Using multiple data sets to populate probabilistic volcanic event trees

The key parameters one needs to forecast outcomes of volcanic unrest are hidden kilometers beneath the Earth’s surface, and volcanic systems are so complex that there will invariably be stochastic elements in the evolution of any unrest. Fortunately, there is sufficient regularity in behaviour that some, perhaps many, eruptions can be forecast with enough certainty for populations to be evacuated and kept safe. Volcanologists charged with forecasting eruptions must try to understand each volcanic system well enough that unrest can be interpreted in terms of pre-eruptive process, but must simultaneously recognize and convey uncertainties in their assessment. We have found that use of event trees helps to focus discussion, integrate data from multiple sources, reach consensus among scientists about both pre-eruptive process and uncertainties and, in some cases, to explain all of this to officials. Figure 1 shows a generic volcanic event tree from Newhall and Hoblitt (2002) that can be modified as needed for each specific volcano. This paper reviews how we and our colleagues have used such trees during a number of volcanic crises worldwide, for rapid hazard assessments in situations in which more formal expert elicitations could not be conducted. We describe how Multiple Data Sets can be used to estimate probabilities at each node and branch. We also present case histories of probability estimation during crises, how the estimates were used by public officials, and some suggestions for future improvements.