A manifesto for Bayes Nets in Earth Sciences
Bayes Nets (BN), or Bayesian Belief Networks (BBN), are directed acyclic graphs which can be used to model multiple strands of uncertain evidence. The talk will describe a number of applications in Earth Sciences where the speaker has used BNs, mainly in the contexts of quick and cheerful hazard and risk assessments: evaluating indirect observations hinting at incipient magmatic eruptions in Montserrat and at Bardarbunga, Iceland; appraising alternative hypotheses of the cause(s) of seismic unrest and deformation at Santorini; estimating contributions of ice sheet melting to sea level rise; evaluating evidence for a supposed major active seismogenic fault close to a Japanese nuclear facility; attribution of induced seismicity in Oklahoma to individual waste disposal wells; lahar forecasting and alerts; resilience of radwaste geological barriers to climate change; retrospective analysis of the 1976 Guadeloupe volcanic crisis. These examples range from the simplest, very basic nets (by WPA) to all-singing all-dancing frameworks of great complexity and sophisticated analytical capability (due to TH). En passant, the talk should encourage the notion that - in the inevitable presence of uncertainties - BNs are a natural rational basis for combining empirical observations and expert judgements. Some available BN/BBN packages for constructing and enumerating such nets will be mentioned.
Willy Aspinall, with extensive contributions from Thea Hincks and others
School of Earth Sciences, Bristol