The ambient seismic noise recorded by seismometers includes microseisms and anthropogenic or cultural noise. While microseisms are caused by the coupling between the ocean and the solid earth at frequencies mostly <1 Hz1, anthropogenic seismic noise, especially in urban areas, includes seismic signals generated by human activities such as moving people and industrial activities2–5. A recent study reported an increase in ambient seismic noise energy in response to an increase in the gross domestic product (GDP), which is associated with the magnitude of human activity6. Since an increase in ambient noise energy can be attributed to a wide range of human activities, ambient seismic noise in urban areas tends to be stronger and more complex. For instance, Díaz et al.5 monitored road traffic and subway trains along an avenue about 150 m away for frequencies >1 Hz (maximum energy between 8–35 Hz) and 20–40 Hz band, respectively. They succeeded in detecting signals generated by various anthropogenic noise including rock concerts, football games, and road traffic. Considering the temporary absence of a specific noise source, Green et al.7 monitored the ambient seismic noise during a complete shutdown of the subway system during an industrial strike, and found clearly evident reduction for frequencies >25 Hz within 100 m distance but no reduction at a distance of ~600 m from the nearest subway line. Hence, the ambient seismic noise includes various anthropogenic noises depending on the frequency, time, and distance8,9.
The outbreak of COVID-19 has immensely impacted and imposed restrictions on the routine lives of people all over the world. After the first report of COVID-19, it has spread rapidly and exponentially across the world, with the World Health Organization (WHO) declaring it a pandemic on March 13, 202010. Many countries, including China and Italy, went into lockdown for several weeks to contain major outbreaks of COVID-19. Although lockdown restrictions were effective in containing the virus' spread, a second or third wave of COVID-19 is a cause for concern. The daily life would change across the world under the outbreak of COVID-1911. The large-scale restrictions on human activities during the pandemic also presented a rare opportunity to clearly distinguish anthropogenic seismic noise from ambient seismic noise. Poli et al.12 and Xiao et al.13 compared ambient seismic noise levels before and during the pandemic-related lockdown by using continuous seismic records for Italy and China, and observed a clear decrease in noise level during the lockdown. Poli et al.12 found that the noise level significantly reduced in the frequency band 1–10 Hz. For China, the decrease in noise level was much larger than that for Italy13. Xiao et al.13 inferred that this was caused by the difference in the strictness of enforcing restrictions by the governments of the two countries. Reduction of up to 50% in the noise level in the frequency range 4–14 Hz up to 50 % were observed world over14. While the reduction was highest in populated areas, the seismic quiescence extended for several kilometers. At the local scale, continuous fiber-optic distributed acoustic sensing recordings showed a spatial reduction in noise level with a COVID-19 response15. A 50% decrease in vehicular noise was observed in one commuter sector immediately following the order to stay indoors, even though traffic near hospitals persisted.
Since the first case was reported on January 28, 2020, the number of confirmed cases in Japan increased. To prevent the spread of infection certain self-restraint advisories, such as school shutdowns and canceling public events, were issued by the national and local governments. In Tokyo, schools remained closed from March 2 to June 30, 2020. In response to the COVID-19 epidemic, the Japanese government declared a state of emergency in Tokyo from April 7 to May 25, 2020, during which many people voluntarily refrained from going out. On the other hand, trains and road traffic did not stop because economic activities still continued.
Regarding the strictness of government advisories, the declaration of the state of emergency by the Japanese government was not legally binding but was more of an appeal for self-restraint. While most people voluntarily refrained from going out, some people and stores did not follow the advisory. In addition, as most economic activities still continued during the emergency, public transport was also continued like the situation in Italy.
With this background we carried out a study to understand anthropogenic seismic noise and estimate the change in human activities during the COVID-19 pandemic by monitoring temporal changes in the seismic noise level around Tokyo.
Figure 1. Arrangement of MeSO-net. Red and green dots and squares show the MeSO-net stations. Dots and squares indicate the stations located in schools and parks, respectively. Blue and magenta lines indicate highways and railways, respectively.
Data Source and Calculation of Power Spectral Density (PSD)
We used the publicly available, continuous, three-component ground acceleration data recorded by the Metropolitan Seismic Observation Network (MeSO-net). MeSO-net stations have acceleration sensors with a sampling rate of 200 Hz and installed at depths of ~20 m to reduce the influence of anthropogenic noise16,17. Around 300 MeSO-net stations are deployed around the metropolitan area, out of which we used data from 118 stations located around Tokyo (Fig, 1). We used the vertical component of the seismic noise from 1 December, 2017 to 31 July, 2020.
To estimate the temporal change in ambient noise before and after the COVID-19 outbreak, we calculated the power spectral density (PSD) following the method of McNamara and Boaz18. PSD as a function of time and frequency is an effective tool to visually distinguish different types of seismic noise. We divided the daily data into 5-min segments, and applied demean and detrend to each data segment. After applying 10% cosine taper, we calculated the PSD for each 5-min data segment. In this study, we used the median value of PSD in the specific term (e.g., all day, daytime, or night) as the daily PSDs. We expressed PSD in units of decibels (dB) with respect to 1 (m2/s4)/Hz to make it easy to compare with previous or future studies.
Figure 2. Time-series and time-dependent PSDs for five stations (a–e). Left panels: time-series PSD from December 2017 to July 2020. Middle panels: time-dependent PSDs on weekdays and Sunday. Right panel: difference between time-dependent PSDs on weekday and Sunday.
PSDs near dominant specific anthropogenic noise sources
We focused on certain stations located near dominant specific noise sources, such as vehicle traffic, train, or industry. To study their spectral features, we estimated two types of PSD, viz. time-series (daily) PSD, and time-dependent PSD. We estimated PSDs by using the median value of all segments each day and at the same time during a day, respectively (Fig. 2). Time-dependent PSDs were calculated using weekday data and Sunday data, respectively. Here, we excluded bank holidays and long holidays (New Year holidays: December 27–January 6, summer holidays: August 11–August 16 and consecutive holidays: April 28–May 6).
Station GHGM was the closest to railway line (~60 m) but far from the highway (~1.5 km), while, conversely, station TKNM was the closest to the highway (~90 m) but far from the railway line (~2 km) (Fig. 1). The time-series PSDs at GHGM were expected to include seismic signals generated by the train and showed some belt-shaped anomalies, with clear peaks in the wide frequency range of ~10–20 Hz and a narrow frequency range of ~2–4 Hz (Figure 2a). Energy was high from 5 AM to 1 AM, likely caused by seismic noise from trains, since this corresponded with the time between the first (05:06) and last (00:55) train. Additionally, the difference in PSDs between weekdays and Sundays was clear due to the difference in train schedule.
Station TKNM, where seismic signals generated by vehicular traffic on the highway was expected to be dominant, showed a large PSD in a wide frequency range >1.5 Hz, with maximum energy over 1.5–20 Hz (Fig. 2b). This includes the frequency range 4–10 Hz, where the seismic signal generated by trains was weak. Although PSD decreased above 20 Hz, it still had a substantial energy in the higher frequencies depending on the day and time. The time-dependent PSDs clearly showed high energy during daytime on weekdays and a large difference in energy between weekdays and Sundays in the frequency range <20 Hz (Fig. 2b).
Station JKPM was located in a park in an industrial estate, and was far from both the highway (~2 km) and railway track (~3 km) (Fig. 1). The PSD in the frequency range 1–30 Hz was relatively large, with maximum energy over 1–10 Hz (Fig. 2c). Continuous dominant spike noises of certain specific frequencies were observed for the entire period or specific time. In particular, the spike noise ~48 Hz was dominant. It is likely that this noise was caused by vibrations generated by motors, transformers, inverters, etc., associated with mechanical equipment such as air conditioners and vending machines operating on 50 Hz commercial power supply19,20. Indeed, a spike noise of ~50 Hz was observed for other stations. Mechanical equipment in the industrial estate could induce additional spike noises at JKPM. The time-dependent PSD for weekdays showed the dominant energy during daytime in the frequency range <50 Hz. The large difference in energy between weekdays and Sundays was observed for the frequency range ~25–35 Hz and 3–10 Hz. Furthermore, the times when the energy increased (~09:00) and decreased (~18:00), and break times (~10:00, 12:00, and 15:00) was clear. These likely correspond to the working time of industries.
Station MBSM was located far from the highway (~2 km) and railway (~1.5 km) (Fig. 1). It showed a relatively low PSD in the entire period compared to other stations (Fig. 2d). A relatively large PSD occurred at a frequency 2–20 Hz, with a peak at ~3 Hz, and it was dominant on weekdays. For frequency range >20 Hz the energy on weekdays was higher around 8 AM and evening, and the energy on Sunday was higher in the morning. Possibly, these high energies were generated by school activities (e.g., going to school and club activities).
GNZM was located downtown and was close to the highway (~80 m) and subway (~110 m). The time-series PSD showed higher energy over a wide frequency range (2–90 Hz) for the entire period compared with other stations (Fig. 2e). High energy was observed in the frequency range >50 Hz except for late at night. A dominant spike noise ~50 Hz was evident, which could be caused by mechanical equipment, like the signals observed at station JKPM. For frequency range 10–30 Hz, we could observe large PSDs even late at night, similar to the signals observed at the station TKNM. The large difference in energy between weekdays and Sundays was evident for the frequency range <10 Hz from around 3 AM to 5 PM.
Frequency ranges
To simplify the frequency-dependent time-series PSD we defined five frequency classes (1–5, 5–10, 10–20, 20–45, and 55–85 Hz) based on the features of PSD, as described above (Fig. 2). In the lowest frequency range 1–5 Hz, PSD was dominant at all stations and also appeared late at night. Considering the signals observed late at night when trains were not running, we deduced that this was most likely caused by truck traffic for logistics. In the frequency range 5–10 Hz, seismic signals that can be mainly generated by trains were weak, and the difference between weekdays and Sundays was large for the stations TKNM, JKPM, and GNZM. PSD in the frequency range 10–20 Hz was dominant for all of these stations, especially GHGM, TKNM, and GNZM. In the frequency range >20 Hz, PSD was still high, depending on the time. PSD in the frequency range 20–45 Hz was greater than that for 55–85 Hz, although their dominant times were similar. While it is difficult to confirm that this classification was applicable for other stations, we still observed five frequency ranges from five stations located near specific dominant noise sources as described above.
We calculated the average value of noise level at each station for each frequency class by considering day (weekday and Sunday), daytime (7 AM to 5 PM), and nighttime (2 AM to 5 AM) (Fig. 3). We clearly see the high energy around the centers of the considered stations. These correspond to the youngest sediments21. The noise level strength likely reflected the ease of shaking or amplification due to soft sediments22,23. The energy in the frequency range >20 Hz was much smaller than that for the lower frequency range. The energy was higher during the daytime and on weekdays compared to nighttime and Sundays2,5.
Figure 3. Average noise levels calculated by considering the day and time in five different frequency classes.
Temporal changes in relative noise level
Figure 5. Map with correlation coefficients, which were calculated considering the day and time in five different frequency classes. Black star indicates Tokyo station. Black arrows indicate the station in the park with high correlation coefficient (55–85 Hz in panel a).
Assuming that a correlation coefficient ≥0.6 signified significant association, we observed strong and consistent seasonal effects for frequencies >20 Hz (Fig. 5). In the frequency range <20 Hz, there was no strong seasonal effect except for that on Sunday in the daytime in the frequency range 1–5 Hz. Around the Tokyo station (center of the city), we observed a low correlation for weekdays in the daytime for the frequency range 55–85 Hz. Low correlation can be attributed to weak seasonal effects, irregular local events, and changes in the environment.
Figure 5. Map with correlation coefficients, which were calculated considering the day and time in five different frequency classes. Black star indicates Tokyo station. Black arrows indicate the station in the park with high correlation coefficient (55–85 Hz in panel a).
We defined the annual seasonal change as the mean value of changes from December 2017 to November 2018 and December 2018 to November 2019, and we used only data that showed a correlation coefficient >0.6 (Fig. 6).
The annual seasonal changes for weekdays during daytime for the frequency range >20 Hz showed a clear trend (Fig. 6a), with a large peak (from late November to early March) and two dips (around late March and August). Notably, these dips were not observed for Sundays (Fig. 6b). The two dips coincided with long school holidays (roughly from 25 March to 5 April and from 20 July to 31 August). Therefore, we considered that this noise reduction was caused by the absence of the noise generated by school activities.
Considering seasonal change for the nighttime, we observed a strong seasonal variation for the frequency range 20–45 Hz on both weekdays and Sundays (Fig. 6). This noise increases in summer and winter. Hence, this perturbation likely indicates vibrations generated by mechanical equipment (e.g., air conditioners and water heaters). However, this seasonal change due to mechanical equipment could not be observed at frequencies <20 Hz as the background noise was strong (Fig. 3). In the frequency range <20 Hz, many stations showed high correlations only for daytime on Sundays for frequency range 1–5 Hz (Fig. 6). This seasonal variation showed high energy in winter and low in summer, and it could be masked by other strong noises on weekdays.
Figure 6. Seasonal variation in the relative noise level. Thin lines represent stations which showed high correlation coefficient. Black thick line indicates the median value. The seasonal variations were calculated considering the day and time in five different frequency classes.
Temporal changes in relative noise level during the COVID-19 pandemic
We corrected the temporal change in the relative noise level by deducting the annual seasonal change. The corrected temporal change in the relative noise level from December 2019 to July 2020 clearly decreased after temporary school shutdown (from March 2 to May 31) and declaration of a state of emergency (from April 7 to May 25) to contain the spread of COVID-19 (Fig. 7).
Figure 7. Corrected temporal change in relative noise level. Thin lines represent stations which showed high correlation coefficient. Black thick line indicates the median value. Yellow and pink shaded areas respectively represent the periods of school shutdown and the state of emergency in Tokyo. The corrected temporal changes were calculated by considering the day and time for five different frequency classes.
We could clearly see two reductions in noise level on weekdays for daytime in the frequency range >20 Hz (Fig. 7a). These reductions started following the school shutdown and declaration of the state of emergency. Before declaration of emergency the noise level recovered slightly. This period corresponded to normal school holidays. The noise level during school shutdown was similar to that during school holidays. During the state of emergency, the noise level decreased significantly. After the state of emergency was lifted, the noise level immediately recovered to the level before 2020; furthermore, it increased more than the usual level from July.
Focusing on nighttime, we could clearly detect the noise reduction during emergency for the frequency range 5–20 Hz, especially on Sundays, although the number of stations showing high correlation coefficient was limited.
The noise level gradually decreased after school shutdown for Sunday in the daytime in all frequency ranges, even for school holidays. Interestingly, the daytime noise level on Sunday did not immediately recover like that on weekdays, after emergency was lifted, and recovered gradually until the beginning of July.