Pattern Matching for Volcano Status Assessment: what monitoring data alone can say about Mt. Etna activity
Abstract
The importance of assessing the ongoing status of active volcanoes is crucial not only for exposures to the local population but due to possible presence of tephra also for airline traffic. Adequately monitoring of active volcanoes, hence, plays a key role for civil protection purposes. In last decades, in order to properly monitor possible threats, continuous measuring networks have been designed and deployed on most of potentially hazardous volcanos. Nevertheless, at the present, volcano real-time surveillance is basically delegated to one or more human experts in volcanology, who interpret data coming from different kind of monitoring networks using their experience and non-measurable information (e.g. information from the field) to infer the volcano status. In some cases, raw data are used in some models to obtain more clues on the ongoing activity. In the last decades, with the development of volcano monitoring networks, huge amount of data of different geophysical, geochemical and volcanological types have been collected and stored in large databases. Having such big data sets with many examples of volcanic activity allows us to study volcano monitoring from a machine learning perspective. Thus, exploiting opportunities offered by the abundance of volcano monitoring time-series data we can try to address the following questions: Are the monitored parameters sufficient to discriminate the volcano status? Is it possible to infer/distinguish the volcano status only from the multivariate patterns of measurements? Are all the kind of measurements in the pattern equally useful for status assessment? How accurate would be an automatic system of status inference based only on pattern recognition of data? Here we present preliminary results of the data analysis we performed on a set of data and activity covering the period 2011-2017 at Mount Etna (Italy). In the considered period, we had 52 events of lava fountaining and long periods of Strombolian activity. We consider different state-of-the-art techniques of pattern recognition to try to answer the above questions. Results are objectively evaluated by using a cross-validation approach.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2017
- Bibcode:
- 2017AGUFMNG33A0178C
- Keywords:
-
- 1630 Impacts of global change;
- GLOBAL CHANGE;
- 1914 Data mining;
- INFORMATICS;
- 1942 Machine learning;
- INFORMATICS;
- 3270 Time series analysis;
- MATHEMATICAL GEOPHYSICS