Predictability of Volcano Eruption: lessons from a basaltic effusive volcano

Abstract

Volcano eruption forecast remains a challenging and controversial problem despite the fact that data from volcano monitoring significantly increased in quantity and quality during the last decades.This study uses pattern recognition techniques to quantify the predictability of the 15 Piton de la Fournaise (PdlF) eruptions in the 1988-2001 period using increase of the daily seismicity rate as a precursor. Lead time of this prediction is a few days to weeks. Using the daily seismicity rate, we formulate a simple prediction rule, use it for retrospective prediction of the 15 eruptions,and test the prediction quality with error diagrams. The best prediction performance corresponds to averaging the daily seismicity rate over 5 days and issuing a prediction alarm for 5 days. 65% of the eruptions are predicted for an alarm duration less than 20% of the time considered. Even though this result is concomitant of a large number of false alarms, it is obtained with a crude counting of daily events that are available from most volcano observatories