Journal of Industrial Aerodynamics, 5 (1979) 61--91 61

© Elsevier Scientific Publishing Company, Amsterdam -- Printed in The Netherlands

AN ANALYTICAL APPROACH TO WIND VELOCITY GUST FACTORS

M.E. GREENWAY
Department of Engineering Science, University of Oxford, Oxford (Gt. Britain)
(Received June 15, 1978; accepted in revised form January 23, 1979)

Summary

An analytical expression is derived for the factors (called wind velocity gust factors) by
which the mean wind speed is scaled to obtain a maximum gust speed. The analysis is
based on the assumptions of the horizontal wind velocity spectrum being of the yon
Karman form, and the probability distribution of the horizontal wind velocity being of
gaussian form.
The wind velocity gust factors are obtained as functions of the structural size and hence
can be used to obtain an appropriate design wind speed for any given structure. The
analytical expression for these factors depends on a number of dimensionless parameters
from which conclusions are drawn regarding wind tunnel modelling of wind effects on
structures.
Measurements related to the extremes of a filtered linearized velocity signal from a
hot-wire anemometer in a wind tunnel simulation of the atmospheric boundary layer are
described. These show good agreement with the predictions of the analysis.

1. I n t r o d u c t i o n

Excluding other considerations, such as fatigue, structures and structural

elements have to be designed to bear the m a x i m u m loads that are likely to be
imposed upon them within the lifetime of the structure. Maximum loads due
to wind effects occur when high pressures are correlated over the entire
structure, or structural element, concerned. Large correlated loads tend to
occur when the structure (or element) is engulfed by an eddy of at least the
same size.
Clearly the determination of design wind loads based on the dynamic
pressure obtained from the mean wind speed yields unrealistically low load
values. For this reason most design codes (e.g. [1] ) allow for the effects of
gusts by "factoring u p " the m a x i m u m expected mean wind speed of given
probability. Design data for these factors -- which are called wind velocity
gust factors in this paper -- have been deduced from a number of measure-
ments of wind speed averaged over fixed short time intervals [2, 3]. The codes
generally recommend a value of the short averaging time depending on the
size of the structure concerned. For example, the British wind loading code
of practice [ 1] recommends an averaging time of 3 s for structural elements
62

such as cladding, roofing and glazing; 5 s for structures with the largest
vertical or horizontal dimension less than 50 m; and 15 s for structures with
the largest dimension greater than 50 m. Newberry et al. [4] have pointed
out that this gradation is too coarse. Further, there is wide support for the
suggestion of several authors [4, 5] , that the averaging time, Tav, appropriate
to a structure or structural element of size,Dst,, should be selected by use of
an equation of the form
Tar = constant X D ~ / V
where V is the mean wind velocity.
In this paper wind velocity gust factors are derived analytically. The above
equation for the averaging time is incorporated in the analysis from the begin-
ning, together with the yon Karman analytical description of the wind velocity
spectrum. This spectrum, in non-dimensional form, is a function of the
dimensionless parameter n XL./V, where n is frequency (Hz) and XLu is the
longitudinal integral length scale of the turbulence. The result of this choice
of spectral form and of the averaging time given by the above equation, is a
self-consistent analysis yielding the wind velocity gust factors in terms of the
dimensionless ratio Dstr/XL., and thus relating them directly to the structural
size parameter, D,~.
In the analysis, the wind velocity is further assumed to be a gaussian
random variable. A linear filter is associated with the short-time averaging
process, and hence the filtered wind velocity is also gaussian. This allows the
analysis of Rice [6], Cartwright and Longuet-Higgins [7], and Davenport [8]
to be used to determine an extreme value distribution for the averaged wind
speed.
An alternative approach to the effect of gusts is that which attempts to
derive a "maximum gust profile" (similar to a mean velocity profile), rather
than determine gust factors. An analytical approach to such a profile was
made by Brook and Spillane [9]. They too used the yon Karman spectrum
although they assumed the existence of a characteristic time scale rather than
a characteristic length scale (XL,,) as is adopted here. Also, their gust profile
is based on the percentile points of the filtered wind velocity probability
density, and not on an extreme value analysis.
The motivation for casting this analysis in terms of the dimensionless ratio
D~/XL,,, as opposed to fixed short averaging times such as 3 s, 5 s etc. is
important and is discussed in Sections 2, 3 and 4. The equation for the gust
factors is obtained in Section 5. Section 6 describes some measurements in a
wind tunnel simulation of the atmospheric boundary layer which show good
agreement with the predictions of the analysis.

2. Existing methods of determining a design wind speed

The British Standards Institute's recommended code, CP 3 [1] for the

determination of wind loads on buildings and structures is appropriate to stiff
 63

structures. Structures likely to respond to wind dynamically are deliberately

omitted and the designer is advised to seek specialist assistance. The method
of determining the wind load on structures handled by the code is based on
two separate aspects
(i) the determination of a pressure coefficient appropria~ to the shape of
building or structure and
(ii) the determination of a design wind speed.
The design wind load is then easily obtained from the product of the
pressure coefficient and the dynamic pressure based on the design wind speed,
Using the code CP 3 to arrive at a design wind speed it is necessary to read
off a map a "basic wind speed" appropriate to the geographic location of the
structure and then to multiply this wind speed by three factors which account
for the known characteristics of the wind (see Table 1).
Another source of data concerning the behaviour of wind is that provided
by the Engineering Science Data Unit [10]. Although this is not as simple to
use as CP 3 it follows a similar pattern. An appropriate wind speed can be
determined again from a map of "basic wind speed" and this time five rather
than three factors accounting for the variation of wind speed. Table 1 com-
pares the code CP 3 [1] with the E.S.D.U. data sheets 72026 [10].

TABLE 1

CP3 Wind feature E.S.D.U.

(72026)

Maximum gust speed Map of U.K. showing Extreme mean-hourly wind

likely to be exceeded on basic wind speed speed likely to be exceeded
average only once in 50 contours on average only once in 50
years at 10 m above the years at 10 m above the
ground in open level ground in open level
country country

Local topographic KL, the topographical

features such as exposed factor
S 1 factor
hill tops and sheltered
valleys

(
Surface roughness of sur- Ks, the surface roughness
roundings factor

S~ factor
Height (of structure) Kz, the height factor
above ground

Gust duration appropriate KT, the averaging time

to the size of building factor

Probability of occurrence KN, the probability factor

of design wind speed or
S 3 factor
equivalently building
design life
64

The subject of this paper is the second last parameter listed in Table 1,
the so-called averaging time factor. Before proceeding to examine this factor,
it is considered appropriate to mention the difference in the basic wind data
presented in the code CP 3 and the E.S.D.U. data sheets 72026.

3. Basic wind data

From a n e t w o r k of recording stations distributed all over the U.K., the

meteorological office obtain from anemograph records [11]
(a) the mean-hourly wind speed for each clock hour in the day
(b) the highest gust speed of the day.
All these data are stored on a computer archive from which various statistics
can be obtained. To arrive at the basic wind speed map presented in CP3, the
single m a x i m u m gust speed from the 365 daily highest gust speeds is selected
for each year, for each anemograph station. Annual maxima from successive
years are then analysed to obtain an extreme value distribution following a
m e t h o d due to Gumbel [12]. This analysis yields a wind speed value likely to
be exceeded, with a probability of 0.636, at least once in a period of 50 years.
The collection of these values, for different anemograph stations, forms the
data set from which the map showing constant wind speed contours is drawn.
The reference wind speed map in E.S.D.U. 72026 is obtained in an identical
manner except that it is based on the mean hourly wind speed and not the
gust speed.
While the mean hourly wind speed is well defined, the gust speeds recorded
are effectively averages over a short period which depends on the response
time of the anemometer. T w o types are used by the meteorological office.
They are the Dines pressure-tube anemograph which has been in use for over
65 years, b u t which is being phased o u t in favour of the second type, the
electrical cup-anemograph. Both these instruments are said to have a response
time estimated to be between two and five seconds and so the gusts are con-
ventionally referred to as "3-sec gusts". However, it must be noted that the
cup anemograph has a response time which varies approximately inversely
with the mean wind velocity. Hence the response will be fast in a high wind
speed and slow in a low mean wind speed.

4. The concern with short interval gust velocities

Research into the effects of wind on structures dates back to the Tay
Bridge disaster in 1880. After an inquiry into this disaster, the Board of
Trade issued a ruling that structures should be designed to resist wind pressures
of 56 lb/ft 2 applied over the frontal area. At the time, some engineers con-
sidered this to be unrealistically high, Benjamin Baker [13] pointing out that
many existing structures would have failed under this lateral loading. Because
of a lack of data on wind loads this figure had to be accepted, b u t in connec-
tion with the design and construction of the Forth Railway bridge, Baker
 65

decided to make his own measurements. On an island in the Firth of Forth,

near the site of the proposed bridge, Baker had four wind pressure gauges
erected. The sensing element of one was a square board 300 ft 2 and those of
the other three were circular boards 1.5 ft 2 in area. During the course of
seven years, the m a x i m u m pressure registered on the small boards was 31 lb/
ft 2 and that on the large board was 19 lb/ft 2. Other investigators had ob-
served this "dimensional e f f e c t " but Baker was one of the first to understand
it and offer the explanation that now carries wide support.
The motion of the air in the wind is very turbulent and eddies of differing
energy content covering a wide spectrum of sizes are present. Eddies which
are small compared to the size of the structure impinge upon the structure
successively rather than instantaneously, and hence small eddies are unable to
correlate significant pressures over the whole structure. On the other hand,
eddies which are larger than the structure itself envelop the entire structure
and hence are able to cause well correlated pressures over the surface of the
structure, which contribute to the instantaneous load on the structure.
Other aspects of the turbulent motion of air over a structure such as the
distortion of turbulence in the mean flow field [14] are known to influence
the loads imposed by wind on a structure. However these can be considered
as second-order effects compared to the influence of eddy size.
Clearly then, as has been pointed out by Davenport [15], the important
parameter affecting the response of the load on a structure to the buffeting
of the turbulent eddies in the wind is the ratio of a typical structural dimen-
sion, Dstr, to a representative length scale of the turbulence -- conventionally
the integral length scale XL~.

4.1 Response o f flat plates to a normally incident turbulent flow

The effect of the Dstr/XL. parameter is clearly illustrated by some results
of Bearman [16] showing the aerodynamic admittance of flat square plates
of different side lengths Dsq to a normal turbulent flow of integral length
scale XL,. Bearman plotted his results against n D~q/V, and the data collapsed
more or less onto a single curve. Here t h e y are presented against a n XLu/V
coordinate (see Fig. 1) so that it can be seen that for fixed XLu and V, the
real cut-off frequency depends on D~qfl'Lu, and increases for smaller D~q/*L~,.
Also for the same Dsq and *L~ values, the cut-off frequency increases in
proportion to the mean velocity.
Vickery [17] has performed similar measurements of the aerodynamic
admittance of flat plates. Although his measurements were for plates of dif-
ferent geometric shape (circular, rectangular, etc.) he was able to suggest a
universal admittance function to correlate the experimental data. Based on
the frontal area, A, of the plate the admittance function is
1
66

~ 2.0
D ~ / X L u = 0,67

u :'< 2.67

Fig. 1. Aerodynamic admittance of fiat square plates after Bearman [ 16 ].

The square root of the aerodynamic admittance would correspond to the

amplitude response of the drag force to the velocity fluctuations. The cut-off
frequency, of this response function, defined by the 3 dB point is thus
(nV~/~J)cut.o. = 0.26 (2)
which for square plates becomes
(nD,q/V)eu~off = 0.26 (3)

4.2 Response o f a cup anemometer to turbulent flow

Recent tests on a prototype cup anemometer designed by "Danjay Designs"
reminded the author [18] that the velocity indicated by a cup anemometer
responds to a turbulent flow in much the same way as the load on flat plates.
The frequency response function can be shown to be of the form
1

V
where Dam is the distance constant* of the anemometer defined as "the
length of wind" required to blow over the anemometer before it registers
63% of a sudden increase of wind speed. The cut.off frequency defined by the
3 dB point is given by
(nDam/V)cut~ff = 0.159 (5)

4.3 Previous work on effective gust velocities

Despite Davenport's pointing out the importance of the DstJXLu parameter
as early as 1963, most full-scale work on the effect of gusts on structures has
focussed on the idea of averaging wind speed over short time intervals. Empir-

*In fact Dam is usually a weak function of V.

 67

ical factors relating these short time averaged values to the mean wind speed
V have been obtained [2,3].
There are at least three reasons w h y the time average idea has prevailed.
(i) The meteorological office have quoted response times for their anemom-
eters (2--5 s). (ii) The time duration, Ted, of an eddy and its size, Ded, can be
related b y Taylor's hypothesis, i.e.
Ted ~ Ded/V

Hence it m a y have been thought that averages over a fixed time interval
could thus be related to averages over fixed eddy sizes, b u t obviously this
can only be done at one constant and agreed value of the mean wind velocity,
V. (iii) In an experiment it is easy to average an electrical signal over fixed
time intervals using a low pass filter. Note that the frequency response of an
analogue short-time averaging circuit which performs the mathematical
operation

1 t
v,d t) - f v(r) dr
Tav t - T av

can easily be shown to be

X(n) = Isin(n~ Tav) l/n~ Tar (6)
where Tav is the averaging time. This has a cut-off frequency defined by the
3 dB point given by
ncut.ot ~ = 0 . 4 4 / T ~ (7)

Presumably a low pass filter sharing this cut-off frequency and with a similar
cut-off rate would suffice to produce a short-time averaged signal.

4.4 Terminology for Meteorological Office gust speeds

It can be seen from the preceding discussion that while it would be accurate
to use the term "3 s average velocity" for the gust velocity recorded by a
Dines anemograph, this would not be true for the case of a cup anemograph.
It would be more accurate to speak of an "average length of wind" or to
define instead the distance constant of the anemometer. The U.S. Weather
Bureau do just this, quoting wind speeds corresponding to the "worst mile
of wind".
To provide statistics more applicable to the gust loading of structures, a
wind speed corresponding to the worst say, 10 m, of wind would be more
appropriate. N o w this length is of the order of the distance constant of the
meteorological office cup anemometers and hence the existing m e t h o d of
data collection is a good method. However, the terminology used to describe
recorded gust speeds could be improved.
68

5. Calculation of wind velocity gust factors

Noting the similarities in the frequency responses of cup anemometers and

flat plates to turbulent wind, one could conceive, at least in principle, of an
experiment measuring the peak values of the velocity signal from a series of
cup anemometers with different distance constants, Dam, covering a range of
values corresponding to the range of structural sizes of concern in the gust
loading problem. The ratio of the peak values to the mean wind speed would
provide the factors, KT, used to obtain a design wind speed in the E.S.D.U.
approach, or provide similar information for the code CP3 approach. As
desired, these factors would now be a function of the anemometer distance
constants (equivalently structural size), and not of the averaging time.
An equivalent concept will be applied in the following analysis in order to
calculate the wind speed gust factors. It will be assumed that the horizontal
turbulent wind velocity can be modelled by a signal with an analytical power
spectrum which, in non-dimensional form, is a function of n ~L,,/V. It will
also be assumed that the probability distribution of the turbulent wind velocity
is gaussian. Conceptually, this wind velocity signal will form the input to an
averaging filter. This filter's response function will be a known function of
the product of the dimensionless parameters n XL,,/V and Dstr/XLu, chosen to
model the aerodynamic admittance of a structure.
Because of the linear behaviour of the averaging filter, the output signal
will have spectral characteristics which can be determined, and the output
signal will be gaussian [19].
An extreme value analysis will then be applied to the output signal in order
to determine the average of the extreme values occurring in a given time, and
hence the gust factor from the ratio of this average extreme value to the mean
wind velocity.
The concept underlying the analysis procedure is illustrated schematically
in Fig. 2.

5.1 Power spectrum o f horizontal wind velocity

The power spectrum, Su (n) of the horizontal wind velocity is assumed to
be of the form first derived by von Karman [20] for homogeneous isotropic
turbulence. Its use for modelling the spectra of strong winds in a neutral
atmosphere has been advocated by Harris [21], and it forms the basis of the
E.S.D.U. data 74030 and 74031 [22], where it is expressed as
n S,(n)lo,, 2 = 4 ~/(1 + 70.8 ~2)s/6 (8)
where ~ = n XLu/V and XL,, is the horizontal integral length scale which is a
function only of the terrain roughness and the height above ground, ou 2 is
the variance of the wind velocity.

5.2 Choice o f averaging filter response function

Rather than use the empirically determined aerodynamic admittance func-
 69

CHARACTERISTICS
AMPLITUDE
TIME PROBABILITY SPECTRAL
Wind velocity
nSu(n)
Rel Freq
du2
U(t)
O

t
A
u(t) {/.u (t)
Filter response
X2 (n)

Linear
averaging
filter

~ r - , ]
LO J
Filter wind velocity
0(t) nSv(n)
#

t 9 tL~J
V(t)~V* v ( t )

J Extreme value analysis

Fig. 2. Illustration of the concept underlying the analysis procedure.

tion (1) for square plates to represent, to a first approximation, the aero-
dynamic admittance of a structure or structural element, the analytically
convenient frequency response function of the short-time averaging filter (6)
is selected, with a suitable choice of the averaging time, Tar. On the basis of
the measurements of Newberry et al. [4], Lawson [5] has suggested the value
Tar = 4.5 Dstr/Y (9)
as the relevant averaging time for the measurement of extreme pressure loads
on structures, or structural elements of size, Dst r. This value for Tar will be
adopted here.
Hence the frequency response function of the averaging filter is
70

sin [4.5 ~nD,~/V] sin [4.5 ~(D,c~/XL,)]

X(n) = [4.5 ~nDst~/V] [4.5~(D~r/~L,)] (10)

Because this function has a similar shape to (1) and (4), it could also be
used to represent the response of square plates or cup anemometers to wind,
provided their cut-off frequencies are matched. Matching the cut-off fre-
quency of the short-time averaging filter (7) to that of the square plate admit-
tance function (3) yields
Tar = 1.7 D~/V
and similarly matching that of the cup a n e m o m e t e r response function (5)
yields
T.v = 2.8 Dm/V
Thus by making the substitutions D,tr = 1.7 Dsq/4.5, or Dstr = 2.8 D ~ / 4 . 5 ,
the following analysis can be used as an approximation for the peak response
of square plates or cup anemometers.

5.3 Derivation of the equation for wind speed gust factors

The o u t p u t from the averaging filter is denoted V(t) = V + v(t), where V
is the mean value and v(t) the fluctuation about the mean. Note that in this
paper v(t) should not be confused with the instantaneous lateral fluctuating
velocity component. The power spectrum of v(t) denoted So (n) can be writ-
ten in terms of S,,(n) and the averaging filter's response function, viz.
So(n) = S.(n) X2(n) (11)
and the variance of V(t) can be obtained from this via the relation

av2= J S,(n)dn (12)

0

Substituting in (12) from (8), (10) and (11) yields

1
ao 2 XLu [ ~f sin2 [4.5~r~(Dst,/XL,)]d~] 2
o., 4.5 ~ D.¢r o (1 + 70.8 ~2)s16~2 (13)

The right hand side of the above expression is a function only of D,~/XL.
which is denoted G, i.e.
oo/a., = G(Dstr/XL.) (14)
An expression for the function G in terms of modified Bessel and Struve
functions is obtained in the appendix and recorded in a later section.
Davenport [8] has considered the extreme value distribution of a gaussian
random variable such as V(t). Introducing the simplified dimensionless variable
= (V(t)-V)/av = v(t)/ao (15)
 71

and drawing on previous work b y Rice [6] and by Cartwright and Longuet-
Higgins [7], he was able to show that the probability that the largest maxima
of ~ does n o t exceed a value ~m~ during a period, T, is given b y
exp (- exp ( - y ) ) (16)
where y is related to Wm= b y
~m= = x/2 In v T + y / x ~ 2 In v T (17)
and where v is the zero-crossing rate of v ( t ) given by
v = (1/2~) o0/oo (18)
The zero-crossing rate, v, can be expressed in terms of the power spectrum,
S o ( n ) , by

v = [? n 2 So n n So ,n, . ]n .
1

,1o,
Extreme value distributions of the form (16) are known as Fisher--Tippett
T y p e 1 distributions when the so-called reduced variate y is linearly related
to the extreme ~m~. Equation (17) is usually written in the form
Y = ~ ( ~ m a x - m) (20)
where 1/a is called the dispersion and m the mode of the distribution. Here
m = x/21nvT and 1/a = 1/x/21nvT (21)
The mean value of the distribution is given by
~m~ = V ~ In v T + 7 / x / ~ l n v T (22)
where 7 is Euler's constant (= 0.5772).
The dispersion is a measure of the width of the extreme value distribution,
so it can be seen from eqn. (21) that the width decreases for increasing T.
For this reason, the distribution is usually narrow and one point on the
distribution is regarded as sufficient for the purpose of obtaining the gust
factor. It has to be decided whether to base the gust factor on the mode or
the mean. The mode is more characteristic of the extreme value distribution
and is given by a simpler expression (21) than the mean (22}. However in this
analysis, the gust factor will be based on the mean value given b y eqn. (22}.
This has the advantage of being easy to check experimentally; the mean of a
number of samples of the extreme value distribution is more easily established
than the mode. Just h o w much the values of the m o d e and mean differ
depends, of course, on the value of the dispersion. Those with a preference
for basing the gust factor on the mode rather than the mean can use the
relevant equation which is derived later (see eqn. (25)).
First however, an expression for v must be obtained. Substituting from
(8), (10) and (11) in (19) yields
72

V
~
o
sin 2 [4.5 7¢R(Dstr/XLu)] dR
(1 + 7 0 . 8 R2) s/6
Is = (23)
XLu

--
~
0
sin 2 [4.5 nR (Dst+/~Lu)] dR
+
(1 70.8 R2)s/6 R2

The term in square brackets above is a function only of (Dstr/XLu) which is

denoted F, so'that the product vT becomes
(vT) = ( Y T/=Lu) F (D,tr/':L,) (24)
(In a similar way to the function G, an expression for F in terms of modified
Bessel and Struve functions is obtained in the appendix and recorded in a
later section.)
Now noting from (15) that the extreme values, ~'ma~/V, of the 1?/V distribu-
tion can be written in terms of ~ma~ by
Y,-,,~/Y = 1 + o,, ~,-,.,~/V,
on using eqns. (14) and (24) in (21) and (22) respectively, the following ex-
pressions for the mode, dispersion and mean value of the extreme velocity
distribution are obtained

1 I
Dispersi°n(~-~) =(~-)G( lnl(--~u)F~ ,L,.,
o (D,+, (D.,

In this paper the average ratio of Vm~/V has been called the gust factor
for wind velocities so eqn. (27) is an expression for these gust factors which
will be referred to as the gust factor equation.

5.4 Discussion of the gust factor equation

It can be seen from eqn. (27) that the gust factor is a function of three
dimensionless parameters, namely o./V, V T/XL. and Dstr/XL,,. The gust factor
is a linear function of o . / V - the turbulence intensity of the wind. The
parameter V T/=Lu can be interpreted as the number of turbulence length
scales to blow over the structure (or the point at which Vm~/V is being
measured) during the obsezvation time, T. V/~L, is approximately invariant
 73
u

with height above the ground so that the parameter V T/XL, can be further
considered to be the n u m b e r of integral length scales to blow over a point at
any chosen height above the ground, for example the 10 m height. The func-
tions F and G introduce into the gust factor equation the desired dependence
upon the ratio Dstr/XLu as opposed to the fixed time average.
The three dimensionless parameters discussed above have important impli-
cations for the wind tunnel modelling of the effect of wind on structures. If
extreme values are to be a feature of such tests then the following relation-
ships must hold
(Ou/~7)model = (Ou/Vd)~nscale (28)
x
(Dstr/~ Lu )model = (Dstr/ Lu )~ seale (29)
(V T/~Lu)model = (V T/~L,)~unscale (30)
and because of the dependence of the functions F and G on the shape of the
wind velocity spectrum

[Su (n XLu/V)] model ---- [Su (n XLu/V)] ~n-seale (31)

Equation (29) taken together with eqn. (9) gives the alternative dimension-
less parameter
( Y Tav/XLu)model = ( Y Tav/XLu)full scale (32)
The parameters o~/V and Dstr/XLu are c o m m o n l y modelled in wind tunnel
tests, however some tests have only attempted a partial simulation of the
power spectrum Su (n), that is the simulation has been accurate only over a
certain range of frequencies. Equation (31) stresses the need for a more
accurate simulation of spectra. Equations (30) and (32) can be interpreted as
Strouhal numbers and show how to choose correctly, the averaging time and
observation time in the wind tunnel as a function of wind tunnel mean velocity
and length scale.

5.5 Evaluation of the functions F and G

Equations (13) and (23) can be regarded as the defining equations for the
functions G and F respectively. Both expressions involve integrals which are
evaluated in the appendix by Fourier Transform methods. By using equations
(A18) and (A19) of the appendix, the following expressions for G and F can
be written

G \( DstrxLu/I= 2 s/6 s-' IF (½)1-½ [,~2 -2/3 F (2) s2 {K~ (s)L_~(s) + L,~(s)K2~(s) }

- (21/3 r(~) - s4/3 g_, (s) }1 -~ (33)

3

and
74

where

B(s) =
I (2-2/3r(])- s 1/3 K~ (s)) }J~
F(~) 2-2/3 ~ s 2 (K~(s)L~ (s) + L_~(s)K] (s)) - (21/3 r(-~) - s4/aK~(s)

and s -~3.36 D~u/XL,, K~ is the modified Bessel function of order v, L~ is

the modified Struve function of order v, F(x) is the G a m m a function.
Although the above equations for F and G can be regarded as exact forms,
they involve unusual functions. For this reason the table of values of F and G
presented in Table 2 is useful. Graphs of F and G are also shown in Figs. 5
and 4 respectively.

TABLE2

Table of the f u n c t i o n s F(Dstz/XLu ), G (Dstz/XLu)

Dstr
F G
XL u

0.01 1.611 0.976

0.02 1.026 0.964
0.03 0.790 0.952
0.05 0.569 0.933
0.10 0.367 0.893
0.15 0.284 0.860
0.20 0.237 0.832
0.30 0.184 0.782
0.60 0.119 0.672
1.00 0.086 0.574
2.00 0.057 0.439
4.00 0.039 0.322

The values in Table 2 were calculated using a digital computer. Function

subroutines were written for the Bessel function K, and the Struve function L.
These subroutines generated the values of the function by_means of power
series (for small arguments) or asymptotic expansions (for large arguments).
Power series expansions are obtained in the appendix for the two integrals
involved in the eqns. ((13) and (23)) defining the functions F and G. Use of
these expressions yield the following approximations for small values of
DstJXLu :
2 !
V(Dstr/~L,) -- [1 - 0.9645 (Dstrf'Lu)~] ~ (35)
which is accurate to within 2% for values of D~t~f'L,, up to 0.2; and
2 2

F(D, tr/XL,) ~- 0.0732 (Dst~/XLu) -~ [1 + 0.4823(Dstr/XLu) ~] (36)

which is accurate to within 4% for values of D ~ / " L . u p to 0~1.
 75

5.6 Computation of the gust factors

Once the table for F(D,t,/XLu) and G (D,t,/XL,) has been obtained, the
tabulated values may be used to determine the gust factor for any choice of
o , / V and V T/~L, from eqn. (27), using a pocket calculator. However, for
the purpose of illustration, a table of gust factor values has been computed
for two values of the parameter V T/~Lu (= 200, 2000). These are shown in
Table 3. It can be seen that the gust factors are only a weak function of the
parameter V T/XL,. One of these tables (appropriate to V T/~Lu = 2000) is
displayed graphically in Fig. 3. This contrasts with the empirically based Fig.
4b in the E.S.D.U. data sheets 72026 [10]. The lines on Fig. 3 correspond to
constant values of the parameter D,t,/XLu whereas those in the E.S.D.U. figure

035

03C

Q 2 ~=

i~~ , >

e~
f,
~_ 0 2 C

015

15 20 25
Gust factor

Fig. 3. G u s t f a c t o r s as f u n c t i o n s o f (Ou/~r) a n d (D~t~/XLu) f o r (V T/XLu) = 2000.

TABLE 3 ¢b

Gust Factors as functions of (au/V)and (Dstr/XLu)for (VT/XLu) = 200 (Table 3a), and (V T/XLu)ffi2000 (Table 3b).

TABLE 3a

Dstz/XLu
au/V 0.01 0.02 0.03 0.05 0.I0 0.15 0.20 0.30 0.60 1.00 2.00 4.00

0.075 1.261 1.249 1.240 1.228 1.210 1.197 1.186 1.170 1.138 1.113 1.081 1.056
0.100 1.348 1.331 1.320 1.305 1.279 1.262 1.248 1.227 1.184 1.151 1.108 1.074
0.125 1.435 1.414 1.400 1.381 1.349 1.328 1.310 1.283 1.231 1.189 1.135 1.093
0.150 1.523 1.497 1.481 1.457 1.419 1.393 1.372 1.340 1.277 1.226 1.162 1.112
0.175 1.610 1.580 1.561 1.533 1.489 1.459 1.434 1.397 1.323 1.264 1.190 1.130
0.200 1.697 1.663 1.641 1.609 1.559 1.524 1.497 1.453 1.369 1.302 1.217 1.149
0.225 1.784 1.746 1.721 1.685 1.629 1.590 1.559 1.510 1.415 1.340 1.244 1.167
0.250 1.871 1.829 1.801 1.762 1.699 1.655 1.621 1.567 1.461 1.377 1.271 1.186
0.275 1.958 1.912 1.881 1.838 1.768 1.721 1.683 1.624 1.507 1.415 1.298 1.205
0.300 2.045 1.994 1.961 1.914 1.838 1.786 1.745 1.680 1.553 1.453 1.325 1.223
0.325 2.132 2.077 2.041 1.990 1.908 1.852 1.807 1.737 1.599 1.491 1.352 1.242
0.350 2.219 2.160 2.121 2.066 1.978 1.917 1.869 1.794 1.646 1.528 1.379 1.260

TABLE 3b

DstzlXLu
oulV 0.01 0.02 0.03 0.05 0.10 0.15 0.20 0.30 0.60 1.00 2.00 4.00

0.075 1.305 1.293 1.285 1.273 1.254 1.240 1.229 1.211 1.175 1.146 1.108 1.076
0.100 1.406 1.391 1.380 1.364 1.339 1.320 1.306 1.282 1.234 1.195 1.143 1.101
0.125 1.508 1.488 1.475 1.456 1.423 1.400 1.382 1.352 1.292 1.243 1.179 1.127
0.150 1.610 1.586 1.570 1.547 1.508 1.481 1.458 1.423 1.351 1.292 1.215 1.152
0.175 1.711 1.684 1.665 1.638 1.593 1.561 1.535 1.493 1.409 1.341 1.251 1.177
0.200 1.813 1.781 1.760 1.729 1.677 1.641 1.611 1.564 1.468 1.389 1.287 1.203
0.225 1.914 1.879 1.855 1.820 1.762 1.721 1.688 1.634 1.526 1.438 1.323 1.228
0.250 2.016 1.976 1.950 1.911 1.847 1.801 1.764 1.705 1.585 1.487 1.358 1.253
0.275 2.118 2.074 2.045 2.002 1.931 1.881 1.840 1.775 1.643 1.535 1.394 1.279
0.300 2.219 2.172 2.140 2.093 2.016 1.961 1.917 1.846 1.702 1.584 1.430 1.304
0.325 2.321 2.269 2.235 2.184 2.101 2.041 1.993 1.916 1.760 1.633 1.466 1.329
0.350 2.422 2.367 2.330 2.275 2.186 2.121 2.070 1.987 1.819 1.681 1.502 1.355
 77

are for constant averaging times. Figure 3 here applies to one value of V T/XL,
and the position of the lines would vary slightly with different values of
V T/XL,. Though it has been argued here that the averaging time formulation
of gust factors should be abandoned, it is possible to cast the preceding
analysis in terms of the averaging time, Tar. When this is done the following
approximate relationship for the gust factor based on the 3-s average velocity
is obtained

This equation yields values generally less than those of the 3-s line in E.S.D.U.
Fig. 4b.
On the basis of experimental data, Deaves and Harris [23] suggest the fol-
lowing formula for the gust factor Va-s/V recorded b y a standard U.K.
anemograph
V3.s/V = 1 + 3.7 ( o , / ~z)
For large values of (a,/V) this expression also yields values less than the 3-s
line in the E.S.D.U. Fig. 4b. Note that neither the E.S.D.U. figure nor the
above formula account for the influence of the (V T/XL,) parameter.
It has yet to be established whether the present analysis holds for measure-
ments made in the atmospheric wind. Whether this is the case or not will
depend to a large extent on h o w closely the assumptions of the analysis hold
in the atmospheric boundary layer -- in particular, whether the probability
distribution of the wind velocity approaches the assumed gaussian form. If
there is any marked deviation from the gaussian form it is likely to occur
near the ground, where the turbulence is known to become strongly non-
isotropic, and also where the turbulence intensity is high. These effects might
also be expected to be reproduced in a wind tunnel simulation of the atmo-
spheric boundary layer. For this reason, and as a check on the analysis, some
measurements were made in a wind tunnel simulation. These are described in
the next section.

6. Measurements in a wind tunnel simulation of an atmospheric b o u n d a r y layer

6.1 Description of the wind simulation

Measurements of the function F and G and the wind velocity gust factors
were made in the Oxford University L o w Speed Wind Tunnel. This tunnel,
which was specifically built for architectural and industrial aerodynamic
testing, has a working section with the dimensions 2 m X 4 m in cross section,
and 14 m in length. The construction and performance of the tunnel has
been fully described by Wood [24].
The wind simulation in which the measurements were made was that set
up to model, at a scale of 1/75, the full scale measurements made by the
78

Building Research Establishment on the Aylesbury low rise housing estate.

The mean velocity profile, turbulence intensity and spectra were designed to
agree with the standard E.S.D.U. data [10,22] for open level farmland, with
the empirical roughness factor, z0, chosen to be 0.07 m full scale. This target
simulation was achieved by a combination of coarse graded grid, fences and
floor roughness elements. For a detailed description of this wind simulation
refer to Greenway and W o o d [25].

6.2 Measurement of the length scale, ~L,,

The experiments were c o n d u c t e d at a position 2 m upstream of the turn
table centre, on the centre-line of the tunnel, at t w o different heights above
the floor; one at 133 mm (10 m full scale), and one at 666 m m (50 m full
scale). At b o t h these positions the p o w e r spectrum of horizontal wind speed
was measured, in a manner similar to that described in Greenway and Wood
[25], using a DISA 55 K series hot-wire anemometer, and a data acquisition
programme on an on-line P.D.P. 11/10 computer. Fast Fourier transform
techniques were used when processing the data to obtain the spectra. Using
the computer's on.line plotting facility, these spectra were plotted in nS (n)
vs. log n form to a chosen scale. By superimposing a transparent trace of the
E.S.D.U. spectrum in nS(n) vs. log (n XL,,/V) form, drawn to the same scale,
until a best fit was obtained, the ratio ~L,,/V could be deduced from the shift
in the scales of the t w o superimposed spectra. A separate measurement of V,
made at the data acquisition stage enabled ~L,, to be estimated. The following
values were measured
At position (2, 0, 0.133) XL./V = 0.135 s
V = 4.68 m/s
_ * (37)
"L./V = 0.120 s
- I
At position (2, 0, 0.666)
V = 6.35 m/s

6.3 Measurement of the F(D,tr/XLu) and G(D,tr/XLu) functions

Recall from eqns. (12), (13) and (14) that the function G is the ratio of
the deviations of the filtered and un-filtered velocity signals. To measure G,
the linearized velocity signal was first amplified and low pass filtered at a cut-
off frequency chosen to correspond to the Dst~/~L, values in Table 2. A Kemo
variable filter t y p e V B F / 8 K was used. This had a cut-off rate of 48 dB/octave.
The cut-off frequency could be set to 3 figures within the frequency range
0.001 Hz to 10 kHz. The cut-off frequencies corresponding to the chosen
Dstr/~L, values were easily determined from eqn. (7) and (9), and the
measured ratio of ~L,,/V, (37) i.e.
ncut.o ff ---- (0.44/4.5)/(~Lu/~z) (Dst,/~L.)
The deviation o f the filtered velocity signal was determined b y the use of an
on-line c o m p u t e r programme which sampled the signal at a selected rate and
 79

dynamically displayed a probability density on an oscilloscope operating in

X - - Y mode. The visual display of the probability density allowed the amplifi-
cation factor to be chosen so as to avoid either t o o narrow a probability
density, or a signal o u t of range of the A--D converter. Once an acceptable
display was obtained, the c o m p u t e r programme calculated the mean and
deviation of the signal from the moments of the probability density.
Using sampling techniques a low pass filter is inevitably used to satisfy the
Nyquist criterion. For this reason the deviation of the unfiltered velocity
signal was not actually measured in these tests. However, it's value was de-
duced on the assumption that the value for G (0.01} in Table 2 is correct.
From the measurement of the deviation of the filtered velocity signal, with
filter setting corresponding to D~tr/:'L, = 0.01, the following values of o, for
the t w o different heights were obtained

(2, 0, 0.133), au = 1.15 m/s I (38)

(2, 0, 0.666), au = 1.23 m/s

Hence with the relevant values for the mean velocity from (37), the turbulence
intensity at each measuring position was

(2, 0, 0.133}, o,IV_ = 0.245 (39)

(2, 0, 0.666), o , / V 0.194 1
Measured values of G were thus obtained from the measured values of o~ for
different values of the D,t~/XL, parameter by dividing by the relevant value
of o,, from (38}. For ease of comparison, the measured values for both heights,
and the predicted line (from values in Table 2) are plotted in Fig. 4.
Values of the F function were measured in two different ways. Note from
eqn. (18) and (24) that the F function is related to the ratio of the deviations
of differentiated filtered velocity signal and the filtered velocity signal, i.e.
F = ( 1 / 2 . ) (o~/o~) (XLu/V~)

In the first m e t h o d an analogue differentiator was used to measure a0 in

much the same way as av was measured to determine the G function. Values
of the F function determined from the ratio of a~ and av in this way are
plotted in Fig. 5, together with the predicted curve from the values in Table 2.
It can be seen that the measured values deviate from the predicted line
considerably at large values of Ds,~/:'Lu. When the measurements at large
Dstr/XL. were made, the probability density display of the differentiated
filtered velocity signal was observed to be noticeably non-gaussian. The
probability density display appeared gaussian at first, then flat-topped and
finally actually had a double peak shape as the low pass filter cut-off fre-
quency was decreased to correspond to D~tr/"L~, changing from 0.01 to 4.0.
The deviation of the points at large DsJXL,, from the predicted line, and the
distortion of the probability density may have been due to an imperfect
80

05--

I I I I I I I I
0,0~ (302 Q05 0,1 Q2 05 10 20 50

Dstr /XL u

Fig. 4. Comparison of measured and predicted values of the function G (Dst r IXL u );
Predicted line (eqn. (33)); ~ Measurements at position (2, 0, 0.133); o Measurements
at position (2, 0, 0.666).

z~

1"5i~

1.0-

0.5

I I I I I - ~ --I~t-- -Q- j
0.01 0.02 0,05 0.1 0.2 0.5 1.0 2.0 5.0
Dstr/XLu
Fig. 5. Comparison of measured and predicted values of the function F(Dstr/XLu );
Predicted line (eqn. (34)); V Measurements at position (2, 0, 0.133) with differentiator
method; v Measurements at position (2, 0, 0.133) with zero crossing rate method; ~ Measure-
ments at (2, 0, 0.666) with differentiator method; o Measurements at (2, 0, 0.666) with
zero crossing rate method.
 81

frequency response of the differentiator at low frequencies. Another possible

cause could have been a p o o r signal to noise ratio; a differentiator is an in-
herently noisy device.
In the second m e t h o d the zero crossing rate, v, of the filtered velocity sig-
nal was measured. Recall from eqn. (24) that F is given in terms of v by
F = v XLu/V

To measure the zero crossing rate, the linearized filtered velocity signal was
low pass filtered in the way described for the measurement of the G function.
A high pass filter of low cut-off frequency (0.03 Hz) was also used to remove
the mean. This filtered signal formed the input to a Schmitt trigger. The
o u t p u t of the Schmitt trigger switched rapidly from 5 V to 0 V whenever the
input signal crossed 0 V with positive slope. (It could also be set to trigger on
negative slope.) A small c o m p u t e r programme was used in the P.D.P. 11/10
computer to set up an accurate prescribed time interval on a crystal controlled
computer clock. During this interval the computer was enabled to receive
interrupts (occurring at random time intervals) from the o u t p u t of the Schmitt
trigger. By counting the number of interrupts during the prescribed interval,
an average zero crossing rate could be determined.
The values of F obtained from these measurements are also plotted in Fig.
5. The agreement with the predicted values at large values of D~t~/XL~ is much
better than those determined b y the first method. This presumably implies
that the large deviation of the points measured at large D~t~/XL~ values by the
first m e t h o d is due to the experimental technique or equipment rather than
being due to a poor simulation of the wind velocity spectrum.

6.4 Gust factors determined from extreme velocity measurements

A series of experiments were performed to determine the wind velocity
gust factors at both the previously mentioned heights in the simulated
b o u n d a r y layer. Measurements were made for values of Dstr/XLu between
0.01 and 1.0 shown in Table 2, for the parameter V T/XLu = 200. To check
on the influence of the observation time T, further measurements were made
for Dstr/XLu = 0.01 and 0.1 for V T/XLu = 600, 1000, 1400, 1800.
To measure the gust factor for any given value of Dstr/XLu and V T/XLu it
was necessary to measure the extreme values of short time averaged velocity
signal, with the averaging time chosen to depend o n D s t r [XLu through eqn.
(9). To achieve this, another c o m p u t e r programme was written. This pro-
gramme sampled the velocity signal periodically with sampling interval A t.
By storing the last m samples of the signal an average over the last m a t seconds
could be determined every time a new sample was taken. By comparing this
average value with the largest previous average value, the extreme average in
any observation period T could be recorded. Effectively then, the programme
acted as a digital filter performing a "running average" on the samples of the
velocity signal. It can be shown that the frequency response of such a digital
filter is
82

X(n) = --1 Isin(nm~ht)[= Isin(n~Tav) I

m sin(n 7r At) m~m(n-~T,v/m)
Provided m i> 4 this frequency response function corresponds closely to that
of the ideal analogue short time averager given in eqn. (6) in the frequency
range up to (2At) -I i.e.the Nyquist frequency. W h e n the programme was run,
the values of m, At, and T were entered from a Teletype control terminal.
The value of T was chosen to satisfy P T/~L. = 200 (say). m and At were
chosen so that m a t = Ta~, with m ~ 4, and Tar = 4.5 (D,t,/XLu)/(V/~L,). The
velocity signal was low pass filtered to satisfy the Nyquist sampling criterion.
Because the extreme of the filtered velocity signal is a random variable,
the value occurring in an observation period, T, is merely one sample of the
extreme value distribution, so more samples must be taken so that the m o d e and
dispersion parameters can be estimated. For this reason the programme was
run within a loop so that 16 extreme values occurring in 16 independent time
intervals of duration T were measured. These 16 extreme values were used to
calculate the m o d e and dispersion parameters b y use of the Best Linear Un-
biased Estimator (BLUE) procedure described b y Lieblein [ 26].
Once the m o d e and dispersion of the extreme velocity distribution were
determined, the mean value could be calculated (mean = m o d e + 0.577 X
dispersion), and all these values expressed in dimensionless form b y division
b y the mean velocity, V. These experimentally determined values are
tabulated in Table 4a and b for V T/~L,, = 200 for the heights 0.133 and
0.666, respectively. Also shown in the table are the values predicted b y eqns.
(25), (26) and (27) calculated with a,,/V given by (39). The mean of the
extreme velocity distribution made non-dimensional by dividing b y the mean
velocity, V, was earlier called the gust factor. The graph in Fig. 6 plots the
gust factor at both measuring positions and the predicted line (from equation
(27)) for ease of comparison.
The measurements for D,t~/XL. = 0.01 and 0.1 for different values of the
parameter V T XL,. are shown in Table 5. Half the square of the dimensionless
m o d e values are plotted in Fig. 7, against the coordinate log (V T / " L , ) , from
which an approximate linear behaviour is observed.

6,5 C o m m e n t s on the comparison o f predicted and measured results

It can be seen from the preceeding discussion that the measurements of
the F and G functions and the extreme velocities all depend on the experi-
mentally determined value of XL,/V. This parameter is difficult to determine
accurately. The error in the measurement of X L , / V is probably a b o u t 10%.
Any error in this value affects the cut-off frequency chosen for the low pass
filter to correspond to the given Dst~/XL. value. This error would cause smaller
errors in the G function measurements as well as the extreme velocity
measurements. However, it will cause a comparable error in the F function
measurements because F is determined from a product of another measured
parameter (e.g. zero-crossing rate) and the ~ L . / V value. Although the trend
TABLE 4

Table of measured and predicted values of the dimensionless mode, mean, and dispersion, of the extreme filtered wind velocity
distribution for (V T XLu ) = 200, showing the variation with (Dstr/XLu); Table 4a: Measurements at position (2, 0, 0.133); Table 4b:
Measurements at position (2, 0, 0.666)

TABLE 4a

Dstr/XLu
0.01 0.02 0.03 0.05 0.10 0.15 0.2 0.3 0.6 1.0

Dimensionless measured 1.84 1.87 1.78 1.74 1.61 1.59 1.58 1.50 1.38 1.37
mode predicted
eqn. (25) 1.81 1.77 1.74 1.70 1.64 1.60 1.57 1.51 1.41 1.34

Dimensionless measured 1.91 1.92 1.81 1.79 1.66 1.63 1.64 1.53 1.41 1.40
mean i.e. predicted
gust factor eqn. (26) 1.85 1.81 1.78 1.75 1.68 1.64 1.61 1.56 1.45 1.37

Dimensionless measured 0.123 0.085 0.058 0.080 0.076 0.067 0.096 0.043 0.055 0.065
dispersion predicted
eqn. (27) 0.070 0.072 0.073 0.074 0.075 0.074 0.073 0.071 0.065 0.059

TABLE 4b

Dstr/XLu
0.01 0.02 0.03 0.05 0.10 0.15 0.2 0.3 0,6 1.0

Dimensionless measured 1.69 1.60 1.61 1.58 1.48 1.45 1.45 1.42 1.34 1.29
mode predicted
eqn. (25) 1.64 1.61 1.59 1.56 1.51 1.47 1.45 1.41 1.33 1.27

Dimensionless measured 1.73 1.64 1.64 1.62 1.51 1.49 1.48 1.45 1.37 1.31
mean i.e. predicted
eqn. (26) 1.68 1.64 1.62 1.59 1.54 1.51 1.48 1.44 1.36 1.29

Dimensionless measured 0.076 0.082 0.045 0.065 0.061 0.061 0.052 0.043 0.045 0.037
dispersion predicted
eqn. (27) 0.056 0.057 0.058 0.059 0.059 0.059 0.058 0.056 0.052 0.047 GO
O0
84

O O

14-- "" x 0 0

~-r9

I I I I I I
002 005 01 Q2 05 1.0
Ost r /XL u

Fig. 6. Comparison of measured and predicted values of the gust factor for ( V T/XLu ) ffi
200; - - Predicted, o measured values at position (2, 0, 0.133); . . . . Predicted,
measured values at position (2, 0, 0.666).

of the predicted and measured values of both the F and G function corre-
spond very closely, the possibility of error in the XL,,/V parameter accounts
for the poorer agreement in the magnitudes of the measured and predicted F
function values, compared with those for the G function.
It can be seen from Fig. 7 that the measurements of the gust factors at
both positions (2, 0, 0.133) and (2, 0, 0.666) are consistently high for Dstr/
XLu = 0.01. This is most likely caused by too much energy in the high fre-
quency end of the simulated wind velocity spectrum.

7. C o n c l u s i o n s

An equation has been derived for the wind velocity gust factor appropriate
to a structure or structural element of size Dstr, which depends on the ratio
Dstr/xL u rather than on an averaging time. The gust factors have been found
to be linearly dependent on the turbulence intensity, (o,,/V). Also, as a result
of using an extreme value analysis to determine the wind velocity gust factors,
the expected increase of the gust factor with an increase in the observation
time, T, is revealed in the _gust factor equation, with T incorporated in the
dimensionless parameter (V T/XLu). From the expression of the gust factor
equation in terms of dimensionless parameters it has been deduced that it is
TABLE 5
Table of the m e a s u r e d a n d predicted values of the dimensionless m o d e , m e a n , a n d dispersion of t h e e x t r e m e filtered wind velocity
distribution for (Dst~ffLu) = 0.01, a n d 0.1, s h o w i n g the variation with (V T f f L u ) ; Table 5a: M e a s u r e m e n t s at position (2, 0, 0.133);
Table 5b: M e a s u r e m e n t s at position (2, 0, 0.666)

T A B L E 5a

Dstr/XLu = 0.01 Dstr/XLu = 0.10

V T]XLu 200 600 1000 1400 1800 200 600 1000 1400 1800

Dimensionless measured 1.84 1.99 1.99 2.04 2.10 1.61 1.70 1.80 1.82 1.84
mode predicted
eqn. (25) 1.81 1.89 1.92 1.94 1.95 1.64 1.72 1.75 1.77 1.79

Dimensionless measured 1.91 2.06 2.04 2.08 2.14 1,66 1.75 1.82 1.86 1.87
m e a n i.e. predicted
gust factor eqn. (26) 1.85 1.92 1.95 1.97 1.99 1.68 1.76 1.79 1.81 1.82

Dimensionless measured 0.123 0.120 0.090 0.078 0.067 0.076 0.083 0.039 0.068 0.047
dispersion predicted
eqn. (27) 0.070 0.064 0.062 0.061 0.060 0.075 0.067 0.064 0.062 0.061

T A B L E 5b
Dstr/XLu = 0.01 Dstr/XLu ~ 0.10

V T/XLu 200 600 1000 1400 1800 200 600 1000 1400 1800

Dimensionless measured 1.69 1.77 1.78 1.82 1.82 1.48 1.56 1.61 1.65 1.63
mode predicted
eqn. (25) 1.64 1.70 1.73 1.74 1.76 1.51 1.57 1.60 1.61 1.63

Dimensionless measured 1.73 1.80 1.81 1.85 1.86 1.51 1.60 1.64 1.67 1.66
m e a n i.e. predicted
gust factor eqn. (26) 1.68 1.73 1.76 1.77 1.78 1.54 1.60 1.62 1.64 1.65

Dimensionless measured 0.076 0.056 0.063 0.046 0.060 0.061 0.060 0.039 0.036 0.052 C]O
dispersion predicted e~
eqn. (27) 0.056 0.051 0.049 0.048 0.047 0.059 0.053 0.050 0.049 0.048
86

2~5m

[3
[] /

2.0
/
/ "
t~ / j /

=o ~5

O . ~ "
.__....-A/

J _------r Z~ I I I I
VT/XLu=200 600 1000 1400 1600

200
1 ,
500
I
1000
I I la2000 I
5000
- ×
(VT/ L u )

Fig. 7. Plot of the measured and predicted values of half the square of the dimensionless
mode of the extreme filtered wind velocity distribution against log ( V T/XLu), showing
the influence of the observation time, T. At position (2, 0, 0.133): measured values o
Dstr/XLu = 0.01, o Dstr/XLu ffi0.1; predicted lines - - - - - Dstr/XLu = 0.01 ; - - - - Dstr/XLu ffi
0.1. At position (2, 0, 0.666): measured values V Dstz/XLu ffi 0.01, ~ Dstz/XLu ffi 0.1;
predicted lines --- Dstz/XLu= 0.01, Dstr/XLuffi0.I

necessary to model the parameters given by eqns. (28), (29), (30) and (31) if
extreme velocity measurements are to be a feature of any wind tunnel tests.
Good agreement has been f o u n d between the predictions of the analysis
and measurements made in a wind tunnel simulation of the atmospheric
b o u n d a r y layer. The analysis can therefore be used with confidence to yield
a wind velocity gust factor for use in wind tunnel measurements once XL u
and a u / V have been measured. The analysis has yet to be checked with full
scale wind velocity data and this is considered to be a worthwhile exercise.

Acknowledgements

The author wishes to t h a n k Mr. R.I. Harris of the Environmental Sciences

Research Unit, Cranfield for his helpful suggestions regarding the evaluation
of the integrals in the appendix. Thanks are due also to Mr. R.E. Belcher for
his able assistance with the experiments.
 87

References

1 British Standards Institute Code o f Practice, CP3, Ch. V, P~. 2, Wind Loads, 1972.
2 C.D. Durst, Wind speeds over short periods of time, Met. Mag., 89 (1960) 181.
3 E.L. Deacon, Wind gust speed: averaging time relationship, Austral. Met. Mag., 51
(1965) 11.
4 C.W. Newberry, K.J. Eaton and J.R. Mayne, Wind Loading on tall buildings - - further
results from R o y e x House, BRE Current Paper CP 29/73, 1973.
5 T.V. Lawson, The design of cladding, Building and Environment, 11 (1976) 37.
6 S.O. Rice, Mathematical analysis of random noise, Bell System Tech. J., 23 (1944)
282--332; 24 (1945) 46--156.
7 D.E. Cartwrigbt and M.S. Longuet-Higgins, Statistical distribution of the maxima of a
random function, Proc. Roy. Soc. A, 237 (1956) 212.
8 A.G. Davenport, Note on the distribution of the largest value of a random function
with application to gust loading, Proc. Inst. Civil Engrs., 28 (1964) 187.
9 R.R. Brook and K.T. Spillane, On the variation of maximum wind gusts with height,
J. Appl. Meteorol., 9 (1970) 72.
10 Engineering Sciences Data Unit, Item number 72026, 1972, amended 1974.
11 H.C. Shellard, The estimation of design wind speeds, Proc. Conf. on Wind Effects on
Buildings and Structures, N.P.L. Teddington, 1963.
12 E.J. Gumbel, Statistical theory on extreme values and some practical applications,
Appl. Maths. Ser., Nat. Bur. Stand., 33 (1954).
13 B. Baker, The F o r t h Bridge, Engineering, 38 (1884) 213--215.
14 P.W. Bearman, Some measurements of the distortion of turbulence approaching a two-
dimensional bluff body, J. Fluid. Mech., 53 (1972) 451.
15 A.G. Davenport, The buffeting of structures by gusts, Proc. Conf. on Wind Effects
on Buildings and Structures,N.P.L. Teddington, 1963.
16 P.W. Bearman, An investigation o f the forces on fiat plates in turbulent flow, N.P.L.,
Aero. report 1290, 1969.
17 B.J. Vickery, On the flow behind a coarse grid and its use as a model of atmospheric
turbulence studies related to wind loads on buildings, N.P.L. Aero. report 1143, 1965.
18 M.E. Greenway, Performance tests on a p r o t o t y p e cup-anemometer and wind vane for
Daniay Designs, Oxford University Engineering Laboratory Report, 1224/77, 1977.
19 W.B. Davenport, Probability and Random Processes, McGraw-Hill, New York, 1970,
p. 361.
20 Th. von Karman, Progress in the statistical theory of turbulence, Proc. Nat. Acad. Sci.
U.S., 34 (1948) 530.
21 R.I. Harris, The nature of the wind, Proc. of Seminar at the Institution of Civil
Engineers, June 1970, CIRIA Pub. 1971, paper 3.
22 Engineering Sciences Data Unit, Items number 74030 and 74031, 1974.
23 D.M. Deaves and R.I. Harris, A mathematical model of the structure of strong winds,
Environmental Sciences Research Unit Report No. 24, Cranfield, 1976.
24 C.J. Wood, The Oxford University 4 m × 2 m Industrial Aerodynamics Wind Tunnel,
Oxford University Engineering Laboratory, Report Number 1188/77, 1977.
25 M.E. Greenway and C.J. Wood, Wind tunnel pressure measurements on the Aylesbury
Low Rise ttousing Estate, Part 1, Simulation design and mean pressures, Oxford
University Engineering Laboratory, Report Number 1213/77, 1977.
26 J. Lieblein, Efficient methods of extreme value methodology, NBS1R 74-602,
National Bureau of Standards, Washington DC, 1974.
27 M. Abramowitz and I.A. Stegun, Handbook of Mathematical Functions, Dover, New
York, 1965.
28 A. Erd~lyi (Ed.), Tables of Integral Transforms, Volumes I and II, McGraw-Hill, New
York, 1954.
88

Appendix

Evaluation o f the integrals in equations (13) and (23)

The integral in eqn. (13) is denoted J,, i.e.

sin2[4.5 n ~(Dstr/XLu)]d~
J' = (1 + 70.8 ~2)s/6 ~2 (A1)
0

and similarly' the integral in the numerator of eqn. (23) is denoted J2, i.e.

sin 2 [4.5 ~ ~(Dstr/~Lu)]dVt

J2 -- (1 + 70.8 ~2)s/6
(A2)
0

To simplify the appearance of the integrals J1 and J2, the following substi-
tutions are made
4.5n Dstr/XLu = b (A3)
and for the constant 70.8 which is actually an approximate value for
4,[r(-~)y[r(-~)] -2 = a -2 (A4)
where F is the Gamma function, defined for example in Abramowitz and
Stegun [27]. On taking out a fractional power of a the integrals can be
written

,0 ~ sin2(bff) d~ (A5)
J1 = a-U ~2 (a 2 + ~2){
0

,o ~
j2 = a-C sin2(b~) ~ d~
J (a 2 + ~2)~ (A6)
0

To evaluate these integrals use will be made of Parseval's formulae for Fourier
sine and cosine transforms.
If the functions f(x) and g(x) have Fourier sine transforms, denoted Fs(y )
and Gs(y), defined for example by

Fs(y) = } f(x) s i n ( x y ) d x y > 0

0

or Fourier cosine transforms, denoted Fc (y) and Gc (y), defined for example
by

Fc(y) = "f f(x) cos(xy)dx y >0

gt
 89

then Parseval's formulae for Fourier sine transforms takes the form

f(x)g(x)dx =- Fs (Y)Gs (y)dy (A7)

7r
0 0

and for Fourier cosine transforms, the form

f(x)g(x)dx = - - Fc(y)Ge(y)dy (A8)

7r
0 0

Note the following Fourier sine transforms taken from Erdelyi [28]

Ref. [28] f(x) Fs(y)

_1

Vol. I x ( x 2 + a2) ~-~ ~ - (2a)'[r(~ - v ) l - ' y l - V K , ( a y )

p. 69 Re v > - 1 , Rea>0
2.3
(11) (A9)
?I
=~ 0<y<2b
Vol. I sin 2 (bx)
p. 78 x ?l
= -- y = 2b
2.6 b>0 8
(8) = 0 y>2b (A10)

and also the following Fourier cosine transforms

Ref. [281 f(x) Fc(y)

1 1 ~ 1 -1
Vol. I (x 2 + a2) -v-~ (~y/a)~Tr~[r(v + ~)] Kv(ay)
1
p. l l Rea>0, Rev>-~
1.3
(7) (All)
Vol. I sin 2 (bx)
p. 19 x2 t ~"(b-y/2) y<2b
1.6 b>0
(8) 0 y > 2b (A12)

Using the cosine transforms ( A l l ) and (A12) with Parseval's formula for
cosine transforms, J1 becomes on putting y = Y/a
90

t 2ab 4

Jt - 2]~bF(s) ~oby~K,(Y) fo Y~K,_(Y) (A13)

Similarly using Parseval's formula for sine transforms, the sine transforms (A9)
and (A10) yields
2ab
ffi - ~ 2~[r(~)] -' Y~ K, (Y) dY (A14)
j~ a4 , ~ fo '

K~ in (A13) and (A14) is the modified Bessel function of order v.

To evaluate the integrals in (A13) and (A14) the following standard forms
taken from Erdelyi [28] are required

Ref. [28]

Vol. II
p. 367
"f xVgv(x)dx ffi 2v-l~F(v+½)a[Kv(a)L~l(a) + Lv(a)K~-,(a)] (AI
§ 19.5
(21) 0 Re u >-½

(22) fx
a

TM K.(x)dx ffi 2 " r ( v + l ) - a T M Kv+, (a) (A16)

o Re u > --21

(23) fx'-VK (x)dx = 2 - v r ( 1 - v ) - a'-~Kv_,(a) (A17)

0 Rev< 1

Using (A15) and (A16) in (A13), and noting that K.,,(x) = K,,(x)
Jl = 1tab2 [K, (2ab)L.,_ (2ab) + L,_ (2ab)K,_ (2ab)]
3 $ 3 3

4 1 4.
+ [(2ab)~K._ (2ab)/2~r(~) - 1 ]/12a 2 (A18)
3

(L. is the modified Struve function, see for example Abramowitz and Stegun
[27].) Using (A17) in (A14), J2 becomes

J2 = [1 - (2ab)~K, (2ab)/2-]r(~)]a ~ r ( ~ ) / 4 r ( ~ ) (A19)

3

Note from (A3) and (A4) that the argument 2ab can be written
I
2ab = [4.5 n ~ [D /XLu] ~- 3.36 Dstr/XLu.
Clearly, then, values of the functions F(Dstr/XL.) and G (Dm/XL,,) can be
 91

generated from these expressions for J1 and J2 provided values of the mod-
ified Bessel functions, K, and modified Struve function, L, can be deter-
mined. Values of K and L can be generated on a computer by use of the
power series (for small arguments) and asymptotic expansions (for large
arguments) which are given in Abramowitz and Stegun [27]. A power series
for K is not given explicitly in [27] but can easily be obtained from that
given for the modified Bessel function Iv (x) using the property
7r I_~(x) - IAx)
Kv(x) = --
2 sin n v
Using the power series thus obtained for K, in (A13) and integrating term
by term, the following power series is obtaine+d for J1

b2 l 27 r(-~) (ab)~2 + (ab) 2 279 r(-~) (ab)~ + f (A20)

J' = 4 1 20 r(-~) 4 308 r(-~)

which is valid for small values of ab. Using the same power series for K, in
expression (A19), the following power series for J2 results 3
1 /3r(~) ~ 3(~b) ~ 9 r(-~) , t
= + (ab)~ + . . . . I (A21)
2 4

again valid for small values of ab. Expressions (A18), (A19), (A20) and (A21)
are all referred to in the main part of this paper.