number, type of recording, origin time, denomination of maximum effects area, reference code, number of intensity observations, maximum intensity, epicentral intensity, location of the epicenter (latitude and longitude), magnitude, source zone, depth. Figure 4.8 shows the distribution of about 45,000 earthquakes from instrumental recordings, from small to large events for the period 1997 to Figure 4.8. Hypocemtral distribution of about selected events. Colour scale, continuously varying, indicates the depth of events (blue colours for the crustal seismicity and red colours for the mantle seismicity). The different size of circles is given by the magnitude scale indicated on the lower right corner. [Chiarabba et al 2005]. Meletti et al. [2000] elaborated a map of potential sources in Criteri e procedure per la compilazione di un inventario speditivo delle sorgenti potenziali di terremoti distruttivi finalizzato alla compilazione di una nuova mappa delle zone sismogenetiche per l area 35

54 Chapter 4. Seismic hazard in northern Italy italiana. In this map, due to the fact that in northern Italy the relationship source-geological evidence is very poorly constrained, some sources were identified from intensity data (see figure 4.9). According to what included in the inventory, geological survey alone cannot be considered exhaustive, especially because it is difficult to observe clear surficial fault activity in Italy for magnitudes less than 6. Destructive earthquakes, however, can be generated by fault whose rupture does not reach the surface (due to an insufficient energy level). Figure 4.9. Map of seismic sources determined from intensity data (geological evidence was not available) [Meletti et al. 2000]. Figure 4.10 shows events collected in the catalogue CPTI with magnitude more than 5.5, the different symbols indicate events associated to a source which could be individuated by geological survey. The colour indicates magnitude classes. As we can see, in the Lombardia Region it is possible to identify seismic sources with magnitude between 5.5 and 5.7 in the surroundings of Bergamo and Brescia. Towards the boundary with Veneto Region we can also see few sources which can be associated to events with magnitude until 6.4. It is easy to observe how known active geological structures, assiciable with event of magnitude 5.5 or more, are concentrated in central and southern Italy, where surveys on active faults have been more developed. 36

55 Chapter 4. Seismic hazard in northern Italy Figure Map of CPTI catalog events with M 5.5 [Meletti et al. 2000]. Since the literature relating to the active sources in the Lombardy Region is very poor, for this work we will refer to well constrained faults as earthquake scenario. In the following chapters we will describe these sources on which we will base our simulation. Indicatively, however, satisfactory information has been collected about the zone surrounding the town of Brescia, close to Garda Lake (Salò 2004 earthquake) and the Emilia Romagna Region (Reggio Emilia 1996 earthquake). This last source is outside Lombardy, but its effects can affect a lot our area of study. Figure 4.11 shows a map of the main seismogenic sources of northern Italy, which are capable of generating earthquakes in Lombardy region (from DISS 3.0 database [2006]). 37

56 Chapter 4. Seismic hazard in northern Italy Figure Seismogenic sources in northern Italy [DISS 3.0, 2006]. 38

57 Chapter 4. Seismic hazard in northern Italy 4.4 Seismic hazard maps Figure 4.8 shows the Italian Seismic Hazard Map proposed by INGV [2006]. Figure Map of Italian seismic hazard expressed in terms of maximum acceleration with 10% in 50 years of probability to be exceeded for stiff soil (Vs > 800 m/s; cat. A, punto del 30 D.M ). Reference: Ordinanza PCM del 28 aprile 2006 n.3519, All.1b. [INGV 2006]. We can see that the highest hazard areas are located along the Appenninc belt axes, from the Messina strait to the Garfagnana and in the Friuli area. In particular, about the Lombardia Region, we can observe very low maximum expected acceleration values: most part of the Regional territory seem to belong to the first classes of hazard, with maximum expected values of PGA in the Brescia-Bergamo zones ( g). 39

58 Chapter 4. Seismic hazard in northern Italy Unfortunately the area we are referring to in this study is very poorly defined due to lack of information. This comes evidently out from figure 4.8, where seismic intensity is represented. (see Analisi e confronti verso la nuova mappa delle massime intensità macrosismiche osservate, Camassi et al [1999]). Also for this quantity expected values are pretty low for the considered area, generally less than 5, but with peaks till 9 in the mentioned zones. Figure Maximum observed intensities from 1000 to 1992 [Camassi et al. 1999]. 40

59 Chapter 5. Simulation procedure 5. GENERATION OF SYNTHETIC SEISMOGRAMS 5.1 Overview This chapter focuses on the procedures that lead to simulation of ground motion generated from a seismic source used to carry out the analysis. Through a simulation code (Hisada [1996]) it has been possible to obtain synthetics seismograms at different observation points; this allowed for evaluating the temporal separation between the first arrivals of P and S waves and thus to define pre-alarming time scenarios at main urban areas of Lombardy. We will describe in details the characteristics of the code used in this work to represent the time histories of the ground motion. The results will be expressed in terms of velocigrams since the frequency band we are looking at refers to medium value range. In particular we will describe the general characteristics of the simulation procedure, the physics fundamentals on which it is based, the moments which marked it, and the parameters required for the computation. 5.2 Characteristics of the simulation procedure The ground motion generated from an earthquake is the result of a complex physical system which can be resumed into three main parts: - Source process, in which the seismic waves are generated by the deformation energy released from the fault rupture. - Wave propagation, during which the seismic waves travel through the crust towards the ground surface - Site effects, the seismic waves undergo further modifications while propagating through shallow geologic formations. 41

60 Chapter 5. Simulation procedure Figure 5.1. Ground motion evaluation: illustration of the three processes which characterize the generation and the propagation of seismic waves [Stewart et al. 2001] Source process: a kinematic model The modelling of the seismic source is based on the elastic rebound theory, proposed by Reid in According to this theory, as relative movement of the plates occurs, elastic strain energy is stored in the materials near the boundary as shear stresses increases on the fault planes that separate the plates. When the shear stress reaches the shear strength of the rock along the fault, the rock fails and the accumulated energy is released. The effects of the failure depend on the nature of the rock along the fault. If it is weak and ductile, the little strain energy that could be stored will be released relatively slowly and the movement will occur aseismically. If, on the other hand, the rock is strong and brittle, the failure will be rapid. Rupture of the rock will release the stored energy explosively, partly in the form of heat and partly in the form of the stress waves that are felt as earthquake [Kramer 1996]. The rupture starts from the hypocenter and propagates along the fault at a velocity close to that of shear waves in the rock. When the rupture front finds a point on the fault, slip movement starts and, in a finite time (around some seconds) it reaches its final value and stops. The time between the beginning and the end of the slip is called Rise Time. The simulation procedure employ a kinematic source model to describe the slip process, the main parameters of this model are the rupture geometry, represented by the rupture surface area, the strike and dip angles (see figure 5.2), the hypocentral depth, the rupture velocity, the direction of rupture (represented by the rake angle), and a function slip-time, including the information about the rise time and the magnitude of the final displacement. 42

61 Chapter 5. Simulation procedure Figure 5.2. Sketch for the spatial definition of the fault plane and of the slip direction [Stein and Wysession, 2003]. The displacement process on the fault is complex both in spatial and temporal terms. The movement coherent components generate long period waves, while the small scale components control the high frequency wave excitation. These small scale components are too complicated to be exactly specified, thus they are treated as casual phenomena and characterized stochastically. In order to implement source stochastic processes into the simulation procedure, the small scale casual heterogeneities are imposed on source specific parameters. For example casual perturbations can be applied at a constant rupture velocity in order to generate an irregular rupture front, so that accelerations or decelerations can generate high frequency waves. Also rise time and rake angle are treated similarly Wave propagation Wave propagation effects in the Earth crust include wave amplitude attenuation due to geometrical spreading and anelastic energy absorption, reflection and refraction at the interfaces between different lithologies (large scale heterogeneities) and wave dispersion due to small scale heterogeneities in the crust. To fully describe wave propagation effects in a crust model in a simulation procedure, we use Green s functions. Each of these functions represents the motion generated at a certain observation point by a unitary instantaneous slip on an individual fault element (see figure 5.3). Green s functions don t depend on rupture characteristics, but only on geologic structure and fault-observation point geometry [Faccioli, 2005]. 43

62 Chapter 5. Simulation procedure Figure 5.3. Sketch representing Green s functions (from Faccioli [2005]) Since the problem is quite complex, Green s functions can be calculated analytically by introducing some approximations. There are crust models considering wave propagations in a homogeneous media and others more realistic based on a layered structure. To model twodimensional or three-dimensional complex structures, finite element and finite difference methods are used, but they are still computationally slow. [Stewart et al., 2001]. An alternative approach used to obtain Green s functions for earthquake simulation is represented by the empirical Green s functions: in this case instrumental recordings of small earthquakes generated by ruptures on confined portions of the potential fault are considered as Green s functions. In fact it is legitimate to assume that the rupture mechanism of a small earthquake is simple enough to well approximate the instantaneous unitary slip which is the necessary condition in the Green s function calculation. The Green s function addition method has the advantage of integrating the characteristics of the source with those of the crust portion in which seismic waves propagate [Faccioli 2005]. This property makes the method very interesting for high frequency motion simulations, which are strongly influenced by small scale heterogeneities, beyond description by analytical way. A potential problem of empirical Green s functions is represented by the signal-noise ratio of the recordings, which is often inadequate in low frequency simulations. Practical difficulty, in addition, is represented by the acquisition of a number of small shakes recordings sufficient to cover the entire surface of the fault [Stewart et al., 2001] Representation of the ground motion: near field and far field approximation Given the fault slip model and the Green s functions, in order to calculate the global motion generated by a finite source at a site, the Representation Theorem is used. The numerical implementations of this theorem are represented by an integration which can be evaluated as 44

63 Chapter 5. Simulation procedure the triple summation of weighted and time and space delayed Green s functions over a subfaults two-dimensional grid. This last one represents the overall fault discretization. The delays are applied to Green s functions to take into account rupture propagation and the time seismic waves need to cover the distance between the fault portion and the site [Stewart et al., 2001]. 5.3 Simulation code The earthquake scenario simulations have been carried out by GRFLT [Hisada, 1996] code. This program allows us to compute the three components of the ground motion generated by an arbitrary seismic source at a generic point of a given crust model, made up of plane parallel layer series. The computation of Green s functions of a plane parallel layered half-space on which the method is based on, is partly analytic and partly numeric. The innovation introduced by Hisada in his code, with respect to previous methods, consists in expressing the integrands of Green s functions through suitable asymptotic expressions. This makes particularly efficient the computation and allows for reasonable time duration complex problem analysis (such as the one here treated). The program is constituted by three calculation codes to be executed in cascade: phs3sq, grflt12f and grfftsp2. The first code computes phase and group velocities, secular functions and the main eigenvectors for Love and Rayleigh waves. In the input file we give as parameters data for delta time, duration and period (minimum period and the imaginary omega for Phinney). The second input parameter is represented by all the characteristics of the layered half-space: these include the number of layers of the considered media and all the physical and geometrical properties, such as density, longitudinal wave velocity, quality factor for P waves, shear wave velocity, quality factor for S waves and layer thickness. The code grflt12f computes the displacements of the seismic source, it receives as input from phs3sq phase velocities data, which are the poles for the integrands in the wave number integration. In addition to circular frequencies and media characteristics, we must input other parameters such as fault characteristics. We can either considerate the source process as unique (in this case we must input single source data) or as multiple, that is composed by several fault segments, so different slip sources, each one to be individually described. In our case, due to the fact that information about the source was not exhaustive, we decided to carry out the simulation in the simple way, by considering the fault rupture as a unique process. For each source we must indicate the delay from the rupture origin, the position of the origin in terms of X-Y coordinates (corresponding to the Easting and Northing UTM coordinates), the depth, the strike, slip and rake angles, the rupture surface area (represented by its length and width), the total displacement, the rise time and the rupture velocity. 45

64 Chapter 5. Simulation procedure Finally the code grfftsp2 is used to invert the result obtained with grflt12f by Fast Fourier Transform (FFT). The output of grfftsp2 is the file wave.dat, containing the results of the simulation in terms of displacement, velocity or acceleration in the three components of the UTM coordinate system. By Matlab routines we could represent graphically these results, as we will show in the following chapter. 5.4 Comparisons of synthetic seismograms with real recordings For both the selected earthquake scenarios, we compared the synthetic results with real recordings at seismic station. For the case of 1996 Reggio Emilia earthquake recorded data referring to the main shock at a borehole station were available, while for the 2004 Salò earthquake, the only available real recordings were referred to the aftershocks of the main event. For the first case we carried out a simulation of the ground motion generated at the seismic station by the main event, while for the 2004 Salò earthquake we individuated suitable parameters to carry out a simulation of the ground motion generated by the aftershocks. In the following chapter we are going to show the results of these comparisons. 46

65 Chapter 6. Ground shaking scenarios 6. GROUND SHAKING SCENARIOS 6.1 Overview As we already mentioned, in order to study the feasibility of an Early Warning system in the Lombardy Region (northern Italy), we carried out simulations of two occurred events, of which we could find quite exhaustive information in literature: - M w 5.4 Reggio Emilia 1996 earthquake - M w 5.0 Salò 2004 earthquake This allows us to test the procedure on real and well representative cases, in order to make the study the most possible reliable. According to the seismic zoning we described in chapter 4, these two events fall in the eastern and southern zones respectively: ZS 913 for Reggio Emilia and ZS 907 for Salò. The first of the two sources is outside the region, nevertheless we decided to base our simulation on it since quite good information about the mechanisms and characteristics of the seism were available and the proximity to the Regional boundaries ensure the fact that Lombardy would be affected by an event generated by this source. In the following paragraphs we are going to detail all the characteristics of the two scenarios in terms of seismic sources parameters, event magnitudes, focal mechanisms and generated slip. On this basis we set up the simulation. 6.2 Reggio Emilia 1996 earthquake Seismogenic model As mentioned, the Reggio Emilia seismic sequence occurred on the southern edge of the Po plain, in the transition zone between the active extension zone of the Appenninic chain and the buried compressive structures of the Po plain [Selvaggi et al., 2001]. Studies about the tectonic evolution of the Tyrrhenian-Appennine system have shown that, since the late Miocene, an E to NE migration of the contemporaneous extension-compression process towards the Adriatic foreland, interested the system as a result of the retreat of the subducting Adriatic lithosphere (see Malinverno and Ryan [1986], Royden et al. [1987]). Extension affects the western area, while compression, accommodated by thrust faults presently forming an almost continuous front, is confined to the eastern sector with respect to the Appenninc 47

66 Chapter 6. Ground shaking scenarios chain. Figure 6.1 shows a simplified seismotectonic model for northern Italy, including thrust fronts of the Po plain, instrumental seismicity of the past 15 years and historical seismicity. P- axis directions are also indicated for the main seismic sequences. Figure 6.1. (a) Simplified seismotectonic setting of northern Italy. Red lines are the thrust fronts of the Po Plain (PTS, Pede-apenninic thrust front; MTF, Monferrato thrust front; EFTF, Emilia±Ferrara±Romagna folded arcs). Small circles are instrumental seismicity of the past 15 yr recorded by RSNC and empty squares are historical seismicity (from Boschi et al. 1995). Thick bars are the P-axis directions of the Parma 1971, Parma 1983 and Reggio Emilia 1996 earthquakes and the Caorso (Ca) and Cesena±Forli (Fo) seismic sequences. Thin lines are traces of sections in Fig. 2. Filled squares are the main towns of the region (Bo, Bologna; Fe, Ferrara; RE, Reggio Emilia; Pr, Parma). From Selvaggi et al. [2001] The two main compressional structures are the Pede-Appenninic thrust front, in the foot-hills of the chain, and the arcuate and buried outer arcs (the Monferrato, Emillia and Ferrara- Romagna folded arcs). Through geophysical reflection profiles, these structures have been imaged, appearing to be continuous down to depths of 8-10 km (see figure 6.2). Their continuation at greater depths in the basement is mainly inferred from structural considerations rather than from direct observations in the seismic profiles [Pieri and Groppi, 1981]. Historical seismicity shows that the Pede-Appenninc foothills are characterized by an almost continuous and narrow band of seismicity with estimated magnitude generally of about or lower [Boschi et al., 1995] (see figure 6.1). Only one earthquake had a magnitude of almost 6.0. Two earthquakes of magnitude 5.7 and 5.1 occurred in this region in 1971 and 1983 respectively. The focal mechanisms of the 1971 event was computed by Anderson and Jackson [1987], showing a reverse solution with a large strike-slip component. The 1983 fault 48

67 Chapter 6. Ground shaking scenarios plane solution shows a reverse mechanism with a small strike-slip component. The P-axes of both the 1971 and the 1983 earthquakes are subhorizontal and strike approximately NW [Selvaggi et al., 2001]. Figure 6.2. Subsurface geological structure interpreted from active seismic data (modifed from Pieri & Groppi 1981). From Selvaggi et al [2001]. The 1996 Reggio Emilia earthquake is the largest instrumentally recorded earthquake in this region. The main shock caused moderate and severe damage in Reggio Emilia and other small town in the Po plain. The macroseismic intensity was estimated to be of degree VII on the MCS scale [Selvaggi et al., 2001], with local magnitude of 5.1 and moment magnitude 5.4. Further details about the source are given in the following paragraphs Source characteristics and model parameters The focal mechanism solution of the 1996 Reggio Emilia earthquake main shock, computed by the Harvard group, shows a dominant compressive mechanism with a strike-slip component. The main source parameters, are the following (see Selvaggi et al. [2001]): - Nodal plane 1: strike 94 dip 54 rake 132 ; - Nodal plane 2: strike 217 dip 53 rake 47 ; - Fault plane solution: sub-horizontal P-axis oriented NW-SE, parallel to the Appenninc chain and perpendicular to the compressive front of this part of the Po plain; - Scalar seismic moment from CMT solutions: 1.46 x N m (M w = 5.4). In order to carry out the simulation of the ground motion produced at given observation points by this seismic source, we applied the empirical relationship found by Wells and Coppersmith 49

68 Chapter 6. Ground shaking scenarios [1994]. These regression relationships allow for prediction of the mean values of the main source parameters through correlations with seismic moment (and thus moment magnitude). They are based on statistical analysis of worldwide historical earthquake data. In particular, since we are dealing with a reverse fault with strike-slip component, and since the condition of reverse mechanism provide the most burdensome case, we applied relationships suitable for reverse movement: - Surface rupture length: L R = 3.5 km, obtained from: log L 0.63M 2.86 log = 0.20 = w Where log indicate logarithm on base ten. - Rupture area: A = 20 km 2, from: log A 0.98M 3.99 log = 0.26 = w - Maximum surface displacement D = 0.53 m, from: log D 0.29M 1.84 log = 0.42 = w Earthquake location at depth comes from analysis of well-located aftershocks, Selvaggi et al. [2001] indicates the following: - Main shock hypocenter location: N E - Main shock hypocenter depth: 15 km From aftershock locations we could also define a value for the rupture area in terms of fault length and width, they resulted to be: - Fault length: L = 9 km - Fault width: W = 3 km We decided to use these values as input for the simulation code instead of the area obtained from Wells and Coppersmith [1994] relationships since they appeared to be more representative for the case we are considering. - Rupture velocity: 2700 m/s In order to the code to run properly, we decided to carry out the simulation within an intermediate range of frequencies, we will therefore show the ground motion results in terms of velocities, which are the ground motion parameter that better represent the selected frequency range. The maximum frequency resolved by the code is thus 2.5 Hz, with a imaginary omega for Phinney of 0.1. The time step was chosen to be of 0.2 seconds, and the number of time steps was 1024, for a signal duration of seconds. 50

69 Chapter 6. Ground shaking scenarios As concerning to the crust model we applied for the simulation, we referred to what mentioned in chapter 3.6, we defined a 6 layer model, with a series of sedimentary deposits at shallow depths and limestone basement. In absence of precise value regarding the P and S wave velocities in the different layers, suitable values were obtained from literature. Table 6.1 summarizes all the parameters we used for the different layers. Table 6.1. Parameters of the crust model used for the simulation of the 1996 Reggio Emilia earthquake Layer Density P wave velocity P wave quality factor S wave velocity S wave quality factor Layer thicknes Synthetic seismograms In this part we are going to show the results of the simulation, supposing to have a series of sensors, disposed at constant spacing between each other, along a straight line crossing roughly S to N all the Regional Territory, from Reggio Emilia to Milan. The maximum distance for which we did the computation was 142 km. As mentioned before, the result we got from the simulation of the ground motion generated by the selected scenario at given observation points is expressed in terms of velocity, which is the best representative parameter in case of an intermediate frequency values range. We simulate the ground motion at 21 observation points, aligned at constant distance from each other, from the epicenter (close to Reggio Emilia) to Milan The result is a series of velocigrams virtually recorded at a distance of around 7 km one from the other. The following pictures show the velocigrams obtained at each observation point in the three components: horizontal N-S, E-W and up-down. From the pictures it is easy to see the effect of the distance in the wave propagation, P wave and S wave arrivals are pretty well distinguishable, in particular in the case of the horizontal components, followed, at higher distances from the epicenter by surface waves. 51

70 Chapter 6. Ground shaking scenarios Figure 6.3. Horizontal (N-S) component of the velocigrams at different observation points. Figure 6.4. Horizontal (E-W) component of the velocigrams at different observation points. 52

71 Chapter 6. Ground shaking scenarios Figure 6.5. Vertical (U-D) component of the velocigrams at different observation points. In our analysis we also considered the effect of the direction on the synthetic velocigrams: in this case we fixed a distance of 50 km away from the epicenter and we simulate a circular disposition of receivers. Figure 6.6 shows the result of the simulation, the origin of directions coincides with the rake of the fault. The north direction is oriented upwards, it is easy to recognize the direction of rupture looking at the peak of shaking. We can observe as, following the rupture sense, velocigrams show the highest peaks of ground motion, but the duration of the shaking is a little shorter than in the direction opposite to the rupture propagation. Along directions orthogonal to the rupture the ground motion is much weaker. In this case thus, town located N-NW of the epicenter will be affected by the highest level of ground motion and centres located S-SE of the epicenter will be struck by a shaking a little weaker, but of more cycles. Towards the East and West the ground motion has long duration and intermediate level of magnitude, while centres locates in NE-SW direction will be affected by a very low level of ground motion. 53

72 Chapter 6. Ground shaking scenarios Figure 6.6. Directivity effect on the synthetic velocigrams and fault mechanism Comparison with recorded data Since recorded data were not currently available, we obtained information about real data from the work of Selvaggi et al. [2001], where seismic signals on the N-S and the E-W components recorded by the borehole station for the main shock are reported. Hereafter we show these real data compared with our synthetic velocigrams. Figure 6.7. shows the location of the borehole station with respect to the epicenter of the Reggio Emilia 1996 earthquake. 54

73 Chapter 6. Ground shaking scenarios Figure 6.7. Map of the area showing the location of the epicentres and of the seismic stations; focal mechanisms for the main shock and the aftershocks are reported. Grey triangles indicate the recording sites of the AGIP network, while the bigger square represents the borehole station. (From Selvaggi et al. 2001). Figure 6.8 shows the recorded signals, while in figure 6.9 we reported the results of the simulation of the effects generated by the main shock at the borehole station. We can observe good agreement between the amplitudes of the two signals. Figure 6.8. Seismic signals on the N-S and the E-W components recorded by the borehole station for the main shock. Amplitude units are m/s. (From Selvaggi et al 2001). Figure 6.9. Synthetic velocigrams on the N-S, the E-W and the U-D components simulated by Hisada [1996] code at the borehole location for the main shoch. 55

74 Chapter 6. Ground shaking scenarios Spatial attenuation of synthetic PGV: comparison with attenuation relationships from the literature We studied the attenuation of the synthetic velocigrams and accelerograms (obtained by differentiation from the formers) in order to compare with existing relationships from literature,in particular we selected the relationships of Sabetta and Pugliese [1996] and of Ambraseys [1996] as the most representative for the Italian case. The results we found are showed in figures 6.10 and 6.11 for the accelerograms and for the velocigrams respectively. As we can see there is pretty good agreement between the simulated data and the relationship of Sabetta and Pugliese for what regards the velocity attenuation, while the agreement is quite bad for the case of Ambraseys for both ground motion parameters. This can be due to the fact that the relationship of Ambraseys can be considered more suitable in case of acceleration values, here obtained just by differentiation from the simulated velocities, while Sabetta and Pugliese relationship is more convenient in an intermediate range of frequencies, which is what we are taking into account in our analysis. For this reason we considered the results as pretty satisfactory. Figure Attenuation of the ground motion in terms of acceleration, a) vertical component of synthetic acceleration compared with the relationship of Ambraseys [1996]; b) horizontal components of synthetic acceleration compared with the relationships of Sabetta and Pugliese [1996] and of Ambrasyes [1996]. 56

75 Chapter 6. Ground shaking scenarios Figure Comparison between the attenuation of the synthetic velocities with the attenuation laws of Sabetta and Pugliese [1996] and Ambraseys [1996]. We simulated the ground motion along different directions, in order to observe how the attenuation of synthetic PGV and PGA varies with direction. The following figures show the results for the different component of motion. We can see that the attenuation is quite homogeneous for the three components as direction varies. Figure Attenuation of synthetic velocity and acceleration with respect to direction (the angles represent azimuth with respect to the north); horizontal N-S component. 57

76 Chapter 6. Ground shaking scenarios Figure Attenuation of synthetic velocity and acceleration with respect to direction (the angles represent azimuth with respect to the north); horizontal E-W component. Figure Attenuation of synthetic velocity and acceleration with respect to direction (the angles represent azimuth with respect to the north); vertical up-down component. 58

77 Chapter 6. Ground shaking scenarios Temporal separation of P and S phases: scenario of pre-alarming times The aim of the study was to evaluate the feasibility of an Earthquake Early Warning system, the objective was therefore to analyse the temporal separation between P wave and S wave arrivals at different observation point in order to evaluate whether a suitable warning can be issued to facilities. To get the temporal separation between P-wave and S-wave arrivals we carried out the picking of first arrivals on the synthetic velocigrams by Matlab code. In each case we selected the component for which the arrivals were the most clearly identifiable. Figure 6.15 shows the result of this operation. Figure Picking of first P-wave (in pink) and S-wave (in red) arrival on the vertical up-down component of ground velocity. Figure 6.16 shows the relation between the time separation between the two arrivals and the distance of the observation point from the epicenter. As we can see, we could evaluate an almost linear relationship between time separation and distance from the epicenter, this relation can be expressed as following: T D e Where D e is the distance of the observation point from the epicenter. 59

78 Chapter 6. Ground shaking scenarios Figure Time separation between P-wave and S-wave arrivals at different locations as function of the distance of the observation point from the epicenter. As we can see from figure 6.16, there is an almost linear relationship between separation time and distance from the epicenter at a given observation point. This linear relationship can be expressed as following: T D e Where D e is the epicentral distance. We can hypothesize to install a seismic station near the epicenter location, as reference for recording the first p-wave arrival (we fix at this instant the origin of time), estimating the magnitude of the event and transmitting the alert to the hospitals. In this way we can roughly evaluate what could be the warning time the system would be able to issue to the different urban areas. Fixing the zero time at the moment the first wave strikes this hypothetic station and considering an average transmission time of 3 seconds, we collected in table 6.2 the time of P-wave arrival, temporal separation, time of S-wave arrival and warning time (all referred to the imposed origin of times) for the main towns in the Lombardy Region. 60

79 Chapter 6. Ground shaking scenarios Table 6.2. Distance, main instants and warning times issued to the major towns in Lombardy after the first recording at the reference station. Distance from Salò (km) First P-wave arrival (s) Delta T (s) First S-wave arrival (s) Warning time (s) Cremona Pavia Bergamo Milano As we can see in the previous table, the warning time becomes relevant for distances larger than 50 km from the seismic station, where the pre-alert can be issued around 15 s ahead the arrival of stronger motion. Once estimated the warning that could be issued to the sensitive objectives of the study, we evaluated, what would be the level of shaking at the same observation points considered in the study of temporal separation, in order to evaluate the real effectiveness of an EEWS. Figures 6.17 and 6.18 show this relationship (in terms of PGA and PGV respectively), picking out a drawback which is characteristic of low-seismicity regions as the one considered in this work. As showed in figure 6.10, in fact, a delta time of 8 seconds (as that we estimated for Cremona for example) corresponds to a PGA of less than g, which is absolutely non-damaging, making useless the alert. 61

80 Chapter 6. Ground shaking scenarios Figure Relation between the time separation between arrival of P and S waves at a given point and the ground shaking level suffered by the same location. The level of shaking in fact, decreases fast getting further away from the epicenter, this means that centres which could be issued with a good warning will not actually suffer damages from the ground motion, whilst town located closer to the epicenter, and thus subjected to much higher levels of shaking, will not be alerted with sufficient anticipation. This decreases strongly the effectiveness of an EEWS in this kind of regions. Figure Relation between the time separation between arrival of P and S waves at a given point and the ground shaking level suffered by the same location. The study we have carried out until now war referred to the selected earthquake scenario, with moment magnitude M w = 5.4. We tried to vary this parameter keeping constant the linear relationship we found between temporal separation and distance from the epicenter. We simulate the variation of temporal separation for three different values of moment magnitude: 4, 5 and 6. Figure 6.19 shows the results of the simulation. Lines of the same colour are the boundaries in between events represented by standard deviation of 1 or 2 can fall. As we can see, only for greater earthquakes the installation of an EEWS could be actually effective. 62

81 Chapter 6. Ground shaking scenarios Figure Variation of temporal separation between P-wave and S-wave arrivals for different moment magnitude values Comparison with a simple geometric model As mentioned before, by picking the first arrivals of P- and S-waves at observation points on synthetic velocigrams, we could define the temporal separation between the two events, identify a nearly linear law governing its relationship with the distance from the seismic source and finally define pre-alerting time scenarios at different locations in Lombardy. We wonder then whether the same result could have been found by simply applying a geometric model, based on Snell s law for plane waves propagating into a layered medium and geometric ray theory (see figure 6.16). This approach, studying wave propagation using ray paths, does not fully describe important aspects of wave propagation, for example it would not be able to provide information regarding the amplitude of the ground motion striking a given observation point. Its employment in determining temporal separation could however be of great utility due to the simplification it would bring to the whole analysis. We therefore defined a model with the characteristics of the layered medium in terms of thickness and P- and S-wave velocity in each layer. We applied the Snell s law to find the path followed by the rays propagating from the seismic source to the ground surface, given by the following expression: V i Vi 1 = sin sin i i 1 Where Vi is the wave velocity in the layer i and i is the incidence angle of the ray travelling from the layer i to the layer i-1. 63

82 Chapter 6. Ground shaking scenarios Figure Sketch illustrating a ray path travelling from a seismic source to the ground surface in a layered medium. Applying the Snell s law to the crust model previously defined, we found the relationship between travel time from the source to the ground surface of P- and S- waves and the distance from the epicenter of the observation point at surface. Figure 6.21 illustrates this relationship. Figure Geometric model: travel times of P- and S- waves from source to ground surface as function of the epicentral distance. 64

83 Chapter 6. Ground shaking scenarios We applied then the previous model to the configuration defined in the simulation of ground motion generated by the 1996 Reggio Emilia earthquake. That is, we calculated the temporal separation between P- and S- wave first arrival at given observation points with the simplified model and we compared the results with what we obtained before by carrying out a rigorous simulation. Figure 6.22 shows the comparison between the results of the two approaches. Figure Comparison of temporal separation between P- and S-waves first arrivals as function of the distance from the epicenter obtained by rigorous simulation and by simplified geometric model. We can observe as the agreement between the two models is very good for intermediate to large epicentral distances, while in the near field the two outcomes cannot be compared. This is due to the fact that with a geometric model we consider the source as punctual and we are not taking into account the rupture process along the fault. In the simulation code, in fact, we hypothesized a rupture velocity of 2700 m/s, while in this last case the velocity corresponds always to that of the layer in which the ray is travelling, which is larger tan the rupture velocity for the two deepest layers. This effect is weakening as distance increases. This explains the higher time difference between the two arrivals of the geometric model at short distance from the epicenter. We concluded therefore that a geometric model is too simplistic to fully reproduce the ground motion produced at a given observation point and cannot be employed as substitute of a rigorous simulation in EEWS analysis. 65

84 Chapter 6. Ground shaking scenarios The geometric model, however, can be used to evaluate the sensitivity to the parameters of the whole analysis as we will show in the following paragraph Sensitivity to the model parameters In order to understand which parameters are affecting stronger the simulation of the ground motion produced by a given earthquake at certain observation points, we carried out a study of sensitivity by means of the geometric model we described in the previous paragraph. With this aim, we wrote a Matlab routine which allowed for randomly varying thickness and P- and S- wave velocities of each layer, leaving unvaried the model geometry. The code starts with the configuration defined in the simulation and then wit varies all parameters by a 10% in each cycle for 3000 cycles. For each combination of the parameters it computes the temporal separation between P-and S-wave arrivals at the same observation points we defined for the computation with the simulation code [Hisada, 1996]. In figure 6.23 we show the travel time curves for all the parameter combinations taken into account in this sensitivity study, while figure 6.24 illustrates the underground features for all combinations, highlighting combinations corresponding to smaller and larger temporal separations. As we can see, smaller time differences are given by faster layer combinations and vice-versa. Figure All Temporal separations between P- and S-waves first arrivals at given observation points deployed on the ground surface that can be obtained varying the main parameters in the geometric model. 66

85 Chapter 6. Ground shaking scenarios Figure Set of the whole parameter combinations used in the geometric model for the sensitivity study. In blue all the combinations, in red the ten fastest and in green the ten slowest. By representing the time arrival difference as function of the main parameters, we are able to evaluate the influence of each of them in the model. In particular we observed how the delta time varies with respect either to S-wave or P-wave velocity of the different layers. In the following pictures we show these relationships. In particular, if we consider the near field, most of the influence on the synthetic result comes from the characteristics of the two thickest layers (in this case the forth and the fifth), as shown in figure 6.25 and 6.26, while the correlation is less strong for the bedrock characteristics. The dependency on the wave velocities of the shallow layers is irrelevant. We can also see that the correlation is stronger if we refer to the S-wave velocities than to the P- wave velocities. As getting further from the epicenter, on the other hand, we can observe how the temporal separation between P- and S-first arrivals is more and more linked with the characteristics of the bedrock, especially with the S-wave velocity of the layered media. This is shown in figures 6.27 and Similar correlation could be found between temporal separation between P- and S-wave first arrivals and P- and S-wave velocities in the layered media: in the near field we observed good correlation between the two quantities in the thickest layers, while in the far field the strongest influence was coming from the bedrock characteristics. 67

86 Chapter 6. Ground shaking scenarios Figure Near field: correlation between temporal separation and P-wave velocity in the layered media. Figure Near field: correlation between temporal separation and S-wave velocity in the layered media. 68

87 Chapter 6. Ground shaking scenarios Figure Far field: correlation between temporal separation and P-wave velocity in the layered media. Figure Far field: correlation between temporal separation and S-wave velocity in the layered media. 69

88 Chapter 6. Ground shaking scenarios The following two figures show better the previous correlations. Figure Near field: correlation between temporal separation and S- (in blue) and P-wave (in red) velocity in the layered media. Figure Far field: correlation between temporal separation and S- (in blue) and P-wave (in red) velocity in the layered media. 70

89 Chapter 6. Ground shaking scenarios We could not find any relevant correlation between the temporal separation and the thickness of each layer in the medium. The previous analysis allows us to determine which can be the variability of the gap between the first arrivals of P- and S-waves for different crust models: between the extreme cases the temporal separation difference is up to 3-4 seconds in the near field and to seconds in the far field. In conclusion we can therefore state the validity of the prediction provided by the geometric model in terms of temporal separation between P- and S-waves for intermediate to large distances from the epicenter. The main utility of this model is the possibility to carry out a sensitivity analysis and individuate which parameters and which combinations of parameters are most highly affecting the case in analysis, in order to predict in a simplified way the gap between the two events. We demonstrated how the temporal separation in near field is mostly affected by the characteristics of the two thickest layers, the following picture is aimed at emphasizing this dependency by showing the relationship between time difference cumulated in these two layers and the temporal separation between P- and S-wave first arrivals at observation points located in near field. Figure Near field, relationship between the temporal separation between P- and S-waves cumulated in the two thickest layers (fourth and fifth) and the temporal separation between P- and S-wave first arrivals at an observation point located in proximity of the epicenter on the ground surface. 71

90 Chapter 6. Ground shaking scenarios In conclusion we can therefore say that the parameter which really affects the simulation is the average velocity along the vertical direction, independently from the values in the shallowest layers. In this way the results of the simulation can be considered pretty plausible and the estimations of temporal separation adequately reliable. As we can see from the previous picture in fact, the uncertainty linked to the temporal separation for different soil configurations is around seconds. In the far field, on the other hand, we observed how the temporal separation was mostly affected by the characteristics of the underlying bedrock, in the following picture we highlight this property by showing the relationship between the delta V of P- and S-wave velocities in the deepest layer and the temporal separation at an observation point located 142 km away from the epicenter on the ground surface. Figure Far field, relationship between the velocity difference of P- and S-waves in the bedrock and the temporal separation between P- and S-wave first arrivals at an observation point located 142 km away from the epicenter on the ground surface. The previous observation allows us to conclude that, at large distances from the epicenter, the parameter that most strongly affects the model is the maximum velocity of seismic waves in the layered medium, it is therefore straightforward to estimate the sensitivity and the uncertainties of the model. In this case the uncertainty related to the temporal separation is bigger than in the previous case, around seconds. 72

91 Chapter 6. Ground shaking scenarios Globally we can hence affirm that the rigorous simulation is absolutely useful in order to get information about the arrival times of seismic waves and on the amplitude of the ground motion at given observation points. On the other hand, however, the geometric model allows us for quick and pretty simple model calibration. Having demonstrated that the temporal separation depends mostly on the characteristics of the layers at larger depths, a good calibration of the model could be achieved by means of only two recordings at different stations. In the following picture we summarize the contribution of the main parameters of the model to the results in the Montecarlo simulation we carried out. As mentioned before, during that phase we allowed all main parameters to freely vary. For each layer we show the distribution of events as function of a given parameter. Figure Montecarlo simulation: bar distributions of the variable parameters for each layer Comparison of synthetic and real pre-alarming times 6.3 Salò (Brescia) 2004 earthquake Seismogenic model On November 24, 2004 a moderate earthquake struck the western side of Garda Lake, in northern Italy, with a moment magnitude Mw = 5.0 (see Esposito et al. [2005] and Pessina et al. [2006]. This area is located within the active fold and thrust belt of Southern Alps, which formed in the framework of the Africa-Europe convergence since the Miocene. The Garda Lake area is affected by the Giudicarie fault system, related to NNE-SSW trending thrusts and transpressional structures. The main active structures of the Southalpine sector are usually blind thrust, geomorphological and paleoseismological investigations carried out in this area, in fact, did not identify the major fault which can be responsible for large earthquakes (see Galadini et al., [2001]). Focal mechanisms [Slejko et al., 1989; MEDNET, 2006] and 73

92 Chapter 6. Ground shaking scenarios minimum horizontal stress from breakout data [Montone et al. 2004] and GPS measurements [D Agostino et al., 2005] show kinematics consistent with the activity of these thrusts, with minor oblique component. According to Pessina et al. [2006], the November 2004, Salò earthquake confirms the mechanism of tectonic deformation in the South-Alpine chain, and the style of seismogenic faulting of the Giudicarie system region, as suggested in the DISS. They therefore hypothesized that the event ruptured a portion of the Giudicarie thrust system located on the same thrust fault of the 1901 earthquake seismogenic source at larger depth. This agreement between the two event locations can be seen in figure 6.16, showing a seismotectonic sketch of the area. Figure Seismotectonic sketch of the Garda Lake area (from Pessina et al. [2006]). The earthquake was felt in the whole northern Italy and abroad (Switzerland for instance), the epicentral area includes Salò and its surroundings, where significant damage occurred and more than 200 people were left homeless [Esposito et al., 2005]. 74

93 Chapter 6. Ground shaking scenarios Figure Map of the area including instrumental epicenter of the November 2004 Salò earthquake, focal mechanisms and field stations where observations were collected [Esposito et al. 2005]. On the base of the damages at the villages of Clibbio and Pompegnino, epicentral intensity was preliminary evaluated of VII-VIII in the MCS scale. Figure 6.17 shows the location of the epicenter (from INGV), focal mechanisms (from the MEDNET Database) and the field stations where observations on coseismic ground effects have been collected Source characteristics and model parameters The focal mechanism solution of the 2004 Salò earthquake, was computed by INGV-Harvard group [MEDNET, 2006], it shows the northwest-dipping plane in agreement with the geometry of the thrust of the Giudicarie fault system. Large uncertainties affect the location of the hypocenter at depth, whose position could actually be between 5 and 15 km depth (Augliera, personal communication). This is also confirmed by our results: we carried out simulations locating the hypocentral depth at 5, 8 and 10 km depth, obtaining the same ground motions. This can be due to the low dip angle individuated by seismotectonic settings. The main source parameters, applied in the simulation are the following (from INGV-Harvard group and Pessina [2006]): - Nodal plane 1: strike 246 dip 24 rake 113 ; - Nodal plane 2: strike 42 dip 68 rake 80 ; - Scalar seismic moment from CMT solutions: 1 x dyn cm (M w = 5.0). 75

94 Chapter 6. Ground shaking scenarios As in the previous case, for the simulation of the ground motion produced at given observation points by this seismic source, we applied the empirical relationship found by Wells and Coppersmith [1994]. We are dealing again with a reverse fault with strike-slip component, however we preferred to keep the most burdensome condition, given by relationships individuated for reverse solutions: - Surface rupture length: L R = 2.0 km, obtained from: log L 0.63M 2.86 log = 0.20 = w Where log indicates logarithm on base ten. - Rupture area: A = 8.1 km 2, from: log A 0.98M 3.99 log = 0.26 = w - Maximum surface displacement D = 0.41 m, from: log D 0.29M 1.84 log = 0.42 = w - Hypocenter location: N E - Hypocenter depth: 5 km According to Pessina et al. [2006], suitable values for fault length and width resulted to be: - Fault length: L = 2.6 km - Fault width: W = 2.5 km We decided to use these values as input for the simulation code instead of the area obtained from Wells and Coppersmith [1994] relationships since they appeared to be more representative for the case we are considering. - Rupture velocity: V R = 2700 m/s Again we based the simulation on an intermediate range of frequencies, expressing the results in terms of ground velocities. The maximum frequency resolved by the code is thus 2.5 Hz, with an imaginary omega for Phinney of 0.1. The time step was chosen to be of 0.2 seconds, and the number of time steps was 1024, for a duration of the signal of seconds. We are interested in the synthetic ground motion that can be expected at the hospitals of Lombardy, which are mainly located in the central part of the region; we therefore applied the same crust model as in the previous case, with 6 layer composed by a series of sedimentary deposits at shallow depths and limestone basement. Table 6.3 summarizes all the parameters we used for the different layers. Table 6.3. Parameters of the crust model used for the simulation 76

95 Chapter 6. Ground shaking scenarios Layer Density P wave velocity P wave quality factor S wave velocity S wave quality factor Layer thicknes Synthetic seismograms As before, the results of the simulation, are showed here after, we carried out the computation supposing to have a series of sensors, disposed at constant spacing between each other, along a straight line crossing roughly W to E all the Regional Territory, from the epicenter (Salò) to Novara (Piedmont). The maximum distance was therefore around 150 km from the epicenter. The following pictures show the synthetic velocigrams at all observation points, for the three component of motion. In each velocigram the ground motion value has been normalized with respect to the maximum value at the same point. Figure Synthetic velocigrams (N-S horizontal component) from the epicenter (Salò) to Milan (142 km far from the epicenter). 77

96 Chapter 6. Ground shaking scenarios Figure Synthetic velocigrams (E-W horizontal component) from the epicenter (Salò) to Milan (142 km far from the epicenter). Figure Synthetic velocigrams (U-D vertical component) from the epicenter (Salò) to Milan (142 km far from the epicenter). 78

97 Chapter 6. Ground shaking scenarios We can easily recognize the first arrival of P-waves in all the velocigrams, but the vertical component is the one which better shows the S-waves arrivals. Again we considered the directivity of synthetic by fixing a distance of 50 km away from the epicenter simulating a circular disposition of 20 receivers. Figure 6.39 shows the result of the simulation, the origin of directions coincides with the rake of the fault. The north is oriented upwards, as showed in the previous case, the peak of shaking is along the rupture direction, with shorter duration along the rupture and higher peak than contrary to the rupture. Longer duration but still lower level effects are observed in the orthogonal directions. In this case thus, town located SE of the epicenter will be affected by the highest level of ground motion and centres located NW of the epicenter will be struck by a shaking a little weaker, but of more cycles. Centres locates in NE-SW direction instead will be affected by a very low level of ground motion. Figure Directivity effect on synthetic ground motion Comparison with recorded data From Pessina et al we obtained acceleration data recorded at Vallio Terme during the November 2004 Salò earthquake. This centre is located 13.3 km away from the epicenter, figure 6.40 shows the recorded accelerograms. 79

98 Chapter 6. Ground shaking scenarios Figure 6.40 Acceleration data recorded at Vallio Terme, (13.3 km epicentral distance) during the November 2004 Salò earthquake. (From Pessina et al. 2006). Unfortunately we noticed the fact that the beginning of the shaking does not appear in the recordings, maybe due to triggering delays of the sensor. By differentiating the ground velocity values obtained from the simulation, we got synthetic accelerograms at the same epicentral distance as the recorded data. The three components of ground acceleration are shown in the following pictures: Figure Synthetic accelerogram, horizontal N-S component. From. we obtained also recordings of ground motion during the 27 November 2004 aftershocks of the Salò earthquake at different seismic stations. For example, we got the data referred to the station Salò Scuola, with the following characteristics: 80

99 Chapter 6. Ground shaking scenarios Table 6.4. Characteristics of the data recorded at Salò Scuola station during aftershocks Latitude N Longitude E Epicentral area (INGV) Garda Lake zone Latitude epicenter N Longitude epicenter E Epicentral distance (km) Hypocentral distance (km) Depth (km) 5.00 M L 2.2 M d 2.7 The recordings are expressed in terms of ground acceleration, they are shown in picture We simulated the same event at the same location in order to compare synthetic results with real recorded data. Since our simulation procedure does not allow for reaching such high frequencies, we integrated the real recordings in order to get calculated data of ground velocity coming from real recordings. In this way we can compare the synthetic and the real dataset. As we can see from picture 6.42 and 6.43, we could reproduce quite well the magnitude of the ground motion, while there not such a good agreement for the waveforms. Figure Recorded data at Salò Scuola station during the aftershocks of the 2004 Salò earthquake. 81

100 Chapter 6. Ground shaking scenarios Figure Velocigrams calculated by integration from recorded data at Salò Scuola station during the aftershocks of the November 2004 Salò earthquake. Figure Synthetic velocigrams obtained by simulation at Salò Scuola station due to the aftershocks of the November 2004 Salò earthquake. 82

101 Chapter 6. Ground shaking scenarios Spatial attenuation of synthetic PGV: comparison with attenuation relationships from the literature We calculated accelerograms by differentiation from velocigrams and we compared with existing relationships from literature (Sabetta and Pugliese [1996] and Ambraseys [1996]). The results we found are showed in figures 6.45 and 6.46 for the accelerograms and for the velocigrams respectively. Again we notice quite good agreement between the simulated data and the relationship of Sabetta and Pugliese for what regards the velocity attenuation, while the agreement is quite bad for the case of Ambraseys for both ground motion parameters. This can be due to the fact that the relationship of Ambraseys can be considered more suitable in case of acceleration values, here obtained just by differentiation from the simulated velocities, while Sabetta and Pugliese relationship is more convenient in an intermediate range of frequencies, which is what we are taking into account in our analysis. For this reason we considered the results as pretty satisfactory. Figure Attenuation of peak ground acceleration (vertical and horizontal components) as function of the distance from the epicenter, comparison with relationships from literature. 83

102 Chapter 6. Ground shaking scenarios Figure Attenuation of peak ground velocity (horizontal components) as function of the distance from the epicenter, comparison with relationships from literature. In both cases, attenuation of horizontal PGA and PGV, are in a better agreement with relationships from literature in near field, while the coupling get worse as distance increases. This can be due to the fact that the code we used to carry out the simulation (Hisada [1994]) is appropriate for near field analysis. In this work we decided to extend its validity also to the far field considering reasonable the results. This hypothesis is supported by the fact that we need a more rigorous analysis in the near field, while a large amount of uncertainties and variables can come up when treating with bigger distances from the fault, with the present day available tools is in fact impossible to carry out a rigorous analysis also in far field. We considered a possible directivity effect also regarding the attenuation law, so we repeated the simulation for different directions with respect to the rupture. Pictures 6.47, 6.48, and 6.49 show the result for all the components of ground motion. As we can see from these pictures, the attenuation is stronger for the two horizontal components than for the vertical component of simulated ground motion. 84

103 Chapter 6. Ground shaking scenarios Figure Attenuation of peak ground velocity and acceleration (respectively left side and right side) for the N-S horizontal component with respect to the distance from the epicenter for different directions (azimuth are referred to the north). Figure Attenuation of peak ground velocity and acceleration (respectively left side and right side) for the E-W horizontal component with respect to the distance from the epicenter for different directions (azimuth are referred to the north). 85

104 Chapter 6. Ground shaking scenarios Figure Attenuation of peak ground velocity and acceleration (respectively left side and right side) for the U-D vertical component with respect to the distance from the epicenter for different directions (azimuth are referring to the north) Temporal separation of P and S phases: scenario of pre-alarming times From the synthetic velocigrams obtained by simulation at different observation points, we picked the first arrivals of P and S waves in order to get the temporal separation between the two events. Figure 6.50 shows the result of the picking, while figure 6.51 reports delta time between the two arrivals as function of the distance from the epicenter. Figure Picking of P and S waves first arrivals. 86

105 Chapter 6. Ground shaking scenarios Figure Temporal separation between P-wave and S-wave first arrivals as function of the distance from the epicenter. As we can see from figure 6.51, there is an almost linear relationship between separation time and distance from the epicenter at a given observation point. This linear relationship can be expressed as following: T D e Where De is the epicentral distance. If we think of using the seismic station of Salò as reference for recording the first p-wave arrival (we fix at this instant the origin of time), estimating the magnitude of the event and transmitting the alert to the hospitals, we can roughly evaluate what could be the warning time the system would be able to issue to the different urban areas. Fixing the zero time at the moment the first wave strikes Salò station and considering an average transmission time of 3 seconds, we collected in table 6.5 the time of P-wave arrival, temporal separation, time of S- wave arrival and warning time (all referred to the imposed origin of times) for the main towns in the Lombardy Region. 87

106 Chapter 6. Ground shaking scenarios Table 6.5. Distance, main instants and warning times issued to the major towns in Lombardy after the first recording at the reference station. Distance from Salò (km) First P-wave arrival (s) Delta T (s) First S-wave arrival (s) Warning time (s) Brescia Bergamo Piacenza Milano Pavia As we can see in the previous table, the warning time becomes relevant for distances larger than 50 km from the seismic station, where the pre-alert can be issued around 20 s ahead the arrival of stronger motion. Once estimated the warning that could be issued to the sensitive objectives of the study, we evaluated, what would be the level of shaking at the same observation points considered in the study of temporal separation, in order to evaluate the real effectiveness of an EEWS. Figures 6.51 and 6.52 show this relationship (in terms of PGA and PGV respectively), picking out a drawback which is characteristic of low-seismicity regions as the one considered in this work. The level of shaking in fact, decreases fast getting further away from the epicenter, the level of shaking at zones which could be issued with a relevant warning, will be struck by a ground motion level which could not be considered damaging. This decreases strongly the effectiveness of an EEWS in this kind of regions. We can see in fact, that points located further away from epicenter, where the delta time is higher than 6 seconds, would be struck by PGA of less than g. 88

107 Chapter 6. Ground shaking scenarios Figure P- and S-wave temporal separation as function of PGA at observation points. Figure P- and S-wave temporal separation as function of PGV at observation points. 89

108 Chapter 6. Ground shaking scenarios In practice, therefore, the towns indicated in the previous table 6.4, for which we estimated interesting times of pre-alert, would be struck by a ground shaking of almost no relevance for the anticipated purposes. Concerning to this point, we can demonstrate how the delta time would vary for different magnitudes of events, making effective the EEWS for higher magnitudes. To do this we kept the hypothesis of a linear relationship between delta time and epicentral distance, obtained from the graph in picture 6.51 and we computed variation of warning time at virtual observation points for different magnitudes, considering a possible variation of one standard deviation (all events fall in between same colour lines). Results are represented in figure 6.54, we can observe as an EEWS becomes more and more effective as earthquake magnitude increases. Figure Realtionship between temporal separation of P and S wave first arrivals and PGV as function of different earthquake magnitudes. Lines of the same colour refer to events represented by standard deviations of 1 or 2. 90

109 Chapter 6. Ground shaking scenarios Comparison of synthetic and real pre-alarming times 6.4 Future earthquake scenario: re-activation of the North Giudicarie line Overview We tested further our EEWS in the Lombardy region by defining a virtual earthquake scenario represented by a re-activation of a tectonic fault. This allows us to test the system for possible future earthquakes not yet occurred. The scenario we selected in this case is related to the North Giudicarie line, which is part of the Insubric line, in northern Italy Seismogenic model The central part of the Alpine chain hosts an important transpressive fault zone: the North Giudicarie line, a NNE trending fault which offsets the dextral Insubric line with an apparent left-lateral displacement of about 70 km. Figure 6.55 shows a simplified tectonic map of the central southern Alps and of the Austroalpine nappes. This line represent the major irregularity of the nearly E-W striking central Alpine chain [Prosser 1998]. Several evidences indicate that the Giudicarie line is a good example of a transpressive fault, developed along an inherited weakness zone. Thrust and left lateral transpression along the Giudicarie line were contemporaneous with south-directed shortening in the southern Alps during the Miocene [Laubscher, 1990 and Castellarin et al., 1992]. 91

110 Chapter 6. Ground shaking scenarios Figure Simplified tectonic map of the central southern Alps and of the Austroalpine nappes west of the Tauern Window (after Thöni [1981], Castellarin and Vai [1982], Rossi and Rogledi [1988], Schmid and Haas [1989], and Beserzio and Fornaciari [1994]). Bold lines represent the main tertiary faults of the central Alpine chain. The shear sense of these faults has been reported following Martin et al. [1991] and Schmid and Foitzheim [1993]. From Prosser [1998]. Prosser [1998], summarizing the information relating to this system, writes that the north Giudicarie line seems to be a good candidate for a transpressive fault zone controlled by strain partitioning due to pre-existing Jurassic normal faults. The footwall is characterized by N-S striking strike-slip faults, which reactivate extensional faults of early Jurassic to late 92