Automatic classification of dome instabilities based on Doppler radar measurements at Merapi volcano, Indonesia: Part I

  • artificial neural network
  • Doppler radar
  • explosive volcanism
  • fuzzy mathematics
  • image classification
  • instability
  • lava dome
  • remote sensing
  • volcanology
  • Asia
  • East Java
  • Eurasia
  • Greater Sunda Islands
  • Indonesia
  • Java
  • Malay Archipelago
  • Merapi
  • Southeast Asia
  • Sunda Isles
  • Monitoring lava dome instabilities is crucial to efficiently monitor active dome building volcanoes. The Doppler radar technique provides a unique opportunity to gather information about the number of instability events occurring at the growing dome and about the dynamic processes that take place during different types of instabilities. So far, three different kinds of processes have been identified: sliding material, gravitational break-offs and explosive outbursts. In addition, Doppler radars provide rain measurements, which can be used to investigate possible correlations between rainfall and dome activity. Two radar systems have been installed at Merapi volcano in October 2001 and January 2005 to continuously monitor dome instabilities. Due to the large number of instability events that occur during times of high activity, manual processing and analysis of instability events is not practical for monitoring purposes. Therefore, an automatic classification system has been developed, which is capable of identifying different kinds of instabilities as well as rainfall. Two different kinds of classifier models have been applied: (1) neural network and (2) K-nearest-neighbour classifier model. Both classify Doppler spectra according to the underlying dynamic process, that is, rain, sliding material, gravitational break-off or explosive outburst. The classifiers are able to identify disturbances, which have no physical source, but are merely artefacts from the radar device itself. Because radar events are sequences of Doppler spectra, a rule set has been defined, which finally determines the event class. All classifiers have been trained and tested on independent data sets to estimate the classification performance. The overall classification rate is about 90 per cent. Discrimination of instabilities and non-volcanic events reaches about 98 per cent accuracy. © 2007 The Authors Journal compilation © 2007 RAS.
  • info:eu-repo/semantics/closedAccess
Forschungsinformationssystem der UHH

Interne Metadaten