An investigation to improve community resilience using network graph analysis of infrastructure systems

Abstract

Disasters can have devastating effects on our communities and can cause great suffering to the people who reside within them. Critical infrastructure underpins the stable functioning of these communities and the severity of disasters is often linked to failure of these systems. Traditionally, the resilience of infrastructure systems is assessed by subjecting physically based models to a range of hazard scenarios. The problem with this approach is that it can only inform us of inadequacies in the system for the chosen scenarios, potentially leaving us vulnerable to unforeseen events. This thesis investigates whether network graph theory can be used to give us increased confidence that the system will respond well in untested scenarios by developing a framework that can identify generic system characteristics and hence describe the underlying resilience of the network. The novelty in the work presented in this thesis is that it overcomes a shortcoming in existing network graph theory by including the effects of the spatial distribution of geographically dispersed systems. To consider spatial influence, a new network generation algorithm was developed which incorporated rules that connects system components based on both their spatial distribution and topology. This algorithm was used to generate proxy networks for the European, US and China air traffic networks and demonstrated that the inclusion of this spatial component was crucial to form the highly connected hub airports observed in these networks. The networks were then tested for hazard tolerance and in the case of the European air traffic network validated using data from the 2010 Eyjafjallajökull eruption. Hazard tolerance was assessed by subjecting the networks to a series of random, but spatially coherent, hazards and showed that the European air traffic network was the most vulnerable, having up to 25% more connections disrupted compared to a benchmark random network. This contradicts traditional network theory which states that these networks are resilient to random hazards. To overcome this shortcoming, two strategies were employed to improve the resilience of the network. One strategy ‘adaptively’ modified the topology (crises management) while the other ‘permanently’ modified it (hazard mitigation). When these modified networks were subjected to spatial hazards the ‘adaptive’ approach Page i produced the most resilient network, having up to 23% fewer cancelled air routes compared to the original network, for only a 5% change in airport capacity. Finally, as many infrastructure networks are flow based systems, an investigation into whether graph theory could identify vulnerabilities in these systems was conducted. The results demonstrated that by using a combination of both physically based and graph theory metrics produced the best predictive skill in identifying vulnerable nodes in the system. This research has many important implications for the owners and operators of infrastructure systems. It has demonstrated the European air traffic network to be vulnerable to spatial hazard and shown that, because many infrastructure networks possess similar properties, may therefore be equally vulnerable. It also provides a method to identify generic system vulnerabilities and strategies to reduce these. It is argued that as this research has considered generic networks it can not only increase infrastructure resilience to known threats but also to previously unidentified ones and therefore is a useful method to help protect these systems to large scale disasters and reduce the suffering for the people in the communities who rely upon them.