An approach is presented to calibrate and use regional P-S amplitude ratios to improve seismic-event characterization capabilities with regard to monitoring the Comprehensive Nuclear-Test-Ban Treaty. Data for presumed earthquakes are used to estimate distance corrections for Pn-Sn and Pn-Lg ratios in the 6-8-Hz passband for tectonic and stable-region types. The regional phase-amplitude ratios are further corrected for path variations using simple kriging. Simple kriging is derived using a Bayesian approach. A correction surface is determined for each type of amplitude ratio at each station as an optimal linear combination of existing amplituderatio data at the station, giving greater weight to calibration data nearer to the correction location. A corresponding uncertainty surface is also estimated in terms of the residual variance of the data and a calibration variance. For well-calibrated locations, the correction converges to the mean of nearby data, and the uncertainty converges to the residual variance. For locations far from calibration data, the correction surface converges to the worldwide average, with larger uncertainty. With these correction and uncertainty surfaces, corrected values of Pn/Smax(6-8 Hz) are obtained and used to define a hypothesis test that fixes the significance level with respect to misclassifying explosions. The criterion is applied to 140 explosions at known nuclear-test sites and to 4173 Reviewed Event Bulletin (REB) events above m b 3.5 (presumed to be mostly earthquakes) with regional recordings between 3Њ and 17Њ, Pn signal-to-noise ratio (SNR) Ͼ2.0, and Sn or Lg SNR Ͼ1.3. At a 0.005 significance level, none of the 140 explosions at any of the known nuclear-test sites are screened out, whereas about 78% of the REB events are screened out. Correcting regional P-S ratios for spatial variations improves the screening performance by about 25% over just correcting for distance. The screening results are fairly insensitive to estimates of parameters (correlation length, calibration variance, and residual variance) that are used, along with data, to compute the correction and uncertainty surfaces at each station.

No abstract

No abstract

We present methods to do fast online anomaly detection using scan statistics. Scan statistics have long been used to detect statistically significant bursts of events. We extend the scan statistics framework to handle many practical issues that occur in application: dealing with an unknown background rate of events, allowing for slow natural changes in background frequency, the inverse problem of finding an unusual lack of events, and setting the test parameters to maximize power. We demonstrate its use on real and synthetic data sets with comparison to other methods.

No abstract