Professional Documents
Culture Documents
www.elsevier.com/locate/apacoust
a,*
Received 6 February 2004; received in revised form 28 February 2005; accepted 1 March 2005
Available online 26 April 2005
Abstract
Environmental noise levels can vary over a wide range as a result of the diversity of site
conditions and activities occurring during eld measurements. This variability, extremely
important for a correct measurement interpretation, is often a source of disagreement when
applying standards and regulations, as there is no consensus about how to estimate and present it. The paper contributes to this technical debate by investigating the statistical variability
associated with a large measurement database acquired under eld conditions. The database
consists of 2 weeks noise recordings at each of 50 separate locations in residential areas
aected mainly by road trac noise. The results show increased variability (standard deviations) at the lower values of either logarithmic or arithmetic mean LAeq over the time periods
investigated. It is anticipated that the observed relationships may be of assistance when estimating the noise level variability and the uncertainty associated with a noise measurement
aected by road trac or other environmental noise sources.
2005 Elsevier Ltd. All rights reserved.
Keywords: Statistical variability; Uncertainties; Measurement; Trac noise; Standard deviation
Corresponding author. Tel.: +44 23 8059 4955; fax: +44 23 8059 2728.
E-mail addresses: ja@isvr.soton.ac.uk (J. Alberola), ian.indell@virgin.net (I.H. Flindell), andrewbullmore
@acoustics.hoarelea.com (A.J. Bullmore).
0003-682X/$ - see front matter 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.apacoust.2005.03.001
1181
1. Introduction
It is of course well known that road trac noise levels vary over time and space,
but there is no consensus as to how this variation should be taken into account. In
respect of variation over time, the recent Environmental Noise Directive [1] species
strategic noise mapping using the Lden, Lday, Levening, and Lnight noise level indicators which are to be determined over a relevant year as regards the emission of
sound and an average year as regards the meteorological circumstances. The Directive
is not specic as to how a relevant year and an average year should be understood
in these terms. Most people tend to interpret this as meaning that the indicators
should average out over the actual variation in emission of sound by the source
(or sources) over an entire year relevant for the assessment. Variation in acoustical
propagation from source to receiver arising under dierent meteorological conditions is slightly dierent, in that the intention seems to be that the data should be
adjusted or corrected to account for long-term average meteorological conditions
rather than the actual range of meteorological conditions occurring over the particular year relevant for assessment. While the intention is clear, that the indicators
should be representative of long-term average conditions dened in a particular
way, there is little or no acknowledgement in the Directive of the practical diculties
likely to ensue from attempting to follow the specication too rigorously. It is also
clear that any failure to apply the specication rigorously will lead to uncertainty
which could, on occasion, lead to inappropriate noise management decisions being
taken.
Regarding variation over space, the Environmental Noise Directive is quite
specic in dening the height above ground (4.0 0.2 m) for either calculations
or measurements, but is not at all specic about position in plan in relation to
potentially noise sensitive receivers. These uncertainties arise chiey because,
and notwithstanding that experienced acousticians normally have their own ideas
about how best to deal with the problem, there is no real consensus and there
very little information available in the published literature which is of much practical help.
This general concern has already been identied. Ten Wolde [2], as one of the
chief architects of the Environmental Noise Directive, has already suggested that
the estimation and management of the uncertainties involved in noise measurement
and prediction should be one of the priorities for future research. There is a clear risk
of inaccurate or erroneous interpretation leading to noise action plans focussing on
the wrong priorities, unfair applications of noise management restrictions, or unexpected adverse health consequences.
A number of authors have already made signicant contributions in this eld. Craven and Kerry [3], for example, recommends the uncertainty budget method for estimating uncertainties in environmental noise measurements. The separate
uncertainties associated with each of the variables aecting the measured noise level
are added together to derive a combined overall uncertainty. Because of limited time
and resources, each component of the overall uncertainty must normally be estimated
based on scientic judgement or practical experience rather than be determined from
1182
the results of a large set of repeated observations. Cravens approach follows the recommendations of the Guide to the expression of uncertainty in measurements
(GUM) [4], which states that uncertainty estimations could be obtained based either
on a professional judgement or on real data. Indeed other authors have also adopted
this methodology to estimate uncertainties in noise prediction methods (see, for
example, Manvell et al. [5]).
Notwithstanding the above, it is clear that uncertainty estimates should reect as
closely as possible what it is actually happening in real measurements. Professional
judgement is just that and could easily be wrong. Following this principle, other
authors such as Farrelly and Brambilla [6], have attempted to estimate actual uncertainties based on statistical analysis of real data by expanding what would otherwise
be considered as rather small data sets. However, there is no way that any method
based on assumed extrapolation can compete in terms of underlying accuracy
against estimates based on real long-term data.
For this paper, we have adopted a dierent approach where, based on a large
measurement database, we rst identify which input variables are most closely associated with (and hence predictive of) measured variability as described by standard
deviation, and then develop predictive models which could be used to estimate variability and hence uncertainties associated with future measurement or calculation
exercises. It is anticipated that this work might help to expand knowledge about
the range of uncertainties associated with environmental noise measurements and
thereby also help those acoustic professionals who base uncertainty estimates on scientic or expert judgement.
2. Measurement database
The analysis reported in this paper exploits a large measurement database which
was originally collected for planning and environmental impact assessment purposes.
The database comprises the results of continuous noise monitoring for 2 weeks at
each of 50 separate locations within a large area potentially aected by noise from
a projected new main road development. The detailed characteristics of the 50 separate locations are summarised in Table 1. The main source of environmental noise
existing at the time of measurement was typically road trac noise, but some locations were also aected by other noise sources as well. The measurement locations
were selected to be as generically representative as possible of a wide range of dierent types of suburban and rural residential areas at varying distances from existing
main roads and other noise sources, with the exception that there were no inner-city
areas included in the overall sample. The detailed selection of measurement locations
could not be fully randomised because it was necessary to ensure relatively even coverage throughout the projected route of the new main road and because of constraints imposed by potentially limited accessibility and landowner consents for
the 2 week measurement survey period at each monitoring location. Several authors
(see, for example, Craven and Kerry [3] or Gaja et al. [7]) agree that a 2 week
1183
1184
1185
80
Log. (Trend line)
R2 = 0.4156
75
33
16
70
11
42
65
13 2
6
8
60
26
37
55
22
36
25
35
18
45
43
50
20
45
40
0.5
1.5
2.5
(a)
3.5
4.5
5.5
75
70
33
16
11
42
65
13
2
6
60
26
37
55
22
36
25
50
18
35
43
45
20
45
40
0.5
(b)
1.5
2.5
3.5
4.5
5.5
Fig. 1. (a) Long-term average sound level (dBA) over 2 weeks at weekdays (06:0019:00) vs. standard
deviation of LAeq, 1 h values over the same period of time for each of the 50 sites. (b) Arithmetic mean of
LAeq, 1 h values over 2 weeks at weekdays (06:0019:00) vs. standard deviation of LAeq, 1 h values over the
same period of time for each of the 50 sites.
reason for looking at the arithmetic mean in addition to the logarithmic mean is
the theoretical possibility of a closer relationship to subjective response. For example, this may have been one of the reasons why the arithmetic mean was selected
1186
for some existing standards and regulations such as the current C.R.T.N. [8] applicable to the calculation of road trac noise in England. These regulations specify
that the 18 h arithmetic average of the separate hourly values of LA10 from 0600 to
2400 h should be used as the indicator of the amount of road trac noise present.
For most road trac noise, the hourly LA10 indicator reects the physical magnitude of the noise in a similar way to the hourly LAeq (except that absolute values
are typically 23 dB higher), and the calculation algorithms dened in the ocial
method [8] eectively treat the LA10 as if it was a true energy average anyway.
The theoretical possibility that the arithmetic mean could have a closer relationship
to subjective response is based simply on the well-known fact that humans are
more sensitive to proportional than to absolute or linear changes in physical
quantities.
This argument essentially depends on the fact that there is no consensus on the
averaging time typically applicable to human subjective response. The presumption
for annual (logarithmic) averages implicit in the Environmental Noise Directive implies that short-term peaks and dips in noise levels are only important for subjective
response to the extent that they aect the annual average. Most of the available evidence suggests the contrary, that individual listeners tend to be more concerned about
specic noisy events or noisy periods and are not therefore as concerned about the general level of continuous background noise. On the other hand, long-term averaging is
more convenient for regulatory and administrative purposes. In view of this debate, it
might still be appropriate to recommend alternative types of statistical averaging (such
as the arithmetic average of hourly or even 1 min LAeq) if there was some demonstrable benet such as more consistent statistical behaviour to be gained.
Figs. 1(a) and (b) show a clear tendency for the standard deviation of the hourly
values to increase as the overall average sound level decreases. There is no obvious
dierence between arithmetic and logarithmic averaging although the R2 coecient
indicates a marginally stronger relationship for the arithmetic mean. There is only
one measurement location with high standard deviation and high overall average
sound level (location No. 7) and there are various location specic factors which
can explain this outlier. This is a useful nding as it is consistent with the widely held
assumption that conventional road trac noise mapping becomes increasingly less
accurate at lower sound levels or at increasing distances from either modelled or
measured road trac noise sources.
Figs. 2(a) and (b) show the eect of looking at the 1 min LAeq values instead of the
1 h LAeq values as shown in Figs. 1(a) and (b). The overall pattern of the results is
very similar. It should be noted that the marginally higher R2 coecients are more
likely to be a consequence of the signicantly increased size of the database rather
than any greater strength in the underlying relationships.
Fig. 3 compares the trend lines for the logarithmic mean and the arithmetic means
of the 1 h and 1 min values against the standard deviations of the hourly values. The
three trend lines diverge at the lower overall average sound levels where the dierences between the three methods of averaging increase as the standard deviations
increase. The dierences between the arithmetic means of the 1 h values and the
arithmetic means of the 1 min values can be explained mathematically by the
1187
80
Log. (Trend line)
R2 = 0.4692
75
33
11
70
16
42
65
13
2
6
8
60
26
37
55
22
36
25
18
50
35
45
43
20
45
40
0.5
1.5
2.5
(a)
3.5
4.5
5.5
75
33
11
16
70
42
65
13
6
60
8
26
55
37
22
25
50
36
18
43
35
20
45
45
40
0.5
(b)
1.5
2.5
3.5
4.5
5.5
Fig. 2. (a) Long-term average sound level (dBA) over 2 weeks at weekdays (06:0019:00) vs. standard
deviation of LAeq, 1 min values over the same period of time for each of the 50 sites. (b) Arithmetic mean of
LAeq, 1 min values over 2 weeks at weekdays (06:0019:00) vs. standard deviation of LAeq, 1 min values over
the same period of time for each of the 50 sites.
1188
75
70
65
60
55
50
45
40
0.5
1.5
2.5
3.5
4.5
5.5
Fig. 3. Long-term average sound levels (dBA) and arithmetic means of LAeq, 1 h (dBA) and LAeq, 1 min
(dBA) vs. standard deviation of LAeq, 1 h values over 2 weeks at weekdays (06:0019:00) for each of the 50
sites.
Leq1 min; 1
Leq1 min; 2
; . . . ; 10
Leq1 min; 60
10
, correspond-
then working out both sides of the above expression results in:
10
Leq1 h
10
P 10
Leq
1 min
10
Fig. 3 provides good evidence that for this type of environmental noise, the differences between logarithmic and arithmetic averaging (of either 1 h or 1 min values)
are of no practical signicance except at the lower overall average sound levels where
the standard deviations are much greater.
Fig. 4 shows the eect of comparing the standard deviations of the 1 h and 1 min
values against the overall logarithmic means. As would be expected, the standard
deviation of the 1 min values always exceeds the standard deviation of the 1 h values,
but the gure shows that there is no tendency for the dierence to increase at the
lower sound levels with the higher standard deviations (measured by either method).
There are some measurement locations with much greater dierences between the
two standard deviations than the norm and the particular properties of these locations are addressed in Section 4 of this paper below.
1189
80
1-min standard deviation
1-hour standard deviation
75
33
11
70
42
65
13
2
6
8
60
26
37
55
45
18
50
20
45
40
0.5
1.5
2.5
3.5
4.5
5.5
Fig. 4. Long-term average sound levels (dBA) vs. both 1 h and 1 min standard deviations for each of the
50 sites.
The equivalent data from the most recent UK national noise incidence study (UK
NIS [9]) is shown in Figs. 5(a) and (b). Fig. 5(a) shows the data for all 1157 measurement locations included in the study and Fig. 2(b) shows the data for just those locations (323) coded as having road trac noise as the most dominant source present.
Each data point represents the arithmetic average of only a single 24 h worth of
hourly measurements and the corresponding standard deviation. The standard deviations therefore include the eects of daynight variation and do not take into account any variation from one day to the next. The gures show the same
tendency for the standard deviation of the hourly values to increase as the overall
average sound level decreases, although the strength of the relationship is much
weaker, as evidenced by the much lower R2 coecients. The R2 coecient is marginally higher for the data set for just those locations where road trac noise was considered to be the most dominant source. For the national survey, the overall
relationship between the logarithmic average and the standard deviation is very similar to that of the arithmetic average but the R2 coecient is slightly smaller. This is
similar to the ndings from the 50 location database.
1190
75
R2 = 0.2012
70
65
60
55
50
45
40
35
30
0
10
(a)
80
75
70
65
60
55
50
45
40
(b)
Fig. 5. (a) 13 h average sound level (dBA) from 06:00 to 19:00 vs. standard deviation of LAeq, 1 h values
over the same period of time for all UK NIS measurement locations [9]. (b) Descriptors 3 (urbanresidential), 4 (predominantly residential but with some light industry and main roads) and 5 (general
industrial area intermediate) of the UK NIS [9].
investigation of the 1 min LAeq sequence data for location No. 26, with the highest
dierence between the 1 min and 1 h standard deviations, suggested a previously
undetected instrument malfunction with successive 1 min values cycling over a 5
10 dB range for no obvious reason. Location No. 2 was approximately 30 m from
1191
200
180
160
140
120
100
80
60
40
49
47
45
43
41
39
37
35
33
31
29
27
25
23
21
19
17
15
13
11
20
Receivers
Fig. 6. Relative increments in percentage terms from LAeq, 1 h to LAeq, 1 min standard deviations for each of
the 50 receivers.
a commuter railway line. Detailed comparisons of the 20 s LAeq sequence data for
this location against the equivalent time histories for location Nos. 1 and 3 (at 45
and 70 m from the same railway line) showed that location No. 2 had been signicantly aected by railway noise, whereas location Nos. 1 and 3 had not.
Location Nos. 11, 14, 23, and 33 were all within 30 m of main roads carrying
intermittent trac ows, intermittent because of the relative proximity of the road
sections to junctions, and with no intervening structures or topography capable of
providing any signicant acoustic screening. It is interesting to note that for free
owing trac, whereas the standard deviation of instantaneous sound levels tends
to increase with increasing overall average sound level caused by reduced distance
from source to receiver, the standard deviation of hourly sound levels tends to reduce
with increasing overall average sound level. This is not necessarily the case for intermittent trac ows where successive stops and starts to the trac ow can increase
the standard deviation depending on the averaging time of the measurement relative
to the trac ow cycle time.
Variability due to dierences in meteorological conditions tends to increase at
increasing distances because of the dierent eects on acoustical propagation [10].
For typical suburban and rural locations aected by distant road trac noise,
changes in source to receiver vector wind speed prole above the ground can cause
small reductions in excess attenuation from some directions subject to increasingly
downwards sound ray curvature and correspondingly much bigger increases in excess attenuation from other directions subject to increasingly upwards sound ray curvature. These eects can become even more marked when interactions with
1192
structures and topography providing acoustical screening are taken into account.
For the measurement locations sitting lowest below the trend lines shown on Fig.
1(a) and (b) (i.e., locations 18, 20, 22, 36, and 43) the most likely explanations for
the otherwise low standard deviations are that increases in excess attenuation from
a distant road trac noise source in one direction tend to be compensated for by decreases in excess attenuation from distant road trac noise sources in opposing
directions. Measurement locations 68 tend to show opposite behaviour with relatively high standard deviation for the overall average sound level compared to the
general trend. These locations were all particularly sensitive to dierences in meteorological conditions because of their elevated position in respect of the most significant road trac noise source for those locations. Fig. 7 compares corresponding
hourly LAeq data for locations 8 and 9. Location 9 was eectively a close-in
(30 m) reference position location 8 which was elevated and 150 m away from the
same main road. The relationship between the two sets of data is much weaker
(i.e., more variability) for upwind conditions (small triangles) than for downwind
conditions (small circles).
Other noise sources present can also aect the strength of the relationship between
variability measured by standard deviation and overall average sound level. The signicant contribution made by railway noise at location No. 2 which has already been
discussed above could be identied from the detailed sound level time histories, but
identication also depends on detailed knowledge about individual site characteristics. For example, there was a continuous low level background noise source aecting location No. 25. This background noise would have made only a small
contribution during the day-time but would have set a noise oor which ambient
75
73
Downwind conditions
Upwind conditions
71
69
67
65
63
61
59
57
55
45
47
49
51
53
55
57
59
61
63
65
Fig. 7. Correlation between the LAeq, 1 h levels for site 9 (close to the dominant source) and site 8 (far from
source) under downwind and upwind conditions.
1193
noise levels could not have dropped below at night, even if the main road trac noise
sources in the vicinity had fallen completely silent. This factor may have contributed
to the lower than expected standard deviations observed at this location as compared
to the general trend. As another example, location Nos. 6, 7, and 37 were all found to
80
Raw data
Modified data
Log. (Trend Line)
2
R = 0.7015
75
70
7
65
6
7
8
60
6
8
26
37
37
55
50
45
40
0.5
1.5
2.5
(a)
3.5
1-min standard deviation(dBA)
4.5
5.5
80
Raw data
Modified data
Log. (Trend Line)
2
R = 0.7606
75
70
65
6
7
8
60
6
2
2
37
55
37
26
50
45
40
0.5
(b)
1.5
2.5
3.5
1-min standard deviation(dBA)
4.5
5.5
Fig. 8. (a) Logarithmic mean vs. standard deviation of LAeq, 1 min values with some modications in 6
receivers. (b) Arithmetic mean of LAeq, 1 min values vs. standard deviation of LAeq, 1 min values with some
modications in 6 receivers.
1194
5. Conclusions
Preliminary analysis of a large database of outdoor environmental noise measurements demonstrated a strong inverse relationship between measured standard deviations of hourly LAeq values and overall average noise levels measured in a variety of
dierent ways. More detailed analysis demonstrated marginally stronger inverse relationships between standard deviations and arithmetic mean noise levels as compared
to the more technically correct logarithmic or true mean noise levels. A similar, but
weaker, inverse relationship can also be observed in the most recent UK National
Noise Incidence Survey database.
By using the observed dierences between the standard deviations of the 1 h and
1 min LAeq values, it was then possible to identify and then isolate measurement locations not conforming to the general pattern for some specically identiable local reason. When the eects of (a) suspected equipment malfunction (location No. 26), (b)
unusual upwind propagation conditions (location Nos. 68), (c) signicant railway
noise contributions (location No. 2), and (d) school playground noise (location
1195
Nos. 6, 7, and 37) had all been deleted from the analysis the strength of the inverse
relationship was signicantly improved.
Subject to the caveat that extrapolation of these results to conditions not generically represented in the overall measurement database may be of dubious value,
the general nding of relevance to the estimation of uncertainty in noise measurements is that measured road trac noise variability increases from typical standard
deviations of 1 min LAeqs of around 1.5 dB at overall average noise levels of around
70 LAeq and above to around 4.5 dB at lower average noise levels of around 50 LAeq.
Any measured variability above the general trend is likely to be associated with signicant non-road trac noise sources, unusual propagation conditions, or otherwise
unsuspected equipment malfunctions. It is anticipated that these ranges of observed
variability may be of assistance when commenting on the range of uncertainty likely
to be associated with environmental noise measurements. Furthermore, these results
could be of value when, as explained in the GUM [4], a professional judgement is
used for estimating uncertainties associated with both predicted or measured noise
levels.
References
[1] European Commission, Environmental Noise Directive 2002/49/EC, Ocial Journal of the European
Communities L189; 2002. p. 1225.
[2] Wolde TT. The EU noise policy and its research needs. Revista de Acustica 2002;23(3&4):1520.
[3] Craven NJ, Kerry G. A good practice guide on the sources and magnitude of uncertainty arising in
the practical measurement of environmental noise. DTI Project: 2.2.1 national measurement system.
Programme for Acoustical Metrology.
[4] ISO Publication 6461-3:1995. General metrology part 3: guide to the expression of uncertainty in
measurement (GUM), BIPM/IEC/IFCC/ISO/IUPAC/IUPAP/OIML; 1995.
[5] Manvell D, et al. Matching noise maps with reality reducing error through validation and
calibration. In: The IOAs conference: noise mapping which way now? 2002.
[6] Farrelly FA, Brambilla G. Determination of uncertainty in environmental noise measurements by
bootstrap method. J Sound Vib 2003;268:16775.
[7] Gaja E, Gimenez A, Sancho S. Sampling techniques for the estimation of the annual equivalent noise
level under urban trac conditions. Appl Acoust 2003;64(1):4353.
[8] Calculation of road trac noise (C.R.T.N.). Department of Transport, UK.
[9] Skinner CJ, Grimwood CJ. The national noise incidence study 2000/2001, UK: volume 1 noise
levels, BRE client report number 206344f. Public on DEFRA.
[10] Alberola J, Bullmore AJ, Flindell IH, Peral R. Variability of outdoor sound propagation arising
under dierent meteorological conditions. In: 11th International congress on sound and vibration, St.
Petersburg; 2004.