11:00 AM - 1:00 PM
[SSS13-P06] Toward high-density, broadband, and long-term monitoring of seismogram amplitude
Keywords:seismogram, continuous record, dataset
The continuous observation record by the high-density, high-sensitivity seismograph network has been attracting attention as a record of other than ordinary earthquakes. These continuous records contain information on a wide variety of vibrations caused by the interaction between the hydrosphere and the solid earth, and various noises caused by human activities. However, these records are very diverse, and it is essential to understand the steady-state characteristics of the continuous records that have been accumulated in order to discover and understand further phenomena from them. In this study, we attempted to create a basic dataset for further extracting features from all records of Hi-net, which is a highly sensitive seismic observation network.
In this study, all stations of Hi-net, a highly sensitive seismic observation network of the National Research Institute for Earth Science and Disaster Prevention (NIED) as of the end of 2021, are used to extract their amplitude values by the following standard signal processing for the records since April 1, 2004. First, the continuous waveforms of the three components at each station were cut out using 9:00 to 8:59 of the following day in Japan Standard Time as the unit of the day, with a margin of 5 minutes before and after each. The amplitudes of the data were muted by applying a cosine taper at the beginning and end of the data after removing the mean. Daily artificial signal for instrumental health monitoring at around 9:00 JST also muted off. Then, time-domain convolution and deconvolution filters were used to convert the data into an equivalent seismograph record with a natural frequency of 0.025 Hz. To these records, the amplitudes of three-components in seven frequency bands (0.1-0.25 Hz, 0.25-0.5 Hz, 0.5-1.0 Hz, 1.0-2.0 Hz, 2.0-4.0 Hz, 4.0-8.0 Hz, and 8.0-16.0 Hz) were averaged over a one-minute time window.
The continuous waveform record of Hi-net consumes about 600-800 GB per month, but the record output by this process is about 1.6 GB per month in the HDF5 format, which means that the data volume has been compressed to ~ 1/350. The analysis is being conducted, but as the continuous records to be analyzed are enormous of ~130 TB, the processing of records from April 2004 to the end of 2012 has been completed at the time of writing this abstract. If all goes well, the processing of records up to the end of 2021 is expected to be completed by the time of the presentation, and the results of the processing using the latest data will be presented at the presentation.
As a preliminary analysis based on these records, we monitored the day-elapsed change of the seismic amplitude. First, we defined the daytime and nighttime periods as 10:00 to 16d:00 and 22:00 to 04:00 Japan time, respectively, and extracted the median values in each interval as the daytime and nighttime values, and examined their spatio-temporal distributions. In the low-frequency range at a frequency of approximately 1 Hz and lower, the amplitude pattern with spatial continuity and its annual variation, which is considered to be caused by changes in weather conditions, was observed. While in the higher frequency range, there was a marked difference between day and night, weekday and holiday, and the amplitude varied greatly depending on the location. The amplitude patterns and their diurnal and annual variations were significantly different in the high-frequency region. As an example, the nighttime amplitude of 2011/3/17, immediately after the Tohoku earthquake, is shown in the figure. The median amplitude of the high-frequency tremor is clearly visible because of the continuous aftershocks, but its range of influence is from southern Kanto to central Hokkaido. On the other hand at lower frequencies, it can be seen that a large amplitude region is spreading widely along the coast of the Sea of Japan, uncorrelated with the earthquake. In the future, we will try to extract more features based on this dataset.
In this study, all stations of Hi-net, a highly sensitive seismic observation network of the National Research Institute for Earth Science and Disaster Prevention (NIED) as of the end of 2021, are used to extract their amplitude values by the following standard signal processing for the records since April 1, 2004. First, the continuous waveforms of the three components at each station were cut out using 9:00 to 8:59 of the following day in Japan Standard Time as the unit of the day, with a margin of 5 minutes before and after each. The amplitudes of the data were muted by applying a cosine taper at the beginning and end of the data after removing the mean. Daily artificial signal for instrumental health monitoring at around 9:00 JST also muted off. Then, time-domain convolution and deconvolution filters were used to convert the data into an equivalent seismograph record with a natural frequency of 0.025 Hz. To these records, the amplitudes of three-components in seven frequency bands (0.1-0.25 Hz, 0.25-0.5 Hz, 0.5-1.0 Hz, 1.0-2.0 Hz, 2.0-4.0 Hz, 4.0-8.0 Hz, and 8.0-16.0 Hz) were averaged over a one-minute time window.
The continuous waveform record of Hi-net consumes about 600-800 GB per month, but the record output by this process is about 1.6 GB per month in the HDF5 format, which means that the data volume has been compressed to ~ 1/350. The analysis is being conducted, but as the continuous records to be analyzed are enormous of ~130 TB, the processing of records from April 2004 to the end of 2012 has been completed at the time of writing this abstract. If all goes well, the processing of records up to the end of 2021 is expected to be completed by the time of the presentation, and the results of the processing using the latest data will be presented at the presentation.
As a preliminary analysis based on these records, we monitored the day-elapsed change of the seismic amplitude. First, we defined the daytime and nighttime periods as 10:00 to 16d:00 and 22:00 to 04:00 Japan time, respectively, and extracted the median values in each interval as the daytime and nighttime values, and examined their spatio-temporal distributions. In the low-frequency range at a frequency of approximately 1 Hz and lower, the amplitude pattern with spatial continuity and its annual variation, which is considered to be caused by changes in weather conditions, was observed. While in the higher frequency range, there was a marked difference between day and night, weekday and holiday, and the amplitude varied greatly depending on the location. The amplitude patterns and their diurnal and annual variations were significantly different in the high-frequency region. As an example, the nighttime amplitude of 2011/3/17, immediately after the Tohoku earthquake, is shown in the figure. The median amplitude of the high-frequency tremor is clearly visible because of the continuous aftershocks, but its range of influence is from southern Kanto to central Hokkaido. On the other hand at lower frequencies, it can be seen that a large amplitude region is spreading widely along the coast of the Sea of Japan, uncorrelated with the earthquake. In the future, we will try to extract more features based on this dataset.