CREDIT-X1local: A Reference Earthquake Dataset for Machine Learning from ChinArray Covering the Southern North-South Seismic Zone

Abstract

High-quality datasets are critical for developing advanced machine learning algorithms in seismology. Here we present one earthquake dataset built from the records of ChinArray Phase I (X1), which was deployed in southern North-South Seismic Zone during 2011-2013 with 355 portable broadband seismic stations. As the first release of ChinArray Reference Earthquake Dataset for Innovative Technique (CREDIT), CREDIT-X1local organizes comprehensive information of 105,455 local events occurred in southern North-South Seismic Zone (20°-32°N, 95°-110°E) during array observation in one single HDF5 file. The original 100 Hz sampled three component waveforms are organized by each event for stations with epicenter distance up to 1,000 km, each waveform contains record of at least 200 seconds. Two kinds of phase labels are provided. The first includes manually picked labels for 5,999 events with magnitude larger than 2.0, providing 66,507 Pg, 42,310 Sg, 12,823 Pn and 546 Sn phases. The second contains automatic labeled phases for 105,442 events with magnitude ranging from -1.6 to 7.6. These phases are picked using one RNN phase picker and qualified using corresponding travel-time curves, retaining 1,179,808 Pg, 884,281 Sg, 176,089 Pn and 22,986 Sn phases. Additionally, first-motion polarities are also attached to 31,273 Pg phases. The events and station locations are provided, so that deep learning networks for both the conventional phase-picking and phase association can be trained and validated. Benefit from the dense array and high seismicity, CREDIT-X1local dataset can also serve as a basis for advanced data-driven innovative technique development such as focal mechanism inversion and seismic tomography.

Keywords: Earthquake dataset, Machine Learning, Pg/Sg/Pn/Sn phases picking, P-wave first-motion polarity