Review of techniques and research for gust forecasting and parameterisation

Abstract

Numerous techniques are used by national met services and researchers to predict wind gusts based on output from large scale forecasts. For forecasting purposes, severe gusts are most often divided into those originating in convective, and in non-convective environments. The former are generally associated with convective downdrafts and the vertical mixing associated with deep convection, and attempts to parameterise them focus on representing these mechanisms. The latter are typically associated with transport of turbulence within the boundary layer, using measures of TKE, boundary layer wind, and in some cases stability to estimate surface gust. Perhaps the most sophisticated of these is the widely used “Wind Gust Estimate” method of Brasseur (2001). Convective and non-convective gust parameterisations operate exclusively.

Contrasting with these physically/heuristically-based parameterisations are empirical/statistical models generally derived from the variation of the behaviour of observations with different static and meteorological factors. Here, a number of predictors are usually tested in regression formulae to model the overall gust behaviour without specific reference to gust-producing mechanisms, though some account of these may be implicit in the choice of predictors.

Various orographic flow phenomena such as downslope winds, mountain waves, flow channelling, rotors, wakes are not treated explicitly in gust parameterisations. Again they may perhaps be present implicitly in statistical models trained using observations within orography, or in physically/heuristically-based models insofar as the source NWP model is able to represent terrain-affected airflow. Nevertheless, scope exists for building in new parameterisation components which deal more directly with mountain-induced effects.

Full text

Review of techniques and research
for gust forecasting and
parameterisation
Forecasting Research Technical Report 570
April 2011
P. Sheridan

Abstract
Numerous techniques are used by national met services and researchers to predict wind gusts based
on output from large scale forecasts. For forecasting purposes, severe gusts are most often divided
into those originating in convective, and in non-convective environments. The former are generally
associated with convective downdrafts and the vertical mixing associated with deep convection, and
attempts to parameterise them focus on representing these mechanisms. The latter are typically asso-
ciated with transport of turbulence within the boundary layer, using measures of TKE, boundary layer
wind, and in some cases stability to estimate surface gust. Perhaps the most sophisticated of these is
the widely used “Wind Gust Estimate” method of Brasseur (2001). Convective and non-convective gust
parameterisations operate exclusively.
Contrasting with these physically/heuristically-based parameterisations are empirical/statistical mod-
els generally derived from the variation of the behaviour of observations with different static and mete-
orological factors. Here, a number of predictors are usually tested in regression formulae to model the
overall gust behaviour without specific reference to gust-producing mechanisms, though some account
of these may be implicit in the choice of predictors.
Various orographic flow phenomena such as downslope winds, mountain waves, flow channelling,
rotors, wakes are not treated explicitly in gust parameterisations. Again they may perhaps be present
implicitly in statistical models trained using observations within orography, or in physically/heuristically-
based models insofar as the source NWP model is able to represent terrain-affected airflow. Never-
theless, scope exists for building in new parameterisation components which deal more directly with
mountain-induced effects.
1 Introduction
Wind gusts, when severe, represent hazards to property, people and transport. The processes lead-
ing to their formation, such as boundary-layer turbulence, deep convection, mountain waves and wake
phenomena are generally not resolved in NWP models, so frequently parameterisations and diagnos-
tic formulae are used to predict gusts based on more coarse grained output from NWP and/or limited
observations. Gusts are, however, chaotic and difficult to predict, often the result of a combination of
processes which may not be fully understood. Hence much effort is ongoing to develop viable tech-
niques.
The current state of gust forecasting naturally divides along several lines:
(i) convective and non-convective gusts;
(ii) statistical models trained using observations, and physical/heuristic models designed around trans-
port processes;
(iii) routine, established operational forecasting techniques used by met services, and research into
gust forecasting and turbulence.
1

(i) depends on the existence or otherwise of deep convection, (ii) represents different approaches/philosophy
when modelling gusts, (iii) the difference between practical solutions and development research.
This review first covers physical/heuristic models of convective and non-convective gusts, of the kind
which met services have so far used for gust forecasting, in section 2, including the gust diagnoses
used in the Met Office operational suite. Then, the topic of statistical and empirical gust models, which
form an area of relatively new development, is discussed in section 3. Subsequently orographic flow
processes in stable flow which contribute to gustiness, but which are unresolved by NWP models and
not treated explicitly by current gust models, are summarised in section 4. A summary and possible
directions for future work are given in section 5. The details included are derived from published papers,
reports, conference presentations and material from the web at the time of writing (April 2011).
2 Physical/heuristic models
These models are based on some hypothesis about the process involved in gust formation, such as
turbulent vertical transport within the boundary layer, or downdrafts within deep convection, and informed
by the general body of practical field experience. The discussion is not exhaustive of all models or Met
services, but takes in a broad selection.
2.1 Non-convective
Models of wind gust in non-convective conditions rely on estimates or calculations of boundary layer
turbulence and vertical transport of momentum. The effects of this are expressed in different ways for
different models. A summary of the different gust models that have been used with different NWP models
/ by different groups (whether operationally or otherwise), relevant formulae and references is given for
non-convective gusts in Table 1.
The first group of users, including the MetUM, in Table 1 use the relationship of near-surface wind
variability to u∗ (or equally, drag coefficient CD) after Panofsky et al (1977) and Panofsky and Dutton
(1984). The MetUM and ECMWF/AEMET gust diagnostics also take into account variations in BL stabil-
ity. The Met Office Virtual Met Mast (VMM) calculates mean wind using a correction to remove the effect
of orographic roughness on the near-surface wind, and to account for difference in height between the
(4km resolution UK) model surface and the true surface. In recent versions mean wind is first calculated
using a 3-D linear turbulent flow theory (Wilson et al, 2010) which predicts direction dependence due to
the surrounding topography; a u∗ based approach is then used to determine gusts (Wilson and Vosper,
2011). Note the COSMO-EU diagnostic is the maximum of the non-convective and the convective gusts
(see later) output.
The second group, which includes the algorithms used in the Met Office NIMROD and UKPP sys-
tems, represent the simple tactic of searching upwards, either to the first stable layer, the boundary layer
2

� �
top (often amounting to the same thing), or some set level to determine heights from which momentum
may be transported to the surface. This is based on the argument that beneath these heights, there
is essentially no obstruction to downward mixing of momentum. The gust diagnostic then reflects the
maximum wind encountered within these levels.
In the third group, the turbulence intensity, which often forms the basis of the model of boundary layer
vertical mixing in these models, decides the strength of the gust, based on the obvious reasoning that
gusts directly represent the degree of turbulence in the boundary layer. Of the latter group, the boundary
layer stability is also taken into account in the KNMI HIRLAM method, via the normalised gust, g, which
reflects an assumed stability-dependence of the turbulence spectrum (Schreur and Geertsema, 2008).
The fourth and largest group uses the “Wind Gust Estimate” (WGE) method of Brasseur (2001). This
is similar to the second group, except with a more intelligent selection of the level from which momentum
may be transported to the surface. Qualifying levels must satisfy,
0 0
1 Δθv(z)
E(z)dz ≥ g dz (1)
zparcel zparcel zparcel Θ(z)
i.e. TKE must be large enough to overcome any intervening bouyant inhibition. Hence this method
also has something in common with ‘group three’ in Table 1, in particular the KNMI HIRLAM method,
which is based on similar information. The WGE method is argued to be more ‘physically based’ than
the other non-convective gust prediction methods; also limits are offered to bound the estimate of gust
provided. The bounding limits are determined, for the lower bound, by using a local instead of vertically
averaged measure of TKE in the above equation, and for the upper bound, as simply the maximum wind
within the boundary layer.
A number of authors have cited the usefulness of WGE (e.g. Olafsson and Agustsson (2007), Agusts-
son and Olafsson (2009), Goyette et al (2003), Nilsson et al (2007), Adams (2004), Cheung et al (2008),
Szeto and Chan (2006), Chan (2011), Chan et al (2011)), but only limited comparisons with other tech-
niques have been published (Brasseur (2001) find some improvement relative to a rudimentary local
gust factor determined from observations, while Brasseur (2001) and LaCroix (2002) find comparable
behaviour to the standard ‘surface-layer-deflection’ (SL) method used in the AFWA MM5 configuration
(Table 1), a method which is unlikely to be the strongest of those listed). Also, there is some suggestion
that the algorithm may have a tendency to overestimate gusts (Pinto et al, 2009).
Only one (regional) climate model is included in Table 1. Rockel and Woth (2007) discuss a multi-
model ensemble of RCMs, of which only two diagnose gusts. The diagnostics described are rather
simple, but clearly eessential. The authors not only recommend use of (preferably more sophisticated
and physically based) gust parameterisations, but also highlight problems simulating extremes in moun-
tainous areas where topography may be poorly resolved compared to NWP models, suggesting the
need to account for sub-grid orography.
3

2.2 Convective
Physical/heuristic convective gust models revolve around basic concepts concerning air motion within
convective cells. Downdrafts possess vertical momentum which, on approaching the ground, is de-
flected to the horizontal. These downdrafts are considered to be driven by negative bouyancy as a
result of (i) latent cooling, due to melting of frozen precipitation as it descends through the freezing level,
and evaporation/sublimation from falling precipitation, and (ii) loading of the air with precipitation (similar
to e.g. dust in pyroclastic flows). In addition, these downdrafts also transport parcels with large hori-
zontal momentum from high levels to the surface, further enhancing gust strength. Surface winds may
also be accelerated by pressure perturbations induced by the convective or frontal system. Lastly, the
momentum of the system itself may contribute to the precise magnitude of gusts.
Different convective gust models are summarised in Table 2. Since this review was motivated origi-
nally by the requirement to better predict orographically-induced gusts during stable (i.e. non-convective)
conditions, only a handful of operationally used convective gust parameterisations have been listed.
The Nakamura et al (1996)-based scheme used by the Met Office is a fairly typical method dealing
with some of the above physical concepts. The three terms in the equation in the first line of Table 2,
calculated using model profile data, represent the momentum of the parcel at the top of the downdraft,
the latent heat-induced negative bouyancy of the air, and the precipitation loading effect. Tdef icit is the
difference between the Tmean and Tsurf ace . The top of the downdraft is considered to be the highest out
of 500m and H(Tw = 0). This assumption, consistent with the much earlier Fawbush and Miller (1954)
technique for non-frontal convection (Met Office 1993, Forecasters’ Reference Book), may be flawed:
Nakamura et al (1996) state that the downdraft may originate at higher levels if driven by sublimation,
and suggest instead that a standard height above cloud base be used for better accuracy and to avoid
spurious seasonal variations (the above limit of 500m was introduced to try and address this). Pierce
et al (1996) cast doubt on the predictive skill of the algorithm and Ashton (2004) advises care in using
the latest, improved version which still retains substantial RMS and bias errors (Hand, 2000). The
COSMO-EU diagnostic is also based on Nakamura et al (1996). Holleman (2001) proposes using
the NIMROD gust algorithm at KNMI, with modifications using upper air and radar observations for
nowcasting purposes.
Two older techniques, termed ‘T1’ and ‘T2’, based on Fawbush and Miller (1954), are still a common
rule of thumb, for instance with aviators, though only thermal downdraft potential is treated. The Fawbush
and Miller (1954) technique depends on the difference between the wet bulb potential temperature at the
Tw = 0 level and the predicted maximum temperature at the surface (a measure of potential negative
bouyancy). It also forms part of the basis for the technique of Bartha (1994), used in the MEANDER
nowcasting system (Simon et al, 2011). In the latter, standard radar reflectivity output and surface
pressure gradient also supplement the information used to diagnose gusts. Steen (1999) also reviews
some early convective gust forecasting techniques.
5

The WINDEX method (McCann, 1994) is based on a vertical momentum equation for convective
downbursts which takes account of downdraft potential due thermal, precipitation loading and pressure
perturbation effects, with empirical modifications. Kuhlman (2006) compares WINDEX with the T1 and T2
methods for a two month period in areas of the US, finding T1 and WINDEX perform comparably, better
than T2 (Kuhlman provides a full description of the techniques and reproductions of convective dynamics
diagrams from Wakimoto (2001)). Meanwhile the Naval Aerographer (http://www.tpub.com/weather3/6a-
21.htm, “Calculations of Convective Wind Gusts”) suggests use of T2 for frontal thunderstorms and
prefrontal squalls, and T1 for airmass thunderstorms. Geerts (2001) finds that WINDEX is significantly
improved by a further term to account for horizontal momentum transport (from a nominal 500mb level),
and rescaling, resulting in the “GUSTEX” method.
Bukharov et al (2008) describes a method based on the nearest standard level wind (‘principal wind
speed component’), and contributions for isobar curvature, vertical momentum exchange and a basic
treatment of thermal and precipitation induced downdraft potential. The vertical momentum exchange is
considered relatively local (based on the nearest standard level). All terms are based on NWP except
for the thermal part of the downdraft term, which is derived from satellite data.
The ECMWF algorithm (Bechtold and Bidlot, 2009) takes a similarly local approach to vertical mo-
mentum exchange, based on the shear forecast at low levels, only applying the algorithm where deep
convection is diagnosed within the NWP model. The authors state that this is found to produce better
results than downdraft-based models. Though this simple approach is somewhat empirical, it is included
here for completeness and comparison with the other current operational methods.
A number of research studies look for the (dominant) origins of severe convective gusts. Kuchera
and Parker (2006) compared analysis vertical profiles (from NCEP Rapid Update Cycle, RUC) against
reports of damaging wind, looking for skill in different profile parameters for predicting the damage oc-
currences. They found the best skill for a combination of the wind at the top of the convectively unstable
layer and a downdraft CAPE parameter derived from the level of minimum equivalent potential temper-
ature aloft (Table 2). They also found that wind blowing from warm to cold over a front, or blowing along
the line of convective development diminishes damaging wind occurrence. In the former case, bouyant
parcels rise over stable, colder air and convection is relatively isolated from the surface at higher levels.
In the latter, the system tends to slow and thus additivity between descending gusts and system speed
results in smaller values. Mahoney and Lackmann (2009) study rear-inflow jets in “realistic” idealised
3D simulations of mesoscale convective systems, showing that downward momentum transport asso-
ciated with these jets dominates the production of wind extremes. Other authors take a more detailed
view of convective system structure in estimating gust potential. Kwon and Kareem (2009) describe
research around a gust-front model containing time dependent, height-varying mean and fluctuating
wind terms at a given location, developing a ‘gust front factor’ relating to the enhancement of bound-
ary layer gust magnitude by convective gust fronts. The physical basis of different gust front models
may vary in rigour, but they essentially rely on the concept of a gravity current (e.g. Qian et al (1998)).
6

Evans et al (1995) describe a gust front model component of a cumulus parameterisation illustrating
its dependence on system/downdraft speed. Zeng et al (2010) studied observed gust ‘wavepackets’
accompanying the passage of a cold front. They decomposed the variation of wind into mean quasi-
stationary flow, coherent gust flow (∼ 1-20 minute fluctuations dominated by anti-correlated vertical and
horizontal components) and isotropic turbulence. They found that flux of momentum to the surface is
mainly through the mean term, but with the coherent gust and turbulent terms contributing substantial,
roughly equal amounts.
7

3 Statistical/empirical models and related observational studies
A large proportion of gust models place little emphasis on a physical mechanism for the understanding
of gusts, either relying on established empirical observations, or performing regressions based on ob-
served gusts in different locations under different conditions. Previously, these might represent simple
gust factors at a site (the average ratio between peak wind and mean wind within a given time interval),
determined empirically or based on simple assumptions concerning the surrounding environment (Met
Office 1993 Forecasters’ Reference Book). More recently, however, in parallel with the advent of GIS,
such methods tend to take inputs in the form of mapped model data (variables successfully trialed as
predictors and with some intuitive connection to wind behaviour) and terrain/land use characteristics,
and result in mapped products on the same grid. Due to their use of regression methods, and proba-
bilistic treatment of gusts, these are often termed statistical models. Statistical models are well suited to
predicting the probability of gusts exceeding some threshold.
The statistical approach has advantages and disadvantages. For instance, since it is not constrained
by a particular physical process model, it has a good chance of capturing the net effect of all relevant
processes to some extent, insofar as they are represented by the variation of the predictor variables.
However, the absence of a physical process understanding may mean that they are more difficult to
evaluate meaningfully and improve.
Field observations have been used to study gust factors, often with the focus on how these factors
are affected by surrounding terrain. For instance, Agustsson and Olafsson (2004) show a direction
dependence in the gust factor at a location adjacent to a large, isolated mountain. The gust factor
varies between 1.4 and 1.6 for mean wind speeds greater than 10ms−1, with larger values related to
upstream disturbance created by the mountain. A similar effect occurs for other stations depending
on mountain height and distance. Little dependence is found on stability (N, Ri), possibly due to the
compensating effect of enhanced mountain effects, combined with decreased downward momentum
mixing, with increasing stability. A deficiency of using gust factors to derive gust forecasts from the
mean wind is that they typically decrease with increasing windspeed (Agustsson and Olafsson, 2004),
but this decrease is not generic (Naess et al, 2000). However, they are a standard measure used by
wind engineers in assessing the vulnerability of standing structures. Typical mean values of gust factors
seem to lie generally between 1 and 2 (Met Office 1993 Forecasters’ Reference Book, Naess et al
(2000), Agustsson and Olafsson (2004)).
Barrett and Short (2008) developed a tool to predict peak wind strength by using observations at
the Kennedy Space Centre/Cape Canaveral tower network and soundings. The most successful com-
bination of predictors proved to be the maximum wind up to 3000ft and the surface inversion strength
and depth. The latter reflects the decoupling of surface winds from the flow aloft by strong near-surface
stability.
Gray (2003) describes a statistical approach to convective gust forecasting, using the Met Office
9

LEM, with initialisation and lateral boundary conditions provided by two series of typical radiosonde
profiles for convective conditions and driven by a prescribed surface heating function, to simulate con-
vection. After finding that the Nakamura et al (1996) method predicts the peak near-surface wind in the
domain well, Gray goes on to fit Gaussian functions to the distributions of perturbation gusts (calculated
for individual grid points within a given time window) in each simulation at different times. Gray finds
empirical relationships of these distributions’ mean and standard deviation to the product of the maxi-
mum gust and the mean wind, and the maximum gust, respectively. Given the maximum gust and the
mean wind may be gained from NWP and gust diagnostics, a forecast of the distribution of gusts may
be made, and thus of probabilities that different thresholds are exceeded.
Friedrichs et al (2009) describe the use of Generalised Linear Models (GLMs) to describe the be-
haviour of winds detected by 139 stations of the DWD observation network. GLMs involve the as-
sumption that the probability of different values of the dependent variable (i.e. gust) conforms to an
exponential-type distribution (e.g. Poisson) whose mean is a linear function of the predictor variable.
Friedrichs et al (2009) also compare the use of Generalised Extreme Value (GEV) distributions to model
the behaviour of gusts. A GEV distribution represents the distribution of maxima of a sequence of inde-
pendent, identically distributed random variables, and hence is suited to modelling gusts (the extreme
value wind measurements within a given time period). The aim is to obtain, for a given mean wind,
a probability distribution for the gust strength, and therefore the probability that some threshold is ex-
ceeded. Both types of model are tuned by regression against observations. Using a Brier score with
only the mean wind as predictor, it is found that GEVs of gust perturbation offer the best combination of
stability (reproducibility for different training periods) and skill. Adding further predictors to this method
such as CAPE and DCAPE (downdraft CAPE), winds at different levels, the best results are found using
the mean wind and the highest gust measured in the surrounding 100km. From the point of view of fore-
casting, using ECMWF model 10m wind offers generally comparable skill; CAPE, DCAPE and winds at
different levels make little relative impact as predictors.
Etienne et al (2010) use Generalised Additive Models (GAMs) to model regional wind extremes in
Switzerland from a climatological point of view. GAMs are an extension of GLMs, whereby the response
function for the dependent variable (gust) need not be confined to a simple linear function. The study
focuses on the 98th percentile of daily maximum winds, found to be a convenient analogue to days with
damaging winds. Observations from seventy stations are used. Etienne et al (2010) employ GIS topog-
raphy data - location, elevation, slope and slope orientation, curvature (resolved into along slope and
transverse components) and landform category (e.g. canyon, midslope drainage) calculated on a range
of scales from 50m to 2000m. Different combinations of these predictors are studied; combinations of
predictors are only permitted if their mutual correlation coefficient is sufficiently small. A combination of
curvature, slope, landform and elevation at 1km scale is found to give the best results. Predictor char-
acteristics show the highest wind extremes correspond to high elevations, exposed landforms, shallow
slope values, and negatively curved (convex) surfaces. The authors remind that their method only takes
10

into account static factors; complex terrain flow processes emerge from multiple static and dynamic
factors on a range of scales, and cannot be tagged simply to terrain characteristics.
Sallis et al (2011) explore different machine learning approaches, including neural networks, for
predicting gusts based on simple meteorological variables. The best performer is found to be a simple
classification and regression tree (CART, has some features in common with a GAM). Some tests are
performed with 30min lagged data to test prediction of future gusts based on present observations;
significant skill remains at this lead time. Sallis et al (2011) use an unusually specific definition of gust,
corresponding to standard US weather observing practice, concerning peak wind, wind variability and
rate of acceleration, and duration; it is prediction of this, rather than some threshold exceedance, which
is tested. In a similar vein, Kretzschmar et al (2004) evaluate the potential of neural network classifiers
based on lagged wind/gust data and ECMWF analysis data from 24 hr previous to predict gusts finding
benefits from inclusion of both kinds of data.
Sanabria and Cechet (2010) fit the behaviour of wind extremes at Sydney Airport to a Generalised
Pareto Distribution (GPD, similar to a GEV distribution, but applying to a range of values close to the
extreme, rather than the extreme value alone) in order to compare this to a newly developed Monte
Carlo technique: using observations to develop gust factor distributions for different mean wind cate-
gories (ignoring as trivial, data where mean wind is below 5 ms−1), these distributions are sampled in
a Monte Carlo process, with care to ensure the same overall distribution of peak wind strengths. This
produces over 2500 effective years of data which compare well statistically to the original dataset and
can be used to estimate bounds upon long return periods for different wind extremes. The Monte Carlo
process is justified by citing that the use of the gust factor, which relates turbulence to mean wind, repre-
sents the physical process of gust formation by transport of turbulence from higher levels. As an aside,
the authors highlight the deficiency of typical anemometer instruments, developed to measure mean
winds, in detecting the true intensity of the extreme, short-lived gusts typically associated with damage.
The technique has been applied to output from high resolution climate models, where hazardous wind
occurrences were found to increase under climate change as a function of emmissions (Sanabria and
Cechet, 2011). Sanabria and Cechet (2007) attempt to directly account for the process producing the
gust by grouping data according to past/present weather type at the time, so that return periods for gusts
with different sources may be determined.
Glahn and Dallavalle (2006) discuss gridded Model Output Statistics (MOS) products under devel-
opment at NWS. MOS involve a (here multiple linear) regression between model and observational data
to correct empirically for differences in local detail. Rudack (2006) describes the application of this tech-
nique to gusts. Using predictors (for mean and gust wind) including model u, v and wind speed at the
10-m, 925-mb, 850-mb, 700-mb, and 500-mb, relative vorticity, relative humidity and some wind speed
observations (at short lead times), along with the first and second harmonics of the day to account for
the seasonal variation of wind gusts throughout each 6-month season. Also added specifically for gusts
are the gust speed observed 3 hours after initial model time, the difference between the GFS 850-mb
11

temperature forecast valid at a specific projection time and 12 hours later, a BL mixing potential param-
eter, and the ratio between the 925-mb and the 10-m model wind speeds. The prediction equations
developed are “regionalised”, applying to a number of stations in a region. Within the interpolation in-
herent in the approach the “lapse rate” of different variables is taken into account (i.e. the simple height
dependence). In a separate effort, Cook et al (2008) describe a method using of mixed layer momentum
transfer to forecast gusts, carried out through the NWS “BUFKIT” profile analysis platform, with good
results for the NAM model.
Connor et al (2003) describe a ‘stratified’ (according to synoptic wind direction) MOS technique for
predicting the detailed wind field in Sydney Harbour based on typical standard level meteorological
variables, local lapse rates, land-sea thermal contrast and other derived variables. They also develop a
gust prediction technique, citing the surrounding complex terrain as a likely source of turbulence. The
optimal technique found involves a nonlinear regression in terms of the stratified-MOS-predicted mean
wind speed and direction which explains 98% of the observed variation of gusts.
K. Herring (Met Office internal reports, PostProc tickets 96, 364) describes a method which uses
gust observations more directly, to modify model-based gust predictions. The modification acts at the
nowcasting timescale; predictions converge with the model-based gust after six hours. A system based
on this method was recently implemented in UKPP.
4 Unresolved orographic flow processes which lead to wind ex-
tremes
Few of the models so far discussed take account of surrounding topography in the calculation of gusts.
The overall effects of the local terrain are implicit in statistical models tuned against observations, while
gust factors and methods such as the VMM and stratified MOS technique of Connor et al (2003) (can)
incorporate a direction dependence reflecting the surrounding landscape. Note also engineering stan-
dards take account of surrounding topography (Ngo and Letchford, 2008). None, however, attempt to
specifically model orographic flow processes (associated with stable, or unstable flows) which are not re-
solved or represented in current NWP models. Such processes include lee waves, downslope winds and
rotors (Doyle and Durran (2002), Vosper (2004), Mobbs et al (2005), Sheridan et al (2007), Doyle et al
(2009)), Foehn (Mayr et al (2002), Zangl (2003)), Bora (Belusic et al (2004), Belusic et al (2007), Gohm
and Mayr (2005)) and channeled flows (Whiteman and Doran (1993), Mayr et al (2007), Sheridan and
Bedford (2010), Sheridan et al (2010)). Meanwhile, efforts at the Met Office to forecast lee waves and
rotors and their effect on near-surface flow (Vosper (2003), Vosper (2004), Mobbs et al (2005), Sheridan
et al (2007)) have resulted in the development of operational forecasting tools. There is also interest in
accounting for such processes in road wind hazard risk products (P. Murkin, personal communication).
Datasets from the Falklands Islands (Mobbs et al, 2005) and the Pennine hills (Sheridan et al, 2007)
12

offer the possibility of a more detailed examination of the effect of lee waves and rotors on gust strength.
5 Summary and discussion
A wide range of approaches for the forecasting of gust strength exists, developed for routine forecasting
or tailored to specific locales. These may be based on a specific treatment of the physical process of
boundary layer mixing or convective vertical transport, or developed with a primary basis on statistical
similarity to observations in relative ignorance of the underlying processes. It is not clear for general
forecasting techniques if the degree of realism in the underlying science necessarily adds significantly
to the accuracy of results (compare WGE (Brasseur, 2001) to the less sophisticated non-convective
gust methods, or likewise the simple shear-based ECMWF convective gust formulation to other more
physically-based treatments of convective flow structure). Statistical methods attempt to sidestep the
need to directly address different atmospheric and orographic processes, while implicitly building in their
effects within the overall response to predictor variables. There are clear advantages in this somewhat
“blind” approach, though the degree of success seems to reflect the complexity of the model, its degree
of local focus, and the extent to which care has been taken to account for physical processes that are
expected to be important. For instance, in the scheme described by Connor et al (2003), 98% of the
variation in gusts is explained using a system with a large number of predictors, including a stratification
in terms of synoptic conditions, tuned to a small area (Sydney Harbour), and which takes account of
the effect of the surrounding topography on turbulence. Such systems are harder to generalise since
the whole process must be repeated for a new area. Meanwhile, simpler physically-based treatments
have the advantages that they are based on reproduction of actual processes, which should be broadly
applicable, while being relatively easy to understand, and therefore to improve. Statistical methods are
well suited to predicting threshold exceedance, though Gray (2003) shows how a physical model can
also be adapted to produce statistical predictions.
To date, models do not attempt to directly address specific orographic processes (such as lee waves
or orographically generated convection). It is possible that the enhancement of the mean wind by
these processes, and subsequent application of existing gust forecasting techniques, would be suf-
ficient. Other suggestions for the future might involve synthesis between the advantages of statisti-
cal/empirical/mapped and physical approaches.
It is clear that methods of gust prediction based on model output have a crucial role to play in climate
change projections as well as weather forecasting.
13

References
Adams, N., 2004: A numerical modeling study of the weather in east antarctica and the surrounding
southern ocean. Wea. Forecast., 19, 653–672.
Agustsson, H., and H. Olafsson, 2004: Mean gust factors in complex terrain. Met. Zeit., 13, 149–155.
, and , 2009: Forecasting wind gusts in complex terrain. Meteorol. Atmos. Phys., 103, 173–185.
Ashton, H., 2004: The nimrod wind gust algorithm. Technical report, Met Office.
Barrett, J. H., and D. A. Short, 2008: Peak wind tool for general forecasting, nasa contractor report
nasa/cr-2008-214743. Technical report, NASA Applied Meteorology Unit, Kennedy Space Center,
Florida.
Bartha, I., 1994: Development of a decision procedure for forecasting maximum wind gusts associated
with thunderstorms. Meteorol. Appl., 1, 103–107.
Bechtold, P., and J.-R. Bidlot, 2009: Parametrization of convective gusts. ECMWF Newsletter No. 119.
Beljaars, A. C. M., 1987: The influence of sampling and filtering on measured wind gusts. J. Atmos.
Oceanic Technol., 4, 613–626.
Belusic, D., M. Pasaric, and M. Orlic, 2004: Quasi-periodic bora gusts related to the structure of the
troposphere. Q. J. R. Meteorol. Soc., 130, 1103–1121.
, M. Zagar, and B. Grisogono, 2007: Numerical simulation of pulsations in the bora wind. Q. J. R.
Meteorol. Soc., 133, 1371–1388.
Brasseur, O., 2001: Development and application of a physical approach to estimating wind gusts. Mon.
Wea. Rev., 129, 5–25.
Bukharov, M. V., V. M. Losev, and B. E. Peskov, 2008: Automated estimation of the maximum speed of
surface wind gusts by taking into account information obtained from the geostationary satellite. Russ.
Met. Hyd., 33, 753–759.
Calvo, J., and G. Morales, 2009: Verification of wind gust forecasts. 10th Aladin/HIRLAM All-Staff
Meeting.
, J. A. Lpez, F. Martn, G. Morales, and R. Pascual, 2010: Expolosive cyclogenesis of extra-tropical
cyclone klaus and and its effects in spain. a case study of hurricane force gusts. 11th Aladin/HIRLAM
All-Staff Meeting.
Chan, P. W., C. C. Lam, and P. Cheung, 2011: Numerical simulation of wind gusts in intense convective
weather and terrain-disrupted airflow. Atmsfera, 24, 287–309.
14

, 2011: A significant wind shear event leading to aircraft diversion at the hong kong international
airport. Meteorol. Appl., DOI: 10.1002/met.242.
Cheung, P., C. C. Lam, and P. W. Chan, 2008: Numerical simulations of wind gusts in terrain-disrupted
airflow at the hong kong international airport. P2.34, 13th Conference on Mountain Meteorology,
Whistler, BC, Canada, AMS.
Connor, G. J., E. Spark, and T. M. Dunsmuir, 2003: Statistical forecasting techniques to describe the
surface winds in sydney harbour. Aust. Meteor. Mag., 52, 101–115.
Cook, K. R., B. Gruenbacher, and L. D. Williams, 2008: Assessment of science and methodologies to
forecast wind and wind gust speed. 12th Annual High Plains Conference, Hays, KA, AMS/NWA.
Della-Marta, P. M., H. Mathis, C. Frei, M. A. Liniger, J. Kleinn, and C. Appenzellera, 2009: The return
period of wind storms over europe. Int. J. Climatol., 29, 437–459.
Doyle, J. D., and D. R. Durran, 2002: The dynamics of mountain-wave-induced rotors. J. Atmos. Sci.,
59, 186–201.
, V. Grubisic, W. O. J. Brown, S. F. J. D. Wekker, A. Dornbrack, Q. Jiang, S. D. Mayor, and M. Weiss-
mann, 2009: Observations and numerical simulations of subrotor vortices during t-rex. J. Atmos. Sci.,
66, 1229–1249.
Etienne, C., A. Lehmann, S. Goyette, and M. B. J.-I. Lopez-Moreno, 2010: Spatial predictions of extreme
wind speeds over switzerland using generalized additive models. J. Appl. Meteorol. Climatol., 49,
1956–1970.
Evans, J. L., W. M. Frank, and G. S. Young, 1995: Integrated cumulus ensemble and turbulence (icet):
An integrated parameterization system for general circulation models (gcms). Fifth Atmospheric Ra-
diation Measurement (ARM) Science Team Meeting, San Diego, California.
Fawbush, E. J., and R. C. Miller, 1954: A basis for forecasting peak wind gusts in non-frontal thunder-
storms. Bull. Am. Meteorol. Soc., 35, 14–19.
Friedrichs, P., M. Gober, S. Bentzien, A. Lenz, and R. Krampitz, 2009: A probabilistic analysis of wind
gusts using extreme value statistics. Met. Zeit., 18, 615–629.
Geerts, B., 2001: Estimating downburst-related maximum surface wind speeds by means of proximity
soundings in new south wales, australia. Wea. Forecast., 16, 261–269.
Glahn, B., and J. P. Dallavalle, 2006: Gridded mos–techniques, status, and plans. 2.1, 18th Conference
on Probability and Statistics in the Atmospheric Sciences, Atlanta, AMS.
Gohm, A., and G. J. Mayr, 2005: Numerical and observational case-study of a deep adriatic bora. Q. J.
R. Meteorol. Soc., 131, 1363–1392.
15

Goyette, S., O. Brasseur, and M. Beniston, 2003: Application of a new wind gust parameterization:
Multiscale case studies performed with the canadian regional climate model. J. Geophys. Res., 108,
D13, 4374.
Gray, M. E. B., 2003: The use of a cloud resolving model in the development and evaluation of a
probabilistic forecasting algorithm for convective gusts. Meteorol. Appl., 10, 239–252.
Hand, W., 2000: An investigation into the nimrod convective gust algorithm, met office technical report
no. 321. Technical report, Met Office.
Higuchi, K., C. Yuen, J. Klaassen, S. Eng, H. Auld, and T. Gan, 2008: Simulation of extreme wind and
precipitation patterns associated with a squall line passage in southern ontario on august 2, 2006.
A13A-0228, Fall Meeting, San Francisco, AGU.
Holleman, I., 2001: Wind gusts. Technical report, KNMI, Netherlands.
James, M. H., and C. B. Block, 1998: Wind gust forecasting for the advanced transportation weather
information system (webpage). Technical report, Regional Weather Information Center, John D. Ode-
gard School of Aerospace Sciences, University of North Dakota, Grand Forks, North Dakota.
Kretzschmar, A., P. Eckert, D. Cattani, and F. Eggimann, 2004: Neural network classifiers for local wind
prediction. J. Appl. Meteorol., 43, 727–738.
Kuchera, E. L., and M. D. Parker, 2006: Severe convective wind environments. Wea. Forecast., 21,
595–612.
Kuhlman, C. J., 2006: Evalulation of convective wind forecasting methods during high wind events.
Masters thesis, Naval Postgraduate School, Monterrey, CA.
Kwon, D. K., and A. Kareem, 2009: A framework for gust-front factor. 11th Americas conference on
wind engineering, San Juan, Puerto Rico, American Association for Wind Engineering.
LaCroix, K. W., 2002: Application of the wind gust estimate and comparison to the afwa mm5 wind gust
algorithm. Masters thesis, Department of the Air Force, Air University, Air Force Institute of technology,
Wright-Patterson Air Force Base, Ohio.
Mahoney, K. M., and G. M. Lackmann, 2009: The role of the trailing stratiform region in convective mo-
mentum transport and mesoscale convective system motion. 23rd Conference on Weather Analysis
and Forecasting/19th Conference on Numerical Weather Prediction, Omaha NE, AMS.
Mayr, G. J., J. Vergeiner, and A. Gohm, 2002: An automobile platform for the measurement of foehn
and gap flows. J. Atmos. Oceanic Technol., 19, 1545–1556.
, L. Armi, A. Gohm, G. Zangl, D. R. Durran, C. Flamant, S. Gabersek, S. Mobbs, A. Ross, and
M. Weissmann, 2007: Gap flows: Results from the mesoscale alpine programme. Q. J. R. Meteorol.
Soc., 133, 881–896.
16

McCann, D. W., 1994: Windex: A new index for forecasting microburst potential. Wea. Forecast., 9,
532–541.
Mobbs, S. D., S. B. Vosper, P. F. Sheridan, R. Cardoso, R. R. Burton, S. J. Arnold, M. K. Hill, V. Horlacher,
and A. M. Gadian, 2005: Observations of downslope winds and rotors in the falkland islands. Q. J. R.
Meteorol. Soc., 131, 329–351.
Naess, A., P. H. Clausen, and R. Sandvik, 2000: Gust factors for locations downstream of steep moun-
tain ridges. J. Wind Eng. Ind. Aerodyn., 87, 131–146.
Nakamura, K., R. Kershaw, and N. Gait, 1996: Prediction of near-surface gusts generated by deep
convection. Meteorol. Appl., 3, 157–167.
Ngo, T., and C. Letchford, 2008: A comparison of topographic effects on gust wind speed. J. Wind Eng.
Ind. Aerodyn., 96, 2273–2293.
Nilsson, C., S. Goyette, and L. Brring, 2007: Relating forest damage data to the wind field from high-
resolution rcm simulations: Case study of anatol striking sweden in december 1999. Global and
Planetary Change, 57, 161–176.
Olafsson, H., and H. Agustsson, 2007: The freysnes downslope windstorm. Met. Zeit., 16, 123–130.
Panofsky, H. A., and J. A. Dutton, 1984: Atmospheric Turbulence: Models and Methods for Engineering
Applications. Wiley and Sons, New York.
, H. Tennekes, D. H. Lenschow, and J. C. Wyngaard, 1977: The characteristics of turbulent velocity
components in the surface layer under convective conditions. Boundary-Layer Meteorol., 11, 355–361.
Pierce, C. E., C. G. Collier, P. J. Hardaker, and E. Archibald, 1996: Forecasting wind gusts using a
mesoscale numerical model and a conceptualmodel of deep convection with radar and satellite data,
final report to eurotunnel. Technical report, Met Office.
Pinto, J. G., C. P. Neuhaus, A. Kruger, and M. Kerschgens, 2009: Assessment of the wind gust estimate
method in mesoscale modelling of storm events over west germany. Met. Zeit., 18, 495–506.
Qian, L., G. S. Young, and W. M. Frank, 1998: A convective wake parameterization scheme for use in
general circulation models. Mon. Wea. Rev., 126, 456–469.
Rockel, B., and K. Woth, 2007: Extremes of near-surface wind speed over europe and their future
changes as estimated from an ensemble of rcm simulations. Climatic Change, 81, 267–280.
Rudack, D., 2006: Gfs-based mos wind gust guidance for the united states, puerto rico, and the u.s.
virgin islands, mdl technical procedures bulletin no. 06-01. Technical report, NOAA, U.S. Dept. of
Commerce.
17

Sallis, P. J., W. Claster, and S. Hernndez, 2011: A machine learning algorithm for wind gust prediction.
submitted to Computers & Geosciences.
Sanabria, L. A., and R. P. Cechet, 2007: Severe winds hazard modelling using generalised pareto
distributions. 12th International Conference on Wind Engineering, Cairns, Australia.
, and , 2010: Severe wind hazard assessment using monte carlo simulation. Env. Model.
Assess., 15, 147–154.
, and , 2011: Severe wind hazard under current and future climate. 13th International Confer-
ence on Wind Engineering, Amsterdam, Netherlands.
Schreur, B. W., and G. Geertsema, 2008: Theory for a tke based parameterization of wind gusts.
HIRLAM Newsletter no 54.
Schulz, J.-P., and E. Heise, 2003: A new scheme for diagnosing near-surface convective gusts. COSMO
Newsletter No. 3.
, 2008: Revision of the turbulent gust diagnostics in the cosmo model. COSMO Newsletter No. 8.
Seity, Y., P. Brousseau, S. Riette, E. Wattrelot, P. Marquet, E. Perraud, and C. Lac, 2010: What’s new in
arome-france configuration? 11th Aladin/HIRLAM All-Staff Meeting.
Sheridan, P. F., and S. Bedford, 2010: Forecasting near-surface winds for uav operations close to
camp bastion, northern helmand. Battlespace Atmospheric and Cloud Impacts on Military Operations
(BACIMO), Omaha, Nebraska.
, V. Horlacher, G. G. Rooney, P. Hignett, S. D. Mobbs, and S. B. Vosper, 2007: Influence of lee
waves on the near-surface flow downwind of the pennines. Q. J. R. Meteorol. Soc., 133, 1353–1369.
, S. Bedford, and S. B. Vosper, 2010: Forecasting near-surface winds in northern helmand,
afghanistan. P2.19, 14th Conference on Mountain Meteorology, Olympic Valley, Lake Tahoe, CA,
AMS.
Simon, A., A. Kann, M. Nestiak, I. Meirold-Mautner, A. Horvath, K. Csirmaz, O. Ulbert, and C. Gruber,
2011: Nowcasting and very short range forecasting of wind gusts generated by deep convection.
General Assembly 2011, Vienna, Austria, EGU.
Steen, T. A., 1999: Forecasting downdraft windspeeds associated with airmass thunderstorms for pe-
terson airforce base, colorado, using the wsr-88d. Masters thesis, Department of the Air Force, Air
University, Air Force Institute of technology, Wright-Patterson Air Force Base, Ohio.
Szeto, K., and P. Chan, 2006: Numerical simulation of a severe squall event in hong kong. P11.2, 23rd
Conference on Severe Local Storms, St Louis, MO, AMS.
18

Vosper, S. B., 2003: Development and testing of a high resolution mountain-wave forecasting system.
Meteorol. Appl., 10, 75–86.
, 2004: Inversion effects on mountain lee waves. Q. J. R. Meteorol. Soc., 130, 1723–1748.
Wakimoto, R. M., 2001: Convectively driven high wind events. Number 28 in Severe Convective Storms,
AMS, 255–298.
Walser, A., M. Arpagaus, and C. Appenzeller, 2006: The impact of moist singular vectors and horizontal
resolution on short-range limited-area ensemble forecasts for two european winter storms. Mon. Wea.
Rev., 134, 2877–2887.
Whiteman, D. C., and J. C. Doran, 1993: The relationship between overlying synoptic-scale flows and
wins within a valley. J. Appl. Meteorol., 32, 1669–1682.
Wilson, C., and S. Vosper, 2011: Notes on virtual met mast development: Local wind maps and turbu-
lence diagnostics. Technical report, Met Office.
, S. Webster, S. Vosper, P. Clark, A. Brown, P. Murkin, A. Leonard-Williams, T. Butcher, and R. Har-
rison, 2010: The met office virtual met mast. Technical report, Met Office.
Zangl, G., 2003: Deep and shallow south foehn in the region of innsbruck: Typical features and semi-
idelized numerical simulations. Meteorol. Atmos. Phys., 83, 237–261.
Zeng, Q., X. Cheng, H. Fei, and Z. Peng, 2010: Gustiness and coherent structure of strong winds and
their role in dust emission and entrainment. Adv. Atmos. Sci., 27, 1–13.
19

Met Office Tel: 0870 900 0100
FitzRoy Road, Exeter Fax: 0870 900 5050
Devon, EX1 3PB enquiries@metoffice.gov.uk
UK www.metoffice.gov.uk