Complex systems simulation for risk assessment in flood incident management

This paper proposes a complex systems methodology for risk assessment in emergency management and response systems. The methodology is applied to a concrete case, the flood incident management system (FIM) operated by the British Environment Agency. Flood incidents are one category of natural catastrophes, disasters or emergencies. The occurrence of emergencies is determined by the operation of complex systems. The management of and response to emergencies are also determined by the operation of complex systems. Flood incident management (FIM) systems are one kind of emergency management system, so a complex systems based approach is necessary for effective flood incident management because: • flood incidents emerge from properties and behaviour of complex systems; • flood incidents and procedures for response to flood incidents are complex systems; • emergent behaviour of these systems may not be obvious from more simplified modelling – complex systems modelling allows emergent behaviour of the system to be identified; • it is essential to assess not only the probabilities and consequences of the failure of individual components, but the impact on the whole, complex system of individual component behaviour; • such an approach will help to decide how to mitigate or manage the risk and uncertainties inherent in the whole system of FIM. For FIM a practical approach to risk management from a complex systems point of view is required to model real systems. To realise such a practical approach the following recommendations are made. • Use multi-agent simulation to model, in an intuitive way, the systems involved. Multi-agent simulation is particularly suitable to simulate complex systems, as it offers a natural way to describe system components and their interactions and can be used effectively to simulate such systems to study their behaviour. An emergency response system can also be described in a very natural way as a system of interacting agents; therefore, multi-agent simulation can be used to gain insight into such a system. • Combine multi-agent simulation with Bayesian decision networks to model probabilistic decision making by agents. Bayesian decision networks are suitable for describing probability and risk aspects of an emergency response system, because they offer an intuitive way to define probabilities of different outcomes on the level of individual system components. These probabilities need not be objectively defined, but can be based on the judgement of experts and FIM staff with practical knowledge of the system. They offer a method to integrate component-level knowledge in an outcome at the whole-system level. A proposal is made for a possible implementation of such an approach for the FIM system of the Environment Agency. • Use evolutionary computation to improve FIM by optimising agent behaviour in relation to whole system behaviour. • Consolidate the above approaches in an extended Emergent Models methodology suited to emergency management and response in general and to FIM in particular.