Search

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, Comment: 4 pages, 4 figures

This paper introduces a multiscale analysis based on optimal piecewise linear approximations of time series. An optimality criterion is formulated and on its base a computationally effective algorithm is constructed for decomposition of a time series into a hierarchy of trends (local linear approximations) at different scales. The top of the hierarchy is the global linear approximation over the whole observational interval, the bottom is the original time series. Each internal level of the hierarchy corresponds to a piecewise linear approximation of analyzed series. Possible applications of the introduced Multiscale Trend Analysis (MTA) go far beyond the linear interpolation problem: This paper develops and illustrates methods of self-affine, hierarchical, and correlation analyses of time series., Comment: 37 pages, 19 figures