Discrimination of teleseismic events in Central Asia with a local network of short period stations

Abstract

The difficult problem of distinguishing underground nuclear explosions from earthquakes at teleseismie distances was approached using short period seismic data from 6 stations in South and Central Finland. The events were nuclear tests mostly from the Semipalatinsk and Lop Nor test sites and earthquakes from adjacent areas. The magnitude range of the events was from 4.1 to 6.6. The features of the two classes of events were examined by computing spectral ratio, third moment of frequency (TMF) and complexity from P wave signals. The spectral discrimination parameters were extracted from spectra computed in 5 different ways in order to obtain all possible information even from weak events. The standard FFT spectra were computed from. raw data, after noise adaption and data adaption, from correlograms and using combinations of adaption and correlation: methods. This was done to employ not only the spectral differences of the events but also the temporal variation of energy and lack of it as a function of frequency. The optimum frequeney windows for spectral ratio and TMF were defined using stacked spectra of about 10 events from both classes. No single discriminant could classify all the events. Their performance varied significanfly for different stations, but on average the spectral discriminants had slightly higher discrimination capability than complexity. The distributions of all discriminants were studied and a group separation function was formed using an optimum set of discriminants. Instead of discriminant values their relative positions in the corresponding distributions of nuelear tests and earthquakes were used as inputs to the function. A weight for each discriminant was derived from the amount of overlap in the distributions of earthquakes and nuelear tests. All 75 events in the data set were correctly classified with the method. The testing was performed with a jack knife method to create an independent test data base