Physics
Scientific paper
Jun 1999
adsabs.harvard.edu/cgi-bin/nph-data_query?bibcode=1999pepi..113..131f&link_type=abstract
Physics of the Earth and Planetary Interiors, Volume 113, Issue 1-4, p. 131-142.
Physics
2
Scientific paper
In monitoring of local seismicity, the occurrence of many chemical explosions poses sever practical problems of two kinds: (i) such recordings add significantly to the analyst workload and (ii) in extreme cases, pollute the seismicity data base to the extent of rendering it useless for serious scientific studies. In some countries, the local seismicity is equivalent to felt earthquakes but the problem remains since both earthquakes and explosions are and will be recorded by local stations. These events will therefore enter the network data center processing system, and thus, be subjected to further analysis. Source classification schemes are not always well suited for this kind of needed analysis at local distance ranges (not easily transportable). Besides, epicenter determinations may be less accurate in cases of few station reportings. The common denominator for failures is the modest usages of the information potential contained in seismic recordings being represented by a few time/amplitude parameters for the Pn- and Lg-phases. On the other hand, seismic waveform similarities for closely spaced earthquakes and explosions in particular are well established observationally. In this study period, we explore the possibility of using single station three-component (3C) covariance matrix traces from a priori known explosion sites (learning) for automatically recognizing subsequent explosions from the same site. To ensure adequate sampling, we used the nine different complex covariance time domain elements in combination with a suit of 12 bandpass filters equivalent to 108 observation pieces for a single event recording. We used a neural net scheme for teaching the computer to recognize new explosion recordings from a specific site through scanning of hundreds of detector segmented waveform files. No epicenter information was used in this analysis. The output was a single number between Zero and Four (log-scale) with an acceptance threshold (repeated explosion) of 1.2-analyst friendly usage. Actual tests were performed on two explosion sites of western Norway for which ground truth information was available. The 100% correct decisions were obtained between `site explosions' and hundreds of non-site events. A brute force test on the relative merits of the 12 frequency bands used gave that the 2-4 Hz and 8-12 Hz spectral parts were most informative. Experiment tied to two quarry sites for which ground truth information was lacking gave less sharp separation of site and non-site events. Since covariance elements comprise P- and S-waveform jointly, good spatial sampling is ensured even for single site 3C stations; beyond 10 km from the epicenter the scores were in the non-site category. The above approach to event discrimination is very flexible as we may combine several 3C stations with even z-component stations and besides, no analyst action in any way will be needed.
Fedorenko Yury V.
Husebye Eystein S.
Ruud Bent O.
No associations
LandOfFree
Explosion site recognition; neural net discriminator using single three-component stations does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.
If you have personal experience with Explosion site recognition; neural net discriminator using single three-component stations, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Explosion site recognition; neural net discriminator using single three-component stations will most certainly appreciate the feedback.
Profile ID: LFWR-SCP-O-1801234