The robustness of volcanic hazard forecasts under uncertainty : a thesis presented in partial fulfilment of the requirements for the degree of Doctor of Philosophy in Earth Science at Massey University (Manawatū campus), Aotearoa New Zealand

Abstract

Forecasting volcanic hazards is crucial for mitigating the impacts of volcanic activity; however, substantial uncertainty exists in model outputs due to both inherent variability in eruptive processes and limited observational data. Computational volcanic hazard models translate physical processes into simulations, but differences in model structure, input requirements, and assumptions can strongly influence forecast outcomes. Furthermore, the effect of prior knowledge about a volcano’s eruptive history on forecast accuracy remains poorly understood.

This thesis investigates how model choice, input parameter uncertainty, and volcano-specific knowledge affect the accuracy and usefulness of volcanic hazard forecasts. A structured review and classification of existing models establishes a framework for understanding how differences in model design and computational complexity influence forecasting under uncertainty. Global sensitivity analyses using two volcanic ash transport and dispersion models, Tephra2 and Fall3D, identify which input parameters most strongly influence output variance. Case-study simulations of the 17 June 1996 eruption of Mount Ruapehu, Aotearoa New Zealand, compare forecasts generated with informed versus uninformed input parameter distributions, evaluating both mass-based and impact-based accuracy.

Results show that input parameters such as median grain size, diffusion, and plume shape consistently have the largest influence on forecast variability, while others, such as plume height, demonstrate limited individual influence but become influential through higher-order interactions with other input parameters. Case-study forecasts indicate that reliance on prior eruption data does not systematically improve forecast accuracy and can introduce biases. These findings provide guidance for prioritising input parameters, selecting suitable models, and balancing complexity with uncertainty to improve forecast accuracy and efficiency. By linking model structure, input sensitivity, and data bias, this work contributes to more accurate, actionable forecasts, supporting decision-making for volcanic hazard mitigation and emergency response.