Abstract
The Comprehensive nuclear-Test-Ban Treaty (CTBT), which was recently adopted by the UN General Assembly and signed by President Clinton, has lowered the testing yield limit to zero and raised the profile of small earthquakes and mining blasts. Because relatively small events are very common, there is a strong need for automated algorithms that can be used to screen out the events that are obviously chemical or natural and identify those few curious enough to warrant closer scrutiny. The primary objective of this article is to assess the utility of high-frequency spectral modulations for the discrimination of mining blasts from earthquakes and single explosions at near-regional distances. Mining blasts commonly yield spectral modulations that are independent of time and the recording component. This article describes an automated discriminant that looks for these delay-fire diagnostics in data recorded on one or three components by single stations, arrays, or regional networks. Distinct deployments in central Asia, Europe, and North America are used to assess the transportability of the approach. The discriminant tests give misclassification probabilities, estimated with multivariate statistics, ranging from 0.5 to 3.5%. Discrimination using time-frequency expansions does not rely on expert interpretation but is quite routine.
The article explores likely causes of the occasional discrimination outliers. Factors that can eliminate spectral modulations from a delay-fired event include attenuation, detonation anomalies (where deviations from the designed, regular, shot sequence occur), and waveform variability. Some natural events and single explosions will exhibit spectral modulations that most likely result from propagation resonance. Data from Kyrgyzstan and Nevada is used to illustrate these effects; however, inadequate ground truth information and lack of calibration explosions in most of these datasets keeps definitive conclusions, regarding why the method will sometimes fail, out of reach. These observations underscore the need to train this algorithm to most effectively deal with these processes and pair it with other, complementary, discriminants to allow accurate characterization of all small events.
