Research digital skills training 2021
Develop short-term eruption warning systems for Whakaari and other volcanoes
Dr David Dempsey, Senior Lecturer, Engineerig Science
On 9th December 2019, at 2pm, Whakaari volcano erupted unexpectedly, killing 21 tourists and guides on the island. The high cost underlines the hazards in our backyard and the difficulties facing those estimating volcanic risk. Since this tragedy, we have been working to develop short-term eruption warning systems for Whakaari and other volcanoes. Whakaari/White Island is an andesite stratovolcano in the offshore Bay of Plenty, part of a chain of volcanoes extending across the North Island to Mt Ruapehu. In spite its regular eruptions – five in the last decade – it has also been a popular tourist destination, with regular boat and helicopter tours from Whakatane. GNS Science monitor volcanic activity at the island using seismometers that continuously stream data to the mainland. We took nine years of seismic data from Whakaari and extracted the component that is particularly sensitive to the volcano state, called tremor. We then searched for patterns in the tremor that were especially unusual in the days before eruptions. We found strong 4-hour bursts of seismic energy in the hours before many of these eruptions, which we interpret as hot magmatic fluids entering the shallow groundwater. Once mixed, they cause heating and pressurisation that eventually leads to an explosive eruption.
The Real-time alert system
It would not have been possible without assistance from the Centre for eResearch (CeR). The key requirement was a continuous computing resource that could operate for months on end without interruption. CeR helped us set up a virtual machine on the Nectar Cloud to run our forecaster, and then configured a web portal so that key scientists could access the current state anywhere in the world, at any time. This work was a collaboration between the Faculty of Engineering (Dr David Dempsey and Dr Andreas Kempa-Liehr) and the School of Environment (Prof. Shane Cronin). The next step is to see if it could work at other volcanoes in New Zealand and around the world.
Recognising suspicious seismic activity through modelling
We taught a model to recognise these patterns and to raise alerts when it sees suspicious seismic activity that could indicate a future eruption. This model – trained using four eruptions and tested with a fifth – could anticipate most eruptions, and would have given 16 hours warning of the fatal 2019 eruption. We have been operating the forecasting system in a 24/7 real-time manner since February 2020. Every 10 minutes, it receives new data from the GNS seismometer and uses this to update its guess about whether an eruption is imminent. If it thinks an eruption is likely, an email alert is automatically sent to key scientists.