In this paper, we constructed an earthquake ontology and Linked Open Data for seismology. Earthquake ontology defines data on seismic waveforms such as seismic intensity and occurred time, data on observation stations where seismic waveforms were observed, and classes and properties such as the size and depth of the hypocenter of the observed waveforms. By using the earthquake ontology, it is possible to assign URIs to "earthquakes" that cannot be observed unlike seismic sources and observed waveforms, making it possible to use the necessary earthquake data based on observation information. We developed the seismic dataset not only the world's seismic data for machine learning represented by STEAD but also data publicly available in Japan on a limited basis, such as JMA and NIED prevention, as Linked Open Data using earthquake ontology.
Japan is one of the most earthquake-prone countries in the world. Around the Japanese archipelago, four plates collide with each other, and more than 100,000 earthquakes occur per year, averaging more than 300 earthquakes per day, including those that are not felt.
Seismic motion is observed as waveform data of acceleration and is used in various research
such as calculation of seismic intensity, determination of hypocenter, emergency earthquake
warning, and predicted seismic intensity. In recent years, it has been used as training data for
research using machine learning, such as predicting the seismic intensity at a specific observation
station, whether the observed waveform is a seismic waveform, and identifying the
P-wave/Swave of an earthquake. Since machine learning requires a large amount of high-quality training
data, seismic observation networks are useful. However, one of the networks K-NET[
In general, the word "earthquake" refers to events such as tremors felt by people on their own, but in reality, it refers to the rapid displacement of the bedrock due to the pushing and pulling of the underground bedrock. Shaking occurs as a result of bedrock displacement and is recognized by us at the ground surface. Because earthquakes occur underground, it is dificult to actually observe them. Therefore, information on the waveforms observed at each observation station is important, such as the hypocenter estimation and calculating the seismic intensity.
Observations of seismic activity are conducted in many countries. The International
Federation of Digital Seismograph Networks (FDSN)[
In Japan, seismic waveforms observed by observation networks such as K-NET and
Kiknet, which are based on data from observation stations established by NIED, JMA, and local
governments, can be obtained. However, although the acquired data can be used for analysis
and other purposes, it cannot be redistributed, and only some of the JMA’s observation station
and data are registered in the FDSN. Other earthquake data is available in the Earthquake
Monthly Report (Catalog Edition)[
However, there is no list of which stations observe which earthquakes, although multiple stations must observe the same earthquake to be selected when searching by the station. In addition, when searching from the hypocenter, it is not known whether the observation stations observe the earthquake that occurred at that hypocenter or not without searching the data and making a list. Furthermore, since the observation stations are diferent from each other, it is dificult to retrieve the seismograms of earthquakes that occurred at the same hypocenter from multiple observation stations because the IDs are not assigned to each earthquake.
To solve the problem in section 2, we aim to make the observed waveforms publicly available and searchable in the form of Linked Data, which links data together. First, we organized the vocabulary related to earthquakes. The JMA’s Earthquake Monthly Report (Catalog Edition) does not provide data on observed waveforms, but it does provide metadata on observed earthquake ground motions. The data in the Earthquake Monthly Report (Catalog Edition) include source data, measured data, first motion mechanism solution data, CMT solution data, seismic intensity data, tsunami data, etc. In this paper, the seismic intensity data file was first selected as the target. The seismic intensity data contains a record called the hypocenter record, which contains information on the hypocenter, and information on the observation points where the earthquake motion that occurred at the hypocenter was observed. Therefore, we created an ontology as an earthquake vocabulary based on the data contained in the Earthquake Monthly Report (Catalog Edition) of JMA. The seismic intensity data contains a record called the hypocenter record, which contains information on the hypocenter, and information on the observation stations where the earthquake motion that occurred at the hypocenter was observed.
Since the earthquake itself cannot be observed, it is important to show the relationship between the waveform information actually observed at the observation station and the hypocenter and magnitude estimated from the observed waveform as the semantics of the earthquake. Figure 1 shows the hypocenter and observed waveforms graphically.
The earthquake ontology was constructed based on the hypocenter, seismic motion, observed
waveforms, and observation station that constitute an earthquake. The seismic motion and
the observation station that observe the waveforms at the ground surface were described
using the SOSA (Sensor, Observation, Sample, and Actuator) ontology of the SSN (Semantic
Sensor Network)[
We converted the available data from the JMA’s Earthquake Monthly Report (Catalog Edition) and FDSN earthquake events into Linked Data based on Earthquake Ontology. Since FDSN includes observation networks registered with ISC and STEAD, data outside Japan are using FDSN. Networks in FDSN are United States National Seismic Network, Hawaiian Volcano Observatory Network, Montana Regional Seismic Network, Southern California Seismic Network, Nevada Seismic Network, Pacific Northwest Seismic Network - University of Washington, USGS Northern California Seismic Network, Alaska Geophysical Network, Oklahoma Seismic Network, University of Utah Regional Seismic Network, Raspberry Shake, Alaska Volcano Observatory, Texas Seismological Network, Puerto Rico Seismic Network & Puerto Rico Strong Motion Program, US Geological Survey Networks, Lamont-Doherty Cooperative Seismographic Network, Geological Survey Networks, National Tsunami Warning Center Alaska Seismic Network.
In this paper, we organized earthquake data registered in the Earthquake Monthly Report from JMA and data from FDSN published. We also created an ontology by organizing vocabulary related to earthquakes and published it in the form of Linked Open Data by assigning URIs to earthquakes based on information on observed waveforms and hypocenters.
In the future, we aim to create an infrastructure for multiple observation networks by converting the latest data published in the JMA’s seismic intensity database, data in NIED’s seismic observation network, and seismic data in FDSN into LOD. Furthermore, we will promote the availability of an earthquake catalog format that can be used as earthquake data and learning data, and LOD conversion of data observed by our own observation network. Benchmarking based on the same dataset is important for source determination, calculation of predicted seismic intensity, and training data for machine learning, but it is believed that datasets for reproduction are not distributed due to the fact that Japanese data cannot be redistributed and IDs are not assigned. By using the earthquake ontology created in this paper to describe the datasets used in earthquake research in LOD, it is expected that the availability of datasets for reconstruction will be improved.