HR: 0800h
AN: V31E-0712    [Abstracts]
TI: Current State of an Intelligent System to Aid in Tephra Layer Correlation
AU: * Hanson-Hedgecock, S
EM: seh5@buffalo.edu
AF: Department of Geology University at Buffalo, 876 Natural Science Complex, Buffalo, NY 14260,
AU: Bursik, M
EM: mib@buffalo.edu
AF: Department of Geology University at Buffalo, 876 Natural Science Complex, Buffalo, NY 14260,
AU: Rogova, G
EM: rogova@rochester.rr.com
AF: Encompass Consulting, 9 Country Meadows Drive, Honeoye Falls, NY 14472,
AB: We are developing a computer based intelligent system to correlate tephra layers by using the lithologic, mineralogic, and geochemical characteristics of field samples, to aid geologists in interpreting eruption patterns of volcanic chains and fields. The intelligent system is used to define groups of tephra source vents by utilizing geochemical data, and to correlate tephra layers based on lithostratigraphic characteristics. Understanding the eruption history of a volcano from stratigraphic studies is important for forecasting future eruptive behavior and hazards. In volcanic chains and fields with a complex eruptive history and no central vent, determining the spatio- temporal eruption patterns is difficult. Sedimentologic and chemical variability, and sparse sampling often result in relatively large variances and imprecision in the dataset. Lithostratigraphic and geochemical interpretation also depends on ones' level of expertise and can be subjective. The processing of lithostratigraphic features is conducted by a hybrid classifier, composed of supervised artificial neural networks (ANNs) combined within the framework of the Dempster-Shafer theory of evidence. Since lithostratigraphic features vary with distance from source, hypothetical vent locations are determined by using expert domain knowledge and geostatistical methods. Geochemical data are processed by a suit of fuzzy k- means classifiers. Each fuzzy k-means classifier assigns observations to multiple clusters with various degrees, called membership coefficients. The assignment minimizes a function of the total distance between the centers of clusters and the individual geochemical data patterns weighed by the membership coefficients. Improved clustering results of geochemical data are achieved by the fusion of individual clustering results with an evidential combination method. Lithostratigraphic data from individual tephra beds of the North Mono eruption sequence are used to test the effectiveness of the intelligent system for tephra layer correlation. Geochemical data from tephra bedsets of the Mono and Inyo Craters, CA, are used to test the effectiveness of the intelligent system for eruption sequence correlation. The intelligent system aids correlation by showing matches and disparities between data patterns from different outcrops that may have been overlooked in initial interpretations. Initial results show that the lithostratigraphic classifier is able to accurately differentiate known layers 76% of the time. Output from the lithostratigraphic classifier can furthermore be plotted directly as isopleth maps that can aid in rapid recognition of tephra layers as well as determination of eruption characteristics, e.g. eruption volume, plume height, etc. The intelligent system produces a useful recognition result, while dealing with the uncertainty from sparse data and the imprecise description of layer characteristics.
DE: 0555 Neural networks, fuzzy logic, machine learning
DE: 8404 Volcanoclastic deposits
DE: 8428 Explosive volcanism
DE: 8455 Tephrochronology (1145)
DE: 8494 Instruments and techniques
SC: Volcanology, Geochemistry, Petrology [V]
MN: 2007 Fall Meeting