Field Measurements of Strong Winds (With Interpretation) | Wind Engineering

In this article we will discuss about:- 1. Introduction to Strong Winds 2. Historical Perspective of Strong Winds 3. High Wind Speeds 4. Description of the Atmospheric Boundary Layer 5. The Structure of Winds and Its Measurement 6. Description of Statistics 7. Some Significant Measurements 8. Conclusion.

Introduction to Strong Winds:

The occurrence of strong winds, to their measurement and to the interpretation of those measurements in the context of the way such data are used for the purpose of the design of structures.

In this paper high wind speeds are considered to be those that, at a height of 10 metres have a sustained mean in excess of 20 metres per second.

The number of full-scale measurements made of wind in the planetary boundary layer is relatively small, and those that are made fall into two broad categories. Firstly, most countries maintain a database of wind measurements, with anemometers placed at strategic locations.

This generates the largest number of data about boundary layer winds. Secondly, and much more rarely, full-scale measure­ments of wind data in the field for specific purposes are sometimes mounted. These experiments are often for the purpose of investigating specific situations so as to ascertain the characteristics of the wind in such circumstances.

The distinction between light winds and strong winds has an important bearing on measurements as we shall see later. In particular the statistical techniques currently used, assume that stability exists in the atmosphere, both in terms of random fluctuations achieving stationarity, and of the atmosphere itself being neutrally stable. If the adiabatic lapse rate is – 1°C/100 metres then the atmosphere is considered to be neutrally stable.

This temperature gradient fundamentally affects the structure of wind in the boundary layer. In light winds, fluctuations in the flow may be supplemented by convective action or vertical transportation may be blocked by thermal inversion. These factors may have considerable influence on the turbulent fluctuations in the atmosphere under light winds.

There is considerable difficulty associated with the measurements themselves. Not only are there problems for the physical integrity of the anemometers, but also the precision of the measuring devices has been called into question. The recent attempts to use novel transducers have shown that the measurement of short term gusts (say of one second duration) has been impossible until the latter half of the last decade.

Historical Perspective of Strong Winds:

The rotating cup anemometer was invented by John Robinson in 1846, and consisted of three or four rotating cups. This is still the most common form for the measurement of wind speeds in the boundary layer. The use of such devices was researched by W.H. Dines at the end of the 19th Century.

His experiments with the simultaneous measurement of wind speeds close to the top of his house showed that the different anemometers did not indicate the same speeds, and so he progressively raised them until at a height of 4.5 metres above roof level they were all reading similar values.

During the last 50 years a revolution in data handling and collection has made possible enormous strides in field measurements of the behaviour of wind. A number of technological developments were made which have subsequently become common­place in the allied fields of meteorology and atmospheric physics. These develop­ments have centered on the use of various types of laser, sonic or microwave devices for monitoring the movement of air. Many measurements are now made from orbiting satellites.

The major problem for measurements of the boundary layer using remote sensing is one of resolution. It is quite possible to use a microwave source and to monitor winds at different heights through the measurement of reflected images backscattered from different levels. The problem is that at present the minimum height at which such measurements can be made is 500 metres, and with a resolution of 500 metre intervals.

At the time of writing the most secure forms of data capture inside the boundary layer are still those associated with readings that are made from a device that is supported from the ground. This effectively limits such readings to a height of 300 to 400 metres. Even so mounting experiments in which the structure of the wind is monitored throughout this height range is sufficiently difficult and expensive that the number of attempts is exceedingly small.

High Wind Speeds:

The highest speed ever measured was that registered at the USAF Air-Force base at Thule Greenland at which a peak gust of 92.5 m/s was monitored. It is generally reckoned that the windiest location on Earth is the Commonwealth Bay George V Coast of Antarctica. At this location a daily mean wind speed of 48.3 m/s was monitored on 21st/22nd March 1951.

Most countries maintain records for their local area. The original purpose of monitoring the wind speeds was to try to predict the forthcoming occurrence of very strong winds that could be dangerous to shipping. This situation is quite typical of the early impetus for the monitoring of wind activity and in turn reflects the fact that winds over the oceans are stronger than those over land areas. Early measurements then were more likely to be instituted by seafaring nations.

High surface winds are themselves a function of entire weather systems tracking over the surface of the planet or of locally induced phenomena. These mechanisms may include cyclonic weather systems, fronts and depressions, tornados, and thunder­storms and may be modified by such effects as Lee waves, or mechanisms associated with gravity or thermally induced movements of air in mountainous regions.

Description of the Atmospheric Boundary Layer:

For the purposes of modern design it is necessary to describe the nature of the wind in terms of its structure at a location, and of its probability of recurrence. Perhaps the most important factor though, is that in keeping with the spirit of scientific inquiry, such models must be tested against full scale measurements and must earn their status through verification.

An impetus to the measurement of the structure of strong winds was given by the quantum leap in the conceptualisation of the processes at work through the work of Davenport in the early 1960’s. The description of the nature of wind in terms of its structure close to the surface was the single most important concept in wind engineering during this century. Several experiments were quickly mounted to test the theory and the outline was quickly substantiated and enlarged by other significant contributions to the field.

These methods require a description of the variation of the mean wind profile with height, the turbulence associated with strong winds, the intensity of that turbulence, the gust maximum velocity, the relationship between the mean wind-speed and the peak gust and the spectrum of wind gusts with respect to their frequency content.

In addition, the size of gust eddies was of particular concern because there was an early understanding that the interaction of structures with the gusts was crucially dependent on the ratio of the structure to gust size. Ground friction forces, turbulent eddies and shear forces at the lowest levels are affected by increases in the mean wind velocity and measurements at high speeds are therefore essential.

The Structure of Winds and Its Measurement:

The atmospheric boundary layer in the first 2000 metres is now discussed and the parameters are considered in a little more detail together with comments on their measurement.

i. Velocity and Direction:

Measurements of the velocity of the wind and its direction of attack are seemingly straightforward. Unfortunately this is not the case. Despite the design of several different types of anemometer the rotating cup type mentioned above and introduced through the work of John Robinson, or the mechanically similar Gill-type propeller and vane type are easily the most robust types for use in rugged field environments.

Of other types introduced recently sonic devices that effectively measure wind velocity by measuring the time taken for a sonic pulse to travel between transmitter and receiver, are perhaps the most common. The harsh outdoor environment is particularly hard on researchers who do not respect the contrariness of nature.

The rotating cup or Gill type propeller anemometer is itself a mechanical device with mechanical properties. A force acting on such devices may be in error by as much as 2 m/s. Also reported is an error of as large as 6% in the reading obtained from a rotating cup anemometer if the mounting is within 15 diameters of the boom.

Similarly a fin behind a propeller or underneath a group of rotating cups also takes a finite time to respond to a change of direction. In this case a reducing force as the vane acquires the “correct” direction leads to a dead zone that increases the apparent variability of the direction of attack.

Radar type devices have recently been used to study the effect of winds higher in the atmosphere. Very recently the different techniques have resulted in devices that are close to being usable in the wind engineering field rather than purely for meteorologi­cal research purposes.

An example is given of German experience in which a SODAR device has been developed and is usable to resolve wind speeds at 10 metre intervals from ground level upwards rather than from the 1 Km upwards previously attainable. Unfortunately the work of Vogt and Thomas shows that the use of the acoustic Doppler SODAR device whilst promising is still prone to differences from more conventional measuring techniques of 30%.

Such devices can be used to generate an electrical analogue of both speed and velocity that can be converted into digital form and stored in a computer. Wind records of this type can then be scanned for both mean and peak data with appropriate averaging times.

If the wind speed and direction data are sampled at a rate of once a second, then for a single anemometer over 63 million data must be archived for one year’s worth of readings. For this reason hourly mean data are normally archived, and the majority of high quality data is discarded.

Davenport’s work of the early 60’s opened the study of dynamic behaviour in which the gustiness of the wind must be defined if the effect on structures is to be predicted rationally. This description must, of necessity, be statistical in nature.

From the basic measurement of wind-speed and of direction can be deduced, through computer analysis several important parameters, such as:

1. Intensity of Turbulence (I_u):

For a wind record the mean velocity is calculated appropriate to a preselected averaging period. The standard deviation of the fluctuations of the wind velocity about this mean value is then computed.

2. Gust Factor (K):

Gust Factor (K) is the ratio of the peak gust to the mean wind speed.

3. Peak Factor (g):

Peak Factor (g) is a related quantity and is expressed as g = (K – 1)/I_u.

The provision of a record of wind data in a digital computer allows further analysis to be conducted and a series of other measures of the statistical properties are possible.

ii. Spectrum of Horizontal Gustiness:

Since the energy in the wind impinges on a structure through the mechanism of an interaction of gusts of wind with the structure then it is important to know how much energy is present as a function of frequency. A spectrum may be generated directly from wind speed data.

Alternatively the spectrum of pressures may be presented through a linear transformation, or the ordinates may be weighted by the frequency to produce a nS(n) spectrum in which the power in the signal is represented as a function of frequency or of reduced frequency. The reason for presenting the data in this fashion is that in the nS(n) form the value Of the energy in the wind can be fed directly into equations for the response of a structure to wind.

This implies that the response of a structure to a wind load can be thought of in terms of two parts. The first is essentially quasi static and may be thought of as the response to the mean wind. The second part is dynamic and is the structure’s response to the sub inertia part of the spectrum at higher frequencies. There are two implications. The measurement of the wind speed and its representation in the frequency domain requires the standard assumptions for spectral analysis to be true.

The data set forming the record of fluctuating wind speeds must therefore form a stationary sample if the spectrum of wind fluctuations as a function of frequency is to be valid. Secondly the separation of the sub inertial range of the spectrum from the inertial range must be valid.

The spectral gap must exist in order to remove trends from the data sets. This spectral gap is itself assumed to exist largely as a result of investigations such as that at Brookhaven. The amount of decoupling between the microclimatological sub inertia regime, and the macro-climatological low frequency regime is exaggerated in such plots since the ordinate is multiplied by frequency.

In wind tunnel tests the separation of the micro and macro climatological parts of the spectrum is induced. This in turn creates a condition in which stationarity of the remaining turbulent spectrum is much easier to achieve. The investigator in the field has no such luxury.

He must attempt to remove the non-stationary data before beginning the analysis. It is a salutary warning that in reviewing many of the largest studies of wind observations the mention of tests for stationarity is one of the rarest events reported. Some authors have attempted to circumvent the problem by using the autoregressive moving average method. However, it is difficult to determine variance errors in this way.

iii. Vertical Profiles:

Measurements made at a series of vertically separated heights can provide a direct measurement of the vertical profile of the boundary layer either in terms of means or of gusts. Averaging over periods such as one hour will produce (normally) well behaved vertical velocity profiles.

Turbulence intensity variation with height can also be presented in this manner. If the averaging process is reduced to the region of a few seconds then a gust profile may be constructed. In the field, vertical transportation of wind gusts tends to make the gust velocity profile and the short term turbulence intensity profile rather variable quantities. Nevertheless they are such important parameters that several attempts have been made to measure them.

In addition to the direct measurements noted above there are some inferences that can be made:

Integral length scales *L_i, *L_i, *L_i.

Integral length scales are a measure of the average size of a gust and are made in the x, y and z senses. There are nine integral scales, and they are of particular interest in scaling wind tunnel measurements.

The measurement of the integral length scale in the u direction can be achieved by integrating the autocorrelation function from a single point (or anemometer). The autocorrelation is itself the inverse Fourier transform of the spectrum. Alternatively, it can be determined by measuring the maximum in an nS(n) spectrum.

This is because the autocorrelation function effectively plots the correlation of a shaft of air moving past that point with itself. A zero value of the autocorrelation function would indicate that a coherent mass of air was exhausted at that distance and would therefore effectively define the limit of the coherent gust.

Unfortunately the autocorrelation function correlates poise signals as well as the useful information and the function often does not decay to zero. Under such circumstances it is common practice to invoke Taylor’s hypothesis to convert from an integral time to an integral length. The peak in a spectrum merely indicates the most likely wavelength associated with gusts. Energy at other wavelengths is also present.

Many difficulties have been experienced with the measurement of integral lengths. This is perhaps not surprising, given the fact that it is rare to test for stationarity, but additionally the autocorrelation function averages information at all frequencies present.

Considering that gust sizes are associated with different frequencies in the microclimatological peak, and these are mixed with trends from low frequency energy, it is perhaps surprising that no attempts to measure integral lengths as a function of frequency appear in the literature.

Values of ^yL_i and ^ZL_i may be obtained in a similar manner with the exception that cross correlations between anemometers in the vertical or the lateral sense must be performed.

The roughness length Z_o, The friction velocity U_*, Surface drag C_D can all be deduced from the measurement of mean and gust speeds, via simple relationships.

The parameters described above can be determined only if the atmosphere is well behaved, the data are stationary and a stable terrain exists. Clearly there are many situations when this is not the case yet measurements from such situations must still be sought.

Description of Statistics for Strong Winds:

Moving from the meteorological to the wind engineering domains, we encounter a field that draws heavily from both. The description of statistics about the return of strong winds normally gives as an end product, a description of a wind in terms of its probability of occurrence.

To obtain this basic wind-speed suitable for any particular location analysis of the wind data collected over a number of years is normally effected. Such techniques have developed and are epitomised by the description by Cook and Prior (1987). Recording stations are normally composed of a series of anemometers that are run not only by government agencies.

The data must be corrected to an “effective height” before incorporating normally as measures of mean wind speeds averaged over a suitable period such as one hour. The statistics of extremes may then be used to provide a description of the likelihood of the occurrence of a particular wind-speed at some location.

This basic technique can be supplemented to produce more efficient estimates drawn from fewer data. Firstly the data relevant to storms can be accumulated rather than for instance yearly data. Secondly, the Best Linear Unbiased Estimator (Lieblein BLUE) can be used to give more emphasis to the data for shorter periods thereby estimating the extremes more efficiently.

In general terms there seem to be some general principles that reduce confusion in the estimation of extremes. Firstly it is better not to mix mechanisms. If strong winds are generated by cyclonic weather systems and by thunderstorms, separation of the two types of data reduces errors.

A study of thunderstorm activity showed that when the mechanisms were separated then the winds generated by thunderstorms at the stations that are located far from the coast are underestimated if the other mechanisms are not separated before the analysis.

The use of Fisher Tippett type I distributions are theoretically applicable to only one mechanism at a time. The use of a Fisher Tippett type II distribution to simulate mixed mechanisms is merely pragmatic.

There is also another very useful method for establishing wind-speeds when a very strong storm has passed through an area causing damage or severe damage. This method that may be termed “reductionism” is epitomised by the approach after Cyclone Tracy or after Hurricane Hugo.

In the latter Sparks assembled data from several military bases and airports mostly with anemom­eters at nonstandard heights to establish the probable area in which the strongest winds occurred. He then used the diaries of some of the local inhabitants to establish the most probable wind speeds.

Sparks and his coauthors, commented that the observation of sea height was of particular use since its rise was correlated with the occurrence of the strongest winds. He was able to ascertain that at Bulls Bay in South Carolina the maximum gust was approximately 65 m/s, and the mean at the height of the storm was 38 m/s. Sparks and his coauthors have produced a list of types of damage correlated with gusts of various speeds in unprotected areas.

Some Significant Measurements of Strong Winds:

Davenport’s contributions (1961, 1965) to the understanding of the structure of wind and in recommending an efficient way in which to represent this structure, were based on a review of measurements made in many different conditions and in several different countries and terrains.

Accordingly he recommended values for the power law index, the gradient height and the surface drag coefficient from his review of these measurements. He described three terrain categories with rising values for the three parameters in each case.

The early theoretical input coupled with references to many measurements led to several investigators mounting experiments to verify or modify the original descrip­tions.

Harris (1971) was able to use the Rugby radio station that is located in level terrain with little obstruction anywhere nearby to make measurements at four different heights. He then went on to extend the areas of investigation.

Deaves and Harris (1980) extended this work with measurements conducted at Cardington and were to present considerable information about profiles, spectra, the variation of turbulence intensity with height and integral length scales. They showed integral length scales to be a function of the terrain roughness, although the difficulties with analysis of the autocorrelation functions was apparent.

Duchene Marullaz (1975) conducted a series of experiments at Nantes. There were three independent masts each of which had anemometers at 10,20,40 and 60 metres height. Measurements of Gust factors, Integral lengths, spectra, U_*, z_o, and were made.

Results were presented for two gales. The velocity profile was found to give a value of in the range of 0.3 to 0.36, z_o in the range 1.15 m to 1.65 metres and U, in the range 1.6 to 2.2 m/s. Values of turbulence intensity were noted to reduce with increasing height dropping from 0.3 at 10 metres to 0.2 at 60 metres.

Integral length scales showed a consistent variation with height. However, the claim that the values obtained were similar to those measured by Harris (1971) was only true in that they were of the same order of magnitude. When the reported value of 0.3 to 0.36 was taken into consideration then the value of ^XL_u would be 190 metres rather than 300 metres at a height of 60 metres. Spectra had a similar form to that proposed originally by Davenport.

A large effort had, in the meantime been mounted in Hong Kong at Cape D’Aguilar. This consisted of a ten story experimental station located on the foreshore at the south easterly tip of Hong Kong. Velocity and direction measure­ments were obtained from five vertical lines of anemometers at heights of 13, 28, 43 and 61 m. Mackey and subsequently Choi (1983) reported on measurements made during five different typhoons.

The mean velocity profile provided a steady value of 0.19 for wind speeds of 10 m/s up to nearly 40 m/s. However, in the case of the gust profile it was found that a time lag generally existed.

A gust at the upper levels would be followed after a delay by a maximum at the lower levels. Plots of the vertical profile during typhoons when maxima are reached at each recording level were made. Instantaneous values of fitted to such profiles vary considerably (from 0.08 to 0.4).

The implication is that is that during very strong winds there is considerable vertical transportation, and that if, over the duration of a storm, the maximum gusts as a function of height are presented then the gust velocity profile reaches an asymptote soon after a height of 10 metres.

Despite problems with the analysis of autocorrelation functions, integral length scales were proposed which were claimed to agree very well with the Rugby measurements. However, these values are 3 to 4 times larger than those established later.

Ishizaki (1983) summarised a considerable amount of work conducted in Japan and which had until that time been published only in Japanese. The work of Arai et al conducted form 1964 to 1967 monitored six typhoons at six different heights near the coast, using aerovane anemometers.

Of particular concern was an attempt to establish a relationship between the mean wind-speed and (n) the inverse of, the power law exponent. There was so much variation in the relationship that the authors were forced to dismiss a rational relationship. Subsequently the work coordinated by the Architectural Institute of Japan at Tarama Island, Okinawa, between 1975 and 1978 was reported.

Rotating Cup anemometers were used, and a calibration established that they followed a 4 second periodicity with about 90% accuracy. Once again an attempt to study the relationship between (n) and mean wind speed for typhoons was made, and on this occasion a tentative relationship was established.

When the hourly mean values were taken then there is a linear increase in the value of (n) as the mean wind speed increases. There is, however, a good deal of scatter. Turbulence intensity showed considerable scatter at low wind speeds.

However above 20 m/s the variation was limited to 8 to 15% and approached an asymptote of about 10%. This result was obtained from the tests at Tarama and was confirmed by other tests at Satoura from measurements made at a height of 40 metres. These data established a good linear relationship between the gust factor and I_u.

These data led Ishizaki to comment that turbulence is an inherent property of an individual typhoon. It is modified by the terrain roughness, but this is not essential to produce the turbulence. This was a very clear indication that each typhoon has its own character. Notwithstanding this the data were used as the basis for standard values for typhoon wind profiles and for values of, Z_o, and I_u for winds of mean speed up to 60 m/s at a height of 10 metres.

By the late 70’s meteorologists established the velocity profile to a height of 10,000 metres (Powell 1979), through a combination of buoys, ships, aircraft and radar measuring devices. These measurements were generally able to discriminate only to 500 or 1000 metres height resolution, and in the absence of measurements in the first 500 metres the assumption was that the power law held for a neutral atmosphere.

The data from Wallops Island were obtained at 15 metre intervals at 5 levels. The usual cup and vane anemometers were supplemented with the use of split film devices for the accurate measurement of turbulence properties. The measurement of temperature profiles was of particular significance in that the authors were able to identify low level inversions that created a separated regime.

This in turn led, at certain times to the blocking of the low frequency wind excitation. The experiment addressed the problem of the measurement of integral length scales. A comparison of the measurement of integral lengths through the integration of the autocorrelation function and through the establishing of the wave number at the peak of the Von Kantian spectrum led the authors to identify that the two methods only gave similar results when the blocking of the low frequency range winds occurred.

Since the spectral method represents measurements only in the sub inertial higher frequency range and the autocorrelation method includes the low frequency data as well, the clear implication of this result is that much of the variation in previous (and subsequent) measurements is caused by a lack of checking for stationarity of data, since data recorded for only a short period and having significant low frequency components in them must test positive for trend-induced non-stationarity.

Measurements were made at Nakagawa and at Satowa at 5 levels, 130m to 150m from the beach and in an array of 5 towers with anemometers all located at 40 metres height in a 190 metre line along the coast respectively. Above 20 m/s at both sites the value of I_u closely approached 10% as an asymptote. The trend continued to 60 m/s.

Teunissen (1983) reported on experiments conducted under the aegis of the Atmo­spheric Environment Service into flow over three dimensional hills in neutrally stable conditions. This started with the Kettles hill experiment and continued with the Askervein experiment.

At Kettles hill ten ten metre towers were placed on the hill in a straight line array, and were supplemented by ten movable 3 metre tall towers. A sonic anemometer at the summit of the hill was used for turbulence measurements. Mean wind speeds in the range 15 to 20 m/s were measured.

These experiments were essentially to underwrite the basis for design using either wind tunnels or the Jackson and Hunt formulation. Both of the modelling techniques were found to overestimate the full scale measurements, the wind tunnel by more than the mathematical model.

Maeda and Makino (1988) reported on measurements made in Fukuoka City and included measurements from the time of a typhoon. In this case the Gill-type anemometers were supplemented with the use of ultra-sonic anemometers for the measurement of short term fluctuations.

The measurements were made at a height of 38 metres and the typhoon maximum gust was 51 m/s with an hourly mean of 24.7 m/s. Measurements of integral scales and of the turbulence intensity were made using an Autoregressive technique to establish the spectral ordinates.

The spectra for the U, V and W directions were compared with theoretical curves based on the Von Karman description of a spectrum for high wind speeds. The authors reported that in the u direction the agreement was good, but that in the v and w directions agreement was only obtained if values of ^vL_x and ^wL_x were substituted for ^uL_x.

Naito (1988) reported on measurements taken from an offshore tower located 1 Km offshore with two sonic anemometers located 6.85m and 23 metres above sea level. The measurements were made for a large number of lower velocity winds and for a typhoon.

During this period of stronger winds the characteristics of gust factor were very similar to the low velocity regime until there was large velocity excursion up to 50 m/s. The gust factor then rose from a range of 1.2 to 1.5 up to 2.0 to 3.2 for a period of about 50 minutes.

Turbulence intensity measurements for high wind speeds showed small values and measurements of spectra showed the peak value to be considerably lower than for those made over land. The implication is that over the ocean large scale turbulence dominates.

Flay and Stevenson used eight towers over a 315 metre fetch to make measurements of integral lengths at wind-speeds of up to 20 m/s. They observed that the use of the spectral peak method always gave values that were less than those obtained using the autocorrelation method, and went on to suggest ways in which the autocorrelation could be used to obtain similar results.

The main method of measuring the integral time over the period up to a reduction to 1/e effectively excludes the very low frequency data, although there would also be an element of very high frequency as well. This high frequency noise comes from the electronics and correlates at all-time delays leaving an unwanted constant offset in the autocorrelation function.

The authors went on to show comparisons of results for ^XL_u, ^xL_v, ^xL_w, ^yL_u, ^yL_v, ^yL_w, with measurements quoted by Counihan, ESDU, Shiotani and Teunissen. Their claim that the comparison was reasonable must be seen within the context of 100% differences in some values.

Two towers in Tokyo at 31 metres and 55.7 metres height, and one in Yokohama at 86 metres were reported. Ultrasonic transducers were used and the authors criticised the older mechanically based devices as not being capable of resolving short duration gusts.

The authors show measurements of gust factors for 1 second averages although there is a great variability to these measurements at lower wind-speeds. However, as the speed rises then the values of I_u and gust factor reach stable values. Additionally the U direction spectrum is quoted and agrees well with the Von Karman formulation.

Snaebjornsson et al (1992) presented measurements made using a 38 metre high tower with anemometers of the Gill-type located at 10 metres and 37 metres. The lava field that surrounded the site gave surprisingly low values for z_o despite the fact that the elements themselves were up to 2 metres in diameter.

The maximum velocity measured was 29 m/s and this was interesting because many of the measured parameters settled down to relatively constant values after a velocity of 20 m/s was passed. I_u settled to 18% and I_v settled to 10%. The authors reported on integral length scales for u, v and w that settled to 250 metres, 125 metres and 37.5 metres respectively although the authors note that the variability in the measurements may be caused by non-stationarity of the data.

Miyashita et al (1993) reported on measurements from a communications tower in suburban Tokyo. Their measurements came from four heights (45, 80,115 and 150 metres) at each of which were located three cup anemometers and a sonic anemom­eter. The authors selected a typhoon (typhoon 9011 August 8 1990) and a depression (October 11 1991) to make comparisons of the measurement of parameters in the two different regimes.

The maximum mean wind-speed at the 80 metre level in the typhoon was measured as 23 m/s whilst for the depression it was 15 m/s. The comparison showed that the value for remained unchanged, the mean value of I_u was similar whilst measurements of *L_u was variable in both cases.

Power spectra for the u direction were similar in both cases and agreed well with the Von Karman formulation. The values measured for *L_u actually agree well with measurements made during the Askervein experiment.

There are many other measurements that have been made on winds and some of these have been notable (particularly Aylesbury, Askervein and at Texas Tech) however, all experiments are at the mercy of the random process associated with the generation of winds near the surface. The artificial selection of a cutoff point at 20 m/s in this consideration has excluded some of these measurements however they serve to show standards against which to compare measurements at higher wind speeds.

There are other types of strong winds that are not considered here. For instance tornados, with maximum wind-speeds of 150 m/s and thunderstorms with maxima of 75 m/s present significant loading to structures. However, the study of such local phenomena is relatively young and details of structure are less well developed.

Conclusion:

Much effort has gone into the measurement of wind data from full scale experiments so as to underpin theoretical models. Unfortunately it is clear that the atmosphere presents many difficulties to the investigator. In particular very low frequency activity often introduces bias and non stationarity into data records. More modern techniques to avoid such problems are available and are being used in isolated cases.

Whilst the major components of parametric models are reasonably well defined for many different situations there still exist major problems to establishing consistent integral length scales. This problem exists not only because of problems with stationarity but also because the averaging process takes in different gust sizes on different occasions.