RESEARCH ARTICLE On Beamforming of DAS Ambient Noise Recorded in an

10.1029/2022JB026339
Urban Environment and Rayleigh-To-Love Wave Ratio
Key Points:
• G eophone and distributed acoustic
Estimation
sensing (DAS) ambient noise data are Yumin Zhao1 , Yunyue E. Li1,2 , and Bei Li1
simulated and their differences and
implications on beamforming results 1
Department of Civil and Environmental Engineering, National University of Singapore, Singapore City, Singapore,
are analyzed 2
Department of Earth, Atmospheric, and Planetary Sciences, Purdue University, West Lafayette, IN, USA
• Beamforming results from DAS
ambient noise data are confirmed
reliable with careful processing when
noise sources are unidirectional Abstract Ambient noise sources in urban environments are generally not perfectly uniformly distributed
• Rayleigh and Love waves are or travel exactly along the axial direction of the receiver array. The horizontally deployed distributed acoustic
identified from the DAS field data
and the Rayleigh-to-Love wave ratio
sensing (DAS) fiber-optic cable is able to record both Rayleigh and Love waves. Therefore, resolving the source
is resolved propagation direction and the content of Rayleigh and Love waves from the DAS ambient noise data is essential
to resolve a reliable near-surface shear-wave velocity model. However, a standard geophone-based beamforming
Supporting Information:
procedure may not apply to DAS ambient noise data due to the complex measurements of DAS. To address
Supporting Information may be found in this issue, we investigate the validity of beamforming results by analyzing the simulated seismic ambient noise
the online version of this article. recorded by a DAS array close to an L-shape under different source conditions. Synthetic examples demonstrate
the importance of correcting amplitude imbalances and polarity reversals in the DAS ambient noise data to
Correspondence to: ensure a reliable beamforming analysis. In addition, we estimate the Rayleigh-to-Love wave (R/L) ratio from
Y. E. Li, the DAS ambient noise data with the resolved source direction, the known array geometry, and the measured
[email protected] amplitude information. Finally, we apply the method to two 2 min DAS ambient noise data recorded by the
Stanford DAS Array-1. The results suggest that the analyzed data were mainly generated from the local traffic
Citation: and dominated by Love waves.
Zhao, Y., Li, Y. E., & Li, B. (2023). On
beamforming of DAS ambient noise
recorded in an urban environment and
Plain Language Summary Human activities are the primary source that generates seismic ambient
Rayleigh-to-Love wave ratio estimation. noise at higher frequencies (above 1 Hz). Resolving the main source propagation direction is essential to extract
Journal of Geophysical Research: Solid a reliable near-surface shear-wave velocity model from the seismic ambient noise recorded by the horizontal
Earth, 128, e2022JB026339. https://doi.
components of seismic sensors. Beamforming is a commonly used method to resolve the ambient noise
org/10.1029/2022JB026339
source propagation directions. However, it is designed for seismic ambient noise data recorded by the vertical
Received 8 JAN 2023 component of traditional seismic sensors. Due to the complex measurements of distributed acoustic sensing
Accepted 5 AUG 2023 (DAS), beamforming may not apply to DAS ambient noise data. We investigate the validity of beamforming
results resolved from DAS ambient noise data with several synthetic examples. We also improve the resolution
Author Contributions: of beamforming results with careful processing. Besides, the horizontally deployed DAS records Rayleigh
Conceptualization: Yumin Zhao, and Love waves, both of which would be used to invert the shear-wave velocity model since they cannot be
Yunyue E. Li
separated without enough information about the sources. We propose to estimate the Rayleigh-to-Love wave
Formal analysis: Yumin Zhao
Funding acquisition: Yunyue E. Li (R/L) ratio based on the resolved source direction, the known array geometry, and the measured amplitude
Methodology: Yumin Zhao, Bei Li information. Finally, we apply the method to the field DAS ambient noise data recorded by the Stanford DAS
Resources: Yunyue E. Li
Array-1.
Software: Yumin Zhao
Supervision: Yunyue E. Li
Validation: Yumin Zhao
Visualization: Yumin Zhao 1. Introduction
Writing – original draft: Yumin Zhao
Seismic ambient noise wavefield travels through Earth's layers and contains rich information about the subsurface
at different scales. Compared with the costly vibroseis and environmentally harmful explosives, ambient noise
sources are free and environmentally friendly. Therefore, seismic ambient noise becomes increasingly important
© 2023. The Authors. for urban environment monitoring. The typical frequency of seismic ambient noise ranges from 0.001 to 100 Hz
This is an open access article under (Nakata et al., 2019). It mainly results from natural vibrations (below 1 Hz) and anthropological activities (above
the terms of the Creative Commons
Attribution-NonCommercial-NoDerivs 1 Hz) (Díaz et al., 2017). The single-station ambient noise data recorded in the urban environment appear random
License, which permits use and due to the random nature of the ambient noise sources. The assumptions in passive seismic interferometry are that
distribution in any medium, provided the ambient noise sources are uniformly distributed, and their generated wave modes are equipartitioned. Based on
original work is properly cited, the use is
non-commercial and no modifications or that, we can extract coherent signals from the seismic ambient noise recordings by cross-correlating the signals
adaptations are made. recorded by a specific sensor (virtual source) with those recorded by the other sensors. The cross-correlation

ZHAO ET AL. 1 of 17
 21699356, 2023, 8, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022JB026339 by National Institutes Of Health Malaysia, Wiley Online Library on [19/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2022JB026339

Writing – review & editing: Yumin result is referred to as empirical Green's function, which approximates the response of each sensor to an impulse
Zhao, Yunyue E. Li, Bei Li
source at the virtual source (Claerbout, 1968; Wapenaar, 2004). The extracted coherence signals, also known
as the common virtual-shot gather, can be used to extract the dispersion curve (e.g., Dou et al., 2017; Tribaldos
et al., 2021; Zhang et al., 2019). This information can then be used to invert the near-surface shear-wave velocity
model and soil properties (e.g., Dou et al., 2017; Park et al., 1999; Zhang et al., 2019).

However, ambient noise sources in urban environments are usually not perfectly uniformly distributed since the
roads and traffic patterns are preferentially oriented according to geographic and urban development constraints.
Besides, as the demand for real-time near-surface monitoring is rising, shorter periods of seismic ambient
noise recordings are preferred to resolve the near-surface structural information with high temporal resolution.
Although long-period (month- or year-long) seismic ambient noise recordings may show more randomness, such
source averaging effect does not apply to short-period (minute-long) recordings. Seismic ambient noise recorded
in the urban environment within a short period usually comes from a fixed location (e.g., Y. E. Li et al., 2020;
Nilot et al., 2019; Zhang et al., 2019). The non-uniform distribution of anthropological noise sources leads to
asymmetrical amplitude and travel times of the causal and anticausal parts of the extracted common virtual-shot
gather (Stehly et al., 2006). The dispersion curves extracted from both parts of the common virtual-shot gather are
apparent seismic wave phase velocities. Directional correction should be applied to the apparent phase velocities
to obtain the true velocity information of seismic waves (Cheng et al., 2016; Morton et al., 2021; Yan et al., 2022).

Distributed acoustic sensing (DAS) is a cutting-edge technology that enables continuous and long-term
near-surface monitoring. The DAS system consists of an interrogator unit and a fiber-optic cable. The length
of the fiber-optic cable can be tens of kilometers (Ajo-Franklin et al., 2019). Compared to conventional seismic
sensors, DAS has a much higher spatial resolution and lower per-channel cost. It has been successfully applied
to near-surface characterization in urban environments, where fiber-optic cables are usually horizontally and
nonlinearly deployed (e.g., Dou et al., 2017; Fang et al., 2020; Spica et al., 2020). Dispersion curves or shear-
wave velocity models were extracted from minute-long (Nilot et al., 2019; Zhang et al., 2019), hour-long (Luo
et al., 2020; Tribaldos et al., 2021; Zeng et al., 2017), day-long (Ajo-Franklin et al., 2019; Dou et al., 2017;
Martin et al., 2017), and month-long seismic ambient noise data (Martin, 2018; Martin & Biondi, 2018; Spica
et al., 2020). The common virtual-shot gathers extracted from most of these studies are asymmetric, indicating
insufficient source averaging and preferential source directionality. However, very few of these studies applied
the directional correction that is essential for a reliable near-surface shear-wave velocity inversion, especially
when dealing with DAS ambient noise.

The propagation direction and velocity of the seismic ambient noise wavefield recorded by the vertical compo-
nent of the geophone are commonly resolved by the beamforming technique (e.g., Brooks et al., 2009; Zhang
et al., 2019). However, it is rarely applied to DAS ambient noise data due to the complexity of DAS measure-
ments. One problem that hinders resolving a reliable source direction from the DAS ambient noise data is the
amplitude imbalances between the data recorded by two nonparallel orientations of the horizontally deployed
DAS array (Lindsey et al., 2017). These differences are caused by the directionality of fiber-optic cables and
geometry effect (Mateeva et al., 2012). Besides, polarity reversal may appear at the corner of the array when
DAS records Love waves when the angle of intersection between the propagation direction of the source and one
orientation of the array is acute and that with the other is obtuse (e.g., Benioff, 1935; Lindsey et al., 2017; Yu
et al., 2019). These two features of DAS ambient noise make beamforming analysis a challenging task. Lindsey
et al. (2017) resolved the direction of propagation of the teleseismic signals recorded by an L-shape DAS array.
Fang et al. (2020) detected the propagation direction of the quarry blast signals recorded by the Stanford DAS
Array-1 (e.g., Biondi et al., 2017; Martin et al., 2017; Yuan et al., 2017). The DAS ambient noise data in these two
cases show coherent events instead of randomness since each was generated from a unique known far-field source
from a fixed direction. However, it is more challenging when the source is unknown, moving, and close (within
a few hundred meters) to the array. Recently, there have been several studies on beamforming with DAS data.
For example, van den Ende and Ampuero (2021) attempted to enhance beamforming results by converting DAS
strain rates into particle velocities. Additionally, Näsholm et al. (2022) developed a framework for DAS steered
response estimation that takes into account both cable directivity and gauge-length averaging.

The horizontal components of the receiver array record both Rayleigh and Love waves. As the physics of Rayleigh
and Love waves are different, prior knowledge of the contents of each surface wave type in the common virtual-shot
gather is essential for a reliable inversion of the near-surface velocity model. Rayleigh and Love waves recorded

ZHAO ET AL. 2 of 17
 21699356, 2023, 8, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022JB026339 by National Institutes Of Health Malaysia, Wiley Online Library on [19/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2022JB026339

by a geophone can be separated by rotating the two horizontal components

in the radial/transverse direction. However, those recorded by DAS cannot
be separated if there is insufficient information about the sources. Previous
studies show that when the virtual source is colinear with the DAS array,
the extracted common virtual-shot gather mainly contains Rayleigh waves
(Luo et al., 2020; Martin & Biondi, 2017; Martin et al., 2017; Snieder, 2004).
The theory assumes that ambient noise sources are random and uniformly
distributed, and the recording time is long (months or even years). However,
the assumptions are unsuitable for real-time monitoring where a short period
of data is utilized.

This paper aims to resolve the ambient noise source direction and estimate
the content of Rayleigh and Love waves (i.e., Rayleigh-to-Love wave [R/L]
Figure 1. Simple illustration of the ith plane wave generated by a distant ratio) from DAS ambient noise data recorded in an urban environment.
ambient noise source and recorded by a 2D receiver array (two blue lines). We Following the acquisition setup of the Stanford DAS Array-1, we simulate
assume the plane wave first arrives at the sensor at the corner of the array and
DAS ambient noise data under different source conditions with varying
use this information as a reference to obtain the seismic data recorded at the
other traces. Take y-line as an example, θi is the intersection angle between combinations of Rayleigh and Love waves. We normalize the amplitude of
the propagation direction of the ith plane surface wave and y-line. The two the DAS ambient noise data recorded by the two orientations and extract the
red dots represent the locations of two channels, one on each orientation, common virtual-shot gather from it. By observing the polarity of the common
which are x1 and x2, respectively. x-line is connected to the distributed acoustic virtual-shot gather at the corner of the DAS array, we determine whether
sensing interrogator.
there is polarity reversal and apply a polarity flip to the common virtual-shot
gather when polarity reversal occurs. After that, we apply multiple signal
classification (MUSIC) beamforming (Gelius et al., 2013; Godara, 1997;
Goldstein & Archuleta, 1987; Kirlin, 1992; Zhang et al., 2019) on the extracted common virtual-shot gathers. We
demonstrate that beamforming results obtained from the DAS ambient noise data after amplitude normalization
and polarity correction are reliable when the source direction is unique. Based on the resolved source propagation
direction, we calculate the R/L ratio from the noise amplitude measured by the fiber-optic cable deployed in two
different directions. Additionally, when ambient noise arrives at the array from multiple directions, beamforming
cannot provide universally reliable information for all scenarios, thus making the apparent velocity correction and
R/L ratio inversion challenging. We apply the method to two 2 min DAS ambient noise data recorded by Stanford
DAS Array-1. The results suggest that the two 2 min DAS ambient noise were mainly generated from local traffic
and dominated by Love waves.

2. Methodology
2.1. Urban Seismic Ambient Noise Simulation

We assume a receiver array is deployed close to the L-shape, and it records seismic ambient noise generated by
ambient noise sources at fixed locations far away from the array. We assume each source generates Rayleigh
and Love waves simultaneously, and the recorded ambient noise is the superposition of plane surface waves.
To simulate DAS ambient noise, we first simulate the seismic ambient noise recorded by one orientation of the
three-component geophone array as displacement, and then we derive the DAS ambient noise as strain. Without
considering the attenuation, the seismic ambient noise recorded by the selected orientation of the vertical compo-
nent of the geophone array is as follows:
𝑁𝑁𝑠𝑠

(1) 𝑎𝑎𝑅𝑅𝑅𝑅 𝑒𝑒−𝑖𝑖𝑖𝑖(𝑡𝑡𝑖𝑖 +𝑥𝑥 cos𝜃𝜃𝑖𝑖 ∕𝑣𝑣(𝜔𝜔)) ,

∑
𝑑𝑑(𝑥𝑥𝑥 𝑥𝑥) = 𝐴𝐴(𝜔𝜔)
𝑖𝑖=1

where x denotes the distance between the current receiver and the one at the corner of the L-shaped array, and
ω is the angular frequency. A(ω) is the amplitude spectrum of the source wavelet, and Ns is the number of plane
surface waves generated by the ambient noise sources during the recording period. aRi is the pulse amplitudes
of Rayleigh waves generated by the ith source, i = 1, 2, …, Ns. ti denotes the time at which the ith plane surface
wave is generated. aRi and ti follow uniform distribution. θi is the intersection angle between the direction of prop-
agation of the ith plane surface wave and the selected orientation of the array (Figure 1). v(ω) is the surface wave
phase velocity at angular frequency ω.

ZHAO ET AL. 3 of 17
 21699356, 2023, 8, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022JB026339 by National Institutes Of Health Malaysia, Wiley Online Library on [19/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2022JB026339

The ambient noise recorded by the horizontal components contains both Rayleigh and Love waves, where the
Rayleigh wave components are:
𝑁𝑁𝑠𝑠

(2) 𝑎𝑎𝑅𝑅𝑅𝑅 cos𝜃𝜃𝑖𝑖 𝑒𝑒−𝑖𝑖𝑖𝑖(𝑡𝑡𝑖𝑖 +𝑥𝑥 cos𝜃𝜃𝑖𝑖 ∕𝑣𝑣𝑅𝑅 (𝜔𝜔)) .

∑
𝑑𝑑𝑅𝑅 (𝑥𝑥𝑥 𝑥𝑥) = 𝐴𝐴(𝜔𝜔)
𝑖𝑖=1

The Love wave components can be expressed as follows:

𝑁𝑁𝑠𝑠

(3) 𝑎𝑎𝐿𝐿𝐿𝐿 sin𝜃𝜃𝑖𝑖 𝑒𝑒−𝑖𝑖𝑖𝑖(𝑡𝑡𝑖𝑖 +𝑥𝑥 cos𝜃𝜃𝑖𝑖 ∕𝑣𝑣𝐿𝐿 (𝜔𝜔)) ,

∑
𝑑𝑑𝐿𝐿 (𝑥𝑥𝑥 𝑥𝑥) = 𝐴𝐴(𝜔𝜔)
𝑖𝑖=1

where aLi is the pulse amplitudes of Love waves generated by the ith source, i = 1, 2, …, Ns.
To approximate the DAS measurements, we first compute point-wise strain by taking the spatial derivative of the
horizontal displacements of the Rayleigh and Love waves components:
𝑁𝑁𝑠𝑠
𝑖𝑖𝑖𝑖cos2 𝜃𝜃𝑖𝑖
(4) 𝑎𝑎𝑅𝑅𝑅𝑅 𝑒𝑒−𝑖𝑖𝑖𝑖(𝑡𝑡𝑖𝑖 +𝑥𝑥 cos𝜃𝜃𝑖𝑖 ∕𝑣𝑣𝑅𝑅 (𝜔𝜔)) ,
∑
𝑠𝑠𝑅𝑅 (𝑥𝑥𝑥 𝑥𝑥) = 𝐴𝐴(𝜔𝜔) −
𝑖𝑖=1
𝑣𝑣(𝜔𝜔)

𝑁𝑁𝑠𝑠
𝑖𝑖𝑖𝑖 cos𝜃𝜃𝑖𝑖 sin𝜃𝜃𝑖𝑖
(5) 𝑎𝑎𝐿𝐿𝐿𝐿 𝑒𝑒−𝑖𝑖𝑖𝑖(𝑡𝑡𝑖𝑖 +𝑥𝑥 cos𝜃𝜃𝑖𝑖 ∕𝑣𝑣𝐿𝐿 (𝜔𝜔)) .
∑
𝑠𝑠𝐿𝐿 (𝑥𝑥𝑥 𝑥𝑥) = 𝐴𝐴(𝜔𝜔) −
𝑖𝑖=1
𝑣𝑣(𝜔𝜔)

To include the gauge length effect of the DAS interrogator (Dean et al., 2017), we approximate the DAS measure-
ments as the average of the point-wise strain over the gauge length Lg. Then DAS Rayleigh waves measurements
can be expressed as follows:
𝑥𝑥+𝐿𝐿 ∕2

𝐿𝐿𝑔𝑔 ∫𝑥𝑥−𝐿𝐿𝑔𝑔 ∕2
𝑔𝑔
1
(6)
𝑠𝑠̄𝑅𝑅 (𝑥𝑥𝑥 𝑥𝑥) = 𝑠𝑠𝑅𝑅 (𝑙𝑙𝑙𝑙𝑙)𝑑𝑑𝑑𝑑𝑑

and DAS Love waves measurements can be expressed as follows:

𝑥𝑥+𝐿𝐿 ∕2

𝐿𝐿𝑔𝑔 ∫𝑥𝑥−𝐿𝐿𝑔𝑔 ∕2
𝑔𝑔
1
(7)
𝑠𝑠̄𝐿𝐿 (𝑥𝑥𝑥 𝑥𝑥) = 𝑠𝑠𝐿𝐿 (𝑙𝑙𝑙𝑙𝑙)𝑑𝑑𝑑𝑑𝑑

Similarly, DAS measurements of the other orientation of the array can be derived. By varying angle θi, we can
obtain the final DAS measurements of the whole close L-shaped receiver array. When the fiber-optic cable takes a
90° corner, Rayleigh waves strain can change significantly in their amplitudes without ever changing their polarity.
In contrast, Love waves strain always flips its polarity without amplitude changes. These conclusions are consist-
ent with those obtained in the literature (e.g., Benioff, 1935; Lindsey et al., 2017; Yu et al., 2019). Such measure-
ment differences in Rayleigh and Love waves are exploited in our methodology to extract their relative amplitudes.

2.2. R/L Ratio Estimation

The horizontally deployed DAS array is able to record both Rayleigh and Love waves. As we do not have suffi-
cient knowledge about the sources, they cannot be separated. Thus, we estimate the R/L ratio in the data based
on the resolved source propagation direction, array geometry, and amplitude of the seismic ambient noise data.
When all ambient noise sources travel from one direction, we can rewrite Equations 6 and 7 as follows:

𝑖𝜔cos2 𝜃
𝑠̄𝑅 (𝑥, 𝜔) = −
(8) 𝐴(𝜔)𝑅(𝑥, 𝜔),
𝑣𝑅 (𝜔)

𝑖𝑖𝑖𝑖 cos 𝜃𝜃 sin 𝜃𝜃

(9)
𝑠𝑠̄𝐿𝐿 (𝑥𝑥𝑥 𝑥𝑥) = − 𝐴𝐴(𝜔𝜔)𝐿𝐿(𝑥𝑥𝑥 𝑥𝑥),
𝑣𝑣𝐿𝐿 (𝜔𝜔)

where
𝑁𝑁𝑠𝑠
𝑥𝑥+𝐿𝐿𝑔𝑔 ∕2 ∑

∫𝑥𝑥−𝐿𝐿𝑔𝑔 ∕2 𝑖𝑖=1
(10)
𝑅𝑅(𝑥𝑥𝑥 𝑥𝑥) = 𝑎𝑎𝑅𝑅𝑅𝑅 𝑒𝑒−𝑖𝑖𝑖𝑖(𝑡𝑡𝑖𝑖 +𝑦𝑦 cos 𝜃𝜃∕𝑣𝑣𝑅𝑅 (𝜔𝜔)) 𝑑𝑑𝑑𝑑𝑑

ZHAO ET AL. 4 of 17
 21699356, 2023, 8, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022JB026339 by National Institutes Of Health Malaysia, Wiley Online Library on [19/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2022JB026339

𝑁𝑁𝑠𝑠
𝑥𝑥+𝐿𝐿𝑔𝑔 ∕2 ∑

∫𝑥𝑥−𝐿𝐿𝑔𝑔 ∕2 𝑖𝑖=1
(11)
𝐿𝐿(𝑥𝑥𝑥 𝑥𝑥) = 𝑎𝑎𝐿𝐿𝐿𝐿 𝑒𝑒−𝑖𝑖𝑖𝑖(𝑡𝑡𝑖𝑖 +𝑦𝑦 cos 𝜃𝜃∕𝑣𝑣𝐿𝐿 (𝜔𝜔)) 𝑑𝑑𝑑𝑑𝑑

Then, the DAS measurements at frequency ω recorded at locations x1 and x2 (Figure 1) are:

(12)
𝐷𝐷1 (𝑥𝑥1 , 𝜔𝜔) = 𝑠𝑠̄𝑅𝑅 (𝑥𝑥1 , 𝜔𝜔) + 𝑠𝑠̄𝐿𝐿 (𝑥𝑥1 , 𝜔𝜔),

and

(13)
𝐷𝐷2 (𝑥𝑥2 , 𝜔𝜔) = 𝑠𝑠̄𝑅𝑅 (𝑥𝑥2 , 𝜔𝜔) + 𝑠𝑠̄𝐿𝐿 (𝑥𝑥2 , 𝜔𝜔).

Then, we can obtain the power spectrum ratio of D1(x1, ω) to D2(x2, ω):
∗
|𝐷𝐷1 |2 𝐷𝐷1 (𝑥𝑥1 , 𝜔𝜔)𝐷𝐷1 (𝑥𝑥1 , 𝜔𝜔) (𝑠𝑠̄𝑅𝑅 (𝑥𝑥1 , 𝜔𝜔) + 𝑠𝑠̄𝐿𝐿 (𝑥𝑥1 , 𝜔𝜔))(𝑠𝑠̄𝑅𝑅 (𝑥𝑥1 , 𝜔𝜔) + 𝑠𝑠̄𝐿𝐿 (𝑥𝑥1 , 𝜔𝜔))∗
(14) = = .
|𝐷𝐷2 |2 𝐷𝐷2 (𝑥𝑥2 , 𝜔𝜔)𝐷𝐷2∗ (𝑥𝑥2 , 𝜔𝜔) (𝑠𝑠̄𝑅𝑅 (𝑥𝑥2 , 𝜔𝜔) + 𝑠𝑠̄𝐿𝐿 (𝑥𝑥2 , 𝜔𝜔))(𝑠𝑠̄𝑅𝑅 (𝑥𝑥2 , 𝜔𝜔) + 𝑠𝑠̄𝐿𝐿 (𝑥𝑥2 , 𝜔𝜔))∗

Assuming =𝐴𝐴𝛼𝛼(𝜔𝜔), |𝑅𝑅| = 𝑚𝑚, the power spectral ratio of D1(x1, ω) to D2(x2, ω) can be obtained according to
𝑣𝑣 (𝜔𝜔)
𝐿𝐿
𝐴𝐴 𝑣𝑣 (𝜔𝜔)
𝑅𝑅 |𝐿𝐿|

Equations 8–11 and 14:

|𝐷𝐷1 |2 𝑎𝑎2 𝑚𝑚2 𝛼𝛼(𝜔𝜔)2 |𝐿𝐿|2 + 𝑏𝑏2 |𝐿𝐿|2 + 2𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎(𝜔𝜔)(𝑅𝑅𝑟𝑟 𝐿𝐿𝑟𝑟 + 𝑅𝑅𝑖𝑖 𝐿𝐿𝑖𝑖 )
(15) = ,
|𝐷𝐷2 |2 𝑐𝑐 2 𝑚𝑚2 𝛼𝛼(𝜔𝜔)2 |𝐿𝐿|2 + 𝑑𝑑 2 |𝐿𝐿|2 + 2𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐(𝜔𝜔)(𝑅𝑅𝑟𝑟 𝐿𝐿𝑟𝑟 + 𝑅𝑅𝑖𝑖 𝐿𝐿𝑖𝑖 )

where a = −cos 2θ1, b = −cos θ1 sin θ1, c = −cos 2θ2, and d = −cos θ2 sin θ2. θ1 and θ2 denote the intersection angle
between the ambient noise source propagation direction and two orientations of the DAS array, respectively. Rr
and Lr denote the real parts of the Rayleigh and Love waves, respectively. Ri and Li are the imaginary parts of the
Rayleigh and Love waves, respectively.

We simplify Equation 15 by assuming that the interactions between the plane surface waves generated by differ-
ent sources are negligible. Then Equation 15 can be approximated as follows:

|𝐷𝐷1 |2 𝑎𝑎2 𝑚𝑚2 𝛼𝛼(𝜔𝜔)2 + 𝑏𝑏2 + 2𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎(𝜔𝜔)cos(𝜔𝜔𝜔𝜔1 𝑒𝑒1 𝛿𝛿𝑣𝑣 (𝜔𝜔))

(16) 2
≈ ,
|𝐷𝐷2 | 𝑐𝑐 2 𝑚𝑚2 𝛼𝛼(𝜔𝜔)2 + 𝑑𝑑 2 + 2𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐(𝜔𝜔)cos(𝜔𝜔𝜔𝜔2 𝑒𝑒2 𝛿𝛿𝑣𝑣 (𝜔𝜔))

where e1 = cos θ1, e2 = cos

𝐴𝐴 θ2, 𝐴𝐴(𝜔𝜔) = 𝐴𝐴, and 𝐴𝐴𝑣𝑣 (𝜔𝜔) = .
𝑣𝑣𝐿𝐿 (𝜔𝜔) 1 1
𝑣𝑣𝑅𝑅 (𝜔𝜔) 𝑣𝑣𝑅𝑅 (𝜔𝜔)
− 𝑣𝑣𝐿𝐿 (𝜔𝜔)

We only consider the fundamental mode surface waves in the DAS ambient noise data and assume vR(ω) and
vL(ω) are constants at higher frequencies. We use the DAS ambient noise data at higher frequency bands to esti-
mate the R/L ratio. The range of vR(ω) at higher frequencies can be obtained from the dispersion curve extracted
from the seismic ambient noise. Additionally, a value range can be assigned to α(ω) based on the local geological
condition. With the grid search method, we can solve the R/L ratio m.

3. Synthetic Data Examples

We use the MUSIC beamforming technique (Gelius et al., 2013; Godara, 1997; Goldstein & Archuleta, 1987;
Kirlin, 1992; Zhang et al., 2019) to estimate the ambient noise source propagation direction. The beamforming
spectrum can be calculated from the seismic data or the corresponding common virtual-shot gather. This section
presents the beamforming results resolved from common virtual-shot gathers extracted from the synthetic DAS
ambient noise recordings with different R/L ratios when the ambient noise sources come from one direction,
two directions (Figures S1 and S2 in Supporting Information S1), and multiple uniformly distributed directions
(Figure S3 in Supporting Information S1).

The receiver array is designed according to the southwest corner of the Stanford DAS Array-1 (e.g., Biondi
et al., 2017; Martin et al., 2017; Yuan et al., 2017). The two orientations of the array are about 85° from each
other (Figure 2a). We use the published 1-D shear-wave velocity model (Spica et al., 2020) to calculate the phase
velocities (Figure 2b). The velocity model was inverted from month-long DAS ambient noise data recorded
by the same DAS array (Spica et al., 2020). The phase velocities are calculated using the Thomson-Haskell
method (Buchen & Ben-Hador, 1996; Haskell, 1953; Thomson, 1950) with the open-source software evodcinv

ZHAO ET AL. 5 of 17
 21699356, 2023, 8, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022JB026339 by National Institutes Of Health Malaysia, Wiley Online Library on [19/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2022JB026339

Figure 2. (a) Layout of the receiver array for the synthetic seismic ambient noise data examples, (b) phase velocities of Rayleigh (blue curve) and Love waves (red
curve) calculated from the published 1-D shear-wave velocity model (Spica et al., 2020). They are used to simulate the seismic ambient noise data.

(Luu, 2019). Overall, the phase velocities of Love waves (red curve) are slightly higher than that of Rayleigh
waves (blue curve). We use Rayleigh waves phase velocities (blue curve) to simulate both the displacement
recorded by the vertical component of geophone and DAS Rayleigh waves measurements. We use Love waves
phase velocities (red curve) to simulate DAS Love waves measurements.

The time interval of the synthetic seismic ambient noise data is 2 ms, the same as the field data to be discussed.
During the 2 min recording period, we assume 100 plane surface waves propagate from designated directions
for each data. The amplitude of the source wavelet is obtained from the preprocessed field DAS ambient noise
data to be addressed in the field results section. The gauge length is set as 7 m based on the field settings (Martin
et al., 2017). We define the direction of propagation toward the East as 0°, and the angle increases to 90° from
East to North. Channel 55 is chosen as the virtual source for calculating the common virtual-shot gather, as it is
located almost equidistantly from both orientations of the array and also sufficiently far from the corner, thereby
minimizing the impact of the complex strain measured at the bending portion (Luo et al., 2020). We apply a
2–10 Hz bandpass filter to all synthetic data.

3.1. Single Surface Wave Type With One Source Direction

When all the plane waves generated by the motor vehicles propagate from the directions of 110°, the intersection
angle between the wave propagation direction and the x-line and y-line are 58° and 37°, respectively. We simulate
the vertical displacement (Figure 3a), horizontal DAS Rayleigh waves measurement (Figure 3d), and horizontal
DAS Love waves measurement (Figure 3g), respectively. The DAS Rayleigh waves measurement (Figure 3d)
shows the amplitude imbalances between the data recorded by different orientations of the DAS array, which is
caused by the unequal projection of Rayleigh waves. There is almost no amplitude imbalance in the DAS Love
waves measurement (Figure 3g), which is consistent with the conclusions we have obtained from the equations
in the methodology section.

Figures 3b, 3e, and 3h show the corresponding common virtual-shot gathering of Figures 3a, 3d, and 3g, respec-
tively. We can also observe the amplitude imbalances on the common virtual-shot gather obtained from DAS
Rayleigh waves measurement (Figure 3e). It is difficult to determine if there is a polarity reversal on the raw
ambient noise recordings. However, we can see from the common virtual-shot gathers, which maintain the polar-
ity information, that the polarity is consistent at the corner for DAS Rayleigh waves measurement while there is
a polarity reversal for DAS Love waves measurement. Therefore, we apply a polarity flip to the DAS Love waves
measurement recorded by one orientation of the array before beamforming. Figures 3c, 3f, and 3i display the
corresponding beamforming results resolved from Figures 3b, 3e, and 3h, respectively. The estimated ambient
noise source directions match well with the reference directions for all three beamforming spectra. However, the
resolution of the beamforming result resolved from the DAS Rayleigh waves measurement is low (Figure 3f).
This is due to the unequal amplitude projection of Rayleigh waves onto the two orientations of the DAS array.

ZHAO ET AL. 6 of 17
 21699356, 2023, 8, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022JB026339 by National Institutes Of Health Malaysia, Wiley Online Library on [19/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2022JB026339

Figure 3. When the propagation direction of the ambient noise sources is 110°, (a) synthetic seismic ambient noise data recorded by the vertical component of the
geophone array, (b) its corresponding common virtual-shot gather, (c) multiple signal classification beamforming spectra calculated from (b). Note that if there is a
polarity reversal in the data, a polarity flip will be applied to it before calculating the beamforming spectra. (d), (e), and (f) are the same as (a), (b), and (c), respectively,
except that (d) is distributed acoustic sensing (DAS) Rayleigh waves measurement. (g), (h), and (i) are the same as (a), (b), and (c), respectively, except that (g) is DAS
Love waves measurement.

When the plane waves come from the direction of 185°, the intersection angle between the wave propagation
direction and the x-line and y-line are 17° and 112°, respectively. The source propagation direction is almost
parallel to one orientation of the array while almost perpendicular to the other. Thus, the amplitude projec-
tion is severe, and the amplitude imbalances are obvious in the DAS Rayleigh waves measurement (Figures 4e
and 4f). As a result, the corresponding beamforming spectrum (Figure 4g) is blurred. To solve this problem, we
apply amplitude normalization to the DAS Rayleigh waves measurement using the root-mean-square ratio of the
data recorded by the two orientations of the array. This involves selecting several traces from each orientation,
calculating the root mean square amplitudes (Ax and Ay) for the x-line and y-line, respectively, and scaling the
y-line data by the ratio Ax /Ay.

MUSIC beamforming spectra obtained from the common virtual-shot gather calculated from the normalized data
are shown in Figures 4d, 4h, and 4l, respectively. These results demonstrate that beamforming resolves the main

ZHAO ET AL. 7 of 17
 21699356, 2023, 8, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022JB026339 by National Institutes Of Health Malaysia, Wiley Online Library on [19/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2022JB026339

Figure 4. When the propagation direction of the ambient noise sources is 185°, (a) synthetic seismic ambient noise data recorded by the vertical component of the
geophone array, (b) the corresponding common virtual-shot gather of (a), (c) multiple signal classification (MUSIC) beamforming spectra calculated from (b), and (d)
MUSIC beamforming spectra calculated from the common virtual-shot gather obtained from the normalized version of (a). (e), (f), (g), and (h) are the same as (a), (b),
(c), and (d), respectively, except that the (e) is distributed acoustic sensing (DAS) Rayleigh waves measurement. (i), (j), (k), and (l) are the same as (a), (b), (c), and (d),
respectively, except that (i) is DAS Love waves measurement.

source direction from each data. In addition, Figures 4k and 4l show that amplitude normalization is not critical
for high-resolution beamforming of DAS Love waves measurement compared to DAS Rayleigh waves measure-
ment. However, when the ambient noise sources come from multiple directions, beamforming may not resolve
all the directions (Figures S1–S3 in Supporting Information S1) due to the limitation of beamforming resolution
(Johnson & Dudgeon, 1993) and the complexity of the data.

3.2. Two Surface Wave Types With One Source Direction

Field DAS ambient noise recordings are often more complex when Rayleigh and Love waves are mixed up. When
all plane waves come from one direction (185°), we test three cases where Rayleigh and Love waves occupy
varying proportions in the ambient noise data: 75% Rayleigh waves and 25% Love waves, 50% Rayleigh waves
and 50% Love waves, and 25% Rayleigh waves and 75% Love waves. Figures 5a, 5e, and 5i display the synthetic
DAS ambient noise data for the three cases, respectively. We observe significant amplitude imbalances in the two
orientations of the seismic ambient noise data. The common virtual-shot gathers (Figures 5b, 5f, and 5j) show
no polarity reversal when there are more Rayleigh waves than Love waves and vice versa. The resolution of the
beamforming spectra (Figures 5c and 5g) calculated from Figures 5b and 5f is very low. As a result, it is difficult
to identify the source propagation direction from them. On the other hand, the high-resolution results displayed
in Figures 5d, 5h, and 5i suggest that the beamforming spectra are reliable for all three cases where beamforming
is applied to the data with amplitude normalization and polarity flip (only if there is polarity reversal).

Tables 1 and 2 summarize the beamforming results obtained from DAS ambient noise with varying portions
of Rayleigh and Love waves when the ambient noise source propagation directions are 110° and 185°,

ZHAO ET AL. 8 of 17
 21699356, 2023, 8, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022JB026339 by National Institutes Of Health Malaysia, Wiley Online Library on [19/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2022JB026339

Figure 5. Same as Figure 4 except that the data are synthetic seismic ambient noise data recorded by the horizontally deployed distributed acoustic sensing array,
which contains (a) 75% Rayleigh waves and 25% Love waves, (e) 50% Rayleigh waves and 50% Love waves, and (i) 25% Rayleigh waves and 75% Love waves.

respectively. As we saw in Figure 3d, when the wave propagation direction is 110°, the amplitude imbalances
in the DAS Rayleigh waves measurement are relatively slight since the wave's propagation direction has
similar intersections angles with both directions of the receiver array. In contrast, when the wave propagation
direction is 185°, the amplitude imbalances in the DAS Rayleigh wave measurement are severe (Figure 4).
Table 1 suggests that when the amplitude imbalances are slight, amplitude normalization and polarity correc-
tion are not critical for beamforming. Table 2 shows that when the data is not normalized, a decrease in the
R/L ratio results in improved source direction resolution through beamforming. This can be attributed to the
decrease in amplitude imbalances between the ambient noise data recorded by the two orientations of the DAS
array as the contribution of Rayleigh waves decreases. It also suggests that when the amplitude imbalances
are significant in the DAS ambient noise data, amplitude normalization and polarity correction could signif-
icantly improve the accuracy of the beamforming results. Thus, careful preprocessing of the data is essential.
Although the resolved source propagation direction may not be exactly the same as the reference, the errors
are negligible. The results can be used for further studies, such as apparent velocity correction and R/L ratio
estimation.

Table 1
Multiple Signal Classification Beamforming Results (Reference Source Propagation Direction: 110°)
R:L 10:1 5:1 2:1 1:1 1:2 1:5 1:10
Polarity reversal No No Yes Yes Yes Yes Yes
Source direction (raw) 110° 111° 113° 108° 109° 110° 110°
Source direction (normalized) 110° 111° 111° 108° 109° 109° 110°

ZHAO ET AL. 9 of 17
 21699356, 2023, 8, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022JB026339 by National Institutes Of Health Malaysia, Wiley Online Library on [19/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2022JB026339

Table 2
Multiple Signal Classification Beamforming Results (Reference Source Propagation Direction: 185°)
R:L 10:1 5:1 2:1 1:1 1:2 1:5 1:10
Polarity reversal No No Yes Yes Yes Yes Yes
Source direction (raw) − − − − 184° 184° 185°
Source direction (normalized) 185° 186° 181° 184° 184° 184° 185°

3.3. R/L Ratio Estimation

To evaluate the effectiveness of the proposed method, we conduct simulations of DAS ambient noise data with
varying R/L ratios. We assume there are 100 sources, the source propagation direction is 110°, each source
generates Rayleigh and Love waves simultaneously at random times, and the recording period is 2 min. We
assume the pulse amplitude of Love waves generated by each source is an independent sample from the uniform
distribution within the range of aL ∼ U(−1, 1). For the same source, the generated pulse amplitude of Rayleigh
waves is aR = (1 + ϵ)maL, where ϵ ∼ U(−0.2, 0.2). These assumptions result in an averaged R/L ratio of m across
all sources.

The estimation of the R/L ratio m requires knowledge of the phase velocity of Rayleigh waves and the ratio
between the phase velocities of Rayleigh and Love waves. To obtain the phase velocity of Rayleigh waves, we
extract the dispersion curve from the data, which provides a rough range of phase velocities at higher frequen-
cies (8–10 Hz). Additionally, we assume that the ratio between the phase velocities of Love and Rayleigh waves
follows a uniform distribution within the range of α ∼ U[1.0, 1.3]. Furthermore, additional constraints on m can
be imposed based on the polarity information at the corner of the array and the amplitude ratio between the two
orientations. For instance, if there is a polarity reversal at the corner and the amplitude ratio is close to 1, the data
is likely to be dominated by Love waves, and m should be close to 0. With this information, we can estimate the
R/L ratio through the grid search method.

Figure 6 shows the relationships between the true R/L ratios and those estimated from the synthetic DAS ambient
noise data. Evidently, the estimated ratios are consistent with the references, demonstrating the applicability of
the proposed inversion method for a wide range of Rayleigh and Love waves combinations.

4. Field Data Results

The layout map of the Stanford DAS Array-1 (e.g., Biondi et al., 2017; Fang et al., 2020; Martin et al., 2017;
Yuan et al., 2017) is displayed in Figure 7. We use two pieces of 2 min DAS ambient noise data to calculate the

Figure 6. Comparison between the estimated Rayleigh-to-Love wave (R/L) ratios and the references when the designated (a) R/L ratios are larger than or equal to 1,
and (b) Love-to-Rayleigh wave (L/R) ratios are larger than 1.

ZHAO ET AL. 10 of 17
 21699356, 2023, 8, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022JB026339 by National Institutes Of Health Malaysia, Wiley Online Library on [19/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2022JB026339

Figure 7. (a) Layout of the Stanford distributed acoustic sensing (DAS) Array-1. The red and blue points represent two loops of the fiber-optic cable (Martin
et al., 2017). The data recorded by the fiber-optic cable marked by the yellow dashed line are used to estimate the source propagation direction. “IU” stands for
interrogator unit and marks the location of the unit. The red numbers label the channel numbers of the DAS array.

beamforming spectrum. These data were recorded by the section of the DAS array marked by the yellow dashed
line (Figure 7) and were less contaminated by the near-field traffic noise. We filter out the low-frequency quarry
blast signals (Fang et al., 2020) in the data with a 2–10 Hz bandpass filter and then remove the near-field traffic
noise with the processing procedure published in Zhao et al. (2019). Figures 8a and 8b show two 2 min preproc-
essed DAS data recorded between 11:30 a.m. and 1:30 p.m. (local time) on different days. Figures 8c and 8d
are the corresponding common virtual-shot gathers calculated from Figures 8a and 8b, respectively. The virtual
source is at channel 93, about 15 m to the corner. From the common virtual-shot gathers (Figures 8c and 8d), we
observe clear polarity reversal around the corner of the DAS array. Thus, the data must contain Love waves. In
addition, both the data (Figures 8a and 8b) and the common virtual-shot gathers (Figures 8c and 8d) show two
types of amplitude imbalances. For each orientation, we observe amplitude variations over different traces due to
variations in fiber coupling. For example, the weaker amplitudes between Channel 70 and 90 are likely due to the
presence of two manholes in this area (Martin, 2018). These channels are excluded from the R/L ratio estima-
tion. Comparing overall amplitudes between the data recorded by the two orientations, we observe systematic
amplitude imbalances that suggest an uneven projection of Rayleigh waves energy to these two different angles.

Figures 9a and 9b show the beamforming results resolved from the two DAS data (Figures 8a and 8b), respectively.
The energy of the beamforming spectra is concentrated. The resolved source propagation directions from the two data
are 111° and 112°, respectively. The two directions are consistent, suggesting that the dominant traffic noise during the
evening rush periods came from the NW direction. The Campus Drive is a major road closest to the array, at a distance
of about 500 m (Figure 7). The next major road in proximity is Sand Hill Road, located about 1 km away from the array.
Considering the attenuation, the ambient noise wavefield would be much more seriously attenuated if they came from
the Sand Hill Road. Thus, the strongest noise should come from the Campus Drive. The reliability of the resolved ambi-

ZHAO ET AL. 11 of 17
 21699356, 2023, 8, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022JB026339 by National Institutes Of Health Malaysia, Wiley Online Library on [19/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2022JB026339

Figure 8. Two-minute preprocessed distributed acoustic sensing (DAS) ambient noise data recorded by the Stanford DAS Array-1 during evening rush periods on (a) 3
January 2017 and (b) 12 April 2017. (c) and (d) are the common virtual-shot gathers extracted from (a) and (b), respectively. We use the data recorded at channel 93 as
the virtual source. Rayleigh-to-Love wave ratios are estimated with the data labeled by the red boxes.

Figure 9. Multiple signal classification beamforming spectra resolved from (a) Figure 8a, (b) Figure 8b.

ZHAO ET AL. 12 of 17
 21699356, 2023, 8, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022JB026339 by National Institutes Of Health Malaysia, Wiley Online Library on [19/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2022JB026339

Figure 10. Synthetic distributed acoustic sensing ambient noise data (a) when the source propagation direction is 111° and Rayleigh-to-Love wave (R/L) ratio is 0.68,
and (b) when the source propagation direction is 112° and R/L ratio is 0.47, (c) and (d) are the corresponding common virtual-shot gathers of (a) and (b), respectively.

ent noise propagation direction is further supported by the time lag observations on the common virtual-shot gathers
(Figures 8c and 8d). Specifically, the time lag decreases as the channel number increases, which is consistent with the
expected behavior based on the resolved direction of the ambient noise propagation.
With the estimated source directions and the known array geometry, we estimate the R/L ratios from the ampli-
tude measurements at higher frequencies (8–10 Hz) bounded by the red boxes in Figures 8a and 8b. For the R/L
ratio estimation, the search range for the phase velocity ratio between Rayleigh and Love waves α(ω) is also set
to [1.0, 1.3]. The estimated R/L ratios for these two datasets are 0.68 and 0.47, respectively. These results suggest
that, during the selected two 2 min recording periods, the DAS array recorded more Love waves than Rayleigh
waves propagating from the Campus Drive intersection in the northwest of the array.
Based on the estimated source propagation direction and R/L ratio, we simulate the corresponding DAS ambient
noise data (Figures 10a and 10b) and extract the common virtual-shot gathers (Figures 10c and 10d) to verify
their agreement with the two sets of field data. We observe the polarity reversal and amplitude imbalances
between the data recorded by the two orientations of the DAS array, which is consistent with those obtained from
the field data. However, the time lags in Figures 10c and 10d deviate from those obtained from the field data.
This is because the shear-wave velocity model (Spica et al., 2020) we utilized to simulate the dispersion curves
may not precisely match the true one.
Figure 11a shows the average amplitude spectra of the data recorded on 3 January 2017, labeled by the red box in
the upper panel (red curve) and lower panel (blue curve) in Figure 8a, respectively. The average amplitude spectra
are smoothed using a weighted moving average smoothing filter with a window size of 3. Figure 11c shows the
amplitude ratio between the two curves. Similarly, Figures 11b and 11d show the spectra and their ratio of the

ZHAO ET AL. 13 of 17
 21699356, 2023, 8, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022JB026339 by National Institutes Of Health Malaysia, Wiley Online Library on [19/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2022JB026339

Figure 11. Averaged amplitude spectrum of the field data recorded by the two orientations of the distributed acoustic sensing array displayed in (a) Figure 8a and (b)
Figure 8b, (c) and (d) are amplitude ratios calculated from (a) and (b), respectively.

DAS data recorded on 12 April 2017. Figures 11c and 11d suggest that the average amplitude ratio between
the two orientations stays roughly constant at around 0.5 across the chosen frequency range. The amplitude ratios
calculated from the two synthetic data within the same bandwidth are shown in Figure 12. The overall agreement
between the field amplitude ratios (Figures 11c and 11d) and the synthetic amplitude ratios (Figures 12c and 12d)
demonstrates the convergence of our inversion method.

5. Discussions
The distribution of ambient noise sources in urban environments is often non-uniform and can be dominated
by anthropological activities at fixed locations. Accurately resolving the propagation direction of ambient noise
sources is crucial in order to mitigate the overestimation of surface wave phase velocities. Amplitude normali-
zation and polarity flipping (when polarity reversal is observed) can enhance beamforming spectrum resolution
when the ambient noise wavefield propagates from a single direction. The improvement is significant when there
are large amplitude imbalances between the two orientations of the DAS array. When there are two directions of
propagation, the resolution of beamforming results can be affected by various factors, including the angle differ-
ence between the two directions, the intersection angles between the two directions and the array (Figures S1 and
S2 in Supporting Information S1), receiver spacing, and the geometry of the receiver array, and so on. Addition-
ally, in cases where ambient noise sources are uniformly distributed, beamforming with an L-shaped DAS array
may show a strong bias toward four parallel and orthogonal angles to the mid-line between the two orientations
of the array (Figure S3 in Supporting Information S1).
Although there are a few studies on Love waves tomography/imaging from 3C geophone/seismometer
cross-correlations (e.g., Galetti et al., 2017; H. Li et al., 2010; Lin et al., 2008), these applications to DAS
ambient noise data are rarely seen. Because DAS is a single-component receiver, Rayleigh and Love waves
cannot be separated without enough information about the sources. Calculating the R/L ratio is a reasonable and

ZHAO ET AL. 14 of 17
 21699356, 2023, 8, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022JB026339 by National Institutes Of Health Malaysia, Wiley Online Library on [19/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2022JB026339

Figure 12. Same as Figure 11 except for the results are calculated from the synthetic distributed acoustic sensing ambient noise data displayed in Figures 10a and 10b.

easy-to-implement method to get information on the content of Rayleigh and Love waves. The R/L ratio has the
potential to be used to invert the near-surface shear-wave velocity model from both Rayleigh and Love waves
simultaneously. Additionally, it is important to note that the dominant surface wave type and the R/L ratio can
experience fluctuations across both time and space. This variability arises due to the complex urban environment,
in which the ambient noise sources exhibit temporal and spatial variations (Movie S1).
We only focus on the cases when the receiver array is close to the “L” shape, which is most commonly seen for
fiber-optic cable deployment in urban settings. However, when the array is in other shapes, more complex ampli-
tude and polarity responses may be observed from DAS recordings, particularly when the source distribution
is complex and unknown. Besides, our method is based on the assumption of a 1D shear-wave velocity model,
where the near-surface is uniform and isotropic in the horizontal directions, without any lateral variations in the
near-surface structure. For small-scale investigations, such as site characterization for a building foundation, the
assumption may be reasonable. However, for large-scale investigations, such as regional seismic hazard analysis
or exploration for oil and gas reservoirs, the assumption may be too simplistic, and more sophisticated models
may be necessary. In these cases, one must take more consideration as the effectiveness of the proposed methods
may need to be further verified under these scenarios.
We assume uniform coupling at the selected locations of the fiber-optic cable and attribute the observed ampli-
tude and polarity variations to the fiber's response to different types of surface waves. However, the accuracy
of the R/L ratio may be compromised if this assumption is not satisfied. Such estimation becomes less reliable
when the coupling effect dominates the amplitude response of DAS. Therefore, data quality control for fiber
coupling is critical, particularly for real-time DAS monitoring applications.

6. Conclusions
Ambient noise sources generate both Rayleigh and Love waves. It is commonly assumed that only Rayleigh
waves can be extracted from a linear DAS array when the virtual source is co-linear with the array. Nonetheless,

ZHAO ET AL. 15 of 17
 21699356, 2023, 8, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022JB026339 by National Institutes Of Health Malaysia, Wiley Online Library on [19/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2022JB026339

our research indicates that Love waves may also be present in the extracted common virtual-shot gather. This
may be due to the non-uniform distribution of ambient noise sources in urban environments, where Love waves
cannot be canceled out by cross-correlation. Therefore, accurately estimating and utilizing both Rayleigh and
Love waves is crucial for subsequent shear-wave velocity inversion, particularly in urban environments where the
non-uniform distribution of ambient noise sources is prevalent.

The direction of source propagation can provide useful information for estimating the content of Rayleigh and
Love waves. However, the reliability of beamforming results on DAS ambient noise can be influenced by complex
measurements of DAS. Beamforming may not accurately resolve all directions, especially in cases where sources
are complex, such as those coming from multiple directions. Our findings suggest that careful data processing,
including amplitude normalization and, if necessary, polarity flip (in the case of a polarity reversal), can make
beamforming results reliable in cases where noise sources mainly come from one direction.

Despite the fact that Rayleigh and Love waves cannot be conclusively separated due to limited information on the
noise sources, the R/L ratio can still be estimated using the estimated source propagation direction, the known
array geometry, and the amplitude of the recorded DAS ambient noise data. The estimated R/L ratio has the
potential to contribute to a more accurate near-surface shear-wave velocity model by allowing for simultaneous
inversion of both Rayleigh and Love waves.

Data Availability Statement

The data used in this study is available at: https://doi.org/10.5281/zenodo.7885909 (Zhao et al., 2023b). The
associated codes are available at https://doi.org/10.5281/zenodo.7885934 (Zhao et al., 2023a).

Acknowledgments References
We thank Professor Biondo Biondi for
sharing the DAS data. We thank Dr. Gang Ajo-Franklin, J. B., Dou, S., Lindsey, N. J., Monga, I., Tracy, C., Robertson, M., et al. (2019). Distributed acoustic sensing using dark
Fang and Dr. Eileen Martin for the useful fiber for near-surface characterization and broadband seismic event detection. Scientific Reports, 9(1), 1–14. https://doi.org/10.1038/
discussions. We acknowledge Cambridge s41598-018-36675-8
Sensing Pte. Ltd. for its financial support. Benioff, H. (1935). A linear strain seismograph. Bulletin of the Seismological Society of America, 25(4), 283–309. https://doi.org/10.1785/
We thank the open-source software bssa0250040283
Madagascar (http://www.ahay.org/) and Biondi, B., Martin, E., Cole, S., Karrenbach, M., & Lindsey, N. (2017). Earthquakes analysis using data recorded by the Stanford DAS array. In
evodcinv (https://pypi.org/project/evod- SEG Technical Program Expanded Abstracts 2017 (pp. 2752–2756). Society of Exploration Geophysicists.
cinv/1.0.0/), with which we preprocess the Brooks, L. A., Townend, J., Gerstoft, P., Bannister, S., & Carter, L. (2009). Fundamental and higher-mode Rayleigh wave characteristics of ambi-
data and simulate the dispersion curves. ent seismic noise in New Zealand. Geophysical Research Letters, 36(23), L23303. https://doi.org/10.1029/2009gl040434
Buchen, P., & Ben-Hador, R. (1996). Free-mode surface-wave computations. Geophysical Journal International, 124(3), 869–887. https://doi.
org/10.1111/j.1365-246x.1996.tb05642.x
Cheng, F., Xia, J., Luo, Y., Xu, Z., Wang, L., Shen, C., et al. (2016). Multichannel analysis of passive surface waves based on cross-correlations.
Geophysics, 81(5), EN57–EN66. https://doi.org/10.1190/geo2015-0505.1
Claerbout, J. F. (1968). Synthesis of a layered medium from its acoustic transmission response. Geophysics, 33(2), 264–269. https://doi.
org/10.1190/1.1439927
Dean, T., Cuny, T., & Hartog, A. H. (2017). The effect of gauge length on axially incident P-waves measured using fibre optic distributed vibration
sensing. Geophysical Prospecting, 65(1), 184–193. https://doi.org/10.1111/1365-2478.12419
Díaz, J., Ruiz, M., Sánchez-Pastor, P. S., & Romero, P. (2017). Urban seismology: On the origin of earth vibrations within a city. Scientific
Reports, 7(1), 1–11. https://doi.org/10.1038/s41598-017-15499-y
Dou, S., Lindsey, N., Wagner, A. M., Daley, T. M., Freifeld, B., Robertson, M., et al. (2017). Distributed acoustic sensing for seismic monitoring
of the near surface: A traffic-noise interferometry case study. Scientific Reports, 7(1), 1–12. https://doi.org/10.1038/s41598-017-11986-4
Fang, G., Li, Y. E., Zhao, Y., & Martin, E. R. (2020). Urban near-surface seismic monitoring using distributed acoustic sensing. Geophysical
Research Letters, 47(6), e2019GL086115. https://doi.org/10.1029/2019gl086115
Galetti, E., Curtis, A., Baptie, B., Jenkins, D., & Nicolson, H. (2017). Transdimensional Love-wave tomography of the British Isles and
shear-velocity structure of the East Irish Sea Basin from ambient-noise interferometry. Geophysical Journal International, 208(1), 36–58.
https://doi.org/10.1093/gji/ggw286
Gelius, L.-J., Tygel, M., Takahata, A. K., Asgedom, E. G., & Serrano, D. R. (2013). High-resolution imaging of diffractions—A window-steered
MUSIC approach. Geophysics, 78(6), S255–S264. https://doi.org/10.1190/geo2013-0047.1
Godara, L. C. (1997). Application of antenna arrays to mobile communications. II. Beam-forming and direction-of-arrival considerations.
Proceedings of the IEEE, 85(8), 1195–1245. https://doi.org/10.1109/5.622504
Goldstein, P., & Archuleta, R. J. (1987). Array analysis of seismic signals. Geophysical Research Letters, 14(1), 13–16. https://doi.org/10.1029/
gl014i001p00013
Haskell, N. A. (1953). The dispersion of surface waves on multilayered media. Bulletin of the Seismological Society of America, 43(1), 17–34.
https://doi.org/10.1785/bssa0430010017
Johnson, D. H., & Dudgeon, D. E. (1993). Array signal processing: Concepts and techniques. Prentice Hall.
Kirlin, R. L. (1992). The relationship between semblance and eigenstructure velocity estimators. Geophysics, 57(8), 1027–1033. https://doi.
org/10.1190/1.1443314
Li, H., Su, W., Wang, C.-Y., Huang, Z., & Lv, Z. (2010). Ambient noise Love wave tomography in the eastern margin of the Tibetan plateau.
Tectonophysics, 491(1–4), 194–204. https://doi.org/10.1016/j.tecto.2009.12.018

ZHAO ET AL. 16 of 17
 21699356, 2023, 8, Downloaded from https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2022JB026339 by National Institutes Of Health Malaysia, Wiley Online Library on [19/05/2024]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Journal of Geophysical Research: Solid Earth 10.1029/2022JB026339

Li, Y. E., Nilot, E., & Feng, X. (2020). Observation of guided and reflection P-waves in urban ambient noise cross-correlograms. In SEG Techni-
cal Program Expanded Abstracts 2020 (pp. 2100–2104). Society of Exploration Geophysicists.
Lin, F.-C., Moschetti, M. P., & Ritzwoller, M. H. (2008). Surface wave tomography of the western United States from ambient seismic noise: Rayleigh
and Love wave phase velocity maps. Geophysical Journal International, 173(1), 281–298. https://doi.org/10.1111/j.1365-246x.2008.03720.x
Lindsey, N. J., Martin, E. R., Dreger, D. S., Freifeld, B., Cole, S., James, S. R., et al. (2017). Fiber-optic network observations of earthquake
wavefields. Geophysical Research Letters, 44(23), 11–792. https://doi.org/10.1002/2017gl075722
Luo, B., Trainor-Guitton, W., Bozdağ, E., LaFlame, L., Cole, S., & Karrenbach, M. (2020). Horizontally orthogonal distributed acoustic sensing
array for earthquake- and ambient-noise-based multichannel analysis of surface waves. Geophysical Journal International, 222(3), 2147–
2161. https://doi.org/10.1093/gji/ggaa293
Luu, K. (2019). evodcinv: Inversion of dispersion curves using evolutionary algorithms (Version 1.0.0) [Software]. Retrieved from https://github.
com/keurfonluu/evodcinv
Martin, E. R. (2018). Passive imaging and characterization of the subsurface with distributed acoustic sensing, (Ph.D. thesis). Stanford University.
Martin, E. R., & Biondi, B. L. (2017). Ambient noise interferometry across two-dimensional DAS arrays. In SEG Technical Program Expanded
Abstracts 2017 (pp. 2642–2646). Society of Exploration Geophysicists.
Martin, E. R., & Biondi, B. L. (2018). Eighteen months of continuous near-surface monitoring with DAS data collected under Stanford University.
In SEG Technical Program Expanded Abstracts 2018 (pp. 4958–4962). Society of Exploration Geophysicists.
Martin, E. R., Castillo, C. M., Cole, S., Sawasdee, P. S., Yuan, S., Clapp, R., et al. (2017). Seismic monitoring leveraging existing telecom
infrastructure at the SDASA: Active, passive, and ambient-noise analysis. The Leading Edge, 36(12), 1025–1031. https://doi.org/10.1190/
tle36121025.1
Mateeva, A., Mestayer, J., Cox, B., Kiyashchenko, D., Wills, P., Lopez, J., et al. (2012). Advances in distributed acoustic sensing (DAS) for VSP.
In SEG Technical Program Expanded Abstracts 2012 (pp. 1–5). Society of Exploration Geophysicists.
Morton, S. L., Ivanov, J., Peterie, S. L., Miller, R. D., & Livers-Douglas, A. J. (2021). Passive multichannel analysis of surface waves using 1D
and 2D receiver arrays. Geophysics, 86(6), EN63–EN75. https://doi.org/10.1190/geo2020-0104.1
Nakata, N., & Nishida, K.(2019). Body wave exploration. In N. Nakata, L. Gualtieri, & A. Fichtner (Eds.), Seismic ambient noise (Vol. 479, p. 480).
Näsholm, S. P., Iranpour, K., Wuestefeld, A., Dando, B. D., Baird, A. F., & Oye, V. (2022). Array signal processing on distributed acous-
tic sensing data: Directivity effects in slowness space. Journal of Geophysical Research: Solid Earth, 127(2), e2021JB023587. https://doi.
org/10.1029/2021jb023587
Nilot, E., Zhang, Y., Li, Y. E., & Feng, X. (2019). Deep bedrock detection based on ambient noise recorded by a short geophone array: A Singa-
pore case study. In SEG Technical Program Expanded Abstracts 2019 (pp. 3121–3125). Society of Exploration Geophysicists.
Park, C. B., Miller, R. D., & Xia, J. (1999). Multichannel analysis of surface waves. Geophysics, 64(3), 800–808. https://doi.org/10.1190/1.1444590
Snieder, R. (2004). Extracting the Green’s function from the correlation of coda waves: A derivation based on stationary phase. Physical Review
E, 69(4), 046610. https://doi.org/10.1103/physreve.69.046610
Spica, Z. J., Perton, M., Martin, E. R., Beroza, G. C., & Biondi, B. (2020). Urban seismic site characterization by fiber-optic seismology. Journal
of Geophysical Research: Solid Earth, 125(3), e2019JB018656. https://doi.org/10.1029/2019jb018656
Stehly, L., Campillo, M., & Shapiro, N. (2006). A study of the seismic noise from its long-range correlation properties. Journal of Geophysical
Research: Solid Earth, 111(B10), B10306. https://doi.org/10.1029/2005jb004237
Thomson, W. T. (1950). Transmission of elastic waves through a stratified solid medium. Journal of Applied Physics, 21(2), 89–93. https://doi.
org/10.1063/1.1699629
Tribaldos, V. R., Ajo-Franklin, J. B., Dou, S., Lindsey, N. J., Ulrich, C., Robertson, M., et al. (2021). Surface wave imaging using distributed
acoustic sensing deployed on dark fiber: Moving beyond high-frequency noise. Distributed Acoustic Sensing in Geophysics: Methods and
Applications, 197–212.
van den Ende, M. P., & Ampuero, J.-P. (2021). Evaluating seismic beamforming capabilities of distributed acoustic sensing arrays. Solid Earth,
12(4), 915–934. https://doi.org/10.5194/se-12-915-2021
Wapenaar, K. (2004). Retrieving the elastodynamic Green’s function of an arbitrary inhomogeneous medium by cross correlation. Physical
Review Letters, 93(25), 254301. https://doi.org/10.1103/physrevlett.93.254301
Yan, Y., Li, J., Huai, N., Guan, J., & Liu, H. (2022). Two-station analysis of passive surface waves with continuous wavelet transform and
plane-wave-based beamforming. Journal of Applied Geophysics, 197, 104526. https://doi.org/10.1016/j.jappgeo.2021.104526
Yu, C., Zhan, Z., Lindsey, N. J., Ajo-Franklin, J. B., & Robertson, M. (2019). The potential of DAS in teleseismic studies: Insights from the
Goldstone experiment. Geophysical Research Letters, 46(3), 1320–1328. https://doi.org/10.1029/2018gl081195
Yuan, S., Martin, E., Chang, J., Cole, S., & Biondi, B. (2017). Catalog of Northern California earthquakes recorded by DAS. SEP-Report.
Zeng, X., Lancelle, C., Thurber, C., Fratta, D., Wang, H., Lord, N., et al. (2017). Properties of noise cross-correlation functions obtained from a
distributed acoustic sensing array at Garner Valley, California. Bulletin of the Seismological Society of America, 107(2), 603–610. https://doi.
org/10.1785/0120160168
Zhang, Y., Li, Y. E., Zhang, H., & Ku, T. (2019). Near-surface site investigation by seismic interferometry using urban traffic noise in Singapore.
Geophysics, 84(2), B169–B180. https://doi.org/10.1190/geo2017-0798.1
Zhao, Y., Li, Y. E., & Fang, G. (2019). Extracting subsurface information based on extremely short period of DAS recordings. In SEG Technical
Program Expanded Abstracts 2019 (pp. 958–962). Society of Exploration Geophysicists.
Zhao, Y., Li, Y. E., & Li, B. (2023a). Music beamforming and Rayleigh-to-Love wave ratio estimation for DAS ambient noise [Software]. Zenodo.
https://doi.org/10.5281/zenodo.7885934
Zhao, Y., Li, Y. E., & Li, B. (2023b). Two-minute DAS ambient noise data recorded by Stanford DAS Array-1 [Dataset]. Zenodo. https://doi.
org/10.5281/zenodo.7885909

ZHAO ET AL. 17 of 17