Analysis of seismic noise of broadband seismological stations installed along the Western Ghats

The Western Ghats (WG) is one of the great escarpments that extend ~1500 km parallel to the west coast of India in the direction of NNW-SSE. We deployed a network of seven broadband seismological stations along the WG to decipher its evolution. In the present study, we investigate the characteristics of different kinds of noises at the stations by utilizing the power spectral density (PSD) measurements. Further, the results are compared with the global standard noise models to assess the data quality. The PSD results reveal that the short period (0.1–1 s) cultural noise is more prominent at stations AGMB, SDPR and MGLI when compared to the other stations, especially during day hours of Indian Standard Time (IST) zone since these sites were in the proximity of roads. The seasonal variations are observed especially in the microseismic period range and noise levels are more prominent in the months of July to August since the western part of India experiences peak monsoon during this period. These variations are observed especially at PCH and KNUR stations as their locations were near the coastline. Furthermore, the results indicate that the noise levels are more prominent in vertical components than that of horizontal components in the microseismic period range whereas it is reversed in the short and long period ranges. Later, the results indicate that the noise levels at all the localities of stations are within the global standard noise models, which suggests that our installation of broadband seismological stations has been successful at all seven locations and has good data quality.


Introduction
A seismogram contains signals generated by the earth's vibrations, including those from natural (e.g., earthquakes) and/or anthropogenic sources. Seismic noise is a continuous vibration of ground owing to a multitude of causes, it varies according to the frequencies of the vibration and manifests with temporal and spatial changes. Seasonal weather changes, daynight variations and local conditions of the area affect the spectral patterns of the seismic data. The noise hinders the uncovering of useful information hidden in a signal. The ambient noise is a composition of various frequency surface waves and can be natural and/or anthropogenic. The most common natural sources could be wind, rain, rivers, ocean microseisms and earthquakes Webb and Lee 2002;McNamara and Buland 2004;Wilson et al. 2002;Elouai et al. 2016;Beauduin et al. 1996;Zürn et al. 1995), which may differ at different localities. Moreover, anthropogenic noise sources include human activity, industrial sources, vehicular traffic, etc. The noise baseline is determined with respect to the geographical locations since it may vary from site to site, type of sensor and installation stage. A signal is considered to be optimal if the noise content is minimal, which leads to better detection of useful information like earthquake detection. The sampling rate and majorly the type of sensor (long period or short period or broadband) installed are the key factors in determining the minimum and maximum observable periods in the recorded data. Noise sources in the local neighbourhood, i.e., within a radius of 15 km to 60 km (Willmore 1979;Webb and Lee 2002;Bormann and Wielandt 2013), determine the noise level at each period (Webb and Lee 2002) and its analysis can help to evaluate the site conditions in terms of data quality at a certain station as well as predominant noise sources recorded at that station (de la Torre and Sheehan 2005; Abd e-l Aal 2013).
The noise spectrum is classified into three frequency bands as per McNamara et al. (2009) and McNamara and Buland (2004). The frequency within the range of 0.01 Hz to 0.1 Hz is defined as longperiod noise. The frequency band of 0.1 Hz to 1 Hz fall in the microseismic period range and that of 1 Hz to 10 Hz or higher is classified as short-period noise (Uthaman et al. 2022). Further, this can be classified as short-period noise (0.1-1 s), short-period microseism (1-4 s), secondary microseism (4-10 s), primary microseism (10-16 s) and other long-period noise (16-100 s) (Rastin et al. 2012). The long-period noise is mainly caused by atmospheric effects like wind, storms, tilt and pressure, which can affect seismic mass similar to the ground acceleration (Rastin et al. 2012;Wielandt 2002). Majorly, the long-period recordings of seismometers are more sensitive to atmospheric pressure variations (Alejandro et al. 2020;Kroner et al. 2005;Steffen H., 2006). The noise in this range usually affects the horizontal components of the seismometer than the vertical component. In general, it is accepted that the tilts generated by atmospheric pressure and wind create the predominant horizontal noise in a long-period range (Zürn et al. 2007;Webb and Lee 2002;Wielandt 2002). Furthermore, the seismic noise is more predominant in the microseismic period range. The noise levels in this range both in the primary and secondary microseismic range could be due to the interaction of ocean waves with the coast. The primary microseism, also called the single frequency microseism (0.05-0.1 Hz) is generated by the interaction between the ocean waves and the topography of the seafloor (Huang et al. 2022;Ardhuin et al. 2015;Hasselmann 1963). Whereas the secondary microseism, also called the double frequency microseism (0.1-0.5 Hz) is generated by the oppositely propagating swells non-linear interaction with a similar wavelength and the resulting standing wave generates an impulse of pressure impinges the seafloor with ocean waves of double frequency (Huang et al. 2022;Bromirski 2009;Cessaro 1994;Hasselmann 1963;Longuet-Higgins and Jeffreys 1950). The microseism may affect the coastal sites more than the continental sites (McNamara and Buland 2004), which are affected mainly by cultural noise. Also, the local weather conditions affect the noise of over 1 Hz (Peterson 1993;Webb 1988). The noise generated by anthropogenic sources (manmade) is usually referred to as cultural noises, it falls within the short-period range (>1 Hz). Its characteristics include propagation as high-frequency surface waves, attenuation in the small distance range and less depth (McNamara and Buland 2004). There is a significant difference in day and night noise levels in this category and it has characteristic frequencies depending on the source of the disturbance. Another type of locally generated noise is by the wind and swinging of towers or masts Withers et al. 1996). Noise generated by objects moved by wind, falls in the high-frequency range, while those by swinging objects generate low-frequency signals. These are different from cultural noise as the latter is generally periodic with diurnal variability, thus allows to be isolated from wind noise (Ringdal and Bungum 1977;McNamara and Buland 2004). The running water, surf and volcanic activity can also contribute to the seismic noise and the temperature effects also produce noise in the seismometer recordings (Alejandro et al. 2020;Jana et al. 2017;Rastin et al. 2012;Stutzmann et al. 2000). The ground fluctuations generated by the heating during the day and cooling during the night time can induce the tilt and long-period noise (0.01 to 0.05 Hz) in horizontal components (Stutzmann et al. 2000;McNamara and Buland 2004). A comparison of horizontal and vertical noise models can decipher these kinds of noises. In addition, seismometer recordings also contain the noise generated by electrical and mechanical instrument noise (Bormann 2002;Wielandt 2012).
As a part of the study to decipher the Western Ghats (WG) evolution, National Centre for Earth Science Studies (NCESS) deployed a network of seven broadband seismological stations along the WG, details are shown in Table 1 and Fig. 1. Most of the stations are in remote locations and close to the coast. The stations are equipped with RefTek 151B Observer sensor with RefTek DAS except for the PCH station, which is equipped with Trillium 240 instrument. All the stations are recording continuous data at 100 samples per second. The primary objective of this experiment is to decipher the lithospheric structure and mantle deformation along the WG to shed light on its evolution. The WG is one of the great escarpments that extend ~1500 km parallel to the west coast of India in the NNW-SSE direction with an elevation of ~1.2 km. The linear extent of these Ghats begins from Gujarat in the north, south of the Narmada rift and ends at Kanyakumari in the south. The low altitude coastal plains banked by the Arabian sea are on the west side of the WG and the elevated plateau of WG comprising of Southern Granulite Terrain (SGT) in the south, Western Dharwar Craton (WDC) at the centre and Deccan Volcanic Province (DVP) in the northern part. Out of our seven stations, two stations were installed in the region of SGT and four were installed in the WDC region and one station is in the region of DVP. To achieve our primary objective in this experiment, it is essential to run the stations with proper functionality. Thus, in the present study, we analysed the station performance by analysing noise levels at all the stations (Table 1 and Fig. 1) and compared them to the standard noise models.

Estimation of power spectral density
To analyse the seismic noise, recorded at broadband seismological stations established along the WG, we utilized the power spectral density (PSD) and the probability density function (PDF) of each category of background noise. This is the standard method for quantifying seismic background noise. The PSD can be obtained by means of averaging as seismograph recordings are stochastic signals. In the present study, we utilized an open-source software PQLX, which was developed by USGS for evaluating the seismic station performance and data quality (McNamara et al. 2009). This software requires the input waveform data and instrument response files to compute the PSD and PDFs. These measurements are computed based on the algorithm by McNamara and Buland (2004). The advantage of using this tool is that it accepts the raw data and removes the instrument response using the input response files. There is no need to remove earthquakes, system glitches or general data traces in the PQLX method since these are low-probability occurrences that do not contaminate high-probability ambient seismic noise. Moreover, Anthony et al. (2020) show that earthquakes introduce bias in PSD estimates and decrease the frequency resolution and thus likely distort the features of interest. However, the PQLX method uses the PDF, calculated from the PSD, which considers system transients into the low background and ambient RefTek 151B-120 Observer 2018-till date noise into high probability. Earthquakes are observed in the PDFs as low probability signals at short and long periods (McNamara et al. 2009;McNamara and Buland 2004). This provides the advantage of representing the true ambient noise levels rather than a simple minimum. In this tool, contiguous data can be processed without removing earthquakes and other data glitches. Further, the instrument response is removed by utilizing the input response files to produce the ground acceleration. It truncates the number of samples in time series data to the next lowest power of 2. The hour-long time series data is divided into 15 minutes segments with 75% overlap. The overlap time series data segments are used to reduce the variance in the PSD estimate (Cooley and Tukey 1965). These are processed by removing mean, longperiod trends to eliminate large-scale distortions in spectral processing. Later, tapering with 10% of the sine function (McNamara et al. 2009;McNamara and Boaz 2006) is applied to smooth and minimize the discontinuity effect between the beginning and end of the time series data and then the Fast Fourier Transform (FFT) is applied. The averaged value for these segments provides the 1-h time series PSD after deconvolving the seismometer instrument response. The PSD estimate is converted into decibels (dB) with respect to acceleration (m/s 2 ) 2 /Hz, the unit of intensity of the random vibration signal vs frequency. Later, to compute the PDF, the raw frequency distribution is constructed from the individual PSDs by binning the periods into 1/8 octave intervals and binding the power in 1 dB intervals. The process reduces the number of frequencies by a factor of 169 (McNamara and Boaz 2006). The power is averaged between a short-period (high frequency) corner (Ts) and a long-period (low frequency) corner Tl =2*Ts, with a centre period Tc = √(Ts*Tl) is the geometric mean period within the octave. The averaged power for that octave, period ranging from Ts to Tl, is stored with the centre period of the octave, Tc. Ts is incremented by 1/8 octave such that Ts = Ts*2 0.125 , to compute the average power for the next period bin. Powers are averaged within the next period range, i.e., the updated Ts to Tl and the process continues until it reaches the longest resolvable period of a given time series data. This process is repeated for every 1-h PSD estimate. Then these raw frequency bins are normalized by the total number of PSDs to estimate the PDF. The probability of a given power occurrence at a particular period is compared with the Peterson New High Noise model (NHNM) and New Low Noise model (NLNM) (Peterson 1993).

Results and discussion
We computed the PSD and PDFs for seven stations installed along the WG ( Fig. 1 and Table 1). Results reveal that the noise levels are within the limits of the NLNM and NHNM (Fig. 2). The short-period noise is more prominent at stations AGMB, SDPR and MGLI when compared to the other stations PCH, KNUR, SBMN and JODA (Fig. 2). The primary reason could be due to the station locations; i.e., the stations AGMB, SDPR and MGLI were a bit close to the road, which is having significant vehicular traffic as compared to the other stations. Further, we checked the time series data at two stations (AGMB is close to the road than KNUR) to verify it and observed that the bursts and more noise at AGMB than at KNUR (Fig. S1). Low power in a short-period refers the less cultural noise. We observed more noise levels in the microseismic period at stations PCH and KNUR as they were very close to the shoreline. Since the data is not filtered to remove earthquake signals, the body and surface wave noise can be seen in PSD/PDF data. For example, the frequency higher than 1 Hz or a period lower than 1 s, the low probability high power events can be attributed to the body waves of earthquakes. Similarly, for frequency less than 0.1 Hz or a period greater than 10 s, can be attributed to earthquake surface waves. On average, the noise levels at stations PCH, KNUR, SDPR and JODA indicate in the middle of NLNM and NHNM. However, the noise levels at stations AGMB and MGLI are closer to the NHNM in the low-period cultural noise band.
The noise levels at station SBMN are closer to the NLNM, which indicates less human-induced noise in the recordings (Fig. 2). The predominance of cultural noise at stations AGMB and MGLI can be attributed to their proximity to relatively busy roads. The 90% of noise at AGMB lies close to NHNM, especially at the period range of 0.5 to 0.9 s, similar things were observed at MGLI also. The difference between the 90th percentile and 10th percentile of the noise level (vertical component) in the month of April 2019 at SBMN is ~7 dB in the cultural noise range and similar values were observed at the AGMB station. The corresponding values at stations PCH, KNUR, SDPR, JODA and MGLI are the primary microseism. At PCH stations, the double and single frequency peaks are observed more prominently as compared to SBMN where it is less owing to being a bit farther from the coast. The effect of microseismic noise in the period range of 1 to 20 s is observed at all the stations with stability except for the seasonal variations.

Diurnal variations
We observed the variations in the noise levels over day and night (defined based on the Indian Standard Time (IST)) at all the stations, especially in the shortperiod cultural noise band (Fig. 3). The noise levels at the MGLI station are high during the day-time (~06:00 to 18:00 hrs IST) and then decrease till 03:00 hrs (Fig. 3). This could be due to human activity and the passage of vehicles from the nearby roads. Though the difference between the day and night noise levels at the JODA station is very small, the noise levels in the day-time are bit high. This could be justified by the location of the station, i.e., which is located in a forest area. At SDPR station, owing to its location nearby a local road and state highway, the noise is higher in the day-time and decreases after 20:00 hrs. The noise levels are quite strong at AGMB station over day-time and continue till mid-night and decrease for a few hours from mid-night to the early morning hours. This could be due to the fact that this station is located near the state highway, which is busy with vehicles and public transport throughout the day and even till midnight. The station SBMN is located in a habitat region and surrounded by forest, thus the noise levels are quite low and even the diurnal variations are also significantly low at this station. The location of KNUR station is in a small village and there is not much disturbance from heavy vehicles except human activities. Therefore, we observed low noise levels even over the day-time; however, the day-time noise levels are a bit high when compared to the night-time noise levels. A similar pattern in noise levels is observed at the PCH station, which is also located in a similar locality. High cultural noise levels were observed at AGMB and MGLI stations and out of which the noise levels are more prominent at AGMB station even during the several hours of the night. All the stations indicate stable noise levels with very minor diurnal variations in the period range of microseism. However, noise levels are a bit higher side at PCH and KNUR stations since these stations were very close to the shoreline. In the long period also, no significant variations are observed in the noise levels over day and night. Figure 3 shows the diurnal variations of day averaged noise PSD for three components at seven stations, which is calculated by using the four months of data (April, May, June and July 2019) at each station and the diurnal variations in a short-period (cultural noise) range is shown in Fig. 4.
We observed that the short-period noise is much stronger in the vertical component and the day-night variations seem to be smaller than that in the horizontal components, especially at AGMB. In the long period range, the horizontal components seem to be having higher noise levels than the vertical components. This could be due to the effect of thermal and tilt since long-period noises are sensitive to temperature changes and the tilt of the sensor. The longperiod noise is low in general if the sensor is insulated from the surrounding environment by using an insulating cover. The variations induced by tilt are mostly seen in the horizontal components when compared to the vertical component, corroborating with the observations from this study.

Seasonal variations
The significant seasonal variations were observed in the period range of microseism rather than the shortand long-period ranges (Figs. 5 and S2). The noise levels at the PCH station indicate higher values in the monsoon period when compared to the other seasons. Moreover, not much variation in the difference between 90th and 10th percentile over the four seasons in both horizontal and vertical components especially in short and microseismic period ranges, while this difference is more in the long-period range. However, the mode values of PSD lie closer to the NLNM in the long-period range. Similar patterns of seasonal variations were observed at KNUR station and there is an increase in noise levels in the season of monsoon, especially in the microseismic period range. The difference in 90th and 10th percentile across all

Monthly variations
To understand the monthly variations of noise levels, we analysed the PSDs of monthly data (Figs. 7 and S3). Results reveal that no significant variations in the short-period (cultural noise) range over the 12 months at all the stations. However, the noise levels in the microseismic period range were quite strong in the months of June to September. The western part of India experiences monsoon during these months with onset in the month of June. Among them, July-August seems to be the noisiest in this period range. This could be due to the strongest period of monsoon along the western parts of India. The noise  levels decay back to that of the pre-monsoon period. Majorly, the short-period microseism and secondary microseismic period ranges show significant noise variations. This could be explained by the stormy conditions in the sea causing violent crashing of waves on shores. The long-period noise remains stable over all the months though the high temperature in the months of summer and wind conditions may cause movement of trees and poles, which gets transferred to the ground as long-period vibrations, which may cause the long-period noise. The monthly variations at station KNUR are shown in Fig. 7 Fig. S3. Furthermore, the variation of PSD median values for all 12 months at station KNUR is shown in Fig. 8 and at other stations is shown in Fig. S4. We evaluated the monthly change of noise peak mode value in the period ranges of short (1-4 s), secondary (4-10 s) and primary (10-16 s) microseism (Figs. 9 and S5) for a better understanding of noise level variations over the 12 months. Results reveal that the noise levels are high in the months of July and August, especially in the secondary microseismic period range and fall down starts in the month of October, i.e., the post-monsoon period. Interestingly, we observed the shifting of peaks from June-July at the southern stations to July-August at the northern stations. This could be due to the hitting of monsoon; i.e., a monsoon hits first in the southern side of WG and it travels towards the northern side. The monthly peak PSD values and their corresponding periods in the short-period microseism show that noise levels shift towards the higher side with the advent of monsoon and gradually decline as monsoon recedes. This variation is very less in secondary microseism and lies mostly in the 4 to 5 s range. In primary microseism, monthly PSD peaks were at 10.2 s (Figs. 9 and S5).

Dominant period vs peak noise
We analysed the monthly data to decipher the dominant period in all the noise period ranges. The observed results reveal that the dominant noise period in short and secondary microseismic period ranges, shifts towards higher values of PSD during the months of monsoon. Moreover, the peak period in the shortperiod (cultural noise) ranges remains nearly the same. The dominant periods in different period ranges at KNUR station during the February, May, July and December months of 2019 with their respective PSD values are shown in Fig. 10 and at other stations are shown in Fig. S6. These months fall in the spring, summer, monsoon and winter seasons, respectively. The curve in the part (b) of the figure illustrates the PSD vs probability of occurrence at peak period over the months in the short-period (cultural noise) ranges. This reveals that during the months experiencing the monsoon (July), the probability of occurrence is scattered in PSD values, while during the other months, it is nearly focussed around −140 to −150 dB. Furthermore, we also evaluated the monthly peak values and their corresponding periods. We observed that in the short-period and secondary microseismic range, the periods corresponding to peak values shifted towards the higher side of PSD values in monsoon (Fig. 10).
A similar trend has been observed at the SDPR station as well. In the three components, the period corresponding to peak PSD shifts to the higher side during monsoon in a short-period and secondary microseism noise ranges. The noise levels in the cultural noise band remain within the same power limits, with the noise around 0.7 to 1 s dominating. At AGMB, the peak noise period shifts to the higher side in monsoon and winter. The station AGMB experiences heavy rain, especially in monsoon, while winter noise seems to be higher side because of traffic as a result of tourist influx and vehicle movement on the highway nearby. In summer, the curve narrows, which indicates only a particular period has more probability. This may be due to lesser vehicle movement from tourism due to extreme heat. At station JODA, results show not much variation across all seasons, except for the minor increase in noise levels in monsoon. It shows similarity with AGMB in summer in terms of the probability of peak noise period. Similar levels were observed at MGLI station slightly on the higher side, as a consequence of daily vehicular movement along the road nearby.
Results at the PCH station show peak noise period in primary and secondary microseism shifted to higher noise levels, while no significant variations were observed in the other noise ranges. PCH also shows less scattering of peak periods, which may indicate the dominance of particular noise levels, particularly those of the microseismic, owing to proximity to the sea.

Vertical vs horizontal noise
To understand the differences in the noise levels in vertical and horizontal components of the seismological data, we analysed the difference in seasonal noise PSD mode of vertical and horizontal components. The difference in PSD mode for two consecutive seasons for the vertical and horizontal components is shown in Fig. 11. The negative values (< 0) indicate that the noise PSD in the horizontal component is more than that of the vertical component and vice versa. Results reveal that the horizontal component noise PSD level is a bit higher when compared to the vertical component in the short-period (cultural) range. Moreover, the vertical component has higher PSD in the period range of microseism. Also, the long-period noise is dominant in horizontal components, similar to the short-period ranges. This could be due to the tilt and/or thermal effect or long-period sources such as wind, swinging of poles or trees, etc. Interestingly, the noise levels in the long-period range are slightly more in summer, this could be due to the thermal effect. The maximum difference which can be seen is slightly above 40 dB. Results are corroborating the observations of Webb and Lee (2002), indicating that the horizontal component long-period noise is ~10-30 dB higher than the vertical component at the surface installation of stations. The highest difference is observed at the JODA station, which could be due to the noise generated by the swinging of trees and local temperature conditions. The positive difference is observed at PCH and KNUR stations up to 24 s in the long-period range, which may be due to the effect of the ocean. While the difference at KNUR station is higher in summer and lower in monsoon, at PCH shows very little difference in PSD values in both seasons. At station AGMB, noise in the vertical component is greater in the cultural noise band initially, however, eventually becomes lesser than horizontal with increasing period. The probability percent of PSD for the corresponding period of peak mode values. The three graphs (c) on the bottom left illustrates the probability vs PSD at each period band peak period, while those on the bottom right (d) illustrate the probability vs PSD for cultural noise vs corresponding periods ◂ Moreover, in the short-period microseism and primary and secondary microseism, vertical component noise is higher than that in horizontal components. Horizontal noise remains higher than vertical throughout primary microseism as well as in the long-period noise range. Results at JODA and SDPR exhibit a similar trend in summer, while in monsoon its horizontal components show a bit more PSD values in the secondary microseism. Results at the MGLI station show more noise in the horizontal component than in the vertical. Results at SBMN show different behaviour as compared to other stations. It shows positive Vertical vs Horizontal differences for short-period, primary as well as secondary microseism. Moreover, in the long-period range, it also shows horizontal component noise to be dominating with a smaller difference, similar to KNUR.

Conclusions
The importance of noise analysis is to check the quality of the data and understand the variations of noise levels at different localities of the stations. Results reveal that the cultural noise at AGMB, SDPR and MGLI stations is more prominent during the day hours as per the Indian Standard Time (IST). This could be due to the station locations; i.e., these three stations are in the proximity of the road, having significant vehicular traffic as compared to other stations. Further, the seasonal variations were observed at almost all the stations, especially in the period range of microseism; however, noise levels are more significant at PCH and KNUR stations as they were very close to the shoreline. Interestingly, we observed the shifting of noise peaks from June-July at the southern stations to July-August at the northern stations. This could be due to the hitting of monsoon; i.e., a monsoon hits first in the southern side of WG and it travels towards the northern side. Later, the results indicate that the noise levels are a bit high in horizontal components than that in vertical components especially in the period ranges of short and long, whereas, in the microseismic period range, the noise levels are more prominent in the vertical component. Overall, the noise levels at all the stations are within the global standard models (NHNM and NLNM), which yields workable data quality at the stations installed along the WG.