Earthquake Reaearch in China  2017, Vol. 31 Issue (3): 289-308
Temporal and Spatial Images of Ambient Noise Intensity in the Chinese Mainland
Zhou Lianqing1,2,3, Song Xiaodong2, Zhao Cuiping1
1. Key Laboratory of Earthquake Prediction, Institute of Earthquake Science, CEA, Beijing 100036, China;
2. University of Illinois at Urbana-Champaign, Urbana IL 61801, USA;
3. Institute of Geophysics, China Earthquake Administration, Beijing 100081, China
Abstract: Successive waveforms of the vertical component recorded by 888 broadband seismic stations in the China Seismography Network from January, 2010 to June, 2011 are used to investigate the temporal and spatial distribution of ambient noise intensity, and the images of ambient noise intensity at the period of 10s in the Chinese Mainland are obtained. The temporal variation of ambient noise intensity shows some seasonal and periodic characteristics. The maximum ambient noise intensity occurred from January, 2011 to March, 2011. The spatial distribution images of ambient noise intensity show obvious zoning features, which doesn't correlate with surface geology, suggesting that the noise field is stronger than the site factors. The strength in southeastern coastal areas reaches its maximum and generally decreases toward to inland areas, and arrives at the minimum in the Qinghai-Tibetan Plateau. The zonal intensity distribution is probably correlated with ocean tides from the Philippine Ocean and the Pacific Ocean. It also shows that the influence from the Indian Ocean seems small. However, the ambient noise intensity increases to a certain degree in the Xinjiang area, indicating that the main source of ambient noise in the western area of the Chinese Mainland is not derived from the East and South China Sea, but rather from the deep interior of the Eurasian continent. The ambient noise intensity obtained in this study can supply reference for seismology research based on ambient noise correlation. Moreover, it can supply basic data for attenuation research based on ambient noise, and thus help achieve the object of retrieving the attenuation of Rayleigh waves from ambient noise.
Key words: Ambient noise     Noise source     Amplitude of surface wave     Attenuation

INTRODUCTION

Seismic ambient noise is usually used to describe signals generated by all unknown or uninterested sources in a seismogram.Unlike instrument noise or noise produced by environmental factors near a seismometer that can be controlled or reduced to some extent, ambient noise usually refers to signals generated by those noise sources beyond human control. These noise sources can be seen as a random contribution to the recorded ground motions. The frequency range of the earth's seismic noise is usually between 10Hz and 10000s (Peterson J.R., 1993). The physical process of seismic noise generation is still unclear in a certain frequency range (Ekström G., 2001). Short-period ambient noise usually refers to microseism, which is generally considered to be caused by the interaction of tidal fluctuation in coastal areas. Two types of relatively strong short-period seismic ambient noise are usually observed in two periodic frequency bands, of which, the first-arrival microseism is at the period of 10s-20s, and the second-arrival microseism at the period of 5s-10s (Bormann P. et al., 2009; Yang Yingjie et al., 2008; Lee W.S. et al., 2013). High-frequency seismic noise is usually generated by natural phenomena such as wind (e.g. wind friction on rough roads and oscillation or vibration of trees and vegetation in the wind, etc.) and running water (e.g. waterfall or fast-flowing rivers and streams) etc. The noise generated by wind is broadband, at a frequency band of 0.5Hz to about 15Hz-60Hz. However, the main source of high-frequency noise is human activity (such as machine rotation or hammering, road or railway traffic etc.) (Bormann P. et al., 2009). Most of these noise sources are scattered, static or kinetic. Long-period seismic noise refers to the hum of the earth (Nawa K. et al., 1998), usually the noise at a period above 100s, which can be observed in the persistent background free oscillations in the low-frequency seismic spectra. Early studies (Ekström G., 2001; Tanimoto T., 1999) suggested that long-period noise is correlated with atmospheric motion. Sorrells G.G. et al. (1971) attributed long-period ambient noise in ground motion to regional fluctuations of the pressure field. Although the change of pressure field could produce horizontal and vertical displacement, Sorrells G.G. (1971) considered that the tilt of horizontal components was the main source of ambient noise. However, the latest research (Rhie J. et al., 2004, 2006; Tanimoto T., 2005) has found that the source of long-period noise is more likely to be related to ocean infragravity waves (long-period gravity waves). Rhie J. et al. (2004) holds that long-period seismic noise includes the mutual coupling processes of atmosphere, oceans and seafloors. Also, the stability of ambient noise varies in different frequency bands. Short-period seismic ambient noise at the period below 1s often shows large fluctuations at different seismic stations in the same seismic network, while relatively long-period ambient noise at the period of above 10s is a more stable function with respect to time (Peterson J.R., 1993). Unlike body waves, the amplitude of seismic noise surface waves decreases exponentially with depth. As the penetration depth of surface waves increases with the wave length, high-frequency noise attenuates faster with increasing depth (Bormann P. et al., 2009).

Early seismologists often believed that records of seismic ambient noise were difficult to understand and useless, but recent studies found that records of continuous ambient noise at two seismic stationscontained information from the same noise sources, and by the method of cross-correlation of ambient noise, the empirical Green's function of the surface wave could be extracted, and corresponding traveling time of its group velocity was basically the same as that of the seismic surface wave traveling on the same path (Shapiro N.M. et al., 2004). Weaver R.L. (2005) inferred that seismic noise correlation may provide powerful tools for study in many fields. Later, the study of seismology by means of cross-correlation of the ambient noise surface wave has been rapidly developed and widely used in tomography of the velocity structure (Shapiro N.M. et al., 2005; Zheng Sihua et al., 2008; Sun Xinlei et al., 2010; Xu Z.J. et al., 2013; Bao Xuewei et al., 2015; Zhou Longquan et al., 2012; Zheng Xian et al., 2012; Fang Lihua et al., 2010; Tang Youcai et al., 2011; Zheng Yong et al., 2011; Yao Huajian et al., 2006; Yang Yingjie et al., 2010; Zheng Xian et al., 2015; Wang Weilai et al., 2014; Tang Xiaoyong et al., 2011; Yao Huajian et al., 2009; Lin Fanchi et al., 2008), study of media anisotropy (Lin Fanchi et al., 2011) and monitoring of seismic wave velocity changes in typical areas such as volcanoes or faults (Brenguier F. et al., 2008a, b).

Relevant researches on the earth's structure based on ambient noise are mostly studies of seismic velocity using elastic attributes of the earth's media. In fact, earth media also has an important inelastic attribute, namely a seismic wave attenuation structure. The attenuation of seismic waves is usually studied by extracting the amplitude of seismic waves to observe the changes of amplitude. Although there are few studies on the extraction of surface wave amplitude attenuation from seismic ambient noise at present, some encouraging research has been achieved in recent years. Research shows that it is possible to recover seismic surface wave amplitude by using ambient noise data, and then calculate the attenuation structure of media. Prieto G.A. et al. (2008) used the ambient noise field to predict strong ground motion, and the results were basically in line with the results obtained from forecasts using earthquakes. Although the accuracy of the results of some research methods remains to be further verified (Tsai V.C., 2009, 2011), many studies have successfully extracted amplitude of surface waves from ambient noise data (Cupillard P. et al., 2010, 2011; Lin Fanchi et al., 2011; Prieto G.A. et al., 2009, 2011) for the inversion of the attenuation structure (Lawrence J.F. et al., 2011). Weaver R.L. (2011) theoretically proved the feasibility of extracting amplitude and attenuation information from ambient noise, and accurately extracted surface wave attenuation from the ambient noise field using numerical data. Research also shows that the intensity of ambient noise sources is very important in the inversion of the attenuation structure of surface waves, and a new method of accelerating cross-correlation convergence has been put forward.

Lin Fanchi et al. (2011b) calculated the average attenuation coefficient of medium in the western part of the Continental United States using ambient noise data recorded by all US Array stations in the western part of the Continental United States from October, 2004 to April, 2010. Research found that the intensity of noise source in the west of the Continental United States recorded using a seismic station for reference was strongly dependent upon azimuthal angle, and the average intensity of noise source in different directions was calculated. The intensity of the noise source in the direction of maximum amplitude can be more than 7 times the intensity of the noise source in the direction of the smallest amplitude. Therefore, it is necessary to correct the influence of the non-uniformity in direction for the calculation of the media attenuation structure. The noise surface wave attenuation factor obtained by amplitude correction is basically consistent with the seismic surface wave attenuation factor. Song X. using ambient noise data from more than 300 stations in the Chinese Mainland from 2007 to 20092, obtained the variation of the intensity of different noise sources in Chinese Mainland with the change of azimuth angle after calculating the amplitude of empirical Green's function at the period of 10s-20s. Statistical results showed that the intensity of noise sources at different azimuth angles recorded by stations of the Chinese Mainland varied significantly, and the intensity of noise sources in the direction of the maximum amplitude can be up to nearly 10 times the intensity of noise sources in the direction of the smallest amplitude. Yang Yingjie et al. (2008) studied the intensity and directional variation of ambient noise sources in the Continent of Europe using data at the period of 8s and 14s respectively, and also found that ambient noise sources at these two cycles both showed significant directional variations.

2 Song X. Ambient Noise Attenution, 2011.

This study attempts to explore the spatial and temporal distribution characteristics of ambient noiseintensity in the Chinese Mainland and discuss its influence on the empirical Green's function. In spatial terms, in addition to the intensity changes with the variation of azimuth, how much difference might the intensity of ambient noise in all regions have? What are the laws of intensity distribution? How fast does noise intensity change in time? Are there seasonal changes? Is there a periodicity?

1 DATA

In this article, successive waveforms recorded by 144 broadband seismic stationsin the China National Digital Seismic Network and 744 broadband seismic stations of regional seismograph networks are used to investigate the temporal and spatial distribution of ambient noise intensity in the Chinese Mainland. The instrument at each station has long period frequency band above 60s, and the distribution of stations is shown in Fig. 2, of which, stations in the southeast of China are densely distributed, with average station spacing of 50km-100km and less than 50km in some areas, while the density of stations in the Qinghai-Tibetan Plateau, Xinjiang and Inner Mongolia is relatively low, with station spacing even greater than 300km in some areas Stations distribution is also uneven due to the influence of terrain conditions and other factors. Vertical components of successive waveforms recorded in the 18 months from January, 2010 to June, 2011 are used in this article. The vast majority of records from these stations were started before January, 2010. All stations are equipped with a three-component seismometer, with relatively complete continuous waveform records, and there are few cases of long time data breaks. Most of the 144 seismic stations of the China National Digital Seismic Network have instruments with bandwidth of more than 120s and better base environment, thus the quality of continuous waveform recording is relatively high. While influenced by various factors, continuous waveform recording at some stations in the regional seismic network shows signal instability, excessive base noise and more interference, which has brought some difficulty to our study. These anomalies are handled appropriately in this article to minimize their impact on the results.

 Fig. 2 Demonstration of removing earthquakes and spikes in the continuous waveform (a) Examples of one-day continuous waveform record.(b)Same waveform after removing earthquakes and spikes

Successive waveforms recorded by a seismometer contain not only seismic signals, but also various ambient noise from the natural world. The main source of these noise records is very complex, and the main energy comes from ocean tides and current, typhoon, rainstorm and other natural phenomena, which might be more concentrated at long periods. Human activity is also an important source of ambient noise, which can cause more short-period effects, and the sphere of influence is also much smaller than that of natural phenomena (Bormann P. et al., 2009). When we use ambient noise cross-correlation to retrieve the surface wave Green's function, ambient noise sources also come from various noise caused by nature and human activity. Because noise is transmitted to different stations through different paths and distances, which results in the correlation of earthquakes with the same noise sources recorded at different seismic stations, and there may be some time delay at the same time, which brings us the possibility of velocity structure inversion (Weaver R.L., 2005; Shapiro N.M. et al., 2004, Shapiro N.M. et al., 2005). In plain terms, after removing earthquakes and some abnormal interference from waveform records, we get ambient noise records at the seismic station, and ambient noise records at each sampling point reflect the intensity of ambient noise at the seismic station in its position at that moment. Therefore, we will make a proper treatment of successive waveform records to keep clean ambient noise records, and then define a time window as the unit to calculate the amplitude root-mean-square value (rms) within the time window as the ambient noise intensity at the seismic station.

We split successive waveforms recorded by each station into everyday records and preprocess the vertical component of one-day successive waveforms recorded by each station. Firstly, the original waveform is sampled with a sampling rate of 1s, then after removing the mean value and tilt, 5s-150s filtering is applied to the processed waveform and instrument response is deducted. Because high-frequency noise mainly comes from human activity, which attenuates very quickly with increased depth and distance (Bormann P. et al., 2009; Webb S.C., 1998), high-frequency noise is not considered in this study. Long-period noise in the frequency band of the Earth's hum can spread farther and reflect medium depth deep into the upper mantle, however, it is also not discussed in our study because there is no seismic activity in the upper mantle, where lateral heterogeneity of attenuation is also smaller. This study is concerned with ambient noise in the microseismic frequency band, namely 5s-20s frequency band, the depth of which is mainly concentrated in the crust (Sun Xinlei et al., 2010; Fang Lihua et al., 2010; Zheng Yong et al., 2011; Yang Yingjie et al., 2010), which is the most important area for the study of seismology. Because the intensity of noise source changes dramatically with frequency (Bensen G.D. et al., 2007), we need to carry out narrow-band filtering on the continuous waveform records at a specific period. In this article, centering on a period of 10s, we carry out narrow-band filtering using the period of 8s and 12s, so the distribution of ambient noise intensity at the period of 10s is obtained. The obtained waveform is shown in Fig. 2(a). Specific methods for the calculation of spatial and temporal distribution of ambient noise intensity based on preprocessed data are as follows:

(1) Daily rms value is calculated for all stations and a threshold value is set for normal records. When an all-day rms value for a station is greater than the threshold value, it can be identified as an exception record to be removed. Other abnormal signals may appear in many seismic stations in regional seismic network. Three typical abnormal signals as shown in Fig. 3 often appear in some seismic stations, and successive waveforms recorded by instruments are not normal noise or seismic records. The existence of these signals also affects the accuracy of calculation results of the distribution of ambient noise intensity. Appropriate methods are also adopted to remove these records according to the characteristics of these abnormal records.

 Fig. 3 Three typical examples of problematic continuous waveforms

(2) Because seismic signals, especially signals of large earthquakes, are far stronger than the intensity of ambient noise, they will seriously affect the calculation of ambient noise intensity. Therefore, we need to remove seismic or interference signals that may exist in the successive waveform records. The deduction method is to use the ratio between amplitude rms in a short time window and amplitude rms in a long time window to indentify seismic signals, and then by sliding judgment, until seismic signals are deducted from everyday successive waveforms. A 20-minute window is chosen as the time window to identify seismic signals, which we call the short time window. The rms value of the amplitude of signals in this short time window, namely RS, is calculated.

(3) A one-hour window is chosen as the time window for noise signals, which we call long time window. The rms value of the amplitude of signals in this long time window, namely RL, is calculated.

(4) RS/RL value is calculated. It is found through experiments that, when RS/RL > 6, it can be recognized as a seismic signal, which should be removed from the waveform records. Interference signals may appear in the records of some seismic stations, which are usually manifested as much larger in amplitude than the average rms value of the station of the day in the seismograms and will affect the average rms value of the whole set of records if untreated, thus affecting the calculation of ambient noise intensity. These abnormal signals are removed in a similar way to the removal of seismic signals.

(5) The rms value is calculated for all stations in each time window (e.g.2 hours). Some seismic stations contain bad data which are removed in the previous processing. The amount of data in some time windows may be small due to the deduction of seismic and interference signals, and too little data can lead to large errors in results. We count the sampling points of all stations, and if the sum of the sampling points is less than 10, 800 (the number of sampling points corresponding to 3 times the time window length), we will regard it as invalid data and deduct it. Next, the rms value of all stations is used to normalize the rms value of each station, and finally the rms value of each station in each time window is obtained.

After the above data processing steps, we can think that the successive waveform obtained is the record of pure ambient noise, and the rms value of each station we obtained after normalization in corresponding time windows represents the intensity of ambient noise at the station in the time period.

2 RESULTS

By calculating the everyday rms value of each station, we obtain the rms values of 888 broadband stations in the study area used in this article for 18 months from January, 2010 to June, 2011. The diagram of temporal variation of ambient noise intensity for all stations nationwide is obtained (Fig. 4) by the average rms value of all stations. The intensity of ambient noise in the Chinese Mainland can be seen to fluctuate greatly with time. Ambient noise intensity is relatively stable and weak from January, 2010 to October, 2010, while the overall ambient noise intensity is higher from October, 2010 to June, 2011, especially in individual months, for example, during October, 2010 to November, 2010 and in March, 2011, ambient noise intensity is significantly higher than that in previous months. The temporal distribution curve of the rms value also shows that the intensity of ambient noise in the Chinese Mainland presents obvious seasonal variation, which, for example, may be stronger in autumn and winter than in spring and summer. The intensity of ambient noise also shows some periodic characteristics, but due to the limited study period, the periodicity is not very obvious. By taking 3 different time window lengths, 1 hour, 2 hours and 1 day, to calculate the temporal distribution of the noise source, we find that the distribution characteristics of the noise source is not affected by various time window lengths and the noise source intensity also changes slightly, indicating that the noise source intensity usually does not change very much in a relatively short period of time (such as 1 day). Therefore, we believe that compared to using a shorter time window, such as 1 hour, a one-day time window can be used to calculate ambient noise intensity, which can increase the computation speed without significantly affecting the accuracy of ambient noise intensity calculation.

 Fig. 4 Temporal variation of average intensity of ambient noise for all stations The colors indicate the window lengths (labeled) used in averaging

We divide the study area into 0.5°×0.5° grids. Centering on the center of a grid point, we search for all stations in the circle with a radius of 300km, and calculate the average rms value of all stations on the same data as the rms value for the grid point of the day. For a grid point, when the number of seismic stations within the given range is less than 3, we will not calculate the rms value at this grid point. The spatial distribution of ambient noise intensity in the whole study area of the day can be obtained by calculating the rms value for all grid points, namely the daily spatial distribution map of ambient noise intensity (Fig. 5). Taking the spatial distribution map of ambient noise intensity on January 5, 2010 as an example, the red color in the map indicates high ambient noise intensity, blue is low ambient noise intensity and white indicates no data is given. Due to the sparse distribution of stations in the Qinghai-Tibetan Plateau, the rms value at many grid points does not meet the conditions, we therefore can't obtain the distribution of ambient noise intensity at these grid points. The daily distribution of ambient noise intensity is very unstable, and corresponding ambient noise intensity distribution on a different date may differ significantly, which also can be seen in the temporal variation diagram of ambient noise intensity.

 Fig. 5 Example map of spatial distribution of ambient noise intensity for one day (January 5, 2010) across the Chinese Mainland The rms amplitude is equivalent to the square root of station autocorrelation. It represents azimuthally averaged noise intensity. The unit of the rms value is μm

In order to obtain a relatively stable distribution of ambient noise intensity and to observe the distribution characteristics of ambient noise intensity in the Chinese Mainland, the superposition of the distribution of ambient noise intensity is done by seasons. Each of the subgraphs in Fig. 6 represents the distribution of ambient noise intensity in every single quarter. We see that the intensity of ambient noise varies significantly in different quarters. There are also significant differences in the spatial and amplitude distribution of noise sources in different quarters. Among them, the distribution of ambient noise intensity is similar in the second and third quarter of 2010 and the second quarter of 2011, the overall shows characteristics of higher intensity along the southeast coast, including the Hainan region, which has the highest ambient noise intensity. The distribution of noise source in the first quarter of 2010 is similar to that in the fourth quarter of 2010, showing the characteristics of being strongest along the southeast coast, weakening towards the inland areas, weakest in the Qinghai-Tibetan Plateau, and stronger again in the Xinjiang region. The distribution of ambient noise intensity in the first quarter of 2011 is very different from that in other quarters. In terms of amplitude, the ambient noise intensity in the first quarter of 2011 is the strongest, with a rms value higher than 40 in large areas, which is significantly higher than in other quarters. In terms of spatial distribution, it also presents the characteristics of strongest along the southeast coast and in the southwest and weakest in inland areas, especially in the Qinghai-Tibetan Plateau. It can be seen from the superposition of spatial distribution of ambient noise intensity in the period from January, 2010 to June, 2011 (Fig. 7) that, in this period of time, the distribution of ambient noise intensity presents obvious zoning characteristics, among which, ambient noise intensity is the strongest along the southeast coast, gradually weakened in the NW direction, reaching the weakest along the Qinghai-Tibetan Plateau and a certain degree of rebound appears in the Xinjiang region.

 Fig. 6 Maps of spatial distribution of ambient noise intensity (RMS amplitude) in the Chinese Mainland by seasons over the 1.5 years (from January 2010 through June 2011)

 Fig. 7 Map of spatial distribution of ambient noise intensity (rms amplitude) in the Chinese Mainland over the whole 1.5 years
3 DISCUSSION AND CONCLUSION

According to Weaver R.L.'s (2011) theory, supposing that the two-dimensional wave field is composed of dispersive attenuation surface waves of noise source with directional smooth shift, then the amplitude of cross-correlation empirical Green's function from station i and j can be expressed as

 ${X_{i \to j}} = {S_i}{S_j}{B_i}\left( {{{\mathit{\boldsymbol{\hat n}}}_{i \to j}}} \right)\sqrt {2\pi c/\omega \left| {{{\mathord{\buildrel{\lower3pt\hbox{$\scriptscriptstyle\rightharpoonup$}} \over r} }_i} - {{\mathord{\buildrel{\lower3pt\hbox{$\scriptscriptstyle\rightharpoonup$}} \over r} }_j}} \right|} {\rm{exp}}\left( { - \int_{{{\mathord{\buildrel{\lower3pt\hbox{$\scriptscriptstyle\rightharpoonup$}} \over r} }_i}}^{{{\mathord{\buildrel{\lower3pt\hbox{$\scriptscriptstyle\rightharpoonup$}} \over r} }_j}} {\alpha {\rm{d}}x} } \right)$ (1)

Where, Si and Sj indicate site response of station i and j, and c, ω and α are respectively velocity, corner frequency and attenuation factor of Rayleigh wave. ${{{\mathord{\buildrel{\lower3pt\hbox{$\scriptscriptstyle\rightharpoonup$}} \over r} }_i}}$ represents the distance between empirical Green's function and station i. ${B_i}\left( {{{\mathit{\boldsymbol{\hat n}}}_{i \to j}}} \right)$ indicates the intensity of ambient noise field at station i in the direction from i to j (Fig. 8), which is the key parameter to solve equation (1). Because the positions of two stations are different, the noise intensities at their locations also differ, therefore, the amplitude of surface wave obtained by the same path is often asymmetric. The attenuation structure in the study area can be extracted from cross-correlation of ambient noise by solving equation (1). In this article, azimuthal average value of ${B_i}\left( {{{\mathit{\boldsymbol{\hat n}}}_{i \to j}}} \right)$ in corresponding equation (1) of intensity of ambient noise field at the position of each station is calculated, which can provide important reference data for the calculation of attenuation structure.

 Fig. 8 Amplitudes of the Green functions retrieved from the noise sources in different directions

In order to verify the reasonability of the temporal and spatial distribution of ambient noise intensity, we conduct experiments from different angles. Firstly, we select several paths for station pairs in different regions and directions in the study area to verify relative sizes of ambient noise intensity at the same stations in different directions. We acquire empirical Green's function for these different paths based on Weaver R.L.'s (2005) and Shapiro N.M. et al's (2004b) theories and Bensen, Ritzwoller, Barmin, Levshin, Lin, Moschetti, Shapiro and Yang's (2007) method of retrieving empirical Green's function from a pair of seismic stations (Fig. 9). Because we adopt non-linear operations such as one-bit in data processing, the obtained amplitude of empirical Green's function cannot reflect real path attenuation, but the relative amplitude of the empirical Green's function retrieved from the same pair of station can reflect the relative sizes of intensity of ambient noise from different directions. Taking the station pair between station FJQZH and LNYKO as an example, the size of amplitude of empirical Green's function in the negative axis reflects the intensity of ambient noise propagating from the position of station FJQZH to the LNYKO station, while the size of amplitude of empirical Green's function in the positive axis reflects the intensity of ambient noise propagating from the position of station LNYKO to the FJQZH station. Because there are often massive strong noise sources such as typhoon along the southeast coast, it can be concluded that ambient noise intensity at the FJQZH station at the southeast coast is stronger than that of LNYKO station located in the Bohai Sea, and this feature is also revealed by the relative amplitude of empirical Green's function. Similar laws can be found at other stations, therefore, this experiment can verify to some extent the reasonability of ambient noise intensity we obtained.

 Fig. 9 Amplitudes of the Green functions with different propagating directions

Ten seismic stations of CDSN are used in this study, which are equipped with instruments with bandwidth mostly over 120s, with long running process and good stability. And compared to stations of regional networks, these stations have better stylobate and maintenance conditions. To verify the accuracy of the calculation results obtained from regional networks in this study, we select two stations at the same location. One is station CDSN, and the other is a station of the regional network. Taking station BJT as an example, we compare the curves of temporal variation of ambient noise intensity respectively obtained from CDSN station and the regional network (Fig. 10). It is worth noting that the influence of site effect is not removed from the rms value obtained in our study, therefore, different sites have a certain influence on the rms value, and the site amplification effect is smaller for stations with better stylobate conditions, thus a lower rms value can be obtained from the station with better stylobate at the same location. In this study, we do not consider the site amplification effect because we're unable to accurately know the amplification factor of stylobate in advance. By comparison, it can be found that curves obtained from two calculation results are basically the same, but the rms value obtained from the regional network is slightly smaller than that from station CDSN. This may because different stations are influenced by different site effects, but the differences between them are not large, and the final image of spatial and temporal distribution of ambient noise intensity are not influenced much. Therefore, we believe that ambient noise intensity obtained by use of successive waveforms from the regional seismic network is also reliable.

 Fig. 10 Temporal curves of ambient noise intensity in regional stations (LBP, LLM) and station CDSN (BJT) Blue curves are temporal curves of ambient noise intensity obtained from CDSN station (BJT), and the red and green curves are temporal curves of ambient noise intensity obtained from regional stations (LBP and LLM)

 Fig. 11 Examples of the temporal variations of ambient noise intensity obtained by adjacent stations in the GD station network and the BJ station network separately

In theory, ambient noise intensity obtained from different stations at nearby locations should be similar.To verify the results of this study, curves of temporal variation of ambient noise intensity obtained from 3 adjacent stations in the Guangdong Digital Seismic Network and 5 adjacent stations in the Beijing Digital Seismic Network are compared to verify the reliability and stability of the results (Fig. 1). It can be seen that the temporal variations of ambient noise intensity obtained from the three stations in Guangdong are basically consistent, and curves of temporal variations of ambient noise intensity obtained from the five stations of the Beijing Digital Seismic Network are much closer, indicating that distribution of ambient noise intensity obtained from different stations at nearby locations is stable and reliable and the overall differences in the amplitude at some stations may be caused by different site effects. It is worth noting that the temporal variation of ambient noise intensity obtained from the three stations of the Guangdong Digital Seismic Network shows a certain periodic characteristics, of which ambient noise intensity was strongest in October, 2010, with the largest rms value being approximately 450, far higher than the ambient noise intensity in other months. Temporal variation of ambient noise intensity obtained from five stations of the Beijing Digital Seismic Network also shows a certain periodicity, however, the periodic features differ significantly from the temporal variation of ambient noise intensity obtained from the Guangdong Digital Seismic Network. The ambient noise intensity is higher overall from September, 2010 to March, 2011. Although the ambient noise intensity has ups and downs during the period, the overall ambient noise intensity is higher than in other periods of time. Compared with the Guangdong Digital Seismic Network, ambient noise intensity obtained from the Beijing Digital Seismic Network is much lower than that from the Guangdong Digital Seismic Network. The likely reason is that Guangdong, which is closer to the ocean, is more affected by strong noise from the sea.

 Fig. 1 Distribution of seismic stations (+) used in this study

The image of temporal variation of noise intensity shows (Fig. 4) that there are two prominent peaks in October to November, 2010 and March to April, 2011, with corresponding maximum rms value of about 350, which is significantly higher than the rms value in other months. Ambient noise intensity is relatively higher during March to April, 2011, which is likely to be related to the Japan MW9.0 earthquake on March 11, 2011 and its prolonged strong aftershocks. The earthquake on March 11 struck near Honshu, Japan, and the earthquake caused the Japan ocean line to move eastward by 5m and raised the over 15, 000km2 sea surface by 5m (Lay T. et al, 2011). The rupture extended about 200km along the dip angle, and displacement near the epicenter reached about 25m (Lay T. et al., 2013). The earthquake triggered enormous tsunami that swept hundreds of kilometers of ocean lines. The height of the tsunami reached 3m-15m and seeped into inland areas by 10m (Lay T. et al., 2011). The earthquake triggered prolonged high-intensity aftershocks, the largest with a magnitude of 7.9 (Nettles M. et al., 2011), and the length of aftershock belt along the subduction zone was more than 500km (Lay T. et al., 2013). Seismicity was increasing in nearly all regions of Japan (Hirose F. et al., 2011). A large number of aftershocks could also generate stronger ambient noise, which also resulted in the enhancement of ambient noise intensity in the Chinese Mainland. The spatial variation of ambient noise intensity (Fig. 6) shows that ambient noise tends to present distinct differences in different seasons. Yang Yingjie et al. (2008) also found that noise sources in Europe also showed seasonal changes. Noise intensity in winter is much higher than in summer, and seasonal variation of noise intensity and noise level in winter are also significantly stronger than in summer, which is consistent with the stronger ocean conditions in winter (Webb S.C., 1998). In this study, it is also found that ambient noise intensity in winter (October, 2010-March, 2011 and January, 2010-March, 2010) is much higher than in summer (April, 2010-September, 2010 and April, 2011-June, 2011) in the Chinese Mainland.

According to the spatial distribution characteristics of ambient noise intensity from January, 2010 to June, 2011, we divide the study area into four zones from southeast coast to inland areas, and examine the temporal variation of ambient noise intensity for all stations in corresponding subareas in this period. Curves of temporal variation of ambient noise intensity (Fig. 12) show that ambient noise intensity is higher overall in October, 2010 and pril, 2011 and the lowest from April, 2010 to September, 2010. Among them, curves of temporal variation of ambient noise intensity in region 1 and region 2 are most consistent, but the average rms value in region 1 is 38.47, which is the highest of the four zones. The average rms value in region 2 is 31.22, which is significantly lower than in region 1. Zoning features of noise intensity reveal that ambient sea noise in the microseismic frequency band is the main source of noise, and the closer to the sea, the stronger the noise intensity tends to be. Curves of temporal variation of ambient noise intensity in region 3 and region 4 differ very much from curves of region 1 and region 2, suggesting that the source of noises in inland areas may differ evidently from that in coastal regions. Sea noise from the Southeast China Coast may have less impact on inland regions and may not be the main source of noises in inland areas. The average rms value in region 3 is the lowest, which is 24.27, while it increases to 28.78 in region 4. The ambient noise intensity in the Xinjiang region corresponding to region 4 is obviously higher than that in the Qinghai-Tibetan Plateau corresponding to region 3. From the perspective of study periods, the distribution of strong noise intensity in Xinjiang region presents seasonal characteristics. Especially from January to March of each year, such as from January to March, 2010 and January to March, 2011, the presence of dramatically high intensity distribution is observed, and in other months, this feature is not obvious (Fig. 6). However, the source of noise in inland regions is very complex and presently we do not have sufficient evidence to infer the origin of the noise source with seasonal characteristics.

 Fig. 12 Temporal variations of ambient noise intensity in different zones in the study area

Fourier transform is implemented to the obtained rms temporal variation data in a one-hour time window in Fig. 4, and we get rms amplitude spectrum (Fig. 13). We see that as the period increases, the amplitude of rms shows an increasing trend, especially at the period of 50 hours (about 2 days), and the amplitude is greatly increased, indicating that the energy of ambient noise intensity is mainly concentrated in a period of over 2 days, in other words, the energy of noise source usually changes a little for about 2 days on average. Therefore, the selection of the time window for the calculation of noise intensity should be no more than 2 days, so as not to filter out a noise source with strong energy.

 Fig. 13 Spectrum of the ambient noise intensity

The vertical component of successive waveforms recorded by 888 broadband seismic stations of China National Digital Seismic Network and regional seismic networks are used to investigate the temporal and spatial distribution of ambient noise intensity from January, 2010 to June, 2011, and by the curves of temporal distribution of ambient noise intensity, we find that ambient noise intensity shows some seasonal and periodic characteristics. By the stack of ambient noise intensity in different seasons, it can be found that ambient noise intensity is highest from January, 2011 to March, 2011, several times the ambient noise intensity of other periods of time. The image of spatial distribution of ambient noise intensity shows that ambient noise intensity presents obvious zoning characteristics in the Chinese Mainland. The strength in southeastern coastal areas achieves the maximum, which may be related to the ocean tides, generally decreases toward the inland areas, and arrives at the minimum in the Qinghai-Tibetan Plateau. The zonal intensity distribution is probably correlated with ocean tides from the Philippine Sea and the Pacific Ocean. It also shows that the influence from the Indian Ocean seems small. However, the ambient noise intensity increases to a certain degree in the Xinjiang area, indicating that the main source of ambient noise in the western area of the Chinese Mainland is not derived from the East and South China Sea, but rather from the deep interior of the Eurasian continent. The ambient noise intensity obtained in this study can supply some fundamental data maps for the research of microseismic intensity in many regions. Moreover, it can supply some reference for seismology research based on ambient noise correlation. For example, the distribution of ambient noise intensity can explain why the amplitude of surface wave is often asymmetrical on the positive and negative half axis in the cross-correlation of Green's function and why the surface wave is very strong on a half axis and almost completely undeveloped on the other half axis. Moreover, the distribution of ambient noise intensity can supply basic data for the extraction of surface wave attenuation based on ambient noise, accurately deduct ambient noise intensity around all stations, and achieve the object of retrieving surface wave attenuation from ambient noise.

It is worth noticing that the variation characteristics of ambient noise obtained in this study are based on a period of 10s. Ambient noise intensity at different cycles may vary greatly and have different laws in temporal variations (Bensen G.D. et al., 2007), therefore, when studying surface wave attenuation of ambient noise at other periods, it is necessary to calculate the distribution of ambient noise intensity at the corresponding period. Also, we collected data of successive waveforms from January, 2010 to June, 2011 for over 1.5 years, and the time period of ambient noise intensity may be longer than the total time of our study, therefore, successive waveforms over a longer period of time are needed in order to fully study its periodicity, and this is also one direction of our further research.

ACKNOWLEDGEMENTS

Many thanks to research professor Zheng Xiufen from the State Earthquake Information Service-Data Management Center, who provided successive waveforms for our study (2009). The completion of the work involved a beneficial discussion with researcher Richard Weaver from the University of Illinois, USA, and here I express my sincere gratitude. In addition, thanks to the anonymous reviewers for their valuable suggestions.

This paper has been published in Chinese in the journal of Earthquake, Volume 37, Number 2, 2017.

REFERENCES
 Bao Xuewei, Song Xiaodong, Li Jiangtao. High-resolution lithospheric structure beneath Mainland China from ambient noise and earthquake surface-wave tomography[J]. Earth and Planetary Science Letters, 2015, 417: 132–141. DOI:10.1016/j.epsl.2015.02.024. Bensen G.D., Ritzwoller M.H., Barmin M.P., Levshin A.L., Lin F., Moschetti M.P., Shapiro N.M., Yang Y. Processing seismic ambient noise data to obtain reliable broad-band surface wave dispersion measurements[J]. Geophysical Journal International, 2007, 169(3): 1239–1260. DOI:10.1111/gji.2007.169.issue-3. Bormann P., Engdahl B., Kind R. Seismic Signals and Noise[M]. Potsdam, Germany: GeoForschungs Zentrum, 2009 Brenguier F., Campillo M., Hadziioannou C., Shapiro N.M., Nadeau R.M., Larose E. Postseismic relaxation along the San Andreas fault at Parkfield from continuous seismological observations[J]. Science, 2008a, 321(5895): 1478–1481. DOI:10.1126/science.1160943. Brenguier F., Shapiro N.M., Campillo M., Ferrazzini V., Duputel Z., Coutant O., Nercessian A. Towards forecasting volcanic eruptions using seismic noise[J]. Nature Geoscience, 2008b, 1(2): 126–130. DOI:10.1038/ngeo104. Cupillard P., Capdeville Y. On the amplitude of surface waves obtained by noise correlation and the capability to recover the attenuation:a numerical approach[J]. Geophysical Journal International, 2010, 181(3): 1687–1700. Cupillard P., Stehly L., Romanowicz B. The one-bit noise correlation:a theory based on the concepts of coherent and incoherent noise[J]. Geophysical Journal International, 2011, 184(3): 1397–1414. DOI:10.1111/gji.2011.184.issue-3. Ekström G. Time domain analysis of Earth's long-period background seismic radiation[J]. Journal of Geophysical Research, 2001, 106(B11): 26483–26493. DOI:10.1029/2000JB000086. Fang Lihua, Wu Jianping, Ding Zhifeng, Panza G.F. High resolution Rayleigh wave group velocity tomography in North China from ambient seismic noise[J]. Geophysical Journal International, 2010, 181(2): 1171–1182. Hirose F., Miyaoka K., Hayashimoto N., Yamazaki T., Nakamura M. Outline of the 2011 off the Pacific coast of Tohoku Earthquake (MW9.0)-Seismicity:foreshocks, mainshock, aftershocks, and induced activity[J]. Earth, Planets and Space, 2011, 63(7): 513–518. DOI:10.5047/eps.2011.05.019. Lawrence J.F., Prieto G.A. Attenuation tomography of the western United States from ambient seismic noise[J]. Journal of Geophysical Research, 2011, 116(B6): B06302. Lay T., Kanamori H. Insights from the great 2011 Japan earthquake[J]. Physics Today, 2011, 64(12): 33–39. DOI:10.1063/PT.3.1361. Lay T., Fujii Y., Geist E., Koketsu K., Rubinstein J., Sagiya T., Simons M. Introduction to the special issue on the 2011 Tohoku Earthquake and Tsunami[J]. Bulletin of the Seismological Society of America, 2013, 103(2B): 1165–1170. DOI:10.1785/0120130001. Lee W. S., Lee J., Han S. Seismic Ambient Noise and Its Applicability to Monitor Cryospheric Environment[M]. [S. l. ]: INTECH Open Access Publisher, 2013. Lin Fanchi, Moschetti M.P., Ritzwoller M.H. Surface wave tomography of the western United States from ambient seismic noise:Rayleigh and Love wave phase velocity maps[J]. Geophysical Journal International, 2008, 173(1): 281–298. DOI:10.1111/gji.2008.173.issue-1. Lin Fanchi, Ritzwoller M.H., Yang Yingjie, Moschetti M.P., Fouch M.J. Complex and variable crustal and uppermost mantle seismic anisotropy in the western United States[J]. Nature Geoscience, 2011a, 4(1): 55–61. DOI:10.1038/ngeo1036. Lin Fanchi, Ritzwoller M.H., Shen Weisen. On the reliability of attenuation measurements from ambient noise cross-correlations[J]. Geophysical Research Letters, 2011b, 38(11): L11303. Nawa K., Suda N., Fukao Y., Sato T., Aoyama Y., Shibuya K. Incessant excitation of the Earth's free oscillations[J]. Earth, Planets and Space, 1998, 50(1): 3–8. DOI:10.1186/BF03352080. Nettles M., Ekström G., Koss H.C. Centroid-moment-tensor analysis of the 2011 off the Pacific coast of Tohoku Earthquake and its larger foreshocks and aftershocks[J]. Earth, Planets and Space, 2011, 63(7): 519–523. DOI:10.5047/eps.2011.06.009. Peterson J.R. Observations and Modeling of Seismic Background Noise[M]. Washington DC: U.S. Geological Survey, 1993 Prieto G.A., Beroza G.C. Earthquake ground motion prediction using the ambient seismic field[J]. Geophysical Research Letters, 2008, 35(14): L14304. DOI:10.1029/2008GL034428. Prieto G.A., Lawrence J.F., Beroza G.C. Anelastic Earth structure from the coherency of the ambient seismic field[J]. Journal of Geophysical Research, 2009, 114(B7): B07303. Prieto G.A., Denolle M., Lawrence J.F., Beroza G.C. On amplitude information carried by the ambient seismic field[J]. Comptes Rendus Geoscience, 2011, 343(8/9): 600–614. Rhie J., Romanowicz B. Excitation of Earth's continuous free oscillations by atmosphere-ocean-seafloor coupling[J]. Nature, 2004, 431(7008): 552–556. DOI:10.1038/nature02942. Rhie J., Romanowicz B. A study of the relation between ocean storms and the Earth's hum[J]. Geochemistry, Geophysics, Geosystems, 2006, 7(10): Q10004. Shapiro N.M., Campillo M. Emergence of broadband Rayleigh waves from correlations of the ambient seismic noise[J]. Geophysical Reserach Letters, 2004a, 31(7): L07614. Shapiro N.M., Ritzwoller M.H. Inferring surface heat flux distributions guided by a global seismic model:particular application to Antarctica[J]. Earth and Planetary Science Letters, 2004b, 223(1/2): 213–224. Shapiro N.M., Campillo M., Stehly L., Ritzwoller M.H. High-resolution surface-wave tomography from ambient seismic noise[J]. Science, 2005, 307(5715): 1615–1618. DOI:10.1126/science.1108339. Sorrells G.G. A preliminary investigation into the relationship between long-period seismic noise and local fluctuations in the atmospheric pressure field[J]. Geophysical Journal International, 1971, 26(1/4): 71–82. Sorrells G.G., Mcdonald J.A., Der Z.A., Herrin E. Earth motion caused by local atmospheric pressure changes[J]. Geophysical Journal International, 1971, 26(1/4): 83–98. Sun Xinlei, Song Xiaodong, Zheng Sihua, Yang Yingjie, Ritzwoller M.H. Three dimensional shear wave velocity structure of the crust and upper mantle beneath China from ambient noise surface wave tomography[J]. Earthquake Science, 2010, 23(5): 449–463. DOI:10.1007/s11589-010-0744-4. Tang Xiaoyong, Fan Wenyuan, Feng Yongge, Tang Youcai, Chen Y.J., Zhu Lixia. Phase velocity tomography of Rayleigh wave in Xinjiang from ambient noise[J]. Chinese Journal of Geophysics, 2011, 54(8): 2042–2049. Tang Youcai, Chen Yongshun, Yang Yingjie, Ding Zhifeng, Liu Ruifeng, Feng Yongge, Li Peng, Yu Chunquan, Wei Songqiao, Fan Wenyuan, Wang Haiyang, Zhou Shiyong, Ning Jieyuan. Ambient noise tomography in north China craton[J]. Chinese Journal of Geophysics, 2011, 54(8): 2011–2022. Tanimoto T. Excitation of normal modes by atmospheric turbulence:source of long-period seismic noise[J]. Geophysical Journal International, 1999, 136(2): 395–402. DOI:10.1046/j.1365-246x.1999.00763.x. Tanimoto T. The oceanic excitation hypothesis for the continuous oscillations of the Earth[J]. Geophysical Journal International, 2005, 160(1): 276–288. Tsai V.C. On establishing the accuracy of noise tomography travel-time measurements in a realistic medium[J]. Geophysical Journal International, 2009, 178(3): 1555–1564. DOI:10.1111/gji.2009.178.issue-3. Tsai V.C. Understanding the amplitudes of noise correlation measurements[J]. Journal of Geophysical Research, 2011, 116(B9): B09311. Wang Weilai, Wu Jianping, Fang Lihua, Lai Guijuan, Yang Ting, Cai Yan. S wave velocity structure in southwest China from surface wave tomography and receiver functions[J]. Journal of Geophysical Research, 2014, 119(2): 1061–1078. Weaver R.L. Information from seismic noise[J]. Science, 2005, 307(5715): 1568–1569. DOI:10.1126/science.1109834. Weaver R.L. On the amplitudes of correlations and the inference of attenuations, specific intensities and site factors from ambient noise[J]. Comptes Rendus Geoscience, 2011, 343(8/9): 615–622. Webb S.C. Broadband seismology and noise under the ocean[J]. Reviews of Geophysics, 1998, 36(1): 105–142. DOI:10.1029/97RG02287. Xu Z.J., Song Xiaodong, Zheng Sihua. Shear velocity structure of crust and uppermost mantle in China from surface wave tomography using ambient noise and earthquake data[J]. Earthquake Science, 2013, 26(5): 267–281. DOI:10.1007/s11589-013-0010-7. Yang Yingjie, Ritzwoller M.H. Characteristics of ambient seismic noise as a source for surface wave tomography[J]. Geochemistry, Geophysics, Geosystems, 2008, 9(2): Q02008. Yang Yingjie, Zheng Yong, Chen J., Zhou Shiyong, Celyan S., Sandvol E., Tilmann F., Priestley K., Hearn T.M., Ni J.F., Brown L.D., Ritzwoller M.H. Rayleigh wave phase velocity maps of Tibet and the surrounding regions from ambient seismic noise tomography[J]. Geochemistry, Geophysics, Geosystems, 2010, 11(8): Q08010. Yao Huajian, Van Der Hilst R.D., de Hoop M.V. Surface-wave array tomography in SE Tibet from ambient seismic noise and two-station analysis-I. Phase velocity maps[J]. Geophysical Journal International, 2006, 166(2): 732–744. DOI:10.1111/gji.2006.166.issue-2. Yao Huajian, Van Der Hilst R.D. Analysis of ambient noise energy distribution and phase velocity bias in ambient noise tomography, with application to SE Tibet[J]. Geophysical Journal International, 2009, 179(2): 1113–1132. DOI:10.1111/gji.2009.179.issue-2. Zheng Sihua, Sun Xinlei, Song Xiaodong, Yang Yingjie, Ritzwoller M.H. Surface wave tomography of China from ambient seismic noise correlation[J]. Geochemistry, Geophysics, Geosystems, 2008, 9(5): Q05020. Zheng Xian, Zhao Cuiping, Zhou Lianqing, Zheng Sihua. Rayleigh wave tomography from ambient noise in Central and Eastern Chinese Mainland[J]. Chinese Journal of Geophysics, 2012, 55(6): 1919–1928. Zheng Xian, Zhao Cuiping, Zhou Lianqing, Zheng Sihua. 3D shear-wave velocity structure beneath the southeastern Tibetan Plateau from ambient noise[J]. Bulletin of the Seismological Society of America, 2015, 105(3): 1371–1382. DOI:10.1785/0120140211. Zheng Xiufen, Ouyang Biao, Zhang Dongning, Yao Zhixiang, Liang Jianhong, Zheng Jie. Technical system construction of Data Backup Centre for China Seismograph Network and the data support to research on the Wenchuan earthquake[J]. Chinese Journal of Geophysics, 2009, 52(5): 1412–1417. Zheng Yong, Shen Weisen, Zhou Longquan, Yang Yingjie, Xie Zujun, Ritzwoller M.H. Crust and uppermost mantle beneath the North China Craton, northeastern China, and the Sea of Japan from ambient noise tomography[J]. Journal of Geophysical Research, 2011, 116(B12): B12312. DOI:10.1029/2011JB008637. Zhou Longquan, Xie Jiayi, Shen Weisen, Zheng Yong, Yang Yingjie, Shi Haixia, Ritzwoller M.H. The structure of the crust and uppermost mantle beneath South China from ambient noise and earthquake tomography[J]. Geophysical Journal International, 2012, 189(3): 1565–1583. DOI:10.1111/gji.2012.189.issue-3.