9:15 AM - 9:30 AM
[STT41-02] Enhancement of the earthquake early warning system using hypoTD
Keywords:hypoTD, EEW, Hypocenter location, Registered template earthquakes
1. Introduction The Earthquake Early Warning (EEW) system locates hypocenters when only a few stations detect P-wave arrival times, enabling the prompt transmission of EEW information. However, there are cases of large location errors in the EEW system due to the heterogeneity of the velocity structure. HypoTD by Horiuchi et al. (2025) registers numerous template earthquakes and conducts relative location.This report presents whether hypoTD can improve the accuracy of hypocenter locations in EEW system.
2. Data We used P-wave arrival times for earthquakes with a magnitude of 4 or greater that occurred across the Japanese archipelago between 2019 and 2024. We obtained the arrival time data by downloading the Japan Meteorological Agency (JMA) unified earthquake catalog. We used P-wave arrival times within 3 seconds from the P-wave arrival at the closest station. Events with fewer than four P-wave arrivals within 3 seconds were not used. There were 2,178 earthquakes with a magnitude of 4 or greater, and 1,689 of these included data from more than 4 stations within 3 seconds. There were also earthquakes with depths greater than 300 km and earthquakes occurring in areas distant from the seismic network. We registered 220,000 template earthquakes, which are used by Horiuchi et al. (2025). These template earthquakes are distributed across the entire Japanese islands. Before conducting hypocenter location, we stored in memory the hypocenter parameters, station codes, and P-wave arrival times at 10 stations close to the epicenters of these template earthquakes.
3. Hypocenter Location HypoTD selects template earthquakes located close to the initial location of event being located. We determined this by comparing the similarity with the data of template earthquakes stored in memory as follows. (1) Select template earthquakes that have more than four same station codes. (2) Determine the value of arrival time difference between the event being located and the selected template earthquake by applying the least-squares method to the arrival time difference data for the same stations. Also, calculate residuals and the residual sum of squares (RMS) in the least-squares solution. (3) If there are arrivals in the template earthquake that are not observed in the event being located and they arrive earlier than those of the event being located, add a value corresponding to the time differences to the RMS. (4) Select the template earthquake with the minimum RMS and assume its hypocenter location as the initial hypocenter. We locate hypocenters using the obtained initial location with applying HypoTD. If the solution using HypoTD is unstable, we use the initial value as the hypocenter.
4. Results We conducted four kinds of hypocenter locations using P-wave arrival times within three seconds and compared them with the JMA unified earthquake catalog. Figure 1 shows cumulative frequency distributions of the horizontal differences for the four kinds of location methods: (1) HypoTD (red), (2) Initial locations used in HypoTD, as shown above (brown), (3) Generating two adjacent earthquakes in the hypocenter list and locating the two earthquakes simultaneously using HypoTD (blue). We generate the two earthquakes so that arrival times to the closest station are the same, (4) Ordinal hypocenter location method (black).
As shown in the figure, 10% of earthquakes have an error of over 40 km in the ordinal location method, whereas the error reduces to 10 km with HypoTD. Additionally, we were able to determine reliable hypocenters from the search for similarity with template earthquakes. This figure demonstrates that the present method can accurately locate hypocenters even during the occurrence of two earthquakes. This study utilized observation data from the JMA, NIED, various universities, and research institutes. We express our gratitude for their contributions.
2. Data We used P-wave arrival times for earthquakes with a magnitude of 4 or greater that occurred across the Japanese archipelago between 2019 and 2024. We obtained the arrival time data by downloading the Japan Meteorological Agency (JMA) unified earthquake catalog. We used P-wave arrival times within 3 seconds from the P-wave arrival at the closest station. Events with fewer than four P-wave arrivals within 3 seconds were not used. There were 2,178 earthquakes with a magnitude of 4 or greater, and 1,689 of these included data from more than 4 stations within 3 seconds. There were also earthquakes with depths greater than 300 km and earthquakes occurring in areas distant from the seismic network. We registered 220,000 template earthquakes, which are used by Horiuchi et al. (2025). These template earthquakes are distributed across the entire Japanese islands. Before conducting hypocenter location, we stored in memory the hypocenter parameters, station codes, and P-wave arrival times at 10 stations close to the epicenters of these template earthquakes.
3. Hypocenter Location HypoTD selects template earthquakes located close to the initial location of event being located. We determined this by comparing the similarity with the data of template earthquakes stored in memory as follows. (1) Select template earthquakes that have more than four same station codes. (2) Determine the value of arrival time difference between the event being located and the selected template earthquake by applying the least-squares method to the arrival time difference data for the same stations. Also, calculate residuals and the residual sum of squares (RMS) in the least-squares solution. (3) If there are arrivals in the template earthquake that are not observed in the event being located and they arrive earlier than those of the event being located, add a value corresponding to the time differences to the RMS. (4) Select the template earthquake with the minimum RMS and assume its hypocenter location as the initial hypocenter. We locate hypocenters using the obtained initial location with applying HypoTD. If the solution using HypoTD is unstable, we use the initial value as the hypocenter.
4. Results We conducted four kinds of hypocenter locations using P-wave arrival times within three seconds and compared them with the JMA unified earthquake catalog. Figure 1 shows cumulative frequency distributions of the horizontal differences for the four kinds of location methods: (1) HypoTD (red), (2) Initial locations used in HypoTD, as shown above (brown), (3) Generating two adjacent earthquakes in the hypocenter list and locating the two earthquakes simultaneously using HypoTD (blue). We generate the two earthquakes so that arrival times to the closest station are the same, (4) Ordinal hypocenter location method (black).
As shown in the figure, 10% of earthquakes have an error of over 40 km in the ordinal location method, whereas the error reduces to 10 km with HypoTD. Additionally, we were able to determine reliable hypocenters from the search for similarity with template earthquakes. This figure demonstrates that the present method can accurately locate hypocenters even during the occurrence of two earthquakes. This study utilized observation data from the JMA, NIED, various universities, and research institutes. We express our gratitude for their contributions.