Skip to content

Hidden Markov Random Fields as scaffolding around Indigenous Knowledge at Ruapehu volcano

Melody Whitehead 1, Mark Bebbington1, Jonathan Procter1, Hēmi Whaanga2, Hollei Gabrielsen3, Murray Lark4

  • Affiliations: 1Volcanic Risk Solutions, School of Agriculture and Environment, Massey University, Palmerston North, Aotearoa-NZ; 2Te Putahi-a-Toi, Massey University, Palmerston North, Aotearoa-NZ; 3Te Papa Atawhai, Department of Conservation, Palmerston North, Aotearoa-NZ; 4Faculty of Science, University of Nottingham, Nottingham, UK

  • Presentation type: Poster

  • Presentation time: Monday 16:30 - 18:30, Room Poster Hall

  • Poster Board Number: 226

  • Programme No: 2.4.16

  • Theme 2 > Session 4

Abstract

Accurate eruption forecasts are vital for emergency management decisions. Volcanic eruption forecasting is intrinsically uncertain, with uncertainty stemming from the incomplete knowledge of system dynamics through to incomplete and temporally biased eruption records. This lack of data severely hinders business as usual forecasting methods and eruption forecasting continues to show room for improvement. Here, a conceptual solution is presented for Ruapehu volcano, coupling the long-lived experiential knowledge of Māori iwi and hapū of the region to instrumental monitoring records using flexible and adaptive statistical models. This approach avoids over smoothing methods and instead retains all characteristics and context of this multi-dimensional qualitative and quantitative information. We describe the opportunities and pitfalls that we have encountered so far during the integration of such wildly diverse data and knowledge into Hidden Markov Random Fields (HMRFs), and discuss the less esoteric issues of reliability, noise-to-signal ratios, environmental-interference, and how we might move towards causal representation learning in volcanology.