Journal of Wind Engineering

and Industrial Aerodynamics 83 (1999) 35}47

Numerical modelling of extreme tropical

cyclone winds
Bruce A. Harper*
Systems Engineering Australia Pty. Ltd., 7 Mercury Court, Bridgeman Downs QLD 4035, Australia

Abstract

The paper addresses the issue of numerical and statistical modelling of extreme tropical
cyclone winds in coastal regions, leading to the estimation of design return periods of gust wind
speeds and the downstream application of risk models. The basic technique is outlined and
examples of comparisons with anemometer records for the Queensland coast are given.
A methodology is presented for the inclusion of varying surface terrain e!ects and an example
is given of its application in estimating insurance losses across a complex suburban
region. ( 1999 Elsevier Science Ltd. All rights reserved.

Keywords: Simulation; Tropical cyclone; Return period; Terrain; Satellite; Insurance loss

1. Introduction

Traditional measured data extrapolation techniques for the estimation of extreme

wind speeds, such as those generated by tropical cyclones (hurricanes or typhoons) are
often inadequate for use in specialist engineering design. Firstly, the period of reliable
data record at many sites is still relatively short (rarely exceeding 50 yr). Secondly, the
range of potential storm intensities can be high, their occurrence is typically very low
and their area of in#uence relatively small. Even where reliable recorded data is
available, extrapolation to long return periods is problematical and retention of an
associated parameter such as direction or persistence to any level of accuracy is
di$cult. Also, the statistical wind record itself does not contain guidance on questions
such as climate change, and is often not suitable for `downstreama design functions
such as the determination of extreme wave heights and currents for o!shore structural

* Tel.: #61-7-3353-0288; fax: #61-7-3353-0288.

E-mail address: [email protected] (B.A. Harper)

0167-6105/99/$ - see front matter ( 1999 Elsevier Science Ltd. All rights reserved.
PII: S 0 1 6 7 - 6 1 0 5 ( 9 9 ) 0 0 0 5 9 - 8
36 B.A. Harper / J. Wind Eng. Ind. Aerodyn. 83 (1999) 35}47

design. Knowledge of the wind generating system then becomes of paramount import-
ance rather than the isolated analysis of speci"c, and sometimes inadequate, anemom-
eter records. This paper presents an overview of techniques designed to overcome
these types of di$culties.

2. Estimation techniques without reliable measured data

In the absence of reliable measured environmental data in the 1960s the `design
storma approach arose as a useful technique for establishing some type of regional
benchmark. This would be based on an assumed return period of a particular regional
storm intensity, using nominal scales of storm in#uence. Aligning such a storm along
the most unfavorable track relative to the site of interest leads to a `worst casea
scenario for a given storm intensity. This yielded a potentially very conservative
estimate of the associated return period of the wind being experienced at a speci"c
point. In some cases it may have been erroneously assumed that the return period of
the resulting wind speed was similar to that of the return period of the regional storm
intensity. Clearly, the joint probability of intensity and proximity to site is not
considered by this method.
As computer modelling became more readily available through the 1970s, the
`design storma was replaced by the `hindcasta technique. This would involve selecting
a number of extreme events from the record that would have been likely to have

Fig. 1. Example tropical cyclone tracks in North Eastern Australia.

 B.A. Harper / J. Wind Eng. Ind. Aerodyn. 83 (1999) 35}47 37

a!ected the site in question, but were not captured by a recording system at the time.
Modelling of these events using historical track information then yielded a synthetic
time series data set which, with su$cient samples, was amenable to traditional
statistical extreme value extrapolation techniques. This technique is reasonable when
an adequate historical track record is available but fails in cases where the record is
sparse, and can be very site and data sensitive.
In the 1980s and 1990s, simulation techniques came to be developed to try and
overcome the de"ciencies of the other methods. Their attraction lay in an ability to
represent the processes at work rather than just the outcomes and, speci"cally, the
capacity to correctly consider joint probability. This shifted the focus away from the
strict mathematical art of statistical extrapolation of scalar data and towards a closer
look at the possible underlying environmental mechanisms. There have been abuses of
the technique, like any other, through a failure to adequately calibrate and verify
where possible. It is the author's view however that simulation techniques o!er not
only the best way of estimating extreme wind speed climates but also the only
reasonable way of doing so out to the long return periods (in excess of 2000 yr)
increasingly being demanded by today's risk managers [1]. This paper presents some
examples of a simulation technique for estimation of tropical cyclone extreme wind
speeds applied to sites in tropical Australia.

3. An outline of the simulation method

The initial step is the one of identifying the environment that leads to extreme
winds. This entails a study of the regional climatology of severe storms, as shown in
Fig. 1, which presents historical tropical cyclone tracks in North Eastern Australia
since 1959. It is essential that the climatological study adequately identify and separate
the main statistical populations contributing to the risk. This leads to knowledge
regarding variation in frequency, intensity, track, forward speed and spatial scale
within a suitable control volume. Typically, the control volume (e.g., 500 km radius of
site) must deliver an adequate sample size without exceeding the regional climatologi-
cal scale. One should also be aware of the accuracy of historical tropical cyclone data,
much of which is inferred by satellite interpretation in many parts of the world.
Physical limits to intensity are also known to exist based on sea surface temperature
and regional atmospheric soundings [2]. This aspect of a study requires attention to
details and reasonably extensive experience in the interpretation of meteorological
data sets.
The next step should be to reassure oneself that reasonable deterministic modelling
of wind speeds is possible in the region, using a model of choice. In this case the
Holland analytic model [3,4] has proved itself to be one of the most capable models
for representing the broad features of the tropical cyclone. An example of its use is
given in Fig. 2 for Cyclone Althea with an estimated central pressure of 952 hPa and
55 ms~1 measured wind gusts, which caused extensive damage to the North Queens-
land city of Townsville in 1971 [5,6]. Fig. 3 shows the comparison between measured
and modelled winds at the Townsville airport, based on #10 m winds. In this case
38 B.A. Harper / J. Wind Eng. Ind. Aerodyn. 83 (1999) 35}47

Fig. 2. Deterministic wind "eld modelling of tropical cyclone Althea.

Fig. 3. Wind model calibration for cyclone Althea.

 B.A. Harper / J. Wind Eng. Ind. Aerodyn. 83 (1999) 35}47 39

Fig. 4. Wind model calibration for cyclone Orson.

the model prediction of mean winds `over watera has been modi"ed to include gust
estimates over land [4]. For further comparison, an example of a much more intense
storm is presented in Fig. 4. This shows measured and modelled winds at an ocean site
during the passage of the 905 hPa Cyclone Orson o! the West Australian coast in 1989
[7]. The twin anemometers failed in their second passage through the eye wall. In this
case the analytic vortex model has been enhanced by the addition of a broad scale
geostrophic gradient that more e!ectively matches the early approach phase of the
storm and can be critical for extreme wave development.
Stochastic modelling of the assumed environment then follows, using a Monte
Carlo or other sampling technique to allow the full range of environmental para-
meters to be explored. A numerical model is constructed which will allow the
generation of synthetic storm tracks across the region of interest, with parameters
such as intensity, angle of approach and scale radius being selected from the probabil-
ity distributions derived earlier. Using the deterministic wind model, entire synthetic
time histories of wind speed and direction may then be generated anywhere within the
region, subject to the assumptions of the various model elements. A model `clocka is
used to provide a base probability reference and the simulation proceeds, with typical
durations in excess of 10,000 `yrsa being required to obtain stable exceedance
statistics at return periods of interest (e.g., 1000 yr). Accumulation of the synthetic
40 B.A. Harper / J. Wind Eng. Ind. Aerodyn. 83 (1999) 35}47

exceedance statistics then directly yields an estimate of the long-term wind speed risk
at any point. Since all major wind-generating parameters can be included in the
simulation, their relative sensitivities can next be examined and veri"cation tests can
be undertaken. Where reasonable measured data exist, the model predictions should
be compared with that data.

4. Example wind speed predictions

The "rst example is given in Fig. 5 for the North Queensland city of Townsville,
where 54 yr of measured peak tropical cyclone only wind gust data from the airport is
shown, together with the independently derived simulation result. For convenience,
the partial data series has been plotted on a simple logarithmic scale according to
Gringorten [8] with an applied Poisson scale shift. The highest recorded gust at
Townsville is due to Cyclone Althea, being approximately 55 ms~1. The current
regional recommendation of the AS1170.2 wind loading code (Region C) is also shown
[9]. It can be seen that the modelled wind speed is a very good match to the measured
data set overall, allowing for the drift of the Althea result towards higher return
periods as the data set grows. The AS1170.2 recommendation is shown here to be
potentially more conservative.
What is of interest beyond the range of data veri"cation of Fig. 5 is the predicted
wind speeds at longer return periods. The e!ect of the reducing parameter space
results in the formation of a natural limit to regional gust speeds. Beyond the 1000 yr
return period, for example, increasing wind speeds result primarily due to di!erences
in forward speeds of tropical cyclones alone, rather than an increase in intensity. This

Fig. 5. Example modelled and measured extreme winds comparison } Townsville.

 B.A. Harper / J. Wind Eng. Ind. Aerodyn. 83 (1999) 35}47 41

Fig. 6. Example modelled and measured extreme winds comparison } Cairns.

Fig. 7. Example modelled and measured extreme winds comparison } Mackay.

is because the maximum regional storm intensity will be limited [2] and in this case
tends towards a return period less than the wind speed return period of interest.
A purely statistical treatment of this data set would be unable to depict this type of
potential behavior at long return periods.
Similar studies have been conducted for a number of coastal locations throughout
Queensland. Fig. 6 presents the equivalent prediction for Cairns, compared with 52 yr
of tropical cyclone gust data and Fig. 7 presents Mackay against 35 yrs of measured
42 B.A. Harper / J. Wind Eng. Ind. Aerodyn. 83 (1999) 35}47

Fig. 8. Example modelled and measured extreme winds comparison } Brisbane.

tropical cyclone gusts. The Mackay anemometer location is subject to some local
topographic e!ects and this is believed to account for the poorer match at low wind
speeds, versus Townsville and Cairns, which are both on #at ground at standard
height with Category 2 roughness. Each city prediction is based on a separate
assessment of the regional tropical cyclone climatology, which slowly varies along the
length of the Queensland coast.
Fig. 8 presents the comparison between measured and modelled data for the case of
Brisbane, situated at the southern extremity of the state. This is an example of
a situation where the record of signi"cant tropical cyclone in#uence at a site is
actually very low but nevertheless the potential exists for a severe event to occur. The
graph shows summer half-year measured gust data for a period of some 44 yr. During
this time there were only nine occasions when peak gusts due to tropical cyclones
exceeded 15 ms~1, with the highest recorded gust being only 30 ms~1. The remaining
data are due to local severe thunderstorms. However, using the simulation approach,
it has been possible to construct a continuous representation of the regional risk of
extreme winds due to tropical cyclones and to develop the predicted curve as shown.
This suggests that tropical cyclone-generated winds are signi"cant only beyond
a return period of about 10 yr at Brisbane, with thunderstorm events predominating
at all return periods. The AS1170.2 (Region B) recommendation brackets the peak
thunderstorm data set. Although tropical cyclone wind risks are clearly lower than
thunderstorms, the much increased scale of in#uence of tropical cyclones means that
the total damage across the region will be much greater than for the equivalent
thunderstorm wind speed, which is more localised. Later model developments which
include severe thunderstorm events [10] have shown that a simulation model can also
exactly match the recorded thunderstorm data series.
 B.A. Harper / J. Wind Eng. Ind. Aerodyn. 83 (1999) 35}47 43

5. Extensions of the basic simulation method

Over the past several years, the author has successfully applied the above tech-
niques to the estimation of coastal and ocean environmental design criteria [1].
Integrated extreme wind, wave and current predictions can be formulated through
simulation, with full joint probability information at very long return periods. This
provides critical information for the dynamic design of structures subjected to multi-
directional loadings at a range of applied frequencies. Lifetime directional fatigue-
loading data that are exactly consistent with the designed extreme conditions can also
be obtained directly from such a model.
More recently the technique has been applied to the problem of estimating insur-
ance losses for domestic style construction [6,11,12] which are especially sensitive to
the incident wind speed. In this case, extensive use was made of satellite imagery to
provide the basis for estimating the regional variation in surface wind e!ects across
large and complex metropolitan regions. Tropical cyclone winds, in the main, can be
assumed to be well mixed and neutrally stable, allowing the development of a turbu-
lent boundary layer structure. Satellite imagery was then combined with a digital
elevation model at a base pixel resolution of 30 m (see Fig. 9). This provided a means
of mapping several city areas in terms of simpli"ed linear boundary layer multipliers,
broadly using the provisions of AS1170.2 [9].
Firstly, the surface roughness was determined based on delineation of the actual
surface features (both manmade and natural), and zoned using 4 Mz,cat levels in
a stepped fashion inland from the coast. Assuming maximum structure heights (z) of
7 m for domestic housing, terrain categories typically assigned were Category 2 within
1 km of the sea; Category 2.5 between 1 and 2 km from the sea; Category 3 for
remaining suburban areas and Category 4 for CBD areas. Similar terrain adjustment
sequences were applied at the land limits of each city where Category 2 or 2.5
boundary conditions were typically used. Allowance was also made for embedded
changes in roughness within suburbs where, for example, golf courses, extensive
sporting "elds or water supply reservoirs provided opportunities for boundary layer
adjustment. The technique used to date is omni-directional, which is conservative for
a landfalling tropical cyclone where both on and o!shore winds will exist, but the
technique is readily extendable to a directional variation. Shielding of houses was also
incorporated based on AS1170.2 recommendations for Ms.
While delineation of `hillsa within the de"nitions of AS1170.2 were extensively
tried, these remained impractical in the majority of real terrain situations examined
over large areas. Accordingly, only the variation in terrain slope was considered as
representing areas of accelerating #ow. A total of seven separate Mt slope multipliers
were developed based on recommendations in Part 2 of AS1170.2. Overall, this
approach is deemed to be conservative since it accounts for upslope and downslope
areas in all cases. It also ascribes the highest multiplier applicable to each slope
category without regard to transitional e!ects, proximity to crest or the `heighta of
the `hilla.
Finally, a separate information code was attributed on the basis of whether or not
the image pixel was judged to be a part of the natural or the built environment. This
44 B.A. Harper / J. Wind Eng. Ind. Aerodyn. 83 (1999) 35}47

Fig. 9. Landsat image of Townsville.

ensured that the wind analyses were constrained to areas of infrastructure interest.
This is particularly relevant along the coastal fringe where typically large areas of
swampy ground might exist next to estuaries, river mouths and the like. The combina-
tion of the chosen terrain and topographic zoning at the 30 m resolution allows for
a very comprehensive representation of the regional variability of wind speeds across
a city. This is depicted visually in Fig. 10 where each color level represents a separate
wind multiplying factor M to be applied relative to the local `over watera modelled
wind speed at any instant. These are made up from up to 56 combinations of
built/non-built Mz,cat, Mt and Ms. The proportion of any built area exposed
to a particular wind multiplier (such as a postal code region) is then obtained
directly by the summing of coded pixels in that area and can be assigned to the
insured value at risk in the same area. The base topographic model was then
incorporated into a Monte Carlo risk model of tropical cyclone winds of the type
already described.
 B.A. Harper / J. Wind Eng. Ind. Aerodyn. 83 (1999) 35}47 45

Fig. 10. Composite terrain and topographic zoning of Townsville.

An example of the outcome of such an analysis is given in Fig. 11 where wind speed
has been successfully converted to expected regional insured loss through the use of
empirical wind-damage transfer functions [13]. The graph shows the veri"cation of
the damage loss model for Townsville based on estimated construction quality in the
city at the time of cyclone Althea. The modelled prediction of the variation of loss with
wind speed (mean and standard deviations) is compared with the speci"c hindcast of
Althea and the industry recorded gross insured loss "gure. The agreement for this
speci"c event is excellent, due in part to Althea being one of the historical cyclones
used to formulate the empirical damage functions. However, the model prediction can
be seen to be well behaved across this range of wind speeds. Furthermore, Fig. 12
illustrates that the total loss prediction for any wind speed is the weighted average loss
from a number of postcodes in the region. Heaviest damage is indicated by the model
to be occuring along the coastal fringes. The general spatial variation in predicted
damage appears to be broadly consistent with experience at the time (D. Lloyd,
46 B.A. Harper / J. Wind Eng. Ind. Aerodyn. 83 (1999) 35}47

Fig. 11. Insurance loss model veri"cation.

Fig. 12. Spatial loss variation in Althea derived from regional topographic zoning.

personal communication). A storm surge inundation model also forms part of the
overall risk model for this region.

6. Conclusions

Simulation techniques, when carefully constructed, can provide essential insight

into the complex mechanisms that underscore extreme winds in a region and may
 B.A. Harper / J. Wind Eng. Ind. Aerodyn. 83 (1999) 35}47 47

provide the only reasonable basis for extrapolation to very long return periods. They
provide an excellent framework for extension to other allied "elds of analysis such as
o!shore design criteria, insurance loss and even emergency planning. Climate change
scenarios can be tested and wind directionality and duration can be readily included
to permit true joint probability estimates to be made at very long return periods. The
cooperation of SUNCORP General Insurance for the use of some illustrations is
gratefully acknowledged.

References

[1] B.A. Harper, K.F. Lovell, B.D. Chandler, D.J. Todd, The derivation of environmental design criteria
for goodwyn &A' platform, Proceedings of the nineth Australian Conference Coastal and Ocean
Engineering, IEAust, December 1989.
[2] G.J. Holland, The maximum potential intensity of tropical cyclones, J. Atmospheric Sci. 54 (1997)
2519}2541.
[3] G.J. Holland, An analytic model of the wind and pressure pro"les in hurricanes, Monthly Weather
Rev. 108 (8) (1980) 1212}1218.
[4] B.A. Harper, G.J. Holland, An updated parametric model of the tropical cyclone. Proceedings of the
23rd Conference Hurricanes and Tropical Meteorology, American Meteorological Society, Dallas,
Texas, 10}15 January 1999.
[5] James Cook University, Cyclone Althea } Part I: E!ects on Buildings, Dept. of Civil and Systems
Engineering, 1972.
[6] B.A. Harper, Tropical cyclone risk model study for North Queensland, SUNCORP Insurance and
Finance, March 1995, 400pp.
[7] B.A. Harper, L.B. Mason, L. Bode, Tropical cyclone Orson } a severe test for modelling, Proceedings
of the 11th Australian Conference on Coastal and Ocean Engineering, IEAust, Townsville, August,
1993, 59}64.
[8] I.I. Gringorten, A plotting rule for extreme probability paper, J. Geophys. Res. 68 (3) (1963) 813}814.
[9] Standards Australia, AS1170.2 } 1989 SAA Loading Code. Part 2: Wind Loads, 96pp.
[10] B.A. Harper, J. Callaghan, Modelling of severe thunderstorms in South East Queensland, Proceedings
of the Sixth Australian Severe Storms Conference, Bureau of Meteorology, Brisbane, August, 1998.
[11] B.A. Harper, Risk modelling of cyclone losses, Proceedings of the Annual Engineering Conference
IEAust, Darwin, April, 1996.
[12] B.A. Harper, The application of numerical modelling in natural disaster risk management, Proceed-
ings of the Conference Natural Disaster Reduction NDR'96, Gold Coast, September, 1996.
[13] G.R. Walker, Personal Communication, 1994.