Applied Acoustics 95 (2015) 27–36

Contents lists available at ScienceDirect

Applied Acoustics
journal homepage: www.elsevier.com/locate/apacoust

Estimation of impedance model input parameters from in situ

measurements: Principles and applications
Gwenaël Guillaume a,⇑, Olivier Faure a,b, Benoit Gauvreau a, Fabrice Junker b, Michel Bérengier a,
Philippe L’Hermite a
a
LUNAM Université, Institut français des sciences et technologies des transports, des aménagements et des réseaux (Ifsttar), Laboratoire d’acoustique environnementale (LAE),
Centre de Nantes, Route de Bouaye, CS4, 44344 Bouguenais Cedex, France
b
EDF R&D, Département Analyses Mécaniques et Acoustique (AMA), 1 avenue du Général de Gaulle, 92141 Clamart Cedex, France

a r t i c l e i n f o a b s t r a c t

Article history: The impact of ground physical parameters variations on the previsions of recent outdoor sound
Received 6 August 2014 propagation numerical models can be substantial. Among these parameters, the air flow resistivity and
Received in revised form 7 January 2015 the effective thickness of numerous soil types (in particular grass lawns) have to be characterized
Accepted 13 January 2015
in situ. Thus, a two-microphone technique and a semi-automatic procedure have been developed in order
Available online 14 March 2015
to estimate the ground impedance parameters. The estimation method is based on a narrow-band fitting
between the experimental sound level attenuation spectra and the analytical solutions for the
Keywords:
measurement configuration. This approach is applied to three types of ground, both in winter and
Ground effect
Space and time variabilities
summer conditions. Results are analyzed and discussed in terms of space and time variabilities as well
Impedance models as the impact of such spatial and seasonal variations on sound pressure levels.
Semi-automatic fitting procedure Ó 2015 Elsevier Ltd. All rights reserved.
In-situ measurements
Two-microphone technique

1. Introduction the technique used for outdoor sound propagation prediction,

sound attenuation is strongly dependent upon the ground absorp-
Environmental noise and more particularly transportation noise tion and more precisely, the surface impedance. Several physical
are considered important risk factors for people living nearby [2,3,31] and semi-empirical [10,26] models are available and
industrial sites, ground transportation networks or airport areas. widely used. However, their physical characteristic parameters
For many years, this problem has been considered by governmen- have to be accurately determined in order to minimize the uncer-
tal institutions and new regulations have been issued. Directive tainty on the resulting sound pressure levels [22,27].
2002/49/EC [11] requires producing noise maps for the largest The necessity of estimating physical parameters used as input
cities and the largest transportation networks using adapted engi- data of the impedance model for each encountered situation leads
neering models. The accuracy of these models can be addressed by to develop an adapted procedure based on in situ sound level dif-
comparisons with either experimental data or numerical predic- ference measurements. Different authors already implemented
tions from reference models. such methods [4,23,29,30]. The [1] standard also applies this tech-
During many years, the most commonly used models were nique but only in third-octave bands. Some of the best results were
based on analytical approaches [5]. In parallel, numerical modeling obtained after a manual best fit. In order to improve the technique,
has also been widely studied. After having developed techniques we propose in this paper a robust and semi-automatic procedure
in the frequency domain mainly based on Finite and Boundary which allows us to estimate the two-input data of a semi-empirical
Elements Methods [28] or Parabolic Equation Approximation impedance model: the airflow resistivity of the ground and its
[6,16,25], time-domain approaches have been more recently inves- effective thickness. This method is based on a narrow band fitting
tigated such as PSTD [19,20], FDTD [13] or TLM [17,18]. procedure performed on experimental sound level attenuation
Those last models are particularly sensitive to input data such spectra for a given measurement configuration.
as ground properties through the impedance models. Whatever After implementation of the method (Section 2), a discussion on
the respective influence of the geometrical measurement config-
uration and the impedance models is carried out (Section 3). The
⇑ Corresponding author. time evolution and the spatial dispersion of the estimated parame-
E-mail address: [email protected] (G. Guillaume). ters are also investigated along the paper.

http://dx.doi.org/10.1016/j.apacoust.2015.01.024
0003-682X/Ó 2015 Elsevier Ltd. All rights reserved.
28 G. Guillaume et al. / Applied Acoustics 95 (2015) 27–36

Fig. 1. Measurement configuration.

8000 propagation distance is larger than k1 (k standing for the wave-
6000 length), the sound pressure at the receiver R1 in Fig. 3, located at
4000 a distance r d from the source, is formulated by the Weyl–Van der
2000
0
Pol equation which can be expressed by using a time dependence
−2000 exp ðixt Þ as:
−4000
 
expðik0 rd Þ expðik0 rr Þ
0 0.005 0.01 0.015 0.02 0.025 pðR1 Þ ¼ P0 þQ ; ð1Þ
rd rr
where rr represents the distance crossed by the reflected wave, P0 is
8000 the source amplitude and k0 stands for the air wavenumber. Q is the
6000
spherical-wave reflection coefficient given by:
4000  
2000 Q ¼ Rp þ 1  Rp F ðwÞ; ð2Þ
0
−2000 with Rp the plane-wave reflection coefficient which is given for a
−4000 localized reaction by:
0 0.005 0.01 0.015 0.02 0.025
sin h  Z=Z 0
Rp ¼ ð3Þ
sin h þ Z=Z 0
Fig. 2. Example of impulse responses obtained at the microphones (top) R1 and
where Z and Z 0 are the ground and the air impedances respectively,
(bottom) R2 .
and h corresponds to the incidence angle of the sound wave with
respect to the normal to the ground. The Faddeeva function F ðwÞ
2. Ground impedance characterization in Eq. (2) is defined by:
pffiffiffiffi
2.1. Experimental setup
F ðwÞ ¼ 1 þ i pw expðw2 ÞerfcðiwÞ; ð4Þ
where erfc stands for the error function and w is the numerical
The experimental setup, depicted in Fig. 1, consists of an distance given by [14]:
omnidirectional sound source and of two microphones R1 and R2 .  2 !
2
The source emits burst-chirp series (average on 50 realizations) 2k0 r r Z0 k0 2
w ¼ i 2 1 cos h ; ð5Þ
at zS ¼ 0:6 m high and at a distance d ¼ 4 m from the receivers. 1  Rp cos2 h Z k
The microphone R1 is located at a height zR1 ¼ zS above the ground
with k the ground wavenumber.
whereas the microphone R2 is placed on the surface (i.e. at
Two impedance models are considered in this study: the
zR2 ¼ 0 m). This measurement configuration has been chosen in
Delany–Bazley and the Miki models. The semi-empirical model
order to maximize the ground dip and thus to improve the post- proposed by Delany and Bazley [10] requires the knowledge of
treatment for the fitting procedure [4,7]. one ground parameter only: the air flow resistivity r. The normal-
The cross-correlations between the signals received at the ized impedance and wavenumber are respectively expressed as:
microphones R1 and R2 and the source signal lead to the impulse  0:754  0:732
responses hðR1 Þ and hðR2 Þ respectively as depicted in Fig. 2. f f
Z=Z 0 ¼ 1 þ 9:08 þ i11:9 ; ð6aÞ
These impulse responses are then windowed by a Blackman– r r
Harris type window in order to avoid any spurious reflection from  0:70  0:595
f f
potential obstacles. The spectra of these windowed impulse k=k0 ¼ 1 þ 10:8 þ i10:3 : ð6bÞ
r r
responses, HðR1 Þ and HðR2 Þ, are computed afterward by applying
a Fourier transform on the frequency range [100–6400] Hz. Miki [26] has proposed to modify the Delany–Bazley model
which can lead to a negative impedance real part in low frequency.
Finally, the energy ratio DLexp ¼ jHðR2 Þj2 =jHðR1 Þj2 is calculated.
Besides, contrary to Delany–Bazley, the Miki model can be
implemented in time-domain models [12]. It can be expressed as:
2.2. Fitting procedure
 0:632  0:632
f f
Considering the sound propagation over an absorbing ground Z=Z 0 ¼ 1 þ 5:50 þ i8:43 ; ð7aÞ
r r
(Fig. 3), the sound wave received at a microphone combines the  0:618  0:618
f f
direct wave emitted by the source, the sound wave reflected by k=k0 ¼ 1 þ 7:81 þ i11:41 : ð7bÞ
the ground and the surface wave radiated by the ground. If the r r
 G. Guillaume et al. / Applied Acoustics 95 (2015) 27–36 29

Fig. 3. Sound propagation over an absorbing ground for the measurement configuration.

In the case of a natural ground (e.g. a grassy ground), a correc- procedure is performed by using the fmincon function of the MATLABÒ
tion can be applied on the impedance in order to take into account Optimization Toolbox [8].
the effect of the effective thickness e. This correction consists in The error on the estimation of the values of the air flow
substituting an impedance Z e to the impedance (given by Eq. (6a) resistivity r and of the effective thickness e is calculated as:
or (7a) for instance) such as:
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uP
u f DL2exp =th ðf Þ
Z e ¼ Z coth ðikeÞ: ð8Þ errðr; eÞ ¼ t P 2 : ð10Þ
f DLexp ðf Þ
Consequently, the input data needed by both impedance
models require the knowledge of a couple of parameters ðr; eÞ.
The proposed semi-automatic fitting procedure is based on five
The fitting procedure presented hereafter precisely consists in
main steps:
estimating this pair of parameters from the impulse responses
measurement described in Section 2.1.
1. Loading of the experimental spectra DLexp ðf Þ;
Previously, the fitting process was manually executed by an h i
operator. Leroy [24] has studied the behavior of a few expert 2. Manual definition of the fitting frequency ranges f min1 ; f max1
h i
operators during this fitting procedure with the same set of
and f min2 ; f max2 ;
experimental spectra. A similar approach has been overall
observed for all operators, that is: 3. Fitting process loop (n ¼ 1 to N, with N the total number of
iterations):
 
 close attention is paid to the low frequency part of the (a) Random selection of the couple of values rn0 ; en0 as start-
spectrum, in particular around the first destructive interference ing points of the minimization algorithm;
pattern (i.e. the ground dip) as the two-parameter impedance (b) Constrained minimization procedure of Eqs. (9a)–(9d));
models can not properly describe the ground effect at high (c) Calculation of the error (Eq. (10)) in the defined frequency
frequency; ranges;
 the first ground dip position according to frequency (obtained (d) Storage of the pair of values ðrn ; en Þ and of the associated
experimentally and numerically) must match; error err n ðr; eÞ;
 
 the discrepancies between the experimental and the numerical 4. Selection of ropt ; eopt among the N pair of values which lead to
spectra are minimized on either side of the ground dip. the minimal error, that is:
 
This approach is recreated in the proposed fitting algorithm err ropt ; eopt ¼ min ðerrðrn ; en ÞÞ; for n½1; N; ð11Þ
presented hereafter. Thus, a constrained minimization algorithm
is performed in the least squares sense by minimizing the 5. Display the fitted theoretical spectrum over the experimen-
error between the experimental attenuation spectrum DLexp tal spectrum with a graphical user interface tool allowing to
(Section 2.1) and the theoretical one DLth in the frequency ranges manually modify the values ropt and eopt .
h i h i
f min1 ; f max1 and f min2 ; f max2 around the ground dip, while con-
Fig. 4 illustrates the previous procedure.
straining the values of the air flow resistivity r and of the effective
ground thickness e, that is:
8 2.3. Experimental sites
0 1
>
>
f max
X1 f max
X2
>
> @
>
> min 2
DLexp =th ðf Þ þ DLexp =th ðf ÞA ! 0;
2
ðaÞ Two experimental campaigns have been realized in order to
>
>
>
< f ¼f min1 f ¼f min2 characterize the influence of the ground type and of the seasonal
0 < r < r max ; ðbÞ ð9Þ evolution on the impedance properties of grassy grounds. Three
>
> types of grounds are considered:
>
> 0 < e < emax ; ðcÞ
>
>
>
> h i
> DL
: 8f 2 f
ðf Þ P 0;
exp =th max1 ;f
min2 ; ðdÞ 1. a synthetic lawn1 (Fig. 5(a)–(d)): perfectly flat and homogeneous
ground;
where DLexp =th ðf Þ ¼ DLexp ðf Þ  DLth ðf ; r; eÞ; rmax is defined according 2. a grass lawn (Fig. 5(b)–(e)): almost flat and homogeneous
ground;
to the surface under investigation (here, rmax ¼ 990 kN s m4 ) and
3. a natural ground (Fig. 5(c)–(f)): rough and inhomogeneous
emax ¼ 0:1 m, the thickness having the same effect as an infinite
ground.
thickness ground from this value (the hyperbolic function
coth ¼ 1 in Eq. (8)). Constraint (9d) ensures that the theoretical
spectrum fits below the minimum of the first ground dip of the
experimental spectrum when this latter is noisy. The minimization 1
The synthetic ground consists in artificial grass set down on a porous drain layer.
30 G. Guillaume et al. / Applied Acoustics 95 (2015) 27–36

Fig. 4. Semi-automatic fitting procedure.

Fig. 5. Impedance measurements over (a)–(d) the synthetic lawn, (b)–(e) the grass lawn and (c)–(f) the natural ground.

These three grounds were chosen because of the presumption experimental setup described in Section 2.1 along 50 m lines as
of the increasing spatial inhomogeneity of their respective prop- shown in Fig. 6 (one measurement line for the synthetic lawn,
erties. The first campaign was held at the end of May 2012 when two parallel measurement lines for the grass lawn and the natu-
the grounds were dry (summer campaign in the following). The ral ground). For each case, the ground is almost continuously
second campaign was carried out in mid-December 2012 described along the measurement line as most of the successive
when the grounds were moist due to winter rainfall events measurements overlap. The experimental data were processed
(winter campaign thereafter). During each campaign and for each by using the semi-automatic fitting procedure described in
ground, 15 impedance measurements were performed with the Section 2.2.
 G. Guillaume et al. / Applied Acoustics 95 (2015) 27–36 31

Fig. 6. Experimental protocol.

Table 1 3.2. Results

Statistical values of the estimated parameters r (kN s m4) and e (m) with both the
Delany–Bazley and Miki models for the synthetic and grass lawns in summer. Tables A.2–A.4and (Appendix A.2) report the estimated values
Ground Synthetic Grass
of the parameters r and e along the measurement lines for each
ground during both campaigns (i.e. in summer and in winter).
Model Delany–Bazley Miki Delany–Bazley Miki
The statistical values of these results, i.e. the means, the standard
r Mean 275 369 262 354 deviations (SD) and the relative standard deviations (RSD), are
SD 36 58 57 82
summarized in Table 2 for each ground and both experimental
RSD 13% 16% 22% 23%
campaigns.
e Mean 0.0130 0.0118 0.0165 0.0157
The real and imaginary parts of the corresponding impedances
SD 0.0007 0.0007 0.0054 0.0047
RSD 5% 6% 33% 30% are plotted for each ground in Fig. 8(a), (c) and (e) for the first cam-
paign (summer) and in Fig. 8(b), (d) and (f) for the second cam-
paign (winter).
3. Application of the method
3.3. Discussion
3.1. Discussion on the empirical impedance models 3.3.1. Influence on the mean values of the impedance
For the first campaign (i.e. in summer), Table 2 shows that the
The fitting procedure described at Section 2.2 is first applied on synthetic and the grass lawns have almost the same mean charac-
the summer experimental data for both the synthetic and grass teristic impedance and so, the same absorption properties. The
lawns. The estimation is carried out by using successively the natural ground seems to be more absorbent as the mean value of
Delany–Bazley and the Miki models. Table A.1 (Appendix A.1) gives r is smaller and the mean value of e is higher.
the estimated values of the parameters r and e with the Delany– Concerning the second campaign (i.e. in winter), the mean value
Bazley model along the measurement line for both the synthetic of r is higher and the mean value of e is smaller than during the first
and grass lawns in summer. Similar data are presented in Tables campaign for the three kinds of grounds. Between summer and win-
A.2 and A.3 (Appendix A.2) with the Miki model. The correspond- ter, the air flow resistivity increases and the effective thickness of the
ing statistical values, i.e. the means, the standard deviations (SD) absorbing layer reduces. Therefore, the grounds are more reflective
and the relative standard deviations (RSD), are gathered in in winter. These expected tendencies are the consequence of the sur-
Table 1 for both grounds. Such results highlight the impedance face humidity [9,21]. This decrease of the grounds absorption clearly
model-dependency of both the air flow resistivity and thickness appears in Fig. 8(b), (d) and (f) when considering the amplitude of
which correspond with effective (and not physical) ground the imaginary parts of the impedances which is higher than in
parameters. Fig. 7 illustrates the discrepancy observed between Fig. 8(a), (c) and (e) respectively. In winter, the grass lawn has a
impedance spectra for the two empirical impedance models. The higher mean r (732 kN s m4) than the synthetic lawn
unsuitability of the Delany–Bazley model at low frequency (i.e. (484 kN s m4) whereas they are almost similar in summer (about
below 100 Hz) is observed for both grounds as the real part of 360 kN s m4). This shows that the grass lawn is more sensitive to
the impedance is negative. Thereafter, only the Miki model is used surface wetness. Besides, the decrease of e and the increase of r is
for the fitting process. lower for this natural ground than for the synthetic and grass lawns.

x 104 x 104
1 1

0 0

−1 −1

−2 −2

−3 −3

−4 −4

−5 −5
102 103 102 103

(a) Synthetic lawn (b) Grass lawn

Fig. 7. Example of fitted impedance curves obtained with both ( ) the Delany–Bazley and ( ) the Miki models for (a) the synthetic lawn and (b) the grass lawn.
32 G. Guillaume et al. / Applied Acoustics 95 (2015) 27–36

Table 2 3.3.2. Influence on the space variability of the impedance

Statistical values of the estimated parameters r (kN s m4) and e (m) with the Miki The relative standard deviations (RSDs) of the results given in
model for the three kinds of grounds and the two experimental campaigns.
Table 2 describe both the spatial dispersion and variability of the
Season Summer Winter grounds properties. For the first campaign (i.e. in summer), the
Ground Synthetic Grass Natural Synthetic Grass Natural RSDs of both r and e increase for the three kinds of grounds.
These results confirm the hypothesis of an increasing spatial
r Mean 369 354 212 484 631 256
SD 58 82 63 74 108 81 inhomogeneity. This variability of the values of r and e is observed
RSD 16% 23% 30% 15% 17% 32% on the imaginary parts of the impedance curves in low frequency.
e Mean 0.0118 0.0157 0.0154 0.0063 0.0060 0.0150 At 100 Hz, the spatial variability is weak for the synthetic lawn
SD 0.0007 0.0047 0.0021 0.0003 0.0012 0.0036 (Fig. 8(a)). For the grass lawn (Fig. 8(c)), the variability is higher
RSD 6% 30% 14% 5% 20% 24% and even more so for the natural ground (Fig. 8(e)).

Summer Winter
10 000 10 000
5 000 5 000
0 0
−5 000 −5 000
Synthetic lawn

−10 000 −10 000

−15 000 −15 000
−20 000 −20 000
−25 000 −25 000
−30 000 −30 000
−35 000 −35 000
−40 000 −40 000
−45 000 −45 000
−50 000 −50 000
102 103 102 103
Frequency (Hz) Frequency (Hz)
(a) (b)

10 000 10 000
5 000 5 000
0 0
−5 000 −5 000
−10 000 −10 000
Grass lawn

−15 000 −15 000

−20 000 −20 000
−25 000 −25 000
−30 000 −30 000
−35 000 −35 000
−40 000 −40 000
−45 000 −45 000
−50 000 −50 000
102 103 102 103
Frequency (Hz) Frequency (Hz)
(c) (d)
10 000 10 000
5 000 5 000
0 0
−5 000 −5 000
Natural ground

−10 000 −10 000

−15 000 −15 000
−20 000 −20 000
−25 000 −25 000
−30 000 −30 000
−35 000 −35 000
−40 000 −40 000
−45 000 −45 000
−50 000 −50 000
102 103 102 103
Frequency (Hz) Frequency (Hz)
(e) (f)
Fig. 8. Estimated impedances using the Miki model with the configuration d ¼ 4 m over 50 m in (left) summer and (right) winter for (a) and (b) the synthetic lawn, (c) and (d)
the grass lawn and (e) and (f) the natural terrain. Thick dotted lines ( ) represent the mean values of the real and imaginary parts of the estimated impedances plotted as thin
solid lines ( ).
 G. Guillaume et al. / Applied Acoustics 95 (2015) 27–36 33

50 m 500 m
10 10

0 0

Synthetic lawn −10 −10

SPL (dB)

SPL (dB)
−20 −20

−30 −30

−40 −40

−50 −50
102 103 102 103
Frequency (Hz) Frequency (Hz)

(a) (b)

10 10

0 0

−10 −10
Grass lawn
SPL (dB)

SPL (dB)
−20 −20

−30 −30

−40 −40

−50 −50
102 103 102 103
Frequency (Hz) Frequency (Hz)

(c) (d)

10 10

0 0
Natural ground

−10 −10
SPL (dB)

SPL (dB)

−20 −20

−30 −30

−40 −40

−50 2 −50 2
10 103 10 103
Frequency (Hz) Frequency (Hz)

(e) (f)
Fig. 9. Sound pressure levels relative to free field computed with the Weyl Van der Pol formula using the impedances estimated during ( ) the first campaign (i.e. in
summer) and ( ) the second campaign (i.e. in winter) with zS ¼ zR ¼ 2 m. For each ground and distance, the solid line corresponds with the sound pressure levels obtained
with the mean value of the grounds parameters and the dashed lines with this mean values more or less the standard deviations.

This increase of the spatial variability also occurs for the second source and the receiver are located at a height zS ¼ zR ¼ 2 m and
campaign (i.e. in winter). Moreover, the values of the RSDs of both are separated by distances d ¼ 50 m and d ¼ 500 m. The results
r and e are very close to the ones of the first campaign for the are shown in Fig. 9 for the three kinds of grounds and for both
synthetic lawn, while they are significantly different for grass propagation distances respectively.
and natural grounds. This indicates that natural grounds are more The results of the first campaign (red curves) are firstly consid-
sensitive to the seasonal moisture behavior. ered. The variability of the calculated spectra increases with the
ground ‘‘complexity’’ (i.e. the synthetic lawn, the grass lawn and
3.3.3. Influence on the predicted sound levels and their variability then the natural ground respectively). Indeed, the area between
The excess attenuations relative to free field have been com- the two dashed curves becomes larger on the left of the ground
puted in homogeneous propagation conditions with the measured dip. This is expected as the same spatial variability has been
values of r and e for each ground and for each campaign using obtained for the three kinds of grounds (see Section 3.3.2). For each
the Weyl–Van der Pol formula (Eq. (1)) in order to estimate the ground, the variability obviously increases with the distance. The
variability of the predicted sound pressure levels (SPL) due to variability is weak for the synthetic lawn and the standard devia-
the spatial dispersion of the impedance. For the calculations, the tion is of ±5 dB near the ground dip for both considered distances.
34 G. Guillaume et al. / Applied Acoustics 95 (2015) 27–36

The variability of the SPL becomes large for the grass lawn and the of the reflection incidence angle on the estimation of the surface
natural ground. Notably, for the natural ground with a propagation wave influencing the global sound attenuation and more particu-
distance d ¼ 500 m, the standard deviation reaches up to ±10 dB at larly, the ground dip magnitude and shape which represent a
the minimum of the ground dip and to ±15 dB between 200 Hz et key-point in the fitting procedure. A second point concerns the
250 Hz. selected impedance model. In the present paper, only two two-
Concerning the second experimental campaign (blue curves), parameter empirical models have been implemented. Knowing
the increasing variability with respect to the ground type also that other models are based on three or more parameters, it would
occurs. For the three types of grounds, a shift of the position of be interesting to implement the procedure with those models.
the mean ground dips towards higher frequencies is observed, However, the question is: Is it relevant if we consider the effort
what is due to the grounds dampness which modifies the absorp- to produce those parameter values in regard to the accuracy of
tion properties (the grounds are more reflective). The impact of the measured data and soil variability even on a few meters?
this season effect on the mean predicted sound levels is large for Finally, in our approach, we always assume flat grounds. What
the synthetic and the grass lawns: the position of the mean ground would happen if soil roughness were introduced? This latter topic
dip rises by 200 Hz compared to the dry grounds. This effect is less has recently been studied by Faure [15]. All these particular points
important for the natural ground (the shift is about 40 Hz), for have to be further investigated.
which the roughness effect leads to a weaker variation of the mean
impedance properties with the season. By the way, the season has
Appendix A. Estimated ground parameters
a reduced influence on the variability of these predicted sound
levels: The area between the blue dashed curves as well as
A.1. Estimated ground parameters with the Delany–Bazley model for
between the red dashed curves is almost the same for each ground,
the summer campaign
as if the blue dashed curves were a translation of the red dashed
curves towards higher frequencies. Then, this variability can still
Table A.1
be important at long distance for both the grass lawn and the natu-
ral ground [27]. However, regarding the mean values of the SPL
attenuations (Fig. 9(d) for instance), huge discrepancy can be Table A.1
observed between predictions in summer and winter periods Estimated ground parameters r (kN s m4) and e (m) with the Delany–Bazley model
(about 40 dB at 300 Hz). for both the synthetic and grass lawn in summer.

Position (Line 1) Synthetic Grass

e r e r
4. Conclusion
0–4 m 0.0132 231 0.0115 270
3–7 m 0.0124 301 0.0142 331
Outdoor sound propagation prediction over inhomogeneous 6–10 m 0.0119 288 0.0118 348
soils is always a challenge because of the variation of the ground 10–14 m 0.0127 268 0.0121 372
intrinsic parameters in space and over time. Whatever the prop- 13–17 m 0.0126 293 0.0120 336
16–20 m 0.0127 305 0.0344 246
agation distance, sound attenuation is strongly dependent on the
20–24 m 0.0125 357 0.0165 258
estimation of ground characteristic impedance. To lower uncer- 23–27 m 0.0124 336 0.0160 250
tainties as much as possible, it is essential to use an adapted 26–30 m 0.0126 258 0.0151 255
estimation procedure and a large amount of real in situ measure- 30–34 m 0.0142 250 0.0191 181
ments covering all the possible time and space ground variabilities. 33–37 m 0.0138 244 0.0177 202
36–40 m 0.0145 242 0.0184 224
The semi-automatic fitting method implemented along this 40–44 m 0.0126 268 0.0158 222
paper was carried out using two empirical impedance models 43–47 m 0.0130 231 0.0153 195
[10,26] on data recorded during two experimental campaigns over 46–50 m 0.0133 259 0.0176 233
summer and winter periods, along different linear profiles and for Mean 0.0130 275 0.0165 262
three kinds of soils presenting an increasing ‘‘complexity’’. SD 0.0007 36 0.0054 57
After obtaining the various values of the characteristic para-
meters, i.e. the airflow resistivity r and the effective thickness e
Table A.2
with their respective standard deviations for all the configurations, Estimated ground parameters r (kN s m4) and e (m) with the Miki model for the
a discussion on the impact of these estimated variations was carried synthetic lawn according to the season.
out from the various calculated sound pressure attenuations rela-
Position (Line 1) Summer Winter
tive to free field (SPL) under homogeneous propagation conditions.
e r e r
These results were obtained above the three kinds of grounds for
two propagation distances (50 and 500 m) and particularly, around 0–4 m 0.0124 307 0.0069 311
3–7 m 0.0106 389 0.0063 453
the magnitude and frequency location of the ground dip. Generally,
6–10 m 0.0107 385 0.0066 465
the discrepancy around mean predicted SPL increases with the 10–14 m 0.0115 351 0.0066 443
ground variability. In the case of propagation above a natural 13–17 m 0.0116 398 0.0064 458
ground at a distance of 500 m, this discrepancy around the ground 16–20 m 0.0121 432 0.0067 417
dip in the mid-frequency range [500 Hz1 KHz] can reach ±10 dB. If 20–24 m 0.0123 492 0.0060 495
23–27 m 0.0112 462 0.0059 567
we look at the seasonal effects, mainly related to ground humidity 26–30 m 0.0114 334 0.0062 476
which affects the absorbing quality of soils, on the same type of 30–34 m 0.0132 326 0.0062 476
ground, the impact in terms of attenuation is not negligible, but it 33–37 m 0.0126 318 0.0061 539
can be masked by roughness effects. 36–40 m 0.0126 320 0.0061 584
40–44 m 0.0113 356 0.0064 475
Even if this estimation procedure provides an interesting
43–47 m 0.0116 301 0.0058 487
improvement with respect to previous standardized methods, sev- 46–50 m 0.0119 361 0.0062 621
eral points still must be investigated. The first one, linked to the
Mean 0.0118 369 0.0063 484
geometrical configuration of the source and the microphones SD 0.0007 58 0.0003 74
seems to have a strong influence. This is mainly due to the effect
 G. Guillaume et al. / Applied Acoustics 95 (2015) 27–36 35

A.2. Estimated ground parameters with the Miki model for both
experimental campaigns

Tables A.2, A.3 and A.4

Table A.3
Estimated ground parameters r (kN s m4) and e (m) with the Miki model for the grass lawn according to the season.

Summer Winter
Position (Line 1) e r Position (Line 2) e r Position (Line 1) e r Position (Line 2) e r
0–4 m 0.0102 344 0–4 m 0.0142 378 0–4 m 0.0082 409 0–4 m 0.0066 990
3–7 m 0.0112 453 3–7 m 0.0125 418 3–7 m 0.0067 694 3–7 m 0.0063 913
6–10 m 0.0110 480 6–10 m 0.0122 459 6–10 m 0.0065 619 6–10 m 0.0065 513
10–14 m 0.0116 510 10–14 m 0.0160 418 10–14 m 0.0066 472 10–14 m 0.0057 935
13–17 m 0.0112 455 13–17 m 0.0150 354 13–17 m 0.0052 610 13–17 m 0.0053 697
16–20 m 0.0250 347 16–20 m 0.0250 301 16–20 m 0.0049 700 16–20 m 0.0055 569
20–24 m 0.0250 334 20–24 m 0.0130 392 20–24 m 0.0053 729 20–24 m 0.0040 990
23–27 m 0.0158 336 23–27 m 0.0121 415 23–27 m 0.0038 700 23–27 m 0.0042 990
26–30 m 0.0139 350 26–30 m 0.0133 448 26–30 m 0.0058 756 26–30 m 0.0064 721
30–34 m 0.0187 280 30–34 m 0.0155 319 30–34 m 0.0059 516 30–34 m 0.0056 990
33–37 m 0.0166 258 33–37 m 0.0151 320 33–37 m 0.0068 661 33–37 m 0.0050 990
36–40 m 0.0183 298 36–40 m 0.0172 266 36–40 m 0.0072 718 36–40 m 0.0057 675
40–44 m 0.0159 304 40–44 m 0.0196 247 40–44 m 0.0061 608 40–44 m 0.0063 964
43–47 m 0.0138 246 43–47 m 0.0162 320 43–47 m 0.0045 750 43–47 m 0.0050 990
46–50 m 0.0168 322 46–50 m 0.0180 270 46–50 m 0.0072 515 46–50 m 0.0059 574
Mean 0.0157 354 Mean 0.0157 355 Mean 0.0060 631 Mean 0.0056 833
SD 0.0047 82 SD 0.0034 69 SD 0.0012 108 SD 0.0008 185

Table A.4
Estimated ground parameters r (kN s m4) and e (m) with the Miki model for the natural ground according to the season.

Summer Winter
Position (Line 1) e r Position (Line 2) e r Position (Line 1) e r Position (Line 2) e r
0–4 m 0.0133 221 0–4 m 0.0093 281 0–4 m 0.0124 237 0–4 m 0.0166 190
3–7 m 0.0155 288 3–7 m 0.0176 226 3–7 m 0.0142 186 3–7 m 0.0266 113
6–10 m 0.0154 223 6–10 m 0.0147 185 6–10 m 0.0127 469 6–10 m 0.0241 91
10–14 m 0.0148 241 10–14 m 0.0172 155 10–14 m 0.0133 281 0–4 m 0.0260 75
13–17 m 0.0145 180 13–17 m 0.0250 93 13–17 m 0.0135 285 3–7 m 0.0204 118
16–20 m 0.0148 355 16–20 m 0.0250 73 16–20 m 0.0103 307 6–0 m 0.0195 123
20–24 m 0.0162 195 20–24 m 0.0206 104 20–24 m 0.0113 347 0–4 m 0.0152 244
23–27 m 0.0180 122 23–27 m 0.0250 119 23–27 m 0.0130 259 3–7 m 0.0140 240
26–30 m 0.0195 135 26–30 m 0.0242 86 26–30 m 0.0152 165 6–0 m 0.0124 333
30–34 m 0.0144 207 30–34 m 0.0204 135 30–34 m 0.0192 223 0–4 m 0.0152 266
33–37 m 0.0109 293 33–37 m 0.0250 100 33–37 m 0.0188 222 3–7 m 0.0122 351
36–40 m 0.0144 210 36–40 m 0.0250 128 36–40 m 0.0239 305 6–0 m 0.0167 277
40–44 m 0.0178 156 40–44 m 0.0250 110 40–44 m 0.0166 189 0–4 m 0.0093 551
43–47 m 0.0167 157 43–47 m 0.0225 84 43–47 m 0.0180 169 3–7 m 0.0213 213
46–50 m 0.0144 193 46–50 m 0.0195 116 46–50 m 0.0002 189 6–0 m 0.0157 267
Mean 0.0154 212 Mean 0.0211 133 Mean 0.0150 256 Mean 0.0177 230
SD 0.0021 63 SD 0.0047 58 SD 0.0036 81 SD 0.0052 124

References [10] Delany ME, Bazley EN. Acoustical properties of fibrous absorbent materials.
Appl Acoust 1970;3:105–16.
[11] Directive 2002/49/EC. Directive 2002/49/EC of the European Parliament and of
[1] ANSI/ASA S1.18-2010, 2010. Method for determining the acoustic impedance
the Council of 25 June 2002 relating to the assessment and management of
of ground surfaces. Accredited Standards Committee S1, Acoustics. American
environmental noise. Off J L 2002;189:0012–26.
National Standard (Revision of ANSI S1.19-1999).
[12] Dragna D, Blanc-Benon P. Physically admissible impedance models for time-
[2] Attenborough K. Acoustical impedance models for outdoor ground surfaces. J
domain computations of outdoor sound propagation. Acta Acust United Acust
Sound Vib 1985;99:521–44.
2014;100:401–10.
[3] Attenborough K, Bashir I, Taherzadeh S. Outdoor ground impedance models. J
[13] Dragna D, Blanc-Benon P, Poisson F. Time-domain solver in curvilinear
Acoust Soc Am 2011;129:2806–19. http://dx.doi.org/10.1121/1.3569740.
coordinates for outdoor sound propagation over complex terrain. J Acoust
[4] Bérengier M, Garai M. A state-of-the-art of in-situ measurement of the sound
Soc Am 2013:133. http://dx.doi.org/10.1121/1.4803863.
absorption coefficient of road pavements. In: International congress of
[14] Embleton TFW, Piercy JE, Daigle GA. Effective flow resistivity of ground
acoustics, Rome, Italia; 2001.
surfaces determined by acoustical measurements. J Acoust Soc Am
[5] Bérengier M, Gauvreau B, Blanc-Benon P, Juvé D. Outdoor sound propagation: a
1983;74:1239–44. http://dx.doi.org/10.1121/1.390029.
short review on analytical and numerical approaches. Acta Acust United Acust
[15] Faure O. Analyse numérique et expérimentale de la propagation acoustique
2003;89:980–91.
extérieure: effets de sol en présence d’irrgularités de surface et méthodes
[6] Blanc-Benon P, Dallois L, Juvé D. Long range sound propagation in a turbulent
temporelles (Numerical and experimental analysis of outdoor sound
atmosphere within the parabolic approximation. Acta Acust United Acust
propagation: grounds effects in the presence of surface irregularities and
2001;87:659–69.
time-domain methods). Phd thesis. École Centrale de Lyon. France; 2014.
[7] Carpinello S, L’hermite P, Bérengier M, Licitra G. A new method to measure the
[16] Gilbert KE, White MJ. Application of the parabolic equation to sound
acoustic surface impedance outdoors. Radiat Prot Dosim 2004;111:363–7.
propagation in a refracting atmosphere. J Acoust Soc Am 1989;85:630–7.
[8] Coleman T, Branch MA, Grace A. Optimization toolbox for use with MATLABÒ.
http://dx.doi.org/10.1121/1.397587.
User’s Guide Version 2; 1999.
[17] Guillaume G, Aumond P, Gauvreau B, Dutilleux G. Application of the
[9] Cramond A, Don C. Effects of moisture content on soil impedance. J Acoust Soc
transmission line matrix method for outdoor sound propagation modelling –
Am 1987;82:293–301.
36 G. Guillaume et al. / Applied Acoustics 95 (2015) 27–36

Part 1: Model presentation and evaluation. Appl Acoust 2014;76:113–8. Geosciences, Architecture. Nantes, France. <https://tel.archives-ouvertes.fr/
http://dx.doi.org/10.1016/j.apacoust.2013.07.011. tel-00906913/>.
[18] Guillaume G, Fortin N. Optimized transmission line matrix model [25] Lihoreau B, Gauvreau B, Bérengier M, Blanc-Benon P, Calmet I. Outdoor sound
implementation for graphics processing units computing in built-up propagation modeling in realistic environments: application of coupled
environment. J Build Perform Simul 2014;7:445–56. http://dx.doi.org/ parabolic and atmospheric models. J Acoust Soc Am 2006;120:110–9. http://
10.1080/19401493.2013.864335. dx.doi.org/10.1121/1.2204455.
[19] Hornikx M, De Roeck W, Desmet W. A multi-domain Fourier pseudospectral [26] Miki Y. Acoustical properties of porous materials – modifications of Delany–
time-domain method for the linearized Euler equations. J Comput Phys Bazley models. J Acoust Soc Jpn (E) 1990;11:19–24.
2012;231:4759–74. [27] Ostashev VE, Wilson DK, Vecherin SN. Effect of randomly varying impedance
[20] Hornikx M, Waxler R, Forssén J. The extended fourier pseudospectral time- on the interference of the direct and ground-reflected waves. J Acoust Soc Am
domain method for atmospheric sound propagation. J Acoust Soc Am 2011;130:1844–50. http://dx.doi.org/10.1121/1.3624817.
2010;128:1632–46. http://dx.doi.org/10.1121/1.3474234. [28] Premat E, Gabillet Y. A new boundary-element method for predicting outdoor
[21] Horoshenkov K, Mohamed M. Experimental investigation of the effects of sound propagation and application to the case of a sound barrier in the
water saturation on the acoustic admittance of sandy soils. J Acoust Soc Am presence of downward refraction. J Acoust Soc Am 2000;108:2775–83. http://
2006;120:1910–21. dx.doi.org/10.1121/1.1290512.
[22] Junker F, Gauvreau B, Cremezi-Charlet C, Blanc-Benon P, Cotté B, Ecotière D. [29] Sabatier J, Raspet R, Frederickson C. An improved procedure for the
Classification of relative influence of physical parameters for long range determination of ground parameters using level difference measurements. J
acoustic propagation. In: Internoise, Honolulu, USA; 2006. Acoust Soc Am 1993;94:396–9.
[23] Kruse R. Application of the two-microphone method for in-situ ground [30] Soh JH, Gilbert KE, Frazier WG, Talmadge CL, Waxler R. A direct method for
impedance measurements. Acta Acust United Acust 2007;93:837–42 (6). measuring acoustic ground impedance in long-range propagation
[24] Leroy O. 2010. Estimation d’incertitudes pour la propagation acoustique en experiments. J Acoust Soc Am 2010;128:EL286–93. http://dx.doi.org/
milieu extérieur (Estimation of uncertainties for outdoor acoustic 10.1121/1.3501116.
propagation). PhD thesis. Doctoral School SPIGA Science Engineering, [31] Zwikker C, Kosten CW. Sound absorbing materials. New York; 1949.