Introduction

Distributed acoustic sensing (DAS) is a newly developed technique that transforms telecommunication fiber optic cables into linear arrays of ground-motion sensors1,2,3,4,5,6. Compared with the conventional seismometer, DAS not only make great cost saving, but also have an advantage in dense spatial sampling7,8,9. In the last three decades, DAS has continually been using in gas industry as well as vertical seismic profiling. Meanwhile, the seismology was making progress towards the ocean study10,11,12,13. There are still huge gaps in ocean study nevertheless on account of the aesthetic maintenance cost and recovery difficulty in ocean bottom seismometer (OBS) layout in deep ocean. However, it became effortless to capture the ground-motion signals when the DAS technique by using communication optical fiber cable under the sea came into sight. Above all, the extensive coverage of optical fiber around the Earth surface provides a highly operational observation platform for seismology researchers5,14,15,16. Without seabed burying operation and observation duration limit, the DAS system have a long observation duration up to several months or even one year when the general geophones can just last for the boundary of couple of months. On the other hand, dense array of meter scale and more than 100 km long-distance effective observation distance highlight its advantages with respect to traditional techniques.

The application of DAS in passive earthquake seismology, however, is still at its early stage. Last few years, several ocean DAS experiments have been carried out with the high degree of correlation with a seismograph waveform. For example, Ethan et al. (2019), analyzed the waveforms of 2018-08-19 Mw8.2 Fiji teleseismic deep earthquake recorded by the DAS array in the Southern Bight of the North shallow Sea offshore Zeebrugge, Belgium. They found that DAS waveforms show a high degree of correlation with a seismometer record17. In the MEUST-NUMerEnv project in Toulon, France, Sladen and Lamare et al. also found coherent earthquake waveforms recorded at a dense DAS array in deep ocean showing the high sensitivity as a dense geophone array from a local M1.9 event18. Two experiments show the capacity of DAS system of detecting the earthquake and even the micro earthquake. On the other hand, the longest DAS monitoring distance is up to 131 km. However, some shortages as a newcomer were exposed in most of the shallow ocean DAS experiments: (i). The precise position of each DAS station is hard to calculate. (ii). Coupling condition and seabed incline determine the quality of the DAS data. (iii). The response of DAS is weak in capturing the ground motion of P-wave due to its different vibration characteristic with S-wave. At the same time, the biggest limitation of DAS in submarine is that the shallow sea area is adverse to observe due to the noise of surface gravity waves within 0.03 ~ 0.5 Hz and Scholte waves generated by solid–liquid interface coupling motion within 0.3 ~ 2 Hz18. It not only causes the waste of communication fiber optic cables resources in shallow sea, but also leads to the low utilization of natural local seismic waveforms. So it is necessary to introduce some denoising methods to process the observation data of submarine cable DAS in shallow ocean.

In this article, we compared DAS records of 10 local earthquakes based on submarine fiber-optic cables with the records from broadband seismometers. The optical cable was early used in optical communication19 and its coupling with the seafloor was unknown. In the case, some signals enhancement methods have been taken in the data processing, such as filtering, stacking and Discrete wavelets transform (DWT) as well as Ensemble Empirical Mode Decomposition (EEMD) Transform. Four of ten records show the high-similarity to the broadband seismic waveforms with clear p and s seismic phases, and the seismic signals in another six records are ambiguous. Then, we analyze the ability of local earthquake monitoring by communication cable in submarine, as well as the potential for a long-term and inexpensive observation way in deep ocean. Due to various interference factors, shallow sea DAS has been found to have a low signal-to-noise ratio in seismic signals, making it difficult to conduct earthquake monitoring. This article compares it with broadband data and shows that shallow sea DAS can also record regional small earthquake waveforms, playing an important role in earthquake monitoring.

Methods

Observation system

The observation platform of this study is based on ISAC (Integrated Sensing and Communication) technology, in which the communication modules and sensing modules can be existed in one fiber without any interference20,21. In this way, any optical fiber could be used for sensing especially the ones under the sea whether in using for communication or not. It immensely filled a gap in the field of undersea observation in seismology. Since the 1990s, the research on fiber sensing increased yearly22, such as earthquake detection17,23,24, Illuminating seafloor faults25,26,27, etc. The distributed sensing technology has numerous advantages in seismology observation, but still need more effort to improve the signal-to-noise ratio (SNR) and application result for foundations building28. We used a section of the communication cable in Pearl River Estuary for monitoring test and select one Mw3.2 event in DongYuan, Guangdong province of China for raw signal quality verification.

Figure 1 shows the DAS observation system, which was set in Pearl River Estuary, the northern South China Sea area with the starting point on Sanjiao Island and the ending point on Guishan Island, with the submarine fiber-optic cable about 10 km. The probe pulse was sent from the side of Sanjiao Island, of which the Rayleigh scattering occurred while it travels along the fiber. The backscattering light eventually would be received by the sensing modules installed on the island20. There would be a phase difference of the backscattering light when strain generated by the vibration of the fiber changed. Hence, the accurate location where the fiber warped as well as the amplitude of the vibration can be easily calculated according to the travelling time of the backscattering light and the phase difference. For comparative verification, the records of ZhuHai broadband Seismic Station are used for waveform comparison. In the test, the spatial sampling rate is adjustable in the range of 2 m, 4 m and 8 m, with the gauge length 2 times. So the gauge length is also floating. Gauge length determines the measuring unit length, and long gauge length measures more effectively than short gauge length, but it is pointless for measuring of longer gauge length because of monitoring accuracy.

Fig. 1
figure 1

Map of the observed system. (a) Map of the earthquake events (red star), optical cable and broadband seismic station(triangle); (b) Map of the Pearl River Estuary area and submarine cable area; (c) details of the test cable from San Jiao Island and GuiShan Island with the length of 10 km. The image was created by the author using MATLAB.

Full size image

Results of different methods

Raw signal

The volume of the data per day is an immense number with the sampling rate up to 1000 Hz. It is formidable to process all the data, so we adopted the scheme that selecting DAS records according to the Earthquake catalogue. Here, we collected the records of ten earthquake events for research. This section will introduce the characteristic of raw waveform, and the comparison between it after processing and Zhuhai seismic station record. The representative one of the events is a 2021.9.4T07:51:52 Mw3.2 Earthquake in DongYuan city, Guangdong province, China, whose focal depth is 9 km, as shown in Fig. 1, in which the sampling rate is 100 Hz and the spatial resolution is 2 m. So there are around 5000 observation points were arranged in 10 km.

Because the record of the ocean surface gravity waves is significant and periodic, it must be taken into consideration in our study. As the frequency of the surface waves is between 0.05 and 0.3 Hz, and their amplitude decays rapidly with depth by the linear theory of gravity waves11. In a DAS test, Sladen found that the distributed cable recorded the dynamic pressure produced by the surface gravity wave down to a depth of 100 m, closed to the minimum with the amplitudes as low as 1 nstrain/s (nano-strain per second) were measured11. As shown in Fig. 1, the average depth of the cable in our test is lower than 10 m, Cross-sectional plot is showed in Supplementary S1, so the effect of the surface wave is in an influential range and the signals of the earthquake is buried in the surface-wave acoustic noise with the frequency band below 0.3 Hz. The continuous waveform transform (CWT) spectrum shows the noise level as shown in Fig. 2, in which wave noise with frequency less than 1 Hz, take the main part of the signals and the Pseudo Color(P-Color) figure is mainly recorded the sea wave.

Fig. 2
figure 2

Raw data analysis figures in the 3.2Mw event. (a) P-Color of the DAS signals among 100 s after the origin time; (b) The image of the continuous waveform transform spectrum as well as three waveforms with different frequency band, 0.02–0.5 Hz, 0.5–2 Hz, 2–10 Hz.

Full size image

In Fig. 2, The raw data recorded by the DAS undersea is in low quality, with the reason of the high sensitivity of the sensing fiber and the complex condition in the sea as well as the weak coupling between the fiber and the seafloor. And waveforms show no signal about the earthquake events. In the frequency band of 0.02–0.5 Hz, the amplitude of the surface gravity wave is the main component. It is hard to observe the signals of earthquake in the other two one-channel waveform. As the seismic signal is weak in the records of single channel, the stacking profile with P-Color image has often been used to enhance the signal to noise ratio. The frequency-wavenumber (f-k) domain analysis is used to separate coherent signals by using two-dimensional(2D) Fast Fourier Transform23. Figure 3 shows that the ocean waves with a speed of less than 10 m/s have the strongest energy, and the noise amplitude shows that during these 10 min of recording, the energy of waves in the westward direction is stronger, indicating that the waves are propagating towards the west in this time. The energy of Scholte waves with wave speeds of 100 m/s and 1000 m/s is slightly weaker than that of ocean waves, but it also completely covers the seismic signal, and as a surface wave, Scholte wave energy exhibits a certain dispersion propagation effect.

Fig. 3
figure 3

Frequency wavenumber spectrum (FK Analysis) of the 10-min recorded Mw3.2 magnitude earthquake on September 4, 2021 at 07:51:52 in Dongyuan.

Full size image

Figure 4 shows waveforms with different frequency band filters, Fig. 4b and c indicate that the first 2 km cable successfully receive the S wave, because the part is in a slope with the supporting force of gravity to get the better coupling. This test suggests the normal communication cable have the potential in undertaking the mission for submarine micro-earthquake monitoring, although it has the shortage of the unidentified coupling with the ground. On the other hand, it is easier to monitor a local earthquake with high frequency wave which will not be cover by the surface gravity wave without difficulty by high-frequency filtering. In the contrary, the frequency of the teleseism is usually in 0.02–1 Hz, in which the sea wave is in dominating position.

Fig. 4
figure 4

One-channel P-Color of the DAS signals among 100 s after the origin time and the clear detail views. (a) 0.02–0.5 Hz band-pass filtering p-color shows the information of the surface gravity wave which occupy the main amplitude. (b) 0.5–2 Hz band-pass filtering p-color contains the waveform of the surface gravity wave and weak S wave. (c) 2–10 Hz band-pass filtering p-color in the front part of the cable go down to submarine with tilt support force display kind of S wave.

Full size image

DAS responds to the external vibration or acoustic signal around the sensing fiber, whose phase change is found linear to the external vibration at a certain location. In the course of vibration recording by sensing fiber, Earthquake, acoustic noise, ocean waves, temperature variation and so on, would cause fiber strain change. In order to highlight the earthquake signals, some denoising method need to be proposed in data processing. This section mainly introduces three common methods, Stacking, Discrete Wavelet Transform (DWT) and Ensemble Empirical Mode Decomposition (EEMD) Transform.

Figure S1B shows the map of 10 earthquakes in this study, all of which are local events with magnitudes around 3.0 MW. In the fiber optic interception data of ten earthquake events, the sampling rate is 100 Hz, and the sampling point intervals include 2 m, 4 m, and 8 m, which correspond to 4756, 2378, and 1189 sampling points.

Here we still take the event of 2021.9.4T07:51:52 Mw3.2 Earthquake in DongYuan as an example to demonstrate the data processing and the other events analysis result see in figure S2 in supplementary.

Channel stacking

In the test, every 2 m a sensing unit is along the 10 km optical cable. By the values superposition of several channels, which is also called simple stacking, the time-varying noise and optical noise produced by temperature fluctuation will be removed and the effect of noise suppression will be achieved21,29,30. However, due to differences in the propagation of focus from different directions, the simple stack cannot be allowed to adapt different earthquakes with the condition of excessive stacking quantity leads to insufficient seismic phase enhancement. So the phase weighted stack should also be complied in the waveform processing.On the other hand, when quantity is not enough, overlap stack is usually used in the undersized samples for repetitive and efficient utilization as shown in Fig. 5.

Fig. 5
figure 5

About 10 km DAS array in two different stacking methods. Simple stack is used to universal application for waveform detection with the shortage of capacity of reverting seismic phase. Overlap stack shows the ability of expanding capacity of DAS traces.

Full size image

In passive earthquake seismology, seismic phase analyzing is to deduce the origin of the specific seismic phases, and then determine the seismic source characteristics and the structure and properties of the earth’s internal structures, So the seismic phases are the base of other intensive study. In the case that the signals of the DAS record are not so clear, stacking will suppress noise while eliminating the influence of temperature by gathering multiple traces of data. All the DAS stacking waveforms is arranged by fiber distance, is used for seismic phase analyzing. In the CWT spectrum of 30-traces stacking, 0.01–2 Hz is surface wave and Scholte oriented conforming to the test conclusion showed in Fig. 3, so we use band-pass filtering of 2–10 Hz to keep the high frequency so that the part is shown about the event with P and S wave in Fig. 6. And the P-Color show the clear signals alongside the fiber, but the middle part is not so clear because of the weak coupling possibly by the traditional Hydraulic jet burial method in burying communication cable, and it is a pity without the concrete condition about that.

Fig. 6
figure 6

The waveform and its analysis after 30-traces stacking. (a) P-color reveals the clearer waveform of P wave and S wave. (b) The cwt spectrogram of single trace after stack. (c) Single trace waveform comparison (red: after-stack; grey: original single trace) with the band-pass filter of 2–10 Hz.

Full size image

In Fig. 7, the DAS trace shows the high-similarity in the graph by comparing with the ZhuHai seismic station. However, there are still three shortages of DAS technique: (i). The wavefield recorded by the cable is dominated by S waves, which is inconsistent with the seismographic station record in which the P-waves are visible. Hence, the very small amplitude of the P-wave on the DAS which may be covered by noise is owing the lack of sensitivity (Ethan et al.,2019); (ii). The frequency band of the Earthquake signals of DAS is not as wide as the seismographic station records; (iii). The other seismic phases are fuzzy in DAS record. At the same time, No-signal events still show no waveform after stacking, while the SNR of other weak-signal events got enhanced along the fiber (see Figure S3-S5).

Fig. 7
figure 7

The waveform comparison between DAS fiber and ZhuHai seismographic station record. The cwt spectrogram and the waveform show high-similarity, but high-level noise exist in DAS record.

Full size image

Discrete wavelets transform

The shortages of the traditional FFT (Fourier Fast Transform) is obvious that fitting the time window is rigid and hard to setup for a accurate figure. In contrast, Discrete Wavelet Transform (DWT) was proposed to match a large range of the frequency band for separating variable back-scattering light frequency. Figure 8 shows the DWT result, and both are after stacking and 2 ~ 10 Hz filtering, and in both graph the P and S wave are obvious, but the DWT one wipe off the most of the background noise and retain the seismic phases signals, of which the wavelet is db8 and the method is soft threshold of 0.2. The support length of DB8 wavelet is moderate, which allows it to maintain high computational efficiency while avoiding boundary problems caused by excessively long support lengths when processing signals, and it has symmetry, which can avoid phase distortion, ensure linear phase characteristics. And The selection of threshold determines the signal output quality of discrete wavelet transform, and noise and signal length are the key factors affecting the threshold. This experiment compared fixed threshold and adaptive threshold, and finally chose the threshold of 0.2 that is suitable for the signal in this experiment.

Fig. 8
figure 8

(a) P-color map after stack. (b) P-color map after stack and DWT. Discrete wavelet transform eliminate most of the noise and retain the P and S waveform according to the different frequency bands and threshold parameter. The DWT result shows the malpractice of P wave elimination in the case that the wave amplitude is equivalent to the noise.

Full size image

Through experimentation, discrete wavelet transform can greatly improve the SNR while preserving important seismic signals, which can greatly improve the utilization of data for shallow ocean fiber optic cable observations. But for weak signal P-waves, this method has a certain degree of weakening effect on P-waves, which is also the limitation of this method. Other DWT-processing events result see Figure S6. In the section, we use SNR for expressing the improvement after DWT and EEMD. SNR > 1 in each channel is the evaluation criterion as the effective signal. We compare the quantity of the criterion to quantitative improvement. For example, Figure S7 shows several no-signal waveforms of which the SNR evaluation criterion quantity is 0%.

Ensemble empirical mode decomposition (EEMD) transform

Due to the limitations of discrete wavelet transform, this research attempts another method for deep denoising research, namely ensemble empirical mode decomposition (EEMD). EEMD is a data-analysis signal decomposition method used to decompose complex nonlinear signals into a set of intrinsic mode functions (IMFs), with each IMF representing an inherent vibration mode in the signal31,32,33. The advantage of EEMD lies in its ability to reduce modal aliasing in EMD, enhance the ability to extract local features of signals, and have better robustness to noise compared to traditional EMD. And EEMD is capable of preserving the phase information of signals effectively during the decomposition process. When decomposing signals, EEMD can separate the various frequency components of the signal into their corresponding IMFs without excessive interference with the signal’s phase. As a result, during the reconstruction of the signal, the phase information of the signal can be well preserved, thereby maintaining the original phase characteristics of the signal. Additionally, EEMD is a nonlinear and non-causal signal processing method. During the signal processing, EEMD decomposes the signal into multiple IMF components and then reconstructs these components. Due to the mutual influence of different parts of the signal during the decomposition and reconstruction processes, EEMD does not satisfy strict causality.

The analysis results shown in Fig. 9 show that for the seismic waveform after channels stacking, total 12 IMFs are decomposed. It can be seen that the vibration modes of each IMF have their unique properties, indicating that EEMD has good robustness for the data in this test.

Fig. 9
figure 9

Seismic waveforms after single channel stacking and 12 different vibration waveforms generated after EEMD analysis. IMFs (6–11) are ignored due to micro waveforms.

Full size image

By analyzing the waveform characteristics and amplitude size of each IMF, and also combining the PSD spectrum characteristics of each decomposed IMF for frequency band selection, as shown in Fig. 10, it can be seen that IMF2, IMF3, and IMF4 all have similar characteristics of seismic waves. However, due to the small amplitude of IMF2 and the excessive noise signal of IMF4, and the fact that the PSD peaks are mainly concentrated in the 2–10 Hz IMF, IMF2 and IMF2 + IMF3 were selected as the seismic wave IMFs for this batch of data.

Fig. 10
figure 10

Seismic Waveform after Single Channel Stacking and Power Spectral Density of the First 5 IMFs after EEMD Analysis.

Full size image

Figure 11 show the waveforms after denoising with multiple EEMD channels. It can be observed that this method removes most of the noise, improves the signal-to-noise ratio better than the discrete wavelet transform, and leaves strong seismic waveforms, demonstrating the feasibility of the EEMD method. In addition, by selecting IMF [3] as the signal reference term, it can be seen that in the pseudo color map of IMF [2] + IMF [3], there is more preservation for P-waves, and the signal-to-noise ratio is similar.

Fig. 11
figure 11

Comparison between Multi channel Stacked Waveform and EEMD Denoising Waveform. (a) The figure shows the results of 30 stacked waveforms after EEMD of IMF [2] as seismic waves, including p-color map and the SNR along the fiber as well as the max SNR waveform compared with Zhuhai station. (b) the results of after EEMD of IMF [2] + IMF [3] as seismic waves. The second one contains more details but more noise. In (a) and (b) the vertical axis is normalized by the maximum amplitude.

Full size image

By using EEMD to denoise 5 event waveforms (see Figure S7), it was found that EEMD outperformed discrete wavelet transform in waveform fidelity, but did not significantly improve noise elimination and SNR. Furthermore, seismic signals could not be extracted for corresponding fuzzy events.

Comparing two methods, the similarity lies in the fact that both can significantly improve the SNR and to some extent enhance the signal of the weak coupling section. The difference is that discrete wavelet transform is more effective in eliminating noise from strong signals, while EEMD is more effective in signal extraction. Selecting corresponding methods for processing observation data from different shallow ocean areas can greatly improve the SNR.

Discussions

Possible causes of undetected earthquakes

By comparing the 10 earthquakes selected in the earthquake catalog, it can be found that 5 of them occurred around the Dongyuan Reservoir in Dongyuan County, Heyuan City, Guangdong Province. Therefore, the distance between their epicenters is almost equal. However, it has been proven that only three of these earthquakes can be detected, so the distance between the epicenters is clearly not the most important influencing factor. Additionally, it can also be proved that the characteristics of the medium and the terrain structure cannot be the main considerations for incomplete seismic detection in this test. In addition, due to the lack of data on each seismic source mechanism in this article, the source mechanism can lead to differences in the directionality, wavefront shape, energy distribution, and frequency characteristics of seismic waves, that is, different radiation patterns of different source mechanisms. Therefore, there is currently no evidence to rule out the influence of seismic source mechanisms on fiber optic sensing signals. For the characteristics of the station itself, optical fibers only have a single axial component, while fixed stations have three directional components for recording. This may also have a significant impact on the recording of seismic waves by optical fibers. Therefore, We compare 8 detected seismic events with the waveforms of ZhuHai seismic station to explore the reasons for weak response.

Figure 12 show the waveforms comparison between broadband seismometers and DAS. The clear seismic phases can be detected by the first four optical cables, the amplitude recorded by the seismic station is relatively large. Correspondingly, the DAS also records the clear seismic signals. In the last four records, the amplitude values of the waves at seismic station are relatively weak, and the seismic signals in fiber optic sensors are also buried behind the sensor self-noise signals. Therefore, through comparative analysis, it can be concluded that the clear waveform recorded by DAS signals is directly related to the waveform amplitude recorded by seismic station. The reason for the weak waveform recording at seismic stations is also the reason for the weak fiber optic signal recording.

Fig. 12
figure 12

The comparison of eight DAS records and ZhuHai seismographic station records filtering in 2-10 Hz. (red: DAS array waveform by 2-10 Hz filtering, stacking and DWT; green: Zhuhai seismographic station record by 2-10 Hz filtering), broadband records are in global-max normalization for amplitude comparison. The first four DAS records show the clear Earthquake waveform in the case that the amplitude of the broadband seismographic waveform is in high level. On the contrary, the other DAS records about S wave are covered by noise, when the broadband records are in low level.

Full size image

Based on the above analysis, it is possible that the unclear seismic signals of DAS might be caused by the different focal mechanisms and the combined effects of radiation pattern, focus depth and DAS performance. Meanwhile, due to the single radial component of the optical fiber itself, it is difficult to capture weak seismic energy.

The future application and limitation

A great advantage of DAS is that it only requires an interrogation unit at one end of a telecommunication fiber optic cable, which makes it cost effective and easy to implement. There is already a vast installed base of telecommunication cables around the world; opportunistic DAS deployments utilizing this resource provide a low‐ cost approach for dense array acquisition. In some harsh environment, such as offshore, where conventional seismic deployment is difficult, DAS may be of more value in terms of seismic monitoring, subsurface imaging, and hazard assessment.

One of the current limitations of DAS is its higher noise level than conventional broadband seismometers, The other one is the dependence of DAS to the ground coupling. In most cases the submarine cables are direct burial with multi-fibers because of the usage of communication, while the illustration might be casual with weak coupling with the submarine ground. In view of the difficulty of submarine construction, it is burdensome for reburial and hopeless to ask for communication construction crew burying with the demand for coupling standard. So, as a tentative submarine DAS test, we make the research according to the data quality, and what we can try is to make the best use of high-quality data and avoid the debased ones.

Conclusions

In 2020 and 2021, a new DAS field test was carried out in Pearl River Rstuary, Guangdong province, China. Using 10-events waveform data from the 4 September 2021 to 7 June 2022, we explore four Earthquake events with the high-similarity to the seismographic stations data successfully. There are three conclusions:

  1. (1)

    Fiber optic cable in shallow ocean is affected by surface gravitational waves of 0.03–0.5 Hz and primary and secondary microseismic noise (Scholte waves) generated by solid–liquid coupling motion at 0.3–2 Hz, resulting in seismic signals hidden in two types of higher energy noise signals.

  2. (2)

    By the processing and deep denoising method proposed in this article for optical cable signal processing in shallow sea areas, weak seismic signals can be extracted from high noise environments. Seismic phases include P-waves and S-waves, and seismic signals can be enhanced through waveform stacking, discrete wavelet transform, and ensemble empirical mode decomposition methods while recovering as much seismic phase information as possible.

  3. (3)

    This study processed ten near epicenters in the earthquake directory, and attempted to extract signals through deep denoising. Four clear seismic waveforms were observed. For other earthquakes, fiber optic seismic response was dim. By comparing the waveform with Zhuhai Seismic Station, it can be seen that the seismic station can record clear waveforms, and the submarine optical cable can also record them. For seismic events with low energy recorded by seismic stations, due to radiation patterns and single component optical fibers, the energy is covered by noise and cannot be observed.

For fiber optic sensing technology as a new technology, in addition to using high signal-to-noise ratio land and deep ocean optical cables for observation, communication optical cables in shallow submarine can also be used for earthquake observation and analysis. For the widely distributed shallow-ocean optical cables are expected to be further developed and used.