550/9-77-355
Comparison of
Highway Noise Prediction Models
May 1977
Office of Noise Abatement and Control
U.S. Environmental Protection Agency
Washington, D.C. 20460
-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before completing)
1. REPORT NO.
550/9-77-355
3. RECIPIENT'S ACCESSION-NO,
4. TITLE AND SUBTITLE
Comparison of Highway Noise Prediction Models
5. REPORT DATE
May 1977
6. PERFORMING ORGANIZATION CODE
7. AUTHOR(S)
Kenneth J. Plotkin, R. G. Kunicki
8. PERFORMING ORGANIZATION REPORT NO.
WR 76-25
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Wyle Laboratories/Wyle Research
2361 Jefferson Davis Highway, Suite 404
Arlington, Virginia 22202
10. PROGRAM ELEMENT NO.
11. CONTRACT/GRANT NO.
68-01-3514
12. SPONSORING AGENCY NAME AND ADDRESS
U.S. Environmental Protection Agency
Office of Noise Abatement and Control
1921 Jefferson Davis Highway
Arlington. Virginia 22202
13. TYPE OF REPORT AND PERIOD COVERED
Final
14. SPONSORING AGENCY CODE
15. SUPPLEMENTARY NOTES
16. ABSTRACT
A review and comparison has been conducted of three highway noise prediction models:
NCHRP, TSC, and Wyle. The first two are those approved by the Federal Highway Administration;
the third was developed for EPA. The currents comprising each model are analyzed in detail,
including basic formulation, vehicle noise levels, propagation, treatment of various road
geometries, and shielding by barriers. Significant differences among the models were found.
A series of charts is presented whereby differences among the models may be estimated for particular
input data. Comparison between measured roadside levels and predictions from the three models are
also presented.
17.
KEY WORDS AND DOCUMENT ANALYS'S
DESCRIPTORS
b.lDENTIFIERS/OPEN ENDED TERMS C. COSATI Field/GlOUp
Highway noise
Environmental noise
Prediction models
18. DISTRIBUTION STATEMENT
Unlimited
19. SECURITY CLASS (This Report)
Unclassified
21. NO. OF PAGES
44
20. SECURITY CLASS (This page)
Unclassified
22. PRICE
EPA Form 2220-1 (9-73)
-------
550/9-77-355
COMPARISON OF
HIGHWAY NOISE PREDICTION MODELS
MAY 1977
CONTRACT 68-01-3514
U.S. Environmental Protection Agency
Office of Noise Abatement and Control
Arlington, Virginia 22202
This report has been approved for general availability. The contents of this
report reflect the views of the contractor, who is responsible for the facts and
the accuracy of the data presented herein, and do not necessarily reflect the
official views or policy of EPA. This report does not constitute a standard,
specification, or regulation.
-------
TABLE OF CONTENTS
Page
1.0 INTRODUCTION 1
2.0 MODEL FORMULATION 2
2.1 Basic Formulation 2
2.2 Vehicle Noise Levels 3
2.3 Propagation 9
2.4 Multiple-Lane Geometry 11
2.5 Barriers 13
2.6 Finite Road Elements 15
2.7 Finite Barriers 19
2.8 Miscellaneous Adjustments 21
2.9 Combination of Sources 21
2.10 Conversion Between Models 23
3.0 COMPARISON OF MEASUREMENTS AND PREDICTIONS 25
4.0 CONCLUSIONS 37
REFERENCES 39
- in -
-------
TABLE OF FIGURES
Figure Page
1 Difference Between L,Q and L5Q 5
2 Comparison of Wyle and NCHRP Vehicle Flow Dependence.
TSC Agrees With Wyle Except For Vehicle Levels 5
3 Automobile Maximum Pass-By Levels at 50 Feet 7
4 Truck Maximum Pass-By Levels at 50 Feet 7
5 Differences Between Predictions Due to Vehicle Level Differences 8
6 Comparison of Propagation Relations Used in the Three Models. . 10
7 Comparison of Road Width Adjustment 12
8 Comparison of Barrier Attenuation Curves 14
9 Barrier Insertion Loss, 5-Foot-High Barrier 16
10 Barrier Insertion Loss, 10-Foot-High Barrier 17
11 Barrier Insertion Loss, 15-Foot-High Barrier 18
12 Adjustment for Finite Road Elements 20
13 Site 9: Roadside Barrier & Earthmound: 6-Lane Freeway. ... 27
14 Site 10: Elevated Highway Configuration: 8-Lane Freeway. . . 28
15 Site 11: Elevated Highway Configuration: 8-Lane Freeway. . . 29
16 Site 12: Roadside Structures: 4-Lane Freeway 30
17 Site 13: Depressed Highway Configuration: 8-Lane Freeway . . 31
18 Comparison of Differences Between Measured and Predicted L . 35
eq
TABLE OF TABLES
Table Page
I Miscellaneous Adjustments 22
II Geometry of Sites 1-8 26
HI Comparison of Measurements With Predictions 32
IV Confidence Limits and Standard Deviations 34
- iv -
-------
1.0 INTRODUCTION
There are presently three major highway noise models In use in the United States.
Two of these, known as NCHRP' and TSC, are approved by the Federal Highway
Administration for predicting levels as required for noise analysis of proposed federally
4 56
funded highway projects. The third model, ' developed by Wyle Research under
contract to the Environmental Protection Agency, is a more recent development, which
takes a simpler point of view than the earlier two.
A major difficulty with these models is that quite often they do not predict the
same levels for identical situations. Even the two FHWA-approved models differ sub-
stantially in some cases. This casts serious doubt on the accuracy of the models, and
can also lead to situations where a model is selected for use because it provides favor-
able results.
The present study has been undertaken to examine the discrepancies among the
three models, and to assess their accuracy. There are two parts to this comparison.
Section 2.0 compares the formulation and basic data for the three models. All three
models purport to sum the noise contributions of individual vehicles over a traffic stream.
It is straightforward to compare the formulation, vehicle levels, propagation adjustments,
etc. In this way a "comparison map" for the three models can be developed, where the
differences, due to various assumptions and/or data elements, can be assessed over a
range of conditions.
Section 3.0 compares predictions with measurements. Published data from well-
documented measurements have been collected, and have been compared with predictions
from all three models. Comparisons are analyzed with respect to both the data and the
results of Section 2.0. Section 4.0 presents conclusions of this review.
-------
2.0 MODEL FORMULATION
2.1 Basic Formulation
All three models begin with simple straight-line elements for a stream of traffic.
The Wyle and TSC models use a formulation based upon the equivalent noise level L ,
and are in principle "exact", making no assumptions as to vehicle spacing and level
variations. The NCHRP model is based on a theoretical calculation of LC« (the level
exceeded 50 percent of the time) for a lane of equally spaced identical vehicles. The
basic single-lane elements of the three models are summarized below.
Wyie. In the absence of propagation losses,
L^L^HMo^—
where Q = Vehicle flow, number per unit time
d = Distance from observer to road
dQ = Reference distance for vehicle pass-by levels (usually 50 ft (15 meters))
V = Vehicle speed
L " = Equivalent vehicle level (see Reference 5) — the intensity mean of the
noise distribution for the vehicle population.
TSC. The basic formulation is identical to Wyle, except that vehicle levels are described
in terms of average pass-by level and standard deviation, rather than LCq. This gives
2 Ard2 Q\
L -C + 0.115 / + 10login _2-- (2)
where L = Average vehicle level
"• = Standard deviation of vehicle levels
The TSC model assumes the distribution of vehicle levels to be Gaussian. This is less
general than the use of L®q in the Wyle model, but is of little practical importance when
the models are used as design tools. The relation between the vehicle level definitions
can be seen by comparing Equations (1) and (2).
-2-
-------
g
NCHRP. This is based on Johnson and Sounders' expression for L-n:
L50 =40log10V-10log1()d + 10log1()p+ 10 \O9}Q [tanh (1.19x lO^pd)] + K (3)
where p = Vehicles per unit distance = Q/V
K = Constant, empirically determined
In Equation (3), p has units of vehicles per mile and d is in feet. (Note that Equations
(1) and (2) are valid for any consistent set of units.) Metrics other than L^ are obtained
by assuming a Gaussian temporal distribution of noise, with standard deviation based on
empirical data. Figure 1 shows the adjustment to L5Q to obtain LQ. The measurements
shown in Figure 1 by Johnson and Sounders were performed in England. Also shown are
7 9
typical data measured in the United States. ' The difference in vehicle types in the two
countries apparently does not affect the statistics of variations. Note that the theoretical
curve in Figure 1, obtained from the same model used to develop Equation (3), differs
substantially from the measurements.
Figure 2 shows a comparison of the Wyle and NCHRP predictions of L as a
eq
function of vehicle volume. At large vehicle volumes, both models have the same slope.
(For vehicle spacing less than 4d, the log tanh term in Equation (3) becomes approximately
zero, so that the traffic volume dependence is the same as Equations (1) and (2).) Shown
as dashed lines are linear extrapolations of the NCHRP prediction from its high value
asymptote, with volume dependence the same as for the L models. The difference in
level at high volume is due to differences in vehicle levels (discussed below). The addi-
tional difference at low volume is due to the inadequacy of the Johnson and Sounders
formulation for real traffic flow. Predictions from the TSC model (not shown) differ from
the Wyle model by amounts corresponding to differences in source noise level.
2.2 Vehicle Noise Levels
The Wyle and TSC models have vehicle levels as direct inputs in the form of
50-foot (15-meter) maximum pass-by levels. The NCHRP mode I is not explicitly based on
individual vehicle levels, but rather fits a constant to highway noise data. It is possible,
however, to compare the final NCHRP expression to the final Wyle expression and thereby
extract an equivalent vehicle level from the NCHRP model.
-3-
-------
12
10
CD
TJ
102
NCHRP
1
\ \NN
Johnson and Sounders Theory
Johnson and Sounders Data
BBN Simulation Model10
Wyle Data7
8
1
103 104
Qd/V, feet-vehicles/mile
105
106
Figure 1. Difference Between Liriand L-rt
IU 50
-------
80
70
60
§50
40
30
Wyl
70mph
•20mph
Wyle Model With NCHRP Vehicle Levels
ifcr
1000
Vehicles Per Hour
10,000
Figure 2. Comparison of Wyle and NCHRP Vehicle Flow Dependence.
TSC Agrees With Wyle Except For Vehicle Levels.
.5-
-------
Automobile and truck levels for the three models, expressed as L , are shown
in Figures 3 and 4. Analytic expressions describing the curves are shown on the figures.
The origins of vehicle levels are:
9
Wyie. Automobile levels are based on Olson's roadside measurements. Truck levels
are based on roadside measurements of over seven thousand trucks at high and low speeds
in eight states.
TSC. Automobile levels are based on Olson's data. The discrepancy between TSC and
Wyle levels is apparently due to interpretation and fitting of an analytic curve. The
9
truck noise levels are based on Olson's roadside measurements of 466 trucks, predominantly
3 10
at high speeds, and on measurements of truck noise on the New Jersey Turnpike. ' The
speed dependence of tire noise is specifically neglected.
NCHRP. Automobile levels are based on fitting Equation (3) to Johnson and Sounders'
roadside measurements of traffic noise in England. Truck levels are based on limited road-
side measurements which indicate that trucks at 50 mph are 15 d& greater than automobiles.
The automobile levels are in reasonable agreement over most of the speed range,
except for the NCHRP levels being consistently high. The speed dependency is in reason-
able agreement with theoretical considerations of tire noise. The truck levels are in serious
disagreement, especially with regard to speed dependence. In view of the larger data base
for the Wyle truck noise relation, the TSC and NCHRP truck noise levels are considered to
be unreliable. The influence of different vehicle levels on differences between the three
models is shown in Figure 5. This is a set of charts of contours of differences, as a function
of speed and truck percentage. At typical highway speeds of 55 mph, agreement between
the models is reasonably good. At low speeds, where the truck levels disagree, discrep-
ancies can be 4 dB or more. In some cases, there are sharp changes of difference at small
truck percentages when the dominant noise source changes between cars and trucks.
Differences for the case of no trucks may be obtained from Figure 3.
-6-
-------
£0
-o 75
X
CO
10
8 70
O-
< 65
E
1
1
60
AUTOMOBILES
NCHRP
72.4 +30log10(V/55mph)
TSC
71.7 +38log1()(V/55mph)
'WYLE
71.4 +32.1 log1()(V/55mph)
55
I
20
30
50
60
40
Speed, mph
Figure 3. Automobile Maximum Pass-By Levels at 50 Feet
70
4)
1
X
CO
8
f
E
.1
'x
90
80
TRUCKS
TSC
88.3
NCHRP 86.2
Wyle 83.6
87.4 + 20 log1()(V/55 mph)
20
30
50
40
Speed, mph
Figure 4. Truck Maximum Pass-By Levels at 50 Feet
-7 -
60
70
-------
*
20
Figure 5. Differences Between Predictions Due to Vehicle Level Differences
-8-
-------
2.3 Propagation
In the absence of propagation losses, noise from a straight line source decays as
10 log,nd, or 3 dB per doubling of distance. Air absorption is always present, and can
12
be calculated from standard tables. Air absorption loss for highway noise is typically
about 2 dB/1000 ft., so is often negligible. For propagation near the ground, finite
ground impedance leads to additional attenuation. Theoretical computation of ground
1 *? 1 A.
absorption is quite complex, ' and requires ground impedance data not readily avail-
able. All three models therefore use simplified, empirical representations.
Wyle. Ground absorption is represented in terms of excess dB per doubling, a form sug-
~"~"™~~~~ 1 "\ \A.
gested by the geometrical nature of theoretical calculations. ' Measurements by
Nelson on individual vehicles show that the excess can be from 0 to 4 dB per doubling
of distance, depending on ground surface. Highway noise measurements indicate 1.2 to
1.5 dB per doubling excess attenuation for typical clear terrain situations. The Wyle
5 6
computer model permits the user to specify a value. The Wyle nomogram method uses
a conservative value of 1 dB per doubling, plus 2 dB per 1000 feet air absorption. This
propagation relation is shown in Figure 6.
TSC. Over bare ground, no ground attenuation is used. For propagation over thick grass
and shrubs and through trees, excess attenuation is based on data collected in Reference 16.
These data provide frequency dependent coefficients of absorption per unit distance.
Figure 6 shows the TSC propagation relation over clear terrain (no excess attenuation) and over
the two types of ground cover. Attenuations for automobiles and trucks differ because
of spectral differences. Shown is a worst case example of ground cover extending up to
the highway; the TSC model permits specifying limited patches of ground cover. The value
of ground absorption depends to some degree on how the shape of the area is described to
the program. Reference 16 specifies that ground absorption should not be used within
200 feet of the source. It is not clear whether the TSC program checks for this, or whether
this is the responsibility of the program user.
The ground absorption model used in the TSC model is not considered to be satis-
factory. There is some question as to the validity of the data. The frequency dependence
n \A.
is opposite that predicted by recent we 11-supported theory. ' The grass considered is
-9-
-------
o
I
CO
-o
0)
0)
o
"3
-10
£ -15
-20
-25
50
TSC Clear Terrain
100 200
Distance, F«*t
300 400 500
Figure 6. Comparison of Propagation Relations Used in the Three Models
-------
also very high and dense, and is not representative of lawn-type grass usually used in
highway landscaping. The dense shrubbery considered is also not applicable to sparse
scrub often found in the southwestern United States. Caution should therefore be used
before using the TSC ground cover correction.
Reference 3 describes values of air absorption which are presently obsolete. It
appears, however, that the current version of the TSC program contains data consistent
with that of Reference 12.
NCHRP. Based on empirical data, all attenuation over clear terrain is represented by
1.5 dB per doubling of distance excess attenuation. The NCHRP propagation relation is
shown in Figure 6. Agreement between this and the Wyle model is very good.
Although the Wyle and NCHRP propagation relations show better agreement with data,
they must be taken with some reservations. Propagation does vary with ground cover, so
that usingasingle relation cannotbe adequate for all cases. The computerized version of
the Wyle model permits the user to specify a value of propagation loss, but without local
measurements it is difficult to determine what that value should be. Even with this poten-
tial flexibility, this version of the Wyle model (or a similar modification of the others)
cannot compensate for variations in vehicle levels at 50 feet observed over various paved
and unpaved surfaces. The difference between a paved and unpaved site is on the order
of 2 dB. This added uncertainty must be considered to exist for all models. More basic
work must be done to establish correct highway noise propagation characteristics.
Figure 6 may be used to assess differences between the models due to propagation.
2.4 Multiple-Lane Geometry
The elements discussed so far are for a single-lane road. Multiple-lane roads are
treated as follows:
Wyle. The computer version accounts for actual lane geometry, and combines the con-
tribution of all lanes. The process properly accounts for the traffic distribution among
the lanes, number of lanes, median width, and propagation effects. A nomogram version
of the Wyle method uses an approximate adjustment, based on an average fit to the com-
puter calculation for various numbers of lanes. As shown in Figure 7, the adjustment
-11 -
-------
Wyle, TSC
NCHRP
NCHRP With Wyle Propagation
VV
D
Figure 7. Comparison of Road Width Adjustment
-12-
-------
depends only on road width, median width, and observer distance. For W/D = 5,
accuracy is within + .5 dB of an exact calculation over extremes of traffic distributions
and number of lanes; accuracy is proportionately better at smaller W/D.
TSC. Actual lane geometry is used, just as in the Wyle model. Lane adjustment is thus
similar to the Wyle adjustment shown in Figure 7.
NCHRP. Noise levels are computed on the basis of a single equivalent lane located at
where d and dc are observer distances to the centers of the near and far lanes,
n t
respectively. There is no theoretical basis for this formula. Figure 7 shows the
adjustment based on this equivalent distance. Shown are the adjustment based on
1 .5 dB/doubling excess attenuation (as used in the NCHRP model) and based on
1 dB/doubling excess attenuation, for a fairer comparison with the Wyle adjustment.
Agreement with the multi-lane zero median case (M/W = 0) is reasonable, but
there are significant differences for wide median strips.
2.5 Barriers
Calculation of barrier shielding in all three models is based on Fresnel diffraction
17 18
as represented by Maekawa. The Wyle and NCHRP models use a line source adaptation,
while TSC uses a numerical application of Maekawa 's original point source curve. Point
and line source shielding curves are shown in Figure 8 as a function of path length dif-
ference, 6 (infect), for a 500 Hz tone. This is very close to the curve for typical A-
weighted vehicle spectra. The three models utilize different assumptions of vehicle source
height and other details in applying these shielding curves.
Wyle. Automobile source height is assumed to be two feet (0.6 meter), and truck source
height is assumed to be eight feet (2.4 meters). The line source curve shown in Figure 8
is used. When barriers are present, ground attenuation is neglected.
TSC. Automobile source height is at pavement level, and truck source height is eight
feet (2.4 meters). The point source curve shown in Figure 8 is used, with a maximum of
-13-
-------
50
40
CO
TO
j 30
t5
I
T
1
Source
d —'—^- Observer
6= A + B - d
Point Source,
Maekawa
—— Line Source,
Kurze & Anderson
NCHRP
Path Length Difference S, feet
Figure 8. Gxnparison of Barrier Attenuation Curves
-------
24 dB attenuation at large 6. The roadway and barrier are subdivided by the program
into short elements, where the point source model is reasonable to apply, and the
shielded contributions combined. This provides a numerical approximation of the line
source shielding result. It was found that in cases of large shielding, computational
time could become excessively long due to the program subdividing the road into many
elements. When barriers are present, ground attenuation is neglected.
NCHRP. Automobile source height is at pavement level. Truck shielding is assumed to
be 3 dB less than automobile shielding, with a minimum of zero. There does not appear
to be any substantive support for this assumption. The line source curve in Figure 8 is
used with two modifications: maximum shielding at large 6 is assumed to be 15 dB, and
shielding is assumed to be zero for negative 6 (i.e., line-of-sight exists between source
and observer). These two modifications appear to be intuitive. When barriers are present,
ground attenuation is still included.
Figures 9, 10 and 11 compare shielding for various single lane cases for the three
models, for automobiles and for trucks. The values shown are insertion losses, i.e., the
level computed with the barrier minus the level computed for the same distance and
traffic flow over clear terrain without a barrier. Shielding for multiple lane cases will
lie somewhere between shielding based on the location of the nearest and farthest lanes.
The large differences between the shielding calculations for the three prediction
methods demonstrate the sensitivity of shielding to the assumptions of lane location and
source height. Since there is little question over the correctness of the basic diffraction
calculation, more work must be done to establish correct effective source heights for auto-
mobiles and trucks, and to assess the effect of the ground when a barrier is present.
2.6 Finite Road Elements
The model elements described thus far are straight infinite line sources. These
are obtained by integrating vehicle pass-bys from plus and minus infinity along the road.
If a road has finite length, the integration may be carried out over this length only. All
three models use essentially this method for finite road lengths; curved roads are approx-
imated by a series of straight elements. Differences between the models arise, however,
due to differences in handling propagation losses.
- 15-
-------
w
""o
-10
CO
"O
§ -15
o
I
-10
-15
•-T
Barrier 101 from Road
W
Barrier 50" from Road
I
25 50 100 200 400
Barrier to Observer Distance, ft.
-5
-10
-15
0
-5
-10
-15
Barrier 25' from Road
Barrier 100' from Road
I
I
25 50 100 200 400
Barrier to Observer Distance, ft.
Figure 9. Barrier Insertion Loss, 5-Foot-High Barrier
W = Wyle T = TSC N = NCHRP
-------
Or-
-5
-10
CQ
-o
o
t -15
Barrier 10* from Road
I N
o,-
-5
-10
-15
Barrier 25' from Road
o
CO
-5
-10
-15
Barrier 50' from Road
I
I
I
25 50 100 200 400
Barrier to Observer Distance, ft.
Or-
-5
-10
-15
•W
Barrier 100' from Road
I
I
I
25
50 100 200 400
Barrier to Observer Distance, ft.
Figure 10. Barrier Insertion Loss, 10-Foot-High Barrier
W = Wyle T = TSC N = NCHRP
-------
Or-
-5
-10
«£>
8
O
c
O
••*•
-15
•J5 0
OOgJ
«*E
O
CO
-5
-10
-15
Barrier 25' from Road
Barrier 100' from Road
•W
N
25 50 100 200 400
Bvrrier to Observer Distance, ft.
-5
-10
-15
Oi-
-5
-10
-15
Barrier 50' from Road
•W
Barrier 200' from Road
N
25 50 100 200 400
Barrier to Observer Distance, ft.
Figure 11. Barrier Insertion Loss, 15-Foot-High Barrier
W = Wyle T = TSC N = NCHRP
-------
Wyle. Figure 12 shows the finite length adjustment used in the Wyle model. It
represents the difference between a symmetric finite road within an included angle
+ 6 and an infinite road. By reason of symmetry, the figure also represents the dif-
ference between a semi-infinite road and one from 0 to 6. It is possible by a series
of steps to obtain from this figure the adjustment for shielding of a central portion of
a road; instructions are given in Reference 6. Figure 12 was obtained by integrating
a finite line source with 1 dB/doubling excess attenuation.
TSC. The TSC model considers the adjustment for a finite road element to be:
AL = 10log1() 6/180° (5)
where 6 is the total included angle between observer and road element ends. This result
is obtained by integrating a finite line source with no excess attenuation. The use of
Equation (5) tends to overestimate the noise contribution of distant road elements. The
program does compute propagation losses individually for each road element (based on
the nearest distance for each) so that the results are not always exactly as Equation (5)
would give. This depends, however, on exactly how the road is defined in terms of
separate elements.
NCHRP. The finite element adjustment is based on Equation (5).
Figure 12 also compares Equation (5) with the Wyle adjustment for finite length
roads. Note that the Wyle adjustment becomes negligibly small at smaller angles than
Equation (5); since the length of the corresponding road element is proportional to tan 8,
much smaller road lengths are involved. Only a relatively short segment of road need be
accounted for in detail to obtain accurate predictions. This is an important result, and
leads directly to the much simpler structure of the Wyle model as compared to the others.
2.7 Finite Barriers
Cases of finite barriers are divided into shielded and unshielded elements, and
the finite road adjustment (discussed above) used to compute the noise contribution for
each. Due to differences noted above, the TSC and NCHRP models tend to underestimate
the effectiveness of a finite barrier, relative to the Wyle model. For the TSC model,
this depends to some degree on the specified road geometry.
-19-
-------
20 30 40 50 60
Road half-angle, degrees
80 90
Figure 12. Adjustment for Finite Road Elements
-20-
-------
2.8 Miscellaneous Adjustments
Table I lists various other adjustments considered in the three models. Most of
these are based on limited data, so it is not possible to offer more than general comments
on their accuracy. Subjective judgments are required for application of the road surface
and dense tree correction in NCHRP. The NCHRP roadside structure adjustment is intended
for detached houses; connected row houses could be treated as a barrier, but no quanti-
tative criteria is given for when to do so. The Wyle and TSC roadside structure approaches
are probably inadequate for detached houses.
2.9 Combination of Sources
All three models combine noise from separate sources by decibel addition. For
L this is the correct approach and is, in fact, one of the major computational advantages
eq
of L . Combinations of levels expressed in terms of the statistical metrics requires knowl-
edge of the full temporal distributions. The combination of separate sources even with
18
Gaussian distributions is far from trivial. Nelson provides a simplified procedure for
combining such distributions.
The TSC model includes a mathematically sophisticated approach to obtain statis-
tical metrics. Quantities called "cumulants" are computed for each source element, based
19
on a theoretical model of randomly spaced equal vehicles. Cumulants have the property
of being additive, and are related to the distribution. The cumulants for separate sources
are simply added to give the cumulants for the total noise; the temporal distribution of
the total noise may then be obtained. In this way the total L can be converted to other
eq
metrics. The approach has the disadvantage of being somewhat complicated. Also, a
number of physically questionable assumptions are made to make the calculation tractable.
These assumptions are also included in the calculation for a single road. Since a Gaussian
20
distribution is usually adequate to relate Lfi to LjQ/ the TSC approach offers little
practical benefit, and may even be a disadvantage. This is especially true when the
assumptions in the cumulant analysis introduce errors larger than present when empirical
data such as Johnson and Sounders relation for a are used to fit a Gaussian distribution.
In Reference 20 it is shown that measured values of LJQ - L are generally within a few
tenths of a dB of that predicted from Johnson and Sounders relation.
-21 -
-------
TABLE I
Miscellaneous Adjustment's
WYLE
NCHRP
TSC
STOP-GO
TRAFFIC
+2 to 4 dB*
Up to +4 dB
GRADES
Up to +2 dB
(Trucks Only)
Up to +5 dB
(Trucks Only)
ROAD
SURFACE
+5 dB for Very Rough
or Very Smooth
ROADSIDE
STRUCTURES
Treat as Barriers
-4.5 dB for First Row
of Houses,
-1.5 dB for Each
Additional, Up to
-lOdB Maximum
Treat as Barriers
DENSE TREES
No Correction
-5dB/100Ft.,
Up to -lOdB Maximum
Included in Propagation
Corrections
REFLECTIONS
Up to 3 dB per Reflecting
Surface
* Preliminary. See Reference 25, Appendix A
-22-
-------
The NCHRP model makes no attempt to correctly combine statistical levels. In
fact, for a highway with automobile and truck traffic, L« is computed separately for
automobiles and trucks, and the two combined by decibel addition. This step is mathe-
matically incorrect. It is only because highway noise distributions are similar over a wide
range of conditions, and because of empirical corrections built into the NCHRP model,
that the final results are often correct. Even if it is argued that Lfl is a more accurate
measure of annoyance than L , the benefit is clearly negated by the gross oversimpli-
eq
fication in the NCHRP method of obtaining LQ. (It should be recalled that the basic
NCHRP expression for L-Q, which is subsequently corrected to LQ, is based on an over-
simplified model for traffic which assumes uniform spacing of identical vehicles.)
2.10 Conversion Between Models
It may sometimes be necessary to convert predictions obtained from one model
into those which would have been obtained with another. This can be to determine how
great a difference exists in a specific case, or it may be necessary to check predictions
originally made with a different model. This latter case often occurs in the review of
environmental impact statements, where the preparing agency and the reviewing agency
may use different models.
Differences between the various curves in Figures 5 through 7 and 9 through 12
provide a basis for an approximate conversion from one model to the other. The procedure is:
1. From Figure 5, obtain the difference due to vehicle noise levels. This accounts
for different speed and traffic mix behavior of the three models.
2. From Figure 6, obtain the difference due to propagation. Note that only
limiting cases are shown here for the TSC model. For cases of partial ground
cover, the TSC propagation loss must be either interpolated or estimated from
the discussion in Section 2.3.
3. From Figure 7, obtain the difference due to different road width corrections.
4. Estimate barrier differences using Figures 9, 10 and 11. The procedure is to
take the height closest to the actual barrier height, then find the shielding
-23-
-------
differences corresponding to the nearest lane and the farthest lane. Some
interpolation will usually be necessary. The difference is the average of
these two differences.
5. For unshielded finite roads, obtain the finite road difference from Figure 12,
In each of the above steps, the sign of the difference should be clear from the
figure and previous discussion. The final conversion is the sum of all differences.
-24-
-------
3.0 COMPARISON OF MEASUREMENTS AND PREDICTIONS
The three models have been used to predict levels for thirteen we 11-documented
measurement sites. Measurements at eight sites, numbered 1 through 8, were conducted
by W/le Research. All of these sites are in Southern California, in the area between
Los Angeles and San Diego. These sites were straight level and unobstructed freeways
and arterials. Road geometry and microphone positions are given in Table II. Five sites,
numbered 9 through 13, are from Reference 2 and represent various elevated, depressed,
and shielded configurations. Geometries are given in Figures 13 through 17. These
thirteen sites include eighty separate measurements at various traffic flow, mix, and
receiver distances.
Predictions were made from the three models. The TSC predictions were made
from the computer program presented in Reference 3. The NCHRP predictions were made
21 22
from the batch version of the Michigan computer version of the NCHRP model. '
This program purportedly is equivalent to the method of References 1 and 2, but appar-
ently contains some undocumented differences. NCHRP predictions for Sites 1-8 were
also made with the charts in References 1 and 2. The Wyle predictions were made using
the nomogram method of the Wyle model.
Table III shows the measured and predicted sound levels. Differences between
predictions for the three models are in general consistent with the differences noted in
Section 2.0. General trends are difficult to identify because all factors are combined.
In order to formally assess the accuracy of the three models, statistical comparisons between
measured and predicted values have been performed. Table IV shows the 90 percent
confidence intervals for the differences between predictions and measurements of L .
eq
A positive value indicates a predicted level greater than measured. Also shown is the
standard deviation, a, for each model. The confidence limits are illustrated in Figure 18.
The Wyle model has the best agreement in terms of the standard deviation and the
90 percent confidence interval about the mean. The corresponding quantities for the
other two models are not significantly worse, however. It would not be reasonable on
the basis of standard deviation and confidence interval to select one model as being
better than the others. The NCHRP model has almost zero average deviation, while
-25-
-------
TABLE II.
Geometry of Sites 1 -8
i
to
SITE
No. of Lanes
Highway Type
Median Width
Outer Lanes
<£ to <£ Distance
Inner Lanes
<£ to
-------
Figure 13. Site 9: Roadside Barrier & Earthmound
6-Lane Freeway
-------
Figure 14. Site 10: Elevated Highway Configuration
8-Lane Freeway
-------
Figure 15. Site 11: Elevated Highway Configuration
8-Lane Freeway
-------
-24-*
****¥'J*>-'
U- 48
aKHBfiK
72
I
188
194
• 35-*|
140
334
480
146
Figure 16. Site 12: Roadside Structures
4-Lane Freeway
-------
Figure 17. Site 13: Depressed Highway Configuration
8-Lane Freeway
-------
"TABLE III.
Comparison of Measurements With Predictions
SITE
1
2
3
4
5
6
7
8
9A
B
C
D
E
F
G
H
HWY
TYPE
6A
6A
8F
2A
8F
2A
2A
4A
6F
TRAFFIC
FLOW
(VEH/HR)
1,557
1,200
5,552
275
5,145
228
948
529
588
496
486
456
348
492
618
521
%
TRUCKS
4%
11%
6%
4%
7%
16%
1%
3%
2%
2%
10%
8%
8%
0%
2%
6%
MIC.
DISTANCE
(FT.)
50
56
95
145
245
445.
50
100
200
56
100
200
50
100
200
50
100
200
360
50
100
200
400
99
149
249
99
99
69
99
149
99
149
249
99
149
249
69
MEAS.
Leq
68.9
67.2
71.5
69.1
66.6
62.5
62.0
56.8
52.4
73.2
69.9 x
66.9
68.6
61.7
58.5
59.1
54.8
52.3
47.3
60.3
60.0
55.8
51.6
54.5
51.7
50.2
51.6
53.6
49.3
52.2
51.5
51.2
49.8
49.3
51.2
49.6
49.3
50.6
PREDICTIONS
TSC
72.7
75.1
75.0
73.5
71.6
69.0
66.5
63.5
60.2
78.7
76.6
73.8
71.2
68.2
62.1
65.2
62.4
59.3
57.5
68.9
66.0
62.9
59.5
52.0
52.2
51.1
52.8
57.9
53.3
58.0
58.0
50.3
50.1
48.6
55.2
55.1
53.5
55.6
NCHRP
Program
70.0
75.0
78.0
74.0
70.0
65.0
65.0
80.0
75.0
70.0
72.0
62.0
56.0
52.0
68.0
62.0
58.0
52.0
50.0
49.0
43.0
50.0
54.0
51.0
53.0
51.0
51.0
49.0
46.0
51.0
48.0
43.0
54.0
NCHRP
Nomogrom
73.0
76.9
74.9
73.0
69.2
65.9
64.5
60.0
55.5
78.2
74.9
69.8
72.6
68.1
62.8
66.2
61.5
56.1
51.0
68.7
63.0
57.4
53.1
Wyle
71.0
72.8
74.2
72.2
69.7
66.3
63.9
59.8
55.6
77.4
74.7
71.3
67.5
63.5
59.5
64.3
60.8
57.1
53.5
65.4
62.1
58.4
54.4
51.3
52.1
50.7
52.0
54.7
52.8
55.6
55.6
48.8
47.6
45.5
52.8
52.6
50.9
53.4
-32 -
-------
TABLE III (Cont'd)
SITE
IDA
B
C
MA
B
C
D
E
12A
B
C
13A
B
C
D
HWY
TYPE
8F
8F
4F
8F
TRAFFIC
FLOW
(VEH/HR)
13,650
11,280
13 140
7,440
7,950
3,090
6,000
8,250
2,184
2,028
3,054
6,303
6,318
8,130
5,772
%
TRUCKS
3%
3%
2%
0.5%
1%
3%
2%
3%
17%
24%
9%
7%
9%
8%
9%
MIC.
DISTANCE
(FT.)
131
256
506
131
256
506
131
256
506
131
256
506
131
256
506
131
256
506
131
256
506
131
256
506
T94
334
480
194
334
480
194
334
480
100
140
190
290
100
140
190
100
290
ME AS.
Leq(dB)
60.9
66.3
65.5
60.3
62.7
64.8
60.0
62.6
64.3
61.7
62.4
59.3
64.0
64.6
52.0
60.4
61.4
59.1
62.5
64.6
62.2
65.1
65.8
62.7
68.2
64.5
60.6
64.4
60.6
62.0
66.1
63.0
62.0
75.2
70.8
66.1
59.6
76.7
71.6
67.0
62.1
52.6
PREDICTIONS
TSC
....
77.1
74.5
71.5
71.7
69.2
66.2
72.2
69.5
66.4
69.4
66.8
63.8
71.7
69.0
65.9
73.6
71.0
68.0
71.7
70.3
69.1
68,0
73.3
71.9
70.6
71.7
67.2
NCHRP
Program
73.0
67.0
62.0
72.0
66.0
61.0
71.0
66.0
61.0
64.0
61.0
57.0
66.0
52.0
58.0
64.0
60.0
55.0
66.0
61.0
57.0
70.0
65.0
60.0
61.0
62.0
62.0
71.0
62.0
58.0
54.0
73.0
65.0
60.0
72.0
54.0
NCHRP
Nomogrom
Wyle
59.2
61.8
62.0
59.2
61.5
61.7
59.2
61.6
61.9
59.2
63.0
61.4
59.7
65.3
63.6
57.4.
66.1
63.8
59.7
66.6
64.5
61.5
70.2
68.0
63.3
63.8
62.9
63.7
64.5
63.7
62.0
63.0
61.9
/7.0
65.4
64.1
65.7
77.8
66.5
65.2
67.4
62.7
-33-
-------
TABLE IV
Confidence Limits and Standard Deviations
Model
Wyle
TSC
NCHRP
Mean
Difference
1.26
5.50
.01
90% Confidence Limits
Lower
.61
4.68
-1.01
Upper
1.90
6.31
1.04
Standard
Deviation
3.5
3.9
5.1
-34-
-------
'measured
+7
+5
+2
TSC
^o
^
0)
cr
0)
0
-1
-2
-3 -
Wyle
NCHRP
Notation:
Mean
90o/0 Confidence Limits
Figure 18. Comparison of Differences Between Measured and
Predicted L
eq
-35-
-------
the Wyle model has an average overprediction of about 1 dB and the TSC model has an
average overprediction of about 5 dB. These average differences must be considered
cautiously, however, because they are to a large degree a function of the particular
highway sites considered* The major consideration in comparing the three models must
remain the detailed review of Section 2.0.
One serious point to be noted in Table III is that the two versions of NCHRP
do not always give the same results. Differences on the order of 1 dB are reasonable
because of inherent inaccuracy in reading the charts and because of rounding of com-
puter output. Differences of up to 14 dB occur in some cases, however. In some cases
the program would not run, although the chart method could be used. Error messages
obtained in these cases did not seem pertinent to the input data.
A similar problem was noted, to a much lesser degree, for the TSC program.
It appears that the program uses reasonably correct values of atmospheric absorption,
rather than the obsolete values reported in the original documentation. This problem
could simply be one of updating documentations.
-36-
-------
4.0 CONCLUSIONS
The NCHRP, TSC and Wyle highway noise prediction models have been reviewed
and compared. The following conclusions have been reached:
• The Wyle and TSC models predict L directly and are based on funda-
mentally sound theory. The NCHRP model is based on a model which
assumes a flow of equally spaced identical vehicles.
• Automobile noise levels in all three models are in reasonable agreement
with each other, within a few dB. It would be desirable, however, to
obtain more recent automobile data and select one consistent set for all
models.
• The TSC and NCHRP models use truck levels based on meager data, and
neglect the speed dependence of tire noise. The Wyle model uses truck
levels based on several thousand roadside measurements at high and low
speeds.
• None of the models handles propagation in an entirely satisfactory way.
The TSC model, in particular, uses questionable values for ground atten-
uation. More research on highway noise propagation must be done.
• All three models use essentially correct diffraction theory for shielding
calculations, although intuitive modifications are employed in some cases
at high and low shielding. Varying assumptions as to vehicle source heights
cause great differences in shielding predictions. None of the models ade-
quately documents source heights used. In view of the sensitivity of shielding
calculations to source height, this must be better established.
• Treatment of multiple-lane roads and finite roads is often crudely done,
leading to errors of up to several dB. Inadequate consideration of propa-
gation losses adds to this problem.
• Analysis of differences between measurements and predictions, summarized
in Table IV, shows the Wyle and NCHRP models to have the most consistent
agreement, although none could be said to be entirely satisfactory. Propa-
gation effects and measurement errors are felt to be the greatest source of error.
-37-
-------
• Theoretical calculations of L,n and other statistical metrics require assump-
tions which make the usefulness of such calculations questionable. Since
adequate empirical data exist to relate L to these metrics, there is little
need for a highway model to predict other than L . Other metrics can be
obtained empirically from this measure with an accuracy entirely adequate
within the context of annoyance assessment.
A general difficulty with the computer programs — NCHRP in particular and TSC
to a lesser degree — is that original versions have been revised through the years, often
for cases where it might have been more appropriate to abandon the original method and
start again with a sounder approach. An example is the NCHRP model at low volumes.
Johnson and Sounders' model is clearly inadequate here, but the approach has been to
23
retain the basic formulation and add an empirical correction factor. A much better
approach is employed in the TSC and Wyle models which adopt an L formulation initially.
Finally, it should also be noted that a highway model need not be complicated.
This review has noted certain areas where detail is not important. These include road
width effects at large distances, and distant road elements. Complete detail of these
need not be included. The computerized models tend to include too much detail of road
geometry, without regard for its importance to accuracy. This situation apparently arises
because the use of a computer permits such detailed complications with very little effort
on the part of the user. Inclusion of such detail, however, adds little to the accuracy
of noise prediction and greatly obscures its simplicity.
-38-
-------
REFERENCES
1. Gordon, C.G., Galloway, W.J., Kugler, B.A., and Nelson, D.L.,
"Highway Noise — A Design Guide for Engineers", NCHRP Report 117 (1971).
2. Kugler, B.A., and Piersol, A.G., "Highway Noise — A Field Evaluafion of
Traffic Noise Reduction Measures", NCHRP Report 144 (1973).
3. Kurze, U.J., Levison, W.H., and Serben, S., "User's Manual for the Prediction
of Road Traffic Noise Computer Programs", DOT-TSC-315-1, May 1972.
4. "Noise Standards and Procedures", Federal Highway Administration PPM 90-2,
February 1973.
5. Plotkin, K. J., "A Model for the Prediction of Highway Noise and Assessment
of Strategies for Its Abatement Through Vehicle Noise Control", Wyle Research
Report WR 74-5, for the Environmental Protection Agency, September 1974.
6. Sharp, B.H., Plotkin, K.J., Glenn, P.K., and Slone, R.M., "A Manual for
the Review of Highway Noise Impact", Wyle Research Report WR 76-24, for the
Environmental Protection Agency, May 1977. Also EPA Report 550/9-77-356.
7. Plotkin, K.J., "A Case for Simple Highway Noise Models", Paper KK3,
91st Meeting of the Acoustical Society of America, April 1976.
8. Johnson, D.R., and Sounders, E.G., "The Evaluation of Noise from Freely
Flowing Road Traffic", Journal of Sound and Vibration, 7, 2, pp. 287-309, 1968.
9. Olson, N., "Statistical Study of Traffic Noise", APS-476, National Research
Council of Canada, Division of Physics, 1970.
10. Galloway, W.J., Clark, W.E., and Kerrick, J.S., "Highway Noise Measure-
ment, Simulation, and Mixed Reactions", National Cooperative Highway Research
Program Report 78 (1969).
11. Sharp, B.H., "A Survey of Truck Noise Levels and the Effect of Regulations",
Wyle Research Report WR 74-8, December 1974.
12. Society of Automotive Engineers, "Standard Values of Atmospheric Absorption
as a Function of Temperature and Humidity for Use in Evaluating Aircraft Flyover
Noise", ARP 866, August 31, 1964.
13. Ingard, U., "On Sound Transmission Anomalies in the Atmosphere", Journal of
the Acoustical Society of America, 45, pp. 1038-1039, 1969.
14. Pao, S.P., "Prediction of Excess Attenuation Spectrum for Natural Ground
Cover", Wyle Laboratories Research Staff Report WR 72-3, February 1972.
-39-
-------
REFERENCES (Cont'd)
15. Nelson, P.M., "A Computer Model for Determining the Temporal Distribution
of Noise from Road Traffic", TRRL Report LR611, Environmental Division,
Transport Systems Department, Transport and Road Research Laboratory,
Crawthorne, Berkshire, 1973.
]0< Kurze, U., and Beranek, L.L., "Sound Propagation Outdoors", Noise and
Vibration Control, edited by L. L. Beranek, McGraw-Hill, 1971, Chapter 7.
17 Maekawa, Z.E., "Noise Reduction by Screens", Applied Acoustics, 1,
pp. 157-173 (1971).
18. Kurze, U., and Anderson, G.S., "Sound Attenuation by Barriers", J. Applied
Acoustics, 4, 1, pp. 35-53 (1971).
19. Nelson, P.M., "The Combination of Noise from Separate Time Varying Sources",
J. Applied Acoustics,^, pp. 1-21, 1973.
20 Kurze, U.J., "Noise from Complex Road Traffic", Journal of Sound and Vibration,
1^,2, pp. 167-177, 1971.
21. Plotkin, K.J., "The Assessment of Noise at Community Development Sites.
Appendix A - Noise Models", Wyle Research Report WR 75-6, October 1975.
22. Grove, G.H., "Traffic Noise Level Predictor Computer Program", Research
Report No. R-942, Michigan State Highway Commission, October 1974.
23. FHWA's User Instruction Manual for Converted Michigan Department of State
Highways and Transportation's Traffic Noise Level Predictor Computer Program —
Version No. 10 (10/1/74)
24. Kentucky Department of Transportation Bureau of Highways Prediction Procedure
Correction Factor Nomograph, FHWA Notice N 6640.5, June 1975.
25. Plotkin, K.J., Kunicki, R.G., and Shop, B.S.., "National Impact of Highway
No?se and Effectiveness of Noise Abatement Strategies", Wyle Research Report
WR 76-14.
-40-
• u.s. aovnauNT rnsnira ornci t 197? 0.341-037/39
------- |