Aviation Safety Paper on Operations Under Meteorological Hazard with Early Detection

Predicting Low-Level Wind Shear Using 200-m-Resolution NWP at the Hong Kong International Airport

KAI-KWONG HON

Hong Kong Observatory, Hong Kong, China

(Manuscript received 1 August 2019, in final form 6 December 2019)

ABSTRACT

‘‘Low-level wind shear’’ is a known aviation safety hazard and refers to a sustained change in head wind

encountered by an aircraft during takeoff or landing. Because of their small spatiotemporal scales and high

variability, automatic alerting of wind shears at airports around the world is almost exclusively detection

based (using remote sensing equipment). Numerical modeling studies so far mainly cover individual cases and

lack systematic validation. This paper presents the first statistical evaluation of numerical weather prediction

(NWP) model performance in predicting low-level wind shear at a major international airport over a 2-yr

continuous period. The 200-m-resolution Aviation Model (AVM) of the Hong Kong Observatory is used to

generate runway-specific wind shear forecasts at 1-min output intervals for the Hong Kong International

Airport (HKIA), known for its susceptibility to wind shear occurrence. The AVM forecasts are then validated

against over 800 actual reports of wind shear by aircraft pilots over the two major arrival runway corridors,

07LA and 25RA, at HKIA between 2014 and 2015 using a verification scheme with the same level of

spatiotemporal stringency as operational alerting systems at HKIA. With ‘‘relative operating characteristic’’

analysis, positive skill is consistently observed across both runway corridors throughout the study period and

across all considered forecast lead times out to 6 h ahead. This study serves to establish and document the

current capability of fine-resolution NWP in predicting the phenomenon of low-level wind shear for aviation

weather applications.

1. Introduction

Low-level wind shear, defined as a sustained head

wind (i.e., winds blowing in the opposite direction to

aircraft motion) change of 15 kt (;7.7 m s21) or more at an altitude of 1600 ft (;488 m) or below, is a known aviation safety hazard (ICAO 2005, 2007). In aviation

meteorology, there are in fact a number of phenomena

related to spatial and/or temporal variations in the

three-dimensional wind fields that are of safety interest

or concern. For example, an aircraft during the level

phase of flight at cruising levels (normally ranging from

30 000 to 40 000 ft above sea level; 1000 ft ’ 0.3 km) may be susceptible to clear-air turbulence (CAT), which is

linked to shear and deformation under suitable stability

conditions (Ellrod and Knapp 1992; Ellrod and Knox

2010). Within the boundary layer (typically the lowest

1 or 2 km of the troposphere) there is also emphasis on

the vertical shear of the horizontal winds (ICAO 2005).

In this paper, we focus on the variation, or shear, of the

horizontal wind component along the aircraft trajectory,

specifically during the takeoff and landing phases of

flight (termed ‘‘low-level wind shear’’ or ‘‘wind shear’’ in

subsequent paragraphs; Hong Kong Observatory 2010).

Such wind shear is important from a safety point of view

because it is the head wind that provides aerodynamical

lift for an aircraft in motion. Rapid or large changes in

head wind can bring about significant changes in lift and

hence altitude of the aircraft. During the takeoff and

landing phases of flight, the aircraft normally have rel-

atively low airspeeds (e.g., when compared with the

speed at cruising levels) but are in close proximity to the

ground. This means that there is little room for correc-

tive measures that are based on the speed–altitude

tradeoff (Proctor et al. 2000) that is effective at higher

altitudes.

Wind shear may be caused by a variety of mesoscale

or microscale meteorological phenomena with differ-

ent physical mechanisms and flow regimes, including

Denotes content that is immediately available upon publication

as open access.

Corresponding author: Kai-Kwong Hon, [email protected]

VOLUME 59 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y FEBRUARY 2020

DOI: 10.1175/JAMC-D-19-0186.1

� 2020 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

193

thunderstorm downburst (or, in its most severe form,

known as microburst to the aviation community), which

possesses characteristics of a density current (Hall et al.

1976; Fujita 1978), terrain-disrupted airflow with or with-

out a temperature inversion (Chan and Hon 2016a,b),

frontal passage, tropical cyclones, land/sea-breeze cir-

culation, and so on (Lau 2000; Chan et al. 2010), all of

which are capable of inducing significant gradients in

the horizontal wind components. These wind shear–

inducing meteorological phenomena may occur with or

without precipitation.

The availability of dedicated, high-resolution remote

sensing instruments such as the Terminal Doppler

Weather Radar (TDWR, which works effectively in

rain) (Wilson and Wakimoto 2001) and the Doppler

light detection and ranging (lidar, which works effectively

in clear-air conditions) (Shun and Chan 2008) have greatly

benefitted the understanding and hence detection of low-

level wind shear at airports around the world. There are

various successful, operational examples of automatic low-

level wind shear alerting systems based on remote sensing

and/or in situ measurement equipment (Shun 2004;

Matayoshi et al. 2016; Robinson et al. 2017). So far, these

methods are detection or observation based (i.e., issuing

an alert when the phenomenon is observed) and do not

involve predicting future occurrences of the phenomena

through objective application of known physical laws.

There have also been literature attempts at prediction

or simulation of wind shear conditions using different

forms of numerical modeling techniques, including

computer fluid dynamics (Li et al. 2013), large-eddy

simulations (LES) (Cheung et al. 2011), numerical

weather prediction (NWP) (Clark et al. 1997; Chan and

Hon 2016a), and shallow-water models (Carruthers

et al. 2014). These studies were generally performed

on a case-study basis and covered single or isolated ep-

isodes of reported wind shear events. There is a lack of

long-term, systematic evaluation of the predictive ca-

pability of numerical models in capturing low-level wind

shear occurrences.

In this paper, we investigate the performance of the

Aviation Model (AVM), which is the subkilometer-

resolution NWP suite operated by the Hong Kong

Observatory (HKO), in short-term forecasts of low-

level wind shear at the Hong Kong International Airport

(HKIA) over a 2-yr period. The HKIA is constructed on

an artificial island surrounded by open waters on three

sides but in close proximity to the mountainous Lantau

Island (with an elevation up to over 900 m above sea

level) to its south. The complex orography and intri-

cate land–sea contrast are favorable ingredients for the

occurrence of low-level wind shear at HKIA, which is

well documented by many observational and modeling

studies (Shun and Lau 2000; Szeto and Chan 2006; Tang

et al. 2011a,b). This study presents the first long-term

verification results in low-level wind shear prediction

using an operational NWP model. The spatial resolution

of the model in continuous operational use (200 m) and

the coverage of the study period (2 yr) make this study a

unique contribution to the literature.

This paper is organized as follows. Section 2 gives the

technical description of the AVM. Section 3 describes

the method for generating wind shear forecasts from

the AVM and the verification method including obser-

vational data. Selected cases of wind shear features

as successfully predicted by the AVM are presented in

section 4. Section 5 presents and discusses long-term

verification results of the AVM wind shear forecasts.

Conclusions are given in section 6.

2. Model description

AVM (Chan and Hon 2016b; Hon 2018) is a

subkilometer-resolution implementation of the Weather

Research and Forecasting (WRF) (Skamarock and

Klemp 2008) Model by the HKO for finescale short-

term weather prediction at HKIA. During the course of

this study (spanning 2014–15), version 3.4.1 of the WRF

Model was in use. The AVM comprised two singly

nested domains (Fig. 1): AVM-PRD (‘‘PRD’’ refers

to the Pearl River Delta to the west of Hong Kong) at

600-m horizontal resolution and AVM-HKA (‘‘HKA’’

refers to HKIA) at 200-m horizontal resolution. The

AVM-PRD domain is about 350 km 3 350 km in di- mension (with 581 3 581 horizontal grid points) and covers the whole territory of Hong Kong, China, and its

immediate vicinity. The AVM-HKA domain is about

50 km 3 50 km in dimension (with 253 3 253 horizontal grid points) and is centered around HKIA. As a com-

promise between vertical resolution and constraints in

computational resources, 42 eta levels were used, of

which at least 10 were below 1000 m at the time of model

initialization. It is acknowledged that such a vertical

resolution would be far from ideal—experiments on

subkilometric NWP in other parts of the world (some

including aviation applications but not specific to the

problem of low-level wind shear) have been reported to

utilize between 70 vertical layers [e.g., the ‘‘London

Model’’ at 333-m resolution in Boutle et al. (2016)] to

over 100 vertical layers [e.g., the ‘‘AROME Airport’’ at

500-m resolution in Hagelin et al. (2014); the Icosahedral

Nonhydrostatic Atmosphere (ICON) model at 156-m

resolution in Heinze et al. (2017)]. However, note that

some of the above examples are run on a one-off basis

(i.e., without the operational demand to meet a real-

time forecast schedule), whereas for one of the real-time

194 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y VOLUME 59

FIG. 1. The forecast domains of the 600-m-resolution AVM-PRD and 200-m-resolution

AVM-HKA. (top) The two domains, with background satellite imagery of the visible channel

of the Himawari-8, provided through the courtesy of the Japan Meteorological Agency.

FEBRUARY 2020 H O N 195

examples the authors through testing found little dif-

ference in further refinement of vertical resolution. The

latter is similar to the experience at HKO during the

early testing stages of AVM and also studies in other

parts of the world (Zhang et al. 2013). As will be seen

from the sample cases in section 4 below, the existing

configuration of AVM, although not ideal, is generally

considered to be usable and acceptable for various

weather conditions typically experienced at HKIA, as

based on day-to-day inspection by experienced avia-

tion weather forecasters.

The AVM is run every hour to provide 6–9-h short-

term forecasts for HKIA. Initial and boundary con-

ditions of AVM-PRD are provided by the latest

available forecast (usually initialized 1–2 h earlier) of

the ‘‘RAPIDS-NHM’’ (Wong 2010; NHM indicates a

nonhydrostatic model), which is HKO’s operational meso-

scale model at 2-km resolution. Initial and boundary

conditions of AVM-HKA are in turn provided by the

concurrent AVM-PRD forecast through 1-way nesting.

For both AVM-PRD and AVM-HKA, 3D variational

data assimilation based on the WRF Data Assimilation

System (WRFDA) is used to ingest conventional surface

and upper-air observations as well as the dense network

of surface observations over the inland Guangdong

Province. The normal data cutoff time is 15 min past

every hour (e.g., for the 0000 UTC model run, we wait

until 0015 UTC to collect all available observations valid

at 0000 UTC before performing data assimilation). To

adhere to the hourly schedule, each model run (includ-

ing preprocessing, data assimilation, model integration,

and postprocessing routines) is completed within 60 min

and is normally available to the aviation weather fore-

casters at HKIA around 15 min past the subsequent

hour (e.g., output from the 0000 UTC model run is

viewable/usable around 0115 UTC).

Model settings of the AVM follow that of Chan

and Hon (2016b). Key configurations include the WRF

double-moment six-class microphysics (WDM6) scheme

(Lim and Hong 2010), RRTMG schemes for short- and

longwave radiation (Iacono et al. 2008), and the Noah

land surface model (Niu et al. 2011). Given the high

horizontal resolution, no cumulus parameterization is

used. AVM-PRD uses the MYJ boundary layer pa-

rameterization (Mellor and Yamada 1982; Janjić 2001)

whereas AVM-HKA is run without a boundary layer

scheme (i.e., using the ‘‘LES’’ mode). In place of the

boundary layer scheme, vertical diffusion is handled by

the Smagorinsky approach with the diffusivity diag-

nosed from the deformation term. The default value of

the constant Smagorinsky coefficient in the WRF Model

(i.e., Cs 5 0.18) is used. It is understood that there are inherent limitations in this formulation, particularly

when near a ‘‘wall’’ (e.g., Talbot et al. 2012), which in the

case of AVM-HKA means being adjacent to the pres-

ence of model terrain. Both domains utilize the U.S.

Geological Survey 3-s topography and MODIS land-use

data, with local adaptations to cater to urbanization and

changes in coastline. To provide additional savings in

computational time during model integration, adaptive

time stepping was used for both domains of the AVM.

For the 200-m-resolution AVM-HKA, under the pre-

scribed Courant number limit of 1.05, an integration

time step between 3 and 5 s can be achieved under most

weather conditions, but a smaller step (occasionally

down to 1–2 s) is needed during organized convection or

on particularly windy days.

3. Method and observation data

In this paper, we use forecast output from the 200-m-

resolution AVM-HKA domain. The capability of the

AVM-HKA in capturing the spatiotemporal behavior

of finescale wind features around HKIA has been re-

ported in a number of papers (Chan and Hon 2016a,b).

Following Hon and Chan (2017), forecast head wind

profiles are generated at 1-min output frequency to

emulate the specialized lidar scans in use for operational

wind shear detection at HKIA (Shun and Chan 2008).

These head wind profiles are computed by extracting the

horizontal wind vector on the closest grid points along

the three-dimensional aircraft glide path during takeoff

and landing at HKIA, then decomposing along the

runway direction (heading angle of 708 or 2508) to obtain the head wind component. The glide paths are assumed

to extend along the runway direction at an elevation

angle of 38 for landing or 68 for takeoff flights. To ap- proximate the geometry of the aircraft glide path, wind

vectors extracted from the nearest model levels are in-

terpolated vertically (but not horizontally, to preserve

spatial variability) assuming logarithmic linearity in

height. The resultant forecast head wind profile is a spa-

tial series of the head wind component of the horizontal

wind vectors interpolated onto the three-dimensional,

slanted aircraft glide path for each runway corridor at

HKIA. This is a prediction of the head wind as would be

experienced by an aircraft landing or taking off along

that particular flight path. Conventionally, positive head

wind is taken as winds blowing against the direction of

flight. Negative head wind (or tailwind) is taken as winds

blowing along the direction of flight. Since there are two

runways at HKIA (the north runway and the south

runway), and each runway can in principle be used for

takeoff or landing at either ends (heading angle of 708 or 2508), there are a maximum of eight possible config- urations (or runway corridors). AVM-HKA produces

196 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y VOLUME 59

forecast head wind profiles for all eight runway corridors

(Fig. 2) out to a distance of 4 n mi (1 n mi 5 1.852 km) from the respective runway endpoints. In practice, the

north runway is the predominant landing runway at

HKIA. Landing on the north runway from the western

side (i.e., heading angle of 708) is known as using runway corridor ‘‘07LA’’ while landing on the north runway

from the eastern side is known as using runway corri-

dor ‘‘25RA.’’

To generate a binary prediction of wind shear (‘‘yes’’

or ‘‘no’’) at a particular time, it is necessary to identify

from the forecast head wind profile of AVM-HKA the

existence of any sustained head wind changes occurring

within a certain distance. One possible choice is to

directly apply the ‘‘ramp identification algorithm’’ of

HKO’s lidar wind shear alerting system (Jones and

Haynes 1984) to the forecast head wind profiles. Here

we adopt an alternative approach loosely based on the

concept of head wind gradient similar to Chan et al.

(2011). Given a head wind profile Hi(x), where H is the

head wind (in knots) along the ith runway corridor at a

distance of x nautical miles from the runway endpoint,

the maximum and minimum head wind within a spec-

ified window, Hmax,L and Hmin,L, where L is the dis-

tance of the specified window in nautical miles, are first

identified. The pair of Hmax,L and Hmin,L that maximizes

the value of their absolute difference (i.e., producing the

largest head wind change) is then picked out. The ab-

solute difference of this pair will be labeled DHL, the maximum head wind change under a given L, for this

particular head wind profile. In this way, a binary pre-

diction of wind shear can be generated from a head wind

profile by applying a threshold to DHL. It is necessary to constrain the maximum length over which the head

wind changes occur because gentle gradients (i.e.,

changes spreading out over a long distance) are often

not regarded by aircraft pilots as wind shear events

(Hong Kong Observatory 2010).

This study covers the 2-yr period from January 2014 to

December 2015. Hourly forecasts by AVM-HKA will be

validated against reports of wind shear occurrence by

pilots landing at or taking off from HKIA [pilot reports

(PIREPs)]. As an established practice at HKIA, pilots

of landing and departing aircraft may provide reports of

low-level wind shear (and turbulence, which is consid-

ered to be a separate phenomenon in terms of reporting

and alerting at HKIA) encounters to air traffic control

either directly through onboard radio communication or

by submitting a report form afterward. This is different

from conventional PIREPs of turbulence and icing,

FIG. 2. Display panel of wind shear forecasts by AVM-HKA for 11 Jan 2014 in the form of simulated Doppler lidar return (the color

scale is given on the right). This is a 4-h forecast by AVM-HKA valid at 0600 UTC and initialized at 0200 UTC. (left),(right) Box-and-

whisker plots combining sixty 1-min forecast head wind profiles between 0600 and 0659 UTC for each of the eight runway corridors of

HKIA. (center) The simulated planar scans at elevation angles of 38 and 68. The converging winds resulting from a sea-breeze front are highlighted by the dashed yellow-outlined rectangles. During this hour, there are four PIREPs at runway corridor 07RD.

FEBRUARY 2020 H O N 197

which normally cover the en-route phase of flight. The

HKIA wind shear PIREPs contain information on the

timing, location (normally reported to the nearest nau-

tical mile), altitude (normally reported to the nearest 50

or 100 ft), and perceived intensity [normally reported

to the nearest 5 kt (1 kt ’ 0.51 m s21), e.g., 225 kt, with positive or negative signs representing gain or loss in

head wind, respectively] of an event. All received

PIREPs underwent basic quality control to flag those

reported events that fall outside the scope of HKO’s

low-level wind shear alerting service, namely 1) the

reported absolute magnitude must be 15 kt or greater

and 2) the location of occurrence must be below 1600 ft

(used interchangeably with 500 m) or within 3 n mi from

touchdown/takeoff. Since the majority of PIREPs are

received over 07LA and 25RA, we will focus only on

these two runway corridors in the following paragraphs.

Out of the 584 PIREPs received in 2014, 268 (45.9%)

occurred over 07LA and 129 (22.1%) were over 25RA.

For 2015, of the 558 PIREPs received, 293 (52.5%) oc-

curred over 07LA and 148 (26.5%) were over 25RA.

Verification is performed using the ‘‘relative operat-

ing characteristics’’ (ROC) diagram (Wilks 2006), which

has been employed in different studies on wind shear

detection in the past (Chan et al. 2011; Lee and Chan

2014). A threshold-based approach is applied such that

an alert is triggered for a runway corridor whenever the

value of DHL from a forecast head wind profile for that particular runway corridor exceeds a predetermined

threshold. If a PIREP at the same runway corridor oc-

curs within 4 min after an alert is triggered, a ‘‘hit’’ is

recorded. This is identical to the verification scheme

in use for actual alerts of low-level wind shear (i.e.,

those issued automatically using lidar detection) at

HKIA by HKO. Following Wu and Hon (2018), the

percentage of detection (PoD) and percentage of time

on alert (PoTA), which form respectively the vertical

and horizontal axes of the resultant ROC diagram, are

defined as

PoD 5 No. of hits

Total no. of pilot reports 3 100% and

PoTA 5 Alert duration

Total length of time investigated 3 100%,

where the ‘‘alert duration’’ is the length of time (in

minutes) for which an alert from the AVM forecast is

issued and the ‘‘total length of time investigated’’ is

the length of the study period (also in minutes). This

modified formulation of the ROC diagram has been

in use (since well before the study period of this pa-

per) for verification of HKO’s low-level wind shear

alerts with aviation users’ acceptance, and it is also

reported in a number of other papers (Hon and Chan

2014a,b).

4. Selected cases of wind shear features forecast by AVM-HKA

It might be illustrative to present a few cases of

AVM-HKA forecasts that correspond to actual, re-

ported occurrence of wind shear at HKIA. The three

cases described below all occurred under different

weather conditions.

The first case is related to sea breeze. On 11 January

2014, the prevailing winds at HKIA were light under

weak synoptic forcing. Sea breeze caused by differential

heating of the HKIA island and its surrounding waters

set in during the early afternoon. Figure 2 shows the

wind shear forecast display panel for this case, as made

available to aviation weather forecasters of HKO in real

time. This is a 4-h forecast valid at 0600 UTC, initialized

at 0200 UTC the same day. As described in Chan and

Hon (2016b), the middle columns show the simulated

Doppler velocity distribution around HKIA assum-

ing fixed elevation scans at 38 and 68 by an imaginary Doppler lidar located at the center of the circle. The

legend for interpreting the color contours of the simu-

lated Doppler velocity distributions is shown on the

right of Fig. 2. In this color scale, ‘‘warm’’ colors (e.g.,

pink, yellow, and brown) are used to represent different

strengths of flows away from the lidar (positive Doppler

velocities). Similarly, ‘‘cool’’ colors (e.g., green, blue,

and magenta) are used to represent different strengths

of flows into the lidar (negative Doppler velocities). The

highlighted area shows alternating brown and green

colors on the eastern side of HKIA. Since this area

of brown and green colors occur to the east of the

(imaginary) lidar, the brown colors within the high-

lighted region would translate to winds blowing away

from the lidar toward the east (i.e., westerly winds).

Similarly, the green colors within the highlighted re-

gion would translate to winds blowing toward the lidar

from the east (i.e., easterly winds). This corresponds to

the opposing, or converging, winds resulting from a sea-

breeze front at that position and is a typical signature for

lidar observations of sea-breeze features. The box-and-

whisker plots on the left and right columns of Fig. 2 also

follow the presentation in Chan and Hon (2016b). There

is one plot for each runway corridor of HKIA. Each plot

is the resultant of 60 forecast head wind profiles, at 1-

min output interval, for that runway corridor during

that forecast hour (0600–0659 UTC). The highlighted

regions show potential occurrence of wind shear as

seen from the sustained changes in head wind, rang-

ing from 210 to 110 kt. Since the sea-breeze front is

198 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y VOLUME 59

forecast to occur at the eastern side of HKIA, only the

runway corridors on the right show possible wind shear

features. During this hour, four PIREPs were received

over 07RD, where possible occurrence of wind shear

had been forecast. (Note that at HKIA only one arrival

and one departure runway corridor could be simulta-

neously in use.) The corresponding temporal evolution

of AVM-HKA forecast head wind profiles over 07RD

during this case is illustrated in Fig. 3.

The second case is related to thunderstorm down-

bursts. The capability of the AVM to capture complex

wind features associated with convection is recently re-

ported in Hon et al. (2019). On 4 September 2014, iso-

lated thunderstorms affected the western parts of Hong

Kong in the afternoon. Figures 4a and 4b show the

forecast radar reflectivity by AVM-HKA and actual

radar observations at 0600 UTC. The thunderstorms

occurred around 10 km northeast of HKIA and did not

directly affect the airport. However, thunderstorm out-

flow or downbursts spread southwestward to affect the

eastern runway corridors of HKIA such that around

0600 UTC there was one PIREP at 25RA. It can be seen

from Fig. 4c that the 6-h forecast from AVM-HKA

(initialized at 0000 UTC of the same day) successfully

captured the wind shear feature due to convergence

between prevailing north or northwesterly winds and

thunderstorm-induced easterly winds to the east of

HKIA. The corresponding temporal evolution of AVM-

HKA forecast head wind profiles over 25RA during this

case is illustrated in Fig. 5.

The third case is related to terrain-disrupted airflow,

which is relatively well documented for the AVM (Chan

et al. 2020). On 28 February 2015, Hong Kong was under

the influence of a late-season northeast monsoon, with

surface winds mainly blow from the east that then

gradually veered to southeasterly with height. This wind

regime is known to be conducive to occurrence of sig-

nificant wind shear at HKIA (Leung et al. 2018). In

Fig. 6, it can be seen that significant mountain wake flow

occurred at the western side of HKIA downstream of

the Lantau mountains. In particular, the forecast head

wind profiles for the western runway corridors all de-

picted sharp head wind changes ranging from 210 to 115 kt at different positions. Here the AVM-HKA forecast is valid at 0200 UTC and initialized at 2100

UTC of the previous day. During this hour, there were

10 PIREPs over 07LA. For this prolonged wind shear

episode, the temporal evolution of AVM-HKA forecast

head wind profiles over 07LA is illustrated in Fig. 7.

The above examples show that short-term forecasts

by the 200-m-resolution AVM-HKA are capable of

predicting correctly, under a number of different weather

conditions, low-level wind shear features that are

reported to be significant by pilots landing at or taking

off from HKIA.

5. Verification results and discussion

As described in section 3, verification of the wind

shear forecasts by AVM-HKA will utilize the ROC di-

agram. Figures 8 and 9 are the ROC diagrams for run-

way corridor 07LA over 2014 and 2015, respectively.

Figures 10 and 11 are the same diagrams for runway

corridor 25RA. Forecasts up to 6 h ahead are used in the

verification. These are represented by different colors in

the diagrams as explained in the legends. Since the

forecast head wind profiles by AVM-HKA are gener-

ated in 1-min output intervals, and PIREPs are also

timestamped to the minute, in these diagrams a ‘‘T 1 0’’ forecast by AVM-HKA refers to a forecast head wind

profile with valid time between T 1 0 and T 1 59 min (i.e., the zeroth hour), a ‘‘T 1 1’’ forecast refers to between T 1 1 h and T 1 1 h 59 min, and so on (Unfortunately ‘‘T 1 0’’ forecast data, marked by black dotes, had not been archived for 2015 and are therefore

not available.) For each forecast lead time, a family

of ROC curves is generated following the manner of

Wu and Hon (2018) by applying a range of thresholds to

FIG. 3. Evolution of AVM-HKA forecast head wind profile

during the sea-breeze-induced wind shear event over runway cor-

ridor 07RD as illustrated in Fig. 2. Here the aircraft takes off to-

ward the right-hand side (from the 0 n mi position eastward toward

the 4 n mi position). The same AVM-HKA run as in Fig. 2 is used.

The head wind profiles shown here cover three of the four PIREPs

mentioned in Fig. 2. These three PIREPs occurred successively

within a 5-min window. Evolution of the wind shear feature as a

result of sea-breeze onset is as follows: 10 min before the first report

(blue line) the head wind variations at 07RD are relatively gentle,

between 0 and 10 kt only; by the time of the first PIREP (red line), a

sharp head wind change is observed between 0 and 1 n mi

(changing from 210 to 15 kt; i.e., wind shear of ;15 kt) where the sea-breeze front occurred; this feature sustained in intensity until

the third PIREP (green line); thereafter it weakened (yellow line;

valid at 10 min after the third PIREP) such that the extent of head

wind change decreased to only ;10 kt (changing from 0 to 110 kt) while remaining at the same location (between 0 and 1 n mi).

FEBRUARY 2020 H O N 199

FIG. 4. Isolated thunderstorms as (a) simulated by AVM-HKA and (b) observed by weather radar on the afternoon of 4 Sep 2014. This is

a 6-h forecast valid at 0600 UTC and initialized at 0000 UTC. The coverage of (a) is given by the white-outlined square in (b). The area of

the zoom-in plot of (c) is shown by the black-outlined square marked in (a). It can be seen in (c) that the thunderstorm outflow in the form

of easterly winds of 10–15 kt has spread to the east of HKIA. The line of wind shear–inducing convergence is marked in red. During this

hour, there is one PIREP over runway corridor 25RA that is landing from the east using the north runway of HKIA.

200 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y VOLUME 59

the parameters DHL and L, which have been defined in section 3. The ranges of DHL and L considered are re- spectively head wind differences of 0.5–15 kt and length

scales of 0.5–2 n mi, which are similar to those used in

observation-based studies involving lidars. This leads

to a collection of points in the two-dimensional (PoD,

PoTA) space that are then plotted onto the ROC dia-

grams. If these points (which collectively we refer to

as the ‘‘curve’’ or ‘‘family of curves’’) lie along the di-

agonal, there is no skill in the detection method in dif-

ferentiating between events and nonevents. Proximity to

the top-left corner indicates increasing skill level.

It is immediately obvious that the AVM-HKA fore-

casts exhibit positive skill over both 07LA and 25RA

consistently across the 2-yr study period and throughout

the available forecast hours out to T 1 6. Although there are some subtle differences in the shape of the individual

ROC curves (such as the location or curvature of the

‘‘shoulder’’), all curves are away from the diagonal and

convex toward the top-left (i.e., skillful) corner of the

diagrams. Note that the rightmost points in the ROC

curves may not reach the top-right corner because of

model downtime resulting from system maintenance.

Assuming a characteristic PoTA of 20%, the PoD for

07LA would range between 40% and 60%. The values

are similar for 25RA. Since this is the first time that

NWP forecasts of wind shear have been systematically

validated, there is no readily available ‘‘benchmark’’ for

direct comparison. However, under the context of aviation

weather hazards, we may draw a loose analogy with the

prediction of upper-air CAT by convection-permitting

FIG. 5. Evolution of AVM-HKA forecast head wind profile

during the thunderstorm-induced wind shear event over runway

corridor 25RA as illustrated in Fig. 4. Here the aircraft lands from

the right-hand side (from the 4 n mi position to the 0 n mi position,

which is touchdown). The same AVM-HKA run as in Fig. 4 is used.

The propagation of the wind shear feature is evident from the line

plots taken at different times around the event: the sharp head wind

change (blue line, valid at 7 min before the event) occurring be-

tween 1 and 2 n mi gradually shifted westward to between 0 and 1 n

mi (orange line; valid at 3 min before the event) and finally de-

parted from the runway corridor (green line; valid at 3 min after

the event).

FIG. 6. As in Fig. 2, but for the case of 28 Feb 2015. This is a 5-h forecast valid at 0200 UTC and initialized at 2100 UTC on the previous

day. (center) The region of wind shear–inducing wake flow downstream of Lantau Island affecting the western side of HKIA is highlighted

in red. During this hour, there are 10 PIREPs at runway corridor 07LA.

FEBRUARY 2020 H O N 201

NWP models (Ellrod and Knox 2010). Similar ROC

behavior could be observed for NWP forecasts of CAT

before the application of sophisticated ensemble and

statistical postprocessing techniques (Sharman and

Pearson 2017).

There does not appear to be a definitive trend in

prediction skill across different forecast lead times. For

example, for 07LA in 2015 (Fig. 9), the ‘‘T 1 6’’ forecast (magenta) is the most skillful while the ‘‘T 1 2’’ (red) and ‘‘T 1 3’’ (orange) forecasts appear relatively least skillful. This trend is less clear cut in 2014 (Fig. 8), with

the ROC curves of different lead times generally mixed

together. Prediction skill also displays different behav-

ior across runway corridors. For 25RA, the ‘‘T 1 2’’ (red) and ‘‘T 1 3’’ (orange) forecasts appear to show highest skill level in both 2014 and 2015 while the longer-

range forecasts (‘‘T 1 4,’’ ‘‘T 1 5,’’ and ‘‘T 1 6,’’ respectively in yellow, purple, and magenta) have rela-

tively little skill. The ‘‘T 1 0’’ forecasts (available only in 2014 for both runway corridors) show higher skill level

over 07LA, particular for PoTA below 30%, but the

‘‘T 1 0’’ forecasts for 25RA do not show visibly more skill than the other forecast hours. There are a number

of possible, competing factors contributing to the lack of

observed trend in prediction skill across lead times. One

would naturally expect forecast skill to decay gradually

with increasing lead time, as a simple consequence of

atmospheric predictability (Lorenz 1963). On the other

hand, since AVM-HKA is run on a ‘‘cold start’’ basis

without cycling of model output from previous forecast

hours, the gradual ‘‘spinup’’ of turbulent features during

the first few hours of forecasts (Kealy et al. 2019) might

contribute to an increase of skill level with lead time,

given the relatively short forecast duration of the AVM

system. Last, the assimilation of dense surface obser-

vations at and around HKIA (including a network of

anemometers and weather buoys spaced at around 1 n

mi, or 1.852 km, apart) may have some positive contri-

butions around the time of model initialization. It may

be conjectured that there are complex interplay of

the above competing factors at the microscale, which

would require in-depth analysis (possibly with spectral,

or Fourier transform, methods) to delineate (e.g.,

Skamarock 2004; Mirocha et al. 2014).

Observational studies have indicated that low-level

wind shear at HKIA is a phenomenon with high spa-

tiotemporal variability (Tang et al. 2011a,b). In addition,

individual wind shear events may be transient and spo-

radic in nature (Shun and Chan 2008). These physical

characteristics can be expected to bring considerable

technical challenges in the numerical prediction—as op-

posed to detection—of low-level wind shear. According

to Skamarock (2004), the effective resolution (i.e., the

FIG. 7. Evolution of AVM-HKA forecast head wind profile

during the prolonged terrain-induced wind shear episode over

runway corridor 07LA as illustrated in Fig. 6. Here the x axis is

reversed to reflect that the aircraft arrives from the left-hand side

(from the 4 n mi position traveling eastward to the 0 n mi position,

which is touchdown). The same AVM-HKA run as in Fig. 6 is used.

In this case the terrain-induced wind shear feature (mainly caused

by the turbulent wake of the Lantau mountains upstream of

HKIA) is sustained throughout the forecast hour: since the be-

ginning of this hour (blue line) a sharp head wind change is visible

(from 15 to 118 kt) around 1 n mi; this feature sharpens slightly by the time of the first PIREP (red line); in the subsequent minutes the

head wind subsided a little but the general feature of the sharp head

wind change remained (green line; from 25 to 115 kt) between 0 and 1 n mi; this lasted until the end of the hour (yellow line).

FIG. 8. ROC diagram for wind shear forecasts by AVM-HKA for

runway corridor 07LA for the 1-yr period covering the whole of

2014. Different forecast lead times are marked using different

colors. Each dot is a point on an ROC curve generated by varying

the thresholds on DHL and L to obtain binary forecasts of wind shear events using AVM-HKA forecast head wind profiles at 1-min

output intervals. A ‘‘T 1 0’’ forecast by AVM-HKA refers to a forecast head wind profile with a valid time between T 1 0 and T 1 59 min (i.e., the zeroth hour), a ‘‘T 1 1’’ forecast refers to between T 1 1 h and T 1 1 h 59 min, etc.

202 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y VOLUME 59

wavelength of the smallest eddies in the kinetic energy

spectrum effectively represented by the NWP model) of

the WRF Model would be around 5–7 times the hori-

zontal grid spacing. Here, with the 200-m nominal res-

olution of AVM-HKA, it could be expected that eddies

on the order of 1–1.4 km may be effectively resolved.

This is comparable to the length scale of certain known

wind shear features at HKIA (Tang et al. 2011a; Wu and

Hon 2018), which could explain in part the definitive

positive skill exhibited by AVM-HKA throughout the

scope of this study. On the other hand, more recent

observational studies with even higher-resolution re-

mote sensing instruments (Hon et al. 2014) have re-

vealed other wind shear–inducing flow phenomena of

even smaller spatial and temporal scales (on the order

of a couple hundred meters and lasting only 1 or 2 min).

These phenomena are naturally beyond the prediction

capability of the existing AVM-HKA.

At all times during the study we bear in mind that in

reality the atmosphere is a three-dimensional fluid and

the full vector wind consists of three orthogonal com-

ponents. We recognize the potential theoretical limita-

tions in leaving out the vertical wind component (and

its spatial variations) in our analysis, whether in remote

sensing observations or model simulations of the wind

shear conditions. However, we note too from a practical

point of view that, given the highly inhomogeneous en-

vironment and airflow around HKIA, it would pose

considerable technical hurdles to make reliable mea-

surements of the vertical wind component (and its

gradient) along any stretch of spatial distance to the

same level of accuracy as the lidar head wind profiles.

Without such measurements, it would not be practical

to base event predictions of the NWP model on ver-

tical wind components in the boundary layer, which

might themselves exhibit certain deficiencies (Gibbs

and Fedorovich 2014; Shin and Dudhia 2016). Our

current definition, and choice, of wind shear can be to

an extent justified by the operational successes of

those remote sensing and in situ systems that are

based on the horizontal (or slanted) shear of the

horizontal winds (of which head wind is a subset) at

airports around the world.

FIG. 9. As in Fig. 8, but for the 1-yr period covering the whole of

2015. Note that ‘‘T 1 0’’ forecasts are not available for 2015.

FIG. 10. As in Fig. 8, but for runway corridor 25RA.

FIG. 11. As in Fig. 10, but for the 1-yr period covering the whole of

2015. Note that ‘‘T 1 0’’ forecasts are not available for 2015.

FEBRUARY 2020 H O N 203

The verification scheme used in the current study

follows that of detection-based alerting systems that are

operationally used. This scheme is rather demanding in

terms of simultaneous location and timing accuracy.

High-resolution NWP is known to be sensitive to phase

errors (Ebert et al. 2013) due to incorrect positioning

of weather systems in parent model domains. For

AVM-HKA, a shift of only a few grid points in the

2-km RAPIDS-NHM boundary or initial conditions

is already capable of putting a certain weather phe-

nomenon well away from HKIA (which is only about

4 km across in length). It must be considered quite re-

markable already, that positive skill is consistently ob-

tained throughout the current study. At the time of

writing, the fine-resolution winds and temperature fields

of AVM-HKA are already routinely referenced by

aviation weather forecasters at HKIA in their opera-

tional forecasting process. Postprocessed wind shear

prediction products, such as the condensed display

format in Figs. 2 and 6, could be expected to assist the

forecaster in making broad, qualitative assessments

covering the next few hours.

6. Conclusions

For the first time, statistical evaluation of NWP model

performance in the prediction of low-level wind shear,

a high-impact aviation weather phenomenon, is per-

formed. Forecast head wind profiles, which emulate the

wind components experienced by an aircraft during

the takeoff or landing phase, are generated at 1-min

output intervals using short-term forecasts from the

200-m-resolution AVM-HKA, which is the HKO’s

subkilometer-resolution WRF-based NWP suite serving

the HKIA. The HKIA is situated near complex terrain

and is known for its susceptibility to occurrence of var-

ious forms of low-level wind shear. Predictions of wind

shear specific to each of the eight runway corridors of

HKIA are generated using a threshold-based approach

and compared against actual reports of wind shear by

pilots landing at or taking off from HKIA using a veri-

fication scheme with the same level of spatial and tem-

poral accuracy requirements as operational low-level

wind shear alerting systems at HKIA based on remote

sensing detection. A 2-yr study period from January

2014 to December 2015 is used, covering over 800

events reported over the two major arrival runway

corridors of 07LA and 25RA. These wind shear events

occurred under a number of different meteorological

conditions including sea breeze, convection, and terrain-

disrupted airflow in the vicinity of HKIA. Based on

ROC analysis, positive skill is consistently observed across

both runway corridors throughout the study period and

across all considered forecast lead times out to T 1 6 h ahead. This study serves to establish and document the

current capability of fine-resolution NWP in pre-

dicting the phenomenon of low-level wind shear for

aviation weather applications, which so far is almost

exclusively alerted using high-resolution remote sensing

equipment on a detection basis when done automatically.

Acknowledgments. The Hong Kong Airline Pilots

Association (HKALPA) is gratefully acknowledged for

the permission to use deidentified pilot reports of low-level

wind shear for scientific research purposes. The author

thanks the three anonymous reviewers and Dr. Katherine

Klink for many useful comments and suggestions.

REFERENCES

Boutle, I. A., A. Finnenkoetter, A. P. Lock, and H. Wells, 2016:

The London Model: Forecasting fog at 333 m resolution.

Quart. J. Roy. Meteor. Soc., 142, 360–371, https://doi.org/

10.1002/qj.2656.

Carruthers, D., A. Ellies, J. Hunt, and P. W. Chan, 2014: Modelling

of wind shear downwind of mountain ridges at Hong Kong

International Airport. Meteor. Appl., 21, 94–104, https://doi.org/

10.1002/met.1350.

Chan, P. W., C. M. Shun, and M. L. Kuo, 2010: Latest develop-

ments of windshear alerting services at the Hong Kong

International Airport. 14th Conf. on Aviation, Range, and

Aerospace Meteorology, Atlanta, GA, Amer. Meteor. Soc.,

10.4, https://ams.confex.com/ams/pdfpapers/157792.pdf.

——, and K. K. Hon, 2016a: Observation and numerical simulation

of terrain-induced windshear at the Hong Kong International

Airport in a planetary boundary layer without temperature

inversions. Adv. Meteor., 2016, 1454513, https://doi.org/

10.1155/2016/1454513.

——, and ——, 2016b: Performance of super high resolution numer-

ical weather prediction model in forecasting terrain-disrupted

airflow at the Hong Kong International Airport: Case studies.

Meteor. Appl., 23, 101–114, https://doi.org/10.1002/met.1534.

——, ——, and D. K. Shin, 2011: Combined use of headwind ramps

and gradients based on lidar data in the alerting of low-level

windshear/turbulence. Meteor. Z., 20, 661–670, https://doi.org/

10.1127/0941-2948/2011/0242.

——, ——, and Q. S. Li, 2020: A more comprehensive study of

terrain-disrupted airflow at the Hong Kong International

Airport—Observations and numerical simulations. Weather,

https://doi.org/10.1002/wea.3593, in press.

Cheung, J. O. P., P. W. Chan, and D. Y. C. Leung, 2011: Large-eddy

simulation of the wind flow across a terminal building on the

airfield. Int. J. Earth Sci. Eng., 4, 486–489.

Clark, T. L., T. Keller, J. Coen, P. Neilley, H. Hsu, and W. D.

Hall, 1997: Terrain-induced turbulence over Lantau Island:

7 June 1994 Tropical Storm Russ case study. J. Atmos. Sci.,

54, 1795–1814, https://doi.org/10.1175/1520-0469(1997)

054,1795:TITOLI.2.0.CO;2. Ebert, E., and Coauthors, 2013: Progress and challenges in forecast

verification. Meteor. Appl., 20, 130–139, https://doi.org/

10.1002/met.1392.

Ellrod, G. P., and D. I. Knapp, 1992: An objective clear-air tur-

bulence forecasting technique: Verification and operational

204 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y VOLUME 59

use. Wea. Forecasting, 7, 150–165, https://doi.org/10.1175/

1520-0434(1992)007,0150:AOCATF.2.0.CO;2. ——, and J. A. Knox, 2010: Improvements to an operational clear-

air turbulence diagnostic index by addition of a divergence

trend term. Wea. Forecasting, 25, 789–798, https://doi.org/

10.1175/2009WAF2222290.1.

Fujita, T. T., 1978: Manual of downburst identification for project

NIMROD. University of Chicago Dept. of the Geophysical

Sciences Satellite and Mesometeorology Research Project

Research Paper 156, 104 pp., https://ntrs.nasa.gov/archive/

nasa/casi.ntrs.nasa.gov/19780022828.pdf. Gibbs, J. A., and E. Fedorovich, 2014: Comparison of convective

boundary layer velocity spectra retrieved from large-eddy-

simulation and weather research and forecasting model data.

J. Appl. Meteor. Climatol., 53, 377–394, https://doi.org/10.1175/

JAMC-D-13-033.1. Hagelin, S., L. Auger, P. Brovelli, and O. Dupont, 2014:

Nowcasting with the AROME model: First results from the

high-resolution AROME airport. Wea. Forecasting, 29,

773–787, https://doi.org/10.1175/WAF-D-13-00083.1.

Hall, F., W. Neff, and T. Frazier, 1976: Wind shear observations in

thunderstorm density currents. Nature, 264, 408–411, https://

doi.org/10.1038/264408a0. Heinze, R., and Coauthors, 2017: Large-eddy simulations over

Germany using ICON: A comprehensive evaluation. Quart.

J. Roy. Meteor. Soc., 143, 69–100, https://doi.org/10.1002/

qj.2947.

Hon, K. K., 2018: Simulated satellite imagery at sub-kilometre res-

olution by the Hong Kong observatory. Weather, 73, 139–144,

https://doi.org/10.1002/wea.3100.

——, and P. W. Chan, 2014a: Terrain-induced turbulence intensity

during tropical cyclone passage as determined from airborne,

ground-based, and remote sensing sources. J. Atmos. Oceanic

Technol., 31, 2373–2391, https://doi.org/10.1175/JTECH-D-

14-00006.1. ——, and ——, 2014b: Application of lidar-derived eddy dissipa-

tion rate profiles in low-level wind shear and turbulence alerts

at Hong Kong International Airport. Meteor. Appl., 21, 74–85,

https://doi.org/10.1002/met.1430.

——, and ——, 2017: Frequent-output sub-kilometric NWP models

supporting enhanced runway throughput and performance-

based navigation. Aeronautical Meteorology Scientific Conf.,

Toulouse, France, WMO, https://www.wmo.int/aemp/sites/

default/files/Presentation_Hon_Session1_AeroMetSci-2017.pdf.

——, ——, Y. Y. Chiu, and W. Tang, 2014: Application of short-

range lidar in early alerting for low-level windshear and

turbulence at Hong Kong International Airport. Adv. Meteor.,

2014, 162748, https://doi.org/10.1155/2014/162748.

——, S. M. Tse, P. W. Chan, and Q. S. Li, 2019: Observation and

real-time simulation of a tornado event in Hong Kong on

29 August 2018. Adv. Meteor., 2019, 8571430, https://doi.org/

10.1155/2019/8571430.

Hong Kong Observatory, 2010: Windshear and Turbulence in

Hong Kong–Information for Pilots. 3rd ed. Hong Kong

Observatory, 24 pp, https://www.hko.gov.hk/en/aviat/articles/

files/WS-turb-booklet-eng-3rd.pdf. Iacono, M. J., J. S. Delamere, E. J. Mlawer, M. W. Shephard, S. A.

Clough, and W. D. Collins, 2008: Radiative forcing by long-

lived greenhouse gases: Calculations with the AER radiative

transfer models. J. Geophys. Res., 113, D13103, https://doi.org/

10.1029/2008JD009944.

ICAO, 2005: Manual on Low-Level Wind Shear and Turbulence.

1st ed. International Civil Aviation Organization, 213 pp.

——, 2007: Meteorological Service for International Air Navigation:

Annex 3 to the Convention on International Civil Aviation.

16th ed. International Civil Aviation Organization, 187 pp.

Janjić, Z. I., 2001: Nonsingular implementation of the Mellor-

Yamada level 2.5 scheme in the NCEP Meso Model. NCEP

Office Note 437, 61 pp.

Jones, J. G., and A. Haynes, 1984: A peakspotter program applied

to the analysis of increments in turbulence velocity. Royal

Aircraft Establishment Tech. Rep. 84071, 69 pp.

Kealy, J. C., G. A. Efstathiou, and R. J. Beare, 2019: The onset of

resolved boundary-layer turbulence at grey-zone resolutions.

Bound.-Layer Meteor., 171, 31–52, https://doi.org/10.1007/

s10546-018-0420-0.

Lau, S. Y., 2000: Windshear and turbulence detection and warning

at the Hong Kong International Airport. Proc. Working

Group on Training, The Environment and New Developments

(TREND), Commission for Aeronautical Meteorology,

Hong Kong, China, World Meteorological Organization,

https://www.hko.gov.hk/en/publica/reprint/files/r371.pdf.

Lee, Y. F., and P. W. Chan, 2014: Lidar-based F-factor for wind

shear alerting: Different smoothing algorithms and applica-

tion to departing flights. Meteor. Appl., 21, 86–93, https://

doi.org/10.1002/met.1434.

Leung, M. Y. T., W. Zhou, C. M. Shun, and P. W. Chan, 2018:

Large-scale circulation control of the occurrence of low-

level turbulence at Hong Kong International Airport.

Adv. Atmos. Sci., 35, 435–444, https://doi.org/10.1007/

s00376-017-7118-y.

Li, L., P. W. Chan, L. Zhang, and F. Hu, 2013: Numerical simula-

tion of a Lee wave case over three-dimensional mountainous

terrain under strong wind condition. Adv. Meteor., 2013,

304321, https://doi.org/10.1155/2013/304321.

Lim, K.-S. S., and S.-Y. Hong, 2010: Development of an effective

double–moment cloud microphysics scheme with prognostic

cloud condensation nuclei (CCN) for weather and climate

models. Mon. Wea. Rev., 138, 1587–1612, https://doi.org/10.1175/

2009MWR2968.1.

Lorenz, E. N., 1963: Deterministic nonperiodic flow. J. Atmos.

Sci., 20, 130–141, https://doi.org/10.1175/1520-0469(1963)

020,0130:DNF.2.0.CO;2. Matayoshi, N., T. Ijima, K. Yamamoto, and E. Fujita, 2016:

Development of Airport Low-level Wind Information (ALWIN).

16th AIAA Aviation Technology, Integration, and Operations

Conf., Washington, DC, American Institute of Aeronautics and

Astronautics, AIAA 2016-4362, https://doi.org/10.2514/6.2016-4362.

Mellor, G. L., and T. Yamada, 1982: Development of a turbulence

closure model for geophysical fluid problems. Rev. Geophys.,

20, 851–875, https://doi.org/10.1029/RG020i004p00851.

Mirocha, J., B. Kosović, and G. Kirkil, 2014: Resolved turbulence

characteristics in large-eddy simulations nested within meso-

scale simulations using the Weather Research and Forecasting

Model. Mon. Wea. Rev., 142, 806–831, https://doi.org/10.1175/

MWR-D-13-00064.1.

Niu, G. Y., and Coauthors, 2011: The community Noah land sur-

face model with multiparameterization options (Noah–MP):

1. Model description and evaluation with local–scale mea-

surements. J. Geophys. Res., 116, D12109, https://doi.org/

10.1029/2010JD015139.

Proctor, F. H., D. A. Hinton, and R. L. Bowles, 2000: A windshear

hazard index. Preprints, Ninth Conf. on Aviation, Range,

and Aerospace Meteorology, Orlando, FL, Amer. Meteor.

Soc., 7.7, https://ams.confex.com/ams/Sept2000/techprogram/

paper_16347.htm.

FEBRUARY 2020 H O N 205

Robinson, D. C., D. Collins, J. Brett, J. Klewicki, and P. Murray,

2017: Airport building development: Towards a framework

for managing building-induced wind shear and turbulence

risks. J. Airport Manage., 11, 369–385. Sharman, R. D., and J. M. Pearson, 2017: Prediction of en-

ergy dissipation rates for aviation turbulence. Part I:

Forecasting nonconvective turbulence. J. Appl. Meteor.

Climatol., 56, 317–337, https://doi.org/10.1175/JAMC-D- 16-0205.1.

Shin, H. H., and J. Dudhia, 2016: Evaluation of PBL parame-

terizations in WRF at subkilometer grid spacings: Turbulence

statistics in the dry convective boundary layer. Mon. Wea.

Rev., 144, 1161–1177, https://doi.org/10.1175/MWR-D-

15-0208.1.

Shun, C. M., 2004: Wind shear and turbulence alerting at Hong

Kong International Airport. WMO Bull., 53, 316–320.

——, and S. Y. Lau, 2000: Terminal Doppler Weather Radar

(TDWR) observation of atmospheric flow over complex ter-

rain during tropical cyclone passages. Proc. SPIE, 4152, 42–53, https://doi.org/10.1117/12.410622.

——, and P. W. Chan, 2008: Applications of an infrared Doppler

lidar in detection of wind shear. J. Atmos. Oceanic Technol.,

25, 637–655, https://doi.org/10.1175/2007JTECHA1057.1. Skamarock, W. C., 2004: Evaluating mesoscale NWP models using

kinetic energy spectra. Mon. Wea. Rev., 132, 3019–3032,

https://doi.org/10.1175/MWR2830.1.

——, and J. B. Klemp, 2008: A time-split nonhydrostatic atmo-

spheric model for weather research and forecasting appli-

cations. J. Comput. Phys., 227, 3465–3485, https://doi.org/

10.1016/j.jcp.2007.01.037.

Szeto, K. C., and P. W. Chan, 2006: High resolution numerical

modelling of windshear episodes at the Hong Kong International

Airport. 12th Conf. on Aviation, Range, and Aerospace

Meteorology, Atlanta, GA, Amer. Meteor. Soc., P5.8, https://

ams.confex.com/ams/pdfpapers/100319.pdf.

Talbot, C., E. Bou-Zeid, and J. Smith, 2012: Nested mesoscale

large-eddy simulations with WRF: Performance in real test

cases. J. Hydrometeor., 13, 1421–1441, https://doi.org/10.1175/

JHM-D-11-048.1.

Tang, W., P. W. Chan, and G. Haller, 2011a: Lagrangian coherent

structure analysis of terminal winds detected by lidar. Part I:

Turbulence structures. J. Appl. Meteor. Climatol., 50, 325–338,

https://doi.org/10.1175/2010JAMC2508.1.

——, ——, and ——, 2011b: Lagrangian coherent structure analysis

of terminal winds detected by lidar. Part II: Structure evolution

and comparison with flight data. J. Appl. Meteor. Climatol., 50,

2167–2183, https://doi.org/10.1175/2011JAMC2689.1.

Wilks, D. S., 2006: Statistical Methods in the Atmospheric Sciences.

2nd ed. Academic Press, 627 pp.

Wilson, J. W., and R. M. Wakimoto, 2001: The discovery of the

downburst: T. T. Fujita’s contribution. Bull. Amer. Meteor.

Soc., 82, 49–62, https://doi.org/10.1175/1520-0477(2001) 082,0049:TDOTDT.2.3.CO;2.

Wong, W. K., 2010: Development of operational rapid update non-

hydrostatic NWP and data assimilation systems in the Hong Kong

Observatory. Third Int. Workshop on Prevention and Miti-

gation of Meteorological Disasters in Southeast Asia, Beppu,

Japan, https://www.hko.gov.hk/tc/publica/reprint/files/r882.pdf.

Wu, T. C., and K. K. Hon, 2018: Application of spectral decom-

position of lidar-based headwind profiles in windshear detec-

tion at the Hong Kong International Airport. Meteor. Z., 27,

33–42, https://doi.org/10.1127/metz/2017/0858.

Zhang, H., Z. Pu, and X. Zhang, 2013: Examination of errors in

near-surface temperature and wind from WRF numerical

simulations in regions of complex terrain. Wea. Forecasting,

28, 893–914, https://doi.org/10.1175/WAF-D-12-00109.1.

206 J O U R N A L O F A P P L I E D M E T E O R O L O G Y A N D C L I M A T O L O G Y VOLUME 59

Copyright of Journal of Applied Meteorology & Climatology is the property of American Meteorological Society and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.