Tsunami risk perception in southern Italy: ﬁrst evidence from a sample survey

. The Italian Tsunami Alert Centre of the Istituto Nazionale di Geoﬁsica e Vulcanologia (Centro di Allerta Tsunami, hereinafter CAT-INGV) supported a computer-assisted telephone interview (CATI) survey to investigate tsunami risk perception in two pilot regions of southern Italy. The survey was carried out on a stratiﬁed sample of 1021 interviewees representing about 3.2 million people living in 183 coastal municipalities of the two regions, namely Calabria and Apulia. The main goal of this research is to verify whether and how people’s perception of tsunami hazard com-pares to the results of (PTHA) – probabilistic tsunami hazard assessment (TSUMAPS-NEAM project; Basili et al., 2018). As shown by the results of this project, both investigated regions are characterised by high tsunami hazard. Nonetheless, the long return time of such events could lead people to consider the occurrence of a tsunami in the Mediterranean Sea to be very unlikely. The survey results reveal that people’s risk perception is low: for almost half of the whole sample the occurrence of a tsunami in the Mediterranean Sea is considered quite unlikely, with a clear difference between Apulia and Calabria. In the latter region the risk perception is much

Abstract. The Italian Tsunami Alert Centre of the Istituto Nazionale di Geofisica e Vulcanologia (Centro di Allerta Tsunami, hereinafter CAT-INGV) supported a computerassisted telephone interview (CATI) survey to investigate tsunami risk perception in two pilot regions of southern Italy. The survey was carried out on a stratified sample of 1021 interviewees representing about 3.2 million people living in 183 coastal municipalities of the two regions, namely Calabria and Apulia. The main goal of this research is to verify whether and how people's perception of tsunami hazard compares to the results of (PTHA) -probabilistic tsunami hazard assessment (TSUMAPS-NEAM project; Basili et al., 2018). As shown by the results of this project, both investigated regions are characterised by high tsunami hazard. Nonetheless, the long return time of such events could lead people to consider the occurrence of a tsunami in the Mediterranean Sea to be very unlikely.
The survey results reveal that people's risk perception is low: for almost half of the whole sample the occurrence of a tsunami in the Mediterranean Sea is considered quite unlikely, with a clear difference between Apulia and Calabria. In the latter region the risk perception is much higher than in the former, probably due to the shorter time elapsed since the last event. Also, belonging to different coastal areas 1 appears to have a significant influence on the way tsunami haz-ard is conceived, having a stronger effect on risk characterisation: the interviewees of Tyrrhenian Calabria are indeed more likely to associate tsunami risk with volcanoes than the Ionian citizens. This is coherent considering the presence of active volcanoes and related tsunami precedents in the Tyrrhenian.
Television emerged as the most relevant source of knowledge for almost 90 % of the sample, and the influence of media also results in the way tsunami risk is characterised. In particular, the survey showed that people's perception and understanding of tsunamis are affected by media accounts of large events, such as the 2004 Sumatra and the 2011 Japan tsunamis. At the same time, it is evident that the risk posed by smaller events is underrated. Furthermore, the survey's results show that the word "tsunami" occupies a different semantic space in comparison to the Italian traditional headword maremoto, with differences among sample strata. In other words, the same physical phenomenon would be understood in two different ways by younger, educated people and elders with a low education level. The results of this study, although limited to two regions, provide a first assessment of tsunami risk perception in Italy, also entailing important consequences for both risk communication practice and mitigation policies.

Relevance of tsunami risk on the Mediterranean and Italian coasts
Almost all countries surrounding the Mediterranean have faced the effects of historical tsunamis, with more than 200 events documented for the area, starting from 1630 BCE (Santorini volcano eruption), as shown in the catalogue published by Maramai et al. (2014). According to this catalogue, most of the tsunamis in the area were generated by earthquakes ( ∼ 83 %), a fraction similar to that of other oceanic regions worldwide . Since 1700 CE, historical sources provide evidence of an average of 20 events every 50 years, as reported in the catalogue (Maramai et al., 2014), i.e. one event every 2.5 years (including small tsunamis). In addition to large historical tsunamis, such as Crete 365 CE (Papadopulos et al., 2010), at least two significant events occurred in the Mediterranean in the 20th century: the 1908 tsunami in southern Italy (which hit Messina, Reggio Calabria and the surrounding coasts) due to a magnitude 7 earthquake in the Messina Straits, with run-up as high as 13 m in Pellaro (Tinti and Maramai, 1996) and a large number of fatalities, and the 1956 tsunami in Greece, due to a magnitude 7.7 earthquake which occurred close to the Cycladic island of Amorgos (Greece) and hit the coasts of Amorgos, Astypalaia and Folegandros with run-up values of 20, 10 and 14 m, respectively (Okal et al., 2009), or even up to 30 m according to other sources (Ambraseys, 1960). More recently, in 2003 a relatively small tsunami caused by a magnitude 6.9 earthquake in Boumerdès (Algeria) hit the western Mediterranean coasts, causing damage to properties in at least eight harbours in the Balearic Islands (Vela et al., 2011). Finally, two small tsunamis occurred in the Aegean Sea in 2017 (due to earthquakes with a magnitude of 6.4 and 6.6 close to the Lesvos and Kos islands), along with the most recent one that occurred in the Ionian Sea (Zakynthos) in October 2018 (magnitude 6.8). Other potential sources of tsunamis in the Mediterranean are volcanoes, such as those presently active in the Tyrrhenian Sea. Rosi et al. (2019) have investigated the occurrence of past tsunamis in this area through geological, archaeological and carbon dating, along with historical sources. They identified three large tsunami deposits triggered by landslides which occurred at the Stromboli volcano between the 14th and 16th centuries, possibly including the one observed in Naples in 1343 and described by the poet Petrarch. A database with all the observations related to the tsunamis known in the Italian region has been recently published by Maramai et al. (2019).
Based on these and other geological data, the first probabilistic tsunami hazard assessment (of seismic origin) (S-PTHA) in the NEAM region (North-eastern Atlantic, the Mediterranean and connected seas) has been computed and published (TSUMAPS-NEAM Team, 2018). In an S-PTHA approach, the hazard at any specific point on the coast comes from the various tsunami sources affecting that point, including close and distant sources Grezio et al., 2017;Davies et al., 2017;Volpe et al., 2019). For Italy, it is evident that the most hazardous areas are those exposed to both local and distant earthquakes. In particular, the most active region in the Mediterranean is the Hellenic arc, where strong tsunamigenic earthquakes have occurred in the past (Papadopulos et al., 2010;Maramai et al., 2014). Consequently, the coastal areas of Apulia, Calabria and eastern Sicily facing the Ionian Sea, have the highest hazard in Italy. However, a significant hazard exists for many other coastal areas throughout Italy, such as the Ligurian Sea, the Adriatic Sea, and the Tyrrhenian Sea, due to both local and distant earthquake sources, such as for instance the northern African fault system from Gibraltar to Tunisia.
Despite the high hazard along the Italian coasts, the awareness of tsunamis is weak in Italy, mainly due to the long time elapsed since the most recent deadly event in 1908. During this event, the tsunami significantly increased the already heavy death toll from the earthquake. Due to the lack of tsunami awareness, many people escaping from the earthquake and buildings in Messina and other towns headed to the sea. After more than a century since this tragedy, we do not know if any memory of it is left in the region.
Another recent event that could have modified the perception of tsunami risk in Italy was the collapse of the unstable flank of the volcanic island of Stromboli in 2002. This generated a local tsunami with measured run-up of up to 10 m (Tinti et al., 2005).  (Marriner et al., 2017). The Mediterranean Sea also fosters a thriving maritime economy: according to the estimates of a WWF-BCG report, economic activities related to the Mediterranean are worth USD 450 billion per year (Randone et al., 2017). According to the Italian National Institute of Statistics (IS-TAT), in the 15 Italian coastal regions there are 644 coastal municipalities, i.e. representing 8.0 % of the total number of municipalities in Italy. These municipalities cover an area of 14.3 % of the surface area of the whole country, hosting a population of 28.4 % of the entire Italian population (more than 17 million inhabitants). Furthermore, coastal areas have the greatest population density, with 400 inhabitants per square kilometre, compared to an average of 168 for the internal areas. They also show a population growth rate which is more than twice that of the non-coastal areas, i.e. +3.3 % vs. +1.6 % for the remaining areas (ISTAT, 2016). This means that tsunami risk exposure is relevant and constantly growing, suggesting the need for effective mitigation measures. Following the establishment of ICG/NEAMTWS, in 2013 Italy started to build a tsunami alert centre at the Istituto Nazionale di Geofisica e Vulcanologia (INGV). Although residents in coastal municipalities are not all located in the inundation areas which might be directly hit by a tsunami, they are likely to face indirect consequences of a tsunami event. After a 3-year testing phase, the CAT-INGV became operational in 2016, after being formally accredited as a Tsunami Service Provider (TSP) for the whole Mediterranean area by the ICG/NEAMTWS. Together with CAT-INGV, in 2016 three other centres were accredited as TSPs within the NEAM region, namely CENALT (France) for the northeast Atlantic and western Mediterranean; KOERI (Turkey) for the eastern Mediterranean, Marmara and Black seas; and NOA (Greece) for the eastern/central Mediterranean. Moreover, IPMA (Portugal) is currently a candidate TSP for the Atlantic. Soon after that, the CAT-INGV began providing national alerts within the framework of the SiAM (Sistema d'Allertamento nazionale per i Maremoti di origine sismica), coordinated by the Italian Department of Civil Protection (hereinafter DPC), and together with the Istituto Superiore per la Protezione dell'Ambiente (ISPRA), which manages the national sea level network.
As a Tsunami Service Provider, CAT-INGV sends alert messages to about 15 countries and institutions of the Euro-Mediterranean region in case of potentially tsunamigenic earthquakes. At the national level, CAT-INGV cooperates strictly with DPC and ISPRA for disseminating alert messages to the local authorities and the population. In addition, CAT-INGV is involved in increasing knowledge and people's awareness of tsunami hazard and risk.
Although the Italian tsunami early warning system was established in 2017, a comprehensive risk communication strategy is still under development. The main goal of this study is to contribute to risk mitigation efforts by identifying people's knowledge and perception and helping to define the best strategy.

Background: risk perception studies
Risk perception involves the process of collecting, selecting and interpreting signals about uncertain impacts of events, activities or technologies. These signals can include direct observation or information from others (for example reading about an earthquake in the newspaper). Perceptions may differ depending on the type of risk, the risk context, the personality of the individual, and the social context. Within natural sciences the term "risk" seems to be clearly defined and measured as the probability distribution of adverse effects. However, the everyday use of the word risk has different connotations (Renn, 2008). For social sciences the term "risk perception" has become the conventional standard (Slovic, 1987). Current studies on risk perception, grounded in sensorial human perception research from the beginning of the 20th century (Wagemans et al., 2012), have shown that people's perception is affected by cognitive processes and past experience. The mental models and other psychological mechanisms that people use to judge risks (such as cognitive heuristics and risk images) are internalised through social and cultural learning and constantly moderated (reinforced, modified, amplified or attenuated) by media reports, peer influences and other communication processes (Morgan et al., 2002).

Theoretical references of studies on risk perception
In recent decades, many research studies have been carried out on psychological, social and cultural factors that influence the perception of risk. At present, the perception of risk is considered fundamental to understand what lay people think about risks and to adopt suitable political and communication strategies. Renn and Rohrmann (2000) developed a structured framework that provides an integrative and systematic perspective on risk perception. Figure 1 illustrates this perspective by suggesting four distinct context levels (originally presented by Renn and Rohrmann, 2000: 221; adapted from the generic model of Breakwell, 1994).
The first level, from the bottom centre in Fig. 1, includes the collective and individual heuristics that individuals apply during the process of forming judgements. These heuristics are independent of the specific nature of the risk, personal beliefs, emotions or other conscious perception patterns of the individual. Heuristics represent common-sense reasoning strategies that have evolved over the course of biological and cultural evolution (Ross, 1977;Kahneman and Tversky, 1979;Breakwell, 2007) and may differ among cultures. However, psychological research shows a surprising degree of convergence across different cultures (Renn and Rohrmann, 2000).
The second level refers to the cognitive (knowledge-based) and affective (emotion-based) factors that influence the perception of specific properties of the risk in question. Cognition of a risk source -what people believe to be true about a risk -governs the attribution of qualitative characteristics (psychometric variables) to specific risks (e.g. dread or personal control options) and determines the effectiveness of these qualitative risk characteristics for the perceived se- riousness of risk and the judgement about its acceptability (Slovic, 1992). Recently, psychologists have discovered that emotions play an important role in people's decision processes (Loewenstein et al., 2001;Slovic et al., 2002). People's feelings about what is good or bad in terms of the causes and consequences of risks colour their beliefs about the risk and, in addition, influence their process of balancing potential benefits and risks.
The third level refers to the social and political institutions that individuals and groups associate with either the cause of the risk or the risk itself. Most studies on this level focus on trust in institutions, personal and social value commitments, organisational constraints, social and political structures, and socio-economic status. One important factor in evaluating risk is the perception of fairness and justice in allocating benefits and risks to different individuals and social groups (Linnerooth-Bayer and Fitzgerald, 1996).
Other studies have placed political and social organisations, and their strategies to communicate with other organisations and society at large, as the prime focus of their attention (Clarke, 1989;Shubik, 1991). Press coverage appears to contribute substantially to a person's perception of risk, particularly if the person lacks personal experience with the risk and is unable to verify claims of risks or benefits from their own experience. In contrast to popular belief, however, there is no evidence that the media can create opinions about risks or even determine risk perceptions. In fact, studies on media reception suggest that people select elements from media reports and use their own frame of reference to form their opinions. Most people reconfirm existing attitudes when reading or viewing media reports (Peters, 1991;Dunwoody and Peters, 1992;Breakwell, 2007).
The last level (the external frame in Fig. 1) refers to cultural factors that govern or co-determine many of the lower levels of influence. The most specific explanation for cultural differences about risk perceptions comes from the so-called "cultural theory of risk" (Douglas and Wildavsky, 1983).
All four levels of influence are relevant in order to gain a more accurate understanding of risk perception. In spite of many questions and ambiguities in risk perception research, one conclusion is beyond any doubt: abstracting the risk concept to a rigid formula and reducing it to the two components' "probability and consequences" do not match people's intuitive thinking of what is important when making judgements about the acceptability of risks (Slovic, 1992). The framework of social amplification may assist researchers and risk managers to forge such an integrative perspective on risk perception. Yet, a theory of risk perception that offers an integrative, as well as empirically valid, approach to understanding and explaining risk perception is still missing (Wachinger and Renn, 2010).

Recent research on tsunami risk perception
In many countries where the tsunami hazard is high, such as those in the Pacific and Indian oceans, several studies have been carried out on risk perception and people's understanding and preparedness (Raine, 1995;Bird and Dominey-Howes, 2008;Kurita et al., 2007;Paton et al., 2008Paton et al., , 2017Oki and Nakayachi, 2012;Goeldner-Gianella et al., 2017;Arias et al., 2017, etc.). Unfortunately, research on tsunami risk perception in the Mediterranean is neither numerous nor homogeneous. Nevertheless, one of the oft-cited figures in research is a tendency of coastal populations to underestimate tsunami risk. Recent research has been carried out in the NEAM area (North-eastern Atlantic, the Mediterranean and connected seas) by the EU project ASTARTE, which has investigated tsunami risk perception and community preparedness, through the collection of 1159 questionnaires in six coastal areas of France, Greece, Norway, Portugal, Spain and Turkey. The survey was based on a standardised questionnaire (about 50 questions) and random faceto-face interviews being administered on main beaches, on boats, in ports and in city centres (Papageorgiou et al., 2015;Goeldner-Gianella et al., 2017;Liotard et al., 2017). Despite the precious insights coming from this investigation, such as the perception of negligible tsunami hazard along European shores, the method was based on interviews of a convenience sample (also known as accidental sampling). Such a non-probabilistic approach is based on the availability of participants, their geographical proximity or their willingness to participate rather than on well-defined statistical criteria to select and include subjects, and unfortunately it does not guarantee results' significance and generalisability (Etikan et al., 2016).
The 2004 Sumatra Boxing Day tsunami and its globalscale consequences revealed how cultural and societal resources have actually resulted in different abilities to cope with tsunami risk, thus triggering a fresh new interest in risk perception research in its broadest sense, also including psychological, sociological and anthropological approaches. The lack of information or of cultural memory of past events, including their negative outcomes, may jeopardise the effectiveness of any mitigation programme, while the improvement of knowledge and emergency plans should be prioritised. Such programmes should not be implemented hastily, but must always be placed within a given social context. The communities involved should indeed mediate between agencies' proposals and pre-existing knowledge through a variety of patterns of relationship, which should always include and properly consider the value of participation, self-efficacy, empowerment and trust (Paton et al., 2008). In Italy, despite a long tradition of tsunami research, the issues of tsunami risk perception have yet to be more deeply examined.

Goals and research hypotheses
Our pilot study, conducted in the Apulia and Calabria regions ( Fig. 2), has the main strategic goal of providing empirical data on citizens' understanding and risk perception in a tsunami risk-prone area, also allowing future comparisons with different areas of the NEAM region. Moreover, the re-sults will contribute to identification of key messages, channels and techniques to effectively communicate tsunami risk in the Mediterranean area.
The results of our survey will provide both CAT-INGV and DPC with data on people's perception and preparedness in Italy, contributing to address an improved communication strategy on tsunami risk. A comprehensive and sound risk communication strategy will hopefully improve people's ability to understand phenomena and to enforce both individuals' and communities' response capabilities.
Our research relies on a general assumption: the lack of awareness and the misconceptions about tsunami dynamics and impact may considerably hamper the effectiveness of any mitigation measures. Furthermore, the scope of this study is to provide a general description of what people know about tsunamis and how they perceive this risk. In particular, we formulate the following two hypotheses.
-RH1. Does the tsunami risk perception of Apulia and Calabria match with the hazard assessed by scientific data?
-RH2. Does people's perception in Apulia and Calabria about tsunamis rely upon media representations of catastrophic events such as the ones that occurred in Sumatra and Japan?

Methods and techniques
Following the well-established standard in social science and risk perception research, a questionnaire survey was deemed to be the most suitable method of investigation. This method allows us to obtain valid, accurate and robust data in a reasonably short time as well as to check data quality and validity of empirical evidence as it emerges from the survey. The tsunami risk perception survey was based on a semistructured questionnaire, consisting of 27 items, including closed questions, open questions and Likert scales to measure respondents' attitudes with respect to the topics under investigation (see Supplement). The number of questions was optimised for being administered by telephone, taking duly into account the need for brief and comprehensible questions. To date, despite some problematic elements such as the increasing under-coverage rate (De Vitiis and Righi, 2011), Computer-assisted telephone interview (CATI) is still considered the most suitable technique in order to reach all population strata (i.e. the oldest and the youngest, the most and the least educated), thus obtaining a statistically representative sample. It is no coincidence that the Italian National Institute of Statistics (ISTAT) along with similar organisations across the world, make use of CATIs for their research.
The questionnaire is structured into six areas of interest: (1) socio-demographic data and characteristics of the territory, (2) level of knowledge of the phenomenon and sources used for the collection of information, (3) perception of the tsunami risk in the geographic and social context of the in- terviewees, (4) social representations of the tsunamis, (5) cultural attitude towards the risk, and (6) messages and channels suitable for early warning messages (see SM1).
The questionnaire was developed in the light of preexisting research literature (Lindell and Perry, 2012) on tsunami risk perception (e.g. Bird and Dominey-Howes, 2008) and also considering the need to encompass the different ways in which scientists and lay people are used to understand tsunami risk. A number of papers have been considered to formulate, include or readjust questions to be included in the questionnaire. Questionnaire structure and contents were formulated according to both cited risk perception research and tsunami risk perception studies. As an example, the idea of including some questions about familiarity, understanding and the occurrence of events in the local area came from Raine (1995) and Alam (2016), beliefs about possible regions of tsunami sources are from Bird and Dominey-Howes (2008), the effectiveness of tsunami knowledge and optimal channels of information are from Kurita et al. (2007), risk evaluation of tsunami heights were readapted by Oki and Nakayachi (2012), and so on.
The questionnaire was drafted in a simple and nontechnical language so as to be easily understood by lay people without any knowledge of seismology, geology, physics and related fields. Furthermore, questions were intended to ensure maximum adherence between physical concepts (e.g. maximum inundation height and distance, run-up, and so on) and common language. For these reasons, the questionnaire has been drafted and reviewed in the light of results of two focus groups with (1) scientists and (2) lay people. The first group involved INGV tsunami scientists for a first review and an elicitation of scientific content of the questions. Then, the questionnaire was tested on 40 people with low and medium levels of education and different ages, to assess questions' readability, identifying possible biases in the way questions were formulated. Their feedback was used to adjust and rephrase some questions to make them easier to understand.

Study area and sample characteristics
The research covered two regions of southern Italy, Calabria and Apulia, representing over 3 million inhabitants living in some of the most tsunami-prone areas of the Italian peninsula (Table 1), according to historical tsunami catalogues of Italy (Tinti et al., 2004;Maramai et al., 2014) and S-PTHA studies (Lorito et al., 2008;Basili et al., 2013) (Fig. 2).
The research was carried out on a proportional stratified sample of 1021 respondents, including 474 men and 547 women aged between 18 and 95 years across 138 coastal municipalities of Apulia and Calabria. These two regions have shorelines extending for 865 and 780 km covering 22 % of the Italian coastline and 16 % of the Italian coastal population. The interviews were collected following research team directions about (a) reference universe, (b) sampling strategy, (c) stratification variables, and (d) number of interviews to be implemented and administered. Interviews were carried out using CATI methodology. The sampling plan was aimed at ensuring the best possible representativeness in accordance with the available resources, in order to provide scientists, end-users and civil protection with robust and reliable data to ground mitigation actions, also improving scientific debate on these topics. Sampling operations followed these steps: (a) defining the population, (b) choosing sample size, (c) sorting the population, (d) assigning numbers to cases for each class, (e) calculating the sampling fraction and (f) selecting the sample (Table 2). Interviewees were selected by using three stratification variables, age, gender and coastal areas, so as to guarantee the best possible correspondence between subpopulations in the sample and within the reference universe. A total of 833 questionnaires were administered to landline users and another 188 to mobile phone users, for a total of 1021 questionnaires. The decision to contact mobile phone users was due to the need to include a larger number of young people, who are less likely to use landlines. Data collection was completed between April and May 2018 by a team of over 20 trained interviewers, supervised by highly trained research experts of Questlab S.r.l., a specialised research company based in Venice.
In order to get 1021 complete and valid interviews, a total of 20 248 units were contacted. Among these, 15 564 call attempts were unsuccessful (unanswered phone calls) and 3663 units explicitly refused the interview. The response rate is 21.8 %, calculated as the ratio of the number of positive responses to the total number of responses. Table 3a and b show the response rate by age and gender, and they also indicate whether the contact was made through a landline or mobile phone.

Results and discussion
The results of our research are discussed in two separate sections. The first part is dedicated to the tsunami risk perception, in the strict sense. The second part illustrates and discusses the data on the characteristics of the tsunami phenomenon and examines the data on tsunami knowledge sources.

Tsunami risk perception
Considering the whole sample, the majority of people (49.5 %) consider tsunamis to be rather unlikely in the Mediterranean area. Specifically, 41 % of the interviewees believe that the occurrence of such an event is "unlikely" and 8.5 % consider it "not at all likely". By contrast, the overall percentage of those who think that a tsunami in the Mediterranean is a likely event is 42.5 %. Of these, most consider it "quite likely" (33 % of the total), and 9.5 % even consider it to be "very likely". The leftover 7.9 % consider the occurrence of a tsunami an event "neither likely nor unlikely" (see Fig. 3). It may be useful to recall here that the Likert scale (not at all unlikely, unlikely, neither likely nor unlikely, quite likely, very likely) is to be considered a generally accepted standard in social research (see Likert, 1932). Figure 4 shows in detail which Mediterranean areas are deemed to be subjected to tsunami hazard: 35.6 % of the sample indicate their region of residence and nearby regions (Calabria and Apulia coasts) as the most prone areas, 17.1 % suggest the western Mediterranean, 10.9 % suggest the eastern Mediterranean, 11.2 % suggest the central Mediterranean and 25.3 % declare they do not know.
With respect to the geographical reference area of the sample (the Italian regions of Calabria and Apulia) the question of whether they think the coasts of their municipality could be hit by a tsunami was split in half: those who answered yes made up 44.7 %, whilst no made up 44.8 %, and the remaining 10.6 % answered that they do not know. The answers are slightly different depending on the age group. In particular, tsunami risk perception seems to be slightly higher in the re-    (Fig. 5). The comparison also shows that the way risk is perceived strongly differs among the different regions and coastal areas: respondents from Apulia appear to have a lower perception than those from Calabria (Fig. 5). In particular, only 30 % of Apulian citizens consider their coasts as likely to be hit by a tsunami, whereas the figure for Calabria exceeds 60 %. A possible explanation of this sharp difference is in the frequency and the time interval of tsunamis that have occurred in the two regions. As a mat- ter of fact, looking at data reported in the tsunami catalogue EMTC and in the database of observations ITED (Maramai et al., 2014(Maramai et al., , 2019, it is evident that the occurrence of tsunamis in Calabria is more frequent than in Apulia, at least in the last 2 centuries. In particular, data from Calabria are dominated by the 1908 event, with observations for both coastal areas of the region (Fig. 2) and run-up as high as 13 m. On the contrary, Apulian data are far less numerous and refer to much older events, such as the 1627, 1731 and 1743 earthquakes, with no tsunami reported in the 20th and 21st centuries. The value of probabilistic tsunami hazard assessment (S-PTHA), such as the one made by TSUMAPS-NEAM (Basili et al., 2018), also lies in its ability to consider the contributions from old and/or hidden events; this allows us to get a better, though still imperfect, image of the real hazard. Comparing the results of the survey only with the historical data could provide a misleading image of the perceived risk, whereas comparing them with the estimated hazard gives a more realistic view of perception, confirming that tsunami risk in Apulia is particularly underrated.
Furthermore, the results of our survey also show that Tyrrhenian Calabria's residents have a higher perception of tsunami risk than all other coastal regions, but this point will be discussed in detail later in the text.
We also compared the perception of tsunami risk with the education level of respondents. The data show that the perception of risk increases with the level of education; in particular graduates and postgraduates show a perception of risk higher than the others (Fig. 6). Figure 7 shows the results of tsunami perception for a number of generations residing in the considered coastal municipality. It can be noted that, contrary to what happens for the age, the number of generations of residence has a small effect on tsunami risk perception. This may indicate that, considering the whole sample, there is a low transfer of information and experiences related to the tsunami risk from generation to generation, while recent local events may trigger other patterns of information gathering and seeking.
Collective memory of natural disasters is a relevant issue to take into consideration: although a single definition is still missing, some authors highlight it as a dynamic process of functional adaptation to a changing world (Assmann, 1997). Oral communication could play a key role because it would allow memories to circulate, also connecting historically separate generations that otherwise could not have mnemonic access to each other. This mnemonic transitivity would allow people to preserve memories in the form of oral traditions, passed on from one generation to the next by means of elders and families (Nora, 1982).
A peculiar aspect concerns the perception of risk with respect to the different coastal regions. In this case, as we showed in Fig. 6, Tyrrhenian Calabria has a higher level of risk perception than the other regions.
At the moment, we are not able to explain this particular result but we can formulate some hypotheses. We have noticed that the respondents from Tyrrhenian Calabria are also more likely to consider volcanoes to be a possible tsunamigenic source: 66.2 % of them indicate volcanoes as a possi-  ble cause of the phenomena, while the value is far lower for people from other coastal areas, which varies from a minimum of 41.4 % for those living in Adriatic Apulia to 45 % of Ionian residents. Indeed, the southern Tyrrhenian sea hosts several active and quiescent volcanoes, including the Aeolian Islands (Stromboli, Vulcano, Salina, Lipari and others), and submerged volcanoes like Marsili, Palinuro and other sea mounts (Fig. 2). Therefore, the results outlined above could reflect both people's knowledge of this presence (on clear days people living on the coasts of Tyrrhenian Calabria can see the volcanoes offshore) and the fear of submarine eruptions and tsunamis, particularly from Mt. Mar-sili, to which a strong devastating power is often attributed by media. Moreover, it must be considered that a "volcanic" tsunami did actually occur in 2002 triggered by a collapse of the Sciara del Fuoco flank, on Stromboli island, with runup as high as 10 m on the island and notable effects even in Calabria (Bonaccorso et al., 2003: Maramai et al., 2005Tinti et al., 2005;Chiocci et al., 2008). The fear of volcanic tsunami risk on Tyrrhenian shores is confirmed by recent research on tsunami risk perception in relation to a hypothetical volcanogenic tsunami event, conducted on a sample of 888 interviewees from three Tyrrhenian coastal regions of southern Italy, namely Campania, Calabria and Sicily (Grav- Regarding the perceived size of tsunami waves, one should keep in mind that tsunamis can occur in many ways, and even a small event can cause serious damage and loss of life (by dragging children and even adults into the sea). Only 16 % of the sample believe that a wave size of 50-100 cm would be dangerous for an adult who is near the shore, and those who think that even smaller waves of the tsunami can be a serious threat are even fewer: only 3.2 % (Table 4).
Despite the greater probability of occurrence of small tsunamis and the remarkable danger represented by waves of less than a metre (with speeds up to 10 m s −1 ), it is still probable that people's idea of a tsunami is influenced by the strong and persistent images of great waves and flooding displayed on television.

Tsunami: sources of knowledge and phenomenon characterisation
According to hypothesis 2 (see Sect. 3.1), the way tsunamis and related risk are understood and perceived may be affected by the sources of knowledge which have actually been used by the interviewees to gather information, first and foremost the media. In our survey, we first considered the difference between the Japanese word "tsunami", which dominates the tsunami risk governance field in many languages, and the word maremoto (literally seaquake), which is more common in spoken Italian. Survey results have shown that these two words are associated with two different mental models, in which some given features of the phenomenon are differently recalled and combined together, although with some degree of overlapping. Putting aside minor differences, the idea of "big wave" is strongly associated with the word tsunami (60.8 %) rather than with maremoto (39.5 %). Moreover, the word earthquake (terremoto in Italian) is mentioned as a feature of maremoto (50 %) more frequently than for tsunami (35.4 %). Other differences are found for the association of maremoto with sea storms (23.7 % vs. 17.9 % for tsunami), while sea withdrawal is slightly more associated with tsunami (15.9 %) than with maremoto (10.9 %). In general terms, the majority of the interviewees were more familiar with the Italian word maremoto (53.3 %) than tsunami (46.7 %). Such a difference is more pronounced for elders, for women and for people with low level of education and of course has relevant implications for future risk communication strategies. It would be interesting to verify whether similar differences are present in other languages, where "autochthonous" words such as the Italian maremoto do exist, for instance in Spanish in which the same word is used.
Interviewees were also asked to respond about the possible causes of tsunamis: earthquakes are correctly recalled by 75 % of respondents, while volcanic eruptions were indicated by 46.1 %, meteorological phenomena by 12.2 %, meteorites falling in the sea by 10.1 %, landslides by 9.0 %, and finally other possible causes by 6 % 2 . Bivariate analysis has shown that causes listed above are first influenced by coastal area, then by level of education, age and gender: further analyses are required to better explain these differences. Such percentages reflect in some way the relative distributions of tsunamis' sources worldwide. Although the interviewees had the possibility to select more than one choice (and therefore they could have selected all of them), it is possible that they decided to pick only a few of them, i.e. those that were considered more likely.
The answers to the question about the possible effects of a tsunami on the coasts of one's own region suggest that people are more likely to recall some aspects of the physical phenomenon instead of others, which appear to be less familiar. Data distribution shows that the first five items, arranged in decreasing number of "correct" answers, are consistent with the catastrophic visual imagery of the great tsunamis of Sumatra and Tohoku, as most of the interviewees were able to address physical damage to houses, buildings and infrastructures (92.2 %); negative impacts on economy and occupation (91.6 %); environment (90.4 %); casualties; and injured people (89.4 %). An interesting result emerging from the survey is that people are well aware that fleeing to the beach after a strong shaking is not the right choice (85.1 %).
The greatest difficulty in understanding tsunamis is concerned with some relatively unfamiliar effects, such as the possibility of great tsunamis (> 20 m) occurring even in the Mediterranean (38.6 % think this is a real possibility), tsunamis triggering strong sea currents (37.8 %) and a tsunami wave of only 50 cm actually being dangerous for people staying near the shorelines (only 19.2 % of the sample consider this a real danger). Evidence from this survey is consistent with what happened in the aftermath of recent events that occurred in the Mediterranean. On 21 July 2017, a small tsunami generated by a magnitude 6.6 crustal earthquake hit the island of Kos (Greece) and the nearby coastal city of Bodrum (Turkey) with run-up elevation as high as 2 m (Yalçiner et al., 2017). On that occasion, surveillance cameras on the Kos waterfront captured the way people were reacting to sea level anomalies: they were seemingly calm and curious to see the water inundating quaysides. They were shooting pictures and videos with their smartphones instead of fleeing, thus emphasising that the risk posed by small tsunamis was almost completely ignored.
In order to get a concise and comprehensive picture of knowledge about the phenomenon, a rough but effective knowledge index has been developed, simply calculated as the unweighted sum of the number of correct answers to all the above-listed questions about the physical reality of the event, divided by the number of items considered. Given the average value for the whole sample (0.6952), gender, level of education, age and coastal area differently affect the level of knowledge. It is evident that a higher level of education implies a higher index of knowledge, as expected. Index values appear to be higher (> average value) for women, for middleaged people (35-49), and for residents of the Tyrrhenian Calabria and Ionian Apulia coastal areas. Conversely, elder people (65 and over), less educated people, males in general, and inhabitants of Ionian Calabria and Adriatic Apulia are placed below the average value.
These data could be more usefully scrutinised by considering the main sources of knowledge which have been used by people. Social images of the tsunami and in turn risk characterisations rely on a variety of sources, combined in different patterns according to age, education, gender and coastal areas. We assumed that broadcast media, printed media, the Internet and other sources, including word of mouth through interpersonal networks (relatives, neighbours and friends), would have different impacts on people's understanding of tsunamis, thus resulting in different mental models. It should be noted that the aim of these questions was to explore the most suitable channels to spread information on tsunami hazard to the public.
We asked people to list the sources of information they actually used from a list of 12 items (multiple-response questions), thus analysing both the relative relevance of any single source and of their possible combinations. It is evident that television has a fundamental importance as the first source of information for almost all the respondents, in line with general statistics on cultural consumption in Italy (ISTAT, 2018). Figure 8 shows the results obtained by grouping the channels into homogeneous categories; the overwhelming role of television emerges even more clearly, since it is able to reach almost 90 % of the sample (89.4 %), followed by traditional media (newspapers, books, movies, radio), which reaches just over half (50.8 %), and then by the Internet (17.5 %). This suggests that any effective risk communication campaign must face the overwhelming role of television. It should be considered that any channel may include expert knowledge from institutional science sources: as television ensures a larger audience, it also requires scientists to shape message according to particular visual formats and established narratives, also accepting particular conditions to access the public.
Social networks, which include all the interpersonal channels such as friends, parents and other relatives, together with neighbours and personal acquaintances were found to be a relevant source for 13.4 %, while institutional and scientific sources together with other sources are placed lower in this ranking (9.9 %). It is therefore important for research institutions and civil protection agencies to work in this field, trying to reach more people and giving the correct information about this risk. Unfortunately, the results of our survey highlight the minimal impact that institutional and scientific sources have had on people's understanding of tsunamis (cumulative percentage: 5.6 %). In Italy, the efforts of DPC, along with INGV and other scientific institutions, to raise people's awareness of tsunami risk and inform about best practices of civil protection date back to 2013, through the campaign "Io Non Rischio" ("I don't take risks"), started in 2011 for earthquake risk (see Postiglione et al., 2016). However, only a very limited percentage of municipalities have included the tsunami risk among the activities. In 2018, only in 5 % of the squares where Io Non Rischio took place (a total of 502) was the tsunami risk included. In particular, in the two regions where we launched the survey (Apulia, Calabria), no activity on tsunami risk was carried out during the 2018 campaign in any of the towns and villages involved (source: http://iononrischio.protezionecivile. it/io-non-rischio/dove-si-svolge/, last access: 16 December 2019). The results of our survey seem to indicate that the previous awareness-raising initiatives have proven neither numerous nor effective. The number of respondents who recalled communication campaigns of civil protection/risk is 34 out of 1021 people interviewed, a number too low from which to draw any significant statistical inference (N = 34/1021, that is about 3.3 % of the sample). Furthermore, Io Non Rischio is a campaign that takes place at the municipal level, and our research has not collected enough answers to draw conclusions at this level. Rather, in the future, it would be desirable that the same Io Non Rischio campaign, 10 years after its launch, provides itself with tools to assess its impact and effectiveness. Our results seem to indicate that the enormous media coverage of the events of 2004 and 2011 in the Indian Ocean and in Japan has left a profound impact in Italy on people's understanding of tsunamis. Compared to the past, some of the bestknown images of these events came from accounts of digital eyewitnesses, transmitted through multiple Internet channels, so that the content generated by amateur users quickly became the most important source of transmission news from the most affected areas (Allan and Peters, 2015). Images like those of the big waves approaching the beach of Khao Lak (Thailand) after the withdrawal of the sea, as well as the flood wave that overcomes the shores and crashes on the Miyako seafront in Iwate prefecture (northern Japan), have travelled the world, providing a vivid account of the event, shaping people's understanding of tsunamis globally (Yamori, 2013;Couling, 2014;Goeldner-Gianella et al., 2017).

Conclusions
We carried out a CATI survey in order to retrieve empirical data on tsunami knowledge and risk perception in Italy. The results of the survey show that the tsunami risk is generally underrated, despite the high hazard of the regions under investigation. The level of risk perception seems to be quite low for the whole sample, and it appears to be influenced by education level and gender, as well as by the possibility to access reliable sources of information. Conversely, the interviewees' age and their time of residence in the same coastal area (considering number of generations) do not result in significant differences in risk perception.
The general underrating may be due to the infrequency of damaging tsunami events in the last decades in Italy, amplified by the lack of available information on this risk and by the limited number of tsunami drills and campaigns.
An interesting result emerging from this study is that the inhabitants of Calabria and Apulia have a very different level of tsunami risk perception, although the hazard assessed from scientific data shows similar levels of occurrence probability in the two regions. More than 60 % of the Calabrians consider the occurrence of a tsunami in their region to be quite likely, whereas only 30 % of the Apulians do. We interpret this difference as due to the lower frequency of tsunami observations in Apulia compared to Calabria (Fig. 2) and to the larger time elapsed from the most recent event in the former region. This result confirms the need for raising awareness in areas where the memory of events is loose and the perception of risk is even less pronounced.
Moreover, we find that residents in the Tyrrhenian Calabria coastal area are more likely to consider tsunamis to be actual and impending threats. As discussed in the previous section, this might be related to real or purported sources of tsunamis from the volcanoes in the Aeolian Islands or from submarine volcanoes (Fig. 2). Such a result suggests the need for a thorough analysis on cultural and historical factors that may locally affect the way tsunami risk is perceived and understood.
Mental models of tsunamis, stemming from people's characterisation of hazard, appear to be heavily influenced by media images of the devastating Sumatra (2004) and Japan (2011) tsunamis, since TV news coverage and documentaries of these events are the first source of information in terms of importance for most of our interviewees. Both disasters received huge media coverage, triggering a global-scale "media event", which deeply shaped individual and social understanding of tsunami.
Evidence from our survey provides robust support to this interpretative hypothesis: the way tsunamis are understood is very consistent with such televised imagery, and almost nine out of 10 people cite TV as a primary source of information.
Risk characterisation, which shapes the way hazard is understood, is affected by different factors, including the words that are used to refer to certain phenomena. Our results high-light that this event is differently conceived when using the foreign word tsunami rather than the Italian word maremoto. Although the two terms are equivalent for Italian Earth scientists, according to people's perception, the two words refer to two different events, with some features in common. Our results also show that people appear to be conscious that earthquakes are the most frequent cause of tsunamis. Also, they tend to overestimate volcanoes as a possible cause of tsunamis, while underrating other causes such as landslides. Moreover, there is a poor awareness of some aspects of this hazard: previous tsunami disasters in Italy are part of a distant past, the details of which have faded away. In general, the respondents appear to totally neglect the possible impacts of small tsunamis, as also evidenced in other countries, thus fostering a false sense of subjective immunity. Our research underlines another critical point: people are likely to match information on tsunamis with their personal experience of sea-storm waves to understand and characterise this risk, thus resulting into misleading assumptions about the real hazard posed by tsunamis (Oki and Nakayachi, 2012;Santos et al., 2016;Sutton and Woods, 2016;Wood et al., 2019).
This research is the first of its kind conducted in Italy, and its findings appear to be promising. Future analyses on this data set will probably allow us to better identify the main factors affecting tsunami risk perception in Italy as well as to better understand the reasons behind the differences between regions and coastal areas. Future steps of this research will include the extension to other contiguous coastal regions which are most exposed to tsunami hazard together with Calabria and Apulia.
This research aims at integrating and enriching tsunamirelated literature from social science fields, also providing new data on and insights into the Mediterranean area. Currently, most of the available contributions regard only a few coastal areas in the Pacific and Indian oceans, such as Japan, Indonesia, Chile, western US and the Pacific Islands, where tsunamis are considered both a matter of fact and a historical reality, and the risk posed by tsunamis is fairly well known by local populations. To date, research papers on this topic are noticeably scarce for the NEAM area, with a few local exceptions, for example some Norwegian fjords (Lacasse and Nadim, 2009;Rød et al., 2012;Goeldner-Gianella et al., 2017). Lacking directions on people's perception and understanding of what a tsunami is and its related impact may lead to significant difficulties in setting up sound risk communication strategies. Furthermore, the lack of data from social sciences could result in serious difficulties in fostering people's engagement and participation in the implementation of effective mitigation measures. In general terms, the development of tsunami warning systems should focus not only on managing an ongoing event through crisis communication, but it should also improve individuals' and communities' awareness and preparedness in the long term (Lundgren and McMakin, 2018). This implies a better understanding of targets, messages and channels to be arranged both for in-forming people about the hazard posed by tsunamis and for effectively shaping an alerting strategy to make people conscious about what it is happening and what they should do in case of an event.
The research is first intended to provide viable knowledge about people's perception and attitudes toward tsunamirelated risks in order to improve communication strategies of both CAT-INGV and the Department of Civil Protection (DPC); it also aims to provide useful cues and suggestions for the overall tsunami community in the Mediterranean region and beyond. This is the first extensive study on tsunami risk perception in Italy and the first of this kind (with a large stratified sample and CATIs) completed in the NEAM region.
Any effective, sound risk communication strategy should rely on the integration of theory, empirical research, best practices and careful assessment of outcomes, within an open-ended cycle of research and action. Research results may indeed foster an open discussion on risk and crisis communication strategies to be held and to improve both individual awareness and communities' involvement and participation in risk reduction programmes at national and regional levels.
Risk communication should be integrated with other initiatives of community engagement rather than being conceptualised as a stand-alone process. The relevance and the meaning of the information about tsunamis arise from the way they are interpreted and prioritised within given social contexts; hence to facilitate preparedness any successful communication strategy must consider if and how information is known and whether it is used (Paton et al., 2008). In particular, it would be important to challenge clichés about tsunamis and to consider the actual knowledge and education level of people living in tsunami-prone areas, also bearing in mind which channels are more suitable to reach as many people as possible.
The validity of the data collected and analysed in this paper is limited by definition to the coastal populations of Calabria and Apulia and cannot be generalised to the entire Italian coastal population. As shown by research data, tsunami risk perception may be affected by a number of factors acting at global and local levels such as event history, risk characterisation and presence of known tsunamigenic sources (volcanoes), along with socio-demographic features of the considered population.
In order to improve knowledge and clarify problems, in future investigations one may consider the following alternatives.
a. Use the same sample methodology of this first survey (stratified population sample) to investigate tsunami risk perception in other Italian regions and coastal areas (Basilicata, Molise, Sicily, etc.).
b. Extend the survey to the entire Italian population, considering that many Italians who also live in inland areas spend their holidays in coastal areas.
c. Conduct interviews addressed to specific groups of people representing the coastal population in a specific season of the year, using the same survey questionnaire. However, in order to have a good representative statistical base of the population, this method requires having data on the tourist population and the non-resident population of the coastal municipalities.
The general structure of the questionnaire, the type and number of questions, and the duration of the interviews are strictly designed to be administered by telephone, but the questionnaire could be adapted to be suitable for direct interviews or self-administration. Survey methodology entails an assumption: data about individuals are used to make inferences about social attitudes and beliefs, thus underestimating the influence of both local culture and "group thinking" when facing complex problems. For these reasons the survey should be ideally seen as a first step in a wider research strategy, aimed at providing further developments within a mixed-method approach, "to bring in more robust evidence than either qualitative or quantitative approaches provide when they are used separately" [...] and "to gain a deeper understanding of hazard perception and preparedness" (Alam, 2016). In this way, qualitative research methods (focus groups, interviews, etc.) could be used to implement and explain the results of the survey, contributing to clarify the role of both culture and individual motivations in shaping social response and risk awareness.
The research is indeed conceived as a set of integrated modules, to fit different needs and social context, and it is suitable to be replicated as a whole or in part in other geographical contexts, both in Italy and in the countries of the North-eastern Atlantic, the Mediterranean and connected seas region (NEAM). Data comparison and multivariate analysis may reveal underlying cross-cutting factors of tsunami risk perception predictors, thereby focusing on similarities and differences among coastal areas and countries.
Data availability. Research data can be accessed by contacting Massimo Crescimbene at his e-mail address: massimo.crescimbene@ingv.it.
Author contributions. AC, MC, FLL and AA provided theoretical background, analysed data and wrote the paper. AC and AA revised and supervised the research.