Assessing the effectiveness and the economic impact of evacuation: the case of Vulcano Island, Italy

Evacuation planning and management represents a key aspect of volcanic crises because it can increase 20 peoplepeople’s protection as well as minimize the potential impactimpacts on the economy, properties, and infrastructure of the affected area. Assessment of evacuation scenarios that consider human and economic impact is best done in a pre-disaster context as it helps authorities develop evacuation plans and make informed decisions outside the highly stressful time period that characterizes crises. We present an agent-based simulation tool that assesses the effectiveness of different evacuation scenarios 25 using the small island of Vulcano island ((south Italy) as a case study. Simulation results show that the overall time needed to evacuate people should be analysed together with the percentage of people evacuated as a function of time and that a simultaneous evacuation on Vulcano is more efficient than a staged evacuation. For example, during the touristic (high) season between July and August, even though the overall duration is similar for both evacuation strategies, after ~6 hours about 96% of people 30 would be evacuated with a simultaneous evacuation; in contrast, only 86% would be evacuated with a staged evacuation. We also present a model to assess the economic impact of evacuation as a function of evacuation duration and starting period thatwith respect to touristic season. It reveals that if an evacuation of Vulcano would cause significant economic impact to the tourism industry if lasting more than 3 to 6 months (in case it was initiated at the beginning of the visitor season) to 1 year (in case it 35 was initiatedor at the end of the visitor season). touristic season (i.e., June or November), it would cause a very different economic impact to the tourism industry (about 78-88% and 2-7% of the total annual turnover, respectively). Our results show how the assessment of evacuation scenarios that consider human and economic impact carried out in a pre-disaster context helps authorities develop evacuation plans and make informed decisions outside the highly stressful time period that characterizes crises. must take and foresees the evacuation of the population before the eruption onset in case increasing volcanic unrest is observed. The results of the work presented in this paper as well as the 1015 insights into risk assessment presented in Bonadonna et al. (2021) were used to finalise the emergency plan at national level. However, given that the recent unrest was mostly related to gas hazard and was, therefore, managed at the local level (i.e., by the Municipality) and that the overnight restrictions issued by the municipality did not involve an evacuation of the island, the results of these two analyses could not be tested nor validated.


2021
eruption of Cumbre Vieja, La Palma, Spain; https://volcano.si.edu/volcano.cfm?vn=383010#September2021). However, failure to evacuate in anticipation of an eruption or of the associated primary and secondary hazards can lead to catastrophic outcomes as seen during the 1985 Nevado Del Ruiz eruption in Colombia and during the 2018 Fuego eruption in Guatemala (Voight et al., 2013;IFRCb 2019;Leone et al., 2019). 80 Unlike other emergencies, the duration of volcano-related evacuations can last for days, months or even years depending on the typestyle of eruption and its impacts on the landscape and can result in longlasting or even permanent relocation of communities (e.g., Soufrière Hills, Montserrat and Tungurahua, Ecuador; Barclay et al., 2019). In detail,Evacuations involving long duration of evacuation occursdurations mostly occur because i)periods of elevated unrest can be protracted, ii)or eruptive 85 activity can be protracted, iii) post-eruption activity such as remobilisation of pyroclastic deposits by water (i.e.., lahars) and wind (i.e.., ash storms) can continue threatening communities, and/or iv) the damage can be so overwhelming that people and their government lack the resources to rebuild in a timely period. itself can also be divided in subgroups of costs such as direct (any action taken for mitigation infrastructures), indirect (secondary costs such as economic disruption due to mitigation measures) and intangible (e.g.., environmental damage due to change in agriculture practices) costs (Meyer et al., 2013). In this study, we are mostly concerned with the third type of cost related to the interruption of 110 economic activities (i.e.., tourism) as a result of a prolonged evacuation.
In this context, weWe present here a novel methodology to couple an evacuation model with an assessment of its potential economic impact. We use the island of Vulcano, in the Aeolian Islands of south Italy, to illustrate strategies for the assessment of the effectiveness of an evacuation as well as its the associated economic impact on the island's main source of revenue, i.e. tourism. In the past decade, 115 evacuation and civil protection planning have been underway in Italy for the main active volcanoes (e.g. Vesuvius and Campi Flegrei, www.protezionecivile.gov.it;Baxter et al., 2008;Marzocchi and Woo, 2009 (Selva et al., 2020).
, the low touristic season (i.e., when fewer visitors are on island) occurs during the months of November-125 March, while the high season is in July-August. We developed an agentAgent-based simulationmodelling (ABM) in GIS space using the AnyLogic® software platform to assist emergency managers and assess the effectiveness of specific evacuation parameters, i.e. number of people present on the island (during the low and high touristic seasons), type of evacuation (simultaneous whole community evacuation or sequentially staged evacuation of different areas), eruption probability, 130 exposure, timing (before, during or after the eruptive event). ABM has been used in evacuation simulation extensively (e.g. Bae et al., 2014;Hilljegerdes and Augustijn-Beckers, 2019) and it has many advantages compared to aggregate and static approaches as it allows us to incorporate individual level behaviors, event scheduling, dynamics of agent interactions, flexibility, natural description of evacuation process (Mas et al., 2019). In addition, the platform AnyLogic® allows to visually observe 135 and assess the evacuation scenarios. A strategy to assess the economic impact of an evacuation based on the analysis of the consequences on the main economic activity (i.e., tourism) is also presented.
The next section provides some conceptual background related to effective evacuation, types of evacuation methods, and evacuation modelling, while section three describes the study area. Section four illustrates the methodology adopted in our analysis, while sections five and six present and discuss 140 the results on the assessment of evacuation efficiency as well as the assessment of economic impact of an evacuation considering different durations and starting time of both a total and partial evacuation and(i.e. evacuation of differentindividual areas); the current unrest on Vulcano is also discussed in relation to the work presented here. Section seven provides conclusions. Han et al. (2007) developed and described a four-tier evacuation effectiveness framework, by looking at evacuation time, individual evacuation time, exposure over time, and spatio-temporal exposure measures. Effectiveness of evacuation planning and operations for volcano emergencies can be assessed using this four-tier framework. One of the most common goals of evacuation analysis and planning is 150 to improve the effectiveness of evacuation by reducing evacuation time to minimize the adverse impacts associated with people leaving their place of employment, study, or their homes. Several methods have been proposed to improve the effectiveness of emergency evacuation such as enhancing the outcome of an evacuation order and warning dissemination of warning messages, controlling flows and movements in and out of designated areas, implementing staged evacuation, directing people to the best evacuation 155 routes, and focusing on flexibility to plan a possible evacuation (Abdelgawad and Abdulhai, 2009;Gaudard and Romerio, 2015).

Background on effective evacuation
In this paper we distinguish between "evacuation time", defined as the time required for the last person to evacuate an emergency zone (Urbanik, 2000), and "evacuation duration", which represents the period during which a community has been removed from a risky area. In addition, we define "evacuation 160 effectiveness" as the time required to evacuate a certain fraction of the population (e.g., 95%) (Han et al., 2007). Evacuation time of individuals or families depends on a number behavioural, logistical, perceptual, and communication factors (Tomsen et al., 2014). In order to minimize evacuation time, it is, therefore, important to reduce evacuation warning time (time it takes for the evacuation warning to reach each individual), evacuation preparedness time (time it takes for individuals to prepare for 165 evacuation after receiving an evacuation warning), and evacuation travel time (time it takes for individuals to travel from their residence to designated evacuation designatedassembley areas). Each of these time segments varies from person/ to person and family to person/family depending on their demographic attributes, preparedness levels, and access to information and resources (Jumadi et al., 2019, Lechner andRouleau, 2019). 170

Simultaneous and Staged Evacuations
Evacuation can be implemented using simultaneous or staged methods. In the simultaneous evacuation, people in an exposed area are informed and expected to evacuate simultaneously. In the staged evacuation, the exposed area is divided into several zones, and people in each zone are evacuated 175 according to a specific order (Sbayti and Mahmassani, 2006). Both simultaneous and staged evacuations have been used in past emergencies. Staged evacuations have been frequently used during hurricanes and for the 2002 Los Alamos wildfire in New Mexico (Malone et al., 2001;Farrell, 2005;Wolshon et al., 2006). Simultaneous evacuations are often used during sudden emergencies when rapid evacuation is necessary (e.g., earthquake, landslide, industrial accidents), whereas staged evacuation is considered 180 more effective when sufficient lead time exists to prepare for evacuation or when resources are limited for simultaneous evacuation of the whole population. Chen and Zhan (2008) also found that simultaneous evacuations are more suited in areas of low traffic congestion, whereas staged evacuation may be the most effective in high population density areas and complex street networks. In case of staged evacuation, the number of stages can influence the evacuation effectiveness and thus optimising 185 the number of stages is essential in reducing delays during the evacuation process (Chien and Korikanthimath, 2007). Jumadi et al. (2019) developed a staged evacuation using a spatial multi-criteria analysis for prioritisation of evacuees and found that while the staged evacuation was more effective in reducing potential traffic congestion, the simultaneous evacuation still showed better results in reducing the population at risk. 190
Agent-based modelling (ABM) is emerging as a suitable and promising framework for evacuation analysis and planning in recent years (Chen and Zhan, 2008;Liang et al., 2015;Jumadi et al., 2019).
ABM is appropriate for modelling complex and interactive systems (Gilbert and Bankes, 2002) such as emergency evacuation because it combines behavioural attributes with spatial and environmental data 205 (Brown and Xie, 2006). Moreover, ABM can provide a more realistic evacuation simulation with respect to the aforementioned approaches by incorporating human agents to the geographical environment (Mas et al., 2012;Joo et al., 2013).

Case Study: Vulcano island, Italy 210
The Aeolian Islands, which earned UNESCO World Heritage status in 2000, are a volcanic arc located in the Tyrrhenian Sea (25 km north of Sicily) associated with the subduction of the African plate under the Eurasian plate. Stromboli volcano on the island of Stromboli and La Fossa volcano on the island of Vulcano are the youngest and most active volcanic edifices (Selva et al., 2020). Vulcano is the southernmost of the seven Aeolian Islands located in the Tyrrhenian Sea (25 km north of Sicily). It has, 215 waste-water plant in Porto. (Fig. 1). The road network is limited with only one road connecting Porto and Piano (Galderisi et al., 2013;. 225 Vulcano's predominant economic activity is tourism (Galderisi et al., 2013;Aretano et al., 2013;. The island's economy and urbanization have been growing fast since the 1980s by attracting tourists from Italy and other countries, particularly during the summer season. Vulcano has a floatingfluctuating population passingranging from about 800 residents in the winter to monthly peaks of about 22,000-28,000 visitors in July-August . With increasing number of 230 visitors and seasonal workers, volcanic risk also increases and, therefore, emergency. Emergency management, particularly evacuation planning and preparedness, has also become an important issue for the island.  Milazzo is also shown.

Geological settings and implications for evacuation plannning 235
In terms of geological settings, Vulcano consists of several overlapping volcanic structures including two caldera systems (i.e. Il Piano caldera to the south and La Fossa caldera in the central portion of the island), and a smaller structure (i.e. Vulcanello) in the northern side of the island.north. A stratovolcano (i.e., La Fossa cone) sitsrests within the La Fossa caldera and, while three smaller and coalescing pyroclastic cones sit atop the Vulcanello islet. Subaerial volcanic activity inon the island dates to 135 240 ka and 120 ka (Zanella, et al., 2001), with La Fossa cone (hereafter referred to simply as La Fossa) starting at ~6 ka and being the current most active system (Dellino et al., 2011).;De Astis et al., 2013).
The last eruption of La Fossa was a long-lasting Vulcanian cycle that occurred between 1888-1890 (Mercalli and Silvestri, 1891). The eruption produced emission of ballistics, and tephra fallout, andwhile intense remobilization of the tephra-fallout deposits by rainwater intoproduced lahars (Di Traglia et al., 245 2013). The most likely hazards associated with eruptive activity of La Fossa are tephra fallout (including ballistic projectiles) and pyroclastic density currents as well as gas emissions, volcanic debris flows, lahars, ground deformation, and seismicity that can also occur during the quiescent and unrest states (Selva et al., 2020). However, it is also important to consider that the activity of La Fossa has been characterized by a large variety of eruption styles, including effusive activity and explosive events. 250 Among this variety, hydrothermal events of various intensity have occurred and associated with impactful hazards such as blast, diluted PDCs and ballistic fallout, with the most violent being the Breccia di Commenda eruption dated around 12301300 (Rosi et al., 2018;Pistolesi et al., 2021). It is thus important to distinguish between magmatic events, for which the main driver is the magma rising to the surface, and hydrothermal (or phreatic) events, for which the main driver is the interaction 255 amongst water, rocks, and magmatic heat and gas (e.g.., Barberi et al., 1992;Rouwet et al., 2014;Stix and de Moor, 2018). By their nature, hydrothermal events may be more difficult to predict than magmatic unrests and they can also happen outside the main active vent (as it has often been the case at La Fossa). In fact, the La Fossa system is a permanent and powerful emitter of fluids whose flow is maintained by an elevated gas over-pressure in the subsoil (Selva et al., 2020). Even modest 260 imbalancesdisequilibrium in the supply of fluids can trigger explosive eruptions as the numerous cases that have occurred on Vulcano after the Breccia di Commenda in the last eight centuries demonstrate (e.g.., 1444AD, 1550AD, 1727AD, 1873Selva et al., 2020). This is also the case of other volcanoes that have been associated with recent and sudden explosions such as White Island and Tongariro in New Zealand (Breard et al., 2015;You Lim and Flaherty, 2020), Ontake in Japan (Oikawa 265 et al., 2016) and Turrialba and Poas in Costa Rica (Alvarado et al., 2016;de Moor et al., 2016). While at most Most of these volcanoes mentioned above that are located in remote areas these hydrothermalsohydrothermal events represent a threat mostly for tourists, in close proximity to them, but in Vulcano they represent a serious threat also for inhabitantsresidents that live very close to the volcano (e.g.., Porto area on the north of La Fossa). The main infrastructures (including two ports -Porto 270 di Levante and Porto di Ponente, the telecommunication station, and the main power plant) and the majority of economic and touristic activities are concentrated in the Porto area also located just north of La Fossa cone (Fig. 1). This is whyTherefore, in the case of Vulcano, the potential for evacuation becomes an important issue even in case of weak unrest. Both a hydrothermal explosion and a magmatic eruption would be especially challenging events during which to manage an evacuation if they happened 275 during the high season (July-August) and with little or no warning, as it has been the case for the recent small but deadly eruptions at touristic places mentioned above.

Civil Protection System in Italy
In order to reduce the potential volcanic impact, scientists, including those of the Istituto Nazionale di 280 Geofisica e Vulcanologia (INGV), of selected research institutions ("Centri di Competenza") and of the Consiglio Nazionale delle Ricerche of Italy -Institute for electromagnetic sensing of the environment (CNR-IREA), continuously monitor all active volcanoes in Italy, including La Fossa, periodically transferring information on the state of volcanic activity to the national and regional decision makers each with defined authorities, roles and responsibilities; all these institutions are part of the overall Civil 285 Protection System in Italy (Legislative decree "Codice della protezione civile", 2018). In Italy, in fact, civil protection activities are not assigned to a single body, but represent complementary tasks attributed to an integrated system composed of both public and private and both national and territorial (regional and local) structures. The Italian Civil Protection Department (at the national level) is a structure of the Presidency of the Council of Ministers and coordinates the entire Civil Protection System. 290 Another important aspect to consider is that volcanic risk management in Italy is based on an alert level system and the Italian scientific community have defined four Volcanic Alert Levels based on monitoring parameters that describe the state of activity of each Italian volcano. These levels correspond to four colors (green, yellow, orange and red), which are indicative of the level of activity and to its possible evolution, including the shift from local to national impact scenarios. In fact, volcanic activity 295 can also generate local impact events, which are managed at the local level by the appropriate institutions (i.e. regions and municipalities). Recently, the Italian Civil Protection Department financed detailed studies to improve the current understanding of the volcanic system and the whole range of potential volcanic hazards on Vulcano Island (Selva et al., 2020). Based on these results, an updated version of the Alert Levels for Vulcano were issued at the end of 2021. These Alert Levels have been included in 300 the National Civil Protection Plan for volcanic risk on the island. The plan also describes the "national level operational phases", which include the mitigation actions that all the stakeholders involved in the emergency management must take and foresees the evacuation of the population before the eruption onset in case of increasing volcanic unrest.

Potential consequences of a volcanic crisis on Vulcano
Although it would appear to be a quick and small operation, evacuation of the island under different weather and marine conditions, occurrence of different hazards (e.g.gas, tephra fall, PDCs, lava flows, Formatted: Strikethrough lahars, landslides, tsunami) and various seasons (summer versus winter) could result in different decisions and actions. Moreover, one must also account for unforeseen factors that might limit the 310 availability and efficiency of evacuation (e.g., damaged ports). In addition, forecasting volcanic eruptions and managing volcanic crises represent an important challenge for both scientists working in observatories (e.g. geochemists, geophysicist, geologists, volcanologists) and civil authorities such as those associated with emergency management (i.e. Civil Protection System in Italy). Many impediments may be encountered in interpreting key aspects such as: i) whether or not unrest will lead to an eruption, 315 ii) the nature of explosive activity (magmatic or hydrothermal), iii) the eruptive style (i.e., effusive, explosive or both), iv) the potential activation of lateral vents, v) the eruption magnitude (i.e. erupted volume) and intensity (i.e. the rate of discharge of magma, plume height), vi) the type, extension and timing of hazards with the potential to impact human life and the infrastructure supporting evacuation whether occurring either in the unrest phase or eruptive phase or both. The interpretation of scientific 320 data complicates the decision-making process for the officials (Fearnley, 2013). . damaged harbours).
Scientists including those of the Istituto Nazionale di Geofisica e Vulcanologia (INGV) and of the Consiglio Nazionale delle Ricercheof Italy -Institute for electromagnetic sensing of the environment (CNR-IREA) continuously monitor all active volcanoes in Italy, including La Fossa, periodically transferring information on the state of volcanic activity to the national and regional decision makers 325 each with defined authorities, roles and responsibilities that are part of the overall Civil Protection system in Italy (Legislative decree "Codice della protezione civile", 2018). Evacuees may include local residents, national and international tourists, and seasonal workers. While a general municipal plan for emergency management on Vulcano exists (http://www.comunelipari.gov.it/zf/index.php/serviziaggiuntivi/index/index/idservizio/20015), and an evacuation drill was carried out with the residential 330 population in 1991 after a period of seismic unrest at La Fossa, a detailed and updated evacuation plan for the Vulcano island does not currently exist. Recently, the Italian Civil Protection Department has undertaken a dedicated effort to finance detailed studies of the current understanding of the volcanic system and of the whole range of potential volcanic hazards (Selva et al., 2020). Based on these results the alert level system is being reviewed in collaboration with the scientific communityHigher levels of 335 scientific uncertainty may thus translate in increased difficulty for emergency managers to understand the value of evacuation (measured in terms of human lives saved) and the costs associated with any evacuation associated with unrest that does not result in eruption.
When a volcano begins to show increasing signs of unrest above background level, authorities must deal with uncertainties and decide how to manage a potential crisis (e.g., have people shelter in place or 340 evacuate some or all of the population), as scientists cannot guarantee if the unrest will result in an eruption or not. Although successful forecasts have been made (e.g., Mt St Helen's 1980, USA;Mt Redoubt 1989-1990Pinatubo 1991, Philippines), alarms raised that are not followed by hazards impacting exposed areas sometimes cause both scientists and other officials to lose credibility among the public (e.g., Sparks, 2003;Tilling, 2008). In addition, a volcanic crisis can easily result in an 345 of Tungurahua volcano (Ecuador). In the first case, a dramatic short-term increase in seismicity and ground deformation led to an intensification of disaster-preparedness activities and voluntary evacuations by villagers, which resulted in substantial losses of revenue due to business interruptions 350 and a large cost of emergency preparations; at the end, many people thought that two years of preparation was a waste of money even though public awareness of potential volcanic hazards increased and the community became more resilient (Hastings, 2013;Tilling, 2008). In the second case, an economic crisis at both local and national level resulted from an evacuation that followed a successful forecast as most of the community's economic activity depended on tourism (Lane et al., 2003). This is also valid in the 355 case of Vulcano Island. Even without an evacuation order, the increasing level of unrest may cause the local people to leave the island if they believe that tourism on the island may be affected negatively by the increasing volcanic activity. This is especially true since most business owners are not from Vulcano and they may decide to relocate their activities. In both cases, the economy of the island would be negatively impacted. Additionally, there could be significant negative economic impacts on Vulcano limitation of the number of people on the island. If the increasing activity on the island results in an evacuation and finally in an eruption, still many tourists interested in natural areas and adventure may want to visit the island once the activity is back to pre-eruption and the risk is decreased.

Agent-Based Modelling of pedestrian evacuation
The Vulcano evacuation simulation toolEvacuation Simulation Tool has been developed using the Anylogic platform (version 8.7.5), which provides ABM capabilities as well as GIS spatial data incorporation. Our simulation tool includes four main agents, each of which is described below (i.e. Hazard, Evacuees, Ferries and Ports, Agents' Environment). In order to correctly characterize such 380 agents and tailor the analysis to the specifics of the island without which the tools would be useless, risk factors including hazard, vulnerability and exposure of both the community and critical assets must be Vulcano Island . All the needed aspects and elements required to assess the various indicators are provided there and rely on an extended work of surveys and data collection carried out in the last ten years. The following ABM uses the outcomes of such data collection and risk assessment, especially as far as hazard and exposure are considered. Some aspects of systemic vulnerability are also considered related to the accessibility of the three ports of the Islandisland and 390 their intrinsic characteristics.

Hazard (volcano) agent
We define La Fossa volcano as a physical agent with specific behaviour and states. defined based on scientific evaluation. It is important here to differentiate between the state of the volcano (Quiescence, 395 Unrest, and Impending/Ongoing eruption) with Alert Levels (Green, Yellow, Orange and Red) typically identified by the Civil Protection Department. In this ABM, La Fossa volcano has three main states including background levelQuiescence (i.e. normal conditions), 400 Pre-Alarm, Alarm) and eruptionshallow hydrothermal crisis and deep hydrothermal crisis) and Impending or Ongoing eruptive activity (Fig. 2). It is 405 important to consider that, for simplicity, the sake of illustration, these states in this ABM are general states and do not correspond to the alerthave been simplified with respect to the Alert Level system specific to Vulcano, 410 which is still under evaluation. However (https://rischi.protezionecivile.gov.it/it/vulcanico/vulcani-italia/vulcano). In any case, modifications can be made to reflect the specificities of individual volcanoes. Normal conditions varyBoth the Quiescence and Unrest state are different for each volcano; indifferent volcanoes. In Vulcano, they consistthe Quiescence state mostly consists of fumarolic emissions (mostly concentrated in two main 415 fumarolic fields located in the northern rim of the active crater of La Fossa cone and at the beach of Baia di Levante, in Porto Levante area), ground deformation, earthquakes and accompanying landslides (Barberi et al., 1991;Selva et al., 2020). While), while the Unrest state is mostly related to a shallow or deep activity of the hydrothermal system associated with various levels of increase in the flux and temperature of gas emissions combined with seismicity and deformation of the volcanic edifice. As 420 already mentioned, phreatic eruptions can occur in all these volcano agent can be very complex, here we only include the volcano behaviours and states that impact the evacuation process. As suchwith little or no warning. However, in our model, we assume that the simulation of evacuation starts before an eruption when the volcano is in the AlarmImpending eruptive state (Fig. 2). In the Alarm state, eruption is assumed to be imminent or highly likely such that a mandatory evacuation order is issued. 425 Shift from Alarm state to the Eruption state is handled through a condition transition that is linked to a user-defined table function.2). Forecasting of a volcanic eruption can be based on various functions 430 discussed in literature, e.g. exponential hazard function (Ho, 1992;Cornelius and Voight, 1994;Chastin and Main, 2003;Connor et al., 2003 hazards depends on eruption dynamics (i.e., occurrence of ballistics, tephra fallout, lava flows, blast surge-like PDCs, lahars) as well as topography and atmospheric conditions (e.g., wind speed and 445 direction). Hazard maps for Vulcano exist that describe the potential extent and intensity of tephra fallout and ballistic projectiles (Biass et al. 2016a,b), pyroclastic density currents (Delling et al., 2011) and lahars . However, given that before the actual eruption (hydrothermalphreatic or magmatic) takes place, the extent and intensity of the associated hazards are not known, we consider here the evacuation of certain areas to be based on the worst-case scenario, e.g. occurrence of PDCs and 450 ballistic ejection in the case of Vulcano, which could impact the whole La Fossa Caldera (including Porto area) and part of Piano Caldera (e.g.., Dellino et al., 2011;Biass et al., 2016b). We combined and expanded the Sorensen and Mileti (2014) and Stepanov and Smith (2009) multi-step 455 evacuation process models to include four main time segments: 1) warning issuance, the step from when unrest or evidences of hazard appear to when decision makers decide to issue the warning; 2) warning diffusion, the process from when the warning is issued to when the warning reached the intended audiences; 3) evacuation decision and preparation; and 4) evacuation movement.preparation for evacuation, the process from when the warning reached the intended audience to when they are ready 460 to evacuate (this includes the time required to organize departure and secure the belongings that are left behind, e.g. house, car(s), other vehicles, boats); and 4) evacuation movement. A statechart is used to model the evacuee agent's evacuation behaviour (Fig. 3).

Evacuee Agent
The agent is created, and its initial 465 state is set to "before warning" (or normal). As soon as a warning is issued, the agent's states change from "before warning" to "warning issued", corresponding 470 to an evacuation order. Transition from this state to the ""warning received" is controlled by a time out triggered transition. We use a normal truncated distribution for 475 this transition with minimum and maximum time values that can be set by evacuation planners before running the simulation. Use of this distribution allows us to limit 480 the lower bound to 0 and the upper bound to a finite value. Transition from the warning received state to prepared state is also handled by a truncated normal distribution that can be set by the evacuation planner. However, this transition is triggered only if the evacuation is either a simultaneous evacuation or the evacuee is located in the assigned evacuation stage.
The order of evacuation during a staged evacuation is based on the proximity to the hazard, with the 485 most exposed people being evacuated first. In Vulcano, the Northnorth part of the Islandisland will be evacuated from the Levante and Ponente ports and the Southsouth of the island from the port of Gelso.
In our simulations, people in Porto and Piano will be simultaneously evacuated first, and people in Vulcanello will be evacuated last. However, to provide the emergency planner with more flexibility, the simulation allows the users to set the evacuation order as needed. Evacuation time depends on the 490 evacuees' pedestrian speed and their distance to the closest active port. We consider the walking speed as a uniform distribution, but the model allows the lower and upper bounds of this distribution to be set depending on the environmental situations and population scenarios being analyzedanalysed. We assume here only pedestrian evacuation, but the simulation can be adapted to also include evacuation by vehicles, or a combination of the two. We recognize that while walking may be a more feasible option 495 for those in the north part of the Island, (Vulcanello e Porto), it may be more difficult for the people in the south part of the island. (Piano). Upon arriving at the closest active port, evacuees wait for ferries.
Once the ferries arriveGiven that our simulations are based on the assumption that the evacuation takes place before the eruption, evacuees board and they are considered to be evacuated once the ferries arrive and are boarded. However, in case the evacuation was carried out during the eruption, people should be 500 considered evacuated once the ferries actually leave the ports, as both ports and ferries could be impacted by the eruption.

Ferries and Ports Agents
Ferries transport evacuees from ports on Vulcano southward some 44 km to the large port of Milazzo 505 on the north shore of Sicily (Fig. 11a inset). As an evacuation order is issued, available ferries are mobilized in the Milazzo port. It takes between 40 minutes to 1 hour for ferries (hydrofoils) to reach the Porto Levante in Vulcano from Milazzo. In our simulations, ferries will have athe capacity ranging fromof 200 (hydrofoils) to , 400, 600 and 800 (ferries) passengers (including intermediate capacities of 400 and 600),and an average speed of 50 km/hour, but these variables can be changed. Since the two 510 smaller ports in Vulcano (Porto Ponente and Porto Gelso) are not suitable for large ferries, only boats with small capacities are dispatched to these ports. BoatIn fact, ferry speed depends on the weather and marine conditions that can be set by the users before running the simulation. However, for this study we use an average speed of 50 km/hour that is the regular speed of hydrofoils boats operating between Vulcano and Milazzo. As shipsferries arrive in theirto the port, evacuees start boarding until full capacity 515 is reached, at which point ships willthey travel back to Milazzo. If there are more requests, ferries and boats continue going back to the assigned Vulcano ports, otherwise they stay in Milazzo.
Port agents have two main states in our ABM including Normal and Evacuation states (Fig. 4). As soon as an evacuation order is issued, the state of the ports changes from Normal to Evacuation through a message transition. Inside the evacuation state, two substates demonstrate whether a port has ferries to 520 board evacuees or not. The transition between these two substates is controlled by the interactions between the ferries' agents and ports' agents.

Agents' Environment
Two main GIS networks were created for this study. The first connects the three ports of in Vulcano 525 (Porto Ponente, Porto Levante and Porto Gelso) to the port in Milazzo (Fig. 1a). The second connects buildings in Vulcano (e.g., residential, commercial, hotels, facilities, etc.)) with the road network created based on the existing road network on the OpenStreetMap (Fig. 1a,b1). We illustrate our evacuation simulation tool by setting up two pre-eruption evacuation scenarios taking place during the low and the high touristic seasons. Summary of the scenario's initial conditions are summarised in Table 2 Table 2). In addition, our simulations do not account for variable weather and marine conditions. Note that these parameters were chosen based on the author'sour knowledge of the 545 area and are used only with the purpose of illustrating the functionality of the tool. All parameters can and should be identified by emergency managers based on the availability of information and on the range of conditions to be tested (e.g. ., people with reduced mobility or with health issues, evacuation using a variety of vehicles).

Assessment of the economic impact of an evacuation
When the hazard level is high and human life is at stake, economic losses usually play little to no role 555 in the decision of whether to evacuate. In less extreme situations, however, authorities weigh different factors, and different evacuation plans can be considered. In fact, the management of the crisis will take different courses depending on the evolution of the unrest and the time-dependent evolution of the hazard. Accurate data necessary for a reliable cost-benefit analysis are rarely available, especially in the context of small islands where they are aggregated at the level of the Municipality. Furthermore, in case 560 of relatively simple economic systems such as that of a small island, complex and sophisticated models can be replaced by a set of reasonable hypotheses. Consequently, we present an approach to estimate the loss of revenue caused by a total or partial evacuation of the population on the island at any one time (i.e. residents, seasonal workers and tourists) due to an imminent eruption. Such an analysis is especially important in case of scenarios of long-lasting Vulcanian cycles, such as that of the 1888-90 eruption of 565 La Fossa volcano, that would disrupt the island's economy for a long time (many months to years).
Data collection required to estimate the impact of an evacuation on the island's main source of revenue was carried out between 2014 and 2016. This investigation focused on tourism related business activities. We interviewedspoke with owners and workers of shops, restaurants, hotels, and a tourist office in May 2014 to constrain working seasons, business hours and consumer prices. We also spoke 570 with the tourist office in Lipari to determine the number of tourists visiting Vulcano. This was supported with online research (2014-2016) to assess hotel prices that could not be obtained through discussions with personnel onsite. Several booking websites were used in case the hotel did not have its own website.
While there are two main beaches between Porto and Vulcanello that serve as the main attraction for visitors overall, one of the most popular touristic activities on Vulcano is the mud pool. in the Porto 575 Levante area. The mudpool sits on a fault lineament between La Fossa and Vulcanello and was initially developed around an exploration drilling site for geothermal exploitation drilled in the 1950s (Faraone et al., 1986;Gioncada et al., 1995). Many people visit the island only for this reason. Tourists mostly come to the island during summer and in addition to visiting the mud pool they also like to taste the local cuisine, to and take boat tours around Vulcano and/or around other Aeolian Islands. Hiking to and 580 around the summit of La Fossa and daily visits to other islands are also popular activities. A variety of lodging and accommodation solutions are available on the island (Fig. 1b).  (Fig. 11b). Those open during the middle season are not fully occupied.
On the contrary, fromFrom June until September, they all operate almost at full capacity. In addition to 585 the hotels, there were 21 B&Bs, hostels and residences with two camping areas, as well as 40 apartments. and July, because this is when the vineyard flourishes and becomes more susceptible to pests. They sell the wine mainly in Vulcano to hotels, restaurants and the grocery stores/supermarket and export some product to the mainland Italy and the USA and Japan rather than the other Aeolian Islands.

Methodology to calculate the revenues from touristic business activities in Vulcano 625
Our analysis focused on the turnover created by tourism-related businesses, which provides the main income to the island's economy. The turnover represents the gross revenue that a business generates without considering associated expenses (e.g., food, water, energy and maintenance). The economic impact associated with an evacuation of the island is represented by the loss of this revenue. This revenue must not be confused with the added value provided by the national accounts, which includes the profits, 630 wages, interest and amortizations, but not the intermediate goods and services. We do not consider the revenues from the grocery stores, supermarket and shops due to lack of sufficient reliable data nor the revenues from the cheese and wine factories because they are not tourism-related businesses. The income from maritime transport is also not included, because it does not have a major impact on the local economy. 635 As mentioned in Table 1, different categories of costs are concerned when dealing with impacts from natural events such as those involving volcanic unrest and eruption that might necessitate an evacuation of people from the island. Our focus on Vulcano was identifying tangible business interruption cost related to interruption of touristic activities. The revenues, expressed by the turnover, for all the touristic activities on the island, which will become the loss in case of an evacuation, are calculated for different 640 seasons as part of the cost assessment. The main touristic business on the island can be divided as B&B'sBs, hostels and residences, restaurants and bars, hotels, leisure activities and shops. Each of these are described below.

B&Bs, hostels and residences 645
Data were collected from the internet and field interviews for seven7 B&B, oneBs, 1 hostel and six6 residences (out of 21), but; we were unable to obtain data for seven otherthe remaining 7 structures due to lack of online information.). The revenue for each season is calculated by multiplying the capacity, the price, the total days and the occupancy rate. In the equations below, H and M indicate high and medium season, respectively. 650 RH = C * PH * TH * OH (Eq. 1) RM = C * PM * TM * OM (Eq. 2) RH and RM represent the total revenues for high and middle season, respectively. C representrepresents the total capacity, i.e. maximum number of people that can be accommodated, at a given place. PH and PM are prices per night per person; TH and TM are number of total days estimated in calculations and 655 OH and OM are occupancy rates, i.e. the proportion of available accommodation occupied. As there are no official statistics available, simple assumptions are made for occupancy rate that are based on observations done over more than 10 years of research on the island, which are expected to be reasonable within a margin of 5-10%. During high season, a rate of 100% is estimated and for middle season, the value of 50% is used. 660

Restaurants and bars
Dine-in data (i.e. meal prices and total days open) was collected from discussions with owners/workers at 11 of the 24 restaurants and bars, but "take-away" (dine-out) revenues are not included.  middle and low season, the time a table turns is fixed at '1', considering a restaurant never works on full capacity during these seasons. TH, TM and TL represent the number of total days and OH, OM and OL represent the occupancy rate for high, middle and low season which is 100%, 50% and 15% respectively.

Hotels 685
We were able to collect the required data (capacity, prices, opening season) for 12 out of 17 hotels. As for the other missing facilities, there was no official website or they were not open for us to speak with them when data were collected. In the equations below, H and L indicate high and low season, respectively. M1 and M2 represent the two subgroups ofM represents middle season. season, respectively. In contrast toPH, PM and PL indicate the seasonal classificationprice of restaurants, hostels a room for high, middle and B&B's, low season. Even though the middle season for hotels is dividedcould be subdivided into two subgroups due to high differences in prices. The first subgroup includes May and October, while groups as the second one includesmonths of June and September. These months are considered together because the price per night per person is more or less 700 are busier than those of May and October, the same value was assumed, considering the same. When calculating the revenue for the first subgroup, the price by night is the one that is used for the low season even though we are on middle season. TM1limited data. TH, TM and TM2 areTL represent the number of total days for May-October (62) and June-September (60). OM1OH, OM and OM2 areOL represent the occupancy rate for high, middle and low season which is estimated as100%, 50% and 60% in 705 calculations for May and October and June and September,15% respectively.

Leisure activities 710
While touristic attractions contribute an important amount of revenue to the economy of the island, they TH is the total days of high season (July and August, 62 days), whereas OH is the occupancy rate during the high season, i.e. the proportions of activities occupied, and RH represent the revenue for high season.
The middle season is also divided into two subgroups, as in the case of hotels, but considering different temporal distributions. The first subgroup considers only June (TM1 total days equal to 30) because the revenue is remarkably higher than the total of the rest of the middle season months. The second subgroup 725 consists of April, May, September and October (TM2 total days equal to 122). OM1 and OM2 are the occupancy rate which is estimated as 90% and 40% in calculations for the first and second subgroups of the middle season, respectively. TM is the total days for the middle season (April-May-June and September-October: 152 days) and the occupancy rate is set to 40%. 730

Figure 6. Plots of evacuation simulations for low-season scenario showing: a) a simultaneous evacuation, b) a staged evacuation, c) percentage of people evacuated with time, d) variation of exposure with time
minutes(~8.9 hours for both). For the latter one, although both scenarios have equal evacuation times, their evacuation effectiveness differ (FigsFig. 7c). During the low season, a 95% evacuation effectiveness is reached within 348~5.8 and 392 minutes~6.5 hours for the simultaneous and staged 740 evacuations, respectively (Fig. 6c). For the high season, a similar effectiveness is reached within 365~6.1 hours (simultaneous) and 447 minutes~7.5 hours (staged) (Fig. 7c). These results have two implications. Firstly, the simultaneous evacuation results in less people left exposed to increasing hazard over time, which confirms findings from previous studies (e.g., Chen and Zhan, 2008;Jumadi et al., 2019). Secondly, an increase of population of 360% of population between the low and high seasons 745 results only in an increase in evacuation time of ~12%. In fact, assuming that warning time and preparedness time distributions are independent of population size, the main aspects that could impact the evacuation time are the pedestrian speed and the number and capacity of the boatsferries used for evacuations and the pedestrian speed.evacuation. For the case of Vulcano, the relatively large capacity of the boats can equally accommodate the increase of population during the high season and the 750 pedestrian density in the roads considered under both scenarios does not impact pedestrian speed (the population density in the space, in our case roads, increases beyond 1 person per square meter, which is not reached in Vulcano).increases beyond 1 person per square meter, which is not reached in Vulcano).
In addition, the number and capacity of the ferries can equally accommodate the increase of population during the high season. If a larger number of people had to be evacuated (e.g., 10,000 people as supposed 755 to 4,600), the time needed to evacuate 95% of the population would nearly double because of the number of ferries (10) and associated capacity (200 to 800 passengers) set in the simulation. However, 1,000

Figure 6. Plots of evacuation simulations for low-season scenario showing: a) a simultaneous evacuation, b) a staged evacuation, c) percentage of people evacuated with time, d) variation of exposure with time
people (considered in the low-season scenario) and 4,600 people (considered in the high-season scenario) can be almost equally managed by the capacity and number of ferries used.

Determination of revenue on Vulcano
With the methodology explained in section 4.3, the revenues from four different categories (hotels, hostels-B&Bs-residences;residence, restaurants; hotels, and leisure activities) are calculated. Hotels and 765 restaurants are the only two categories providing revenues during low season ( Table 3). While calculating the monthly revenue by using equations 6 and 10 the number of total days (TL) considered

Revenues in Low Season (€)
Business Activity is 30 and the occupancy rate (OL) is set at 15%. With a monthly amount of 148,275about 153,233 €, the revenue from hotels is 4.54 times greater than the revenue from the restaurants during low season.
From the endbeginning of April, (beginning of middle season), the tourist population starts to increase 770 on the island. Equations 2, 3 and 5 are used to calculate the monthly revenues from hostels-B&Bsresidences and restaurants. The number of total days (TM) for a month considered is 30 and theThe occupancy rate (OM) used for this season is 50%. On the other hand, while calculating monthly revenue for the middle season for hotels and leisure activities, an average is taken due to different occupancy rates (OM) throughout the season. As seen in equations 8, 9, 12 and 13, the season is divided in two 775 subgroups for these two categories. Thus, first the daily revenues are calculated for each month with designated values, e.g. forFor leisure activities 90% of occupancy is considered in June whereas 40% of occupancy is considered for September. Then, an average is taken to determine the daily revenue during middle season. After that, the number of total days (TM) considered to calculate the monthly revenue is 30., whereas 50% of occupancy rate is used for hotels. As seen in Table 4, hotels provide more than 780 half (6258%) of the monthly revenue for middle season with a 1,302,927939,610 €, whereas the restaurants, hostels-B&Bs-apartments and leisure activities provide 22%, 1029%, 12% and 61% of the monthly revenue, respectively. The touristic population reaches its peak point during July and August. Thus, the occupation rate (OH) is considered 100% for all the categories. The total number of days (TH) for a month is taken as 31, i.e. 785 representing July and August. The revenues are calculated by using equationsEquations 1, 3, 4, 7 and 1110. Hotels and restaurants providesprovide the highest revenue for this period with 44% and 46% of total revenue respectively, whereas leisure activities and hostels-B&Bs-residences contribute 4% and 6%, respectively (  the number of times a table is occupied by different groups, is considered 1. However, during July and 790 August, restaurants are full of tourists and a table in a restaurant is served more than once. Thus, the number of TT varies for each restaurant while calculating the revenues for high season.
It should be noted that the prices for hotels, hostels-B&B-residences are not constant during different seasons, and, in fact, they slightly differ for each month. The highest prices throughout the year are applied for the second and third week of August 795 which is considered as summer vacation in Italy.
An average price for each season is calculated based on website data. Additionally, while calculating the revenues for each season, different occupancy rates are 800 considered to obtain a range of revenues. For example, during low season it has been considered as varying between 5% and 15%. More thanrevenue for each season. About 805 half (51%) of the yearly revenue (35,510,78233,692,640 €) comes from hotels ( Table 6). The other half is divided between the remaining three groups, with restaurants providing the second highest revenue after hotels with 37% of total revenue41% of total revenue. Even though in Tables 3, 4 and 5 the monthly calculation is based on 31 days, the number of days considered in Table 6 is related to each individual month. 810

Business Activities
Revenues ( Table 5 Revenues for each business activity during high season. 14 days are used for two weeks and 31 days, representative for July and August, are considered for a monthly revenue.

Business Activities
Revenues ( Table 6 Total annual revenue for Vulcano Island resulting from hotels, restaurants, B&B's (including hostels and residences) and leisure activities 5.3 Analysis of potential economic impact of an evacuation Figure 8a shows that theThe evacuation results in a very different impact in the island's revenue whether it starts at the beginning or at the end of the tourist season. The total loss of revenue (expressed by the turnover) is significant if the evacuation begins in June and lasts for more than one month (i.e., > 1.5 815 million €; Table 7, Fig. 8a). If it starts in November, the impact becomes significant if it lasts more than 6 months-1 year (i.e. >30., >2 million €).€; Table 7, Fig. 8a). The high season represents the critical period. The impact of an evacuation starting in November and June in the two Vulcano main touristic areas (Porto and Vulcanello) is also considered (Table 7, Fig. 8b,c8). A partial evacuation of Piano was not considered because most of the tourist infrastructures are located in Porto and Vulcanello. and, 820 therefore, most of the turnover is related to activities in Porto and Vulcanello. Clearly the evacuation of only Vulcanello would result in a smaller loss of revenue with respect to a partial evacuation of Porto for any of the durations considered (i.e. <15., <10 milllion euros).€). However, in the case of escalating  Table 7. Loss of turnover related to tourism due to total or partial evacuation of Vulcano island (data based on Tables 3, 4 and 5) Table 7 for original data).

Figure 8 a) Total loss of revenue (€) for different evacuation periods starting in November and in June (whole Vulcano Island) and partial evacuation of Porto and Vulcanello starting b) in November and c) in June (see
unrest activity, the safety of people is typically prioritized with respect to economic factors. As a result, the areas that are the most exposed to the hazard (i.e.., Porto) would be evacuated first. 825 6 Discussion

Effectiveness of evacuation
The main objective of our study is to provide decisions makers with an operational tool to investigate various evacuation scenarios. This evacuation simulation tool allows emergency managers to identify 830 and optimize individual and organizational parameters (related to actions, behaviours, policies and resources) that minimize the evacuation time as crises evolve. The tool allows to estimate such key indicators as the minimum time necessary to fully accomplish the evacuation which, in the context of volcanic crises, can be compared to eruption forecasts provided by monitoring networks. Together, these two aspects provide a comprehensive picture of the various components to achievepursue successful 835 emergency management.
However, although the overall evacuation time and the individual evacuation time are vital measures for enhancing the effectiveness of the evacuation process, they do not fully consider the dynamics of hazard and exposure during a volcanic eruption. In volcanic eruptions, hazard and exposure vary in time  Table 7. Loss of turnover due to an evacuation (in 1,000 Euro) Table 7 for original data). and space. In other words, the risk can increase because the probability of eruption might increase with 840 time and because the actual exposure could be significantly higher a few hours after the evacuation order is issued compared to the first hour due to the movement of people towards the evacuation areas (e.g.., ports), which are sometimes closer to the source of the eruption (La Fossa) than where they initiated evacuation (e.g., Vulcanello). Therefore, to reduce exposure the goal should be to evacuate more people faster (Han et al., 2007). The spatial exposure on Vulcano is complicated by to the proximity of the main 845 port (Porto Levante) to La Fossa crater (i.e Porto Levante is located at the foot of the northwest flank of La Fossa). Particularly, for people in Vulcanello moving towards Porto Levante to evacuate requires that they get closer to the hazard source at La Fossa. Evacuating people from these ports can, therefore, increase the exposure in time and space. While optimizing evacuation requires that evacuees move away from the hazard source, evacuation of people in the north side of Vulcano to either Porto Levante or 850

Figure 8 a) Total loss of revenue (€) for different evacuation periods starting in November and in June (whole Vulcano island) and partial evacuation of Porto and Vulcanello starting in b) in November and c) in June (see
Ponente cannot be done without moving people closer to the hazard source, especially when moving people to the Porto di Levante because it is closer to La Fossa than Porto di Ponente. Exposure could be reduced by moving people from the Porto di Levante area to Porto di Ponente, but the latter port cannot accept large shipsferries nor handle large volumes of people. It is, in fact, significantly smaller and characterised by shallower water than the port facility at Levante. Therefore, the planning of an effective 855 evacuation should assess the evacuation time as well as the temporal variation of exposure. For the case of the two evacuation scenarios described above, exposure was assessed based on the distance from La Fossa volcano and was found higher for the staged evacuation (Fig. 6d) during both seasons, with an increasing difference over time during the high season (Fig. 7d).
Some assumptions have been made to carry out our evacuation simulations that should be mentioned: i) 860 peoplethe evacuation starts before the eruption (so evacuation operations are not disrupted by volcanic hazards), ii) people are not allowed to return to the island after the alarm has been issued, iiiii) people are only allowed to evacuate by foot (for the sake of these simulations; however, some people might try to drive to ports causing traffic jamsa combination of evacuation strategies can also be considered in the future (both by foot and road blocks), iiimotorized vehicles), iv) people with disabilities are considered 865 in the simulations by using a low walking speed (; however, other considerations could be made in order to improve the analysis, iv) the "evacuation preparedness time" includes the time required to organize departure and secure the belongings that are left behind (e.g. house, car(s), other., integrating evacuation with dedicated motorized vehicles, boats), v) people might be able to take with them small pets, vi) animals of farming activities (e.g. goats, cows) are not considered here but represent a critical aspect for 870 an island such as Vulcano, vii) evacuation is carried out from the three ports available on the island (i.e. Porto Ponente, Porto Levante and Porto Gelso) even though the only port that can be accessible by large boatsferries is Porto Levante (more studies should be carried out based on the actual evacuation capacities of Porto Ponente and Porto Gelso, and in various weather and marine conditions). Finally), viii) people follow the instructions provided in the evacuation orders (this is particularly important for 875 staged evacuation as people in each community are asked to evacuate according to their turn; the possibility of having a fraction of the population not following the order of staged evacuation can be included in the simulations in order to add a level of uncertainty). In addition, while we did not directly include social vulnerability aspects due to small community size and lack of up-to-date data, the current evacuation simulation tool can be enhanced to include social vulnerabilities, especially if it is going to 880 be used in larger and more complex social systems. The simulation can be parameterized based on more granular detail on socio-demographic characteristics of the agent population. This will allow to includeinclusion of social vulnerability factors related to age, health conditions, gender, language, education, access to resources and information in the evacuation simulation tool. Finally, it is important to consider and discuss some stochasticity and uncertainty aspects of the proposed evacuation simulation 885 tool. Given that most of the distributions we have used to describe the various evacuation parameters are uniform, the stochasticity and uncertainty are relatively low, and the different simulations do not produce significantly different results. The main source of uncertainty in our model is related to the random distribution of population and capacity of the ferries. However, more parameters can be varied in order to explore a wider range of conditions. 890

Assessment of the economic impact of an evacuation of Vulcano island
The loss of revenue due to touristic business interruption associated with an evacuation of Vulcano Island is studied as a function of time, in order to investigate the influence of different touristic seasons, and as a function of space, in order to investigate how a partial evacuation affects the economic loss on 895 the island. According to our results, both the time when the evacuation process is carried out and the duration of the evacuation period have significant impact on tourism. For instance, a short-term evacuation (i.e.., up to three months) during low season (e.g.., November, December to January) causes less than oneabout 0.5 million Euros€ of revenue loss (about 550,000 €).. Should peoplethe island be evacuated for 6 months, the loss could increase up to about 2.5 million € only after six months due to 900 an overlap with the beginning of the middle season when touristic activities start to resume. One year of total evacuation on the island causes about 3534 million € of revenue loss. Only 52% of this loss results from evacuation during low season. (about 0.5 million €). This is due to the fact there are no tourists on the island during these months and most touristic activities ceasedstop. The situation is, therefore, critical if the evacuation needs to be carried out towards the end of the middle season (e.g.., June) and/or 905 during the high season when the population on the island reaches its peak point. In such a case, a monthlong evacuation in June is almostabout 1 million € higher than 63 months of evacuation during the low season (i.e. starting from., November to January). After that, a rapid increase in revenue loss is observed on the island: three months of evacuation starting in June causes up to 2826 million € of revenue loss which corresponds to 8078% of the one-year loss because it includes the high season. 910 InIt should also be considered that an evacuation during the low season could affect or compromise also the high season, due to the typical maintenance works of the touristic infrastructures performed during the low season and the impact on preparation activities (e.g., hotel booking). However, eruptions also attract tourists, as recently shown by the 2021 crises of Cumbre Veja (La Palma, Spain) and Fagradalsfjall (Iceland). As a result, the overall impact on the high season revenue of an evacuation 915 during the low season due to an eruption of La Fossa would be difficult to forecast. Finally, in addition to the high revenue loss that could occur during the high season, it is important to note that the evacuation process becomes more complicated due to the high number of tourists between June and September (in addition to the diversity of languages represented by international tourists and workers), whereas an evacuation between November and April concerns only local people, all of whom would presumably 920 speak Italianmostly concern residents. The loss of revenue on the island is also considered as a function of space. To do this, partial evacuations including only Porto or only Vulcanello are evaluated. The main reason for assessing the partial evacuation is to be able to maintain at least some activities on the island, without interrupting all tourismdependent businesses and also to see which part of the island has the highest impact on the economy. 925 According to our results, during the low season the loss of evacuating Vulcanello is slightly higher than the loss of evacuating Porto (lower than a(<1 million euros).€). Although most of the touristic facilities and all the restaurants are located in Porto, the largest hotels on the island are all situated in Vulcanello and they are open for the whole year (Therasia Resort Seas and Spa and Jera Residence). However, with the beginning of middle season the revenue loss in Porto exceeds Vulcanello. If the evacuation includes 930 July and August, the loss resulting from evacuating Porto is more than double of the loss of evacuating Vulcanello.
Piano is not considered in the partial evacuation scenarios. In; in fact, on this southern side of the island, there are no shops, hotels or any other leisure activities to attract tourists with the exception of a famous lookout (Capo Grillo) and small beaches. Only two B&B'sBs and two restaurants are located in Piano 935 with revenues negligible revenue compared to those located in Porto and Vulcanello. However, this does not mean that Piano has no effects on Vulcano's economy. As mentioned earlier, the wine factory is situated between Piano and Gelso. According to the owner, the vineyard flourishes between March and July. Thus, if an eruption occurs during this period and the area is evacuated, there willwould be at least 60,000 € of loss generating from Piano. AlsoIn addition, the important infrastructurescritical 940 infrastructure that areis not considered in the cost assessment of this study, such as the solar plant located in Piano, may cause problems for other businesses. For example, if the electricity is cut on the island, the restaurants and hotels cannot function and this affects directly the tourism and thus the revenues, even though the evacuation is partial, and that part of the island is not affected.
Cost assessments are also required to conduct Cost-Benefit Analysis of different mitigation measures. 945 Although evacuating the island will cause an economic loss (i.e.., the loss of revenue as the cost), it is a key measure to reduce the impact on public health. It helps ensure the prevention of eruption related injuries and deaths, hence the main components of benefits. Quantifying the value of life is an ethical issue. Although there are studies that try to assign a value to a human life (e.g.., Cropper and Sahin, 2009), here we do not consider it, as this is beyond the scope of our analysis. In any case, if an eruption 950 on the island is imminent, total and/or partial evacuations will be conducted regardless of the cost in order to avoid casualties. However, it is important to evaluate the socio-economic impact on affected communities for authorities, in order to help them to implement informed decisions.
Although this study has provided some significant findings on the tourist sector of the economic system of Vulcano Island, such as the main income activities and the possible loss in case of an evacuation, it 955 does not provide a complete picture for the cost assessments, and some important caveats need to be discussed. The first and the most important limitation concerns the lack of data. Although the municipality of Lipari was visited in May 2014, no access was granted for any official data concerning the economic situation of the Aeolian Islands, let alone Vulcano itself. All the data used to calculate the revenue on the island were based on our field visit in May 2014, official websites and various booking 960 websites were used to complete the data set. Another important point to mention is that the main focus of our study is on the revenuesrevenue originating only from tourism-related businesses, and, therefore, the total cost of evacuation process is not investigated (i.e.., cost of evacuation operations and of relocating people). This loss presents only one part of the total cost associated with an evacuation. An extensive cost assessment requires the consideration of all different types of costs involved with the 965 evacuation process. Municipality of Lipari, defines the mitigation actions that all the stakeholders involved in the emergency management must take and foresees the evacuation of the population before the eruption onset in case increasing volcanic unrest is observed. The results of the work presented in this paper as well as the 1015 insights into risk assessment presented in  were used to finalise the emergency plan at national level. However, given that the recent unrest was mostly related to gas hazard and was, therefore, managed at the local level (i.e., by the Municipality) and that the overnight restrictions issued by the municipality did not involve an evacuation of the island, the results of these two analyses could not be tested nor validated. 1020 i) whether or not unrest will lead to an eruption, ii) the nature of explosive activity (magmatic or hydrothermal), iii) the eruptive style (i.e., effusive, explosive or both), iv) the potential activation of lateral vents, v) the eruption magnitude (i.e. erupted volume) and intensity (i.e. the rate of discharge of magma, plume height), vi) the type, extension and timing of hazards with the potential to impact human life and the infrastructure supporting evacuation whether occurring either in the unrest phase or eruptive 1025 phase or both. The interpretation of scientific data complicates the decision-making process for the officials (Fearnley, 2013). Higher levels of scientific uncertainty may thus translate to increased difficulty for emergency managers to understand the value of evacuation (measured in terms of human lives saved) and the costs associated with any evacuation that is not accompanied by the occurrence of hazards necessitating eruption. 1030

Potential negative and positive economic consequences of a volcanic crisis
When a volcano begins to show increasing signs of unrest above background level, authorities have to deal with uncertainties and decide how to manage a potential crisis (e.g. have people shelter in place or evacuate some or all of the population), as scientists cannot guarantee if the unrest will result in an eruption or not. Although successful forecasts have been made (e.g. Mt St Helen's 1980, USA;Mt Redoubt 1989-1990Pinatubo 1991, Philippines), false alarms that cause both scientists and 1035 officials to lose credibility also occurred in the past (Sparks, 2003;Tilling, 2008). For example, during 1983-1985 volcanic crisis at Rabaul Caldera (Papua New Guinea), the government practiced many evacuation exercises, which led to voluntary evacuations by villagers. They intensified disasterpreparedness activities when intense earthquake swarms begin to occur in September 1983 and continued until April 1984. Although there was a high expectation that an eruption was imminent (i.e. 1040 that eruption would take place) by early 1984, the number of earthquake swarms and their intensities suddenly decreased. The government subsequently dropped the alert level in November 1984 and by mid-1985 the seismicity returned to its pre-1983 levels (Hastings, 2013). Consequently, the volcanic crisis resulted in substantial losses of revenue due to business interruptions with the total cost of emergency preparations exceeding 20 million PNG Kina (~21 million $). At the end, many people 1045 thought that two years of preparation was a waste of money (Hastings, 2013). Nevertheless, some benefits also emerged from this crisis, as public awareness of potential volcanic hazards increased and the community became more resilient (Hastings, 2013;Tilling, 2008).
Unfortunately, successful forecasts followed by evacuations may also cause economic distress for communities located in hazardous areas. As an example, in October 1999, almost 19,000 people were 1050 evacuated from Baños, Ecuador when Mt. Tungurahua renewed activity after a long period of quiescence. Some 95% of the community's economic activity was dependant on tourism (Lane et al., 2003), showing a similar situation to Vulcano Island. After the evacuation, an economic crisis was felt both locally and nationally. In the city of Ambato, where evacuees were rehoused, unemployment was an issue, health costs increased by about 103%, and food and beverage prices increased by about 108% 1055 (Lane at al., 2003). When authorities realized that economic recovery would be hard without tourists, the tourism industry launched an effective campaign to promote positive views of the area by using the volcano's attractiveness, to convince both domestic and foreign tourists that the situation in Baños was back to normal. Even journalists were invited to the town to report on the successful recovery (Lane et al., 2003). Finally, in 2000, Baños attracted approximately 23% of the country's 615,000 foreign 1060 visitors. In November 2001, 56% of all tourists visiting Baños were foreigners (Lane et al., 2003).
As seen in both cases at Rabaul and Baños, a volcanic crisis, if not managed well, can easily result in an economic crisis, with or without an evacuation and an eruption occurring. This is also valid in the case of Vulcano Island. Even without an evacuation order, the increasing level of unrest may cause the local people to leave the island, if they believe that tourism on the island may be affected negatively by the 1065 increasing volcanic activity. This is especially true since most business owners are not from Vulcano and they may decide to relocate their activities. In both cases, the economy of the island would be negatively impacted. Additionally, there could be significant negative economic impacts on Vulcano associated with changes in the volcano alert level even when an eruption or evacuation does not occur, as Peers et al. (2021) described for the protracted unrest at Long Valley Caldera, California, in USA. 1070 Volcanic unrest and eruptions can also have positive impact on economy. As an example, volcano tourism and geotourism has become more and more popular all around the world. It is estimated that between 150 and 200 million people visit volcanic and geothermal environments on an annual basis (Heggie, 2009;Erfurt-Cooper, 2011), because a growing number of tourists seek adventure by planning holidays close to active volcanoes (Brace, 2000;Erfurt-Cooper and Cooper, 2010). As an example, in 1075 2008, 1.2 million tourists visited the active volcanic features in Hawaii Volcanoes National Park, 3 million visited the geysers and hot springs of Yellowstone National Park and in 2004 103 million people visited Fuji-Hakone-Izu National Park in Japan (Heggie, 2009;Erfurt-Cooper, 2011). Other than USA and Japan, geothermal and volcanic activity in Italy and Iceland are also highly attractive destinations for tourists (Heggie, 2009). Research by Bird et al. (2010) in Thorsmork, Iceland near Katla Volcano in 1080 2009 examined the relationship between tourism and volcanic activity. They found that all the participants (tourists) knew that Iceland is volcanically active, but they do not think of volcanic eruptions as hazardous events, hence they lack hazard knowledge. Additionally, most tourists and tourism employees think that tourism will benefit positively after a future Katla eruption. However, according to results of Dominey-Howes and Minos-Minopoulos in 2004 in Santorini, Greece, it is the residents 1085 who fear that a future eruption may have a negative impact on the tourism.
Vulcano Island appeals to a wide range of tourists: some visit to relax and/or for health reasons, whereas others are attracted to volcanic landform and geothermal features. Thus, an increase of unrest may attract more adventure-driven tourists, unless such visits are curtailed by civil authorities as a result of increased likelihood of eruption and resulting limitation of the number of people on the island. If the increasing 1090 activity on the island results in an evacuation and finally in an eruption, still many tourists interested in natural areas and adventure may want to visit the island once the activity is back to pre-eruption and the risk is decreased. This type of tourism should be foreseen and well organized to boost the local economy especially after business disruption due to evacuation and/or eruption.

Conclusions
Evacuation is often the only strategy to save lives in case of extreme volcanic activity and rapidly escalating unrest. This is especially critical for La Fossa volcano whose activity has been characterizedcharacterised by hydrothermal events, which are typically very sudden and unpredictable, and magmatic events with little warning signals (e.g.., 1888-90 Vulcanian cycle). In such a case and 1100 considering the high level of population exposure to dangerous hazards, evacuation shouldmight be considered even in case of weak unrest. Nonetheless, the timing and routing of evacuation is critical to remove people from the hazardous zonehazard zones before it isthey are impacted.
The Vulcano evacuation simulation toolEvacuation Simulation Tool decribed here has been developed to test the effectiveness of ABM simulation in evacuation planning for areas subject to volcanic hazards 1105 on a small island. Based on a pre-eruption simulation at Vulcano, we have demonstrated that the both the simultaneous and the staged evacuation are slightly faster during the low touristic season (401(~6.7 and 427 minutes~7.1 hours to evacuate 1,000 people, respectively; Fig. 6a,b) with respect to the high touristic season (535 minutes(~8.9 hours to evacuate 4,600 people; Fig. 7a,b). Nonetheless, we have also shown that the type of evacuation (i.e.., staged or simultaneous) can optimize the number of people 1110 evacuated in time, with the simultaneous evacuation being more efficient at removing people from the island than the staged evacuation, especially in the . In fact, after 5 hours (300 minutes), during low season, about 84% and 72% of people would be evacuated with a simultaneous and a staged evacuation, respectively, while after ~6 hours (370 minutes), during the high season, about 96% and 86% would be evacuated with a simultaneous and a staged evacuation, respectively (Figs. 66c and 77c). Additional 1115 analyses should be carried out to explore more evacuationsevacuation conditions (e.g.., evacuation by car, evacuation from fewer ports, evacuation after the onset of the eruption) orand the role of social vulnerability. In fact, the proposed evacuation simulation tool can be used to model varying impacts for different scenarios to enable proper allocation of resources required for evacuations and economic support of the affected areas. 1120 We have also shown how, in an island aslike Vulcano whose economy is based on tourism, the timing and duration of evacuation can have a very different impactsimpact. In fact, if the evacuation of the whole island starts inat the lowbeginning of the low touristic season (e.g., November), the impact becomes significant only if it lasts more than 6 months-1 year, (> 7% of annual total turnover), whereas if it starts in June (i.e., at the end of the middle season and approaching the high touristic season in July-1125 August) the impact becomes significant after 1 month (> 5% of annual total turnover) and reaches 78% of the annual total turnover after 3 months. In particular, our Our results also show that a total evacuation starting in June, for a period of 6 months or less, result of various durations results in ~9588-98% more revenue loss than thean evacuation starting in November. This is directly related to the large number of high tourist populationtourists on the island during that period.the high season (July-August). In 1130 addition, if thean evacuation starts in November and lasts for of up to 63 months, there is no in the low season would not produce a large difference in revenue loss between the partial evacuations of Porto and of Vulcanello in terms of revenue loss. On the contrary, if thean evacuation starts in June instead of November, the revenue loss resulting from evacuating Porto is 30-50of 1 year would cause 60% higher than evacuating Vulcanello. Moreover, for an evacuation lasting more than 6 months (e.g. one year), the result of evacuating Porto causes 50% of higher revenue loss than evacuating Vulcanello. Consequently, we can say that for a partial evacuation, evacuating Porto starting from June will cause the largest impact on the island's economy. This is due to the fact that all leisure activities and restaurants with the majority of hotels, hostels and B&B'sBs are located in Porto. However, it is important to stress that human life has to be prioritized over economic losses, therefore being the most exposed area to volcanic hazard, 1140 Porto should be evacuated even though it is Regardless of the beginning of the evacuation and associated with the highest revenue losses. Finally, regardless of the timing of the evacuation and its duration, the total evacuation of the island generates 30-50would generate 28-53% more revenue loss than the partial evacuation of Porto, and 45-7047-73% more revenue loss than the partial evacuation of Vulcanello.
It is important to stress that, during crises, human life is prioritized over economic losses, whatever the 1145 situation. However, our analysis provides insights into the potential economic impact, which could be mitigated if foreseen and integrated into an emergency plan.