Exploring the potential of probabilistic Bayesian event tree tools for volcanic hazard analysis, risk assessment, and hazard and risk communication in New Zealand

Abstract

In recent years, the emergence of free, robust, user-driven Bayesian event tree tools for probabilistic volcanic hazard analysis (PVHA) has introduced new opportunities for building comprehensive hazard analyses for volcanoes worldwide. This thesis investigates the potential for Bayesian event tree tools for volcanic hazard analysis, risk assessment, and hazard and risk communication in New Zealand. A long-term PVHA for one of New Zealand’s most recently active and complex volcanoes, the Okataina Volcanic Centre (OVC), was performed by integrating past eruption records, geospatial analyses, expert elicitation data, and TEPHRA2 models into BET_VH (Bayesian Event Tree for Volcanic Hazards). This represents the first tephra hazard analysis for the OVC that assesses different possible vent areas and eruption styles, quantitatively defines the OVC’s linear vent zones (LVZs), and presents hazard information at multiple levels of uncertainty (average, 10th and 90th percentiles). Results revealed that in order to estimate the full potential distribution and magnitude of tephra hazard, all possible vent locations and both basaltic and rhyolitic Plinian eruption styles should be considered in OVC tephra hazard analyses. A new approach for developing PVHA-based quantitative risk assessments is presented through a study of OVC ashfall risk to farms in the Bay of Plenty (BOP) region. By integrating TEPHRA2 and BET_VH seasonal hazard datasets with agriculture fragility functions and seasonal vulnerability coefficients, a robust risk assessment is achieved, which quantifies potential damage (loss) over a continuum of hazard and vulnerability, at multiple levels of uncertainty, and in the context of fluctuating hazard and vulnerability environments. A risk uncertainty matrix is presented as a scheme to guide evaluation and communication of the level of uncertainty related with such robust quantitative risk assessments, based on combining different levels of uncertainty available in each of the hazard and vulnerability datasets used. Fruit farms were found to be more at risk of OVC ashfall than dairy farms, and farms east of the OVC were found to be more at risk than farms north of the OVC. Detailed assessment revealed that the volcanic ashfall risk at fruit farms is cyclic and fluctuates with time of year and harvest season, with the highest risk experienced during peak harvest season (15 October – 14 April). Fourteen interviews and a survey of 110 organisational stakeholders and scientists in New Zealand revealed that simple visual design properties can have significant influences on the communication and interpretation of hazard information from probabilistic volcanic hazard maps and hazard curves designed with BET_VH data. Such examples include colour scheme and data classification style. Based on these findings, a set of empirically-based recommendations are presented for consideration in the design of probabilistic volcanic hazard maps, the first such recommendations for volcanic hazard maps worldwide. Through these three interdisciplinary and methodological case studies based on the tool BET_VH, this thesis finds that Bayesian event tree tools for PVHA have encouraging potential to contribute to applied volcanological practices in New Zealand.