Statistical discrimination studies for nuclear test verification. Final report, 1 July 1991-31 December 1994

Technical Report · Sat Dec 31 04:00:00 EST 1994

OSTI ID:121835

Shumway, R H; McQuarrie, A D

Robust statistical approaches to the problem of discriminating between regional earthquakes and explosions are developed. We compare linear discriminant analysis using descriptive features like amplitude and spectral ratios with signal discrimination techniques using the original signal waveforms and spectral approximations to the log likelihood function. Robust information theoretic techniques are proposed and all methods are applied to 8 earthquakes and 8 mining explosions in Scandinavia and to an event from Novaya Zemlya of unknown origin. It is noted that signal discrimination approaches based on discrimination information and Renyi entropy perform better in the test sample than conventional methods based on spectral ratios involving the P and S phases. Two techniques for identifying the ripple-firing pattern for typical mining explosions are proposed and shown to work well on simulated data and on several Scandinavian earthquakes and explosions. We use both cepstral analysis in the frequency domain and a time domain method based on the autocorrelation and partial autocorrelation functions. The proposed approach strips off underlying smooth spectral and seasonal spectral components corresponding to the echo pattern induced by two simple ripple-fired models. For two mining explosions, a pattern is identified whereas for two earthquakes, no pattern is evident.

• Research Organization: California Univ., Davis, CA (United States). Intercollege Div. of Statistics
• OSTI ID: 121835
• Report Number(s): AD-A--293572/4/XAB; CNN: Contract F19628-91-K-0033
• Country of Publication: United States
• Language: English

Related Subjects

35 ARMS CONTROL
45 MILITARY TECHNOLOGY, WEAPONRY, AND NATIONAL DEFENSE
99 GENERAL AND MISCELLANEOUS
COMPUTERIZED SIMULATION
NUCLEAR EXPLOSION DETECTION
PROGRESS REPORT
VERIFICATION