With the contribution of data from an increasing number of high-quality seismic stations and networks throughout the globe, the event detection threshold is reduced to smaller and smaller magnitudes. At M4 and below, delay-fired explosions are often commonly observed along with natural seismicity. The addition of this class within an event catalog leads to several questions which much be addressed: (1) can delay-fired mining explosions be discriminated from earthquakes; (2) can the understanding of mining discriminants illuminate the explosion discrimination process in regions where there are no known nuclear explosions; and (3) are mining discriminants regionally dependent, and are there corrections to account for this dependence? A broader question is in understanding the ability to separate source and propagation path effects from regional data. Addressing this question utilizing growing regional dataset has direct application to the problems noted above but more generally to our ability to interpret regional seismic records from all types of events including earthquakes. To address these types of questions, we have assembled a database of ∼2500 mining explosions and ∼1300 earthquakes recorded at a number of regional stations, comprising a total of ∼150,000 waveforms.
The database is focused on two regions, the Western US and the Altai-Sayan region of Russia, which are both areas of prolific mining activity. As part of an extensive collaboration with the largest coal mine in Wyoming and the nation, we have detailed shot information for ∼1000 mining events, classified into six distinct blast types. We have limited information for events in the Altai-Sayan based upon contacts with the ASSE. We have applied three discriminants to data from 11 stations and one array in the WUS. The first discriminant, amplitude ratios, exploits differences between regional phases due to source type. Results are highly station centric, although the largest mining events separate from earthquakes that are <250 km from the mine; as the earthquake dataset expands spatially, discrimination performance degrades. One-dimensional path corrections provide improvement, but additional calibrations are necessary to optimize this discriminant. The second discriminant, time-frequency, capitalizes on the unique spectral signature of delay-fired mining events as a function of time. This discriminant separates the larger types of blasts with the longest source duration at all stations. Smaller blasts do not discriminate because of the shorter shot durations. The third discriminant, time-of-day, assesses the event origin time. Strictly speaking, this type of analysis may have a secondary role in the discrimination of an individual event but may be quite useful in assessing man-made seismic activity in a regional context. Mining events occur between 9 am and 6 pm, while earthquakes are randomly distributed in time. We have utilized waveform correlation techniques to better understand how factors such as mining blast type and location within the mine are manifested in the waveforms. Initial results show good correlation between blast types within two main pits; as the correlation threshold is increased, we are able to resolve spatial location within individual pits for the simplest types of mining blasts.
In the Altai-Sayan region, we calculated these same discriminants for ∼260 earthquakes and ∼850 mining events. The amplitude ratio discriminant shows significant overlap of the earthquake and mining populations. Certain events do separate, but the lack of ground-truth makes is what makes these events difficult to identify. Similar results are seen for the time-frequency discriminant. We do not know if the discriminant itself fails, or if the majority of our data points are from smaller shots that have shorter time durations. Time-of-day results are similar to the WUS in that presumed mining events fall within working hours and indicate the assessment utility of this tool. These unanswered questions illustrate the need for detailed ground-truth information. Future studies of mining discrimination, particularly where large datasets are to be acquired, should involve cooperation with mine operators in order to address ambiguities such as those identified in the Altai-Sayan study.
Although we see mixed results with the amplitude ratio discriminant, there is more success with the time-frequency and time-of-day discriminants. No discriminant individually is able to successfully act as a surrogate for a single-fired explosion. However, the three discriminants, when used in combination, can provide a means of defining a delay-fired population region that could be integrated into a model such as the Event Classification Matrix to aid in identifying events that do not fall within traditional nuclear explosion or earthquake population bounds (Anderson et al., 1997). We have begun testing this methodology by using classification trees and RDA to combine the three discriminants discussed above, which yield a statistical measure of the probability of correctly classifying mining events.