부산대학교 도서관

Volcano observatory best practice recommends using probabilistic methods to forecast eruptions to account for the complex natural processes leading up to an eruption and communicating the inherent uncertainties in appropriate ways. Bayesian networks (BNs) are an artificial intelligence technology to model complex systems with uncertainties. BNs consist of a graphical presentation of the system that is being modelled and robust statistics to describe the joint probability distribution of all variables. They have been applied successfully in many domains including risk assessment to support decision-making and modelling multiple data streams for eruption forecasting and volcanic hazard and risk assessment. However, they are not routinely or widely employed in volcano observatories yet. BNs provide a flexible framework to incorporate conceptual understanding of a volcano, learn from data when available and incorporate expert elicitation in the absence of data. Here we describe a method to build a BN model to support decision-making. The method is built on the process flow of risk management by the International Organization for Standardization. We have applied the method to develop a BN model to forecast the probability of eruption for Mt Ruapehu, Aotearoa New Zealand in collaboration with the New Zealand volcano monitoring group (VMG). Since 2014, the VMG has regularly estimated the probability of volcanic eruptions at Mt Ruapehu that impact beyond the crater rim. The BN model structure was built with expert elicitation based on the conceptual understanding of Mt Ruapehu and with a focus on making use of the long eruption catalogue and the long-term monitoring data. The model parameterisation was partly done by data learning, complemented by expert elicitation. The retrospective BN model forecasts agree well with the VMG elicitations. The BN model is now implemented as a software tool to automatically calculate daily forecast updates.