situs slot gacor cincinbet usaklub slot gacor hari ini

Open Access Mini Review

Agent-Based Simulation Models of Natural Disaster Evacuation Behaviour

John A Black*

School of Civil and Environmental Engineering, UNSW Sydney, Australia

Corresponding Author

Received Date: August 10, 2020;  Published Date: August 25, 2020


This short review, drawing on the author’s experience as a researcher into evacuation behaviour caused by typhoons and coastal inundation and tsunamis and as a peer reviewer of a government task force into flood evacuation studies, considers the role of agent-based simulation models for emergency evacuations. Transport modellers conceptualise that peoples’ behavioural responses in four distinct stages: decision on whether to evacuate or not; departure time choice (if evacuating); destination choice; and route/transport mode choice. The modelling issue is the appropriate determination of the behavioural input parameters. Calibrated models on location-specific areas provide the basis for scenario analyses as demonstrated by a case study of a small town on Shikoku Island, Japan. The applications of emergency evacuation models to disaster management authorities and their role in stakeholder communication are discussed.

Mini Review

Tsunamis, floods and hurricanes (typhoons) are frequently occurring phenomena throughout the world that cause widespread devastation yet differ in terms of the length of the warning times that authorities can issue before people have to prepare to evacuate their homes or workplaces. The shortest elapsed time between a disastrous event and a warning being triggered is for a tsunami – the time is a function of the earthquake epicenter location and the wave propagation reaching coastal areas. Flood warnings are more dependent on rainfall intensity, the length of the river and its catchment characteristics. Peak river heights may be estimated in hours or often days. Meteorological data allow cyclonic conditions to be accurately tracked, often days ahead of the event in terms of timing, landfall location and windspeeds. The science underpinning these forecasts is thoroughly understood whereas how people respond in the face of potentially disastrous emergency situations remains deficient.

This paper concerns modelling such evacuation travel behaviour [1], irrespective of the use of macro models [2] or agent-based models [3] where the focus is on the latter class of models. A roadnetwork GIS model representation of escape routes and traffic loadings is a mature area of transport planning [4] that uses nodes and links (analogous to graph theory) and link capacities, and a macro representation of travel demand behaviour, especially departure times and rates. However, the way that vehicular or pedestrian traffic on this network is represented by agent-based models is relatively new. In a defined geographical region where a natural disaster may occur the two key questions for authorities are: whether there is adequate capacity in the transport network to cope with the surge in the demand to evacuate; and whether there is adequate time evacuees for to reach a safe destination (high ground, emergency shelters or some arbitrarily defined safe haven on the transport network. Mathematical models are constructed using data, such as traffic flow counts collected during evacuations (but obviously not behavioural information), or questionnaire survey results from asking people what they did when evacuating or what they plan to do when a natural disaster does occur (stated and revealed preference surveys).

Transport modellers conceptualise that peoples’ behavioural responses [5] as: decision on whether or not to evacuate; departure time choice (if evacuating); destination choice; and route/transport mode choice. Pre-departure behaviour involves the psychological reaction of individuals and the availability and accuracy of information. Residents decide whether or not to evacuate based on physical conditions, location and knowledge of shelters, local roads and conditions, and knowledge of procedures. The time taken to make a decision and prepare is called milling time. There is a high degree of simplification in the mathematical representations of people’s departure times following warnings and evacuation orders by authorities [6]. For example, the commonly used S-shape of the evacuation departure time curves is based on numerous hurricane evacuation events as published by the US Army Corps of Engineers with typical fast, medium and slow evacuation response rates. Other models of evacuation adopt the Rayleigh distribution, the Uniform distribution, the Poisson distribution and the Weibull distribution [5]. There is only a modest amount of empirical data collected against which to compare these assumed departure time distributions in the modelling process.

Much of the emergency evacuation literature comes from the USA where authors consider hurricane evacuation by private vehicles on public roads where the important factor is lane capacity: an under-supply of road space may lead to congestion and people being trapped (with associated risk to life) by rising water levels. In general, people will want to evacuate the flooded area by the fastest possible means and follow direct routes to destinations. The majority of evacuees take their own private transport (some hauling trailers, caravans and horse boxes) whilst those without a vehicle, or older people living in aged-care facilities, require an offer of a ride or the use of public, or community, transport. In countries in the developing world, there is often no alternative but to walk. Even in a developed country, such as Japan, where walking is the preferred way of escaping tsunamis, consideration has been given to bicycle travel [7]. In the case of floods or typhoons, this travel mode is especially dangerous.

As with any model of the real world, agent-based models are designed for a particular purpose: the development of scenarios to assist in the formulation of evacuation plans. Emergency authorities need an operational model that indicates the scale and timing of peak flooding and identifies those high-risk areas - informed by weather forecasts and hydrological models. Once the agent-based model is calibrated to local circumstances its application provides quantitative output on whether people will reach safety (or not) based on a range of behavioural assumptions, assumptions about the road network (roads blocked; augmentation of road capacity, reversible highway lanes, etc) and assumptions about the location and capacity of shelters.

In the following example [8], an agent-based model was developed on the NetLogo platform, with the necessary input data (July flood of 2018) and assumptions applied to the Chunan area (1587 households in 2017), Manno town, on the island of Shikoku, Japan. This area experiences typhoon flooding and has ancient reservoirs that are vulnerable to collapsing. Figure 1 shows the results of six scenarios, including the base case, more vehicles and changes to the location of shelters, and the travel times for all vehicles to reach their safe destination. The simulations focused on night-time evacuation by car (base 150 cars following a Raleigh distribution of departure times with speeds of 30 km/h with reductions of up to 8km/h to reflect driving in appalling windy conditions) and gave prominence to older people who take more time to prepare to depart.


There are challenges for the further development of agentbased evacuation models. The application of emergency evacuation agent-based models is of value to disaster management authorities and can recreate past events with a reasonable degree of accuracy [9]. In general, economic analyses are performed on each scenario to determine the capital and maintenance costs of interventions and the social benefits (savings in lives and their monetary values) and this is an area for further research as trade-offs are part of the policy process by governments. Another potential area for research is a comparative study of the performance of different agentbased models (resource costs, computational time, reliability of model outputs). Above all, more information of people’s behaviour during stressful emergency evacuations, and how this might be represented in models, is needed.

There are many complex and interdependent behavioural issues when formulating evacuation management plans that must be understood for location-specific modelling using agent-based platforms. For example, a forecast of a more-recent approaching typhoon to Manno City, revealed that residents drove their car to secure shopping malls, walked home and were prepared, if necessary, to evacuate on foot. In other examples, concern for family members and efforts to search and confirm their location will further delay evacuation (not captured in models) thereby putting evacuees at a higher risk of death and injury. Many studies confirm that evacuation decisions and behaviours are not informed by appropriate risk assessment, education and discussion. There is a need at the local level for a detailed evacuation plan and a manual for officials to follow during an emergency. Research in developing countries suggests promotion/education strategies are needed to raise participation of residents in evacuation exercises so that residents gain a greater preparedness and understanding of behaviours they should follow during an evacuation [10]. Preparedness of communities to natural disasters is a key to mitigating more immediate impacts and the visualisation provided by agent-based models (and the speed with which results from alternative scenarios can be produced in community workshops) is helpful in raising community awareness, stimulating debate and ensuring emergency responses are a shared responsibility amongst civil and civic society.


The research program on which this review is based is led by Associate Professor Hitomi Nakanishi (University of Canberra). The author’s research has been supported by a Japan Society for the Promotion of Science Long-term Fellowship and by Urban Research and Planning (URAP) International Social Enterprise Organisation.

Conflict of Interest

The author declares no financial interest and no conflict of interest.


  1. Murray-Tuite P, Wolshon B (2013) Evacuation transportation modeling: An overview of research, development, and practice. Transportation Research Part C: Emerging Technologies 27: 25–45.
  2. US Department of Transportation (2009) Structuring modeling and simulation analyses for evacuation planning and operations. Washington DC, US Department of Transportation.
  3. Horni A, Nagel K, Axhausen KW (2016) The multi-agent transport simulation MATSim. London, Ubiquity Press.
  4. Black J (1981) Urban transport planning: Theory and practice. London, Croom Helm.
  5. Pel AJ, Bliemer MCJ, Hoogendoorn SP (2012) A review on travel behaviour modelling in dynamic traffic simulation models for evacuations. Transportation 39(1): 97–123.
  6. Urena Serulle N, Cirillo C (2017) The optimal time to evacuate: A behavioral dynamic model on Louisiana resident data. Transportation Research Part B: Methodological 106: 447–463.
  7. Takada K, Ikeda K, Aoki K, Murakami H, Koyama M (2017) Experimental study on the effectiveness of bicycle use for tsunami evacuation-case study of Horikiri district in Tahara city. Journal of the Eastern Asia Society for Transportation Studies 12: 158–166.
  8. Nakanishi H, Black J, Suenaga Y (2019) Investigating the flood evacuation behaviour of older people: A case study of a rural town in Japan. Research into Transportation Business and Management 30: 1-14.
  9. Mas E, Suppasri A, Imamura F, Koshimura S (2012) Agent-based simulation of the 2011 Great East Japan earthquake/tsunami evacuation: An integrated model of tsunami inundation and evacuation. Journal of Natural Disaster Science 34(1): 41–57.
  10. Wargadalam, R, Nakanishi H, Vidyattama Y, Black J, Suenaga Y (2019) Tsunami evacuation decisions and behaviour: A case study of Pangandaran, Indonesia. 12th AIWEST-DR Sendai 2019 “Raising Tsunami Awareness, Reducing Tsunami Risk: Sharing Tohoku-Aceh Experience, Knowledge and Culture.” Sendai, Japan, 7-8 November 2019.
Signup for Newsletter
Scroll to Top