SWIP: An Integrated Workflow For Surface-Wave Dispersion Inversion and Profiling

GEOPHYSICS
Downloaded 07/25/17 to 130.56.64.29. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
SWIP: an integrated workflow for surface-wave dispersion

inversion and profiling
Journal: Geophysics
Manuscript ID GEO-2016-0625.R2
Manuscript Type: Reproducible research: Geophysics papers of the future
Date Submitted by the Author: 19-May-2017
Complete List of Authors: Pasquet, Sylvain; University of Wyoming, Department of Geology and
Geophysics
Bodet, Ludovic; Sorbonnes Universitées, UPMC Univ Paris 06, UMR 7619
METIS
Keywords: surface wave, dispersion, inversion, profiling, shear wave (S-wave)
Area of Expertise: Geophysical Software and Algorithms, Seismic Inversion
This paper presented here as accepted for publication in Geophysics prior to copyediting and composition.
© 2017 Society of Exploration Geophysicists.
Page 1 of 55 GEOPHYSICS
1
2
3
4
5 SWIP: an integrated workflow for surface-wave dispersion
6 1
7
8 inversion and profiling
9
10
11 2 Sylvain Pasquet∗ and Ludovic Bodet†
12
13 ∗ University
14 3 of Wyoming, Department of Geology and Geophysics, Laramie, WY, USA
15
16 4
† Sorbonne Universités, UPMC Univ Paris 06, CNRS, EPHE, UMR 7619 METIS, Paris,
17
18 5 France
19
20
21 6 (July 20, 2017)
22
23 7 Running head: Surface-wave inversion and profiling
24
25
26
27
ABSTRACT
28
29
30
31 8 The simultaneous estimation of 2D pressure (P-) and shear (S-) wave velocities (VP and VS ,
32
33 9 respectively) is a promising approach for imaging subsurface mechanical properties. It can be
34
35 10 performed with a single acquisition setup by combining P-wave refraction and surface-wave
36
37
38 11 (SW) analysis. While SW methods are commonly applied for the 1D estimation of VS , 2D
39
40 12 profiling requires the implementation of specific processing and inversion tools not yet widely
41
42 13 available in the community. Here we present an open-source MATLAB-based package that
43
44
45 14 performs SW inversion and profiling (SWIP) so as to retrieve 1D to 2D variations of VS from
46
47 15 any kind of linear active-source near-surface seismic data. Each step of the workflow involves
48
49 16 up-to-date processing and inversion techniques, and provides ready-to-use outputs with
50
51
52 17 quality control tools. First, windowing and stacking techniques are implemented to enhance
53
54 18 signal-to-noise ratio and extract local dispersion images along the line. Then, dispersion
55
56 19 curves are picked for each window with an uncertainty range in phase velocity including
57
58
59 20 higher uncertainties at low frequency. These curves are next inverted using a Monte Carlo
60
1
GEOPHYSICS Page 2 of 55
1
2
3
4
21 approach with various parameterizations (e.g. user-defined, refraction-based). Best models
5
6
7 22 are finally selected according to their fit to the data to build an average final model with a
8
9 23 suggested investigation depth. As an example, we used SWIP to process data collected at a

10
11 24 Yellowstone hydrothermal system. Results show the benefits of estimating VP and VS from
12
13
14 25 a single seismic setup to highlight subsurface gas pathways.
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
2
1
2
3
4
INTRODUCTION
5
6
7
8 26 Seismic methods are classically used for near-surface applications (in general at depths
9
10 27 shallower than 100 m) to determine the main mechanical properties of the subsurface. More
11
12 28 particularly, the joint estimation of pressure (P-) and shear (S-) wave velocities (VP and VS ,
13
14
15 29 respectively) is often proposed for engineering purposes, such as landslide characterization
16
17 30 (Godio et al., 2006; Jongmans et al., 2009; Socco et al., 2010b; Hibert et al., 2012; Uhlemann
18
19 31 et al., 2016), fill compaction control (Uyanık, 2011; Cardarelli et al., 2014) or earthquake
20
21
22 32 site response assessment (Jongmans, 1992; Lai and Rix, 1998; Raptakis et al., 2000; Othman,
23
24 33 2005). More recently, the combined use of VP and VS has been applied to critical zone science,
25
26 34 with an increasing interest in the Poisson’s ratio derived from those velocities. For example,
27
28
29 35 this approach has been used to study subsurface weathering processes (Olona et al., 2010),
30
31 36 image hydrothermal fluids pathways (Pasquet et al., 2016b), characterize aquifer systems
32
33 37 (Turesson, 2007; Grelle and Guadagno, 2009; Mota and Monteiro Santos, 2010; Konstantaki
34
35
36 38 et al., 2013; Pasquet et al., 2015a,b), and perform time-lapse monitoring of shallow water
37
38 39 content (Bergamo et al., 2016a,b; Dangeard et al., 2016; Pasquet et al., 2016a).
39
40
41 40 For these near-surface studies, VP is typically retrieved with P-wave refraction tomography
42
43 41 using a flat plate and hammer source with vertical component geophones (Parsekian et al.,
44
45
42 2015). The use of this method is widespread as it is easily implemented with a one-dimensional
46
47
48 43 (1D) to three-dimensional (3D) coverage, quick to set up and relatively inexpensive. When
49
50 44 applied for the estimation of VS (e.g. Pasquet et al., 2015b), seismic refraction is mostly
51
52
45 carried out using specific sources strenuous to handle (Sheriff and Geldart, 1995; Jongmans
53
54
55 46 and Demanet, 1993; Xia et al., 2002; Haines, 2007) and horizontal component geophones
56
57 47 difficult to install horizontally (Sambuelli et al., 2001). More recently, shear-wave dominated
58
59
60
3
1
2
3
4
48 wavefields have also been recorded using either multi-component geophones with vectorial
5
6
7 49 seismic sources (Schmelzbach et al., 2016) or vertical single-component geophones with
8
9 50 source- and receiver-side gradients (Sollberger et al., 2016). Though these approaches appear
10
11 51 promising, they usually require a supplementary acquisition or specific equipment with limited
12
13
14 52 availability among practitioners. As an alternative, surface-wave (SW) seismic methods are
15
16 53 commonly proposed to infer the 1D VS structure of Earth and subsurface materials (e.g.
17
18 54 Gabriels et al., 1987; Jongmans and Demanet, 1993; Park et al., 1999; Rix et al., 2001;
19
20
21 55 O’Neill et al., 2003; Socco and Strobbia, 2004; Socco et al., 2010a; Bergamo and Socco, 2016).
22
23 56 When these methods are used to characterize near-surface two-dimensional (2D) variations
24
25 57 of VS , they are mainly implemented along linear sections using active-source prospecting
26
27
28 58 in roll-along mode. SW data are basically extracted and inverted from typical seismic shot
29
30 59 gathers, providing a collection of 1D VS profiles merged to obtain a pseudo-2D VS section
31
32 60 (Socco and Strobbia, 2004; Socco et al., 2010a).
33
34
35 61 Though numerous techniques have been proposed to process and invert SW data over the
36
37
38
62 last 15 years, only a few of them have been made available to the practitioners’ community,
39
40 63 especially when it comes to 2D profiling. In order to address this shortcoming, we present
41
42 64 here a free and open-source MATLAB-based software package that performs Surface-Wave
43
44
65 dispersion Inversion and Profiling (SWIP, available at https://github.com/SWIPdev/SWIP/
45
46
47 66 releases). SWIP makes it possible to retrieve 1D to 2D variations of VS from any kind
48
49 67 of active-source near-surface seismic data recorded along linear profiles. It is particularly
50
51
68 adapted (but not limited) to process seismic data originally collected to estimate VP from P-
52
53
54 69 wave refraction tomography. In that case, it allows to extract supplementary VS information
55
56 70 from already acquired datasets, thus improving the understanding of subsurface structures
57
58
71 and processes. Each step of the implemented workflow involves up-to-date processing and
59
60
4
1
2
3
4
72 inversion techniques, mostly available in the literature. These different steps are integrated
5
6
7 73 within five MATLAB modules, each one automatically calling the necessary functions and
8
9 74 software. Specifically, seismic data are handled with the open-source software package Seismic
10
11 75 Unix, while the inversion is performed using the open-source software package Geopsy. In
12
13
14 76 the following, we first give a short review of the existing strategies and methods used for
15
16 77 SW profiling. We then present the selected approach with a detailed description of each
17
18 78 processing step, including corresponding theoretical background, technical elements and
19
20
21 79 associated notations. We finally use field data collected at a shallow hydrothermal system
22
23 80 in Yellowstone National Park (USA) to illustrate how SWIP provides ready-to-use outputs
24
25 81 with extensive quality control tools.
26
27
28
29
30 BACKGROUND METHODOLOGY: FROM 1D TO 2D
31
32
33 82 Surface waves are, basically, guided along the Earth’s surface with their amplitude decreasing
34
35 83 exponentially with depth. Most of their energy is thus confined in the shallow subsurface
36
37
38 84 where mechanical properties usually vary significantly with depth. SW propagation velocities
39
40 85 are therefore different for high frequencies (short wavelengths) confined in the shallow
41
42 86 subsurface than for low frequencies which are influenced by deeper materials. In addition,
43
44
45 87 these velocities are strongly linked to the shear properties of the material through which
46
47 88 SW propagate. The aforementioned velocity dependence on frequency, commonly refered
48
49 89 as dispersion, is then related to VS of the medium. As SW are of higher amplitudes than
50
51
52 90 body waves, this dispersion can be quite easily extracted from typical shot gathers using
53
54 91 a wavefield transform. The shot gathers, originally recorded in the distance-time domain,
55
56 92 are transformed into the frequency-wavenumber (or frequency-phase velocity) domain, in
57
58
59 93 which maxima should correspond to SW propagation modes (McMechan and Yedlin, 1981;
60
5
1
2
3
4
94 Russel, 1987; Mokhtar et al., 1988; Sheriff and Geldart, 1995; Strobbia and Foti, 2006). As
5
6
7 95 1D forward modeling techniques of SW dispersion are well-established, straightforward and
8
9 96 of low computational cost, it is possible to invert these data for a 1D VS structure along
10
11 97 the seismic spread (see for instance reviews of Socco and Strobbia (2004) and Socco et al.
12
13
14 98 (2010a) for more details about theoretical and methodological aspect as well as classical
15
16 99 literature about the topic).
17
18
19 100 In the early 2000’s, SW methods have rapidly grown in popularity among near surface
20
21 101 practitioners with their access to multichannel equipment and the development of dedicated
22
23
24 102 techniques and associated software. Among those techniques, the Multichannel Analysis of
25
26 103 Surface Waves (Miller et al., 1999; Park et al., 1999; Xia et al., 1999) pioneered 2D profiling by
27
28 104 repeating identical single-shot seismic acquisitions along a profile, then extracting dispersion
29
30
31
105 curves from each of these acquisitions. When targeting 2D shallow structures with strong
32
33 106 lateral variability, this method is however limited by the assumption that the probed medium
34
35 107 is horizontally layered below each acquisition setup, as imposed by its 1D inverse problem
36
37
38
108 formulation (Sambridge and Mosegaard, 2002). In the presence of strong lateral variations,
39
40 109 the dispersion image obtained from a classical seismic setup corresponds to a 1D equivalent
41
42 110 medium that can not be used to infer 2D characteristics of the true medium (Bodet et al.,
43
44
111 2005). Though these images could be interpreted in 2D using full-waveform seismic modeling
45
46
47 112 (e.g. Martin and Komatitsch, 2009; Dhemaied et al., 2011), the computational cost of such
48
49 113 approach has led scientists to look for more practical and time-efficient solutions.
50
51
52 114 Taking advantage of redundant seismic data, several techniques have been developed to
53
54 115 overcome these limitations. For example, O’Neill et al. (2003) proposed a more comprehensive
55
56
57 116 strategy providing local dispersion images along a profile, using a fixed acquisition setup and
58
59 117 several sources interspersed between the geophones. The procedure consists in computing
60
6
1
2
3
4
118 dispersion images from identical subsets of the seismic setup illuminated by different shots,
5
6
7 119 then stacking the dispersion images obtained for each subset to increase the signal-to-noise
8
9 120 ratio and enhance the identification of SW propagation modes. In the meantime, Hayashi
10
11 121 and Suzuki (2004) proposed an approach based on the analysis of common midpoints
12
13
14 122 cross-correlations (CMPCC). With this technique, cross-correlations are initially calculated
15
16 123 for each shot between all pairs of traces. Then, those having identical common midpoints
17
18 124 (CMP) are combined and sorted by the distance between each pair of traces, resulting in
19
20
21 125 a shot gather from which a local dispersion image associated with the CMP position can
22
23 126 be extracted. Those two main approaches have subsequently been adapted and applied to
24
25 127 numerous geophysical problems. For instance, Bohlen et al. (2004) used a gaussian moving
26
27
28 128 window to extract the dispersion of Scholte waves from marine seismic data, while Grandjean
29
30 129 and Bitri (2006) combined the stacking tools proposed by O’Neill et al. (2003) to the
31
32 130 cross-correlation technique developed by Hayashi and Suzuki (2004) in order to increase
33
34
35 131 the signal-to-noise ratio of local CMPCC dispersion images. Soon after, Neducza (2007)
36
37 132 proposed a generalization of the stacking and windowing techniques described by O’Neill
38
39 133 et al. (2003), introducing systematic parameters controlling the extraction of dispersion
40
41
42 134 images along a seismic profile. More recently, Boiero and Socco (2011) and Bergamo et al.
43
44 135 (2012) proposed using a series of Gaussian moving windows to extract local dispersion images
45
46 136 from a single seismic setup with a limited number of sources located on either side of the
47
48
49
137 spread, following on the work of Socco et al. (2009) and Boiero and Socco (2010). A similar
50
51 138 windowing technique was also used by Ikeda et al. (2013) to improve the lateral resolution
52
53 139 of the CMPCC method.
54
55
56 140 Once extracted from the dispersion images, dispersion curves are generally inverted
57
58
141 to estimate a 1D VS model below the extraction spread. Forward analytical and semi-
59
60
7
1
2
3
4
142 analytical modeling techniques, such as the Thomson-Haskell matrix propagator (Thomson,
5
6
7 143 1950; Haskell, 1953) or the reflection-transmission coefficients (Kennett, 1974), enable rapid
8
9 144 calculation of modal solutions of the SW dispersion relation in a 1D medium and are
10
11 145 commonly used for SW applications. These forward models are implemented in various
12
13
14 146 inversion schemes, the most common being damped or constrained least-squares techniques
15
16 147 (Lai et al., 2002; O’Neill et al., 2003; Herrmann, 2013). From a practical point of view, these
17
18 148 local optimization methods have a relatively low computational cost which for instance
19
20
21 149 facilitates the implementation of laterally constrained inversion schemes applied to the
22
23 150 2D characterization of the medium (Socco et al., 2009). Alternatively, global optimization
24
25 151 methods have been increasingly proposed to widely investigate the parameter space and
26
27
28 152 provide a more comprehensive solution (Socco and Boiero, 2008; Wathelet, 2008). Several of
29
30 153 these methods have been applied to SW inversion, including genetic algorithms (Lomax and
31
32 154 Snieder, 1994; Nagai et al., 2005; Song et al., 2012, 2015), simulated annealing (Martı́nez
33
34
35 155 et al., 2000; Beaty et al., 2002) or the neighbourhood algorithm (Sambridge, 1999a; Wathelet
36
37 156 et al., 2004).
38
39
40
41 IMPLEMENTED WORKFLOW AND BACKGROUND THEORY
42
43
44
45 157 As stated above, SWIP is mainly designed to retrieve 1D to 2D variations of VS from
46
47 158 typical near-surface seismic data collected along linear profiles with various acquisition
48
49 159 geometries (e.g. off-end shots, successive roll-along). It is particularly adapted (but not
50
51
52 160 limited) to process datasets that were originally designed to estimate VP from P-wave
53
54 161 refraction tomography and thus extract supplementary VS information. Each step of the
55
56 162 designed workflow comes as follow:
57
58
59
60
8
1
2
3
4
163 1. We implemented windowing techniques adapted from O’Neill et al. (2003) and Neducza
5
6
7 164 (2007) to narrow down the lateral extent of dispersion measurements and realistically
8
9 165 consider a 1D medium below each extraction spread, thus achieving the required lateral
10
11 166 resolution for 2D profiling.
12
13
14 167 2. Seismic data are then transformed in the the frequency-phase velocity domain where
15
16 168 phase velocities can clearly be identified. For this step, we implemented a slant stack in
17
18 169 the frequency domain, as described by Russel (1987) and Mokhtar et al. (1988).
19
20
21 170 3. To compensate the loss of spectral resolution caused by windowing the data (Gabriels
22
23 171 et al., 1987), we implemented stacking techniques, also adapted from O’Neill et al. (2003)
24
25 172 and Neducza (2007), to enhance signal-to-noise ratio, limit near-field effects, and give
26
27
28 173 access to larger wavelengths necessary for increasing investigation depth (Russel, 1987;
29
30 174 Forbriger, 2003a,b; O’Neill, 2003; Bodet et al., 2005, 2009; O’Neill and Matsuoka, 2005;
31
32 175 Zywicki and Rix, 2005).
33
34
35 176 4. Dispersion curves are extracted for each window with phase velocity uncertainty taking
36
37 177 into account resolution limitations at low-frequency, following the algorithm described by
38
39 178 O’Neill (2003).
40
41
42 179 5. These curves are then inverted for each window position using the neighbourhood algo-
43
44 180 rithm (Sambridge, 1999a; Wathelet et al., 2004) with different possible parameterizations
45
46 181 (e.g. user-defined, refraction-based).
47
48
49
182 6. Models matching the observed data within the uncertainty range (Endrun et al., 2008)
50
51 183 are selected to build a mean average or misfit-weighted final model and estimate the
52
53 184 investigation depth through their standard deviation or from empirical criteria.
54
55
185 7. 1D VS models obtained for each extraction window are ultimately merged into a pseudo-2D
56
57
58 186 section of VS (Pasquet et al., 2015b).
59
60
9
1
2
3
4
187 As stated above, these processing and inversion steps consist mostly in up-to-date tech-
5
6
7 188 niques available in the literature. However, SWIP users can easily suggest (and implement)
8
9 189 alternative approaches if more appropriate methods are required and/or more specific appli-
10
11 190 cation are targeted. A detailed description of each step, including corresponding theoretical
12
13
14 191 background, technical elements and associated notations is given in the following.
15
16
17
18 192 Extraction of dispersion
19
20
21
22
193 SWIP takes advantage of multi-shot acquisition setups to retrieve the lateral variations of SW
23
24 194 (i.e. Rayleigh or Love depending on the source and geophone component) dispersion using shot
25
26 195 gather windowing and dispersion stacking. Data-handling is achieved using the open-source
27
28
196 software package Seismic Unix (SU, available at http://www.cwp.mines.edu/cwpcodes)
29
30
31 197 after converting standard SEG2 or ASCII data into SU format. Though the wavefield
32
33 198 transform is performed after data windowing in the actual workflow, it is presented first
34
35
199 in the following in order to introduce resolution issues that are critical to understand the
36
37
38 200 relevance of data windowing and dispersion stacking. The dispersion extraction procedure is
39
40 201 illustrated by the flowchart in Figure ??.
41
42
43
44 202 Wavefield transform (Module A)
45
46
47
48 203 Seismic data are generally recorded in the distance-time domain, where the contribution
49
50 204 of SW (S(x, t)) can be described as the superposition of an infinite number of propagation
51
52
53
205 modes (Aki and Richards, 1980; Russel, 1987):
54
55
56
57 +∞ X
1
Z
58 S(x, t) = Am (ω, x) ei(ωt−km (ω)x) dω , (1)
2π −∞
59 m
60
10
1
2
3
4
206 where ω is the frequency, Am (ω, x) is the amplitude spectrum of the seismic signal, and
5
6
7 207 km (ω) the wavenumber for the propagation mode m. Phase velocitiy c and slowness p are
8
9 208 defined for each mode m as:

10
11
12
13
14
15 1
cm (ω) = , (2)
16 pm (ω)
17 ω
18 = . (3)
km (ω)
19
20
21
22 209 While these velocities are difficult to estimate in the distance-time domain, they can
23
24 210 clearly be identified in the frequency-phase velocity domain. It is therefore necessary to
25
26 211 transform the wavefield from one domain to another. For this purpose, we implemented a
27
28
29 212 slant stack in the frequency domain (p − ω stack) as defined by Russel (1987) and Mokhtar
30
31 213 et al. (1988). The p − ω stack of a multimode SW can be written as:
32
33
34
35 N
36
X A(ω, xn )ei(φn +ωpxn )
U (ω, p) = , (4)
37 A(ω, x1 )eiφ1
n=1
38
39 214 where A(ω, xn ) is the amplitude spectrum of the nth trace (n = 1, 2, ..., Nx ) at a distance xn ,
40
41
215 and φn is its phase spectrum. Before this operation, the data are multiplied by the square
42
43
44 216 root of the distance to correct for geometrical spreading, as recommended by Gabriels et al.
45
46 217 (1987). The denominator in Equation 4 is introduced to normalize each trace of the signal
47
48 218 by the first trace. Finally, the obtained dispersion image presents maxima that correspond
49
50
51 219 to seismic events propagating with the slowness p at a frequency ω. For a given mode m
52
53 220 propagating at a slowness pm (ω) and a trace spacing of ∆x , these maxima appear when
54
55 221 (Russel, 1987; Forbriger, 2003b):
56
57
58
59
60
11
1
2
3
4
5
6 2π
7 p(ω) = pm (ω) + n1 , (5)
ω∆x
8
9
10 222 where n1 is an integer. The maximum observed for n1 = 0 is the main maximum corresponding
11
12 223 to the mode m propagating at the slowness pm in the wavefield. The other maxima (n1 6= 0)
13
14 224 are aliased and depend on the spatial sampling ∆x . Aliasing appears in the interval 0 ≤ p ≤
15
16
17 225 2pm at a frequency defined by:
18
19
20
21 1
22 f (pm ) = . (6)
∆ x pm
23
24
25 226 Conversely, the amplitude of the dispersion image is minimal when:
26
27
28
29 2π
30 p(ω) = pm (ω) + n2 , (7)
31 ωNx ∆x
32
33 227 where n2 is an integer, and n2 =
6 0, ±Nx , ±2Nx , etc. The resolution in slowness (half-distance
34
35 228 between two minima) can then be defined as:
36
37
38
39
40 1
∆p = . (8)
41 f Nx ∆ x
42
43
44 229 This relationship implies that the resolution of the dispersion image directly depends
45
46 230 on the length of the seismic array (Nx ∆x ), but also on the frequency f . The resolution
47
48
49
231 decreases with the frequency, but is independent from the phase velocity. The resolution of
50
51 232 a given frequency is thus higher for an event propagating at a lower velocity. Furthermore,
52
53 233 since the low frequency (i.e. large wavelength) dispersion is influenced by the properties of
54
55
234 the deepest layers, the investigation depth is directly linked to the resolution, and hence to
56
57
58 235 the length of the seismic array.
59
60
12
1
2
3
4
236 Data windowing and dispersion stacking (Module A)
5
6
7
8 For 2D applications, the length of the extraction spread is thus a key parameter that needs
237
9
10 238 to be defined with care, based on the desired lateral resolution and investigation depth. The
11
12
13 239 spread has to be short enough to validate the 1D hypothesis (i.e. no or only small lateral
14
15 240 variations below it) required for the inverse problem (Bodet et al., 2005), while alternatively
16
17 241 it has to be large enough to ensure sufficient spectral resolution and record low frequency
18
19
20 242 dispersion data needed to increase the investigation depth (Bodet et al., 2009). Taking
21
22 243 advantage of multi-shot acquisition setups, we implemented data windowing and dispersion
23
24 244 stacking techniques (O’Neill et al., 2003; Neducza, 2007) to narrow down the lateral extent
25
26
27
245 of dispersion measurements and increase the signal-to-noise ratio necessary to perform 2D
28
29 246 profiling. In SWIP, these techniques are performed according to the following workflow
30
31 247 (Figure ??):
32
33
34
35 248 1. Select seismic data subsets centered on a specific position (Xmid) with window sizes
36
37 249 defined by a vector nW vec containing the number of traces of each window.
38
39
250 2. Select shots illuminating the selected subsets with offsets (i.e. distance between the shot
40
41
42 251 and the closest trace) ranging between dSmin and dSmax traces on the left, right or
43
44 252 both sides of each subset.
45
46
253 3. Extract the selected subsets from the shot gathers for each shot/window size pair.
47
48
49 254 4. Transform the wavefield to the frequency-phase velocity domain (dispersion image) for
50
51 255 each selected subsets.
52
53
256 5. Normalize amplitude spectrum at each frequency for each dispersion image.
54
55
56 257 6. Stack all normalized dispersion images computed at the same Xmid.
57
58 258 7. Shift the window of dW traces and repeat steps 1-6 to the next Xmid.
59
60
13
1
2
3
4
259 As mentioned above, it is critical to find the best compromise between investigation depth,
5
6
7 260 spectral and lateral resolution while keeping the 1D assumption valid for each extraction
8
9 261 spread. It is thus recommended to start the analysis with trial-and-error tests (Pasquet
10
11 262 et al., 2012) in order to determine the optimum windowing and stacking parameters, keeping
12
13
14 263 in mind that no perfect criterion has yet been defined (Pérez Solano et al., 2014). The
15
16 264 best window size (i.e. nW vec) should be determined after checking, for different window
17
18 265 sizes, the validity of the 1D approximation for each shot/window size pair centered on a
19
20
21 266 common Xmid (e.g. Bodet et al., 2005; Jongmans et al., 2009). The window size should
22
23 267 be progressively decreased, until the loss in spectral resolution becomes too high to justify
24
25 268 the gain in lateral resolution. As shown in Figure ??, SWIP also allows to stack dispersion
26
27
28 269 extracted from different window sizes to help mitigate those issues. However, this option has
29
30 270 not yet been extensively tested and requires further studies before being widely applied.
31
32
33 271 A similar procedure should be employed for the choice of the shot-window offset range
34
35 272 (i.e. dSmin and dSmax). While stacking is meant to enhance the quality of dispersion
36
37
38
273 images, mitigate near-field effect and facilitate dispersion picking, using noisy far offset shots
39
40 274 can have the opposite effect and deteriorate the final image quality. Finally, the shift between
41
42 275 two successive extraction windows should be defined depending on the expected lateral
43
44
276 variations in the studied medium. While it can range from one trace spacing to several
45
46
47 277 window lengths, using a large overlap (i.e. small dW ) between two adjacent stacking windows
48
49 278 will allow to retrieve smoothly varying dispersion images and help for visual browsing when
50
51
279 picking dispersion curves.
52
53
54
55
56
57
58
59
60
14
1
2
3
4
280 Dispersion picking and uncertainty estimation (Module A)
5
6
7
8 On each stacked dispersion image, the coherent maxima associated with the different propa-
281
9
10 282 gation modes are identified, picked and saved as dispersion curves. These curves can be picked
11
12
13 283 either manually, or with a semi-automatic procedure looking for the closest maxima around
14
15 284 each pick. The measurement uncertainties associated with these dispersion curves depends
16
17 285 directly on the resolution of the dispersion images and should be taken into account during
18
19
20 286 the inversion. When field conditions (e.g. weather, funding, manpower) prevent repeated
21
22 287 measurements, it is impossible to estimate uncertainties with an experimental statistical
23
24 288 analysis. To address such issue, O’Neill (2003) demonstrated that these uncertainties follow,
25
26
27
289 at logarithmic scale, a frequency-dependent relationship proportional to the resolution
28
29 290 envelope (and thus to the spread length), with a gaussian distribution at high frequency
30
31 291 (≥ 25 Hz), and a lorentzian distribution (Bevington and Robinson, 2002) at low frequency
32
33
292 (≤ 25 Hz). Hence, O’Neill (2003) proposed to use the following algorithm:
34
35
36
37 293 1. calculate the resolution in terms of phase velocity derived from the resolution in terms of
38
39
294 slowness (Equation 8),
40
41
42
43

44
1 1
∆c (f ) = 1 − ; (9)

45 c(f ) − 2f N1 ∆ 1
c(f ) + 1
2f Nx ∆x

x x
46
47 295 2. calculate the absolute uncertainty by logarithmic reduction of the resolution,
48
49
50
51
52 δc (f ) = 10−a ∆c (f ) , (10)
53
54
55 296 with a the logarithmic reduction factor, usually about 0.5 (O’Neill, 2003).
56
57
58
59
60
15
1
2
3
4
297 The dispersion curves and their associated uncertainties can then be resampled either in
5
6
7 298 wavelength or in frequency. A discretization in wavelength is generally recommended in order
8
9 299 to invert depth consistent data (Rix and Leipski, 1991; Wathelet et al., 2004). This also
10
11 300 prevent from giving excessive weight to high frequency samples which correspond only to the
12
13
14 301 shallowest part of the medium. The frequency and wavelength ranges of the dispersion curves
15
16 302 can be confined into reasonable boundaries using several criterions. As a first approach,
17
18 303 dispersion curves can be simply limited down to frequencies where the spectral amplitude
19
20
21 304 of the shot gather becomes too low (Pasquet et al., 2015a,b). Several authors (e.g. O’Neill,
22
23 305 2003; Bodet et al., 2005, 2009; Zywicki and Rix, 2005) also mentioned that wavelengths
24
25 306 higher than 50 % of the spread length should not be used in order to mitigate near-field
26
27
28 307 effects and prevent from velocity underestimation at low frequency. These recommendations
29
30 308 are basic rules of thumb useful to prepare inversion parameterization and mostly valid when
31
32 309 using the fundamental mode only. Such limitations in wavelength have to be reconsidered
33
34
35 310 when including higher modes since they have a significant impact on the investigation depth
36
37 311 and contrain the inversion (Gabriels et al., 1987; Xia et al., 2003). In SWIP, the resulting
38
39 312 dispersion curves are eventually presented as phase velocity pseudo-sections (e.g. Strobbia
40
41
42 313 et al., 2011; Haney and Douma, 2012; Boiero et al., 2013; Ezersky et al., 2013; Pasquet et al.,
43
44 314 2015b). This representation is very convenient for quality-control of picked dispersion and
45
46 315 more particularly to check the lateral coherence in mode identification (Zhang and Chan,
47
48
49
316 2003; O’Neill and Matsuoka, 2005; Boaga et al., 2013; Ezersky et al., 2013).
50
51
52
53 317 Inversion of dispersion
54
55
56 318 Parameterization, forward modeling and neighbourhood algorithm inversion are performed
57
58
59 319 within the Dinver tool, part of the Geopsy open-source software package (available at
60
16
1
2
3
4
320 www.geopsy.org). Assuming a horizontally layered (1D) medium below each extraction
5
6
7 321 window, SWIP performs 1D inversion of dispersion curves obtained at each Xmid position.
8
9 322 When a large overlap between two adjacent stacking windows (i.e. small dW ) is chosen,
10
11 323 stacking and windowing operations will naturally smooth lateral changes in dispersion data.
12
13
14 324 In this case (recommended in SWIP), the use of lateral constraints between successive
15
16 325 inversions is not necessary to retrieve smooth and coherent lateral variations of VS (Strobbia
17
18 326 et al., 2011). The dispersion inversion procedure is illustrated by the flowchart in Figure ??.
19
20
21
22
23 327 Inversion parameterization (Module B)
24
25
26 328 An appropriate choice of the parameters is considered fundamental to successfully performing
27
28
329 the inversion (Socco and Strobbia, 2004; Renalier et al., 2010). Usually based on a priori
29
30
31 330 knowledge (presence of weathering gradients, sedimentary layers, low velocity zone, etc.), the
32
33 331 parameterization can be defined with several layers, of fixed or varying thickness, velocities
34
35 332 (VP and VS ), and density. Velocities and density can be defined in each layer with various
36
37
38 333 depth-dependent shapes (e.g. uniform, linear increase or decrease, power law) allowing a large
39
40 334 range of possible models. The maximum half-space depth (HSD), defined by the number of
41
42 335 layers and their maximum thickness, is of great importance since it depends on the poorly
43
44
45 336 known investigation depth of the method. It is usually recommended, as a first step, to fix it
46
47 337 to the half of the maximum observed wavelength (O’Neill, 2003; Bodet et al., 2005). Since
48
49 338 P-wave velocity and density have weak constraint on SW dispersion, it is important to keep
50
51
52 339 in mind that only S-wave velocity profile can be interpreted from the inversion results (Der
53
54 340 and Landisman, 1972; Russel, 1987). While density can most of the time be set as uniform,
55
56 341 it is recommended to use an identical layering for VP and VS .
57
58
59
60
17
1
2
3
4
342 As shown above, there are no specific limitations on the parameterization. An important
5
6
7 343 number of layers (over-parameterization) should be avoided, but the parameterization
8
9 344 should still give some flexibility to the inversion algorithm (i.e. number of layers as low
10
11 345 as possible). Yet, the variability of generated models should remain important enough for
12
13
14 346 the modeled dispersion to fit possibly complex extracted data. In such situation, finding
15
16 347 a good compromise is always a delicate task, particularly when the measured dispersion
17
18 348 curves show variability along the profile. In the absence of strong a priori information, it is
19
20
21 349 recommended to:
22
23
24 350 1. Select several typical Xmid positions along the line in terms of dispersion patterns.
25
26
351 2. For each of these Xmid positions, build the simplest possible parameterization, with the
27
28
29 352 lowest possible number of layers.
30
31 353 3. Run the inversions and, step by step, give more degrees of freedom to the algorithm by
32
33
354 adding layers (and/or by extending parameter ranges) if the dispersion curves are not
34
35
36 355 sufficiently matched (Wathelet et al., 2004).
37
38
39 356 When a priori information about the probed subsurface is available along the line (e.g. from
40
41
42 357 other geophysical, geological or log data), the inversion can consist in the optimization of
43
44 358 an a priori model, rather than the exploration of all possible solutions. In that case, we
45
46 359 recommend to apply the following parameterization strategy:
47
48
49
50 360 1. Select several typical Xmid positions along the line in terms of a priori knowledge.
51
52 361 2. Build velocity structures based on a priori information with reduced thickness and
53
54 362 velocity ranges.
55
56
57 363 3. Perform forward modeling to roughly estimate if these structures present the appropriate
58
59 364 number of layers and velocity ranges.
60
18
1
2
3
4
365 4. Run the inversions and, step by step, adapt degrees of freedom given to the algorithm if
5
6
7 366 the dispersion curves tend to be correctly matched, in order to converge to a satisfying
8
9 367 parameterization according to a priori information.

10
11
12
13 368 When studying variations of VP /VS or Poisson’s ratio (Pasquet et al., 2015b) and when
14
15 369 the acquisition setup allows to perform P-wave refraction tomography, SWIP can create a
16
17 370 semi-automatic parameterization based on refraction results. In such a case, VP soundings
18
19
20 371 are extracted at each Xmid position from the tomography model and resampled according
21
22 372 to the desired parameter space discretization in depth. This average value can then be used
23
24 373 to fix VP in each layer or to estimate a limited and realistic variation range.
25
26
27
28
29 374 Neighbourhood algorithm inversion (Module C)
30
31
32 375 The inversion is performed with the neighbourhood algorithm (NA) developed by Sambridge
33
34
376 (1999a) and implemented for near-surface applications by Wathelet et al. (2004) and Wathelet
35
36
37 377 (2008) within the Dinver tool. Theoretical dispersion curves are computed from the elastic
38
39 378 parameters using the Thomson–Haskell matrix propagator technique (Thomson, 1950;
40
41
379 Haskell, 1953) as implemented by Dunkin (1965). The NA then makes use of Voronoi cells
42
43
44 380 to iteratively sample the parameter space (namely VP , VS , density and thickness of each
45
46 381 layer). At the first iteration, NA randomly generates ns0 models in the parameter space and
47
48 382 calculates the following misfit function (M F ):
49
50
51
52 v
u Nf
uX (V 2
53 cali − Vobsi )
54 MF = t , (11)
Nf σi 2
55 i=1
56
57
383 with Vcali and Vobsi , the calculated and observed phase velocities at each frequency fi ; Nf ,
58
59 384 the number of frequency samples and σi , the phase velocity measurement uncertainty at
60
19
1
2
3
4
385 each frequency fi .
5
6
7 386 The parameter space is then divided in ns0 Voronoi cells centered on each generated
8
9
10 387 model, with boundaries in each parameter direction being equidistant from the nearest
11
12 388 neighbour model. The nr best cells (i.e. with the lowest M F ) are then selected, within
13
14 389 which ns /nr new models are randomly generated. The ns new models are finally added to
15
16
17 390 the previous ones, updating the Voronoi cells distribution. This operation is repeated for nit
18
19 391 iterations until reaching ns0 + (ns nit ) generated models. The use of Voronoi cells allows to
20
21 392 adapt the parameter space along successive iterations, unlike classical Monte Carlo inversion
22
23
24 393 schemes which search for new models along a predefined grid of parameters (Wathelet, 2008).
25
26 394 Depending on the tuning parameters used, the NA can be considered very exploratory (high
27
28 395 ns0 , ns and nr ) or on the contrary tend to local optimization (low ns0 , ns and nr ). The
29
30
31
396 main goal is to find a compromise allowing to generate a large enough number of models
32
33 397 sweeping the whole parameter space, then to converge towards the more probable areas
34
35 398 without reaching local minima too quickly, while limiting the processing time.
36
37
38 399 The NA is consequently not a pure random approach. The generation of models is
39
40
400 iteratively guided toward the area of lowest M F . The advantage is obviously a lower
41
42
43 401 computational cost compared to completely random methods. Yet it has to be used with
44
45 402 care since it does not prevent from falling into local minima of the M F , more particularly
46
47 403 when using a restricted parameter space with poor a priori information. As suggested above
48
49
50 404 regarding the parameterization, we suggest to establish the inversion strategy based on
51
52 405 the strength of available a priori information. In the case of poor knowledge about the
53
54 406 investigated structures, it is recommended to tune the NA as exploratory as possible. Then,
55
56
57 407 when the areas of the M F in which the NA tends to converge seem repetitive with an
58
59 408 important number of models and a large parameter space, the inversion can be re-tuned
60
20
1
2
3
4
409 as more optimizing until finding the best compromise between the number of generated
5
6
7 410 models and processing time. The flexibility offered by the NA in terms of tuning, as well as
8
9 411 the variety of provided outputs and associated estimators, make it a rather versatile and
10
11 412 convenient tool for such SW applications (e.g. Sambridge, 1999a,b; Wathelet et al., 2004;
12
13
14 413 Wathelet, 2008).
15
16
17
18 414 Final 1D VS model estimation (Module C)
19
20
21
415 Thousands of models can be generated for each Xmid with the NA, allowing the appraisal
22
23
24 416 of an a posteriori estimate of the model error. SWIP allows for a detailed presentation
25
26 417 of 1D inversion results at each Xmid position. Computed dispersion data and models are
27
28
418 represented according to their misfit with different color scales for accepted and rejected
29
30
31 419 models (Endrun et al., 2008). This layout is useful to visually estimate the modal dispersion
32
33 420 of the explored parameter space and fairly interpret extracted results. Two options are
34
35 421 offered to select the accepted models: (i) keeping the nbest models with the lowest misfits; (ii)
36
37
38 422 or selecting all models whose calculated dispersion curves fit the observed data within the
39
40 423 uncertainty range. Using all selected models, we then propose to build a final average model
41
42 424 either by taking the mean value of each model parameter, or by weighting the different
43
44
45 425 parameters according to each model’s misfit value. In order to correctly estimate the a
46
47 426 posteriori model error, it is recommended to select at least 500 to 1000 models. In the case
48
49 427 of models selected within the uncertainty range, that can lead to run the inversion with
50
51
52 428 different NA tuning parameters in order to obtain an acceptable number of models fitting
53
54 429 within the uncertainties. Ultimately, this final average model could be used as a starting
55
56 430 model in a linearized inversion scheme, as proposed by Socco and Boiero (2008). It is worth
57
58
59 431 mentioning that SWIP does not yet provide specific tools to quantitatively draw inferences
60
21
1
2
3
4
432 from the estimated misfit (Mosegaard and Tarantola, 1995; Sambridge, 2001; Sambridge and
5
6
7 433 Mosegaard, 2002). As for SW dispersion inversion in general, various approaches existing in
8
9 434 the literature (Sambridge, 1999b; Wathelet et al., 2004; Socco and Boiero, 2008; Wathelet,
10
11 435 2008) could be implemented to exploit the full wealth of NA outputs.
12
13
14
15
16 436 Quality control and pseudo-2D VS section extraction (Module D1 and D2)
17
18
19 437 Despite these limitations, SWIP provides an extensive selection of post-inversion quality-
20
21
438 control tools. The software for instance offers to compare pseudo-sections of picked and
22
23
24 439 calculated phase velocities, along with their residuals at each Xmid position. Such represen-
25
26 440 tation of the data, adapted from electrical resistivity tomography codes, is very usefull to
27
28
441 review the inversion fit along the acquisition line and check for possible misinterpretations.
29
30
31 442 The coherence of the 1D VS models eventually extracted at each Xmid can also be verified
32
33 443 by superimposing theoretical dispersion curves on dispersion images (Pasquet et al., 2014).
34
35 444 This particular representation helps pointing out possible modes misidentification and check
36
37
38 445 if originally discarded higher modes could have been picked and inverted. Furthermore, it is
39
40 446 possible to represent the misfit value for each pair of inverted parameters and characterize
41
42 447 the resolution and sensitivity of each of those parameters (Wathelet et al., 2004).
43
44
45 448 Each final 1D VS model can finally be represented at its corresponding Xmid position
46
47
48 449 in order to create a pseudo-2D section of VS . While the investigation depth of 1D VS models
49
50 450 is usually considered equal to the half of the maximum observed wavelength (Socco and
51
52 451 Strobbia, 2004), we also propose here to take advantage of the Monte Carlo approach to
53
54
55 452 estimate the investigation depth from the standard deviation of all models selected after the
56
57 453 NA inversion. For each final 1D VS model, we propose to define the depth of investigation
58
59
60
22
1
2
3
4
454 when the standard deviation reaches a user-defined threshold above which VS is considered
5
6
7 455 unconstrained. SWIP users should then keep in mind that this final pseudo-2D VS section
8
9 456 only represents merged averages (and hence smooth) of all possible solutions at each Xmid
10
11 457 (Mosegaard and Tarantola, 1995). In most cases, the final models interpreted from such data
12
13
14 458 fitting process will, by definition, only consist in an estimation of the true VS structures.
15
16 459 However, the outputs provided by the inversion can be used to address specific questions
17
18 460 such as the depth or location of specific interfaces, the existence of velocity anomalies (low
19
20
21 461 velocity layers for instance) or the occurrence of strong lateral variations, as shown in the
22
23 462 field example presented in the following.
24
25
26
27 FIELD EXAMPLE
28
29
30
31 463 Description of the geophysical survey
32
33
34 464 We present here the results of a geophysical survey carried out in Yellowstone National
35
36
37 465 Park (USA), in the Obsidian Pool Thermal Area. It is located in the eastern part of the
38
39 466 Yellowstone caldera, within the Mud Volcano thermal area, which mainly consists of rhyolitic
40
41 467 ash flow tuff covered with varying thicknesses of glacial silts, sand, and gravel (Christiansen
42
43
44 468 and Blank, 1975). The area is also characterized by extensive diffuse degassing of CO2
45
46 469 through soils (Werner et al., 2000) and hosts several isolated thermal features, mostly of
47
48 470 acid-sulfate composition with water temperatures between 21.9◦ C and 84.0◦ C (Hurwitz
49
50
51 471 et al., 2012).
52
53
472 Seismic measurements were performed in this site to study shallow hydrothermal systems,
54
55
56 473 characterize fluid pathways and improve understanding of the depths of separation of steam
57
58 474 from liquid water. Seismic data were collected in July 2016 along a SSW-NNE transect,
59
60
23
1
2
3
4
475 crossing a heat flow anomaly (Hurwitz et al., 2012) between 50 m and 120 m and a degassing
5
6
7 476 feature (Pasquet et al., 2016b) between 86 m and 96 m. We used ten 24-channel Geometrics
8
9 477 Geode seismographs with 4.5-Hz vertical component geophones spaced every 1 m, so as to
10
11 478 obtain a 239-m long profile (Figure ??a). 25 shot gathers were recorded every 10 m using a
12
13
14 479 5.4-kg sledgehammer source swung onto a metal plate. The plate was hit 5 times at each
15
16 480 position to increase signal-to-noise ratio. The sampling rate was 0.125 ms and the recording
17
18 481 time was 0.75 s to include full SW wavefield. The hydrothermal features and the acquisition
19
20
21 482 line were GPS surveyed, and the topography was extracted from airborne LiDAR data
22
23 483 (available at http://www.opentopography.org).
24
25
26 484 First arrival times were picked manually on the shot gathers (Figure ??b) and inverted
27
28 485 for P-wave velocity using a MATLAB travel-time tomography code (St. Clair, 2015). In
29
30
31 486 order to estimate the sensitivity and the depth of investigation of our model, we repeated
32
33 487 the inversions for a range of 100 starting models with different velocity gradients and surface
34
35 488 velocities (St. Clair et al., 2015). All models presenting a satisfactory fit to the data were
36
37
38
489 used to build an average final model with a depth of investigation defined using a threshold
39
40 490 on the standard deviation of all accepted models (Pasquet et al., 2016b). The final VP
41
42 491 model shows smoothly varying velocities ranging between 100 m/s and 2000 m/s, with a
43
44
492 ∼5-m-thick, low-velocity layer at the surface (Figure ??c).
45
46
47
48
49 493 Extraction of dispersion
50
51
52 494 After converting raw SEG2 data into SU format, we used SWIP to extract dispersion images
53
54
55 495 from the seismic data. Following the recommendations formulated above, we performed
56
57 496 trial-and-error tests with different window sizes and source-window offsets. We eventually
58
59
60
24
1
2
3
4
497 used a 30-m (nW vec = 31) window with source-window offsets ranging between 0 and 20 m
5
6
7 498 on both sides of the window (dSmin = 0 and dSmax = 20). Though near-field effects can
8
9 499 appear when using very short offsets (i.e. dSmin = 0), stacking dispersion images with
10
11 500 larger offset shots should mitigate this problem (Neducza, 2007). Furthermore, the use of
12
13
14 501 higher modes and larger uncertainties at low frequencies (i.e. where near-field effects perturb
15
16 502 the data the most) also helps dealing with the under-estimation of phase velocities (Bodet
17
18 503 et al., 2009). The first and last windows were centered at 15 m and 224 m, respectively, with
19
20
21 504 3 to 6 shots illuminating each window. For each window position (Xmid), dispersion images
22
23 505 computed from each of these shots were first compared (Figure ??) to confirm the validity
24
25 506 of the 1D approximation below each spread. The clear consistency observed between single
26
27
28 507 dispersion images at each Xmid position authorized the stacking of those images.
29
30
31
508 In the example presented in Figure ?? and all along the line, stacking of dispersion
32
33 509 images clearly enhanced signal-to-noise ratio and helped identification of SW propagation
34
35 510 modes. The window was then shifted 1 m along the acquisition profile (dW = 1) to obtain
36
37
38
511 210 evenly spaced dispersion images at each Xmid position. The large overlap between two
39
40 512 adjacent stacking windows provided smoothly varying dispersion images and helped the
41
42 513 identification of propagation modes during the picking process.
43
44
45 514 We eventually picked coherent maxima associated with the identified propagation modes
46
47 515 on each stacked dispersion image obtained along the line. Adjacent dispersion images were
48
49
50 516 displayed during picking to follow the lateral evolution of different modes and avoid mode
51
52 517 misidentification. Dispersion curves were automatically cut down to frequencies where the
53
54 518 spectral amplitude of the shot gather became too low. We used a threshold of 2.5 % of
55
56
57 519 the normalized amplitude to define this frequency for each single dispersion image, then
58
59 520 used the average frequency at each Xmid to determine the low cut of the dispersion curve.
60
25
1
2
3
4
521 The cut-off frequency ranges between 5 and 10 Hz along the line, with corresponding
5
6
7 522 wavelength varying from 25 to 50 m. These dispersion curves were finally extracted with
8
9 523 their associated uncertainty in phase velocity estimated with Equations 9 and 10, then
10
11 524 resampled in wavelength every 1 m from 0 m to 50 m (Figure ??).
12
13
14 525 The fundamental mode (0) of SW was clearly identified all along the line, while the
15
16
17 526 first higher mode (1) was only identified from 15 m to 47 m, 71 m to 77 m and 125 m to
18
19 527 191 m. In order to visually inspect the lateral consistency of picked modes, we represented
20
21 528 dispersion curves as SW phase velocity pseudo-sections (Figure ??). Both pseudo-sections
22
23
24 529 present smooth lateral variations with phase velocities ranging between 100 m/s and 600 m/s.
25
26 530 Higher phase velocities are observed on the fundamental mode between 60 m and 120 m and
27
28 531 at the end of the line. The whole dispersion extraction procedure (Figure ??) was performed
29
30
31 532 over the 210 Xmid in about 45 minutes using a laptop with a quad-core 2.10 GHz processor
32
33 533 and 16 Gb of RAM (not including the picking process that can take up to a few hours).
34
35
36
37 534 Inversion of dispersion
38
39
40
41 535 Assuming a 1D medium below each extraction window, we then used SWIP to perform 1D
42
43 536 NA inversions of dispersion curves extracted at each Xmid. We applied the trial-and-error
44
45 537 strategy formulated above, testing different parameterizations and NA tuning parameters.
46
47
48 538 We finally used a parameterization with a stack of ten layers overlaying the half-space to look
49
50 539 for smooth non-linear velocity gradients, as expected with regards to P-wave refraction results
51
52 540 and a priori geological knowledge (unconsolidated sediments and weathering gradient). The
53
54
55 541 thickness of each layer was allowed to range from 0.5 m to 2.5 m. We fixed the maximum
56
57 542 half-space depth (HSD) to the half of the maximum wavelength observed along the entire
58
59
60
26
1
2
3
4
543 line (25 m). The valid parameter range for sampling velocities was 10–2500 m/s for VS with
5
6
7 544 velocities constrained to only increase with depth based on a priori geological information.
8
9 545 The range of VP was defined by tomography results after extracting at each Xmid position an
10
11 546 average VP value at each 2.5-m thick slice of the refraction tomography model (Figure ??c).
12
13
14 547 This average value was then used to reduce the range of possible VP in each layer of the
15
16 548 inversion parameterization. Poisson’s ratio was defined between 0.1 and 0.5 in order to
17
18 549 prevent from unrealistic VS values, and density was set as uniform (2000 kg/m3 ). Except
19
20
21 550 for VP which were defined to closely follow variations of the P-wave tomography model, we
22
23 551 used an identical parameterization for all 1D inversions, with no lateral constraints between
24
25 552 successive inversions, relying on the important overlap of adjacent extraction windows.
26
27
28 553 For each Xmid position, we performed two separate runs of NA with nit = 150, ns0 = 100,
29
30
31
554 ns = 75 and nr = 50, so as to generate 22700 models (Figure ??). After each 1D inversion,
32
33 555 we selected models matching the observed data within the uncertainty range (i.e. all samples
34
35 556 of the theoretical dispersion curves calculated from the model fitted the observed data within
36
37
38
557 the uncertainty range). Average parameters of all accepted models were then used to build a
39
40 558 misfit-weighted velocity structure associated with the center of the extraction window. Using
41
42 559 this average VS model and the VP model extracted from tomography results at each Xmid
43
44
560 position, we computed theoretical dispersion curves to check their fit with the observed
45
46
47 561 dispersion (Figure ??) and confirm the acceptability of the average 1D VS .
48
49
50 562 We finally compared observed and calculated phase velocity for both fundamental and
51
52 563 first higher modes at each Xmid position, and computed their residuals to verify the overall
53
54 564 consistency of the inversion along the profile (Figure ??). The final model has a RMS of
55
56
57 565 13.9 m/s, with 94.7 % of the samples with normalized phase velocity residuals less than
58
59 566 7.4 %. The whole dispersion inversion procedure (Figure ??) was performed over the 210
60
27
1
2
3
4
567 Xmid in about 4 hours using a laptop with a quad-core 2.10 GHz processor and 16 Gb of
5
6
7 568 RAM.
8
9
10 569 We then estimated the investigation depth for each Xmid position from the standard
11
12 570 deviation of all selected VS models. We used a threshold of 150 m/s on the standard deviation
13
14 571 of VS to determine the investigation depth and limit the extent of the velocity model in depth
15
16
17 572 (Figure ??a). Finally, each 1D VS model was represented at its corresponding extraction
18
19 573 position to build a pseudo-2D section of VS (Figure ??b). The VS model is characterized by
20
21 574 velocities ranging between 50 m/s and 600 m/s, with higher shallow VS below the heat flow
22
23
24 575 anomaly observed between 50 m and 120 m. Although the VS model has a lower investigation
25
26 576 depth than the VP model, it provides more information regarding the lateral variations of
27
28 577 shallow layers’ velocities due to the intrinsic smoothing of tomographic inversion and the
29
30
31 578 substantial horizontal component of P-wave travel paths.
32
33
579 When a VP model is available from P-wave tomography for instance, SWIP can also
34
35
36 580 calculate Poisson’s ratio, which is known to help identifying different lithology or water/gas
37
38 581 saturation changes in the subsurface (Pasquet et al., 2015a, 2016b). Using the inverted VP
39
40
582 (Figure ??c) and VS (Figure ??b) models, we computed Poisson’s ratio ν such as:
41
42
43
44
45 VP2 − 2VS2
46 ν= . (12)
2(VP2 − VS2 )
47
48
49 583 Poisson’s ratio (Figure ??) shows values ranging between 0.3 and 0.5, which are typical
50
51
584 of non-saturated and saturated media, respectively. Poisson’s ratio is predominantly between
52
53
54 585 0.45 and 0.5, indicating high water content for most of the subsurface, except in the highest
55
56 586 part of the hill. Lower Poisson’s ratio values are also observed at depth below the degassing
57
58
587 area visible at the surface, illustrating the ability of Poisson’s ratio to map the shallow
59
60
28
1
2
3
4
588 “plumbing” structure of hydrothermal systems and efficiently constrain gas vs water saturation
5
6
7 589 at depth. Furthermore, this interpretation is in good agreement with the results of recent
8
9 590 geophysical investigations conducted at the same site by Pasquet et al. (2016b), where rock
10
11 591 physics modeling highlighted lower saturation in the degassing area.
12
13
14
15
16 CONCLUSIONS
17
18
19 592 SWIP is an integrated open-source MATLAB-based package that performs, within the
20
21 593 same framework, SW inversion and profiling for the 1D to 2D imaging of VS . It can be
22
23
24 594 used with any kind of active-source near-surface seismic data collected along linear profiles,
25
26 595 and is particularly adapted (but not limited) to process seismic data originally recorded
27
28 596 to estimate VP from P-wave refraction tomography. Each step of its workflow involves
29
30
31 597 up-to-date processing and inversion techniques, integrated within five MATLAB modules
32
33 598 automatically calling the necessary functions and software.
34
35
36 599 SWIP takes advantage of multi-shot acquisition setups to retrieve the lateral variations
37
38 600 of SW dispersion (i.e. Rayleigh or Love depending on the source and geophone component),
39
40
601 using shot gather windowing and dispersion stacking techniques. These techniques clearly
41
42
43 602 enhance signal-to-noise ratio and make it possible to extract local dispersion images along
44
45 603 the acquisition profile. The dispersion curves are consequently picked for each window
46
47 604 with associated uncertainties in phase velocities taking into account typical low frequency
48
49
50 605 discrepencies due to the limited spectral resolution of the method and near-offset effects. These
51
52 606 curves are then represented as phase velocity pseudo-sections to enable convenient quality-
53
54 607 control of picked dispersion and lateral coherence in mode identification. The recommended
55
56
57 608 use of a large overlap between two adjacent stacking windows provides naturally smooth
58
59 609 changes in dispersion measurements along the acquisition line; hence, inverting dispersion
60
29
1
2
3
4
610 data for a pseudo-2D VS section does not require the use of lateral constraints. Instead,
5
6
7 611 SWIP uses a Monte Carlo inversion approach, with a choice of user-defined or refraction
8
9 612 tomography-based parameterization, in order to retrieve 1D VS models for each extraction

10
11 613 window. Two options are then offered to select accepted models and build a final average
12
13
14 614 model: (i) keeping the nbest models with the lowest misfits; (ii) or selecting all models whose
15
16 615 calculated dispersion curves fit the observed data within the uncertainty range. In both
17
18 616 cases, the final model can be built either by taking the mean value of each model parameter,
19
20
21 617 or by weighting the different parameters according to the misfit value of each model. We
22
23 618 finally merge all the final 1D VS models into a pseudo-2D section of VS , with a suggested
24
25 619 investigation depth estimated from the standard deviation of each Xmid accepted models.
26
27
28 620 Each step of SWIP’s workflow provides ready-to-use outputs with extensive quality control
29
30
31 621 tools, as illustrated in a field example. Seismic data collected with a single acquisition setup
32
33 622 in Yellowstone National Park (USA) were processed to demonstrate the benefits of combining
34
35 623 P-wave refraction tomography and SW dispersion inversion. In this example, the VS model
36
37
38
624 shows strong lateral variations that are not visible on the VP model (due to strong saturation
39
40 625 variations). Furthermore, the Poisson’s ratio calculated from these two models, and more
41
42 626 particularly its contrasts, clearly highlights gas pathways in the subsurface consistent with
43
44
627 degassing observed at the surface. With these results, we demonstrated SWIP’s versatility
45
46
47 628 and robust usage, and showed how it can provide supplementary information from existing
48
49 629 seismic datasets. Nonetheless, some of the implemented features still require additional
50
51
630 investigation before being systematically applied in further studies (e.g. optimum stacking
52
53
54 631 and windowing parameters, refraction-based parameterization, multi-window size stacking).
55
56 632 With this in mind, the open implementation of the software will also allow any user to
57
58
633 suggest (and/or implement) alternative approaches for the extraction (e.g. CMPCC, gaussian
59
60
30
1
2
3
4
634 window, f-k transform,...) or the inversion (e.g. linearized, laterally constrained or other
5
6
7 635 Monte Carlo scheme) of SW dispersion.
8
9
10
11 ACKNOWLEDGMENTS
12
13
14
636 We kindly thank Marine Dangeard, Amine Dhemaied, Ibrahim Rahmania and Russell
15
16
17 637 Callahan for helping to improve SWIP during its development and proof-reading the
18
19 638 manuscript. We also thank Beatriz Benjumea, Carlos Huerta and two anonymous reviewers
20
21
639 for their constructive comments on the manuscript. Finally, we gratefully acknowledge
22
23
24 640 Steven Holbrook and the Wyoming Center for Environmental Hydrology and Geophysics
25
26 641 for providing the seismic data presented in this work. These data are available through the
27
28 642 University of Wyoming Data Repository at https://doi.org/10.15786/M21S3V and can
29
30
31 643 be used to reproduce the figures presented here by following the instructions provided in
32
33 644 SWIP user’s guide. This work was funded by the CNRS, the PIREN-Seine program, the
34
35 645 CRITEX ANR-11-EQPX-0011 project, and the SOERE-H+ hydrogeological network. It was
36
37
38 646 also supported by the Wyoming Experimental Program to Stimulate Competitive Research
39
40 647 and by the National Science Foundation under award EPS-1208909.
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
31
1
2
3
4
REFERENCES
5
6
7 648 Aki, K., and P. G. Richards, 1980, Quantitative seismology: University Science Books.
8
9
10 649 Beaty, K. S., D. R. Schmitt, and M. Sacchi, 2002, Simulated annealing inversion of multimode
11
12 650 Rayleigh wave dispersion curves for geological structure: Geophysical Journal International,
13
14 651 151, 622–631.
15
16
17 652 Bergamo, P., D. Boiero, and L. V. Socco, 2012, Retrieving 2D structures from surface-wave
18
19 653 data by means of space-varying spatial windowing: Geophysics, 77, EN39–EN51.
20
21 654 Bergamo, P., B. Dashwood, S. Uhlemann, R. Swift, J. Chambers, D. Gunn, and S. Dono-
22
23
24 655 hue, 2016a, Time-lapse monitoring of climate effects on earthworks using surface waves:
25
26 656 Geophysics, 81, EN1–EN15.
27
28 657 Bergamo, P., B. Dashwood, S. Uhlemann, R. Swift, J. E. Chambers, D. A. Gunn, and S.
29
30
31 658 Donohue, 2016b, Time-lapse monitoring of fluid-induced geophysical property variations
32
33 659 within an unstable earthwork using P-wave refraction: Geophysics, 81, EN17–EN27.
34
35 660 Bergamo, P., and L. Socco, 2016, P- and S-wave velocity models of shallow dry sand
36
37
38
661 formations from surface wave multimodal inversion: Geophysics, 81, R197–R209.
39
40 662 Bevington, P. R., and D. K. Robinson, 2002, Data reduction and error analysis for the
41
42 663 physical sciences: McGraw Hill Higher Education.
43
44
664 Boaga, J., G. Cassiani, C. Strobbia, and G. Vignoli, 2013, Mode misidentification in Rayleigh
45
46
47 665 waves: Ellipticity as a cause and a cure: Geophysics, 78, EN17–EN28.
48
49 666 Bodet, L., O. Abraham, and D. Clorennec, 2009, Near-offset effects on Rayleigh-wave
50
51
667 dispersion measurements: Physical modeling: Journal of Applied Geophysics, 68, 95–103.
52
53
54 668 Bodet, L., K. van Wijk, A. Bitri, O. Abraham, P. Côte, G. Grandjean, and D. Leparoux,
55
56 669 2005, Surface-wave inversion limitations from laser-Doppler physical modeling: Journal of
57
58
670 Environmental and Engineering Geophysics, 10, 151–162.
59
60
32
1
2
3
4
671 Bohlen, T., S. Kugler, G. Klein, and F. Theilen, 2004, 1.5D inversion of lateral variation of
5
6
7 672 Scholte-wave dispersion: Geophysics, 69, 330–344.
8
9 673 Boiero, D., and L. V. Socco, 2010, Retrieving lateral variations from surface wave dispersion
10
11 674 curves: Geophysical Prospecting, 58, 977–996.
12
13
14 675 ——–, 2011, The meaning of surface wave dispersion curves in weakly laterally varying
15
16 676 structures: Near Surface Geophysics, 9, 561–570.
17
18 677 Boiero, D., L. V. Socco, S. Stocco, and R. Wisén, 2013, Bedrock mapping in shallow
19
20
21 678 environments using surface-wave analysis: The Leading Edge, 32, 664–672.
22
23 679 Cardarelli, E., M. Cercato, and G. De Donno, 2014, Characterization of an earth-filled dam
24
25 680 through the combined use of electrical resistivity tomography, P- and SH-wave seismic
26
27
28 681 tomography and surface wave data: Journal of Applied Geophysics, 106, 87–95.
29
30 682 Christiansen, R. L., and H. R. Blank, 1975, Geologic map of the Canyon Village quadrangle,
31
32 683 Yellowstone National Park, Wyoming: USGS Numbered Series 1192.
33
34
35 684 Dangeard, M., S. Pasquet, L. Bodet, R. Guérin, L. Longuevergne, and J. Thiesson, 2016,
36
37 685 Temporal Variations of Near-surface Seismic Data at the Ploemeur (France) Hydroge-
38
39 686 ological Observatory: Presented at the Near Surface Geoscience 2016 – 22nd European
40
41
42 687 Meeting of Environmental and Engineering Geophysics, EAGE.
43
44 688 Der, Z. A., and M. Landisman, 1972, Theory for errors, resolution, and separation of
45
46 689 unknown variables in inverse problems, with application to the mantle and the crust in
47
48
49
690 southern Africa and Scandinavia: Geophysical Journal of the Royal Astronomical Society,
50
51 691 27, 137–178.
52
53 692 Dhemaied, A., F. Rejiba, C. Camerlynck, L. Bodet, and R. Guérin, 2011, Seismic-wave
54
55
693 propagation modeling in viscoelastic media using the auxiliary differential equation method:
56
57
58 694 Bulletin of the Seismological Society of America, 101, 413–420.
59
60
33
1
2
3
4
695 Dunkin, J. W., 1965, Computation of modal solutions in layered, elastic media at high
5
6
7 696 frequencies: Bulletin of the Seismological Society of America, 55, 335–358.
8
9 697 Endrun, B., T. Meier, S. Lebedev, M. Bohnhoff, G. Stavrakakis, and H.-P. Harjes, 2008, S
10
11 698 velocity structure and radial anisotropy in the Aegean region from surface wave dispersion:
12
13
14 699 Geophysical Journal International, 174, 593–616.
15
16 700 Ezersky, M. G., L. Bodet, E. Akawwi, A. S. Al-Zoubi, C. Camerlynck, A. Dhemaied, and P.-Y.
17
18 701 Galibert, 2013, Seismic surface-wave prospecting methods for sinkhole hazard assessment
19
20
21 702 along the Dead Sea shoreline: Journal of Environmental and Engineering Geophysics, 18,
22
23 703 233–253.
24
25 704 Forbriger, T., 2003a, Inversion of shallow-seismic wavefields: I. Wavefield transformation:
26
27
28 705 Geophysical Journal International, 153, 719–734.
29
30 706 ——–, 2003b, Inversion of shallow-seismic wavefields: II. Inferring subsurface properties from
31
32 707 wavefield transforms: Geophysical Journal International, 153, 735–752.
33
34
35 708 Gabriels, P., R. Snieder, and G. Nolet, 1987, In situ measurements of shear-wave velocity in
36
37 709 sediments with higher-mode Rayleigh waves: Geophysical Prospecting, 35, 187–196.
38
39 710 Godio, A., C. Strobbia, and G. De Bacco, 2006, Geophysical characterisation of a rockslide
40
41
42 711 in an alpine region: Engineering Geology, 83, 273–286.
43
44 712 Grandjean, G., and A. Bitri, 2006, 2M-SASW : Multifold multichannel seismic inversion of
45
46 713 local dispersion of Rayleigh waves in laterally heterogeneous subsurfaces: application to
47
48
49
714 the Super-Sauze earthflow, France: Near Surface Geophysics, 4, 367–375.
50
51 715 Grelle, G., and F. M. Guadagno, 2009, Seismic refraction methodology for groundwater level
52
53 716 determination: “Water seismic index”: Journal of Applied Geophysics, 68, 301–320.
54
55
717 Haines, S. S., 2007, A hammer-impact, aluminium, shear-wave seismic source: Technical
56
57
58 718 Report OF 07-1406, United States Geological Survey.
59
60
34
1
2
3
4
719 Haney, M. M., and H. Douma, 2012, Rayleigh-wave tomography at Coronation Field, Canada:
5
6
7 720 The topography effect: The Leading Edge, 31, 54–61.
8
9 721 Haskell, N. A., 1953, The dispersion of surface waves on multilayered media: Bulletin of the
10
11 722 Seismological Society of America, 43, 17–34.
12
13
14 723 Hayashi, K., and H. Suzuki, 2004, CMP cross-correlation analysis of multi-channel surface-
15
16 724 wave data: Exploration Geophysics, 35, 7–13.
17
18 725 Herrmann, R. B., 2013, Computer Programs in Seismology: An evolving tool for instruction
19
20
21 726 and research: Seismological Research Letters, 84, 1081–1088.
22
23 727 Hibert, C., G. Grandjean, A. Bitri, J. Travelletti, and J.-P. Malet, 2012, Characterizing
24
25 728 landslides through geophysical data fusion: Example of the La Valette landslide (France):
26
27
28 729 Engineering Geology, 128, 23–29.
29
30 730 Hurwitz, S., R. N. Harris, C. A. Werner, and F. Murphy, 2012, Heat flow in vapor dominated
31
32 731 areas of the Yellowstone Plateau Volcanic Field: Implications for the thermal budget of
33
34
35 732 the Yellowstone Caldera: Journal of Geophysical Research: Solid Earth, 117, B10207.
36
37 733 Ikeda, T., T. Tsuji, and T. Matsuoka, 2013, Window-controlled CMP crosscorrelation analysis
38
39 734 for surface waves in laterally heterogeneous media: Geophysics, 78, EN95–EN105.
40
41
42 735 Jongmans, D., 1992, The application of seismic methods for dynamic characterization of
43
44 736 soils in earthquake engineering: Bulletin of the International Association of Engineering
45
46 737 Geology, 46, 63–69.
47
48
49
738 Jongmans, D., G. Bièvre, F. Renalier, S. Schwartz, N. Beaurez, and Y. Orengo, 2009,
50
51 739 Geophysical investigation of a large landslide in glaciolacustrine clays in the Trièves area
52
53 740 (French Alps): Engineering Geology, 109, 45–56.
54
55
741 Jongmans, D., and D. Demanet, 1993, The importance of surface waves in vibration study and
56
57
58 742 the use of Rayleigh waves for estimating the dynamic characteristics of soils: Engineering
59
60
35
1
2
3
4
743 Geology, 34, 105–113.
5
6
7 744 Kennett, B. L. N., 1974, Reflections, rays, and reverberations: Bulletin of the Seismological
8
9 745 Society of America, 64, 1685–1696.

10
11 746 Konstantaki, L. A., S. F. A. Carpentier, F. Garofalo, P. Bergamo, and L. V. Socco, 2013,
12
13
14 747 Determining hydrological and soil mechanical parameters from multichannel surface-wave
15
16 748 analysis across the Alpine Fault at Inchbonnie, New Zealand: Near Surface Geophysics,
17
18 749 11, 435–448.
19
20
21 750 Lai, C. G., and G. J. Rix, 1998, Simultaneous inversion of Rayleigh phase velocity and
22
23 751 attenuation for near-surface site characterization: Technical Report GIT-CEE/GEO-98-2,
24
25 752 Georgia Institute of Technology.
26
27
28 753 Lai, C. G., G. J. Rix, S. Foti, and V. Roma, 2002, Simultaneous measurement and inversion of
29
30 754 surface wave dispersion and attenuation curves: Soil Dynamics and Earthquake Engineering,
31
32 755 22, 923–930.
33
34
35 756 Lomax, A., and R. Snieder, 1994, Finding sets of acceptable solutions with a genetic algorithm
36
37 757 with application to surface wave group dispersion in Europe: Geophysical Research Letters,
38
39 758 21, 2617–2620.
40
41
42 759 Martin, R., and D. Komatitsch, 2009, An unsplit convolutional perfectly matched layer
43
44 760 technique improved at grazing incidence for the viscoelastic wave equation: Geophysical
45
46 761 Journal International, 179, 333–344.
47
48
49
762 Martı́nez, M. D., X. Lana, J. Olarte, J. Badal, and J. A. Canas, 2000, Inversion of Rayleigh
50
51 763 wave phase and group velocities by simulated annealing: Physics of the Earth and Planetary
52
53 764 Interiors, 122, 3–17.
54
55
765 McMechan, G. A., and M. J. Yedlin, 1981, Analysis of dispersive waves by wave field
56
57
58 766 transformation: Geophysics, 46, 869–874.
59
60
36
1
2
3
4
767 Miller, R. D., J. Xia, C. B. Park, and J. Ivanov, 1999, Multichannel analysis of surface waves
5
6
7 768 to map bedrock: The Leading Edge, 18, 1392–1396.
8
9 769 Mokhtar, T. A., R. B. Herrmann, and D. R. Russell, 1988, Seismic velocity and Q model for
10
11 770 the shallow structure of the Arabian Shield from short-period Rayleigh waves: Geophysics,
12
13
14 771 53, 1379–1387.
15
16 772 Mosegaard, K., and A. Tarantola, 1995, Monte Carlo sampling of solutions to inverse
17
18 773 problems: Journal of Geophysical Research: Solid Earth, 100, 12431–12447.
19
20
21 774 Mota, R., and F. A. Monteiro Santos, 2010, 2D sections of porosity and water saturation
22
23 775 from integrated resistivity and seismic surveys: Near Surface Geophysics, 8, 575–584.
24
25 776 Nagai, K., A. O’Neill, Y. Sanada, and Y. Ashida, 2005, Genetic algorithm inversion of
26
27
28 777 Rayleigh wave dispersion from CMPCC gathers over a shallow fault model: Journal of
29
30 778 Environmental and Engineering Geophysics, 10, 275–286.
31
32 779 Neducza, B., 2007, Stacking of surface waves: Geophysics, 72, 51–58.
33
34
35 780 Olona, J., J. Pulgar, G. Fernández-Viejo, C. López-Fernández, and J. González-Cortina, 2010,
36
37 781 Weathering variations in a granitic massif and related geotechnical properties through
38
39 782 seismic and electrical resistivity methods: Near Surface Geophysics, 8, 585–599.
40
41
42 783 O’Neill, A., 2003, Full-waveform reflectivity for modelling, inversion and appraisal of seismic
43
44 784 surface wave dispersion in shallow site investigations: PhD thesis, University of Western
45
46 785 Australia, Perth, Australia.
47
48
49
786 O’Neill, A., M. Dentith, and R. List, 2003, Full-waveform P-SV reflectivity inversion of
50
51 787 surface waves for shallow engineering applications: Exploration Geophysics, 34, 158–173.
52
53 788 O’Neill, A., and T. Matsuoka, 2005, Dominant higher surface-wave modes and possible
54
55
789 inversion pitfalls: Journal of Environmental and Engineering Geophysics, 10, 185–201.
56
57
58 790 Othman, A. A. A., 2005, Construed geotechnical characteristics of foundation beds by seismic
59
60
37
1
2
3
4
791 measurements: Journal of Geophysics and Engineering, 2, 126–138.
5
6
7 792 Park, C. B., R. D. Miller, and J. Xia, 1999, Multichannel analysis of surface waves: Geophysics,
8
9 793 64, 800–808.

10
11 794 Parsekian, A. D., K. Singha, B. J. Minsley, W. S. Holbrook, and L. Slater, 2015, Multiscale
12
13
14 795 geophysical imaging of the critical zone: Reviews of Geophysics, 53, 2014RG000465.
15
16 796 Pasquet, S., L. Bodet, P. Bergamo, R. Guérin, R. Martin, R. Mourgues, and V. Tournat,
17
18 797 2016a, Small-scale seismic monitoring of varying water levels in granular media: Vadose
19
20
21 798 Zone Journal, 15.
22
23 799 Pasquet, S., L. Bodet, A. Dhemaied, A. Mouhri, Q. Vitale, F. Rejiba, N. Flipo, and R.
24
25 800 Guérin, 2015a, Detecting different water table levels in a shallow aquifer with combined
26
27
28 801 P-, surface and SH-wave surveys: Insights from VP/VS or Poisson’s ratios: Journal of
29
30 802 Applied Geophysics, 113, 38–50.
31
32 803 Pasquet, S., L. Bodet, L. Longuevergne, A. Dhemaied, C. Camerlynck, F. Rejiba, and
33
34
35 804 R. Guérin, 2015b, 2D characterization of near-surface VP/VS: surface-wave dispersion
36
37 805 inversion versus refraction tomography: Near Surface Geophysics, 13, 315–331.
38
39 806 Pasquet, S., L. Bodet, L. Longuevergne, A. Dhemaied, F. Rejiba, C. Camerlynck, and
40
41
42 807 R. Guérin, 2012, Surface-wave dispersion stacking on a granite-micaschists contact at
43
44 808 Plœmeur hydrological observatory, France: Presented at the Near Surface Geoscience 2012
45
46 809 – 18th European Meeting of Environmental and Engineering Geophysics, EAGE.
47
48
49
810 Pasquet, S., W. S. Holbrook, B. J. Carr, and K. W. W. Sims, 2016b, Geophysical imaging
50
51 811 of shallow degassing in a Yellowstone hydrothermal system: Geophysical Research Letters,
52
53 812 43, 2016GL071306.
54
55
813 Pasquet, S., G. Sauvin, M. R. Andriamboavonjy, L. Bodet, I. Lecomte, and R. Guérin, 2014,
56
57
58 814 Surface-wave dispersion inversion versus SH-wave refraction tomography in saturated
59
60
38
1
2
3
4
815 and poorly dispersive quick clays: Presented at the Near Surface Geoscience 2014 – 20th
5
6
7 816 European Meeting of Environmental and Engineering Geophysics, EAGE.
8
9 817 Pérez Solano, C. A. P., D. Donno, and H. Chauris, 2014, Alternative waveform inversion for
10
11 818 surface wave analysis in 2-D media: Geophysical Journal International, 198, 1359–1372.
12
13
14 819 Raptakis, D., F. J. Chávez-Garcı́a, K. Makra, and K. Pitilakis, 2000, Site effects at Euroseis-
15
16 820 test—I. Determination of the valley structure and confrontation of observations with 1d
17
18 821 analysis: Soil Dynamics and Earthquake Engineering, 19, 1–22.
19
20
21 822 Renalier, F., D. Jongmans, A. Savvaidis, M. Wathelet, B. Endrun, and C. Cornou, 2010,
22
23 823 Influence of parameterization on inversion of surface wave dispersion curves and definition
24
25 824 of an inversion strategy for sites with a strong contrast: Geophysics, 75, B197–B209.
26
27
28 825 Rix, G. J., C. G. Lai, and S. Foti, 2001, Simultaneous measurement of surface wave dispersion
29
30 826 and attenuation curves: Geotechnical testing journal, 24, 350–358.
31
32 827 Rix, G. J., and E. A. Leipski, 1991, Accuracy and resolution of surface wave inversion:
33
34
35 828 Recent advances in instrumentation, data acquisition and testing in soil dynamics, ASCE,
36
37 829 17–32.
38
39 830 Russel, D. R., 1987, Multi-channel processing of dispersed surface waves: PhD thesis, Saint
40
41
42 831 Louis University, Saint Louis, Missouri, USA.
43
44 832 Sambridge, M., 1999a, Geophysical inversion with a neighbourhood algorithm—I. Searching
45
46 833 a parameter space: Geophysical Journal International, 138, 479–494.
47
48
49
834 ——–, 1999b, Geophysical inversion with a neighbourhood algorithm—II. Appraising the
50
51 835 ensemble: Geophysical Journal International, 138, 727–746.
52
53 836 ——–, 2001, Finding acceptable models in nonlinear inverse problems using a neighbourhood
54
55
837 algorithm: Inverse Problems, 17, 387–403.
56
57
58 838 Sambridge, M., and K. Mosegaard, 2002, Monte Carlo methods in geophysical inverse
59
60
39
1
2
3
4
839 problems: Reviews of Geophysics, 40, 1009.
5
6
7 840 Sambuelli, L., G. P. Deidda, G. Albis, E. Giorcelli, and G. Tristano, 2001, Comparison of
8
9 841 standard horizontal geophones and newly designed horizontal detectors: Geophysics, 66,
10
11 842 1827–1837.
12
13
14 843 Schmelzbach, C., D. Sollberger, S. A. Greenhalgh, H. Horstmeyer, H. Maurer, and J. O. A.
15
16 844 Robertsson, 2016, 9C seismic data acquisition for near-surface applications: recording,
17
18 845 waveform reciprocity and 4C rotation: Presented at the 78th EAGE Conference and
19
20
21 846 Exhibition 2016 - Workshops, EAGE.
22
23 847 Sheriff, R. E., and L. P. Geldart, 1995, Exploration seismology: Cambridge University Press.
24
25 848 Socco, L. V., and D. Boiero, 2008, Improved Monte Carlo inversion of surface wave data:
26
27
28 849 Geophysical Prospecting, 56, 357–371.
29
30 850 Socco, L. V., D. Boiero, S. Foti, and R. Wisén, 2009, Laterally constrained inversion of
31
32 851 ground roll from seismic reflection records: Geophysics, 74, G35–G45.
33
34
35 852 Socco, L. V., S. Foti, and D. Boiero, 2010a, Surface-wave analysis for building near-surface
36
37 853 velocity models - Established approaches and new perspectives: Geophysics, 75, A83–A102.
38
39 854 Socco, L. V., D. Jongmans, D. Boiero, S. Stocco, M. Maraschini, K. Tokeshi, and D. Hantz,
40
41
42 855 2010b, Geophysical investigation of the Sandalp rock avalanche deposits: Journal of Applied
43
44 856 Geophysics, 70, 277–291.
45
46 857 Socco, L. V., and C. Strobbia, 2004, Surface-wave method for near-surface characterization:
47
48
49
858 a tutorial: Near Surface Geophysics, 2, 165–185.
50
51 859 Sollberger, D., C. Schmelzbach, C. Van Renterghem, J. Robertsson, and S. Greenhalgh,
52
53 860 2016, Single-component elastic wavefield separation at the free surface using source- and
54
55
861 receiver-side gradients: SEG 86th annual meeting, Society of Exploration Geophysicists,
56
57
58 862 2268–2273.
59
60
40
1
2
3
4
863 Song, X., H. Gu, L. Tang, S. Zhao, X. Zhang, L. Li, and J. Huang, 2015, Application
5
6
7 864 of artificial bee colony algorithm on surface wave data: Computers & Geosciences, 83,
8
9 865 219–230.
10
11 866 Song, X., L. Tang, X. Lv, H. Fang, and H. Gu, 2012, Shuffled complex evolution approach
12
13
14 867 for effective and efficient surface wave analysis: Computers & Geosciences, 42, 7–17.
15
16 868 St. Clair, J., 2015, Geophysical investigations of underplating at the Middle American Trench,
17
18 869 weathering in the critical zone, and snow water equivalent in seasonal snow: PhD thesis,
19
20
21 870 University of Wyoming, Laramie, WY, USA.
22
23 871 St. Clair, J., S. Moon, W. S. Holbrook, J. T. Perron, C. S. Riebe, S. J. Martel, B. Carr, C.
24
25 872 Harman, K. Singha, and D. d. Richter, 2015, Geophysical imaging reveals topographic
26
27
28 873 stress control of bedrock weathering: Science, 350, 534–538.
29
30 874 Strobbia, C., and S. Foti, 2006, Multi-offset phase analysis of surface wave data (MOPA):
31
32 875 Journal of Applied Geophysics, 59, 300–313.
33
34
35 876 Strobbia, C., A. Laake, P. Vermeer, and A. Glushchenko, 2011, Surface waves: use them
36
37 877 then lose them. Surface-wave analysis, inversion and attenuation in land reflection seismic
38
39 878 surveying: Near Surface Geophysics, 9, 503–514.
40
41
42 879 Thomson, W. T., 1950, Transmission of elastic waves through a stratified solid medium:
43
44 880 Journal of Applied Physics, 21, 89–93.
45
46 881 Turesson, A., 2007, A comparison of methods for the analysis of compressional, shear, and
47
48
49
882 surface wave seismic data, and determination of the shear modulus: Journal of Applied
50
51 883 Geophysics, 61, 83–91.
52
53 884 Uhlemann, S., S. Hagedorn, B. Dashwood, H. Maurer, D. Gunn, T. Dijkstra, and J. Chambers,
54
55
885 2016, Landslide characterization using P- and S-wave seismic refraction tomography —
56
57
58 886 The importance of elastic moduli: Journal of Applied Geophysics, 134, 64–76.
59
60
41
1
2
3
4
887 Uyanık, O., 2011, The porosity of saturated shallow sediments from seismic compressional
5
6
7 888 and shear wave velocities: Journal of Applied Geophysics, 73, 16–24.
8
9 889 Wathelet, M., 2008, An improved neighborhood algorithm: Parameter conditions and dynamic
10
11 890 scaling: Geophysical Research Letters, 35, L09301.
12
13
14 891 Wathelet, M., D. Jongmans, and M. Ohrnberger, 2004, Surface-wave inversion using a direct
15
16 892 search algorithm and its application to ambient vibration measurements: Near Surface
17
18 893 Geophysics, 2, 211–221.
19
20
21 894 Werner, C., S. L. Brantley, and K. Boomer, 2000, CO2 emissions related to the Yellowstone
22
23 895 volcanic system: 2. Statistical sampling, total degassing, and transport mechanisms:
24
25 896 Journal of Geophysical Research: Solid Earth, 105, 10831–10846.
26
27
28 897 Xia, J., R. D. Miller, and C. B. Park, 1999, Estimation of near-surface shear-wave velocity
29
30 898 by inversion of Rayleigh waves: Geophysics, 64, 691–700.
31
32 899 Xia, J., R. D. Miller, C. B. Park, and G. Tian, 2003, Inversion of high frequency surface
33
34
35 900 waves with fundamental and higher modes: Journal of Applied Geophysics, 52, 45–57.
36
37 901 Xia, J., R. D. Miller, C. B. Park, E. Wightman, and R. Nigbor, 2002, A pitfall in shallow
38
39 902 shear-wave refraction surveying: Journal of Applied Geophysics, 51, 1–9.
40
41
42 903 Zhang, S. X., and L. S. Chan, 2003, Possible effects of misidentified mode number on Rayleigh
43
44 904 wave inversion: Journal of Applied Geophysics, 53, 17–29.
45
46 905 Zywicki, D., and G. Rix, 2005, Mitigation of near-field effects for seismic surface wave velocity
47
48
49
906 estimation with cylindrical beamformers: Journal of Geotechnical and Geoenvironmental
50
51 907 Engineering, 131, 970–977.
52
53
54
55
56
57
58
59
60
42
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19 1 Flowchart of the dispersion extraction procedure.
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
2 Data windowing and dispersion stacking workflow.
31
32 75x52mm (600 x 600 DPI)
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
3 Flowchart of the dispersion inversion procedure.
47
48 245x362mm (600 x 600 DPI)
49
50
51
52
53
54
55
56
57
58
59
60
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31 4 (a) Layout of the seismic acquisition setup, with 240 geophones (gray triangles) spaced every 1 m and 25
32 shots (gray stars) spaced every 10 m. (b) Example of shot gather for a source located at 120 m (red star in
33 a). (c) Final VP model obtained from P-wave traveltime tomography. The topography extracted from
34 airborne LiDAR data is represented with a black solid line.
35
36 134x97mm (600 x 600 DPI)
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30 5 Extraction of single dispersion images for a 31-trace window centered at Xmid = 30 m, using shots located
31 at 0 m (a), 10 m (b), 50 m (c) and 60 m (d). On each inset, windowed shot gathers are on the left,
32 corresponding spectrograms are at the bottom right, and computed dispersion images are at the top right.
Red dashed lines on spectrograms and dispersion images correspond to automatic low cut frequencies
33
defined from spectrogram amplitude.
34
35 143x98mm (300 x 300 DPI)
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35 6 (a) to (d), successive stacking of the single dispersion images represented in Figure 5a to 5d.
36 103x85mm (600 x 600 DPI)
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
7 Stacked dispersion images extracted at Xmid = 30 m (a) and Xmid = 110 m (b) with picked dispersion
24 curves (white error bars) of the fundamental (0) and first higher (1) modes. The uncertainty range is
25 defined according to the workflow described in O'Neill (2003). Dispersion curves are limited down to a
26 frequency defined with spectral amplitude threshold of 2.5 % (red dashed line), or up to a wavelength of 50
27 m (blue dashed line).
28
29 59x28mm (600 x 600 DPI)
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25 8 Pseudo-sections of SW phase velocity picked for the fundamental (a) and first higher (b) modes along the
26 line after dispersion stacking, represented as a function of the wavelength (λ) and the spread mid-point
27 position.
28
100x53mm (600 x 600 DPI)
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40 9 Results of 1D NA inversions of dispersion data (black error bars) at Xmid = 30 m (fundamental (a) and
41 first higher (b) modes) and Xmid = 110 m (fundamental mode (d)). Resulting VS models are represented for
42 Xmid = 30 m (c) and Xmid = 110 m (e), along with a misfit-weighted velocity structure (black dashed lines)
43 built from the average parameters of all accepted models. Calculated dispersion and corresponding models
44 are represented with misfit-based color and gray scales for accepted and rejected models, respectively.
45
209x209mm (300 x 300 DPI)
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23 10 Stacked dispersion images extracted at Xmid = 30 m (a) and Xmid = 110 m (b) with picked (white error
24 bars) and calculated (red solid lines) dispersion curves represented for the fundamental (0), the first (1),
25 second (2), and third (3) higher modes.
26
27 59x27mm (600 x 600 DPI)
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39 11 (a) Misfit value calculated with Equation 11 for each 1D NA inversion along the line. (b) Pseudo-section of
calculated phase velocity for the fundamental mode. (c) Pseudo-section of phase velocity residuals for the
40 fundamental mode. (d) Histogram of residuals for the fundamental mode. (e) Pseudo-section of calculated
41 phase velocity for the first higher mode. (f) Pseudo-section of phase velocity residuals for the first higher
42 mode. (g) Histogram of residuals for the first higher mode.
43
44 243x232mm (300 x 300 DPI)
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25 12 (a) Pseudo-2D section of VS standard deviation computed from accepted models at each Xmid position
26 along the line. (b) Pseudo-2D section of average VS computed from accepted models at each Xmid position
27 along the line. The black dashed line corresponds to the depth of investigation estimated with a VS standard
28 deviation threshold of 150 m/s. The topography extracted from airborne LiDAR data is represented with a
black solid line.
29
30 110x58mm (300 x 300 DPI)
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16 13 Poisson’s ratio computed from P-wave tomography VP and SW dispersion inversion VS and masked below
17 the depth of investigation estimated from VS standard deviation. The topography extracted from airborne
18 LiDAR data is represented with a black solid line.
19
20 54x13mm (300 x 300 DPI)
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60

SWIP: An Integrated Workflow For Surface-Wave Dispersion Inversion and Profiling

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

SWIP: An Integrated Workflow For Surface-Wave Dispersion Inversion and Profiling

Загружено:

Авторское право:

Доступные форматы

GEOPHYSICS

SWIP: an integrated workflow for surface-wave dispersion

Manuscript Type: Reproducible research: Geophysics papers of the future

Date Submitted by the Author: 19-May-2017

Keywords: surface wave, dispersion, inversion, profiling, shear wave (S-wave)

Area of Expertise: Geophysical Software and Algorithms, Seismic Inversion

9 23 suggested investigation depth. As an example, we used SWIP to process data collected at a

9 208 defined for each mode m as:

9 367 parameterization according to a priori information.

9 612 tomography-based parameterization, in order to retrieve 1D VS models for each extraction

9 745 Society of America, 64, 1685–1696.

9 793 64, 800–808.

Вам также может понравиться