Earthquake Early Warning Dataset

This dataset is composed of GPS stations (1 file) and seismometers (1 file) multivariate time series<b> </b>data<b> </b>associated with three types of events (normal activity / medium earthquakes / large earthquakes).<b> <br></b><br><b>Files Format</b>: plain text<br><b>Files Creation Date</b>: 02/09/2019<br><b>Data Type</b>: multivariate time series<br><br><b>Number of Dimensions</b>: 3 (east-west, north-south and up-down)<br><br><b>Time Series Length</b>: 60 (one data point per second)<br><b>Period</b>: 2001-2018<br><br><b>Geographic Location</b>: -62 ≤ latitude ≤ 73, -179 ≤ longitude ≤ 25<br><b>Data Collection<br></b><b> </b>- Large Earthquakes: GPS stations and seismometers data are obtained from the archive [1]. This archive includes 29 large eathquakes. In order to be able to adopt a homogeneous labeling method, dataset is limited to the data available from the American Incorporated Research Institutions for Seismology - IRIS (14 large earthquakes remaining over 29). <br> &gt; <i>GPS stations (14 events)</i>: High Rate Global Navigation Satellite System (HR-GNSS) displacement data (1-5Hz). Raw observations have been processed with a precise point positioning algorithm [2] to obtain displacement time series in geodetic coordinates. Undifferenced GNSS ambiguities were fixed to integers to improve accuracy, especially over the low frequency band of tens of seconds [3]. Then, coordinates have been rotated to a local east-west, north-south and up-down system. <br>&gt; <i>Seismometers (14 events)</i>: seismometers strong motion data (1-10Hz). Channel files are specifying the units, sample rates, and gains of each channel. <br><br>- Normal Activity / Medium Earthquakes: &gt; <i>GPS stations (255 events</i><i><i>: 255 normal activity</i>)</i>: High Rate Global Navigation Satellite System (HR-GNSS) normal activity displacement data (1Hz). GPS data outside of large earthquake periods can be considered as normal activity (noise). Data is downloaded from [4], an archive maintained by the University of Oregon which stores a representative extract of GPS noise. It is an archive of real-time three component positions for 240 stations in the western U.S. from California to Alaska and spanning from October 2018 to the present day. The raw GPS data (observations of phase and range to visible satellites) are processed with an algorithm called FastLane [5] and converted to 1 Hz sampled positions. Normal activity MTS are randomly sampled from the archive to match the number of seismometers events and to keep a ratio above 30% between the number of large earthquakes MTS and normal activity in order not to encounter a class imbalance issue.<br>&gt; <i>Seismometers (255 events: 170 normal activity, 85 medium earthquakes)</i>: seismometers strong motion data (1-10Hz). Time series data collected from the international Federation of Digital Seismograph Networks (FDSN) client available in Python package ObsPy [6]. Channel information is specifying the units, sample rates, and gains of each channel. The number of medium earthquakes is calculated by the ratio of medium over large earthquakes during the past 10 years in the region. A ratio above 30% is kept between the number of 60 seconds MTS corresponding to earthquakes (medium + large) and total (earthquakes + normal activity) number of MTS to prevent a class imbalance issue. <br><br>The number of GPS stations and seismometers for each event varies (tens to thousands). <br><b>Preprocessing</b>:- Conversion (seismometers): data are available as digital signal, which is specific for each sensor. Therefore, each instrument digital signal is converted to its physical signal (acceleration) to obtain comparable seismometers data<br>- Aggregation (GPS stations and seismometers): data aggregation by second (mean)<br><b>Variables</b>:- event_id: unique ID of an event. Dataset is composed of 269 events.<br>- event_time: timestamp of the event occurence <br>- event_magnitude: magnitude of the earthquake (Richter scale)<br>- event_latitude: latitude of the event recorded (degrees)<br>- event_longitude: longitude of the event recorded (degrees)- event_depth: distance below Earth's surface where earthquake happened (km)<br>- mts_id: unique multivariate time series ID. Dataset is composed of 2,072 MTS from GPS stations and 13,265 MTS from seismometers.<br>- station: sensor name (GPS station or seismometer)<br>- station_latitude: sensor (GPS station or seismometer) latitude (degrees)- station_longitude: sensor (GPS station or seismometer) longitude (degrees)- timestamp: timestamp of the multivariate time series<br>- dimension_E: East-West component of the sensor (GPS station or seismometer) signal (cm/s/s)<br>- dimension_N: North-South component of the sensor (GPS station or seismometer) signal (cm/s/s)- dimension_Z: Up-Down component of the sensor (GPS station or seismometer) signal (cm/s/s)- label: label associated with the event. There are 3 labels: normal activity (GPS stations: 255 events, seismometers: 170 events) / medium earthquake (GPS stations: 0 event, seismometers: 85 events) / large earthquake (GPS stations: 14 events, seismometers: 14 events). EEW relies on the detection of the primary wave (P-wave) before the secondary wave (damaging wave) arrive. P-waves follow a propagation model (IASP91 [7]). Therefore, each MTS is labeled based on the P-wave arrival time on each sensor (seismometers, GPS stations) calculated with the propagation model.<br><br>[1] Ruhl, C. J., Melgar, D., Chung, A. I., Grapenthin, R. and Allen, R. M. 2019. Quantifying the value of real‐time geodetic constraints for earthquake early warning using a global seismic and geodetic data set. Journal of Geophysical Research: Solid Earth[2] Geng, J., Bock, Y., Melgar, D, Crowell, B. W., and Haase, J. S. 2013. A new seismogeodetic approach applied to GPS and accelerometer observations of the 2012 Brawley seismic swarm: Implications for earthquake early warning. Geochem. Geophys. Geosyst.[3] Geng, J., Jiang, P., and Liu, J. 2017. Integrating GPS with GLONASS for high‐rate seismogeodesy. Geophys. Res. Lett.[4] http://tunguska.uoregon.edu/rtgnss/data/cwu/mseed/[5] Melgar, D., Melbourne, T., Crowell, B., Geng, J, Szeliga, W., Scrivner, C., Santillan, M. and Goldberg, D. 2019. Real-Time High-Rate GNSS Displacements: Performance Demonstration During the 2019 Ridgecrest, CA Earthquakes.[6] https://docs.obspy.org/packages/obspy.clients.fdsn.html[7] Kennet, B. L. N. 1991. Iaspei 1991 Seismological Tables. Terra Nova<br>