Citation

Oven VA, Cakici N. Fire Safety J. 2009; 44(1): 1-15.

Copyright

(Copyright © 2009, Elsevier Publishing)

DOI

10.1016/j.firesaf.2008.02.005

PMID

unavailable

Abstract

aFaculty of Architecture, Gebze Institute of Technology, Cayirova No 101, Kocaeli, Turkey The use of advanced computer models for the analysis of evacuation problems in buildings under fire conditions or terrorist attacks has become an increasingly important research area. Until recently, most safety considerations regarding the evacuation of a building are taken on the basis of some deterministic rules prescribed in fire codes. However, these rules and design principles may not be sufficient to explain the complex interaction between a vast numbers of variables affecting the evacuation process. Also, the characteristics of a fire can differ from building to building and occupants can demonstrate distinctly different behavioural patterns and physiological characteristics. As a result, potential weaknesses, particular to the investigated building, can go unnoticed which, in turn, may result in disastrous consequences during an emergency. The study concentrates on two issues: firstly, what methodology should be pursued to accurately model an evacuation problem and the derivation and extent of parameters needed to fully utilise the potentials of the advanced computer models, in this case, the buildingEXODUS; the second issue is an investigation of the evacuation behaviour in a high-rise office building in Istanbul. It is found that exit knowledge and the preferences of occupants can severely slow down the evacuation process. Fires closer to the ground floor increase the death toll significantly. Failure in the activation of the sprinkler system or the absence of the system altogether can have disastrous effects on the loss of life. Keywords: Evacuation models; Fire safety; BuildingEXODUS; High-rise office buildings The bombing of HSBC Bank headquarter on 20th November 2003 which led to the deaths of 11 people, was an important terrorist attack on a high-rise building in Istanbul. Following the rapid development of the service sector, demand for high-rise office buildings in Istanbul during the 1990s, has led to the completion of 2102 high-rises of over 13 storeys, ranking Istanbul as the 4th largest centre for high rises [7] and [8]. It is estimated that these buildings are populated by 10,000 office workers during office hours [9]. The Turkish fire code which was enacted in 2002, is a set of prescribed rules questioning the evacuation adequacy in terms of a simple procedure adopting occupant flow rates based on occupant density and specifying exit numbers and widths [10]. It assumes a flow rate of 40 people/min from a width of 50 cm and evacuation time from 1 m width is estimated 3 min for non-timber structures. Other deterministic rules involve the minimum width of escape routes, the maximum undivided width of emergency stairs and the travel distances with/without sprinkler systems. These assumptions and simplistic approaches have been challenged by experimental studies that suggested these rules are not being representative of the actual behaviour [11]. Ultimately, the aim of the study is to investigate the evacuation safety of high-rise office buildings in Istanbul. However, it was difficult to realise precisely the full expectations of the project owing to the unwillingness of the building owners to cooperate on the subject for the following reasons: firstly, they claim that the required research activities and data collection in the building may disrupt the office works and endanger the security of the building; secondly, they are afraid of being faced by any potential problem that may undermine their public relations; moreover, they have a prejudgement that their building is utterly safe, and it makes little sense to implement any prospective amendments as a result of any analytical study. With this background in mind, it was intended to analyse one of the most renowned skyscraper projects in Istanbul to obtain a safety viewpoint for one of the best possible examples. The twin block of 39 and 34-storey high Sabancı Towers, the 3rd tallest office building in Istanbul with a height of 140 m, was selected for investigation. The paper describes a rigorous procedure to examine the evacuation problems in depth exemplified by this building. The influence of fire breaking out at different floor levels on the evacuation behaviour is examined. The influence of the sprinkler system activated or not during a fire on the evacuation behaviour is also presented. Deterministic fire codes that have incorporated a set of strict rules to restrict the architectural design have long been in service in many countries. They can be insufficient in providing a safe environment in most cases since they cannot incorporate the new advances in the related specialist fields. Providing easy evacuation routes is the most important fire safety feature in a building. Loss of life depends on speedy evacuation as proven in many occasions. New research in evacuation behaviour has led to the development of comprehensive analytical methods capable of modelling the complex behaviour during an evacuation process [12]. Modern fire safety codes, so-called ‘performance-based codes’, make the necessary provisions to allow for proven analytical approaches to be used in validating the fire safety of buildings without having to resort to prescribed rules [13] and [14]. Performance-based codes allow the incorporation of the latest research findings for a fire safe design and avoid major defects originating from simplistic approaches and restrictive rules. In this respect, evacuation analyses based on advanced computer modelling are of great importance in obtaining a fire safe design. An extensive number of parameters have been reported to influence the evacuation efficiency. These can be categorised into four main groups: Most descriptive fire safety codes take into account only the physical characteristics of the building in evacuation, such as corridor and staircase widths and landing areas [15] and [16]. However, other physical obstacles such as furniture, barriers and other architectural intrusions into the escape routes can pose serious delays and create bottlenecks during an evacuation. Early evacuation models were not capable of modelling the architectural plan features of the building. Instead they used a simple system of nodal network which consists of nodes and cells represented the elapsed time required to move from one node to another based on variable movement times depending on the population density [17], [18] and [19]. Delays due to queuing can be observed but the exact location of the congestion cannot be seen or evaluated [17]. The approach in EVACSIM [18] uses a similar network of nodes and links that adopt a constant movement speed in the evacuation model, utilising the Markov Chain Theory. A slightly more complicated network system based on a stochastic approach utilising either the Donegan or Yoshimura indexes to model the occupant movement was employed by Notake et al. [19]. In all the above models, the occupants are modelled as a single mass; thus; individual preferences and behaviours cannot be tracked. Recently, a number of programmes making the necessary provisions for future extensions to allow for behavioural and environmental parameters were developed [20], [21] and [22]. However, these programmes should incorporate the “architectural aspects” of the evacuation process. For example, SAFE-R [20] specifies all possible paths to the final exit, assuming occupants move with variable walking speeds on the predetermined paths without turning around or passing through the same node. In contrast, programmes with greater extension possibilities and future potentials have been proposed [21] and [22]. SGEM [21] allows evacuation analyses of the architectural drawings that are converted into a network of nodes to calculate traverse time between nodes. The mass movement of the occupants between nodes is undertaken by using a grid of cells each occupied by a single individual who acts according to stochastic conditions. The speed of movement is based on a lattice-gas model; however, it is not clearly stated how mass movement (lattice-gas model) is related to the individual movement (variable individual speeds) in the grid structure. Behavioural and environmental attributes are not incorporated but an adaptive algorithm for future extensions is discussed in great detail. Lizhong et al. [22] presented a simplistic model with a strong emphasis on the geometrical representation of the architectural plan as a grid of cells. It also allows for individual modelling of the occupants, by means of employing different walking speeds and an innovative familiarisation and human intelligence algorithm. It assigns a maximum danger grade for all unwalked cells and then reducing the danger grade for individuals as they walked randomly through particular cells. Thus, different occupants can have different perceptions of the cells and they move according to the availability of neighbouring cells to reach a predetermined safe exit. However, people–people and people–environment interactions are also seen to be incomplete in this model. More advanced computer models, however, concentrate on complex individual movements. This can only be possible if the plan of a building is meshed by a network of individual cells or represented as a continuous space. The occupant can move from one cell to another and a single occupant can occupy only one cell at a time. The greatest advantage of these programmes such as, buildingEXODUS [23], SIMULEX [24] and CRISP [25], is to allow for the movement characteristics by taking into account the furniture layout and location of walls or compartments. In all of these models, highly variable movement speeds for occupants can be incorporated for more realistic predictions. BuildingEXODUS is especially capable to slow down the speed of the occupants faced by a dense smoke to a crawling speed, or simulate a staggered walk through a path of cells along the wall boundary. Behavioural differences of the occupants affect the evacuation time in two distinct stages: (1) pre-movement time and (2) movement time [26]. The pre-movement time is further divided in two segments: The recognition time for the threat and the response time which is defined as the elapsed time between the cognition of the threat and egress. The planning of the escape strategy is also decided upon during the pre-movement time, which is then implemented during egress. Planning the escape route basically defines the escape path an individual is to follow and his/her alternative moves depending on the exit knowledge [12], [25] and [27]. It is well known that during the pre-movement time all sorts of activities (from warning others, to dressing or to extinguishing the fire, etc.) can take place [28] and [29]. Currently only computer models such as, buildingEXODUS [23], SIMULEX [24] and CRISP [25] are capable of modelling most of the above variables. CRISP can evaluate if the occupant is awake or asleep in considering the cognition time [30] and [31]. In SIMULEX and buildingEXODUS different pre-movement times can be assigned to individuals. In buildingEXODUS the time lost during compulsory tasks before egress can be specified. In CRISP, however, the pre-movement time is not user controlled but automatically defined for different types of compulsory activities before egress. Rule-based evacuation depending on the different egress planning strategies is only possible with SIMULEX and buildingEXODUS. Unlike CRISP which is reported to incorporate only shortest egress routes [25], SIMULEX and buildingEXODUS can direct the occupants to specific exits or to the exits of their personal knowledge [23], [24] and [32]. The buildingEXODUS [32] is superior to SIMULEX, in that, it is capable of modelling the patience levels of the occupants in queues, redirecting them to the nearest alternative exit or a second exit of their knowledge. Moreover, it considers occupants with stronger drives to occupy a cell during an egress, if equal conditions occur for other occupants with lesser drives, qualifying to occupy the same cell. The redirection behaviour of the occupants or the perseverance in pressing through the route in the prospect of facing a smoke-filled area is another behavioural characteristic that can only be simulated in buildingEXODUS [32]. Physical fitness of the occupant, gender, age, height and weight affect movement speed as well as his/her ability to tolerate to narcotic gases and heat. Bryan [29] reported that women vacated the area under threat faster than men. The number of women moving into a smoke-filled environment is 20% less than men. Men attempt to intervene with threat while women warn or assist others losing less time than men. The most frequent behavioural responses to fire was found fighting or containing in the fire, calling fire brigade and getting dressed [29]. The agility of the occupants to jump over small obstacles like furniture can also affect the evacuation time. Both SIMULEX and buildingEXODUS are capable of modelling influence of gender and physical fitness on walking speeds. However, the possibility of redirection movement after facing a smoke-filled environment and the jump-over ability with different gender, age and physical fitness can only be modelled by buildingEXODUS[32]. User-specified thresholds for tolerances to heat convection and heat radiation can be set to reflect the physiological characteristics of the occupants. It is well known that the speed of the evacuation process is severely hampered by the smoke exposure of the occupants [33]. A wide range of toxic gases can be generated during a fire depending on the burning contents [34]. However, there is insufficient information about the influence of these gases on mobility parameters [35]. Currently, only the influence of most commonly known narcotic gases (CO, CO2 and HCN) and deprivation of oxygen can be tackled in evacuation analyses [32]. This is achieved by gradually reducing the walking speed on the basis of intake dose for narcotic gases and deprivation rate for oxygen depletion. The thermal conditions and their effects on the mobility are also evaluated by comparing the predicted compartmental values with the user-specified thresholds. Along with buildingEXODUS and CRISP that are discussed above, ALLSAFE, EGRESS 2002, EXITT, E-SCAPE, EvacSim are some of the models that are reported to be capable of taking into account the conditions of the fire environment in the analysis of building evacuation [21], [28] and [36]. From the aforementioned discussion it is apparent that a few computer models such as buildingEXODUS and SIMULEX can provide the necessary means for a realistic evacuation analysis. The other computer programmes mentioned are only suitable to get an insight into a basic egress analysis. Analytical studies using these advanced software as a tool to develop an understanding of the evacuation efficiency and any potential defects of the building were presented in various references [23], [24], [37] and [38]. A similar approach was also adopted in the present paper, but an effort has been made to develop a more robust methodology of collecting data and representing the occupant and environmental characteristics in fire in the computer simulation. The evacuation analyses presented herein have been carried out using the well established and widely accepted computer programme buildingEXODUS [32]. The algorithm utilises the rule-based concept that specifies a series of conditional rules for each of the six aspects known to affect the evacuation behaviour: These concepts utilise a set of predetermined rules that are followed to reach to a decision about the movement possibilities of the occupant at different time increments employing stochastic techniques. During the decision-making process, an extensive amount of interaction takes place between different modules of the programme. Occupant-to-occupant, occupant-to-environmental hazards and occupant-to-building geometry interactions can be handled. The office building selected for analysis, the Sabancı Cooperation Headquarters, consists of two towers of 34 and 39 floors. It is the 3rd tallest office building in Istanbul with a height of 140 m and it was completed in 1993. The 34-storey tower was analysed. The original architectural plan drawings and the office furniture layout represented by adapting square cells size 0.5×0.5 with link (arc) lengths of 0.5 m, as in buildingEXODUS, to investigate the building for each of its 34 floors. The original typical floor plan of the building and its meshing are shown in Fig. 1a and b, respectively. The time increment for each decision and corresponding movement is set to the 1/12 of a second [32]. Currently, 1226 office workers are employed in the analysed tower. Initially, it was planned to map the characteristics of each building user individually and model each of them separately in the analyses. For safety and confidentiality reasons, the profile of office workers have not been revealed and permission to conduct a survey study on each individual by the research team was denied on the grounds of disrupting work and interrupting the complex security system. As a result, the study is limited to the data provided by the administration of the building after long persuasive discussions. Nevertheless, although indicative, it is believed that the study provides invaluable information on how to model an evacuation problem as accurately as possible and what it takes to obtain accurate data and/or to come up with some reasonable assumptions. The typical characteristics of the genders are determined from the survey administrated to a random sample of 27 office workers located in the 3rd and 4th floors of the building. The following physiological and behavioural parameters are determined from the survey: Based on the scale in buildingEXODUS [32], the female and male percentages for two distinct age groups, that is 17–29 and 30–50, are given in Table 1. The predicted total numbers of female and male workers corresponding to age groups mentioned above are extrapolated from the sample percentages and then equally distributed to each floor. Accordingly, it is estimated that, of the 1226 office workers, 44.4% are females of which 11.1% are at the age of 17–29, and 33.3% are at the age of 30–50. The estimated male population is 55.6% of total workers, of which 7.4% are at the age of 17–29, and 48.2% are at the age of 30–50. Different drive and patience values for females and males are specified in the model. Individuals with assertive behaviour (strong drive) can force their way to the exit more decisively than the less assertive individuals. A user-specified drive value from a scale (1–15) can be assigned to override the probabilistic movement procedure to a cell when more than one occupant demands to move to the same cell. A Rathus Assertiveness Inventory Test [39] which consists of 30 questions to express his/her feelings was conducted on each building user in the random sample to evaluate assertive and aggressive behaviour of person. Standard ratings from tests are determined for each gender group. Rathus values are determined from the following procedure: Student's t-test for a confidence level of 95% is conducted on the calculated Rathus values for each gender and the mean interval is specified (Fig. 2). The values outside the mean interval are left out and the arithmetic mean of the remaining data is taken as the Rathus value for the whole body of the gender. The calculated average Rathus values for females and males are 53.7 and 29.17 respectively. The average Rathus ratings are then converted into equivalent drive values of the scale in buildingEXODUS [32]. Similarly, a user-specified patience value to control the waiting time of an individual before redirecting to a different route in a congested area can be specified using a scale (1–1000 s). As there is not a standard behavioural test to quantify the patience of individuals, it was decided to ask the building users in the survey ‘How long would they be prepared to wait should there be an emergency and they had to wait in a queue?’ Each occupant was asked to choose one of the five different time intervals, based on the scale in buildingEXODUS, (0–3, 3–6, 6–9, 9–12, 12–15 and >15 min) as his/her choice [32]. A similar procedure as with Rathus ratings was also adopted to specify the average waiting times. The calculated patience times for females and males are 200 and 180 s, respectively (Fig. 3). The agility of the occupant, specified by a value between 0 and 7, determines the ease of his/her movement over small obstacles (furniture) in the buildingEXODUS. The movement speeds of individuals are automatically readjusted by a mobility factor (0–1) depending on the environmental and physical conditions (toxicity, optical density of smoke, obstacles) of the cell, link and also, the age and gender of the moving occupant. The agility value is assigned by assuming that agility is a function of the body mass ratio (the mass (kg)-to-height (m) squared index) [40]. The calculated mean values for weight, at 95% confidence level, are 61.5 for females and 81 kg for males while mean values for height are found to be 1.66 and 1.74 m, respectively. The corresponding body mass ratios are evaluated as 22.13 for females and 26.75 for males, which are then converted into equivalent agility values as required by the programme, by linear extrapolation (Table 3). The pre-movement times of the occupants significantly affect the evacuation behaviour and the congestion has been created in junctions. Pre-movement times can differ from person to person, and normally should be specified individually in the light of realistically simulated conditions. There is not a widely accepted procedure to specify pre-movement times, but evacuation drills targeting the examined occupants and building can provide valuable data [11], [17], [24] and [37]. In the study, the video recording of a real evacuation drill belonging to a single floor of the examined building where was occupied with 24 workers, has been used to determine the pre-movement times of the female and male occupants and then these times are generalised over the whole population of females and males. The pre-movement time in buildingEXODUS is, in actual terms, the pre-movement time (recognition+response time). The time counter from recordings is used to specify the time elapsed for each occupant to start vacating his/her working area. It is observed that after the audio announce type of alarm was activated. A total of 22% of the workers reacted in 20 s, while the rest continued to work until the power-cut, 95 s into the fire announcement. The number of observed female and male workers with either 20 or 95 s pre-movement times was equal and, therefore, both gender's pre-movement times are assumed to be the same and taken as 20 s for the 22% of the population, and 95 s for the 78% of the population (Table 3). The exit knowledge of the occupant and the number of alternative routes are particularly important in planning the escape route and also redirecting to a different route when the initially planned route is overcrowded. The exit preferences of occupants is determined from the survey study by asking the participant to identify the exit paths he/she would prefer, should there be a need to evacuate the building during an emergency. In the analyses the most preferred staircases leading to the exit by each gender group are then assigned as the target exit of the total population corresponding to each gender. The exit knowledge is also indirectly tested by asking each participant to state the number of staircases they know leading to the exit. The 85% of the occupants know exit door 2, however only 78% and 70% of the occupants know exit doors 1 and 3, respectively. The exit preferences of the occupants are also given in Table 2. In the analyses, only preferred alternative routes matching with declared number of staircases are considered as alternative routes to be used, when waiting times exceeds patience times. Other users defined variables which are used in the analysis, are summarised in Table 3. The environmental characteristics of the simulated fire cannot be determined by buildingEXODUS. Thus, Consolidated Fire and Smoke Transport Model (CFAST) [41] was utilised to predict the heat release rate, temperature, oxygen depletion, CO and CO2 concentrations. CFAST is based on enthalpy relations in the fire compartment between two distinct zones, namely the upper hot layer and the cooler bottom layer which shrinks as the fire progresses. The enthalpy is also considered between the fire compartment and other connected compartments, as long as vents are provided between compartments in the form of openings. The conservation equations for energy mass, momentum and ideal gas flow are solved to determine heat and temperature properties, as well as, toxic fire products. The growth of fire is modelled by the power law using an appropriate constant to simulate slow, medium, fast, ultra-fast growing fires. A typical medium growth fire with a peak heat release rate of 1020 kW was employed (Fig. 4) in the study. The normal floor plan was modelled as five connected compartments, shown in Fig. 1a, with exact volumetric dimensions, furniture contents and lining materials. Inspections in the building revealed that carpets were used in the floor and half of the walls were lined with decorative timber, while the other half was plain plaster. The suspended ceiling was made from mineral wool tiles. The doors were assumed to be open during the fire, while all windows are closed until flashover at which time windows break and air entrainment is allowed into the fire compartment. CFAST is capable of modelling the cooling effect of a sprinkler system on the basis of its sprinkler head locations, activation temperature, pre-movement time index and spray density. Inspections in the building showed that the spacing distances between sprinkler heads were 9 m. The influence of the sprinkler is then solely to reduce the heat release rate of the fire. The present study investigates the influence of the sprinkler system on the evacuation efficiency. Therefore, CFAST and buildingEXODUS analyses are executed using two different scenarios, with and without the influence of the sprinkler system. The results of CFAST analyses, with and without the sprinkler system, are presented respectively in Fig. 5 and Fig. 6, in terms of temperature–time, oxygen–time, CO and CO2 concentration–time plots for each of the five compartments. The heat release rate-time plot for the fire compartment is also provided for the investigated scenarios. The environmental properties of fire are specified in the ‘Hazard Module’ of the buildingEXODUS. The variation of environmental characteristics with time can be approximated using either linear or polynomial mathematical expressions. In the study, it is opted to use discontinuous linear functions to represent the predicted curves from CFAST analyses, on the reason of simplicity. The linear mathematical relation between discontinuity points is obtained by linear regression. As can be seen from Fig. 5 and Fig. 6, the temperature–time plots are predicted for both, top and bottom layers. The chosen programme allows the user to assign two sets of environmental property–time relations for cells, one applicable to bottom and one applicable to upper layer so that when crawling speeds are considered, more realistic bottom layer environmental properties can be utilised. However, for other predicted properties the difference was negligible and a single relation was adopted in the analyses. In buildingEXODUS, the extinction coefficient of smoke is required for the smoke behaviour simulation of the occupants. The extinction coefficient is directly proportional to smoke concentration, which may be assumed to be proportional to the mass concentration of carbon [42]. The increase in extinction coefficient, ‘κ’, may, therefore, be predicted from the mass concentration of carbon, ‘C’, [43] (Eq. (1)). The mass concentration of carbon is obtained using the predicted CO2 fraction, ‘CO2%’, yields from CFAST. The procedure in calculating carbon mass concentration in Eq. (1) is given by the following equations: The study concentrates on two parameters of the evacuation efficiency: (1) the influence of the fire floor and (2) the influence of the sprinkler system. It has been reported by the administration that a fire due to an electric fault broke out in the past but there was early intervention and consequently no harm to any of the occupants or damage to the building. A similar incident also occurred recently from electric circuit. The occupants had to be evacuated and the incident caused great panic among the office workers [44]. On the basis of these two incidents, it may be plausible that an electric fault in the switch-board could be the cause of a prospective fire. The switch-board is located in the office work area, near the doors leading to the centre core (zone two in Fig. 1a) incorporating the lifts, services and toilets. The coordinates of the switch-board are defined as the location of the fire in CFAST analyses. The fire scenarios in the analyses involve the investigation of fires on three different floors, namely the 6th, 16th and 26th floor fires. The investigated building was divided into three parts and then central floor of the each part was selected to examine influence of fire breaking out on different floors on the evacuation behaviour. Each analysis is carried out by assuming that the sprinkler system is active or not active. As a result, six analyses are executed with each repeated at least twice to optimise the results due to the discrepancies of the probabilistic solution algorithm. For comparison purposes, a nominative case evacuation analysis, without the harmful environmental conditions of the fire, is conducted to determine the threshold time for the fastest possible evacuation. The result after simulation of the nominative case evacuation by using buildingEXODUS is given in Fig. 7. It is estimated that the evacuation of 1226 workers takes about 32 min. Results from the analysis of three different floor fires, assuming that the sprinkler system functions, are presented in Fig. 8. It can be seen that almost identical results are observed for all three floor fires. All evacuation times are identical to that of nominative case evacuation. The results of these analyses are not surprising since CFAST simulations used in the buildingEXODUS analyses reveal negligible heat release rates, harmful fire products or/and environmental conditions as seen from Fig. 5. The only noticeable outcome is that, in all three scenarios, the occupants attempted to use the exit route through the floor exit door 2 (Fig. 1). Approximately 994 workers used the floor exit door 2, while only 228 and 4 occupants used the floor exit doors 1 and 3, respectively. The congestion in the staircase led by the exit door 2 is seen to be rather heavy which severely hampers the evacuation efficiency. The sprinkler system may not work owing to sabotage or during a terrorist attack. It is also possible that it can malfunction as a result of inadequate maintenance when it is most needed during a fire. To investigate the influence of this possibility on evacuation, the studies are repeated using fire characteristics from CFAST analyses shown in Fig. 5 and Fig. 6. The figures indicate very severe environmental conditions. The sudden increase in temperatures and oxygen entrainment into the compartment coincide with the flashover 175 s after the fire started. The initial drop and later the subsequent sudden built-up of the toxicant levels are the result of the post-flashover fire characteristics. Even though fire breaks out in compartment 1, in all other compartments the spread of fire temperatures and toxicants are very severe. Particularly worrying is to observe that both fire exit landings are affected by toxicants, smoke and high temperatures. It is also worthwhile mentioning that the staircases are not pressurised to keep the smoke and toxicants out. The results of the analyses without the sprinkler system are presented in Fig. 9, Fig. 10 and Fig. 11. The evacuation of the 26th floor fire shows that 1033–1041 workers manage to vacate the building and 185–193 workers lose their lives (Fig. 9a and b). The evacuation takes 26–28 m. The majority of building users (approximately 81%) recognise the exit door 2 as the primary route to safety. Between 802–810 and 227 people reach safety using the floor exit door 2 and 1, respectively (Fig. 9c and d). Although both exit landings are very similar in toxicant levels, deaths in the landing of exit door 2 are estimated to be around 166–184 people, whilst no casualties are expected among escapees using the exit door 1. This is attributed to the exposure of higher doses of toxicants in the landing of door 2, due to congestion. On the 16th floor fire, the number of casualties rises to an even higher level (Fig. 10). The death toll is predicted to be between 478 and 517 people. Only 709–748 people make it to the safety in between 18 and 25 m (Fig. 10a and b). This is partly because more evacuees have to pass the hazardous landing to reach safety, and partly because they are exposed to higher doses of toxicants due to an increased level of congestion slowing down the evacuation speed, as the crowds from lower floors join the main stream, and the delayed crossing time allows further build-up of hazardous conditions. This behaviour is particularly apparent in the stairwell led by floor exit door 2. The percentages of people preferring exit doors 2 and 1 remain unchanged, as with the 26th floor fire scenario. However, this time, it is estimated that 512–522 and 192–222 people directing to exit doors 2 and 1, respectively, make it to safety (Fig. 10c and d)). Of the users choosing exit door 2, 48–49% are incapacitated, while the percentage of casualties among people using exit door 1 is around 3–16%. The estimated life losses with the 6th floor fire scenario are even worse. According to simulations only 377–476 people manage to evacuate the building, in approximately 15–25 m. Examining Fig. 11a and b it is seen that the sudden increase in the number of overwhelmed occupants coincides with the completion of the evacuation of the survivors. This may support the assertion that most escapees are users of the floors below 6th floor or just a few floors above this level. Similar percentages of people use exits 2 and 1 (81% and 19%, respectively), as with previous scenarios. In parallel with earlier findings, most of the overwhelmed occupants are among those who choose the exit door 2. However, in this scenario only about 186–295 people evacuate safely from the exit 2 and 134–186 people from the exit 1 (Fig. 11c and d). Life losses amount to 66% of the total number of building occupants. Around 696–804 people (71–81% of those using the exit 2) are estimated to perish in the stairwell associated with exit door 2, but only 42–94 (18–41% of those using the exit 1) are estimated to be incapacitated in the stairwell entered by exit door 1. Clearly, the severity of life losses due to most occupants preferring the exit door 2 increases with fires closer to the ground floor. The movement into a smoke-filled environment was observed, as more evacuees accumulated in front of this area and caused further congestion. The possibility of this type of movement is based upon the experimental data provided by Bryan [29]. It was also seen that those who attempted to move into the smoke-filled area were immediately incapacitated as they were exposed to a very high level of toxicant doses. The computer programme buildingEXODUS proves to be a very effective tool in analysing the evacuation behaviour of the complicated structures. However, it requires an elaborate set of data regarding widespread specialist fields. It is fair to say that the algorithm of the programme is well ahead of the research data it requires for realistic simulations. Particularly data about the behavioural characteristics of people and the influence of their physiological characteristics under fire conditions are very scarce and not reliable. For an accurate simulation, the required data should be obtained using the characteristics of the real occupants and realistic simulations of the fire conditions in the investigated building. The results of the study highlight the potential problems that may be encountered in a typical high-rise office building. Keeping the sprinkler system fully operational during a fire is a vital requisite. Fatalities can increase considerably when the system malfunctions or sabotaged during a terrorist attack. The safety of the building cannot be entirely dependent on the operation of a single safety device but should be supported by other additional passive fire measures to alleviate the situation under extraordinary conditions. Pressurisation systems in all stairwells should be installed to keep smoke and fire products at bay, but they have to be also supported by protected lobbies located in front of the stairwells. Both measures are not existent in the analysed building, and in many more in Istanbul, including the bombed HSBC bank headquarters which is known to have no such safety measures. The fire load caused by the decorative timber, lining nearly half of the height of the working offices, is a major planning error in a fire safe design. Similarly, placing the switch-board in the office area and using furniture made of flammable materials are critical decisions that violate the very basic rules of passive fire safety. The use of non-fire-rating doors (mainly wooden) leading to the service core between the two working spaces easily cause smoke and heat transition to other parts of the building, even though the fire breaks out in one of the working spaces. Another passive rule of using fire proof compartmentation in the building is not implemented either. The floor level of the fire is an important determinant in controlling the number of casualties. As the floor level of fire is varied from 26 to 16 and to 6, the percentages of incapacitated occupants corresponding to these fires rise as; 15.09–15.74%, 38.9–42.16% and 75.93–84.82% of the total number of occupants. The polynomial increase in death rate with decrease in fire floor may result from most occupants attempting to use the same exit, namely the stairwell led by the floor exit door 2. This outcome highlights the fact that not only the exit knowledge of the occupants is required for a safe evacuation but also the planning of the evacuation is as important. In this particular case, the fire safety officer should train the building users so that they can be allocated to each available exit in equal numbers. The study clearly shows that an unbalanced use of the exit routes may lead to a great number of fatalities by delaying the evacuation time. Careful examination of the building users and accurate computer modelling may reveal such weaknesses well in advance of any emergency and may prompt purposeful training. It is also striking that none of the occupants considers using the main staircase led by the floor exit door 3. This is also a matter that can be corrected by a proper emergency training. Perhaps another important issue is the slow pre-movement times of the occupants, mounting to a maximum of 95 s. Clearly, the exposure to toxicants of the occupants choosing the floor exit door 2 would have been a lot less if quicker responses had been received. Under current conditions, the inadequate emergency training of the occupants is predicted to claim many lives. Terrorist attacks by light aircraft can cause fires in upper floors, with less severe consequences in the evacuation of building users than a fire caused by bombing of the lower floors. Medium high-rise buildings with no refuge floors are likely to suffer more casualties if they are attacked from their lower floors. In contrast, very tall skyscrapers may be more vulnerable to attacks from higher floors since people are instructed to take refuge in fire rated refuge floors without evacuating the building. This measure may be plausible for most fire incidents, but not where the danger of structural collapse is in prospect. If structural collapse is of low probability; however, the building users, except those near the affected floors, may not be evacuated for lower floor fires/attacks, in order to keep the number of casualties low. The analytical results of this study support this idea since most occupants are predicted to be trapped at the fire affected floor. Building management may train its fire officers to quickly evaluate the situation and make a decision to fully evacuate the building or not, for lower floor fires. Planning strategies for different floor fires should therefore be made in the light of studies presented herein. This outcome, perhaps, better emphasises the importance of personal training based on a specific set of rules validated by scientific evidence and not on some generalised deterministic set of rules. Gebze Institute of Technology Research Fund is gratefully acknowledged for supporting this study. Project No. 02-A-02-01-04. We are also thankful for the help provided by the administration of Sabancı Center. We also sincerely appreciate constructive comments of two anonymous for contribution to the making of this paper.