We assess seismological evidence bearing on claims that North Korea conducted a small nuclear test on 12 May 2010 in the vicinity of known underground nuclear tests (UNTs) in 2006, 2009, 2013, and 2016. First, we use Lg‐wave cross correlation and more traditional methods to locate the 2010 event between about 4 and 10 km southwest of the 2009 test. Second, we compare the relative sizes of regional P and S waves, using stations within 400 km of the known North Korean nuclear tests, to assess the nature of the event.
We measured P/S ratios at different frequencies, at first using data from the open station MDJ in northeast China, for training sets of earthquakes and of explosions. We developed a linear discriminant function (LDF) that, in application to P/S measured at MDJ, is most effective in separating the earthquake and explosion populations. MDJ lacks usable data for the event of interest, but we obtained regional data from stations of the nearby Dongbei Broadband Seismographic Network (DBSN) for the 12 May 2010 event and for nearby UNTs conducted in 2006 and 2009. When our LDF is applied to DBSN data, and to data from stations SMT and NE3C in China, the LDF values measured from P/S ratios from known explosions are explosion‐like; but for the 12 May 2010 event, the LDF values are earthquake‐like for frequencies between 6 and 12 Hz.
Our method for characterizing earthquakes and explosions on the basis of their regional signals can be widely applied. Measurements of P/S based on the three‐component waveform data provide better discrimination power than do those based on vertical‐component data alone.
Electronic Supplement:Tutorial material on the Mahalanobis distance‐squared measure, three‐component linear discriminant function (LDF) analysis, tables of measurements of the log10P/S spectral ratios obtained from waveforms recorded at station MDJ for the two training sets and three‐component discrimination analysis, and figures of log (P/S) values measured at 8 Hz from vertical‐component waveforms at station MDJ for two training sets and probability distributions for D.