A Neurocomputing Approach for Monitoring Plinian Volcanic Eruptions Using Infrasound

• Published in: Procedia Computer Science
• Main Authors: Ham, Fredric, Iyengar, Ishwarya, Hambebo, Bereket Mathewos, Garcés, Milton A., Deaton, John E., Bellesiles, Anna, Williams, Brian
• Format: Conference Object
• Language: English
• Published: 2012
• Subjects: Iceland
• Online Access: http://hdl.handle.net/11141/2114; https://doi.org/10.1016/j.procs.2012.09.109

Plinian volcanic eruptionsInfrasoundParallel neural network classifier bankcepstral feature vectors3D ROC curve Plinian volcanic eruptions can inject a substantial amount of volcanic ash and gas into the stratosphere, which can present a severe hazard to commercial air traffic. A hazardous volcanic ash eruption was reported on April 14, 2010, and London's aviation authority issued an alert that an ash plume was moving from an eruption in Iceland towards northwestern Europe. This eruption resulted in the closure of large areas of European airspace. Large plinian volcanic eruptions radiate infrasonic signals that can be detected by a global infrasound array network. To reduce potential hazards for commercial aviation from volcanic ash, these infrasound sensor arrays have been used to detect infrasonic signals released by sustained volcanic eruptions that can inject ash into the stratosphere at aircraft's cruising altitudes, typically in the order of 10 km. A system that is capable of near, real-time eruption detection and discrimination of plinian eruptions from other natural phenomena that can produce infrasound with overlapping spectral content (0.01 to 0.1 Hz) is highly desirable to provide ash-monitoring for commercial aviation. In this initial study, cepstral features are extracted from plinian volcanic eruption and mountain associated wave infrasound signals. These feature vectors are then used to train and test a two-module neural network classifier. One module is dedicated to classifying plinian volcano eruptions, the other mountain associated waves. Using an independent validation dataset, the classifier's correct classification rate is 91.5%.