11:45 AM - 12:00 PM
[SCG45-20] Trial tremor location using analog seismograms of the Kanto-Tokai Observation Network
Keywords:Slow Earthquakes, tremor, analog seismogram
Slow earthquakes have been studied by many researchers since the discovery of tectonic tremor (Obara, 2002, Science). However, the past activity of tremor before 2000 is still not clear in the Nankai region. This is due to the lack of continuous digital seismic data in this period, as the cost of data storage was very expensive. Though Kobayashi et al. (2006, Zisin) detected short-term slow slip events (SSEs) since 1984, the location of SSEs before 1998 are not clear as only one strainmeter is available. Thus, analog seismograms are essential dataset to reveal the details of slow earthquake activity in that period. In our previous study, we tried to develop a technique for automatic digitization of the seismograms on the recording paper (e.g., Matsuzawa and Takeda, 2021, AGU). Though our method can trace the image data successfully, however, the information of time marks is contaminated in the extracted data. Now, we are developing a technique to reduce the effect of time marks. In this talk, we also show a preliminary result of locating tremor.
National Research Institute for Earth Science and Disaster Resilience operated the Kanto-Tokai Observation Network since late 1970, and has stored analog seismograms on recording paper. The analog seismograms are recorded by ink recorders. Two hours data are divided to four blocks in a sheet. Traces of seismograms in the left and right blocks are 0-35s, 30-65s in each minute, respectively. Upper and lower blocks correspond to the former and the latter hour, respectively. In each trace, time marks for each second are added as streaks. Short offset is added in each minute.
Using scanned image data of these analog recordings, we developed a tool to digitize these data in our previous study. The recovered digital data from images are generally similar to the original data in the case of small amplitude signal, although high frequency components are lacked.
To reduce the contamination of time marks, we modeled the time series of time marks as repeating spike-like signals at the interval of 1 s. Each spike is modeled by a Gaussian function. The amplitude of spikes is estimated by the least-squares method, and the modeled time series are subtracted from the original data to reduce the effect of time marks. In addition, short offset in each minute is also modeled a boxcar function with Gaussian tapers at both ends. By this correction, peaks at 1Hz and 2Hz are less dominant in the power spectrum. Finally, we connected the traces, as the length of each trace is 35 s.
We tried to locate tremor in the Tokai region at 5am on Apr. 27, 1988, using the corrected seismograms at four stations in Tokai, as Kobayashi et al. (2006) detected a short-term SSE from Apr. 26 to Apr. 29 in 1988 in this region, and tremor signals are clearly identified by visual inspection of recording paper. We calculated envelope waveforms smoothed by 3-sec time window, after applying the bandpass filter between 2 and 8 Hz. The differences of arrival times are estimated from the correlation of envelope waveforms with the length of 2 min. We located tremor close to the Shimoyama station in the central Tokai region at 5:33-5:34am, by the grid search technique, fixing the depth of tremor at 30 km.
Though the effect of time marks is successfully reduced in this case, however, our method sometimes causes additional noises, especially when the time interval in the digitized data is not accurate. In addition, stain of ink and notes are also contaminated in the data. Therefore, manual corrections are still necessary in our analysis. Further development of automatic noise reduction would enable us to analyze long-term analog seismograms.
National Research Institute for Earth Science and Disaster Resilience operated the Kanto-Tokai Observation Network since late 1970, and has stored analog seismograms on recording paper. The analog seismograms are recorded by ink recorders. Two hours data are divided to four blocks in a sheet. Traces of seismograms in the left and right blocks are 0-35s, 30-65s in each minute, respectively. Upper and lower blocks correspond to the former and the latter hour, respectively. In each trace, time marks for each second are added as streaks. Short offset is added in each minute.
Using scanned image data of these analog recordings, we developed a tool to digitize these data in our previous study. The recovered digital data from images are generally similar to the original data in the case of small amplitude signal, although high frequency components are lacked.
To reduce the contamination of time marks, we modeled the time series of time marks as repeating spike-like signals at the interval of 1 s. Each spike is modeled by a Gaussian function. The amplitude of spikes is estimated by the least-squares method, and the modeled time series are subtracted from the original data to reduce the effect of time marks. In addition, short offset in each minute is also modeled a boxcar function with Gaussian tapers at both ends. By this correction, peaks at 1Hz and 2Hz are less dominant in the power spectrum. Finally, we connected the traces, as the length of each trace is 35 s.
We tried to locate tremor in the Tokai region at 5am on Apr. 27, 1988, using the corrected seismograms at four stations in Tokai, as Kobayashi et al. (2006) detected a short-term SSE from Apr. 26 to Apr. 29 in 1988 in this region, and tremor signals are clearly identified by visual inspection of recording paper. We calculated envelope waveforms smoothed by 3-sec time window, after applying the bandpass filter between 2 and 8 Hz. The differences of arrival times are estimated from the correlation of envelope waveforms with the length of 2 min. We located tremor close to the Shimoyama station in the central Tokai region at 5:33-5:34am, by the grid search technique, fixing the depth of tremor at 30 km.
Though the effect of time marks is successfully reduced in this case, however, our method sometimes causes additional noises, especially when the time interval in the digitized data is not accurate. In addition, stain of ink and notes are also contaminated in the data. Therefore, manual corrections are still necessary in our analysis. Further development of automatic noise reduction would enable us to analyze long-term analog seismograms.