Earthquake-explosion discrimination using diffusion maps

Discrimination between earthquakes and explosions is an essential component of nuclear test monitoring and it is also important for maintaining the quality of earthquake catalogues. Currently used discrimination methods provide a partial solution to the problem. In this work, wem apply advanced machine learning methods and in particular diffusion maps for modelling and discriminating between seismic signals. Diffusion maps enable us to construct a geometric representation that capture the intrinsic structure of the seismograms. The diffusion maps are applied after a pre-processing step, in which seismograms are converted to normalized sonograms. The constructed low-dimensional model is used for automatic earthquake-explosion discrimination of data that are collected in single seismic stations. We demonstrate our approach on a data set comprising seismic events from the Dead Sea area. The diffusion-based algorithm provides correct discrimination rate that is higher than 90 per cent.