Enhancing emergency response in short-notice bushfire evacuation

摘要

A bushfire, or a wildfire, is a freely burning, uncontrolled and unplanned fire in regional and rural areas. The impacts of bushfire range from destruction of properties and critical infrastructure, supply chain disruptions, to psychological damage, injuries and fatalities of people and wildlife. In the USA, for example, there are 60,000 to 80,000 wildfires burning 3 to 10 million acres of land, causing 4,000 fatalities and 20,000 injuries each year. The Fort McMurray 2016 bushfire in Canada destroyed approximately 2,400 buildings and resulted in evacuation of 80,000 people and the estimated wildfire insurance payouts are around CAD$9 billion. In Russia, the total cost of damage from bushfires in 2010 was about USD$15 billion in addition to 55,526 casualties caused by bushfire heat waves. In Australia, bushfires have claimed hundreds of lives and resulted in billions of dollars of damage. In Victoria alone over the last few decades, 300 people have lost their lives and 4,185 have suffered serious injuries. 32 per cent of all bushfire fatalities in Australian history (176 out of 552 deaths) were associated with short-notice evacuation. The more recent 2009 Black Saturday bushfires resulted in 173 deaths, displacement of more than 7,500 residents, and caused $4.5 billion dollars in financial losses. Notably, over 50 per cent of those who were evacuated on the Black Saturday were last-minute evacuees. Short-notice bushfire evacuation is a complex, dynamic and multifaceted problem. Complexity in evacuation emanates due to multi-stage process, which necessitates operational decisions and actions to be simultaneously performed. Time-sensitive decisions in bushfire evacuation therefore entail assigning and allocating evacuees to secured shelters, selecting suitable vehicles and choosing optimal yet low-risk routes. Uncertainties in time-windows, network disruptions and bushfire propagation make the evacuation problem more dynamic and multifaceted. Any operational planning failure could adversely affect the efficiency and effectiveness of disaster response and hence could increase the risk of human injuries or fatalities. Emergency services agencies therefore require a robust decision support tool that enables simultaneous processing of complex decisions to help minimise risk and cost. This thesis develops multi-objective optimisation models to enhance emergency response and operational planning during a short-notice bushfire evacuation. Four key interrelated research questions are answered as follows: What optimisation approaches can be used to maximise short-notice evacuation under a given set of bushfire scenarios?; What is the optimum allocation of shelters required to maximise spatial coverage of late evacuees in bushfire affected area?; How can the most efficient routes (i.e. safest and shortest) be determined to transfer people from assembly points to designated shelters?; How can vehicle assignment and scheduling be optimised to maximise short notice evacuation within a specified time window? Three key optimisation models are developed to compute solutions to shelter allocation, vehicle assignment and routing problems with time window constraints and disruption scenarios under conditions of uncertainty. (I) The Late Evacuation during Bushfire to Multiple Destinations with Time Windows (LEBMD-TW) is a mixed-integer multi-objective optimisation model. The -constraint method is applied as the solution approach. (II) The Capacitated Multiple Destination Vehicle Routing Problem with Time Window (CMDVRP-TW) is a novel vehicle routing problem- model integrating several VRP variants. A heuristic solution approach is developed to tackle complex vehicle routing problem. The effectiveness of proposed heuristic algorithm is evaluated by comparison with a Meta-heuristic genetic algorithm using set of various computational experiments. Finally, (III) Possibilistic Capacitated Multiple Destination Vehicle Routing Problem with Time Window (P-CMDVRP-TW) is presented as the key contribution of this thesis because of the novelty of integrating the CMDVRP-TW model with fuzzy set theory concepts. These optimisation models are pilot tested in a small numerical experiment in Lake Eildon Park in Victoria, Australia. A real case study context is then presented using the 2009 Black Saturday bushfires in Victoria. Three plausible bushfire scenarios are considered. The baseline scenario represents the propagation of actual bushfires. The minor disruption scenario incorporates shutting down of a high-capacity shelter, whilst the major disruption scenario disconnects a shelter and cut off the main arterial road from evacuation networks. The models generated shortest and safest routes to transfer late evacuees from bushfire-affected areas within the set time windows, taking into account road accessibility and available resources. The LEBMD-TW model efficiently assigned available shelters to absorb all 1,100 late evacuees transferred from assembly points in the bushfire affected areas by a fleet of five buses and twelve vans. The CMDVRP-TW model suggests the evacuation of late evacuees by seven rescue vehicles in four shelters is feasible. A decrease in the number of assigned vehicles is possible, however, it increases the risk of transporting evacuees via high risk routes. The P-CMDVRP-TW model generated optimal routes and evacuation solutions under uncertainty and hard constraints. It was possible to evacuate equal numbers of evacuees by using only six buses and four shelters under low disruption risk. The P-CMDVRP-TW model under disruption scenarios generated an evacuation plan to transfer all late evacuees with seven buses and four shelters. The computed solutions demonstrate that short-notice evacuation is manageable with advanced operational planning. The models are useful in the development of emergency plans and evacuation strategies to enhance rapid response to last-minute evacuation in a bushfire emergency. There are four key implications for short-notice evacuation planning, (1) the capacity and capability of emergency services agencies would be enhanced to identify optimal allocation of shelter to transfer evacuees under any emergency evacuation scenarios; (2) rescue vehicles required to optimise the spatial coverage would be effectively determined; (3) emergency vehicles would be instantaneously scheduled which would lead to an improved efficiency and effectiveness of emergency response; and (4) a comprehensive transit plan can be developed using mapping of routes to mitigate potential risks for each road link in the network. Considering shadow evacuation and background traffic as a key caveat to the emergency evacuation modelling could be an interesting future study as well. With appropriate model calibration and adjustment, this modelling approach could also be applied to other disasters such as flooding and cyclones, which are also widely prevalent in Australia and other countries.