Development of a Bayesian event tree for short-term eruption onset forecasting at Taupō volcano

Abstract

Taupō volcano, located within the Taupō Volcanic Zone (TVZ) in the central North Island of Aotearoa-New Zealand, is one of the world’s most active silicic caldera systems. Silicic calderas such as Taupō are capable of a broad and complex range of volcanological activity, ranging from minor unrest episodes to large destructive supereruptions. A critical tool for volcanic risk management is eruption forecasting. The Bayesian Event Tree for Eruption Forecasting (BET_EF) is one probabilistic eruption forecasting tool that can be used to produce short-term eruption forecasts for any volcano worldwide. A BET_EF model is developed for Taupō volcano, informed by geologic and historic data. Monitoring parameters for the model were obtained through a structured expert elicitation workshop with 30 of Aotearoa-New Zealand’s volcanologists and volcano monitoring scientists. The eruption probabilities output by the BET_EF model for Taupō volcano’s 17 recorded unrest episodes (between 1877 and 2019) were examined. We found time-inhomogeneity in the probabilities stemming from both the changes over time in the monitoring network around Taupō volcano and increasing level of past data (number of non-eruptive unrest episodes). We examine the former issue through the lens of the latest episodes, and the latter by re-running the episodes assuming knowledge of all 16 other episodes (calibration to 2021 data). The time variable monitoring network around Taupō volcano and parameter weights had a substantial impact on the estimated probabilities of magmatic unrest and eruption. We also note the need for improved monitoring and data processing at Taupō volcano, the existence of which would prompt updates and therefore refinements in the BET_EF model.