Comparison of a linear discrimination function and artificial neural networks approach to discriminate the seismic events in Ankara (Turkey)

In this study, natural and blast-induced ground vibrations (namely earthquakes and quarry blasts) in Ankara (Turkey) were analysed and distinguished from each other. A total of 156 digitized vertical component velocity seismograms of seismic events of 2009-2014 with Md <= 3.5 were obtained from the Lodumlu broadband station (LOD) controlled by Bo & gbreve;azi & ccedil;i University, Kandilli Observatory, and Earthquake Research Institute Regional Earthquake-Tsunami Monitoring Center (BU-KOERIRETMC). We examined the following variables: the ratio of the highest amplitude value of the S-wave and the highest amplitude value of the P-wave (Amplitude ratio) of vertical component velocity seismograms, the power ratio (Complexity), the logarithmical value of the amplitude of the S-wave of the seismogram (Log S), and total signal duration of the seismogram. It is the first time that natural and blast-induced ground vibrations were separated from each other using the Fisher's Linear Discriminate Analysis (FLDA) technique and Artificial Neural Networks (ANNs) approach together in Ankara (The capital of Turkey). Ninety-two (59%) of the 156 seismic events studied were identified as earthquakes and sixty-four (41%) of them were described as blast-induced ground vibrations. Then the obtained accuracy percentage results of three pairs of some variables were compared. The amplitude ratio versus complexity and the amplitude ratio versus total signal duration had a high classification accuracy determination coefficient for the LOD dataset (94% for the FLDA technique and 100% for ANNs approach). ANNs approach is more successful than the FLDA technique.