category: literaturenote citekey: onelcinevacuationplanindustrialzone2013 title: "Evacuation plan of an industrial zone: Case study of a chemical accident in Aliaga, Turkey and the comparison of two different simulation softwares" authors: "Onelcin, Pelin; Mutlu, Mehmet Metin; Alver, Yalcin" year: 2013 date: 2013-00-00 2013 doi: 10.1016/j.ssci.2013.07.006 publication: Safety Science url: "https://doi.org/10.1016/j.ssci.2013.07.006" zotero_key: K7ST5TI3 zotero_storage: 7XQGIVCA collections: transpordi_planeerimine folder: 001_artiklid/Road Safety Indicators firstAuthor: "Onelcin, Pelin"

status: converted

Contents lists available at SciVerse ScienceDirect

Safety Science

journal homepage: www.elsevier.com/locate/ssci

Evacuation plan of an industrial zone: Case study of a chemical accident in Aliaga, Turkey and the comparison of two different simulation softwares

Pelin Onelcin ⇑ , Mehmet Metin Mutlu ¹ , Yalcin Alver ¹

Department of Civil Engineering, Ege University, 35100 Bornova, Izmir, Turkey

article info

Article history: Received 13 May 2013 Received in revised form 3 July 2013 Accepted 7 July 2013 Available online 31 July 2013

Keywords: Evacuation Traffic simulation Evacuation time estimate

abstract

Chemical accidents cause great danger for people working in the chemical facilities and residents living close to these facilities where in general evacuation is required in case of an accident. The most efficient way to create evacuation plans is to simulate the transportation network. In this paper, evacuation simulations are carried out for Aliaga, the industrial zone of Izmir, Turkey where petro-chemical companies such as Petkim and Tupras are active. The scenario is based on an explosion in the acrylonitrile (ACN) factory in Petkim. The risk zone is determined considering the features of ACN, the wind speed and the direction of the wind. Total number of evacuees is found 14,472, almost 26% of the population of Aliaga. The second part of this study deals with the simulations executed via mesoscopic simulation software called Cube Avenue and microscopic simulation software called MATSim. The aim of using two simulation softwares is to be able to compare evacuation time estimates and to obtain reliable results from evacuation simulations which are crucial and critical in creating emergency plans. In the final part, a comparison of the two simulation results is presented.

-2013 Elsevier Ltd. All rights reserved.

1. Introduction

The number of man-made disasters has increased in the last decades with the industrial development. One type of these disasters is the chemical accidents which for the most part has an instantaneous characteristic that cannot be estimated beforehand, and appears with explosion and fire.

History witnessed disastrous chemical accidents requiring evacuation such as Seveso, Italy accident in 1976 leading to Seveso Directive fulfill the necessity of a legislation concerning the prevention and control of such accidents developed by European Union in 1982 (European Commission, 2012).

In 1984, the Bhopal, India accident proved the significant role of an evacuation plan. Since the authorities were caught unprepared for this disaster, the consequences were tragic. Several estimations range from 70,000 to 500,000 evacuees (Sriramachari, 2005).

The decision of evacuation or shelter-in-place is made concerning factors such as the available shelters and evacuation time estimates (ETE) (Sorensen et al., 2004). It is apparent that the conditions differ in emergency situations with the entrance of high number of vehicles to the network in a short time leading to an increase in demand. Evacuation time estimate is strongly dependent on the capacity of the evacuation route system being capable of handling the traffic demand (Lindell and Prater, 2007).

Evacuation time can be used in two meanings, one for the time required to move the persons in the evacuation area to safer places, and the other for the time passed from the moment of the evacuation order to the clearance of the evacuation area (Barrett et al., 2000).

The consideration of the behavior of the households is a crucial point in ETE, but, it is generally ignored. In emergency situations some people would tend to enter the risk area to save their children, etc. Thus, the simulation time could give more optimistic results than reality (Murray-Tuite and Mahmassani, 2003).

The number of evacuees shows variations comparing the day and nighttime due to the increase in the population for work, education or special facilities (e.g., shopping centers, etc.) during the day. For nighttime evacuation planning, census information on the number of residents could be sufficient. After determining the number of persons to be evacuated, this information should be translated into the number of evacuating vehicles in order to estimate the area clearance rates and the potential traffic delays (Southworth, 1991). Goldblatt (2004) presents a well-developed list involving the steps in order to identify the best evacuation routes and time estimates.

In emergency situations static assignment models do not allow realistic results as the traffic conditions remain constant during

^⇑ Corresponding author. Tel.: +90 232 388 60 26; fax: +90 232 342 56 29. E-mail addresses: pelin.onelcin@ege.edu.tr (P. Onelcin), mmetinm@gmail.com (M.M. Mutlu), yalcin.alver@ege.edu.tr (Y. Alver).

¹ Tel.: +90 232 388 60 26; fax: +90 232 342 56 29.

simulation period. Dynamic assignment models consider the changes in traffic conditions and those in the level of service (Als nih and Stopher, 2004).

Traffic simulation models have been an assisting tool for evaluating alternative roadway designs, calculating optimum travel times through route analysis or retiming intersection signaling. However, with increase in number of emergency situations, modeling traffic flow via simulation softwares becomes an essential part of transport emergency management plans (Moriarty et al., 2007).

There are three types of models used for modeling the traffic network based on the scale of a geographic area and level of detail; macroscopic, microscopic and mesoscopic models (Hardy and Wunderlich, 2007). Macroscopic models are used to simulate the traffic flow based on speed, flow, and traffic density relationships, and do not model the interactions between individual vehicles whereas microscopic models simulate interactions of individual vehicles (Federal Highway Administration, 2004). The earliest evacuation planning models were due to major accidents and were based on macroscopic scale simulations. Microscopic models are also used for evacuation modeling though the requirement for comprehensive data on the network (i.e., roadway geometry) brings limitations to the extensive use of these softwares (Hardy and Wunderlich, 2007). Mesoscopic models are used to simulate the individual vehicles describing their interactions on aggregate relationships (Federal Highway Administration, 2004).

Mesoscopic models came into use in the sense, filling the gap between the aggregate level approach of macroscopic models and the individual interactions of the microscopic ones. They are able to describe the traffic entities at a high level of detail but, the interactions among them show a lower level of detail which is ideal for prediction applications where a detailed modeling of route choice is sufficient (Burghout, 2005).

There are five mesoscopic simulation models which are being currently used for evacuation modeling: Simulex, TRANSIMS, TransModeler, DYNSMART-P and Cube Avenue (Hardy and Wun derlich, 2007).

This study deals with evacuation of Aliaga, Izmir, where chemical facilities densely take place and presents the evacuation process via simulation models executed in Cube Avenue and MATSim softwares. The need of this study emerged from the lack of an evacuation plan for such an important area in terms of risk being held and the residential areas being located in the vicinity of the chemical facilities conducted by Petkim and Tupras. The factory where the accident would occur was selected as a first step and the evacuation area was determined by taking the risk zones given by Aloha software into consideration. The estimation of the evacuation time was the primary goal of this study. In order to have reliable results, two simulation softwares were used, being MATSim a microscopic simulation model and Cube Avenue a mesoscopic simulation model.

2. Case study: Aliaga

Aliaga is one of the 30 districts of Izmir Province in the Aegean Region of Turkey. The town is situated in about 50 km north of Izmir and has crucial contributions to Turkey's economy. Since Aliaga has one of the greatest petrochemical enterprises of Turkey, called Petkim, the study focuses on this region. In 2011 Petkim was ranked as the 12th greatest enterprise among 500 top industrial enterprises in Turkey (Istanbul Chamber of Industry, 2011). There are 14 plants and seven auxiliary units belonging to Petkim. As the authors live in Izmir city, Aliaga was a perfect target to operate the evacuation studies.

Turkish Statistical Institute announces the population of Aliaga as 68,432 among which 56,440 live in the central neighborhoods and 11,992 live in the villages in year 2011 (Turkish Statistical Institute, 2011).

2.1. Evacuation zone and zone division

Southworth (1991) indicated that applying an evacuation planning model begins with the selection of the evacuation zone boundary. It is believed that outside this zone the hazard in question will not be a threat for people.

The evacuation zone is determined based on a scenario developed for a chemical accident in one of the factories of Petkim. Features of the chemical substances and the National Fire Protection Association (NFPA 704) ratings ranging from 1 to 4, given for flammability, health and reactivity are compared. The NFPA values are shown in Table 1. It is apparent that the most dangerous substance is acrylonitrile (ACN) with the ratings 3, 4, 2 for flammability, health and reactivity, respectively. This is confirmed by the authorities in Petkim (Petkim Petrochemical Holding Seminar, 2011). Thus ACN is decided as the location of the impact. Vinyl chloride monomer, propylene or ethylene could also have been chosen, but, risk zones are smaller and thus lower evacuation times would be obtained.

The radius of the risk zones are found via Aloha software developed by Office of Emergency Management and Emergency Response Division. The software divided the risk area into three zones based on the type of the chemical substance, wind speed and wind direction. The wind data obtained from Aliaga wind measurement station shows that the maximum wind speed is 17 m/s (WolframAlpha, 2012) and the prevailing wind direction is W-NW. The wind blowing from west is the most dangerous one for Aliaga, because it would carry the smoke over residential areas which would increase the number of persons to be evacuated. Aloha software shows threat zones as circles and in the estimation of the evacuation times the clearance time of these circles are used. In this regard not only the west wind was assumed but all wind directions were considered. The risk zones had a radius of 1.7, 2.4 and 3.7 km for the first, second, and the third risk zone, respectively. In Fig. 1, the evacuation zone boundary is shown. All the factories of Petkim and Tupras, a part of the residential buildings, and a part of the E87 (D550) Highway remain within the boundary.

The definition of transportation analysis zone (TAZ) is given as ''spatial entities of at least a few blocks to much larger areas, like a neighborhood'' by Church and Sexton (2002). An origin–destination matrix is assumed between a set of TAZs (Church and Sexton, 2002).

The evacuation zone is divided into 19 TAZs. The trips that are generated from these zones are directed to six destination TAZs of which three of them are health care centers and three of them are gathering-places. Petkim area is divided into six zones where the first zone is in the impact area. Tupras area remains in the

Table 1 NFPA hazard potential ratings.

| | NFPA hazard potential ratings | | | | | | |------------------------|-------------------------------|--------|------------|--|--|--| | | Flammability | Health | Reactivity | | | | | Substances | | | | | | | | Ethylene glycol | 1 | 1 | 0 | | | | | Polyvinyl chloride | 1 | 1 | 0 | | | | | Ethylene | 4 | 1 | 2 | | | | | Polyethylene | 1 | 0 | 0 | | | | | Acrylonitrile | 3 | 4 | 2 | | | | | Vinyl chloride monomer | 4 | 2 | 1 | | | | | Propylene | 4 | 1 | 1 | | | |

Fig. 1. Evacuation planning zone.

second risk area and is divided into two zones due to remarkable vehicle density difference of the area.

2.2. Number of evacuees

The estimation of the evacuation demand requires the assessment of the size of the population which can be achieved through the subdivision of the emergency planning zones (EPZ) population into three groups with permanent residents, transients and special facility population (Urbanik, 2000). In this study, number of evacuees is calculated considering the permanent residents and employees being as transients. The scenario is prepared by taking into account the following conditions:

- The explosion occurs in the evening when there are no students in schools and people are awake.
- The employees in the first risk zone and partially in the second risk zone are taken to the Aliaga state hospital in zone 30, and to the other health care centers in zones 26 and 27. The first risk zone is the most vulnerable zone, thus, persons who need medical intervention in this area should be taken to hospital. The typical behavior of persons in Turkey is to flock to hospitals in emergency situations. In generation of the scenarios this behavior is considered and not only in the first risk zone but also in the second risk zone health care centers are determined as destination zones.
- People living in the residential zones self-evacuate. Since Aliaga is a small town, public transportation service is weak. Moreover, in the evening the frequency of public transportation is low. Therefore, public transportation is not considered in the present study.
- The employees in Petkim and in Tupras work in three shifts. The factories are active for 24 h and there always are employees present in the factories.

The total number of the employees is 2586 for Petkim, and 1297 for Tupras. The number of employees to be evacuated from the factories is computed by means of the following assumptions, which are:

- The total number of the employees is divided into three as they work in three shifts.
- The time of the explosion does not correspond to the time of the shift change.

The number of evacuees from the residential buildings is calculated following these steps:

- The number of the persons living in the neighborhood belonging to the evacuation zone is divided into the number of the buildings in that neighborhood which give the mean number of the persons living in one building.
- The number of the buildings that remains in the evacuation zone is multiplied by the mean number of persons in one building.

Based on the above assumptions the number of employees present at Petkim is 862 and at Tupras is 432 at the time of the accident. Total number of the persons that need to be evacuated from Petkim and Tupras is 1294, and from the residential area are 13,178.

2.3. Number of evacuating vehicles

In literature, number of evacuating vehicles is given on the basis of household (EVHH) for permanent residents. Cova and Johnson (2002) use a Poisson distribution in order to compute EVHH. Surveys are a guiding light in identifying the evacuating vehicle per residential household (EVHH), as well, because they take into account the behavior of the persons (Lindell and Prater, 2007).

In emergency situations people are not expected to behave as they do under normal conditions. Some people would tend to escape on foot. However, considering the risk zones, it would make sense to hypothesize that people would leave by vehicles. Different from vehicles, pedestrians' flexibility for route choice is greater (Federal Highway Administration, 2007). They would use pavements, roads, any space that seems available at that moment, and it is very difficult to model their behaviors since it is not clear which route they would choose. Besides, the available pedestrian behavior simulation softwares in general deal with closed area evacuations (Santos and Aguirre, 2004), and there do not exist many softwares capable of simulating the transportation network under the emergency situations. Pedestrian behavior could be the subject of another study.

Here, to estimate the number of evacuation vehicles, car occupancy ratio rate is used. This rate is 1.57 in normal situations (Istanbul Metropolitan Municipality Directorate of Transportation Planning, 2011) however in emergency situation it is expected to be higher. In this study the car occupancy ratio rate is taken as two. Since the number of evacuees is 14,472, the number of vehicles is computed as 7236.

2.4. Network

The network of Aliaga has been taken from Open Street Map (OSM) and has been converted to a shape file and MATSim network file with NetworkGenerator.class of MATSim. Zones used in generating synthetic population for MATSim were created in Quantum GIS (QGIS). For the simulation executed in Cube Avenue the network was saved as a shape file in QGIS and exported to ArcGIS v.9.3. In Fig. 2, the network in OSM format and the network that is used for simulations are shown.

Functional road classes are used to introduce the transport network. According to a definition given by FHWA ''functional road classification is the process by which streets and highways are grouped into classes, or systems, according to the character of service they are intended to provide''. In urbanized areas there are four functional systems, namely; principal arterials, minor arterial streets, collector streets, and local streets (Federal Highway Administration, 2008).

The transport network that is used for evacuation simulation in this study constitute of a principal arterial, collectors and local streets. The speed limit varies according to the type of the road. For the local streets it is 50 km/h, for the collectors it is 70 km/h, and for the E87 highway (principal arterial) it is 100 km/h.

The capacity values of the roads are a modification of Highway Capacity Manual (HCM) 2010 values. HCM is a widely used source involving methods for analyzing the operations of streets, highways, pedestrian and bicycle facilities. HCM gives the theoretical capacity values per lane. Driver behavior, weather conditions, lane width are all factors that could change the capacity values, hence the local capacity value of the E87 highway is 1050 veh/h/ln, that of the collectors is 450 veh/h/ln, and of the local streets is taken as 300 veh/h/ln in this study.

2.5. Scenarios

Scenarios are generated once the TAZs, the number of evacuating vehicles and the network characteristics are ready to be processed. Three different scenarios have been identified for the evacuation simulation. In Table 2, OD matrices for each scenario can be seen.

The selection of these three scenarios is mainly based on the location of the destination zones and traffic demand. Free spaces which are close to the risk zone are designated as gathering-areas. The time required for the evacuees to reach health care centers is of significant importance in an emergency situation such as in this case. The traffic demand generated for the health care centers is distributed between zone 26, 27 and 30 in the scenarios though the first risk zone is always directed to Aliaga State Hospital for it has the greatest capacity among the health care centers and severely injured persons should be transferred to this state hospital. Evacuating vehicles departing from the second risk zone are directed to health care centers in zone 26 and 27. Changing the number of evacuating vehicles between these zones result in different evacuation times. Thus, these different scenarios could allow the authors to observe the effect of traffic demand on traffic and if this leads to any reduction in evacuation time.

The aim is to provide the arrival of evacuees to health care centers and gathering areas with a minimum time. The simulations based on these scenarios allow to choose the most useful scenario to be implemented in real case.

Fig. 2. Aliaga network files (a) network used in simulations and (b) OSM network.

Table 2 OD matrices for evacuation scenarios.

	Scenario 1						Scenario 2						Scenario 3
	26	27	30	32	33	34	26	27	30	32	33	34	26	27	30	32	33	34
R.Z. 1
1			33						33						33
3			17						17						17
4			33						33						33
6			83						83						83
7			50						50						50
R.Z. 1–2
10			221						221						221
13	16						16						16
14	90						50		40				50		40
R.Z. 2–3
11				837						837						837
15		320					120	200						120			200
17				53						53						53
R.Z. 3
16						660					660							660
18					151						151						151
19					158						158						158
20					216						216						216
21					137						137						137
22					175	280						280						280
23											175						175
24						88						88						88

2.5.1. Scenario 1

In the first scenario, 437 vehicles from evacuation zones 1, 3, 4, 6, 7 and 10 in the first risk zone depart for the Aliaga State Hospital in zone 30. Apart from Aliaga State Hospital, there are two more health care centers outside the boundary of the third risk zone which will accept evacuees from the second and the third risk zones. There are three gathering-areas designated as a priori for the evacuees in case of emergency situations and the individuals are assumed to be informed about their own evacuation destinations. The gathering-areas are designated as zone 32, 33 and 34 which 890, 837 and 1028 vehicles will reach, respectively.

2.5.2. Scenario 2

The second scenario differs from the first scenario in number of persons that will reach health care centers. In this case, 477 vehicles head for Aliaga State Hospital, 186 for zone 26 and 200 for zone 27. The number of vehicles going to gathering-areas remains the same as in the first scenario.

2.5.3. Scenario 3

Destination zones of 200 vehicles in risk zone 3 which were sent to the health care centers in the first two scenarios were changed in this case, and their destination was assigned to zone 33. The aim is to observe if the number of vehicles lead to a decrease in evacuation time.

The number of scenarios can be increased; indeed numerous scenarios can be created. However the scenarios should reflect the possible behavior of the evacuees. It is proved for many times that in Turkey the tendency is to try to reach the health care centers.

In industrial areas like Aliaga, it is important to have previous knowledge about the emergency situations and what should be done under those circumstances. In the present study, persons are assumed to be informed about the destination zones, thus they know where the health care centers are located.

The less risky zones can be diverted to the gathering areas that are identified beforehand. Besides, the health care centers gathering areas are also determined as destination zones.

3. Methodology

A simulation of a transportation network helps to observe various alternatives before implementing them in real world (Frick, 2011).

Simulation plays a significant role in traffic management research in evacuation (Duanmu et al., 2012). In the past decades, for the purpose of evacuation modeling, transportation engineers have become involved in the application of modeling and simulation systems. (Murray-Tuite and Wolshon, 2013).

The use of simulation-based dynamic assignment model provides two fundamental advantages over analytical model. One is the modeling and observation of vehicle-to-vehicle interactions and the other is evaluation of operational strategies at the individual vehicle level (Ben-Akiva et al., 2012).

Simulations were performed using Citilabs Cube Avenue and MATSim softwares. Brief information about the operation principles of both softwares are given as below.

3.1. Citilabs Cube Avenue

In 2005 after Hurricane Rita, Houston-Galveston Area Council (Texas) was asked to prepare a tool for evacuation planning (White and Chanchareon, 2008). Cube Avenue was first used to help the local planners and decision makers during emergency evacuations in Houston- Galveston area to be prepared for incidents such as Hurricane Rita and to come up with alternate solutions for evacuations (Hardy and Wunderlich, 2007).

Avenue is an optional add-onto Cube Voyager that allows planners to study dynamic traffic assignment with mesoscopic simulation (Brown, 2011).

Vehicles are analyzed as ''packets'' and simulated as they move through the network. Packets can be either moving or waiting on a link. A vehicle may have to wait due to capacity constraints; cars leaving a link exceed its exit flow capacity or cars entering a link exceed its entrance flow capacity, or due to storage constraints; no room for a vehicle on the next link. The resulting traffic stream on each link is evaluated using aggregate macroscopic speed-flow relationships.

Table 3 Risk zones evacuation times.

| MATSim simulation results | | | | Citilabs Cube Avenue simulation
results | | | | | |---------------------------|----------------|----------------|----------------|--------------------------------------------|----------------|----------------|----------------|--| | | Risk
zone 1 | Risk
zone 2 | Risk
zone 3 | | Risk
zone 1 | Risk
zone 2 | Risk
zone 3 | | | Scenario
1 | 45 | 83 | 86 | Scenario
1 | 43 | 64 | 93 | | | Scenario
2 | 44 | 82 | 91 | Scenario
2 | 42 | 65 | 106 | | | Scenario
3 | 47 | 86 | 88 | Scenario
3 | 43 | 65 | 80 | |

As an input, Avenue requires a network file in Cube format, OD matrix, storage which is the number of vehicles that can fill the link, distance, capacity and number of lanes.

The storage of the road links are calculated as the multiplication of the number of lanes, vehicle per distance and the distance of the link.

Cube Avenue uses a generalized cost function as the disutility function in the assignment that can include any variable with its own weight. The assignment in Avenue is user equilibrium that utilizes method of successive averages method.

The centroids are automatically assigned and the trips are generated from the centroid of the origin zone and reach to the centroid of the destination zone. The simulations are performed considering the abovementioned scenarios.

3.2. MATSim

MATSim is an agent-based, activity-based, easily extendable, open-source multiagent dynamic simulation toolkit implemented in Java in a modular structure that can be modified or replaced according to various conditions to handle different scenarios (Hor ni et al., 2009). Ferber (1999) describes an agent as an autonomous entity (real or virtual) with its own abilities, resources and goals that can act on and perceive its environment network; has some internal representation of that environment; and can possibly communicate with other agents (Raney, 2005).

The initial demand of plans of activities generated for each agent and is executed in a mobility simulation. Demand is the movement of agent, which is a driver on transportation network in this case, from a location to another on a route. Plan of each agent is scored after simulation iterations, and on the basis of the score, agents adapt their plans according to the conditions and events in the simulation, such as queues in intersections, and link capacities (Bekhor et al., 2011).

The best plan for each agent (i.e., optimum demand for transportation system) is achieved as whole transportation system converges to a relaxed state after adequate iterations (200 iterations for this project) are made using initial demand.

Projecting daily plan of an agent, including mode choice and facilities (e.g., shopping center, hospital, etc.) used depending on agents characteristics (e.g., financial situation, age, car owning, etc.), is possible using MATSim, however, for an evacuation plan, following data is sufficient:

-Initial demand (i.e., OD matrix).
-Road network.

Initial demand refers to synthetic population derived from numbers and locations of evacuees to be transferred to presumed health care centers or gathering-areas according to distance. DemandGenerator module of MATSim is used to generate initial demand, which generates given zone. Zones are modified according to density of population that actually exists in given zone and road links that evacuees will possibly start their trip at a possible chemical accident moment to generate a relatively realistic scenario. MATSim assigns the trip start to closest node to coordinates of agent activity (i.e., home or work) for each agent, but not to the centroid of defined zone, in contrast to Citilabs Cube Avenue.

4. Comparison of the two simulation tools

The clearance time of three risk zones and total arrival time for three different scenarios are listed in Table 3. It can be seen that evacuation times calculated with both softwares are close, especially for the first risk zone.

Cube Avenue distributes the vehicles from the centroids of the origin zones. 10 centroid connectors for each zone are assigned. Trips are generated from the centroids and through the connectors they are assigned to the links inside the zone. Once the trips start, vehicles move towards the centroids of the destination zones.

MATSim performs in a different way. The population zones are set for a group of origin zones which are assigned for a predefined destination. For each agent, random activity coordinates are generated such as home, work, and leisure, or in this case evacuation

Fig. 3. Simulation snapshots (a) Citilabs Cube Avenue simulation snapshot and (b) MATSim simulation snapshot.

Table 4 Evacuation time averages.

| Evacuation time averages | | | | | | | | |--------------------------|-------------|-------------|-------------|--|--|--|--| | | Risk zone 1 | Risk zone 2 | Risk zone 3 | | | | | | Scenario 1 | 44 | 74 | 90 | | | | | | Scenario 2 | 43 | 74 | 99 | | | | | | Scenario 3 | 45 | 76 | 84 | | | | |

zones and hospitals-gathering areas. Departures and arrivals of the agents take place on the nodes which are closest to the activity coordinates, and in the first iteration shortest path is calculated for route choice.

Simulation results of MATSim are analyzed by via - visualization and analysis tool for MATSim by Senozon (via|Senezon, 2008).

In Fig. 3 two snapshots from the simulation scenario 3 in 03:16th min are shown.

5. Conclusion

Regions where chemical facilities are active might confront accidents in any time. Thus, evacuation plans are important in identifying the measures to be taken during emergencies. An evacuation plan needs a well-defined scenario. Once the scenario is selected, the evacuation planning zone can be defined and subsequently the evacuation demand, traffic analysis zones and OD matrices can be evaluated.

The study has been accomplished for a chemical accident in an industrial zone established nearby residential settlements in Aliaga, Izmir, Turkey.

The factory of incident, that is the ACN factory, has been identified based on the features of the chemical substances processed in Petkim and their NFPA ratings. After the destination areas, where the health care centers and available gathering-areas exist, have been designated, the origin and destination zones have been defined. Based on the census and building data, number of evacuees has been calculated and found 14,472, and then has been transformed to the number of vehicles taking the car occupancy rate as two and was found 7236.

The simulation has been performed by Cube Avenue and MAT-Sim softwares. Using two softwares allowed comparing the evacuation times. Emergency plans are important tools in saving lives and preventing losses. Thus, relying only on one simulation might cause inadequate measures. In this study, two approaches – microscopic and mesoscopic – were adopted and compared. Evacuation times of the risk zones calculated with Cube Avenue and MATSim are close, only in the second risk zone Cube Avenue gives lower values which can be a result of distributing the vehicles from 10 centroid connectors and MATSim distributes the vehicles from less number of nodes. Consequently, the averages of the results are taken as the evacuation time and are shown in Table 4.

The final results show that for each scenario the clearance time of the first and second risk zones are close. However, for the first risk zone, 43–45 min clearance times are quite long durations to reach the hospital considering the severity of the event. A corridor with a higher capacity that could be used in emergency situations between Petkim area and Aliaga State Hospital could be designed which could result in lower evacuation times. For the third risk zone, the maximum clearance time is obtained for the second scenario. In this scenario Aliaga State Hospital attracts more vehicles, and not only the health care center in zone 27, but also in zone 26 are assigned as destination zones, and in order to reach zone 26, local streets are used which have lower speed values whereas for zone 27 collectors are used. This difference causes the increase in evacuation time. For zone 26 and 27 it would be proper to place flashing lights along the main roads so that persons could follow these signs to reach health care centers. For gathering areas, the advantage of these lights could be used too.

Overall, the third scenario gives the best results with the minimum clearance time of the entire risk area among the given scenarios. This is a result of diverting 200 vehicles to a gathering area in zone 33 instead of a health care center. This makes sense for the evacuees living in the border of the second risk zone. They may clear the area without being affected by the chemical within a shorter time rather than driving to a health care center.

One significant step of the evacuation plans is defining zones. As the zone dimension and location changes, the demand will change accordingly which will affect the whole transportation system. Especially, in evacuation the network will be used in full capacity which makes the zonal definitions more important. The accessibility of the zones, the capacity of the roads connecting the zones, the traffic demand are all aspects that need to be studied in detail. Roads with low capacities and high traffic demand will result in traffic congestion and high evacuation times which is undesirable. The simulations should be performed in order to determine the location of the destination zones which would avoid the false location choice. An area could be large enough to welcome a high number of persons but if only it is accessible and gives lower arrival times among the alternatives it could be accepted as a destination zone. In the process of preparing simulation data, in this study, it was observed that small modifications in zones and network could lead to great differences in simulation results. Thus, the initial step is the most important step of simulation which is data preparation that must be done in great detail.

The number of scenarios can be increased, considering the time of the day, the impact area, etc. In the present study, the scenarios are generated for a nighttime event and all evacuees are assumed to leave the risk zones by private vehicles with a occupancy rate of two.

Evacuation plans should be prepared and persons should be informed about the plan in peace time. During the time of the emergency, evacuees should already know where to escape and this is only accomplished by a well-functioning public information system. An evacuation plan for Aliaga, Izmir did not exist thus the main goal of this study was to create a basic evacuation plan for further improvements.

References

Alsnih, R., Stopher, P.R., 2004. Review of procedures associated with devising emergency evacuation plans. Transportation Research Record: Journal of Transportation Research Board 1865, 89–97.

Barrett, B., Ran, B., Pillai, R., 2000. Developing a dynamic traffic management modeling framework for hurricane evacuation. Transportation Research Record: Journal of the Transportation Research Board 1733, 115–121.

Bekhor, S., Dobler, C., Axhausen, K.W., 2011. Integration pf activity-based and agent based models: case of Tel Aviv, Israel. Transportation Research Record: Journal of Transportation Research Board 2255, 38–47.

Ben-Akiva, M.E., Gao, S., Wei, Z., Wen, Y., 2012. A dynamic traffic assignment model for highly congested urban networks. Transportation Research Part C, 62–82.

Brown, C.M., 2011.PTP, Citilabs. Dynamic Traffic Assignment and Mesoscopic Simulation with Cube Avenue. SEFL FSUTMS User Group July 8 Meeting. Traffic Analysis Tools and Specialized Applications.

Burghout, W., 2005. Mesoscopic Simulation Models for Short-Term Prediction. Institute of Technology Linköpings Universitet.

Church, R.L., Sexton, R., 2002. Modeling Small Area Evacuation: Can Existing Transportation Infrastructure Impede Public Safety? California Department of Transportation.

Cova, T.J., Johnson, J.P., 2002. Microsimulation of neighborhood evacuations in the urban – wildland interface. Environment and Planning A 34 (12), 2211– 2229.

Duanmu, J., Chowdhury, M., Taaffe, K., Jordan, C., 2012. Buffering in evacuation management for optimal traffic demand distribution. Transportation Research Part E: Logistics and Transportation Review, 684–700.

European Commission, 2012. European Commission Environment. <http:// www.ec.europa.eu/environment/seveso/index.htm>, (accessed 15.07.12).

Federal Highway Administration, 2004. Traffic Analysis Toolbox Volume III: Guidelines for Applying Traffic Microsimulation Modeling Software. US Department of Transportation.