 |
| Titel |
Detection of hydrogeochemical seismic precursors by a statistical learning model |
| VerfasserIn |
L. Castellana, P. F. Biagi |
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| ISSN |
1561-8633
|
| Digitales Dokument |
URL |
| Erschienen |
In: Natural Hazards and Earth System Science ; 8, no. 6 ; Nr. 8, no. 6 (2008-11-10), S.1207-1216 |
| Datensatznummer |
250005891
|
| Publikation (Nr.) |
 copernicus.org/nhess-8-1207-2008.pdf |
|
|
|
|
|
| Zusammenfassung |
| The problem of detecting the occurrence of an earthquake precursor
is faced in the general framework of the statistical learning
theory. The aim of this work is both to build models able to
detect seismic precursors from time series of different
geochemical signals and to provide an estimate of number of false
positives. The model we used is k-Nearest-Neighbor classifier for
discriminating "no-disturbed signal", "seismic precursor" and
"co-post seismic precursor" in time series relative to thirteen
different hydrogeochemical parameters collected in water samples
from a natural spring in Kamchachta (Russia) peninsula. The
measurements collected are ion content
(Na, Cl, Ca, HCO3, H3BO3), parameters (pH, Q, T) and
gases (N2, CO2, CH4, O2, Ag). The classification
error is measured by Leave-K-Out-Cross-Validation procedure. Our
study shows that the most discriminative ions for detecting
seismic precursors are Cl and Na having an error rates of
15%. Moreover, the most discriminative parameters and gases are
Q and CH4 respectively, with error rate of 21%. The ions
result the most informative hydrogeochemicals for detecting
seismic precursors due to the peculiarities of the mechanisms
involved in earthquake preparation. Finally we show that the
information collected some month before the event under
analysis are necessary to improve the classification accuracy. |
| |
|
| Teil von |
|
|
|
|
|
|
|
|