You are on page 1of 9

Journal of Wind Engineering

and Industrial Aerodynamics 90 (2002) 11631171

Extreme wind speed in France:


the 99 storms and their consequences
Christian Sacre! *
Aerodynamic and Climatic Engineering Department, CSTB, 11, rue Henri Picherit BP 82341,
F-44323 Nantes, France

Abstract

Two very strong storms made some important damages in France, in December 1999. The
meteorological stations compared the mean wind speeds and maximum gust speeds recorded
by the French meteorological ofce to previous observations. A new map of 50 years return
period mean wind speed is proposed. r 2002 Published by Elsevier Science Ltd.

1. Introduction

A very deep depression with exceptional violent winds crossed very quickly the
northern part of the country on Sunday morning, 26 December 1999. The trajectory of
that depression follows from west to east an approximate straight line along the 49th
parallel. Hence, it travelled at about 100 km/h. The most violent winds swept south of
#
NormandyIle-de-FranceChampagne-ArdennesLorraineAlsace then Germany.
A second depression travelled also at about 100 km/h, crossed the country from the
afternoon of Monday, 27 December to the following night. The trajectory followed a line
Nantes-Dijon. The most concerned regions were rst the south of Brittany and the Atlantic
coast in the afternoon, then all regions located south of the line La RochelleM#acon.

2. Evaluation of the return period for the velocity observed in December 1999

2.1. The reference mean wind speed

The observed velocities in various meteorological stations have to be compared in


the same conditions. Now, the wind force depends on the close environment of the
*Corresponding author. Tel.: +33-02-40-37-20-21; fax: +33-02-40-37-20-60.
E-mail address: sacre@cstb.fr (C. Sacr!e).

0167-6105/02/$ - see front matter r 2002 Published by Elsevier Science Ltd.


PII: S 0 1 6 7 - 6 1 0 5 ( 0 2 ) 0 0 2 2 9 - 5
1164 C. Sacr!e / J. Wind Eng. Ind. Aerodyn. 90 (2002) 11631171

place where it is observed and obviously every station has a specic environment.
Therefore, it is necessary to correct the measured wind velocity to get the data for
reference conditions. This correction can only be done by using the mean velocities
which are the only wind data at ground level which can be converted from a
particular site to another one.
One denes the open country as the reference environment, also for convenience
sake because the environment of meteorological stations is often close to it.
Therefore, the mean reference velocity V% ref is the 10-min mean wind velocity which
can be measured at 10 m height on a at site without any obstacles with farmland
roughness type z0;ref 0:05 m:
Meanwhile, the environment of meteorological sites is rarely homogeneous. Every
single site is characterised, in a radius of a few kilometres, by its location in the
relief (hill, valley, etc.), by the nature of the terrain (geographic distribution of
urban zones, open country, stretch of water, etc.) and in a radius of a few hundreds
of meters by close obstacles (edges, trees, houses, etc.). Taking into account
these physical elements to estimate the wind mean velocity, V% ref can be summarised
by dening three coefcients of disturbance associated to the measured velocity
V% met :
V% met
V% ref ;
CR CT CS
where CR is the roughness coefcient, CT is the topography coefcient, and CS is the
wake coefcient. The coefcients CR and CS are obtained from the method used in
the WASP software [1,2], the topographic coefcient CT is given from a
bidimensional analytical formulation proposed in Sacre! [3] and deduced from
Refs. [46].
Let us note that the corrections are applied also to turbulence intensity Imet aIref
and therefore to the gust coefcient for every direction sector around the
meteorological measurement mast.
Peak velocity V# met and mean velocity V% met observed during the same period of
time (10 min in French meteorological stations) are linked statistically by the
relationship
V# met
V% met :
Gmet
Gmet is the gust coefcient [7], distinctive of the station site, dened by
V#
G 1 gI;
V
where I is the turbulence intensity in the wind direction and g is the mean peak factor
whose value is 2.86 in French conditions measurements that gives Gref 1:54:
There are about 160 main meteorological stations in France with digital records of
daily maximum instantaneous wind speed, since 1949. But only half of them have
been selected whose local environment was rather easy to correct.
C. Sacr!e / J. Wind Eng. Ind. Aerodyn. 90 (2002) 11631171 1165

2.2. Wind velocity extreme value statistics

To simplify the analysis and for a problem of cost of data, we have not used the
method of independent storms, but the standard Gumbel method applied to annual
maximum of peak wind speed as discussed, for example, by Cook [8,9].
F X is the probability that the maximum velocity x does not exceed the value X
during a year. The distribution function F X of the maximum velocities is tted by
an extreme value distribution law as the Gumbel law
  
X a
F X exp exp  ;
b
where a is a position parameter (mode) and b is a scale parameter proportional to the
standard deviation of the distribution (1=b is the dispersion). The parameters a and b
were calculated for the reference velocity of all selected French meteorological
stations by a least-squares adjustment method. One associate to this, the idea of
return period T which is equal to the average time between two exceeding the
velocity X : T is given by T 1=1  F X ; where T is expressed in number of years.
The reports of Sacre! [10,11] present the cartography of extreme mean reference
velocities of 50 years return period for France, built with this approach.

2.3. Evaluation of the return period for the velocity observed in December 1999

The Me! te! o France observations for the two storms were rst, the daily maximum
mean speed V% met with the associated direction D% met and second the daily maximum
instantaneous velocity V# met with the associated direction D# met as well. The directions
are dened by sector of 101.
The local environment was taken into account for the daily maximum mean
velocity and for the daily maximum peak velocity which did not always happen in
the same period of the day.
Hence
V% met V# met 1
V% ref;1 and V% ref;2 :
CR CT CS CR CT CS Gmet
Due to the random characteristic of the g coefcient, the maximum value of V% ref;1
and V% ref;2 was unchanged. Then, the velocity V# ref;max 1;2 is the theoretical peak value
deduced from the two previous velocities, i.e.
V# ref;max 1;2 MaxV% ref;1 ; V% ref;2 Gref :
It is the peak velocity which would have been observed on average at 10 m altitude in
open country, during the December storms. More often V% ref;2 is higher than V% ref;1 :
For every station the ratio V% ref;max 1;2 =V% ref;T is considered, corresponding to the
return period T of 50 years because the mean velocity V% ref;50 is the reference velocity
in the Eurocode 1Actions of the wind. Map 1 gives the isovalue curves of this ratio
r V% ref;max 1;2 =V% ref;50 : It gives also the isovalue curves V% ref;50 used to set up the wind
map of Eurocode.
1166 C. Sacr!e / J. Wind Eng. Ind. Aerodyn. 90 (2002) 11631171

Map 1. Reference velocities observed during the storms of December 1999 (ratio r), compared to reference
velocities calculated locally (50 years return period V% refeurocode in m/s).

Map 1 underlines the regions that are more concerned by the storms of December
1999; they correspond to a ratio over 1, whose maximum is about 1.2. Table 1 gives
corresponding values of reference velocity ratios and return periods. The calculation
of the return period Ttemp corresponding to V% ref;max 1;2 is achieved with respect to the
extreme value distribution determined for any single station and dened by the
Gumbel law coefcient a and b:
1 1
Ttemp with Y V% ref;max 1;2  a :
1  eey b
One should bear in mind that in the southern part and east of the Paris region and
the north-west of Aquitaine region the return period exceeded 400 years, which
actually represents an increase of only 1520% of the 50 years return period
reference velocity. Moreover, the return periods Ttemp are greater than 400 years,
which is not statistically signicant due to the fact that the initial distributions are
based on a period of about 40 years.
C. Sacr!e / J. Wind Eng. Ind. Aerodyn. 90 (2002) 11631171 1167

Table 1
Return period and value of the ratio r V% ref;max1;2 =V% ref;50

V% ref;max1;2 =V% ref;50 0.90 0.95 1.05 1.10 1.15 1.20

T (year) 15 25 120 250 400 >2000

3. Analysis of destruction from the point of view of wind exposure

3.1. A safety margin

The actions of wind, calculated on the basis of dynamic pressures Q# ref;site ; in


Eurocode [12] are characteristic actions, i.e. they are supposed to be weighted by a
coefcient g over unity (g 1:5) for the verication of ultimate limit state. Knowing
that during the storms of December 1999 the ratio V# ref;max 1;2 =V# ref;50 2 was under 1.5
in the whole country, one can conclude that from the Eurocode point of view, the
loads did not exceed the ultimate limit state.

3.2. Effect of roughness

551 cases were analysed, but the geographic distribution of the cases is not
statistically signicant since some areas, although very much concerned by the
storms, did not answer the enquiry. The questionnaires at least allowed to establish
correlation between damaged building parts and roughness types. The effect of
topography was not studied because the data included in the questionnaires were not
detailed enough to be used in that manner. This is also the reason why topography
was not dealt with in the previous paragraphs.
The damaged building environment was ranked with four roughness types: sea,
open country, suburban areas and city.
Table 2 presents the percentage of buildings being damaged with respect to the
whole building set in the same environment.
The analysis of Table 3 leads to three comments:
The buildings located in roughness sea and open country suffered more structural
damages and outside coating damages than those located in areas of larger
roughness. More severe damages were observed (partial or total roof taking off)
on buildings in front of an open eld with low roughness of open country or
bush type.
The buildings located in roughness city and suburb suffered more damages to the
openings and woodworks than those located in areas of smaller roughness. One can
wonder if the older openings in cities could explain that.
The proportion of damaged roofs and cladding does not show any tendency with
respect to roughness type of the environment.
More severe damages were observed (partial or total roof taking off) on buildings
in front of an open eld with low roughness of open country or bush type.
1168 C. Sacr!e / J. Wind Eng. Ind. Aerodyn. 90 (2002) 11631171

Table 2
Percentage of damages sorted by environment type

Structure Woodwork Openings Outside Small cladding Large cladding Roofs


coatings elements elements

City 14.75 22.13 22.95 5.74 27.05 27.87 6.56


Suburb 9.09 22.89 22.89 5.38 37.04 28.62 12.79
Country 26.98 12.70 17.46 9.52 26.98 50.79 6.35
Sea 42.86 0 0 7.14 28.57 28.57 14.28

Table 3
Sample inuence on extreme wind speed estimation

Colmar time series Sample Coefcient V% ref;50 sV Cond.


size correlation (m/s) interval (m/s)

Without correction 42 0.972 25.6 1.8


Correction before 1975 42 0.988 26.2 1.5
Correction before 1982 42 0.991 27.0 1.6
Correction before 1975, excluding the highest value 41 0.990 25.1 1.4
Partial series after 1975 24 0.946 25.7 1.9
Partial series after 1975, excluding the highest value 23 0.972 23.4 1.4
Mulhouse time series 50 0.990 25.4 1.2

3.3. Problem of openings and roofs

The examination of damages shows that some roofs were heavily damaged (lift of
the framework, taking off of large roong elements) due to the breaking of the
envelope on the windward side by breaking open of openings. In this case, the
internal pressure becomes equal to the external pressure of the windward wall, which
induces a high lift effect on the roof. There is a problem of renement of rules related
to the observed weaknesses of the windward openings, this fragility being possibly
due to the ageing of openings. One could request for the envelope a failure
probability equal to what is requested for the structure.

4. The variability of extreme wind speed distribution

4.1. Statistic quality of the samples

The representativeness of a sample of annual maximal reference wind speed must


be studied by comparison with another sample of the same type. Fig. 1 shows as
example simultaneous time series of wind speeds of two near meteorological stations
(Mulhouse and Colmar), one can admit, that are submitted to the same climatic
constraints. The correlation coefcient between the two series is 0.44, but data of
C. Sacr!e / J. Wind Eng. Ind. Aerodyn. 90 (2002) 11631171 1169

30

COLMAR
COLMAR COR.
REFERENCE WIND SPEED in M/S

MULHOUSE
25

20

15

10
55 60 65 70 75 80 85 90 95 100
YEARS
Fig. 1. Time series for annual maximum reference wind speed.

Colmar show a decit of speed before 1975 could be induced by an unknown change
in anemometer type and location. Series after 1980 seem more reliable; so, by
comparison between the two parts of the series, one multiplies the rst part by a
coefcient in order to homogenise the complete set. A good criteria is the increase of
the correlation coefcient with the time series of Mulhouse. While taking a factor of
1.3, the coefcient of correlation increases up to 0.92.
Then what sample to choose? Table 3 indicates different values of the mean wind
speed of 50 years return period got from different samples while tting by a least
squares method a law of Gumbel applied to the reference mean wind speed. One also
noted the result of the station of Mulhouse that can be used as comparison.
All values of wind speed predicted from data of Colmar are in the interval of
condence to 95% of the wind speed got with data of Mulhouse: 25.472.4 m/s.
Removing the highest value of the series (December 1999) affects the result while
lowering the wind speed evaluation of 1 m/s. Following Cook [9], this value would be
outside Gumbels condence limits.
The homogenisation of the sample while putting to level or in removing a part of
data does not mislead a signicant modication of the result (less than 1 m/s) to the
condition of not having to to correct a too big part of the sample.

4.2. A new extreme wind speed map for France

Map 2 gives a new estimation of extreme mean reference wind speed over France
established from all available data since 1949 to 1999 including the 99 storms
observations [13]. Particular attention has been given to the sample homogenisation
of each wind speed time series as shown in the case of Colmar.
1170 C. Sacr!e / J. Wind Eng. Ind. Aerodyn. 90 (2002) 11631171

Map 2. Isolines of mean reference wind speed for a 50 years return period in m/s.

By comparing the new and the old map visible on the background of Fig. 1, we
can see the inuence of the storms of December 1999: The mean reference wind
speed for a 50 years return period increases on the central part of Atlantic coast
(+1.5 m/s) and the isoline 28 m/s goes more inland.

5. Conclusions

In spite of the mediocre quality of a part of the meteorological data of wind, while
working on the homogeneity of samples, one can get a satisfying spatial extreme
speed distribution. New high values of wind speed have an inuence on the
distribution adjustment; meanwhile, it would be difcult to design a new map at each
new extreme event, as we have to build for more time than the return period of a
strong storm. What do we have to do?

Acknowledgements

This study was supported by the French Minister of Environment and Country
planning, and carried out at the Nantes establishment of CSTB.
C. Sacr!e / J. Wind Eng. Ind. Aerodyn. 90 (2002) 11631171 1171

References

[1] A.M. Sempreviva, S.E. Larsen, N.G. Mortensen, I. Troen, Response of neutral boundary layers to
changes of roughness, Boundary-Layer Meteor. 50 (1990) 205225.
[2] M.D. Perera, Shelter behind two-dimensional solid and porous fences, J. Wind Eng. Ind. Aerodyn. 8
(1981) 93104.
[3] Sacr!e C., Ecoulement de lair au-dessus dun site complexe1"ere partie: Etat des connaissances
actuelles, Rapport CSTB EN-CLI 88.12 L, 1988.
[4] R.E. Britter, J.C.R. Hunt, K.J. Richards, Air ow over a two-dimensional hill: studies of velocity
speed-up, roughness effects and turbulence, Q. J. Roy. Meteorol. Soc. 107 (1981) 91110.
[5] D.R. Lemelin, D. Surry, A.G. Davenport, Simple approximations for wind speed-up over hills,
J. Wind Eng. Ind. Aerodyn. 28 (1988) 117127.
[6] H.W. Teunissen, M.E. Shokr, A.J. Bowen, C.J. Wood, W.R. Green, The Askervein hill project:
wind-tunnel simulations at three length scales, Boundary-Layer Meteor. 40 (1987) 129.
[7] L. Kristensen, M. Casanova, M.S. Courtney, I. Troen, In search of a gust denition, Boundary-Layer
Meteor. 55 (1991) 91107.
[8] N.J. Cook, Towards better estimation of extreme winds, J. Wind Eng. Ind. Aerodyn. 9 (1982)
295323.
[9] N.J. Cook, Improving the Gumbel analysis by using M-th highest extremes, Wind Struct. 1 (1) (1998)
2542.
[10] C. Sacr!e, Estimation des vitesses extr#emes du vent en France m!etropolitaine, Note CSTB, EN CLI
93.9 R., 1993.
[11] C. Sacr!e, Charact!erisation des vitesses du vent observ!ees au cours des temp#etes des 26, 27 et 28
d!ecembre 1999 du point de vue de la statistique des vitesses extr#emes du vent pour la France
m!etropolitaine, Rapport CSTB, EN-AEC 00.26 C, 2000.
[12] Eurocode 1: Basis of design and actions on structures, Part 24: Actions on structures, Wind actions,
ENV 1991-2-4, 1991.
[13] C. Sacr!e, Une nouvelle estimation des vitesses extr#emes du vent en France m!etropolitaine, Rapport
CSTB, EN-AEC 00.111 C, 2000.

You might also like