LEN-DB - Local earthquakes detection: a benchmark dataset of 3-component seismograms built on a global scale

<p>In this study ( <a href="http://www.sciencedirect.com/science/article/pii/S2666544120300010">The paper</a> ) we present a large dataset of 1,249,411 3-component seismograms, recorded along the vertical, north, and east components of 1487 broad-band or very broad-band receivers distributed worldwide, including 631,105 3-component seismograms generated by 304,878 local earthquakes and labeled as earthquakes (EQ), and 618,306 ones labeled as noise (AN). The choice of collecting only local earthquake-data is motivated by the fact that small-magnitude events, which generate relatively small amplitudes and are easily attenuated, are often problematic to detect but provide valuable information about earthquake processes. The labeled data are split into HDF5-Groups: <em>EQ</em> and <em>AN</em>. Each of these groups contains as many HDF5-Datasets as the number of 3-component seismograms; these are labeled in accordance to the format <em>net_sta_starttime</em>, where <em>net</em>, <em>sta</em>, and <em>starttime</em> represent the seismic network, station, and start time of the seismograms. Each HDF5-Dataset (i.e. each triplet of seismograms) has an attribute, which allows accessing the respective metadata. In addition, the HDF5-Group <em>Stations</em> allows accessing stations&rsquo; metadata through as many HDF5-Datasets (which are labeled in accordance to the format <em>net_sta)</em> as the number of receivers employed for collecting the waveforms.</p> <p>This global dataset is intended to be used for carrying out a multitude of seismological and signal processing tasks on single-station recordings, and its size particularly suits machine learning (ML) applications.. Application of ML to this dataset shows that a simple Convolutional Neural Network of 67,939 parameters allows discriminating between earthquakes and noise single-station recordings with high accuracy (93.2%), even if applied in regions not investigated by the training set. We make the dataset publicly available as a unique file in HDF5 data format, intending to provide the seismological and broader scientific community with a benchmark for time-series to be used as a testing ground in seismology and signal processing.</p>