On a global scale, naturally-triggered disasters have killed about 60,000 people per year since 1900 and cost US$ 4,800bn of losses since 1980. Many of these catastrophes are the result of a complex combination of disastrous events. By contrast, natural disaster risk assessment is traditionally carried out on individual hazards in isolation. However, a global scientific interest has been building up over the last few years toward improving the quantification of risk by taking into account the disastrous combinations of hazards, also referred to as multi-hazard events.

This thesis develops and investigates a novel methodology to include the interacting elements of the natural disaster system in risk assessments.

A causal graphical method was found to be a promising framework for multi-hazard risk assessment. A pilot study developed and used such a framework to investigate road impacts following the Kaikōura 2016 earthquake, New Zealand, and validated further development of an iterative graphical approach by providing realistic results in line with the actual events. The potential impacts from the primary and secondary hazards were shown to be uneven for the different road segments. With the aim of leading this framework to a probabilistic risk assessment, the Franz Josef area was selected early on as an interesting case study. Franz Josef township, New Zealand, is located downstream of steep valleys of the Southern Alps and experiences considerable annual precipitation. We identified that the risk from landslide dam blockage and outburst flood could have a strong influence on the multi-hazard risk to the township.

Because of the complexity of the task, pre-emptive quantifiable output associated with landslide dams is often a missing link in the cascading chain of events from an earthquake. An automated tool was developed to forecast potential outburst floods from landslide dam blockages at a regional scale, which could be used for risk reduction and resource allocation. A test study of the Callery river on the West Coast of the South Island showed a strongly positively skewed distribution of the outburst discharges that could be attributed to the known fractal dimensions of mountainous landscapes and river networks. The results point to the need for a comprehensive review of the stopbank inventory in New Zealand.

This tool formed part of the work undertaken to further develop a fully probabilistic risk assessment for the Franz Josef area using a causal graphical framework. The resulting methodology is the ultimate outcome of this thesis: it allows quantification of multi-hazard impacts on a range of vulnerable assets while providing an extensive discrete data output.

This methodology provides several advantages:

i) the unifying aspect of the framework proposed, which allows flexibility to encapsulate multiple elements into the risk “spectrum” (hazard, exposure and vulnerability) with very little friction.

ii) the forward nature of the assessment, which is predictive (use of threshold values and empirical relationships) rather than explanatory (inverse model fitting data).

The application of the framework to Franz Josef township, New Zealand, demonstrates the benefits of multi-hazard assessment. It has shown that differences between single and multi- hazard approaches increase in the low-frequency/high-magnitude tail of the loss distributions where impact aggravation occurs. Because disaster mitigation decisions are often made on higher-frequency (e.g. 100-year), or quantile (e.g. 99th percentile) events, standard risk assessments would miss or, in the best case, underestimate the risk and lead to misplaced mitigation measures.