You are on page 1of 22

Applied Acoust~s, Vol. 51, No. 3, pp. 317~~338.

1997
cj 1997 Elsevier Science Ltd. All rights reserved
Printed in Great Britain
PII: SOOO3-682X(97)00001-7 0003-682X/97 $17.00 + 0.00
ELSEVIER

Ambient Sea Noise Dependence on Local, Regional


and Geostrophic Wind Speeds: Implications
for Forecasting Noise

Douglas H. Cato and Sandra Tavener

Defence Science and Technology Organisation, PO Box 44, Pyrmont, NSW 2009,
Australia

ABSTRACT

Measurements of ambient sea noise in two regions near Australia are pre-
sented as a function of wind speed (a) measured locally (at a buoy, dis-
tance about 120m), (6) measured at weather stations (distances up to
51 km) and (c) predicted from the geostrophic air flow. Estimates of
geostrophic winds from atmospheric pressure contours are used in fore-
casting wind speeds. Good correlation of noise with local wind speed was
observed. Poorer correlations were observed between noise and regional
wind speeds or those predicted from geostrophic wind speeds: trends were
similar but the spread of the data points was greater. Correlations were
better when the windfield had minimal influence by the presence of land up-
wind. The variation of geostrophic wind speed with time generally followed
that of the local wind speed but peaks and troughs were sometimes dis-
placed in time by a few hours, accounting for some of the spread of the
data. These results suggest that geostrophic wind, and thus forecasts of
wind speeds, would be useful in forecasting sea noise. 0 1997 Elsevier
Science Ltd.

Keywords: Ocean noise, sea noise, sea surface noise, wind dependent noise.

INTRODUCTION

Ambient noise in the ocean varies widely over relatively short time scales as a
result of variations in wind, rain or biological activity and, as a consequence,
causes a wide variation in sonar performance. The prediction of sonar per-
formance from a prediction of the ambient noise and the other terms in the
317
sonar equation provides important tactical information for the use of sonar
at sea, and this has led to the development of computer-based tactical
decision aids. While estimation of the likely performance at the time of
prediction is useful, more value would be obtained by providing a ,fbrecast
of sonar performance, particularly in terms of how conditions might
change over scales of an hour, several hours or even days. This, of course,
requires a forecast of the environmental terms in the sonar equation. The
purpose of this paper is to examine some of the factors involved in forecasting
sea surface generated noise, the prevailing component of ambient noise, based
on some controlled experiments at sea. In particular we examine the corre-
lation with geostrophic wind speeds since this is the source of wind speed
forecasts, and sea surface generated noise is correlated with wind speed.
Sea surface generated noise is the prevailing noise of the ocean. and its
importance was identified in the earliest studies of ambient noise.’ It was
originally considered to be a function of sea state, but later studies found
that the noise correlated better with wind speed2.3 and it has since been
known as “wind-dependent noise”. More recent work has shown that the
noise at frequencies above about 100 Hz is generated by the oscillation of the
air bubbles formed by the entrainment of air as waves break.4 s The extent
of wave breaking at the sea surface usually shows a stronger dependence on
wind speed than on properties of the surface waves.
There is also evidence in the above studies, and in studies reported in
Kerman, 9.‘o that the bubbles radiate as monopoles. Thus the far field noise
can be modelled as due to a surface distribution of dipoles, representing the
sources and their out of phase surface images. The dipole axes (of maximum
radiation) would be normal to the sea surface, consistent with the source
directionality inferred by Ferguson and Wylliel ’ from directional noise
measurements.
While surface generated noise shows a dependence on wind speed, it might
be expected that the influence of very short term fluctuations in wind speed
will be small because of the averaging inherent in the development of the
noise field. Part of this averaging is spatial and results from the fact that the
noise is the summation of the contributions from sources spread over a wide
area. This will provide averaging of the effects of temporal fluctuations in
wind speed. In addition, wind speeds are correlated over significant distan-
ces. It might therefore be expected that while the best correlation between
noise and wind speed would result if wind measurements were local, signifi-
cant correlation would also be found with regional wind speeds measured at
some distance from the hydrophone and with geostrophic wind speeds
determined from the atmospheric pressure contours. This may allow the
noise to be predicted from regional wind speed estimates and forecast from
atmospheric pressure movements.
Ambient sea noise 319

Geostrophic wind is the flow of air in response to atmospheric pressure


gradients and can be calculated from charts of pressure contours or isobars.
Pressure gradients cause the air to flow towards regions of lower pressure,
but the Coriolis force deflects the direction of the wind so that it is roughly
parallel to the pressure contours. The actual wind at the earth’s surface
depends on this geostrophic flow but is modified by the deformation and
frictional drag of the surface which reduces the flow speeds near the surface.
This drag reduction is usually greater over land with local effects, such as
sheltering by hills, having an effect.
Forecasting of wind speed is based mainly on estimates derived from con-
tours of atmospheric pressure (isobars), given the expected movements of
these pressure fields. Vessels at sea can be expected to have access to current
isobaric charts. Forecasting of at least the broad features of the wind field,
and thus the ambient noise field, should be possible. In addition there will be
other information available about wind speeds so that, in general, forecast-
ing should- be better than simply what could be obtained from the geos-
trophic flow. Over the open ocean, the geostrophic wind, with the
appropriate correction for the drag effects of the sea surface, should be a
reasonable estimate of the surface wind. The drag effects over land are more
complicated and varied, so a poorer prediction of winds at sea might be
expected if the weather pattern has approached the site from land rather
than from sea.
To test this we compared the correlation of sea noise with (a) wind speed
measured locally, (b) wind speed measured at regional weather stations and
(c) wind speed estimated from the geostrophic wind.

THEORETICAL CONSIDERATIONS

Wind-dependent ambient noise measured at a point is the result of contri-


butions from sources spread over a significant area of the sea surface. The
extent to which the noise correlates with broad scale estimates of wind speed,
such as those obtained from the geostrophic wind, could be considered to
depend on two factors. The first is the extent that the broad scale wind is
correlated with the local wind in the vicinity of the hydrophone. The second
is the spatial scale over which noise sources contribute significantly to the
received noise. This provides spatial averaging of the noise which will have
an effect similar to a spatial averaging of the wind.
Evidence4” indicates that wind-dependent noise sources and their surface
images would effectively radiate as dipoles with vertical (or near vertical)
axes. Consequently, radiation efficiency decreases as the angle relative to the
horizontal of rays from the source becomes less steep. The low angle rays,
320 D. H. Cato, S. Tavener

however, are the ones that propagate with least loss-they are more likely to
be trapped in a duct and usually suffer less bottom reflection loss than higher
angle rays. Thus the area of contributing sources will be a balance between
the angular radiation efficiency of the sources and the angular dependence of
the particular propagation conditions. It will ultimately be limited by absorp-
tion attenuation or the size of the water mass, if not limited by other effects.
This is effectively a spatial weighting of the contribution of sources which
generally decreases with increasing distance from the hydrophone. An idea
of this weighting can be obtained from available models of the ambient noise
field at a receiver (references 12-14, for example). Kuperman and Ingenito12
provide a normal mode model for noise in shallow water. They note that the
discrete modes dominate for conditions of low loss (ducting, low bottom
loss) allowing contributions from large distances whereas, in high loss con-
ditions, the continuous modes tend to dominate, substantially limiting the
source area. HarrisonI presents a ray model and applies it to a shallow
water environment, giving examples of the noise received in and below a
duct. In the case of the receiver within the duct, the weighting decreases only
slightly with range out to the maximum distance shown of 40 km (Fig. 9 of
reference 14). For the receiver below the duct, most of the energy is received
from sources within a few kilometres’ range.
Thus the area of contributing sources may vary substantially depending on
the propagation conditions. The minimum area would occur under condi-
tions of high loss: downward refraction and high bottom loss. Since the
sources radiate preferentially downwards, an idea of this minimum area in
shallow water can be obtained from a simple model which sums incoherently
the contributions from sources distributed over the surface of a non-refract-
ing ocean. Such a model is given by Urick15 (p 227) or could be obtained by
simplifying the more sophisticated models of Chapmani or Harrison.14
Consider a circular area at the sea surface of radius Y centred above the
hydrophone, which is at depth h. An incremental increase dr in the radius
forms an annulus at slant range 1= dm from the hydrophone. The
contribution by direct path from the sources within this annulus, assuming
no refraction, is dl = 2rrrdrZ, sin2 @/I2 where I, is the source strength per unit
area and 8 is the ray angle to the horizontal, i.e. the angle between I and the
horizontal. The term sin28 accounts for the dipole directivity of the sources.
From the geometry, r = h/tan0 and 1= h/sine, also drjde =-h/sin28. The
contribution from all sources within the circular area of radius r centred
above the hydrophone is then
O(r)
Z(r) = -2rrI, sin 19cos 8d6’= rrl, cos* 0(r) (1)
s
n/2
Ambient sea noise 321

where 8(r) is the angle of the ray from the maximum range r. Expressions of
similar form could be obtained by evaluating eqn (6) of reference 13 or eqn
(7) of reference 14, with the appropriate simplifications. For an infinite ocean
I(cc) = nl,, so that the proportion of noise arriving from a source area of
radius Yis

(2)

where e = r/h expresses the maximum range of the contributing circular area
in terms of the water depth.
From eqn (2) it is evident that 90% of the energy by direct path comes
from sources within a circular area with radius three times the water depth
(120 m at the sites of measurement).
The contributions by bottom reflected paths can be modelled using the
method of images as shown by Chapman. l3 For simplicity in the geometry
we assume that the hydrophone is on the bottom and ignore the acoustic
interaction with the bottom at the hydrophone. For n bottom reflections,
and no loss on reflection, the derivation is of the same form as above but
with (2n + 1)h replacing h. Thus the proportion of energy arriving from the
circular area bounded by x for paths with n bottom reflections, is

I@,4 c2
(3)
I(oo,==2+(2n+ 1)2

Note that Z(r,n) is independent of n for no bottom reflection loss. Since we


expect that bottom loss will be significant, the contribution of Z(r,n) can be
expected to decrease with increasing n. For a particular value of
I(r, n)/Z(oo, n), e increases as n increases, so that as the number of bottom
reflections increases, the area of significant contribution by paths with n
reflections expands. For example, 90% of the energy for paths with one
bottom reflection is from horizontal ranges up to nine times the water depth,
for two reflections up to 15 times the water depth, and for 10 reflections up
to 63 times the water depth, etc.
This suggests that the minimum scale of spatial averaging at the sites of
measurements would be about 240 m (based on the minimum area of sources
contributing 90% of the energy by direct path). This is the lower limit-the
actual scale may be very much larger, possibly tens of kilometres.‘2-14 Pre-
vious analysis I6 of the data from Spencer Gulf suggests that the range of
significant contribution is much less than the distance between the Wedge
and Berry sites of 44 km. Wind speeds at Wedge and Berry were generaJJy
well correlated except when the wind at Berry was affected by an afternoon
322 D. H. Calo, S. Tavener

sea breeze that was not evident at Wedge. When the wind differed signifi-
cantly at the two sites, the noise levels followed the local wind speeds and
showed no evidence of correlation with the different wind speeds at the other
site.
The other factor affecting the correlation of surface generated noise with
broad scale estimates of wind speed is the extent to which the broad scale
wind estimates correlate with the local wind over the area of contributing
sources. Wind flow is characterised by gusting due to the passage of eddies in
what is essentially a turbulent boundary layer of the atmosphere. This gust-
ing advects at the mean wind speed. Van der Hover-r” presents a spectrum of
the wind speed fluctuations measured near the ground and this shows a high
frequency spectral peak at about one cycle per min. Significant temporal and
spatial fluctuation in wind speed, and thus wave breaking, is therefore to be
expected over scales of the order of one minute and hundreds of metres.
Since the spatial averaging of the noise will be over distance scales of at least
the same order as these fluctuations, but probably much larger, this will
provide temporal averaging due to the advection of the wind field. To some
extent, therefore, the effect on the received noise of fluctuations in wind
speed and the consequent fluctuations in wave breaking on these scales will
be averaged out. The next peak of temporal fluctuation in the wind speed is
of the order 12 h (reference 17) corresponding to distance scales of hundreds
of kilometres.
Thus, while we can expect noise to correlate best with the local wind speed.
as measured on the nearby anemometer buoy in this experiment, there
should also be significant correlation with regional wind speeds if temporal
and spatial scales are much less than the 12 h or hundreds of kilometres.
Some previous analysis of data at the sites of measurementi has shown
reasonably good correlation of noise and wind speed for separations (of
noise and wind measurements) up to the maximum measured of 55 km spa-
tially and 1.5 h temporally (except at the times of the sea breeze at Berry, as
discussed above). Many of the early measurements that showed correlation
of noise with wind speed used anemometers some distance from the site of
the noise measurements.
This leads to the expectation that regional wind speeds can be used effec-
tively to predict sea surface generated noise, and that forecasts of wind speed
would be useful in forecasting the noise. Forecasting of wind speeds is based
mainly on estimates derived from contours of atmospheric pressure (isobars),
given the expected movements of these pressure fields. Pressure gradients
cause a flow of air towards regions of lower pressure. Coriolis force, due to
the rotation of the earth, deflects the winds so that the flow is broadly par-
allel to the pressure contours (clockwise around regions of low pressure in
the southern hemisphere).18
Ambient sea noise 323

The geostrophic wind speed u in terms of the atmospheric pressure


gradient aplan normal to the isobars is given in reference 19,

u=-- 1 aP (4)
k?fan

where p is the air density, andf= 252 sin4 is the Coriolis parameter, with Q
being the angular speed of the earth and 4 the latitude. Pressure contours are
usually given as the equivalent values at the sea surface for standard tem-
peratures, thus determining the value of the air density for eqn (4).
The earth’s surface causes a drag on the geostrophic flow, slowing the flow
speeds as the surface is approached. Actual wind speeds, therefore, are less
than the geostrophic wind speeds and can be predicted by applying a cor-
rection for drag effects to the geostrophic wind speed. The magnitude of the
drag effect varies, and tends to be higher over land than over the sea. Roll’*
presents two theoretical models of the frictional effect over the sea, and these
suggest that the surface wind would be from 0.55 to 0.61 of the geostrophic
wind speeds in the range 5-20 m SK’ for typical conditions. Petterssen2’ gives
some examples of measured ratios of surface to geostrophic wind speeds and
notes that for strong winds at sea this ratio is about 2/3. Reference 21 notes
that over the sea surface, the wind (at a height of 10m) is reduced to about 2/
3 of the geostrophic wind speed, and that the greater friction over land may
result in a speed of l/3 to l/2 of the geostrophic wind speed.

MEASUREMENTS

Measurements of ambient sea noise and wind speed at a nearby buoy were
made in two regions near Australia (Fig. 1): (a) in Spencer Gulf. South
Australia, at two sites (labelled “Wedge” and “Berry”), water depth 40m
(Fig. 2) and (b) off Perth, Western Australia, at one site in the open ocean
about 25 km west of the coast, water depth 40m (Fig. 3).
Wind speed data were also obtained from weather stations in both regions:
at Neptune Is. about 51 km SW of the Wedge site in Spencer Gulf, and at
Rottnest Is. about 22 km north of the site of the measurements off Perth.
The sites in Spencer Gulf are near the entrance to the Gulf, facing the
Southern Ocean to the south with some obstruction by islands, but little
swell is evident. Traffic noise is very low in Spencer Gulf because shipping
lanes are a considerable distance to the south and propagation loss over the
long distances through shallow water is high. Weather patterns move pre-
dominantly over the ocean but with some influence of land towards the
Spencer Gulf sites.
324 D. H. Cato, S. Tavener

10

120 140 160

Degrees East

Fig. 1. Map of Australia showing the two regions of measurement.

Neptune Is

35.5 -

136 137 138

Degrees East

Fig. 2. Map of the Spencer Gulf region showing the sites of measurement at Wedge and
Berry, and the weather station at Neptune Is.
Ambient sea noise 325

INDIAN Rottnest
OCEAN Island
Perth
.
32 k;s

. B
Recording ,
Degrees
position
South

31
115 116
Degrees East

Fig. 3. Map of the Perth region showing the measurement site and the weather station on
Rottnest Is.

The Perth site faces the open waters of the Indian Ocean for considerable
distances to the west, north and south and is subject to oceanic swell. There
is significant traffic noise at this site because of the shipping lanes to Perth
and other Australian ports. Weather patterns generally move west to east,
from the ocean towards the site off Perth.
Noise measurements were made using bottom moored systems. Each system
comprised an ITC 1032 hydrophone laid on the bottom and connected to a
nearby canister containing a low noise amplifier and a tape recorder. The
noise field on the bottom may be complicated by bottom interaction effects,
but since these would be the same for all correlations of noise with the wind
speed estimates, they should not affect the comparison of the effectiveness of
predicting noise from local, regional or geostrophic wind speeds. Samples of
noise of 20 s duration were recorded at intervals of 15 min over periods of
about six days using timer controlled Sony TCD-D7 digital tape recorders.
System frequency response was 10 Hz to 14 kHz. The data reported here were
sampled over the following periods: Wedge, from 20 to 27 November 1993;
Berry, from 10 to 13 November 1993; Perth, from 10 to 16 November 1994.
Wind speed measurements were made at each site using an R.M. Young
anemometer at a height of 3 m, on a purpose-built buoy moored about 120 m
from the hydrophone. The wind speed was recorded every 20 s as the average
326 D. H. Cato, S. Tavener

of the preceding 20 s in Spencer Gulf, and every 10 s as the average of the


preceding 10 s off Perth, on Unidata data loggers. The measured speeds were
corrected to the equivalent for a height of 10 m using standard wind speed
profiles (effectively multiplying by 1.136). Wind and noise recordings ran
concurrently at each site. Wind speed measurements obtained from the
weather station at Neptune Is. in Spencer Gulf were made at 3 h intervals
from 0300 each day, except that there was no reading at midnight. Each
measurement was the average of the lulls and gusts observed by eye over a
period of one minute. The measurements from the weather station at
Rottnest Is. were made at 1Omin intervals commencing on the hour, each
being the average of the speed over the 1Omin period. The anemometers at
both weather stations were 10 m above ground level, which was 32 m above
sea level on Neptune Is. and 43 m on Rottnest Is.

ANALYSIS

Samples of recorded ambient noise were analysed in one third octave bands,
using a Brtiel and Kjazr frequency analyser type 213 1, and converted to
spectrum levels. The averaging time was 8 s for the Wedge and Perth data
and 16 s for the Berry data, out of the 20 s of sample on tape. The samples
were carefully scrutinised for evidence of noise from non-wind-dependent
sources such as ships, whales or rain and, if present, these samples were not
included in the analysis.
Wind speed recorded on the anemometer buoy was averaged over 20 s (by
averaging two consecutive contiguous 10 s samples, in the case of the Perth
data). The choice of this averaging time was a compromise between being
long enough to match the spatial averaging inherent in the received noise
field, but not so long as to smooth out fluctuations that might correlate with
the noise. The estimated minimum spatial averaging of the noise field would
be over an area of the order of 240 m diameter, which would correspond to a
temporal averaging of the wind of 24 s at an advection speed of 10 m SC’.
The wind and noise samples were synchronised in time using the internal
(crystal controlled) clocks of the anemometer data logger and the tape
recorder. Noise level was plotted as a function of the logarithm of wind
speed so that a dependence of noise intensity on a power of the wind speed
would appear as a straight line. This follows the approach originally estab-
lished by Piggott22 and found to be useful in many later studies. Linear
regression lines were calculated on these plots. Where data at low wind
speeds showed evidence of significant contribution from non-wind-depen-
dent noise, the data at these wind speeds were excluded from the regression
line calculations. Only data at 500Hz are presented in this paper, this fre-
Ambient sea noise 321

quency being representative of the broad peak in the mid frequency wind-
dependent noise. 2 Wind dependence was evident in the data over a broad
range of frequencies, limited by traffic noise off Perth below about 100 Hz
and shrimp noise at both sites above a few kHz.
The noise spectrum level was correlated with the logarithm of the wind
speed measured at the same time at the weather station in each region.
Geostrophic wind speeds were calculated from synoptic charts of standard
surface air pressure contours (isobars) which were derived from measure-
ments of air pressure corrected to the equivalent at sea level for the standard
temperature. The pressure gradient aplan of eqn (4) was approximated by
the difference in pressure between adjacent isobars divided by the distance
between the isobars along a line normal to the isobars. This is a reasonable
approximation where the contour pattern is relatively simple and distance
between contours changes slowly in moving across the contours, i.e. where
the rate of change of pressure gradient is small. It will be a poorer approxi-
mation where the pressure gradient is changing significantly near the position
of interest so that it differs significantly from the average gradient between
isobars. In a small number of cases, the contour pattern was so complicated
that it was not possible to make a reasonable estimate of the pressure gradi-
ent, and these were not included in the analysis.
All estimates of geostrophic wind speeds were corrected for the effect of
drag by reducing the values to 2/3 of the uncorrected values, consistent with
empirical estimates of the ratio of surface (at a mean height of 10m) to
geostrophic wind speeds. We refer to these surface wind speed predictions
obtained from geostrophic winds as the drag corrected geostrophic wind
speeds. The noise spectrum level measured at the time of each synoptic chart
was correlated with the logarithm of the estimated drag corrected geos-
trophic wind speed.

RESULTS

Figure 4 shows noise level at 500Hz as a function of local wind speed mea-
sured at the anemometer buoy at the Wedge site in Spencer Gulf. A clear
dependence of noise on wind speed is evident. Also shown is the linear
regression line calculated for noise level as a function of the logarithm of the
wind speed. Data for wind speeds less than 2.2 m s-’ were not included in the
calculation because of the evidence of non-wind-dependent noise which
would bias the result. This site has very low non-wind-dependent noise and
so provides a good estimate of the dependence of noise on wind speed.
The equation of the regression line is of the form originally established by
Piggott22 for wind-dependent noise:
328 D. H. Cato. S. Tavener

Wind speed (m/s)

Fig. 4. Noise level at 500 Hz versus measured local wind speed at the Wedge site in Spencer
Gulf. The regression line for data above a wind speed of 2.2 m s -’ is also shown.

L = 100 log,,(u) + b (5)

where L is noise level and u is wind speed. This provides a relationship of the
form I cx u’, where I is the noise intensity. In this result, a = 3.35, b = 40.4,
so that noise intensity is approximately proportional to the cube of the wind
speed. There are 525 data points in Fig. 4. Of these, 497 were used in com-
puting the regression line, the standard deviation of the noise level values
from the regression being 2.5 dB.
Figure 5 shows noise level as a function of the measured local wind speed
for the Perth site. The linear regression line was calculated by rejecting data
at wind speeds below 4 m s-l where non-wind-dependent noise was signifi-
cant. The regression line for the Perth data gives a = 2 . 37, b = 45 . 6, which
is not as steep as for the Wedge site and is also lower in level for wind speeds
of 4 m s-l or above. Differences between locations in both the noise level and
the rate of increase in level with wind speed have been found in comparisons
of other studies of wind-dependent noise (see Fig. 3 of reference 23, for
example). There are 257 data points in Fig. 5, of which 241 were used to
calculate the regression line, the standard deviation of points from the
regression line being 1.5 dB.
Part of the reason for the difference in the rates of increase between the
Wedge and Perth data may be due to the presence of significant non-wind-
dependent noise, mainly traffic noise, in the Perth data. This would result in
Ambient sea noise 329

a less steep slope in the regression line since it would increase noise levels at
low wind speeds. Although we rejected data which had evidence of shipping
noise, traffic noise is the nondescript background from distant shipping2 and
is not detectable as such so cannot be removed from the data. Removing
data for wind speeds below 4 m s-’ from the Perth data removes those data
points that are clearly non-wind-dependent, but does not remove points
where non-wind-dependent noise contributes significantly to the measured
noise in the range of wind speeds where the non-wind-dependent and wind-
dependent noises are confused.
If the regression line for the Wedge data is calculated for the same range of
wind speeds as that for the Perth data, i.e. speeds >/ 4m s-l, the resulting
values of a and b are 2.6 and 46.7, respectively, much closer to the Perth
result. This implies that the rate of increase in the Wedge data is greater for
winds below 4 m SC’ than for higher winds, suggesting that the dependence of
noise on wind speed may not be as simple as that given in eqn (5). The
standard deviation of the points from this regression line for wind speeds
>4m s-‘, is 1.8 dB, significantly less than for the line for wind speeds >
2.2 m s-‘. This is also consistent with a change in the rate of increase in the
region of 4 m s- ’ . Non-wind-dependent noise is exceptionally low in Spencer
Gulf, so the anomalous behaviour for wind speeds below 4 m s-’ may not be
evident at many other locations.

90

80

60

Wind speed (m/s)

Fig. 5. Noise level at 500 Hz versus measured local wind speed at the Perth site. The regres-
sion line for data above a wind speed of 4 m S-I is also shown.
330 D. H. Cato. S. Tavener

Figure 6 shows noise at the Wedge site as a function of the wind speed at
the weather station on Neptune Is., 51 km to the south west. While showing
a dependence of noise on wind speed there is significantly greater spread of
data than in Fig. 4 where wind speed was measured within about 120 m of
the hydrophone. The regression line from Fig. 4 is also shown for
comparison. The method of measuring wind speed on Neptune Is. required
averaging the lulls and gusts by eye over a period of one minute and taking
the mean of these values. This is likely to be less accurate than the electronic
averaging used in the other measurements and may partly explain the greater
spread of data in Fig. 6.
Figure 7 shows noise at the Perth site as a function of the wind speed at the
weather station on Rottnest Is. 22 km to the north, and the associated
regression line, for which a=2.44, b-44.2, similar to the values obtained
using the local wind speed (Fig. 5). The standard deviation of the noise level
data from the regression line is 2.2 dB so that there is a greater spread of
data than for the plot with the local wind speed (Fig. 5) but still a good
relationship. The geographical separation of the noise and wind measure-
ments of Fig. 7, and the differences in averaging times (8 s for noise and
IOmin for wind speed) may be expected to result in a greater spread in the
data points than for the plot of noise level with local wind speed in Fig. 5.

..
_;li
&’
. l . .,.
.
:
8:. . l

.
. .
. .
.

/
I /

Wind speed (m/s)

Fig. 6. Noise level at 500 Hz at the Wedge site in Spencer Gulf versus wind speed measured at
Neptune Is., 51 km to the south west. Also shown is the regression line for the plot of noise as
a function of local wind from Fig. 4.
Ambient sea noise 331

The fact that the spread of data points is not much larger in Fig. 7 than in
Fig. 5 suggests that there is significant spatial averaging in the noise field to
compensate for the much longer average of 10min for the wind speed at
Rottnest Is. and the spatial separation between the wind and noise mea-
surements. If the wind field advects at a rate comparable to the wind speed,
this result suggests that the length scale for the minimum area of spatial
averaging is of order 10 km for the range of wind speeds observed.
The greater spread of data in Fig. 6 compared with that of Fig. 7 may be
due to the fact that Neptune Is. is further from the Wedge site than Rottnest
Is. is from the Perth site. However, in an earlier study, noise measured at a
site further seaward off Perth, 55 km from Rottnest Is., showed a better
dependence on wind speed at Rottnest Is. than observed in Spencer Gulf for
similar separation of noise and wind measurement.16
A significant factor in the poorer correlation in Fig. 6 compared with Fig. 7
is thought to be the simpler topographical situation off Perth and the lesser
effect of land on the wind field up-wind of the measurement positions.
Although a variety of wind directions were experienced during the measure-
ments, the wind field was generally more from seaward than over land. Also
the effect of land on the wind is likely to be similar at both the hydrophone
position and Rottnest Is. (see Fig. 3). The Spencer Gulf region has a more
complicated topography with the presence of islands and the peninsulas

90

80

60

Wind speed (m/s)

Fig. 7. Noise level at 500Hz at the Perth site versus wind speed measured at Rottnest Is.,
22 km to the north. The regression line for the data is also shown.
332 D. H. Cato, S. Tavener

likely to affect the air flow. Wind data were from a range of directions in these
measurements. The measurement site and Neptune Is. are likely to be affec-
ted differently by the topography for different wind directions (see Fig. 2).
Figures 8 and 9 show noise as a function of drag corrected geostrophic
wind speed for noise measured at Wedge and Berry sites, respectively, in
Spencer Gulf (over different time periods). Figure 10 shows similar results
for the Perth site. Also shown on each plot is noise versus wind speed mea-
sured locally at the same times as the estimates for the drag corrected geos-
trophic wind speed. There are fewer data points on these figures than in
Figs 4, 5 and 7 because the charts of air pressure contours that provide the
geostrophic wind speeds are produced at 3 h intervals. The drag correction to
the geostrophic wind simply moves the data points a fixed amount to the left
on the graph (since it is a logarithmic scale). It is apparent that there is a
much greater spread of data points in the plots with drag corrected geos-
trophic wind speed, though there is still evidence of a correlation of noise
with wind speed, and the points cluster about the data points for the noise
dependence on the local wind speeds.
The correlation of noise with drag corrected geostrophic wind speed
appears to be better off Perth than in Spencer Gulf, probably because of the
simpler topography off Perth, as discussed above.

$- got

Wind speed (m/s)

Fig. 8. Noise level at 500Hz at the Wedge site versus the wind speed predicted from the
geostrophic wind speed, i.e. after correction for drag effects (n), and versus the local wind
speed measured at the same times (a).
Ambient sea noise 333

11:
0.1 1 10

Wind speed (m/s)

Fig. 9. Noise level at 500 Hz at the Berry site in Spencer Gulf versus the wind speed predicted
from the geostrophic wind speed, i.e. after correction for drag effects (n), and versus the local
wind speed measured at the same times (a).

Wind speed (m/s)

Fig. 10. Noise level at 500Hz at the Perth site versus the wind speed predicted from the
geostrophic wind speed, i.e. after correction for drag effects (n), and versus the local wind
speed measured at the same times (0).
D. H. Cute. S. Tavtwr

Local time (h) from OO:OO,20 November 1993

Fig. 11. Variation with time over eight days of the noise level at 500 Hz (- -a- -), local wind
speed (-•-) and drag corrected geostrophic wind speed (- -O- -) at the Wedge site.

20. I I I , I I I , / I / / I 1 I -110

- 100

lo-
- 90

-20”’ 0 24 48 72 96

Local time (h) from OO:OO,10 November 1993

Fig. 12. Variation with time over four days of the noise level at 500 Hz (- -a- -), local wind
speed (-a-) and drag corrected geostrophic wind speed (- -Cl- -) at the Berry site.
Ambient sea noise 335

O-

- ,’
,*.:
‘1 :’
k
A_
’ y,/
fl
‘-,

I/
.=*r------* \
‘1
.
- 70

- 60
-10 -

- 50

-*o/400 24 48 72 96 120 144 168

Local time (h) from OO:OO,10 November 1994

Fig. 13. Variation with time over seven days of the noise level at 500 Hz (- -@- -), local wind
speed (-a-) and drag corrected geostrophic wind speed (- -a- -) at the Perth site.

The variations with time of the noise level, the measured local wind speed
and drag corrected geostrophic wind speed, are shown in Fig. 11 for the
Wedge site, Fig. 12 for the Berry site and Fig. 13 for the Perth site. All three
variables show similar trends with time, with the local wind speeds and noise
being very similar. There is a tendency for the peaks and troughs in the drag
corrected geostrophic winds to be offset in time compared with the local
wind and the noise, and this is particularly evident in Fig. 13. This has the
effect of spreading the data in the plots of noise level versus drag corrected
geostrophic wind speed. There are a number of points in which the trends in
the geostrophic winds diverge significantly from those of the local wind or
the noise. Some of the divergence occurs as wind changes direction so that
more of the air flow is from the land. In such cases the surface winds would
be overestimated. An example is the peak at about 155 h at the Wedge site in
Fig. 11. The wind had changed direction about 24 h earlier from predomi-
nantly south west to north west, i.e. from the sea to a direction from the
continental land mass.
Figures 1l-13 suggest that the dependence of noise on drag corrected
geostrophic wind speed is better than implied by Figs 8-10. The main trends
in the noise are predicted by the geostrophic wind, with some error in the
times of changes.
336 D. H. Cato, S. Tavener

DISCUSSION

Sea surface noise measured in two regions shows the expected good correlation
with wind speed measured locally. Off Perth, the correlation between 8 s
averages of noise with 10 min averages of wind speed measured 22 km away is
not much less than the correlation with 20 s averaged local wind speed, sup-
porting the theoretical expectation that there is a substantial amount of spatial
averaging in the noise field due to the large area of sources contributing.
The sea surface noise measurements showed a poorer correlation with the
drag corrected geostrophic wind speed, with significantly greater spread of
data. This suggests that while wind forecasting will predict the broad trends,
it will provide poorer prediction of detail. Generally speaking, however, the
need is to forecast the broad trends such as the approach of fronts with a
substantial change in wind speed. It is evident that the drag corrected geos-
trophic winds generally followed the trends of the local wind, but the timings
of peaks and troughs in wind speed were displaced by a few hours, suggest-
ing that major effects could be forecast with some uncertainty about their
actual timing. Forecasting the fact that major changes in wind speed are due
may be more important than forecasting their timing accurately, since the
timing can be refined as the event approaches from other information (e.g.
local changes in wind speed, reports from weather stations).
Part of the spread in the data in plots of noise level versus drag corrected
geostrophic wind speed will be due to the correction for the drag of the sur-
face, which causes surface wind speeds to be lower than the geostrophic wind
speeds. The drag, and thus the correction, is usually greater over land than
over sea. We have used one correction factor (2/3) for all data, which is
typical for winds over the sea. It was generally not practicable to make dif-
ferent corrections for different data since winds were generally affected by
both land and sea. The result, however, is that prediction of surface winds
from geostrophic winds will be overestimates the more there is land influence
on the wind field. Since the relative influence of land and sea varied signifi-
cantly across the data, this would contribute to the spread of data.
Since the data points for the correlation of noise with drag corrected
geostrophic wind speed cluster about the points for the correlation of noise
with local wind speed, it suggests that the drag correction factor of 2/3 is
reasonable. It also seems that the relationship between noise and local wind
speed given in eqn (5) could be used to describe the correlation of noise with
geostrophic wind speed, with the appropriate change in the values of a and b.
The spread in the data reported here is, however, probably too large to make
a reliable estimate of these values.
Comparison of the results for the Spencer Gulf and Perth regions supports
the view that the presence of land complicates the wind field and degrades
Ambient sea noise 331

the accuracy of wind speed predictions. The dependence of noise on wind


speed measured at a distant weather station and on that estimated from the
geostrophic winds were both better off Perth, where the topography is sim-
pler and the wind field generally experiences the same topography upstream
of the measuring positions, especially since the winds are more often from
seaward. The Perth position was close enough to the shore to experience the
afternoon sea breeze. A number of the weather stations supplying the atmo-
spheric pressure readings from which the geostrophic winds were estimated
for the Perth site were along the coast and thus could be expected to show
effects of the sea breeze.
Sites further out to sea than the positions of measurement will be less
affected by the topographical complications, and in this respect the wind
speed predictions should be more accurate. This will depend, however, on
where the atmospheric pressures are measured and how accurately the con-
tours out to sea can be estimated.

ACKNOWLEDGEMENTS

We are particularly grateful to Brian Jones, Doug Bellgrove, Tony White


and Mark Savage for their efforts in the preparation and development of the
equipment and its deployment and recovery at sea. We thank the masters
and crew of the vessels NGERIN and TAMMAR for their efforts in the
work at sea. Thanks are also due to Dr Ian S.F. Jones and Paul Clarke for
help and advice on geostrophic winds. Meteorological data were obtained
from the Australian Weather Bureau, and Kevin Burrows and Roger Tapp
provided useful advice.

REFERENCES

1. Knudsen, V. O., Alford, R. S. and Emling, J. W., Underwater ambient noise.


Journal of Marine Research, 1948, 7, 41&429.
2. Wenz, G. M., Acoustic ambient noise in the ocean: spectra and sources. Journal
of the Acoustical Society of America, 1962, 34, 1936-1956.
3. Perrone, A. J., Deep ocean ambient noise spectra in the North West Atlantic
Ocean. Journal of the Acoustical Society of America, 1969, 34, 762-770.
4. Banner, M. L. and Cato, D. H. Physical mechanisms of noise generation by
breaking waves-a laboratory study. In Sea Surface Sound. Kluwer, Dordrecht,
1988, pp. 429-436.
5. Medwin, H. and Beaky, M. M., Bubble sources of the Knudsen sea noise spec-
tra. Journal of the Acoustical Society of America, 1989, 89, 11241130.
6. Pumphrey, H. C. and Ffowcs Williams, J. E., Bubbles as sources of ambient
noise. IEEE Journal of Oceanic Engineering, 1990, 15, 268-274.
338 D. H. Cato, S. Tavener

7. Updegraff, G. E. and Anderson, V. C., An instrument for the in situ measure-


ments of sea surface noise from a depth of 1 m, under low wind conditions.
Journal of the Acoustical Society of America, 1991, 89, 2253-2263.
8. Updegraff, G. E. and Anderson, V. C., Bubble noise and wavelet spills recorded
1 m below the ocean surface. Journal of the Acoustical Society of America, 199 I,
89, 22642279.
9. Kerman, B. R. (ed.), Sea Surface Sound. Kluwer, Dordrecht, 1988.
10. Kerman, B. R. (ed.), Natural Physical Sources of Underwater Sound: Seu Sur-
face Sound (2). Kluwer, Dordrecht, 1993.
11. Ferguson, B. G. and Wyllie, D. V., Comparison of observed and theoretical
responses of a horizontal line array to wind induced noise in the deep ocean.
Journal of the Acoustical Society of America, 1987, 82, 601-605.
12. Kuperman, W. A. and Ingenito, F., Spatial correlation of surface generated
noise in a stratified ocean. Journal of the Acoustical Society of America, 1980,67,
1988-1996.
13. Chapman, D. M. F., Surface generated noise in shallow water: a model. Pro-
ceedings of the Institute of Acoustics, 1987, 9, l-1 I.
14. Harrison, C. H., Formulas for ambient noise level and coherence. Journal qfthe
Acoustical Society of America, 1996, 99, 2055-2066.
15. Urick, R. J., Principles of Underwater Sound. 3rd edn., McGraw-Hill, New
York, 1983, pp. 227-23 1.
16. Cato, D. H., Tavener, S. and Jones, I. S. F. Ambient noise dependence on local
and regional wind speeds. In Sea Surface Sound ‘94. ed. M. J. Buckingham and
J. R. Potter, World Scientific Publishing, Singapore, 1995, pp. 95-l 11.
17. Van der Hoven, I., Power spectrum of horizontal wind speed in the frequency
range from 0.0007 to 900 cycles per hour. Journal of Meteorology, 1957, 14,
160-164.
18. Roll, H. U., Physics of the Marine Atmosphere, Academic Press, New York,
1965, pp. 216219.
19. Huschke, R. E., Glossary of Meteorology, American Meteorology Society,
Boston, 1959, p. 251.
20. Petterssen, S., Weather Analysis and Forecasting. Vol. 1, pp. 84, 85. McGraw-
Hill, New York, 1956.
21. Handbook of Meteorology, Australian Government Publishing Service, 1977,
p. 28.
22. Piggott, C. L., Ambient sea noise at low frequencies in shallow water of the
Scotian Shelf. Journal of the Acoustical Society of America, 1964, 36, 2 15222 163.
23. Cato, D. H., Ambient sea noise in waters near Australia. Journal of the Acous-
tical Society of America, 1976, 60, 320-328.

You might also like