Journal of Wind Engineering & Industrial Aerodynamics 171 (2017) 121–136

Contents lists available at ScienceDirect

Journal of Wind Engineering & Industrial Aerodynamics

journal homepage: www.elsevier.com/locate/jweia

Estimation of roughness length at Hong Kong International Airport via

different micrometeorological methods
Y.C. He a, b, P.W. Chan c, Q.S. Li a, b, *
a
Dept. of Architecture and Civil Engineering, City University of Hong Kong, Hong Kong
b
Architecture and Civil Engineering Research Centre, City University of Hong Kong, Shen Research Institute, Shenzhen, PR China
c
Hong Kong Observatory, Kowloon, Hong Kong

A R T I C L E I N F O A B S T R A C T

Keywords: Aerodynamic roughness length scale ðz0 Þ is an essential parameter for the parameterization of momentum flux
Aerodynamic roughness length exchanges at land-atmosphere interface. In this paper, several micrometeorological methods are applied for
Wind measurement estimation of z0 based on wind measurements at Hong Kong International Airport (HKIA). The concepts of source
Micrometeorological method area and internal boundary layer are adopted to better understand the measurement results. The validity and
Wind shear alerting prediction accuracy of the estimation methods for z0 are examined and discussed. A map of terrain roughness at
HKIA is established.

1. Introduction
Surface roughness is an aerodynamic property of the earth, which is
related to surface coverage, surrounding obstructions, topographic relief,
and so on. It serves continuously as a momentum sink for the atmospheric
flow (Wieringa, 1993), and plays an important role in governing wind
structures within the atmospheric boundary layer (ABL) (Garratt, 1994).
Conventionally, surface roughness can be best indexed by the aero-
dynamic roughness length z0 which is regarded as an empirical measure
of retarding/disturbing effects that the surface has on near-ground winds.
This parameter is essential for the parameterization of momentum flux
exchange at land-atmosphere interface, and accurate determination of its
value has been identified as a key issue in a wide range of applications in
wind engineering, such as determination of design wind loads on struc-
tures (Irwin, 2006), estimation of diffusion of pollutant plumes (Wong
and Liu, 2013), numerical simulation of environmental problems
(Blocken et al., 2007), mathematical modelling of wind field (Meng et al.,
1995), assessment of wind energy potential (Emeis, 2014), etc. Surface
roughness length is also of great importance to convert wind speeds
associated with different terrains, measurement heights or averaging
periods, and to better understanding of site-specific measurements of
surface wind (Powell and Houston, 1996; Verkaik, 2000; Vickery and
Skerlj, 2005; Harper et al., 2010; Masters et al., 2010a, 2010b; Balder-
rama et al., 2011; Miller et al., 2015; He et al., 2014a, 2016).
Several methods have been developed for the estimation of z0 . These
methods can be categorized into three groups: micrometeorological (or
anemometric) methods (Verkaik, 2000; Powell et al., 2003; Masters
et al., 2010a), classification methods (Davenport, 1960; Wieringa, 1992,
1993), and morphometric (or geometric) methods (Lettau, 1969; Grim-
mond and Oke, 1999). Since morphometric methods are usually only
applicable for built-up terrains, which will not be considered in
this study.
Micrometeorological methods are driven by wind measurements.
Commonly adopted micrometeorological methods include profile
method, variance method and gustiness method. Among these methods,
profile method requires mean speed records collected at multiple height
levels, while variance and gustiness methods need turbulence measure-
ments recorded at a single level. Details about these methods will be
discussed in the following section.
Classification methods rely on existing knowledge of z0 associated
with a group of basic terrain classes. The roughness length for a given
terrain can be subjectively assessed using roughness classes and visual
estimation. The problem is that roughness length suggested in different
literature for the same terrain type may vary distinctly. Wieringa (1992,
1993) reviewed 30 years' roughness data from boundary-layer mea-
surements and compared 5 popular classifications of roughness. It was
found that the local-scale classification of Davenport, (1960) is reliable,
provided that the lowest two roughness classes are adjusted. Table 1 lists

* Corresponding author. Dept. of Architecture and Civil Engineering, City University of Hong Kong, Hong Kong.
E-mail address: bcqsli@cityu.edu.hk (Q.S. Li).

https://doi.org/10.1016/j.jweia.2017.09.019
Received 8 May 2017; Received in revised form 21 September 2017; Accepted 24 September 2017

0167-6105/© 2017 Elsevier Ltd. All rights reserved.

Y.C. He et al.
Table 1
Davenport classification of effective terrain roughness.
Class Landscape description
1 Sea Open sea or lake (irrespective of wave size), tidal flat,
snow-covered flat plain, featureless desert, tarmac
and concrete, with a free fetch of several kilometers
2 Smooth Featureless land surfaces without any noticeable
obstacles and with negligible vegetation; e.g.,
beaches, pack ice without large ridges, marsh, and
snow-covered or fallow open country.
3 Open Level country with low vegetation (e.g., grass) and
isolated obstacles with separations of at least 50
obstacle heights; e.g., grazing land without
windbreaks, heather, moor and tundra, runway area
of airport. Ice with ridges across-wind.
4 Roughly Cultivated or natural area with low crops or plant
open covers, or moderately open country with occasional
obstacles (e.g., low hedges, isolated low buildings or
trees) at relative horizontal distances of at least 20
obstacle heights
5 Rough Cultivated or natural area with high crops or crops of
varying heights, and scattered obstacles at relative
distances of 12–15 obstacle heights for porous objects
(e.g., shelterbelts) or 8 to 12 obstacle heights for low
solid objects (e.g., buildings) (analysis may need zd )
6 Very rough Intensively cultivated landscape with many rather
large obstacle groups (large farms, clumps of forest)
separated by open spaces of about 8 obstacle heights.
Low densely-planted major vegetation like bushland,
orchards, young forest. Also, area moderately
covered by low buildings with interspaces of 3–7
building heights and no high trees (analysis requires
zd )
7 Skimming Landscape regularly covered with similar-size large
obstacles, with open spaces of the same order of
magnitude as obstacle height; e.g., mature regular
forests, densely built-up area without much building
height variation (analysis requires zd )
8 Chaotic City centers with mixture of low-rise and high-rise
buildings, or large forests of irregular height with
many clearings (analysis by wind tunnel advised)
the updated version of Davenport's classification (Davenport et al., 2000)
that is recommended by World Meteorological Organization (WMO,
2008). One may refer to Stewart and Oke (2012) for further information
to better understand the Davenport's classification.
Given the important role of z0 in governing the ABL and the diversity
of estimation methods for z0 , there is a need to assess the prediction
performance of different approaches. However, such works especially
those concerning varied micrometeorological methods are limited, as
wind measurements required for such comparisons are rarely available.
Among few related studies, Barthelmie et al. (1993) compared the pre-
diction results by using classification method, profile method, gustiness
method and speed-variance method. Considerable method-related vari-
ations in z0 were reported, with estimations in the same azimuth sector
varying by a factor of as much as 20. Verkaik and Holtslag (2007)
analyzed the roughness length using three micrometeorological methods.
It was reported that under inhomogeneous conditions, the profile method
might be invalid and estimation results of z0 via different
wind-turbulence based methods differed evidently. It is further note-
worthy that most estimation methods for z0 may suffer from significant
uncertainty. Verkaik (2000) evaluated the performance of two gustiness
methods and found that different parametric settings in the models could
result in great discrepancy of estimation results.
The motivation of this study is twofold: First, to assess the perfor-
mance of several roughness estimation methods through cross compari-
son analysis of the results from these methods; Second, to explore the
characteristics of surface wind at Hong Kong International Airport
(HKIA) and determine associated roughness length so as to advance wind
shear alerting for aircraft operation at the airport.
The remainder of this paper is organized as follows. Section 2
Journal of Wind Engineering & Industrial Aerodynamics 171 (2017) 121–136
introduces the estimation methods considered in this paper. Section 3
describes the observation sites and datasets of wind measurements. The
z0 ðmÞ estimation results are presented in Section 4. Conclusions and main
0.0002 findings of this study are summarized in Section 5.
2. Introduction of estimation methods
0.005
2.1. Source area and internal boundary layer
0.03 Due to adoption of wind measurements, z0 derived via micrometeo-
rological methods depends on both surface covers and arrangement of
measurement systems. It is stressed that a device placed above a site
explores only a portion of its surroundings. Meteorologically, the portion
0.10 of upstream surface that contains the effective sources contributing to the
flux exchanges at the concerned site is termed as the source area (or
footprint). A large source area accounts for a large-scale average of sur-
face properties. The source area is a function of observation height, sta-
0.25 bility condition, and surface roughness (Schmid and Oke, 1990). For
wind measurement, it is elliptical in shape and is aligned in the upwind
direction (~30 in width) from the concerned site (WMO, 2008). Under
neutral condition the area with significant contribution to measurements
0.5 atop a typical mast lies upwind at a distance of several kilometers, while
increasing instability reduces the source area to a region closer to the
concerned site.
In principle, terrain classifications are established for uniform ter-
rains. In reality, however, it is not uncommon to encounter terrain
changes in an area at a scale larger than several hundreds of meters.
Under such heterogeneous conditions, wind structures are dominated by
1.0 both local (or new) surface and upwind (initial) exposure. Internal
boundary layer (IBL) will be formed at the interacting area where at-
mospheres gradually adapt to the new surface (Garratt, 1990). The IBL
depth hI varies with the distance downwind of fetch change x (Powell
2 and Houston, 1996):
hI ¼ cz0R ðx=z0R Þ0:8 (1)
where c (0.28–0.75) is a constant that depends on stability status
(c  0:3 under neutral condition), z0R is the larger value of roughness
length of new and initial fetches. Wind structures above hI are governed
by upwind surface cover, while those below 0:1hI are completely
adjusted to the new terrain. Wind flows in the middle region demon-
strate a blended feature that is influenced by both initial and new ter-
rains: while small eddies are dominated by the immediate action of
local obstructions, large eddies exist in accordance with large-scale
terrain setups.
For an area with evident variation of terrain setups, the effective
roughness length, i.e., the roughness length producing a representative
momentum flux for the concerned area (Fielder and Panofsky, 1972),
may be estimated by analyzing wind measurements recorded at the
blending height, i.e., the height at which the flow is approximately in
equilibrium with the local surface and also independent of horizontal
position (Mason, 1988). Alternatively, it may be estimated by averaging
lnðz0i Þ (called the lnðz0 Þ avearage method hereafter), z0i being the
roughness length of each small-scale homogeneous patch (WMO, 2008).
But the effective roughness length tends to be larger than the average of
z0 values of all patches, due to the fact that it is easier to charge the at-
mosphere with turbulence than to discharge it by dissipation (Wieringa,
1993). Thus, the values of z0 listed in Table 1 constitute a lower limit of
effective roughness that can occur in terrain situations where such sur-
face cover is dominant.
2.2. Wind profile based method
Within the surface layer of neutrally stratified atmosphere, vertical
profiles of horizontal mean wind speed can be depicted by the logarith-
mic law:
122
Y.C. He et al. Journal of Wind Engineering & Industrial Aerodynamics 171 (2017) 121–136

u* z Similarly, for wind direction, one has:

UðzÞ ¼ ln (2)
κ z0 κ cv
σd ¼ (6)
in which, κ  0:4 is the von Karman constant, u* is the surface friction lnðz=z0 Þ
velocity, z denotes the height above the zero-displacement plane, or
where σ u and σ d (unit: rad) are the standard deviation of wind speed and
z ¼ z0  d, z0 and d being the height above ground and the zero-
wind direction, respectively; cu ; cv are coefficients depending upon sta-
displacement, respectively. Under open flat terrain condition, d can be
bility condition. Under neutral condition, cu ¼ 2:2, cv ¼ 1:9 for unfiltered
regarded as zero. But for rough surfaces (i.e., forests and built-up terrain)
measurements of σ u and σ d , while for filtered measurements in which σ u
where wake-interference or skimming flows exist, d should be taken into
is expected to be attenuated by about 12%, and σ d by 2%, cu ¼ 1:94, cv ¼
account so as to obtain reliable estimation of z0 .
1:86 (WMO, 2008). Fig. 1 presents the dependence of z0 on σ u and σ d at
Thus, given mean wind speeds U1 ; U2 at two heights of z1 ; z2 , z0 can
varied height z and with different inputting values of cu and cv .
be computed by:

U1 lnðz2 Þ  U2 lnðz1 Þ 2.4. Gustiness method

lnðz0 Þ ¼ (3)
U1  U2
As a technique driven by measurements of gust factor (GF), the
When wind records at multi-height levels are available, z0 can be
gustiness method has several forms that are characterized by adopted GF
determined as the intercept of the wind profile curve in the altitude co-
model (e.g., Wieringa, 1973; Beljaars, 1987). This study mainly considers
ordinate through fitting.
the method documented by Masters et al. (2010a,b). According to the
definition of GF, one has:
2.3. Variance method
Uτ;max Uτ;max  UT σ u σu
GF≡ ¼1þ ⋅ ¼ 1 þ gðT; τ; zÞ⋅ (7)
The best way of estimating z0 recommended by WMO (2008) is the UT σu UT UT
variance method. According to the similarity theory, the normalized
standard deviation of horizontal wind speed by the surface friction ve- where τ and T denote respectively gust duration and mean speed dura-
locity is a function of stability: tion, and gðT; τ; zÞ is the peak factor which may be expressed as
(Davenport, 1964):
σ u =u* ¼ cu ðz=L; h=LÞ (4) pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi .pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
gðT; τ; zÞ ¼ 2 lnðυTÞ þ 0:5772 2 lnðυTÞ (8)
where L is the Monin-Obukhov length, and h is the ABL depth.
Substituting Equation (2) into the above equation results in:
in which υ is the zero up-crossing rate. Equation (8) is established based
κ cu on the assumption of Gaussian distribution of wind speed records. When
σ u =U ¼ (5)
lnðz=z0 Þ non-Gaussianity of wind records should be taken into account (Balder-
rama et al., 2012), other revised forms, such as that proposed by Kareem

Fig. 1. Dependence of roughness length on variance of wind turbulence.

123
Y.C. He et al. Journal of Wind Engineering & Industrial Aerodynamics 171 (2017) 121–136
∞
σ 2u ðT; τ; z; z0 Þ ¼ ∫ 0 Su ðn; z; z0 Þχ 2 ðnÞdn (15)
and Zhao (1994), may be adopted.
In reality, measurements of σ u and gðT; τ; zÞ are inevitably attenuated In Masters et al. (2010a), the von Karman wind spectrum was adopted
by measurement systems. For a typical anemometry system chain con- to determine the filtered wind speed variance σ u ðT; τ; z; z0 Þ:
sisting of anemometer, transmission line and recorder, spectrum of 
recorded wind turbulence Sr ðnÞ may be expressed as (Beljaars, 1987; n⋅Su ðnÞ 4nLxu U
¼h   2 i5=6 (16)
Verkaik, 2000): σu
1 þ 70:8 nLxu U
Sr ðnÞ ¼ SðnÞ⋅χ 2 ðnÞ (9)
in which turbulence integral length Lxu can be calculated by a combined
χ 2 ðnÞ ¼ T1 ðnÞ⋅Thp ðnÞ⋅Tra ðnÞ (10) usage of Taylor's hypothesis, i.e., Lxu ¼ UTu , and an empirical estimator of
the integral time scale recommended in Engineering Sciences Data Unit
 2 1
(ESDU, 1983), i.e., Tu ¼ 3:13z0:2 (ESDU, 1983), while σ u is the unfiltered
T1 ðnÞ ¼ 1 þ ð2πnλ=UÞ (11) wind variance formulated by the Harris variance model (ESDU, 1985):

 1 16

Thp ðnÞ ¼ 1  1 þ ð2πnTÞ2 (12) 7:5u* η½0:538 þ 0:09 lnðz=z0 Þη

σ u ðzÞ ¼ ; η ¼ 1  6 f z=u* (17)
1 þ 0:156 ln½u* =jfz0 j
 2  2
sinðπnt0 Þ sinðπnΔNÞ The effective roughness length can be computed using an iterative
Tra ðnÞ ¼ ; or Tra ðnÞ ¼ (13)
πnt0 πnΔ method based on the above equations. Fig. 2 shows correlations between
GF and z0 estimated via the above introduced gustiness method, by
Here, SðnÞ is the unfiltered power spectrum, Tl account for low-pass considering the effects of four key inputting parameters, i.e., τ, T, z and λ.
filtering effect due to the inertia of anemometer's rotating components, Without specific notification, the defaulted parameter values involved in
Thp represents high-pass filter associated with the measuring period T, Tra each figure are set as: τ ¼ 3s, T ¼ 600s, U600 ¼ 10m=s, z ¼ 10m, λ ¼ 5m.
stands for running average filtering effects for the analog (first) or
discrete (second) signals, λ is the distance constant of the anemometer, n 2.5. Marine method
is the frequency (unit: Hz), t0 is the analog running-average duration, Δ
and N are the sampling interval and sample number of instantaneous Owing to the mobility of seawater, the state of sea surface, and
readings involved in the discrete average. therefore the marine roughness, depends greatly on wind strength. Other
υ and σ u can be expressed as: factors, such as water depth, can also influence the marine roughness
evidently (Smith et al., 1992). The Charnock model is conventionally
∞ adopted to estimate the roughness length over waters (Charnock, 1955;
∫ 0 n2 Su ðn; z; z0 Þχ 2 ðnÞdn
υ2 ¼ ∞ (14) Garratt, 1977):
∫ 0 Su ðn; z; z0 Þχ 2 ðnÞdn

Fig. 2. Dependence of z0 on varied inputting parameters via gustiness method.

124
Y.C. He et al. Journal of Wind Engineering & Industrial Aerodynamics 171 (2017) 121–136

  pffiffiffiffiffiffiffiffiffiffi
z0 ¼ αu2* g (18) z0 ¼ 10 exp  κ Cd;10 (22)

where α is a constant in range of 0.015–0.035 (Powell et al., 2003). Given in which Cd;10 ¼ u2* =U10
2
is the drag coefficient at 10 m AMSL, which may
the value of α, z0 can be determined using iterative method based on be empirically determined by (Smith et al., 1992):
Equations 2 and 18.
Fairall et al. (2003) examined the results from a number of mea- 103 Cd;10 ¼ 0:27 þ 0:116U10 (23)
surement programs and recommended the following speed-dependent
The above equation was derived based on field measurements at a
form of α:
coastal area where the local water depth was 18 m. Other formulas of
8
< 0:011 for U10  10 Cd;10 can be found in Vickery and Skerlj (2000), Harper et al. (2010) and
α ¼ 0:011 þ 0:000875ðU10  10Þ for 10 < U10 < 18 (19) He et al. (2016).
:
0:018 for 18  U10  25 The above models predict that the marine roughness length increases
monotonously with wind speed. However, some wind profile observa-
Oost et al. (2002) proposed the following an exponential relationship tions over deep oceans (Powell et al., 2003; Vickery et al., 2009) revealed
between α and wave age: that the drag coefficient, and therefore the roughness length, may level
  2:5  off or even reduce when wind speed exceeds a certain value. Thus, it is
α ¼ 50 Cp u* ; Cp ¼ gTp ð2πÞ (20) reasonable to use the capped forms of these estimators. For simplifica-
tion, it may be regarded that Cd;10 derives the maximum value when
where Cp and Tp are respectively the phase speed and period of the U10 ¼ 25 m=s based on the results reported by Vickery et al. (2009).
dominant wave, and Cp =u* is a measure of the wave age. For fully Fig. 3 shows dependence of z0 on mean wind speed using the capped
developed waves in deep waters, Tp ¼ 0:729U10n , U10n being the neutral forms of the above introduced four marine methods.
wind speed at 10 m above mean sea level (AMSL).
Unlike the above studies which focused on modelling of α, Taylor and 3. Introduction of observation sites and datasets
Yelland (2001) modelled the marine roughness directly as a function of
the significant wave height and peak wavelength: 3.1. HKIA
  .
4:5
z0 ¼ 1200Hs Hs Lp ; Lp ¼ gTp2 ð2πÞ (21) As an international metropolis, Hong Kong has one of the busiest
airports in the world. As is well known, low-level wind shear could be
where Hs is the significant wave height and Lp is the wavelength hazardous to departing/arriving aircrafts at the airport. Previous studies
associated with the dominant wave period (thus, Hs =Lp is approximately have shown that airflow disturbances induced by topography/terrain and
the slope of the dominant wave). For fully developed artificial structures (such as airport terminal) are a major cause of sig-
waves, Hs ¼ 0:0248U10n
2
. nificant low-level wind shear at HKIA (Liu et al., 2010; Chan and Lee,
Another kind of estimation method for marine roughness focuses on 2011; Chan, 2012a, 2012b; Li and Chan, 2012). To improve the safety of
modeling the drag coefficient which has a relationship with z0 as follows: airport operations, great efforts have been made to further understanding
of low-level wind shear around HKIA, with a focus on utilizing numerical

Fig. 3. Dependence of marine roughness on mean wind speed.

125
Y.C. He et al.
techniques to facilitate timely forecasting of such wind phenomena.
Although insightful results have been achieved, some basic assumptions
involved in the numerical studies are needed to be validated. One of them
is the value setting of roughness length which not only governs
approaching wind fields during the simulations but also plays an essen-
tial role in treatment of boundary layer condition.
The geographical location of HKIA is shown in Fig. 4. It is located at
Pearl River Estuary, China. The northern and western parts of the airport
are surrounded by seawater, while complicated terrain in Lantau Island
(with peaks rising to ~1000 m AMSL, and valleys ~400 m AMSL in
between) exist to the south of HKIA. Inside the airport, there are two
runways: the north and the south. Both runways have an orientation of
70/250 . The area in the vicinity of the runways is covered by low grass,
that to the south of the runway corridors is occupied by a number of
freight depots whose heights are in a range of several meters to over
20 m, while between them there are airport terminal (at eastern part,
maximum building height is ~30 m), midfield concourse building (at
middle part, maximum building height is ~30 m), and hangars and
warehouses (at western part, building heights are in a range of several
Fig. 4. Location of the study sites and surrounding terrain features.
126
Journal of Wind Engineering & Industrial Aerodynamics 171 (2017) 121–136
meters to 12 m), besides parking aprons with frequently berthed air-
planes. To protect the airport against sea waves and currents, a groyne
was constructed on the north side of HKIA. The slope of the groyne ex-
ceeds 0.3.
3.2. Equipment and datasets
HKIA has been equipped with various meteorological instruments at
different sites. As depicted in Fig. 4, seven masts were erected alone two
runways: A1E, R1E, R1C, R1W, R2E, R2C and R2W (“R1” and “R2” stand
for the south runway and the north runway, while “E” “C” and “W”
denote east, central and west, respectively). Each mast has been equipped
with anemometers and vanes at 10 m height above ground level (AGL) to
measure speed and direction of horizontal wind component. The ane-
mometers are of cup type (Thies 4.3303 model), whose distance constant
is 5 m. The detection accuracies of the anemometer and vane are 0.3 m/s
and 1 , respectively.
Two kinds of datasets from the masts are analyzed in this study:
routine measurements and fine records. The routine measurements
Y.C. He et al.
include 1 min mean wind speed and direction as well as 3 s peak gust in
each minute, which were recorded every minute based on instantaneous
readings that are sampled at 1 Hz. Ten years' (from January 2005 to
December 2014) data of this kind is available for this study. The fine
records consist of 1 s–1 s updated wind speed and direction that were
collected during a period of 44 days between July 2006 and July 2012
when strong wind warning signals (for both tropical cyclones and mon-
soons) were issued by Hong Kong Observatory (HKO). Except R1E, the
fine records are available for this study.
Besides the above introduced mechanical devices, A1E has also been
equipped with a Doppler sonic-detection-and-ranging (Sodar) profiler
(located at several meters far away from the meteorological mast). The
profiler (Aerovironment Model 4000) operates at a central acoustic fre-
quency of 4.5 kHz. It probes three components of wind by emitting three
acoustic beams (i.e., a vertical beam and two oblique beams at about 15
from the vertical) at 0.33 Hz. Mean wind velocity vectors at 39 gate levels
from 10 m to 200 m AGL were recorded every minute based on ensemble
backscatter signals associated with each gate. A 3 years' (2010–2012)
remote sensing dataset from the profiler is available for this study.
All the above three kinds of datasets are utilized in this study. The
records from the profiler are analyzed to obtain the vertical profiles of
mean horizontal wind speed so that the roughness length can be deter-
mined via the profile method. The measurements from the masts are
analyzed to investigate turbulent characteristics of wind near ground.
The fine records can be used to estimate z0 via both the gustiness method
and the variance method. On the other hand, the long-term routine
measurements allow us to examine the dependence of estimation results
on sample size and historical time through comparison analysis.
3.3. Quality control of datasets
To guarantee the quality of the wind measurements adopted in this
study, two kinds of tests are conducted on the raw datasets from the
masts, which aim to examine the thermal stability and stationarity of the
wind records, respectively (He et al., 2013). Measurements which fail to
pass the tests are discarded in the following analysis. Due to absence of
weather information required for the qualification of stratification status,
only neutrally stratified winds are concerned in this study. Practically, it
is regarded that neutral condition can be satisfied in strong wind cases.
Measurement results documented in Wieringa (1973) suggested that
neutral condition can be achieved when mean wind speed exceeds a
threshold level of about 5.5 m/s for daytime and 5.0 m/s for nighttime. In
this study, it is assumed that neutral condition can be achieved when
10 min mean wind speed exceeds 6 m/s.
Tests for stationarity of the wind speed records are conducted
following the way adopted in He et al. (2013) which includes four re-
quirements for each 10 min data run: (1) Standard deviation of wind
direction  15 . (2) Difference between scalar and vector mean
speed  0.51 m/s (or 1 knot). (3) Peak wind speed should not be beyond
5 times standard deviations away from mean. (4) Time series of wind
speed should pass tests for 1st order stationarity at a significant level of
0.05 using the reverse arrangement method.
This study does not examine the validity of remote sensing mea-
surements from the profiler, since relevant studies were conducted by
Chan (2008). The quality control for this dataset aims to examine the
reliability of data at each gate and the completeness of credible gate
records for each wind profile. It is regarded that a gate record is credible
if associated signal-to-noise ratio (SNR)  8 (He et al., 2014b), and a
wind profile is complete if the number of credible gate records contained
in the range of 15–100 m exceeds 14 (there are totally 18 gates in this
range). Here, the lowest gate records (i.e., at 10 m height) are discarded
because interference effect at this level (due to surrounding obstacles) is
generally severe, whilst gate records at a height above 100 m are dis-
regarded since the logarithmic law is only applicable in the surface layer
whose depth is in the order of 100 m in moderate-to-strong wind cases.
The selected profile measurements consisting only of credible gate
127
Journal of Wind Engineering & Industrial Aerodynamics 171 (2017) 121–136
records are further tested according to the following two criteria to
exclude those collected under either rainy or non-neutral conditions: (1)
The horizontal component of mean surface-layer speed  6 m/s. (2) The
amplitude of vertical component of mean surface-layer speed  3 m/s.
Throughout this study, the surface layer denotes the portion of the ABL
below 100 m AGL.
4. Results and discussions
4.1. Comparison of results around A1E
4.1.1. Wind profile and z0;p
Measurements at A1E are analyzed first to investigate the wind
characteristics around HKIA, and particularly to compare the estimation
performance for z0 by different micrometeorological methods. Owing to
unavailability of the refine records from the mast at A1E, those collected
at R1E (distance between R1E and A1E is 270 m) are utilized
substitutively.
Fig. 5 shows the ensemble-mean profiles of horizontal wind speed
normalized by the mean surface-layer speed. The results are stratified
into 12 azimuth sectors, according to the mean surface-layer direction.
The number of qualified profiles (N) in each sector is depicted in the
figure. As can be seen, influenced by nearby buildings and surrounding
topography, mean wind profiles in most sectors do not follow the loga-
rithmic law in the surface layer. In the northern and southern sectors,
there are evident profile kinks around 30–40 m. But the wind profiles in
two western sectors (i.e., 240 –270 and 270 –300 ) associated with an
open-flat exposure are distributed logarithmically.
Based on the above observations, wind profiles in Sector 250 –290
are selected to represent the profiles above open terrain, as shown in
Fig. 6. Two fitting lines are drawn in the figure via the least square
method based on the gate records in the range of 15–50 m (fit-1) and
50–100 m (fit-2), respectively. The estimated values of z0 , denoted as z0;p ,
are 0.05 m (fit-1) and 0.003 m (fit-2), with 95% confidence bounds being
0.007–0.1 m and 0.0006–0.006 m, respectively. The results basically
agree with the recommendations for Class-3 (z0 ¼ 0.03 m) and Class-2 (z0
¼ 0.005 m) in the terrain classification listed in Table 1. The dependence
of z0;p on different fitting ranges indicates the existence of IBL in this
sector (i.e., land area in Sector 250 –290 ) where is associated with a
smooth-to-rough terrain change.
From Fig. 4, the land area in Sector 250 –290 is featured by two
patches of terrain cover: open terrain dominated by runway and low
grass in an area of ~1 km upwind the study site, and an area featured by
occasional large obstacles beyond the above. If the values of z0 for these
patches are assumed to be 0.03 m and 0.1 m in accordance with the
suggestion in Table 1, the effective roughness length at this area
computed via the ln z0 avearage method turns to be 0.055 m, which is
almost identical to z0;p derived on the basis of the gate records below
50 m height. It is then assumed that approaching wind flows experience a
step change (i.e., marine-to-land) in underlying terrains before reaching
to the observation site, with z0 before and after the terrain transition
being 0.0002 m (Class-1 in Table 1) and 0.055 m, respectively. Using
Equation (1) (distance downwind of fetch change  3 km), the IBL depth
is calculated as 100 m. According to the IBL theory, wind flows between
50 and 100 m above the observation site are influenced by both the
overland terrain and marine exposure. Thus, it is reasonable that the
measured z0;p for this range is associated with a Class-2 terrain in Table 1.
4.1.2. GF and z0;GF
Values of GF vary with adopted gust duration τ and mean speed
duration T. In practice, the ratio between 3 s peak gust and 10 min mean
wind speed, i.e., GFð3;600Þ , is most frequently used (WMO, 2008).
Following this convention, 3 s peak gust over 10 min duration is
computed using overlapping average technique to get GFð3;600Þ for each
10 min wind record.
Y.C. He et al.
Fig. 5. Vertical profiles of normalized longitudinal mean wind speed at A1E.
Fig. 7 presents 25th, 50th, and 75th percentiles of GFð3;600Þ and asso-
ciated values of z0 estimated via the gustiness method ðz0;GF Þ, based on
the routine measurements at A1E (A1E-R) and R1E (R1E-R) and the fine
records at R1E (R1E-F). Statistics of the obtained results are not discussed
in terms of mean and standard deviation, as they often demonstrate non-
Gaussian feature (He et al., 2013; Lombardo and Krupar, 2017). The
results are stratified into 12 azimuth sectors according to the mean wind
direction ðθÞ. Data records involved in each sector are generally large
enough in size to make the statistics results meaningful.
The results at A1E-R and R1E-R agree considerably well in Sector
Fig. 6. Wind profiles and z0;p at A1E.
128
Journal of Wind Engineering & Industrial Aerodynamics 171 (2017) 121–136
0 –60 and Sector 150 –360 . The maximum relative difference of me-
dian of GF in these sectors is 2.5%. In Sector 240 –300 which corre-
sponds to a flat exposure, GF values from different datasets almost
coincide with each other, reflecting the credibility of the GF results. The
median of z0;GF in Sector 240 –270 and Sector 270 –300 is ~0.002 m
and ~0.02 m, different from ~0.05 m obtained via the profile method
based on the lower portion of profiles in Sector 250 –290 . From Fig. 4,
the source area in Sector 240 –270 is dominated by the south runway
corridor, while that in Sector 270 –300 is additionally influenced by the
midfield concourse building. Therefore, it is reasonable that these two
sectors respectively belong to Class-2 (featured by “no obstacles”) and
Class-3 (“few isolated obstacles”) stipulated in the terrain classification.
The noticeable difference between z0 estimations via the gustiness
method and the profile method can be explained by the height-
dependence of source area above the observation site. As discussed
previously, the hangars and warehouses located at the western part of
HKIA can influence the flux exchanging process above A1E in a vertical
range of 10–100 m (i.e., 0.1–1hI ). However, below this range, incident
atmospheres are completely adjusted to the local terrain.
In Sector 60 –150 , the results at A1E-R and R1E-R show distinct
discrepancy. Winds measured at A1E are consistently “gustier” than
those detected at R1E. Since A1E is located closer to upwind hills and
civil structures, it is believed that atmospheres over A1E are more
influenced by the disturbing effects of upwind obstacles than those over
R1E. Actually, previous studies of field measurements and numerical
simulations showed that flow separation and vortex may exist downwind
of obstacles (Chan, 2012b; He et al., 2014b). Such wind-turbulence
complex tends to be dissipated during the transferring process both
downwind of the obstacle and downward from the generation point,
resulting in obstacle-relative dependence of GF (He et al., 2013). How-
ever, it is unclear why this dependence of GF becomes less evident in
Sector 150 –210 where the topographic effects should be even more
Y.C. He et al.
Fig. 7. Measurements of GF and z0;GF at A1E and R1E.
significant. A possible explanation is that the observation sites are located
just downwind of the main part of Lantau Island (Fig. 4) in this sector,
and the gradient of GF around A1E and R1E decreases to a level at which
the distance between the two masts becomes insufficiently long
to resolve.
Comparison of the results at R1E-R and R1E-F reveals an interesting
phenomenon: under hilly exposure conditions GFs from the fine records
are consistently larger than those from the routine measurements, while
the above discrepancy vanishes in other sectors associated with less
evident topography. To explore this phenomenon, Fig. 8 examines the
relationship between GF, z0;GF and mean wind speed, based on the
routine measurements at A1E and R1E in two selected sectors (i.e.,
60 –120 and 120 –210 ). As reflected, GF and z0;GF generally increase
with mean wind speed. Because the refine records were collected during
strong wind events, GF and z0;GF in these sectors at R1E-F are reasonably
larger than those at R1E-R. The speed-dependence of wind turbulence
downwind of hills may be attributed to two reasons: First, flow separa-
tion, leeward wave/vortex and other topography-induced phenomena,
are more likely to occur under strong winds; second, when such phe-
nomena occur, the behavior of wind turbulence, such as the frequency of
vortex shedding, is closely related to Reynolds number of approach at-
mosphere or approach wind speed. It is emphasized that the mean wind
speed records depicted in Fig. 8 were collected downwind of Lantau Is-
land where shielding effect is evident. Thus, the speed-dependence of GF
and z0;GF may only be revealed partially.
4.1.3. Variance of wind turbulence and z0;σ
The fine records at R1E are analyzed to derive the standard deviation
of wind speed σ u and wind direction σ d via the technique documented by
Yamartino (1984). Results of 25th, 50th, and 75th percentiles of GFð3;600Þ
and associated z0 values estimated via the variance method ðz0;σu ; z0;σd Þ
are presented in Fig. 9.
The good agreement between the results of σ u and σ d demonstrates
the validity of the collected fine records. The maximum relative
Fig. 8. Speed dependence of GF and z0;GF at A1E and R1E.
129
Journal of Wind Engineering & Industrial Aerodynamics 171 (2017) 121–136
difference between σ u =U and σ d in sectors except 150 –180 and
270 –330 are less than 10%. Comparatively large differences in Sector
150 –180 and Sector 270 –330 are speculated to correlate with the
discrepancy of responding features of anemometers and vanes in non-
equilibrium atmospheres downwind of the mountain or the
airport terminal.
To compare the estimation performance of the gustiness method and
the variance method, results of z0;GF based on the fine records at R1E are
also depicted in Fig. 9. It is observed that while z0;σu and z0;GF agree well
under relatively rough terrain conditions (roughly z0 > 0:03 m), z0;σu
becomes consistently larger than z0;GF under smoother terrain conditions.
As pointed out by Verkaik and Holtslag (2007), the largest eddies of at-
mosphere are those that adapt slowest of all to terrain changes. Because
σ u in the surface layer is mostly determined by the largest eddies, z0;σu is
expected to better represent the large-scale average of surface roughness.
By contrast, peak gust speed, and therefore GF, depends more signifi-
cantly on local surface properties. Thus, the discrepancies between z0;σu
and z0;GF reflect the rough-to-smooth transition in terrain around the
observation site in associated sectors (Fig. 4).
4.2. Results with marine exposure
The speed dependence of GFð3;600Þ and z0 estimated via the gustiness
method and the four marine methods that are respectively depicted by
Equations (19)–(21), (23) is examined in Fig. 10, based on the routine
measurements in Sector 270 –330 at R2E, R2C and R2W. Values of
inputting variables for Equations (19)–(21), (23) are set in accordance
with the information of the measurement systems at the study sites. The
results are stratified into 5 speed groups with 3 m/s intervals. The
number of data runs involved in each group is marked in the figure. For
z0 , measurements at the three sites are composited to enlarge the sample
size in strong-wind groups. Since the observation sites are located only
several meters away from coastal line, wind measurements may be
influenced non-negligibly by the topographic effects caused by the
Y.C. He et al. Journal of Wind Engineering & Industrial Aerodynamics 171 (2017) 121–136

Fig. 9. Measurements of wind variance ðσ u ; σ d Þ and z0;σ at R1E.

Fig. 10. Speed-dependence of GF and marine z0 based on routine measurements.

groyne. Using the method stated in European Standard for wind actions
on structures (EN, 1991-1-4, 2005; Ngo and Letchford, 2008, 2009), the
orography factor for mean wind speed at the observation sites is esti-
mated as co;U  1:09 which means a decrease by 8.3% in measured GF if
topographic effects on peak gust are neglected. If the orography factor for
peak gust is assumed to be co;bu  1:04, there will be a 4.6% decrease in
measured GFs. The results of z0;GF based on the raw dataset (medians
depicted by blue squares) and the records after corrections by consid-
ering only co;U (black circles) and both co;U and co;bu , or co;GF ¼ co;bu =co;U 
1:05 (pink triangles), are presented in the figure.
As demonstrated, the estimation results are extremely sensitive to the
topographic effects. While values of z0;GF based on the raw dataset at this
coastal area are unreasonably (almost one order of magnitude) smaller
than those predicted by the marine roughness models that were estab-
lished based on observations over open oceans, those from corrected GF
measurements by using co;U are several times larger than the model
predictions. But the results after correction for the topographic effects on
both mean speed and peak gust show good agreement with the
predictions from the marine roughness models. The above findings sug-
gest that the adopted gustiness method is able to provide reasonable
estimations of z0 under marine exposure conditions, although the
method is conventionally not utilized in marine environment.
The results based on the fine records at the three sites in Sector
270 –330 are listed in Table 2. Limited by sample size, no classification
of speed group is considered herein. Values of σ u =U, σ d and GF estimated
both with (“w”) and without (“wo”) correction of the topographic effects
on wind measurements are documented. Following the previous discus-
sions, GF is corrected by using co;GF , while σ u is corrected via
co;σu  1 þ ðco;bu  1Þ=g  1:01, with g  3 being the peak factor. Thus,
the orography factor for σ u =U turns to be co;TI  co;σu =co;U  0:93. The
orography factor for σ d is assumed as the same as co;TI .
Similar to the results based on the routine measurements, the medians
of z0;GF without correction for the topographic effects are about one order
of magnitude smaller than those predicted by Equation (23), while the
values based on the corrected measurements agree well with the model
predictions. Values of z0;σu and z0;σd become almost doubled after the

Table 2
Estimations of marine z0 via varied methods based on fine records.

Site Sector 270 –330 25th, 50th, 75th percentiles of turbulence 25th, 50th, 75th percentiles of z0 z0 estimated via marine methods
measurements estimations

No. of runs U (m/s) θ ( ) σu/U σd (wo/w) GF z0,σu z0,σd z0,GF z0,Eq19 z0,Eq20 z0,Eq21 z0,Eq23
(wo/w) (rad) (wo/w) (mm) (mm) (mm) (mm) (mm) (mm) (mm)

R2E 64 10.8 306 0.078/.084 0.070/.065 1.18/1.24 0.48/1.0 0.24/.51 0.001/.062 0.19 0.13 0.48 0.35
0.091/.098 0.080/.074 1.21/1.27 2.0/3.7 0.91/1.8 0.011/.28
0.096/.104 0.097/.090 1.26/1.32 3.1/5.6 4.7/8.0 0.18/1.9
R2C 60 11.0 306 0.082/.089 0.069/.064 1.19/1.25 0.78/1.6 0.21/.44 0.003/.011 0.20 0.14 0.50 0.38
0.091/.098 0.080/.074 1.23/1.29 2.0/3.7 0.91/1.8 0.037/.66
0.104/.112 0.094/.087 1.25/1.31 5.7/10 3.7/6.4 0.11/1.4
R2W 72 10.8 309 0.085/.092 0.070/.065 1.20/1.26 1.1/2.1 0.24/.51 0.005/.18 0.19 0.13 0.48 0.35
0.094/.102 0.082/.076 1.23/1.29 2.6/4.8 1.1/2.2 0.037/.66
0.106/.114 0.103/.095 1.28/1.34 6.6/11 7.3/12 0.42/3.6

130
 Y.C. He et al.
Table 3
Measurement results along airport corridors.

Site Sector-1 Sector-2

(50 –80 at R1E, R1C, R1W; 70 –100 at R2E, R2C, R2W) (240 –270 at A1E, R1C, R1W; 220 –250 at R2E, R2C, R2W)

No. U (m/s) 25th, 50th, 75th percentiles of turbulence 25th, 50th, 75th percentiles of z0 No. of U (m/s) 25th, 50th, 75th percentiles 25th, 50th, 75th percentiles of z0 estimation
of runs & θ( ) measurements estimation runs & θ( ) of turbulence measurements

σu/U σd GF z0,σu z0,σd z0,GF σu/U σd GF z0,σu z0,σd z0,GF

(rad) (mm) (mm) (mm) (rad) (mm) (mm) (mm)

R1E F 245 9.7 0.119 0.110 1.31 14 11 1.3 93 10 0.100 0.091 1.25 4.4 2.8 0.10
67 0.132 0.128 1.36 28 30 5.3 250 0.114 0.106 1.30 11 8.6 1.1
0.150 0.154 1.43 57 79 23 0.130 0.130 1.36 25 33 5.6
R 5685 7.6 – – 1.25 – – 0.13 2430 7.2 – – 1.27 – – 0.33
73 1.30 1.2 250 1.32 1.9
1.38 9.2 1.38 10
A1E R 5352 7.2 – – 1.32 – – 1.8 1361 7.0 – – 1.27 – – 0.33
71 1.37 7.6 252 1.32 1.8
1.45 33 1.38 10
R1C F 386 9.4 0.121 0.110 1.31 16 11 1.4 49 11 0.106 0.095 1.26 6.5 4.0 0.17
66 0.135 0.131 1.37 32 35 6.7 254 0.117 0.121 1.29 13 21 0.75
0.150 0.156 1.44 57 86 28 0.135 0.146 1.40 31 61 12
R 14,956 7.7 – – 1.26 – – 0.24 1619 7.0 – – 1.25 – – 0.18
71 1.31 1.6 250 1.30 1.3
131

1.38 8.6 1.37 7.7

R1W F 400 9.7 0.107 0.096 1.28 7.2 4.1 0.37 54 11 0.091 0.074 1.21 2.0 0.40 0.05
66 0.120 0.111 1.33 16 12 2.6 251 0.105 0.087 1.25 6.3 1.9 0.12
0.136 0.129 1.40 34 31 15 0.132 0.111 1.30 28 12 0.84
R 12,749 7.7 – – 1.25 – – 0.16 1551 7.0 – – 1.20 – – 0.011
70 1.30 1.2 248 1.24 0.12
1.37 7.9 1.30 1.5

Journal of Wind Engineering & Industrial Aerodynamics 171 (2017) 121–136

R2E F 318 12 0.124 0.094 1.30 19 3.5 0.83 405 11 0.108 0.096 1.27 7.7 4.1 0.29
87 0.136 0.104 1.37 33 7.8 3.8 229 0.120 0.108 1.32 15 10 1.8
0.150 0.119 1.40 57 19 15 0.137 0.123 1.38 35 24 8.2
R 20,385 7.7 – – 1.25 – – 0.18 6504 8.0 – – 1.29 – – 0.70
93 1.30 1.0 231 1.35 4.3
1.35 4.8 1.42 21
R2C F 386 12 0.124 0.090 1.30 19 2.6 1.1 441 11 0.103 0.087 1.26 5.5 1.9 0.14
86 0.137 0.101 1.35 34 6.3 4.2 229 0.114 0.097 1.29 11 4.7 0.62
0.151 0.116 1.41 58 17 17 0.129 0.110 1.35 25 11 3.8
R 24,106 7.9 – – 1.27 – – 0.31 9795 7.9 – – 1.24 – – 0.11
92 1.31 1.5 232 1.30 0.87
1.37 6.7 1.36 5.8
R2W F 383 12 0.126 0.092 1.31 21 3.0 1.1 321 11 0.092 0.069 1.22 2.1 0.22 0.014
86 0.139 0.102 1.36 37 6.9 4.9 227 0.105 0.080 1.26 6.1 0.93 0.13
0.151 0.114 1.42 58 15 18 0.119 0.092 1.31 14 3.0 1.2
R 32,008 8.1 – – 1.28 – – 0.50 7927 7.7 – – 1.20 – – 0.012
90 1.33 2.4 228 1.25 0.12
1.38 10 1.30 0.96
Y.C. He et al.
correction process, but both kinds of results are several times larger than
those predicted by Equation (23). As the best estimation of local marine
roughness is unavailable, the validity and prediction accuracy of the
variance method need to be further examined.
4.3. Results along airport corridors
Both routine measurements (“R”) and fine records (“F”) in two sectors
at all the seven study sites are analyzed to investigate turbulent wind
characteristics and roughness length along the two airport corridors. The
results are summarized in Table 3. For R1E, R1C and R1W (i.e., sites
along the south runway), the two azimuth sectors are 50 –80 and
240 –270 , while for R2E, R2C and R2W (i.e., sites along the north
runway), the two azimuth sectors are 70 –100 and 220 –250 .
These results reveal the following features:
(1) Values of z0;GF are consistently smaller than those estimated via
the variance method. The reasons are twofold. First, wind-
variance measurements carry more information of large-scale
eddies. Thus, wind indexed by σ appears to be gustier than that
indexed by GF. Second, as to be discussed later, the estimation
results tend to be more sensitive to model coefficients under
smooth terrain condition, and significantly increased un-
certainties may also account for the distinct discrepancy.
(2) Estimations of z0;GF based on the fine records are noticeably larger
than those derived via the routine measurements in Sector-1 (with
hilly exposure), but the difference tends to decrease with the
distance downwind of the topography, which respectively sug-
gests the dependence of GF and z0;GF on wind strength and hill-
relative position. However, values of z0;GF estimated through the
two kinds of datasets in Sector-2 (with open flat exposure) differ
insignificantly. Meanwhile, estimations of z0 via different
methods in Sector-2 are consistently smaller than their counter-
parts in Sector-1. The above observations reveal the topographic
effects on the obtained results.
(3) Results of z0;GF in Section-2 reveal a marine-to-land terrain change
along each of the two runways. The medians of z0;GF increase
gradually from ~0.1 mm at the western part, to ~0.8 mm at the
middle part, and then to ~2 mm at eastern part, reflecting the
progressively strengthened influence of local terrain on the mea-
surement results. If the local terrain is assumed to belong to Class-
2 (z0 ¼ 5 mm) of the terrain classification in Table 1, the IBL
depths at these three parts turn to be 70 m, 140 m, and 220 m,
respectively. It is clear that the observation points at R1W and
R2W are located in the blending range of the IBL where atmo-
spheres are influenced by both the marine exposure and the local
terrain, whilst those at R1E and R2E are located inside the lowest
one-tenth portion of the IBL where wind structures are in equi-
librium with the local surface. Similar results are also observed
along the north runway in Section-1.
Fig. 11. Evolutions of GF in Sector 50 –80 at two sites.
132
Journal of Wind Engineering & Industrial Aerodynamics 171 (2017) 121–136
From Fig. 4, there is an artificial island located to the northeast of
HKIA. This island was constructed during 2012–2014. Based on the
above discussions, wind flows at the observation heights above R1E and
A1E in Sector-1 experienced a marine-to-land change in terrain before
the appearance of the island, while they should undergo multi-step
changes in terrain afterward. To examine the possible effects of the
artificial island on the measurements, Fig. 11 plots the evolutions of
GFð3;600Þ in Sector-1 at both A1E and R1E (“Aver-1” represents GFð3;600Þ
for each data run, “Aver-50” stands for mean values of every 50 runs,
while “Aver-all” is the ensemble-mean value). As demonstrated, no
evident variation trend can be detected from time history of GFð3;600Þ .
Actually, the above issue featured by multi-step changes in upwind ter-
rains may be simplified according to the recommendations in ESDU
(1982) which suggests that the intermediate island-occupied area inside
the marine region can be ignored since the span of the island is less than
that of the marine area between the island and HKIA.
4.4. Results downwind of obstacle zones
To examine turbulent wind characteristics and surface roughness in
areas downwind of some typical obstacle zones in HKIA, field mea-
surements in two azimuth sectors are selected: Sector-1, i.e., Sector
330 –360 at R1E, A1E and R1W and 0 -30 at R1C, and Sector-2, i.e.,
Sector 150 –180 at R2E and R2W and Sector 120 –150 at R2C. From
Fig. 4, Sector-1 at R1W and Sector-2 at R2W are associated with an
upwind region occupied by scattered buildings and parking aprons.
Sector-1 at R1E and A1E and Sector-2 at R2E are associated with the
terminal building, whilst Sector-1 at R1C and Sector-2 at R2C corre-
spond to the midfield concourse building. Here, measurements in two
sectors around each of the three obstacle zones are analyzed to offer
insights for the validity of the obtained results. Measurements at R1E
and A1E in Sector-1 may further manifest the dependence of wind
turbulence on the distance downwind of the terminal building through
comparison analysis.
The obtained results are summarized in Table 4. Basically, the results
in Sector-1 agree well with their counterparts in Sector-2, taking into
account the topographic effects and other uncertain factors. In most
cases, median of z0 is in the order of 0.2 m. Again, z0;GF derived from the
fine records is found to be typically larger than that based on the routine
measurements. As mentioned previously, turbulent wind structure
downwind of a large obstacle depends greatly on associated Reynolds
number, or incident wind strength and characteristic dimension of the
obstacle. On the one hand, GF results, obtained particularly downwind of
the terminal building, reflect that stronger winds are prone to result in
more turbulent flows at the observation sites. On the other hand, com-
parison of the results at R1E with those at A1E reveals the distance
dependence of GF downwind of the terminal building. The above find-
ings suggest that the disturbing influence of big airport buildings on
departing/arriving aircrafts should be investigated in terms of wind
strength and building-relative position.
 Y.C. He et al.
Table 4
Measurement results downwind of obstacle zones.

Site Sector-1 Sector-2

(330 –360 at R1E, A1E, R1W; 0 –30 at R1C) (150 –180 at R2E, R2W; 120 –150 at R2C)

No. of runs U (m/s) & θ( ) 25th, 50th, 75th percentiles 25th, 50th, 75th percentiles No. of runs U (m/s) & θ( ) 25th, 50th, 75th 25th, 50th, 75th percentiles of z0 estimation
of turbulence measurements of z0 estimation percentiles of turbulence
measurements

σu/U σd GF z0,σu z0,σd z0,GF σu/U σd GF z0,σu z0,σd z0,GF

(rad) (m) (m) (m) (rad) (m) (m) (m)

R1E or R2E F 125 9.3 0.160 0.161 1.45 0.079 0.099 0.036 200 8.5 0.170 0.175 1.49 0.10 0.14 0.059
345 0.181 0.185 1.53 0.14 0.18 0.093 166 0.187 0.199 1.60 0.16 0.24 0.18
0.198 0.211 1.62 0.20 0.30 0.22 0.216 0.231 1.70 0.27 0.40 0.38
R 859 7.2 – – 1.37 – – 0.008 1614 7.2 – – 1.41 – – 0.016
343 1.44 0.028 167 1.52 0.089
133

1.52 0.087 1.66 0.29

A1E R 797 7.4 – – 1.35 – – 0.004 – – – – – – – –
344 1.40 0.015
1.49 0.058
R1C or R2C F 238 7.4 0.154 0.152 1.46 0.065 0.074 0.038 356 10.0 0.146 0.148 1.39 0.049 0.065 0.011
16 0.172 0.181 1.57 0.11 0.16 0.14 131 0.162 0.166 1.46 0.083 0.11 0.037

Journal of Wind Engineering & Industrial Aerodynamics 171 (2017) 121–136

0.194 0.215 1.72 0.18 0.31 0.43 0.184 0.195 1.54 0.15 0.22 0.11
R 1011 6.9 – – 1.39 – – 0.012 4616 7.6 – – 1.31 – – 0.0014
17 1.47 0.048 131 1.37 0.0071
1.59 0.17 1.45 0.032
R1W or R2W F 91 9.8 0.176 0.199 1.49 0.12 0.24 0.061 165 9.6 0.166 0.169 1.47 0.092 0.12 0.043
346 0.203 0.213 1.61 0.22 0.31 0.20 163 0.184 0.193 1.55 0.15 0.21 0.11
0.221 0.230 1.70 0.30 0.39 0.37 0.207 0.223 1.65 0.24 0.35 0.26
R 977 7.1 – – 1.34 – – 0.004 1055 8.1 – – 1.47 – – 0.047
341 1.41 0.018 166 1.56 0.14
1.52 0.082 1.68 0.33
Y.C. He et al. Journal of Wind Engineering & Industrial Aerodynamics 171 (2017) 121–136

measurements to extract records with evident nonneutral/nonstationary

features, the thermal effects and unanticipated noise interference may
still exist in some “qualified” specimens. From Fig. 8, the GF values in
Sector 60 –120 at A1E and R1E demonstrate a decreasing trend with the
increase of wind speed for moderate-to-strong winds, which is usually
regarded as a feature relevant to the thermal effect (Shu et al., 2015).
Meanwhile, field measurements stratified in a certain azimuth sector may
contain records influenced by different kinds of exposure, although great
attention has been made during the stratification process. The way of
azimuthal stratification may result in evident uncertainty in estimations
of z0;p . It is shown that estimations of z0;p in Sector 240 –270 and Sector
270 –300 based on the lower portion of wind profiles (below 50 m) are
0.2 m and 0.02 m, respectively. However, if stricter criteria for data
quality control and/or finer group intervals for data stratification are
adopted, there would be insufficient specimens left in the concerned
sectors, due to the characteristics of regional climate and local topo-
Fig. 12. Roughness map at HKIA. graphic features.
Second, the results from both the fine records and the routine
4.5. Uncertainty analysis measurments demonstrate that the mean GF values are typically 1%–3%
larger than associated medians. Accordingly, z0;GF estimated via mean of
The uncertainty sources can be classified into two groups which are GF is larger than that estimated via median of GF by a factor of 0.2–1. For
related to z0 estimation model and dataset, respectively. Model-related σ u and σ d , the mean value generally differs from the median by a factor of
uncertainty consists of potential errors due to inaccurate value assign- 1%–5%, which is associated with a difference between z0;p and z0;σu by a
ment of some key model parameters, while data-related uncertainty is factor of 0.2–0.8. Such uncertainties related to adopted statistical pa-
largely owing to the way of data selection/sratification and data analysis rameters have been discussed by Lombardo and Krupar (2017).
(such as fitting process). Third, the estimation of z0;p is sensitive to the portion of adopted wind
From Figs. 1–2, the model predictions via either the varance or profiles for data fitting. Different observation heights correspond to
gustiness methods are considerably sensitive to a number of parameters, varied source areas. Thus, narrow range of wind profiles is preferred
especially under smooth terrain condition. From Fig. 1, given σ u =U ¼ during fitting process from the footprint point of view. However, nar-
0:10 at z ¼ 10 m which is consistent with the measurement results with a rowing the fitting range will decrease the number of gate records, and
marine exposure as listed in Table 2, z0;σu computed using different therefore, the accuracy of fitting results. It is suggested by Wieringa
values of cu can vary from 0.4 mm ðcu ¼ 2:5Þ to 6 mm ðcu ¼ 1:9Þ and to (1993) that the least profile levels needed for determination of z0;p within
300 mm ðcu ¼ 1:5Þ. Actually, for simplification purpose, it is sometimes a factor of 2 are 3 for high roughness (z0 ¼ 1 m), 4 for moderate
assumed that cu ¼ 2:5 (thus κ⋅cu  1 in Equation (5)) in practice even for roughness (z0 ¼ 0:1 m), and 5 or more over really smooth terrain.
filtered measurements (cu should be equal to 1.94). The above simplifi-
cation may result in an estimation error up to an order of magnitude in
z0;σu . For the gustiness estimation method, the parameters of gust period, 4.6. Roughness map at HKIA
average wind duration, observation height and distance constant can
affect the estimation results significantly (Fig. 2). Among these parame- A recommended roughness map at HKIA is shown in Fig. 12. The
ters, the distance constant λ, which serves as an index for the dynamic marine roughness length is ~0.5 mm based on the results listed in Table 2
features of an anemometer (Kristensen and Hansen, 2002), may not be which is associated with a fixed incident wind speed of 11 m/s. This
determined accurately. It is also possible that λ varied with historical value is slightly smaller than 0.2 mm for “open-sea” terrain suggested in
service/maintain conditions of the anemometer system. From Fig. 2, the terrain classification (Table 1). But it is identical to that recom-
given GFð3;600Þ (based on the field measurements in Table 2), z0;GF mended by Harper et al. (2010) for the “at-sea” exposure class for tropical
calculated by using different λ values varies from 0.17 mm ðλ ¼ 0Þ to cyclone forecasting purposes. For the runway corridors, the preferred
0.40 mm ðλ ¼ 10Þ and to 1.0 mm ðλ ¼ 20Þ. For smaller GF values, the estimation of z0 is 5 mm, based on a comprehensive consideration of the
difference between estimations of z0;GF via varied λ values should be even measured results listed in Table 3. This value corresponds to the
more significant. Other parameters involved in the estimation model for “smooth” terrain class in Table 1.
z0;GF , such as spectrum of wind turbulence and turbulence integral length The area to the south of HKIA and the western part of the region
scale, may also bring in non-negligible uncertainties for estimation re- between the two runway corridors are dominated by scatter obstacles,
sults especially under rugged-exposure conditions. such as buildings and berthed airplanes, and z0 for both areas is sug-
For data-related uncertainties, three aspects are stressed herein. First, gested to be 0.3 m, based on the results in Table 4. This value is associ-
although a series of data quality tests have been conducted on raw ated with the “rough” landscape class as suggested in the terrain
classification. The effective roughness length for the open area to the
Table 5 west of the midfield concourse building is estimated as 0.03 m, based on
Comparison of z0 results at HKIA from different data sources. the measured results at R1C in Sector 300 –330 . This value corresponds
Surface Type z0 (m)
to an “open” class in the terrain classification. Results around the ter-
minal building and the midfield concourse building are believed to
Liu et al. Leung et al. Li and Chan This study
depend greatly on building-relative location and wind strength. Thus, no
(2010) (2012) (2012, 2016)
specific values of z0 are presented in Fig. 12.
Sea 0.0002 0.015 0.0005 0.0005
By considering the above results, the large-scale average of z0 at A1E
Asphalt 0.0024 0.0015 – –
Runway 0.004 0.00015 – 0.005
in the western sector is calculated as 0.04 m via the lnðz0 Þ average
Building wall 0.02 0.0015 – – method. This value approximates to the measured z0;p (0.05 m) based on
Short grass 0.01 0.015 – – the lower portion of wind profile as shown in Fig. 6. This result is also
Airport island – – 0.005 0.05 consistent to the conventional assumption that an airport area belongs to
Lantau Island – – 0.5 0.5
an “open” terrain with z0  0:03 m.

134
Y.C. He et al.
Table 5 compares the values of z0 obtained in this study with
those assumed in four computational studies on wind characteristics at
near ground level around HKIA. Estimations of z0 for “Lantau Island” in
this study is derived based on the measurement results in Sector
120 –210 at sites along the south runway. The measured median of z0
from the fine records in this sector is in a range of 0.2–0.5 m, compared to
that of 0.1–0.3 m based on the routine measurements. Taking into ac-
count the mountain-relative effect, z0 for “Lantau Island” is adjusted to be
0.5 m which corresponds to the “very rough” terrain in Table 1. It is
emphasized that the above z0 value should not be used to determine the
vertical profile of horizontal wind speed, since the vertical wind profiles
can be distorted severely downwind of hills/mountains.
In light of the fact that the runway corridors consist mainly of asphalt
roads and low grass, values of z0 for “asphalt” and “short grass” may be
estimated via the classification method by further considering the
measured results. According to the terrain classification in Table 1,
“tarmac and concrete with a free fetch of several kilometers” belongs to
Class-1 (z0 ¼ 0:0002 m), while “low vegetation, e.g., grass” belongs to
Class-3 (z0 ¼ 0:03 m). Thus, it is reasonable to assume that the roughness
values for these two terrain types at HKIA are 0:01  0:02 (“short grass”)
and 0.002 m (“asphalt”). It is interesting that the effective z0 for the
runway corridors is calculated to be 0.0055 m using the lnðz0 Þ average
method by assuming z0 ¼ 0:015 m and z0 ¼ 0:002 m for “short grass”
and “asphalt”. This value agrees well with the recommended value of
0.005 m in Table 1.
5. Concluding remarks
Aerodynamic roughness length is an essential parameter for the
parameterization of surface wind. It plays an important role in many
aspects of engineering practice. Thus, accurately estimating z0 is of great
practical significance. However, limited comparative studies have been
reported to examine the validity and estimation accuracy of various
micrometeorological methods, possibly due to unavailability of sufficient
field measurements of winds.
Based on long-term wind measurements from both anemometers
and a Sodar wind profiler system at seven meteorological stations at
KHIA, this paper investigated the characteristics of surface wind and
directional roughness length at HKIA. Several estimation techniques for
z0 , including the profile method, variance method for both wind speed
and wind direction, the gustiness method, classification method, and
four marine methods were adopted in this study. The validity of
collected field measurements and the estimation performances of these
prediction techniques for z0 were examined through comprehensive
comparison analyses. The concept of source area and the IBL theory
were adopted to explain the obtained results. A roughness map at HKIA
was finally drawn. Main findings and conclusions of this study are
summarized below.
(1) Under flat exposure conditions, the results of GFð3;600Þ and z0;GF
derived from the long-term routine measurements agree well with
those from the fine records collected during a much shorter
period. But the two kinds of results differ evidently in sectors with
rugged exposures, owing to the speed-dependence of wind char-
acteristics downwind of hills. Atmospheres in stronger wind
events are consistently “gustier”.
(2) Basically, the results of σ u =U and z0;σu agree well with their
counterparts of σ d and z0;σd , suggesting the validity of wind vari-
ance measurements and the rationality of using wind variance
records to estimate roughness length.
(3) Under heterogeneous terrain conditions, surface wind flows are in
non-equilibrium status. Winds at lower heights are more influ-
enced by local surface cover than those at higher altitudes. At the
observation height, the results reflect that wind variance accounts
for a larger source area than that of GF, and z0;GF tends to be
135
Journal of Wind Engineering & Industrial Aerodynamics 171 (2017) 121–136
smaller than z0;σu or z0;σd in case with a rough-to-smooth terrain
transition.
(4) The results of vertical wind profiles fail to provide the information
on z0;p under rugged terrain conditions, as the profiles are dis-
torted severely by nearby topography. Under flat exposure con-
ditions, estimation of z0;p depends greatly on the fitting range of
wind profile, due to the discrepancy of source area. The z0;p value
derived on the basis of lower portion of wind profile is consistent
with the effective roughness length estimated via the wind tur-
bulence measurements.
(5) Wind measurements collected at sites located close to coastal lines
may be influenced by the topographic effects of groyne. The re-
sults after correction for such topographic effects show that the
roughness length estimated via the gustiness method is compa-
rable to the predictions from the four marine roughness models.
However, values of z0;σ are noticeably larger than those predicted
by the models.
(6) Wind structures downwind of huge obstacles at HKIA, e.g., ter-
minal building, are influenced by both incident wind strength and
observation position. Thus, case studies via either numerical
simulations or wind tunnel tests are needed to analyze the wind
characteristics in the wake region of such obstacles.
(7) There are a number of uncertain factors involved in the obtained
z0 results. Of particular importance are accurate determination of
key coefficients involved in each estimation model for z0 and
reasonable selection and sectionalization of datasets. The model
coefficients may not be simplified arbitrarily under smooth terrain
conditions, since the estimation results become considerably
sensitive to these variables, such as cu , for small z0 . It is also
suggested that medians of GF or σ should be adopted for analysis
of terrain roughness.
(8) A roughness map at HKIA was established. Values of z0 obtained
in this study are compared with those assumed in previous
simulation works. The results of this study provide useful infor-
mation for further numerical studies to advance wind shear
alerting for aircraft operation at HKIA.
In light of the important role of z0 in governing the ABL structure, it is
essential to provide accurate estimation of z0 to avoid excessive ampli-
fication of z0 -related estimation uncertainties and errors due to potential
propagations in subsequent analysis. Therefore, there is a need to
perform further studies to examine the validity and accuracy of different
micrometeorological estimation methods for roughness length under
different terrain conditions.
Acknowledgment
This work is fully supported by a grant from the National Natural
Science Foundation of China (Project No.: 51408520), a grant from the
Research Grants Council of Hong Kong Special Administrative Region,
China (Project No: CityU 118213), a grant from Environment and Con-
servation Fund of Hong Kong (Project No: 9211097 (19/2015)) and a
grant from Fundamental Research Program of Shenzhen Municipality
(Project No. JCYJ20160401100137854).
References
Balderrama, J.A., Masters, F.J., Gurley, K.R., Prevatt, D.O., Aponte-Bermudez, L.D.,
Reinhold, T.A., Pinelli, J.P., Subramanian, C.S., Schiff, S.D., Chowdhury, A.G., 2011.
The Florida coastal monitoring program (FCMP): a review. J. Wind Eng. Ind.
Aerodyn. 99, 979–995.
Balderrama, J.A., Masters, F.J., Gurley, K.R., 2012. Peak factor estimation in hurricane
surface winds. J. Wind Eng. Ind. Aerodyn. 102, 1–13.
Barthelmie, R.J., Palutikof, J.P., Davies, T.D., 1993. Estimation of sector roughness
lengths and the effect of prediction of the vertical wind speed profile. Boundary-Layer
Meteorol. 66, 19–47.
Beljaars, A.C.M., 1987. The influence of sampling and filtering on measured wind gusts.
J. Atmos. Ocean. Technol. 4, 613–626.
Y.C. He et al.
Blocken, B., Stathopoulos, T., Carmeliet, J., 2007. CFD simulation of the atmospheric
boundary layer: wall function problems. Atmos. Environ. 41, 238–252.
Chan, P.W., 2008. Measurement of turbulence intensity profile by a mini-sodar. Meteorol.
Appl. 15, 249–258.
Chan, P.W., 2012a. A significant wind shear event leading to aircraft diversion at the
Hong Kong international airport. Meteorol. Appl. 19, 10–16.
Chan, P.W., 2012b. Observation and numerical simulation of an event of vortex/wave
shedding from a mountain near the Hong Kong International Airport. Meteorol. Appl.
19, 371–384.
Chan, P.W., Lee, Y.F., 2011. Application of a ground-based, multi-channel microwave
radiometer to the alerting of low-level windshear at an airport. Meteorol. Z. 20,
423–429.
Charnock, H., 1955. Wind stress on a water surface. Q. J. R. Meteorol. Soc. 81, 639–640.
Davenport, A.G., 1960. Rationale for determining design wind velocities. J. Struct. Div.
Am. Soc. Civ. Eng. 86, 39–68.
Davenport, A.G., 1964. Note on the distribution of the largest value of a random function
with application to gust loading. Proc. Instit. Civ. Eng. 28, 187–196.
Davenport, A.G., Grimmond, C.S., Oke, T.R., Wieringa, J., 2000. Estimating the roughness
of cities and sheltered country. In: Preprints of the Twelfth American Meteorological
Society Conference on Applied Climatology (Asheville, NC, United States), pp. 96–99.
Emeis, S., 2014. Current issues in wind energy meteorology. Meteorol. Appl. 21,
803–819.
EN 1991-1-4, 2005. Eurocode 1: Actions on Structures – Part 1-4: General Actions – Wind
Actions. The European Union Per Regulation 305/2011, Directive 98/34/EC,
Directive 2004/18/EC.
ESDU, 1982. Strong winds in the Atmospheric Boundary Layer Part 1: Hourly-mean Wind
Speeds. Engineering Sciences Data Unit 82026, p. 61.
ESDU, 1983. Strong winds in the Atmospheric Boundary Layer Part 2: Discrete Gust
Speeds. Engineering Sciences Data Unit 83045, p. 37.
ESDU, 1985. Characteristics of Atmospheric Turbulence Near the Ground Part Π: Single
Point Data for Strong Winds. Engineering Sciences Data Unit 85020, p. 42.
Fairall, C.W., Bradley, E.E., Hare, J.E., Grachev, A.A., Edson, J.B., 2003. Bulk
parameterization of sir-sea fluxes: updates and verification for the COARE algorithm.
J. Clim. 16, 571–591.
Fielder, F., Panofsky, G., 1972. The geostrophic drag coefficient and the effective
roughness length. Q. J. R. Meteorol. Soc. 98, 213–220.
Garratt, J.R., 1977. Review of drag coefficients over oceans and continents. Mon. Weather
Rev. 105, 915–929.
Garratt, J.R., 1990. The internal boundary layer-a review. Boundary-Layer Meteorol. 50,
171–203.
Garratt, J.R., 1994. The Atmospheric Boundary Layer. Cambridge University Press.
Grimmond, C.S.B., Oke, T.R., 1999. Aerodynamic properties of urban areas derived from
analysis of surface from. J. Appl. Meteorol. 38, 1262–1292.
Harper, B.A., Kepert, J.D., Ginger, J.D., 2010. Guidelines for Converting between Various
Wind Averaging Periods in Tropical Cyclone Conditions. World Meteorological
Organization WMO/TD-No. 1555.
He, Y.C., Chan, P.W., Li, Q.S., 2013. Wind characteristics over different terrains. J. Wind
Eng. Ind. Aerodyn. 131, 51–69.
He, Y.C., Chan, P.W., Li, Q.S., 2014a. Standardization of raw wind speed data under
complex terrain conditions: a data-driven scheme. J. Wind Eng. Ind. Aerodyn. 131,
12–30.
He, Y.C., Chan, P.W., Li, Q.S., 2014b. Field measurements of wind characteristics over
hilly terrain within surface layer. Wind Struct. 19, 541–563.
He, Y.C., Chan, P.W., Li, Q.S., 2016. Standardization of offshore surface wind speeds.
J. Appl. Meteorol. Climatol. AMS 55, 1107–1121.
Irwin, P.A., 2006. Exposure categories and transitions for design wind loads. J. Struct.
Eng. ASCE 132, 1755–1763.
Kareem, A., Zhao, J., 1994. Analysis of non-Gaussian surge response of tension leg
platforms under wind loads. J. Offshore Mech. Arct. Eng. 116, 137–144.
Kristensen, L., Hansen, O.F., 2002. Distance Constant of the RisØ Cup Anemometer.
Technical Report RisØ-R-1320(EN). RisØ National Laboratory, Roskilde, Denmark.
Lettau, H., 1969. Note on aerodynamic roughness-parameter estimation on the basis of
roughness-element description. J. Appl. Meteorol. 8, 828–832.
Leung, D.Y.C., Lo, W.Y., Chow, W.Y., Chan, P.W., 2012. Effect of terrain and building
structures on the airflow in an airport. J. Zhejiang Univ. Sci. A 13, 461–468.
Li, L., Chan, P.W., 2012. Numerical simulation study of the effect of buildings and
complex terrain on the low-level winds at an airport in typhoon situation. Meteorol.
Z. 21, 183–192.
136
Journal of Wind Engineering & Industrial Aerodynamics 171 (2017) 121–136
Li, L., Chan, P.W., 2016. LIDAR observation and numerical simulation of vortex/wave
shedding at the eastern runway corridor of the Hong Kong International Airport.
Meteorol. Appl. 23, 379–388.
Liu, C.H., Leung, D.Y.C., Man, A.C.S., Chan, P.W., 2010. Computational fluid dynamics
simulation of the wind flow over an airport terminal building. J. Zhejiang University-
Science A 11, 389–401.
Lombardo, F.T., Krupar, R.J., 2017. Characterization and comparison of aerodynamic
roughness lengths using ground-based photography and sonic anemometry. J. Struct.
Eng. 04017049.
Mason, P.J., 1988. The formation of areally-averaged roughness lengths. Q. J. R.
Meteorol. Soc. 114, 399–420.
Masters, F.J., Vickery, P.J., Bacon, P., Rappaport, E.N., 2010a. Toward objective,
standardized intensity estimates from surface wind speed observations. Bull. Am.
Meteorol. Soc. 91, 1665–1681.
Masters, F.J., Tieleman, H.W., Balderrama, J.A., 2010b. Surface wind measurements in
three Gulf Coast hurricanes of 2005. J. Wind Eng. Ind. Aerodyn. 98, 533–547.
Meng, Y., Matsui, M., Hibi, K., 1995. An analytical model for simulation of the wind field
in a typhoon boundary layer. J. Wind Eng. Ind. Aerodyn. 56, 291–310.
Miller, C., Balderrama, J.A., Masters, F., 2015. Aspects of observed gust factors in
landfalling tropical cyclones: gust components, terrain, and upstream fetch effects.
Boundary-Layer Meteorol. 155, 129–155.
Ngo, T.T., Letchford, C.W., 2008. A comparison of topographic effects on gust wind speed.
J. Wind Eng. Ind. Aerodyn. 96, 2273–2293.
Ngo, T.T., Letchford, C.W., 2009. Experimental study of topographic effects on gust wind
speed. J. Wind Eng. Ind. Aerodyn. 97, 426–438.
Oost, W.A., Komen, G.J., Jacobs, C.M., Oort, C.V., 2002. New evidence for a relation
between wind stress and wave age from measurements during ASGAMAGE.
Boundary-Layer Meteorol. 103, 409–438.
Powell, M.D., Houston, S.M., 1996. Hurricane Andrew's landfall in south Florida. Part I:
standardizing measurements for documentation of surface wind fields. Weather
Forecast. 11, 304–328.
Powell, M.D., Vickery, P.J., Reinhold, T.A., 2003. Reduced drag coefficient for high wind
speeds in tropical cyclones. Nature 422, 279–283.
Schmid, H.P., Oke, T.R., 1990. A model to estimate the source area contributing to
turbulent exchange in the surface layer over patchy terrain. Q. J. R. Meteorol. Soc.
116, 965–988.
Shu, Z.R., Li, Q.S., He, Y.C., Chan, P.W., 2015. Gust factor for tropical cyclone, monsoon
and thunderstorm winds. J. Wind Eng. Ind. Aerodyn. 142, 1–14.
Smith, S.D., Anderson, R.J., Oost, W.A., Kraan, C., Maat, N., Decosmo, J., Katsaros, K.B.,
Davidson, K.L., Bumke, K., Hasse, L., Chadwick, H.M., 1992. Sea surface wind stress
and drag coefficients: the HEXOS results. Boundary-Layer Meteorol. 60, 109–142.
Stewart, I.D., Oke, T.R., 2012. Local climate zones for urban temperature studies. Bull.
Am. Meteorol. Soc. 93, 1879–1900.
Taylor, P.K., Yelland, M.J., 2001. The dependence of sea surface roughness on the height
and steepness of the waves. J. Phys. Oceanogr. 31, 572–590.
Verkaik, J.W., 2000. Evaluation of two gustiness models for exposure correction
calculations. J. Appl. Meteorol. 39, 1613–1626.
Verkaik, J.W., Holtslag, A.A.M., 2007. Wind profiles, momentum fluxes and roughness
lengths at Cabauw revisited. Boundary-Layer Meteorol. 122, 701–719.
Vickery, P.J., Skerlj, P.F., 2000. Elimination of exposure D along hurricane coastline in
ASCE7. J. Struct. Eng. 126, 545–549.
Vickery, P.J., Skerlj, P., 2005. Hurricane gust factors revisited. J. Struct. Eng. 131,
825–832.
Vickery, P.J., Wadhera, D., Powell, M.D., Chen, Y., 2009. A hurricane boundary layer and
wind field model for use in engineering applications. Appl. Meteorol. Climatol. 48,
381–405.
Wieringa, J., 1973. Gust factors over open water and built-up country. Boundary-layer
Meteorol. 3, 424–441.
Wieringa, J., 1992. Updating the Davenport roughness classification. J. Wind Eng. Ind.
Aerodyn. 41–44, 357–368.
Wieringa, J., 1993. Representative roughness parameters for homogeneous terrain.
Boundary-Layer Meteorol. 63, 323–363.
Wong, C.C.C., Liu, C.H., 2013. Pollutant plume dispersion in the atmospheric boundary
layer over idealized urban roughness. Boundary-Layer Meteorol. 147, 281–300.
World Meteorological Organization, 2008. Guide to Meteorological Instruments and
Methods of Observation. WMO-No.8, seventh ed., p. 681
Yamartino, R.J., 1984. A comparison of several ‘single-pass’ estimators of the standard
deviation of wind direction. J. Clim. Appl. Meteorol. 23, 1362–1366.