Fault detection by reflected surface waves based on ambient noise interferometry

Ning Gu ,Hijing Zhng,b,＊ ,Nori Nkt ,Ji Go,b

a School of Earth and Space Sciences,University of Science and Technology of China,Hefei,230026,Anhui,China

b Mengcheng National Geophysical Observatory,University of Science and Technology of China,Hefei,230026,Anhui,China

c Department of Earth,Atmospheric and Planetary Sciences,Massachusetts Institute of Technology,Cambridge,MA,02139,United States

Keywords:Fault detection Ambient noise Cross-correlation functions Reflected surface wave Tanlu fault zone

ABSTRACT Detecting subsurface fault structure is important for evaluating potential earthquake risks associated with active faults.In this study,we propose a new method to detect faults using reflected surface waves observed in ambient noise cross correlation functions.Ambient noise tomography using direct surface waves obtained from ambient noise interferometry has been widely used to characterize active fault zones.In cases where a strong velocity contrast exists across the fault interface,fault-reflected surface waves are expected.We test this idea using a linear array deployed in the Suqian segment of Tanlu fault zone in Eastern China.The fault-reflected surface waves can be clearly seen in the cross-correlation functions of the ambient noise data,and the spatial position of the fault on the surface is close to the stations where the reflected signals first appear.Potentially reflected surface waves could also be used to infer the dip angle,fault zone thickness and the degree of velocity contrast across the fault by comparing synthetic and observed waveforms.

1.Introduction

Fault detection is a very important task for studying active faults(Ben-Zion and Sammis,2003;Scholz,2019;Yang,2015).For some faults,their traces may be clearly seen on the surface,and thus field investigation can be used to detect the faults.However,for many faults,especially for blind faults that do not have clear surface expressions,various geophysical methods are needed to detect their locations,including active seismic exploration (Hole et al.,2001;Mooney and Ginzburg,1986;Zelt et al.,2006;Bleibinhaus et al.,2007),passive seismic methods(Ben-Zion,1998;Li and Leary,1990;Peng et al.,2003),gravity survey(Abbott and Louie,2000;Stierman,1984),and electromagnetic survey(Ledo et al.,2002;Eberhart-Phillips et al.,1995;Unsworth et al.,1999).Different methods have their advantages and disadvantages.For example,active seismic methods have the ability to characterize detailed fault zone structures but generally it is too expensive to conduct a 3-D survey and explosive sources can be restricted to use only in certain areas due to safety concerns.For electromagnetic methods,electrical resistivity may vary too abruptly in the shallow surface to detect the real fault traces.In addition,with the fast development in both urban and suburb areas,background electromagnetic noise sources become too loud,making traditional electromagnetic methods ineffective in these cases.

In addition to active seismic methods,passive seismic methods using local earthquakes around active faults or at teleseismic distances are also widely used for fault characterization (Ben-Zion et al.,2003;Ben-Zion and Sammis,2003;Ozakin et al.,2012;Ross et al.,2020).However,passive seismic methods rely on the existence of earthquake sources,which may not be applicable to faults that are currently locked or inactive.In comparison,ambient noise tomography has emerged as one of the powerful geophysical methods to study fault zone structures at different scales over the last～20 years,which does not require local seismic events.Most ambient noise studies have focused on using direct surface waves extracted among stations to characterize the shallow or deep subsurface velocity structures(e.g.,Shapiro et al.,2005;Yao et al.,2006;Yang et al.,2007,2008;Luo et al.,2012,2013;Liu et al.,2018;Gu et al.,2019;Luo et al.,2021).In addition to direct surface waves,body waves may also exist in cross correlation functions from ambient noise analysis at local,regional and teleseismic scales (e.g.,Nakata et al.,2005;Zhan et al.,2010;Poli et al.,2012;Lin et al.,2013;Feng et al.,2017;Chamarczuk et al.,2021).

In this study,we aim at extracting reflected surface waves with ambient noise interferometry at stations that may emerge from the reflection by some faults with strong velocity contrasts on both sides.With the reflected surface waves extracted from cross-correlations of ambient noise,they can be used to determine the spatial location of a fault interface.In the next sections,we first performed synthetic test to demonstrate the existence of such reflected phases.Next,we extract fault-reflected surface waves using seismic ambient noise data recorded by a linear array deployed in the Suqian segment of the Tanlu fault zone(TLFZ)in Eastern China.

2.Reflected surface waves in a fault zone

We first illustrate through a synthetic model that reflected surface waves may emerge from the reflection by a fault zone.From various studies at different scales (Gu et al.,2019;Bem et al.,2020;Li et al.,2020;Ma et al.,2020;Luo and Yao,2021),it is found that the TLFZ is generally associated with high velocity anomalies.Therefore,we first set up a synthetic fault zone model with high velocity in the fault zone and low velocity on both sides of the fault (Fig.1).In this model,the fault zone consists of high velocity metamorphic rock with S-wave velocity of 2 500 m/s,surrounded by sediments with S-wave velocity of 2 000 m/s.In the synthetic test,the Vp/Vsratio is assumed to be 1.73,and the density is calculated according to the Gardner's relation(Gardner et al.,1974).17 stations(white triangles)are distributed on the surface,and a seismic source (red star) is emplaced at the location of the first station(Fig.1).

Based on the model shown in Fig.1,we perform a forward wavefields simulation by a finite difference method (Madariaga,1976;Virieux,1984,1986).The standard staggered-grid finite difference elastic solver is used to propagate the wavefields (Levander,1988),with the W Formulation with the Adjusted FD Approximations (W-AFDA) scheme adopted for modeling the surface waves(Kristek et al.,2002).In this test,the source on the left side(red star)is excited with a 5 Hz Ricker wavelet for propagating the wavefields.The grid spacing is 25 m and time step is 2 ms.Fig.2 shows the waveform records received by 17 stations.The body wave and the direct surface wave can be clearly seen.Compared to the body wave,the energy of surface wave is relatively stronger.In addition to direct surface waves,the reflected surface waves also appear at the fault location (Fig.2).The apparent velocity of reflected surface wave is almost the same as that of the direct surface wave,but with a negative slope along the time axis.

Fig.1.Synthetic S-wave velocity model for a fault zone with higher velocity anomaly.The yellow high-velocity area represents the fault zone,the white triangles represent the stations,and the red star represents the seismic source.

Fig.2.Simulated Wavefield based on the synthetic model in Fig.1.Different types of wave and fault location are marked.

Considering that fault zones are generally associated with lowvelocity anomalies,we also set another synthetic model with a lowvelocity anomaly of 5% in the fault zone relative to the surrounding rocks(Fig.3).In this model,the S-wave velocity of fault zone is set to be 2 375 m/s,and the surrounding rock has the S-wave velocity of 2 500 m/s.For the waveform simulation,the parameters are the same as those used for the model in Figs.1 and 2.It can be seen that the reflected surface waves also appear at the fault location (Fig.4),but the energy is relatively weaker due to the smaller velocity contrast across the fault.

Through the numerical wavefield simulations,it can be seen that reflected surface waves can occur when there exist strong velocity contrasts across the fault.When the velocity contrast is relatively small,the reflected surface wave may be too weak to be observed.Based on the station where the reflected surface wave first occurs,the fault position can be determined.

3.Data analysis and results

A dense seismic array consisting of 240 short-period (5s-150Hz)seismic stations (EPS-2-M6Q) was deployed in the region of 50 km×60 km around the Suqian segment of TLFZ in Eastern China from February 25 to March 28,2019(Fig.5).There are 6 linear arrays roughly perpendicular to the strike of the TLFZ and the nominal line separation is about 10 km.Each line has 40 stations with the interstation spacing varying from 100 m around fault F4 to 3 km away from it.The dense array recorded data continuously at a sampling rate of 200 Hz.Using this dataset,Gu et al.(2021,under review) conducted ambient noise tomography and obtained a 3-D Vs model for the Suqian segment.One cross section normal to the fault from this 3-D Vs model is plotted in Fig.6,which clearly shows high velocity anomalies are imaged to the west of fault F4.

Fig.3.Synthetic S-wave velocity model for a fault zone with lower velocity anomaly.The blue low-velocity area represents the fault zone,the white triangles represent the stations,and the red star represents the seismic source.

Fig.4.Simulated wavefield based on the synthetic model in Fig.3.Different types of wave and fault location are marked.

Fig.5.Distribution of stations around the Suqian segment of Tanlu fault zone.17 stations enclosed by the red box are shown at the bottom.The red line shows the profile AA’.The insert shows eastern China where the TLFZ is denoted as a solid black line and the study region is marked as the blue rectangle.

In this study,we aim at using a linear array located in the northern edge of the study area to detect potential reflected surface waves related to fault F4.Here we only use 17 stations around fault F4 for subsequent analysis,with a total length of 7.8 km.The station names are marked by 1017,1019,1021,…,1049,respectively(Fig.5).According to the local geological survey data,the F4 fault is near station 1027.

Fig.6.Cross section of the Vs model along the profile AA′ shown in Fig.5 (Gu et al.,2021;in review).

We first assemble the data on a daily basis in the SAC format for each station.Then we follow the ambient noise data processing procedure of Bensen et al.(2007) but with some slight modifications.The vertical component of recorded data at each station is first cut into hourly records and resampled to 50 Hz before demeaning and detrending.Spectral whitening and temporal normalization are then performed on these hourly records in the frequencies of 1–5 Hz.Finally,the cross-correlation functions (CCFs) of these processed hourly data are calculated for each station pair and are stacked to get the final CCFs between stations by using a normalized linear stacking method.

After calculating the CCFs,we can get the common virtual shot gathers similar to the active source exploration.Each station can be regarded as a virtual source,as shown by the red line in Fig.7.The positive part in the CCF represents the recorded wavefield from the virtual source to other stations,while the negative part is the wavefield propagating in the opposite direction.It can be seen that in addition to the direct surface waves,there are also reflected signals with the opposite slope to the direct surface wave (Fig.7).The station first showing reflected signals is around fault F4.

4.Discussion

There are two possible causes for the observed reflected signals.First,it may comes from a local noise source.Second,the reflected signal could be due to the surface wave reflected from the fault.

In order to determine whether the reflected signals come from a local noise source,we calculate the waveform amplitudes for each hour-long record on 17 stations,which are filtered for the frequency band of 1–5 Hz.The normalized amplitude values for each station for four different days are shown in Fig.8.

It can be seen from the time-frequency analysis that at station 1043 the waveform amplitudes are always strongest for the entire time span of deployment.This is because this station is located close to a highway and the high-frequency vehicle noise occurs at all times.However,the reflected waves seen in the CCFs do not appear near station 1043,indicting the noise source close to the high way is not responsible for the observed signal.On other stations,the waveform amplitudes stay roughly at the same level for each day,with the amplitude slightly stronger in the first 10 h,which correspond to the local time between 08:00 and 18:00 in the daytime.Thus,the stronger waveform amplitudes during this time period are caused by various human activities.

To further analyze how the noise could affect the observed reflected signals in the CCFs,we chose two time periods of daytime(06:00–20:00)and nighttime (20:00–06:00),respectively.Here,we used the same processing flow and parameters as before to get the final CCFs.It can be seen that reflected signals can still be observed in the CCFs for both daytime and nighttime periods(Fig.9).This indicates that the reflected signals are not caused by individual noise sources.It is also noticed that the reflected signals are stronger in the CCFs calculated for the daytime period.

Fig.7.Cross-correlation functions for different virtual sources.Red lines mark the location of virtual sources.Red arrows mark the reflected surface waves.

Fig.8.Normalized amplitude values of the filtered waveforms on 17 stations for different days.The frequency range is 1–5 Hz.

Fig.9.Cross-correlation functions by using daytime(06:00–20:00)(a and c)and night(20:00–06:00)(b and d)data for two virtual sources at stations 1017 and 1047.

Fig.10.Cross-correlation functions for virtual source at station 1017.The red dashed lines show the velocity of 300 m/s.The position of fault F4 is marked.

In order to determine that the reflected signal is a reflected surface wave,we first conduct an apparent velocity analysis.As shown in Fig.10,the apparent velocities for the direct surface wave and the reflected signal are both around 300 m/s.Therefore,it can be derived that the observed reflected signal is not body wave,but surface wave caused by the reflection of direct surface wave from fault F4.It can be seen that the reflected signals from different virtual sources all first appeared around station 1029,corresponding well to fault F4.This suggests that we can use the station location where the reflected signal first appears to determine the fault position.

Here we only use reflected surface waves to determine the fault position.Based on wavefield simulations,reflected surface waves could be potentially used to determine other fault parameters,including dip angle,fault zone thickness,and velocity contrast across the fault,in a way similar to fault zone trapped waves (Li et al.,2004).However,further studies are needed to utilize more information from reflected surface waves.

5.Conclusion

In this paper,we use reflected surface waves in the ambient noise CCFs as an indicator for the fault position.Through a synthetic model analysis,it can be seen that the surface waves can be reflected when the fault zone has a strong velocity contrast across the fault zone.The actual data analysis using the ambient noise data recorded on a linear array shows clear reflected signals on the CCFs.The reflected signals are shown to be reflected surface waves of the direct surface waves from fault F4,which have similar apparent velocities but opposite slopes.The stations where the reflected surface waves first appear are close to the surface position of fault F4.Therefore,our study shows that the reflected surface waves in the ambient noise cross-correlation functions could be used as an effective way to detect fault position.

Acknowledgements

We would like to thank Haipeng Li and Jikun Feng for their help during this study and their constructive comments on this paper.This research is supported by the National Key R&D Program of China(2018YFC1504102)and National Natural Science Foundation of China(41961134001).