An analysis of lidar measurements of buoyant plume rise and dispersion at five power stations

Atmospheric Environment Vol. 26A, No. 18, pp. 3249 3263, 1992. Printed in Great Britain.

~

0004 6981/92 $5.00+0.00 I992 Pergamon Press Ltd

AN ANALYSIS OF LIDAR MEASUREMENTS OF BUOYANT PLUME RISE A N D DISPERSION AT FIVE POWER STATIONS M. BENNETT,* S. SUTTONt a n d D. R. C. GARDINER National Power - Technology and Environmental Centre, Kelvin Avenue,Leatherhead, Surrey KT22 7SE, U.K. (First received 28 January 1992 and in final form 3 July 1992)

Abstract--Measurements of wind speed, plume rise and dispersion have recently been made with a rapidscanning, high-resolution Lidar (RASCAL) at four power stations: Fawley, Didcot, Drax and Littlebrook. This has allowed a more accurate determination of the Briggs plume rise coefficientthan was previously possible and has also provided accurate coefficientsfor the rates of lateral and vertical spread arising from the buoyant plume rise. There appears to be a modest, but statistically significant,difference in the plume rise from power stations with and without cooling towers: for direct-cooled coastal stations we find Cl = 1.35, while for indirect-cooled inland stations Ct = 1.64. The former value is consistent with previous measurements at Northfleet. The Drax measurementsconsisted of 76 plume trajectories each sampled every 8 s over about 30 min; this is a very substantial archive of plume behaviour and has permitted partial testing of the classical 2/3 power law equation for plume rise. Linear regression of the logarithm of plume rise against travel time verified the inverse relation between plume rise and wind speed to within a few per cent. Key word index: Lidar, plume rise, Briggs, cooling towers, remote sensing.

become widely used despite a considerable range in the values of C1 obtained from full-scale experiments Simple theories of buoyant plume rise have been well and some discussion as to whether an x 2:3 or x 3/4 established for many years. As classically reviewed by dependence should be more appropriate; Moore Briggs (1975a), the standard theory assumes that a ris- (1974, 1975) prefers the latter, proposing a model in ing buoyant plume entrains ambient air at a rate which the plume breaks up into a series of puffs. proportional to its velocity and cross-sectional area Observations of plumes from gas flares (Leabey and relative to the surrounding air. Assuming that the Davies, 1984), however, would suggest that the deinitial volume flux is small, it may then be shown that pendence should be rather less than x 2/a. Briggs the radius, r, of the plume is proportional to the height (1975a) cites a range of values of C1 from 1.2 to 2.6 risen, z. Solving the vertical equation of motion of though he considers that the more reliable studies a bent-over plume in a neutral atmosphere then gives indicate a true value of around 1.6. Leahey and Davies a plume trajectory of the form: (1984) also recommend Cl = 1.64, but this was ob1 tained by extrapolating from an observed value of {F~V 2 fl = 0.5 for the visible plume radius; because of radiant heat losses, they were not able to estimate Fb and z= -u 1 + ~ - = Cl Br(Fb, x, u), (1) hence C~ directly. It should be noted that real plumes are very different from the uniform cones implicit in where the derivation of Equation (1): aside from breaking up 1 into puffs, in practice the plume from a large point (2) source almost always divides into two counter-rotating vortices as it is bent over. In general, there are several effects which may give with scatter and bias to estimates of plume rise. As inr = flz, (3) dicated by Equation (1), the rise is inversely proporand x is the distance downwind, u the wind speed at tional to wind speed. Given that this increases with plume height, Fb the buoyancy flux and T/2 the ratio height in the boundary layer, it is important that the of momentum to buoyancy flux. Equation (1) has speed be measured at a similar height to the rising plume. Typically, if 40-m wind measurements were used for a plume at 300 m, this could result in the *Present address: Simon Environmental Technology estimated value of Cl being too small by 25%. A reCentre, Department of Chemical Engineering,UMIST, P.O. lated confounding factor is atmospheric stability. This Box 88, Manchester M60 1QD, U.K. ?Present address: 54 Grangefields Road, Jacobs Well, has only a slight effect on the buoyant rise for travel times much less than the inverse of the Brunt-Vaisala Guildford, Surrey GU4 7NP, U.K. !.

INTRODUCTION

3249

3250

M. BENNETTet al.

frequency but may become important at greater distances. Typically, a potential temperature increase of I°C per 100 m gives a plume rise reduction of about 1% at 50 s travel time (Briggs, 1975a, b; Bennett et al., 1992). In practice, the associated wind shear may have a more significant effect on the initial plume dynamics. There is also a range of source-related effects, for example stack or building downwash, topography, initial volume flux or simple uncertainty as to the buoyancy flux, which may bias full-scale measurements of C1. Of more significance for most field measurements of plume rise is the practical difficulty of making large numbers of plume height determinations. Where, typically, the plume rise after 60 s travel time from a large thermal source may be 100 m, individual thermals and downdraughts may easily carry the instantaneous plume centreline up or down by 50 m. A large number of instantaneous measurements over a time scale greater than that of the largest eddies in the boundary layer is therefore required if the plume rise relative to the surrounding air is to be characterized for particular meteorological conditions. Many of these problems were overcome in recent field measurements (Bennett et al., 1992, henceforth referred to as I) using a rapid-scanning Lidar (RASCAL) at Fawley power station on Southampton Water. This Lidar, developed at the former Central Electricity Research Laboratories at Leatherhead, detects aerosol in the plume by measuring light backscattered from it and is capable of obtaining essentially instantaneous cross-sections of plume density, normally to a height and range resolution of 5 m. The laser beam is directed by a plane mirror which may be rotated vertically and horizontally at about 90 ° s - t , thus permitting vertical cross-sections to be obtained at varying distances downwind from a source. Software is also now available to estimate standard statistical measures of each instantaneous plume crosssection, namely centres-of-gravity of height and range, standard deviations of height and range, total backscatter, total cross-sectional area, etc. Further technical details are given in I. At Fawley, it was found that Lidar scans could be repeated every 5 s. Scanning at three azimuths then gave a repetition period of as little as 15 s at a given azimuth. It was subsequently realized that the wind speed could be estimated by comparing the time series of plume heights at different azimuths. Cross-correlations of each pair of time series were calculated for varying delays; when the delay was equal to the advection time, a peak in the correlation was seen. (A typical plot of correlation against delay is shown in I.) From the advection time, a mean wind speed at plume height could be calculated and this speed was then used to obtain a value of Ct unbiassed by wind shear. Stability effects were assessed separately using temperature measurements from model aircraft. The gentle topography of the area was allowed for by assuming that the streamlines flowed parallel to the

surface. Finally, by extrapolating the observed plume heights to zero distance downwind, it was demonstrated that the virtual and geometric source heights agreed to 10 m, i.e. corrections to Equation (1) from building downwash or from the initial bending-over stage of the plume rise were negligible. The sheer quantity of data obtainable by the rapidscanning Lidar is impressive. The results at Fawley were based on measurements of 49 plume trajectories. Each trajectory was sampled at up to three azimuths, between 250 and 1500 m downwind from the stack, with typically 100 scans at each azimuth and the total data set comprised approaching 13,000 plume crosssections. With numbers of this size, it becomes possible to average out the effects of boundary-layer turbulence and indeed it was found that, when occasions of neutral stability were considered, predicted and. observed 30-min mean plume rise agreed with a standard difference of less than 20%. The present paper describes the extension of these techniques to two power stations with cooling towers--Didcot in Oxfordshire and Drax on Humberside (Fig. 1). There was particular interest in characterizing the Drax plume, since this power station is in the process of being fitted with flue gas desulphurization plant and there was thus a need to obtain a baseline against which future changes in the plume 15ehaviour could be assessed. A few measurements at Littlebrook, which is a direct-cooled, oil-fired power station on the Thames Estuary, were also available. This paper also compares these results with the original Lidar measurements of plume rise at Northfleet power station (Hamilton, 1967).

2. EXPERIMENTAL

The general experimental protocol for using the Lidar has already been described in I. Basically,repeated plume crosssections are obtained alternately at several distances downwind from a source. Simultaneous temperature profiles are obtained at heights of up to 500 m using model aircraft (in practice, this may have to be at several kilometres from the power station) while the wind speed at plume height is obtained, in effect, by following plume features downwind. The determination of wind speed has been somewhat improved since the original work at Fawley by scanning with automatic triggering at two azimuths instead of with manual triggeringat three. This reduced the cycletime between scans from 15-20 to around 8 s; the precise value depends on the number of shots per scan and the ranges of azimuth and elevation employed. The improved precision in the lag correlation is shown in Sutton and Bennett (1992). A further elaboration of the technique is to use crosscorrelations of both the height and range of the plume at different azimuths to estimate the wind speed. At Drax, it was possible to measure the wind speeds using both variables for 65 plume trajectories. The ratios of the estimates of the wind speed from height and range variations turned out to be 0.96 + 0.09 (SD) with a correlation coefficientbetween the two of 0.953. The bias between the two measurements appeared to be statistically significant and some effort was put into discoveringits cause (Sutton and Bennett, 1992).We now believe that the discrepancy arises from the skewness of

Analysis of Lidar measurements

f~k )

3251

--

Fig. 1. Locations of power stations irtvolved in this survey.

the turbulent momentum flux in the boundary layer. Thermals tend to carry slow-moving air upwards from close to the ground; they would be detected as rising puffs in the plume when height variations were followed downwind. This analysis tends therefore to underestimate slightly the wind speed at plume height. It was noted that when the atmosphere stabilized, the agreement between the wind speeds obtained at Drax by the two methods tended to improve. Put formally, the measurement of wind speed has relied upon a lag time calculated as the centre of gravity of a crosscorrelation (cf. the Appendix). We find [-

(tA(t) Z~ (t) Z'2(t +tA(t)) ) (z~ (t) zl (t + tA(t)))

(4)

Angle brackets or overbars here denote ensemble means while primes denote variations from these means. Neglecting the decay of perturbations over the Lagrangian time scale, z~ (t) = x l w'(t)/u and z'~ (t + tA) = x2 w'(t)/u, where the true advection time is tA (t) = (X 2 -- X l )/u(t). Considering then variations to first order in 1/u we find f=(x2-xl)

[1
(5)

L

In isotropic turbulence, no bias would arise. Where updrafts are more concentrated than downdrafts, however, the measured value of wind speed will differ from the true mean value. In a very simple model where updrafts and downdrafts have different but constant values of u and w, we find u . . . . ~ / [ 1 + 4ea./t~],

(6)

where a, is the overall standard deviation of u and 2e is the difference in frequency between downdrafts and updrafts. In a convective boundary layer, e will usually be positive. Lenschow and Stephens (1982), for example, give e=0.22 in the

mixed layer over the East China Sea. It may thus be appreciated that biasses to the measured wind speed of order 10% may easily arise through this mechanism. Equivalent biasses from cross-correlations on range are likely to be smaller since (u'v '2) has no particular tendency to be positive or negative. Where wind speeds from both height and range variations are available, the procedure at Didcot, Drax and Littlebrook has been to take their mean to give a best estimate of the mean wind speed at plume height. The reduction in scatter from taking the mean of the two values was believed to outweigh the small bias associated with also using the wind speed from height variations. Overall, the wind speed has probably been underestimated by about 2%. Table 1 summarizes the Lidar data which were available for the present paper. The original data from the Lidar surveys at Northfleet (Hamilton, 1967) were provided by Dr Hamilton and comprised power station loads, plume parameters and meteorological data. This Lidar had a pulse repetition rate of 1 Hz and was swivelled manually in elevation and azimuth. In good conditions, a skilled operator could obtain 20 cross-sections in an hour; these were displayed like a radar trace on an oscilloscope screen, being recorded with a Polaroid camera. The heights of the base and top of the plume were then measured from the photograph. For the present analysis, it was assumed that the plume centreline height was the mean of these two heights. Hourly mean values of wind speed were available from a 187-m tower 7 km from the Lidar. This anemometer malfunctioned for about 25% of the measurement period and was substituted with other measurements from 114 m, scaled assuming a 1/7 power law for the wind speed. Temperature profile data from the tower were unavailable for much of the period of measurements because of an instrument malfunction (Hamilton, 1967).

M. BENNETI"et al.

3252

where g is the acceleration due to gravity, V~the volume emission rate, mt (ra.) the molecular weight of the flue gas (ambient air) and T~ (T,) the temperature of the flue gas (ambient air). Mean ambient temperatures of 10°C were assumed at Fawley and Didcot and 20°C at Drax and Littlebrook. The buoyancy flux at Northfleet was calculated using Hamilton's estimate that the thermal emission from each stack was 1/12 of the electrical output from the plant. The fifth column in Table 3 then lists the ratio of buoyancy flux in m 4 s - 3 to power station load in MW: the variation in this ratio from one plant to another shows the importance of an explicit calculation when trying to make accurate comparisons between the plume rise at different plants. Finally, Table 3 lists the emission velocity, ui and the ratio of momentum to buoyancy flux for the five plants:

The Fawley measurements have already been described in I. The number of scans quoted in Table 1 denotes the number of mean plume height determinations; for RASCAL measurements these are obtained from multiple instantaneous scans, as listed in the fourth column of the table. It is clear that vast quantities of data are available, with around 30,000 plume-height determinations having been made at Drax. At Fawley and Drax, temperature profiles were measured with a model aircraft (Weather and Atmospheric Sampling Platform--WASP) but, because of technical problems, this was only available for 1 day at Didcot. The few Littlebrook measurements were made as part of a plume visibility study and the model aircraft was not employed. Tables 2 and 3 give technical details of the five power stations. The quoted ratios of buoyancy flux to electrical load are correct to better than 10% for sets on full load. Emission parameters were mostly arrived at in discussion with power station staff. In normal operation, the flue gas temperature is a compromise between economy and the need to prevent condensation and acid smut formation in the flue. It should therefore be maintained independently of the ambient temperature or power station load. Temperatures were measured at one of the flues at Drax during the survey; otherwise the quoted values are a nominal value for each plant. Buoyancy fluxes were calculated using the formula

1 Tami~

2 Fro=z=

Tim,/ V~,

(7)

Table 1. Lidar surveys reviewed in present paper Power station Northfleet Fawley Didcot Drax Littlebrook

Period of meagurement

No. scans

Scan mult.

Temp. method

5/7/65-20/5/66 31/5/89-6/7/89 4/12/89-5/12/89 13/3/91-27/3/91 24/6/91-11/7/91 22/7/91-23/7/91

418 130

1 100

tower WASP

36 152 22

150 200 200

-WASP --

Table 2. Technical details of power stations

Power station Northfleet Fawley Didcot Drax Littlebrook

.

For these buoyancy-dominated plumes, variations of this time scale are unimportant. With the exception of Northfleet, the power stations in this study are multiflued so, except on very low load, the emission velocity is maintained close to its nominal value. Excess oxygen (i.e. the 02 content of the exhaust gases) is routinely measured at the back end of the boiler, but leakage around air-heater seals leads to its increasing by 1-2% before final emission of the flue gases. Since each 1% of excess oxygen is approximately proportional to a 5 % change

-

Fb=g \

2ui

Fuel

Capacity (MW)

Thermal efficiency (%)

coal oil coal coal oil

720 1932 1900 3750 2055

31.95 35.17 35.06 37.95 34.77

Cooling method estuary sea cooling towers cooling towers estuary

Table 3. Emission details of power stations

Fb/QM

Power station

Stacks (m)

Tl (°C)

Excess 02 (%)

(m 4 s- 3 MW-1)

ui (m s -1)

_F,,,/& (s)

Northfleet Fawley Didcot Drax Littlebrook

2 x 152 198.1 199.3 259.0 210.3

107 143 130 121 140

6 3 ~ ~ ~

1.90 4.50 4.31 3.47 4.45

20 23 23 23 23

15.9 I0.I 11.8 14.6 11.6

Analysis of Lidar measurements in buoyancy flux, we were concerned to obtain realistic values for Table 3. It should be noted that sets tend to operate on higher excess oxygen at low load. At Didcot, for example, the final excess oxygen for a set on half load may approach 9-10%, implying that the buoyancy flux is twice that from the products of combustion alone. In this case, half the electrical load gives approximately 3/4 of the buoyancy flux on full load. In normal circumstances, electrical demand is mostly satisfied by switching units on or off completely rather than by having several units operating on part load. Nevertheless, inaccuracies may still arise when attempting to relate changes in plume height to changes in electrical load.

3. DIRECT-COOLED POWER STATIONS

R A S C A L measurements were made at two directcooled power stations, Fawley and Littlebrook, and could be compared with the earlier measurements at Northfleet. Summary statistics for plume rise from all power stations are given in Table 4 and for plume spread in Table 5. The second column of Table 4 gives the number of plume height determinations for which the wind speed was also available. At Northfleet, these correspond to individual 3-min samples; at the other stations, they correspond to multiple instantaneous samples over 30 min as listed in Table 1. Wind speed, travel distance and plume heights were then used to calculate the plume rise parameter, C1. The best estirhate of this for each station is listed together with its 95% confidence limits. Given a best estimate of C1, individual measurements of height of course differ

Table 4. Plume rise parameters estimated at five power stations Power station Northfleet

n 418

Fawley

55

Didcot

28

Drax Littlebrook

110 6

C1 + 95%

RSD (%)

r

1.12-1.47 +0.07 1.37 + 0.07 1.60 + 0.08 1.65 +0.05 1.21 +0.16

46.8

0.600

19.4

0.936

12.6

0.944

14.3

0.919

9.8

0.987

3253

from those predicted by Equation (1). The residual standard deviations (RSD) listed in the table are based on the differences of the measured from the calculated heights expressed as a percentage of the calculated rise. Also listed are the correlation coefficients between calculated and observed plume heights. It may be noted that the RSD decreased from around 50% at Northfleet to around 20% at Fawley, probably largely as a result of R A S C A L allowing boundarylayer turbulence to be averaged out. At Didcot and subsequently there was a decrease to 10-15%, probably as a result of the improvement in the estimation of the wind speed through using two azimuths rather than three. A substantial problem with the interpretation of the Northfleet measurements is the existence of a 30-m ridge immediately to the south of the power station. Most of the plume-height determinations were made with south winds and it was believed that the base of the stack was usually in the separated wake of the ridge. Thus, as the streamlines gradually returned to the surface a reduction in the effective height of the stack by an amount comparable to the height of the ridge was to be expected, implying an effective stack height of about 122 m. Subsequent wind tunnel work by Robins (pers. commun.) indicated that a 30-m correction was reasonable. Table 4, therefore, quotes a range of values of C1 corresponding to between zero and 30-m correction, the latter probably being closer to reality. The 95% confidence limits refer to the statistical uncertainty given a reliable value for the streamline deflection. Figure 2 shows the observed values of plume rise, allowing for 30-m deflection, against Br(Fb, x, u). As may be seen, there is too much scatter for the streamline deflection to be estimated directly. The Fawley measurements have already been described in I. For the present analysis, we have only used data where the model aircraft showed that the lapse rate between 200 and 400 m was at least 0.5°C per 100 m, i.e. that the stratification was approximately neutral. This makes little difference to the mean value of C1 but reduces the scatter on the trajectories shown in Fig. 3: this reduction in scatter is probably only marginally due to any real physical effects of stability on plume rise and is more likely to reflect the

Table 5. Plume spread parameters measured with RASCAL Power station

qz

~ly

az/try

n

+95%

+95%

+95%

Fawley

55

Didcot

28

0.43 __+0.03 0.47 __+0.04 0.418 +0.012 0.34 +0.07

0.55 +0.03 0.52 -t-0.05 0.492 +0.013 0.45 +0.19

0.80 +0.05 0.95 +0.07 0.859 +0.025 0.80 +0.18

Drax Littlebrook

109 6

r 0.401 0.777 0.423 0.746

3254

M. BENNEa-ret al. PLume rise a t I

450

NorLhfLeeL,

I

550

I

0

O

O

o

"7-

°O

O

O

0 O0 000 0O

;

I

0

O0 250

1966

I

OO

°o. 05/'::

o"° ° °°

^o

~"~oo,,. v o^ o

o

150

5O

OO

-50

' 0

O

I 100

' Briggs

I 200

' 300

variable

Fig. 2. Plume rise (corrected for 30-m streamline displacement) at Nonhfl~t against 'Briggs variable', Br(Fb,X,U)=(Fb/~r)1/3 x2/3(1 +u~/x)l/3/u, in m. The solid line is the theoretical prediction for C~ = 1.47.

I

500

I

I

I

'

[

I

400 ¸

500,

~_ 2o0

)00.

'

0

tOO

1

200 Briggs

300

voriabLe

Fig. 3. Plume trajectories at Fawley in approximately neutral conditions against Br(Fb,X,U) in m.

model aircraft being unable to fly on occasions with low cloud when the Lidar measurements of plume height might be in error. Only a small amount of data were available for Littlebrook. Measurements were made over 2 days but those on the first day were unfortunately not suitable for our present purposes since the wind was very light. Four trajectories (eight plume heights) were

measured on the second day at distances between 250 and 900 m from the stack. By the end of the period, however, the power station was going off load with an estimated efliux velocity of only 8.6 m s -1 against a wind speed of 8.4 m s - 1. In these circumstances, we should expect the plume to be entrained in the wake of the stack (Hoult and Weil, 1972; Snyder and Lawson, 1991) and indeed an anomalously small plume

Analysis of Lidar measurements rise was observed. A similar occasion had been noted at Fawley. This, then, left six reliable measurements of plume height for use in Table 4. The scatter on the technique is now such that even a small number of measurements should provide useful results. Caution is necessary, however, since the plume was being advected over the Thames Estuary (600 m wide) on this day and there may have been some subsidence. Temperature profile measurements were not available on 23/7/91 but the weather was such that stable stratification was very unlikely. Comparing the values of C1 from these three power stations, it may be seen that they are statistically consistent. Taking the mean of the values for Fawley and Littlebrook, we obtain C~ = 1.35__+0.10, which is well within the range of values that would be expected for Northfleet. The analysis software for RASCAL also provided measurements of the dispersion of the plume. As may be appreciated from Equations (1)-(3), the lateral and vertical standard deviations, try and tr~ of the instantaneous plume should both be proportional to the plume rise and hence to the 'Briggs' variable. Thus, = t#=

Br(Fb, x, u),

(9)

where r/, and r/y are normalized plume spread parameters and are constant insofar as plume dispersion is only caused by the shear of the buoyant plume as it rises relative to the surrounding air. For a uniform circular plume tr z=trr=r/2. Substituting Equations (3), (1) and (2), we thus find 1

(10)

0

3255

and taking the value of C1 = 1.37 for Fawley, the dynamical theory predicts r/= 0.523 and fl = 0.764. Best estimates and 95% confidence limits of r/~ and r/y from the four RASCAL surveys are given in Table 5. Figure 4 shows the observed trajectories of ~/= against ~/y at Fawley. It is generally assumed that the dynamical plume radius, which enters the plume rise formula, is greater than the visible plume radius because of the inertia of the air displaced by the plume. We thus see that r/, at Fawley is indeed smaller than the value predicted on the basis of Equation (10), though this may be partly due to the non-uniform distribution of material within the plume. The lateral spread, r/r, is greater than the dynamical value, presumably as a consequence of the almost invariable splitting of the plume into two counter-rotating vortices as it is bent over by the wind. Analyses were also made of the visible plume area as a function of plume rise. The measured area obviously depends on the zero level assumed for backscatter from the atmosphere; as implemented in the Lidar software, this level is taken as the 5-percentile of backscatter intensity from all elements in a measurement window; this procedure appears to work well so long as a large enough window has been chosen and there is not too much atmospheric haze. Converting, then, the measured plume area into an effective radius and taking its ratio to the plume rise gave fl= 0.64 + 0.05 for Fawley. As expected, this is again less than the value predicted by the dynamical theory. Similarly, the plume radius normalized by Br(Fb, x, u) was found to be 0.85+0.05. This may be noted to be somewhat less than the radius obtained from the measured standard deviations which, assuming a uniform elliptical plume, gave a value of 2r/yr/. = 0.98. The

8

N 06

! ~o4 Z

02

¸

'

0

I

O2

'

I

04

'

I

'

I

O6 Normotiz~l s i ~

08

'

I

'

I 0

y

Fig. 4. Trajectories of vertical against lateral plume standard deviation, both normalized by Br(F~,, x, u), in approximately neutral conditions at Fawley. AE(A) 26:18-B

3256

M. BENNETTet at.

implication here is that mass is concentrated at the outside of the plume, again as a consequence of the plume's double vortex structure. Table 5 also lists the correlation coefficient, r, between individual measurements of rt~ and ~ly. This is instructive in that the simple theory of plume rise and spread in conjunction with instrumental and turbulent noise should give a correlation coefficient of zero. Errors in the estimated wind speed, buoyancy flux or the functional dependence of spread on distance, however, should equally affect ~z and fly and will thus give a positive value to the correlation coefficient. To an extent, Fig. 4 helps to distinguish these possibilities in that errors in wind speed or buoyancy flux should displace similarly all the points on a trajectory; with the exception of one trajectory, this is not seen.

4. I N D I R E C T - C O O L E D

POWER STATIONS

Lidar measurements were made at two inland power stations with cooling towers, Didcot and Drax. A plan of Didcot showing the heights and layout of the various structures is given in Fig. 5. There was no significant topography for the wind directions in which measurements were taken. Measurements were made during an initial 2-week survey in January 1991. These provided very little useful data, since both stack and cooling tower plumes were very slow to evaporate in the cold winter weather and the Lidar could only detect the edges of the plumes. Similar problems were encountered during a 3-week survey in March 1991 but the higher temperatures at least allowed useful measurements to be made on 5 days, on one of which the stack was downwind of the southern cooling towers. Downwind distances sampled were between 500 and 1700 m from the stack. Model aircraft measurements of temper-

ature profiles were only available on one of these days but midday radiosonde ascents at Crawley showed that, with the exception of one trajectory, all the measurements were made in neutral conditions. On the one exception, it may have been slightly stable. The plume-rise data for Didcot are reported in Table 4 and plotted in Fig. 6. Those trajectories where the cooling towers were upwind of the stack have been marked with dashed lines in the figure. Statistical analysis showed that the effect of the cooling towers on the plume rise could not have been very much greater than 10%. Table 5 and Fig. 7 similarly show the normalized plume spread. The figure gives the impression that the plume spread is increased downwind o f the towers; this was found to be on the borderline of significance but was sufficient to give a larger vahre to the correlation coefficient quoted in Table 5. The mean slope of the trajectories in Figs 4 and 7 implies that there may be some dependence of normalized plume spread on distance downwind. At Didcot, it transpires that of the 14 trajectories, only 5 have the smaller value of ~?y for the smaller value of distance downwind. This is not statistically significant. We also find that ~/yfor the 14 points nearer to the stack is 0.53 + 0.07 while that for the 14 farther points is 0.50+0.09. Similar results hold for ~Tz. In aggregate, there are thus no grounds for believing that the dependence of plume spread on distance downwind is significantly different from x 2/3. On individual occasions, however, the rate of plume spread may either be enhanced by intense ambient turbulence or apparently reduced if concentrations on the margins of the plume fall below the detection threshold. The measurements at Drax were programmed for 3 weeks at the height of summer; it was hoped that temperatures would then be high enough to ensure rapid plume evaporation and a high cloud base. The

©©

~m I

I

Cooli~ towersQ 114m /

Q

©©

Stack 199m o

65 m

Fig. 5. Layout of Didcot power station.

......) N

Analysis of Lidar measurements

3257

rise vs Briggs voriolXe ot Didcot, 1991 5OO

I

I

I

I

'

I

400

2OO

0

I

IO0

200

3O0

Br~Jgs voriobte Fig. 6. Plume trajectories at Didcot against Br(Fb, x, u) in m. Occasions when the southern cooling towers were upwind of the stack have been indicated with dashed lines.

Verticol vs lot~rol plume sprl~idot Didcot. i

i

I

i

i

i

i

i

i

'

I

'

i

i

08-N .~ 0.6' S I 0.4"

j,

/J

0.2"

0

'

I

OZ

'

I

O4

'

I

06 OO Normotiz~l sigmo y

I

~0

t 2

Fig. 7. Trajectories of vertical against lateral plume standard deviation, both normalized by Br(Fb, x, u), at Didcot. Occasions when the southern cooling towers were upwind of the stack have been indicated with dashed lines.

model aircraft was flown from Burn airfield, 6 km west of the power station. Conditions were generally favourable and, out of 13 days nominally available for the survey, usable Lidar measurements were made on 10 days. On two of those days, measurements were continued until late in the evening. Mostly, however, measurements were made between 0900 and 1700 G M T . O n safety grounds, the model aircraft was not

allowed to fly after 1600. Hourly mean temperature profiles up to 400 m were also available from the Belmont TV mast (Bennett, 1989, 1992) 70 km southeast of Drax. Given the mast's proximity (27 km) to the coast, however, these were felt to be unreliable in east winds. The layout of the power station is shown in Fig. 8; there is no significant topography in its vicinity. On

3258

M. BENNETT

full load, for a reference ambient temperature of 13.4°C, the sensible heat emission from each cooling tower is 114 MW with an efflux velocity of approximately 5 m s- 1 Most of the cooling towers were operating for most of the period of the survey. It had been estimated that the commissioning of the flue gas desulphurization plant should reduce the plume rise by 12%. Given a 15% RSD on the individual plume height determinations, 25 determinations would be required both before and after commissioning in order to demonstrate such a change conclusively. Ideally the samples should be obtained for wind directions such that the stack plume would not be directly perturbed by the cooling towers: as may be seen from Fig. 8, this implies east or west winds. It was also hoped to be able to compare, as at Didcot, plume rise in the wake of cooling towers to that in the unperturbed flow. In order to characterize the plume rise as thoroughly as possible, trajectories were measured in pairs spanning approximately 250-650 and 600-1000 m on alternate half hours. Overall, 76 plume trajectories with 152 determinations of the 30-min mean plume height were obtained. This corresponded to approximately 30,000 instantaneous plume cross-sections. The peak day was 10/7/91 when, between 0853 and 2141 GMT, 13 trajectories were obtained from 5000 instantaneous cross-sections. Given that time must be spent downloading and backing up data, this was probably close to the practical limit. Plume rise parameters at Drax have been listed in Table 4 and the corresponding plume trajectories plotted in Fig. 9. These refer only to occasions on which we were sure that the lapse rate was at least 0.5°C per 100 m and therefore exclude the measurements made in the evening or on one day (8/7/91)

et al.

when the temperature profile was apparently slightly stable. Plume spread parameters have similarly been listed in Table 5 with vertical against lateral spread being plotted in Fig. 10. It may be seen that the plume spreads are much more tightly clustered than those for Didcot; the single rogue trajectory was obtained on the morning of 11/7/91 and was probably due to potential instability at the top of the plume. These high quality data therefore suggest that the dispersion of the plume from an elevated, highly buoyant source will normally be self-similar out to downwind distances of at least 1000 m. The plume rise parameter at Drax has been measured to within 95% confidence limits of _+3%. It is thus clear that the 12% change when the F G D plant is commissioned should be easily distinguishable.

5. DISCUSSION O F D I F F E R E N C E S

Comparing the plume rise parameters for Didcot and Drax with those for Fawley and Littlebrook, it may be seen that the rise for the indirect-cooled stations is more than 20% greater than that for the direct-cooled stations. Pooling the data for Didcot and Drax, we find that C1=1.64 _+0.06 (95% confidence limits) as against C1 = 1.35_+0.10 for the other two stations. Statistically the difference appears to be highly significant. The rates of plume spread at the different power stations, however, appear to be similar: there is some indication that the rate of lateral spread at Fawley may be 10% greater than at Didcot and Drax, but this does not rise to statistical significance. From measurements of the plume area, a normalized effective plume radius of 0.856_+0.024 is found at Drax. This is statistically indistinguishable

200m I,__j

®®® ®®®

) N

Stack, 25@m O

Boiler house, 77 m

Cooling towers, 117 m

Fig. 8. Layout of Drax power station.

® ®® ®® ®

Analysis of Lidar measurements PLume rise

vs

Briggs

i

500

Drax,

at

variable

3259 1991

i

i

400

3Oo

/

j

.

:~ 200

IOO

I 0

'

I

200

IO 0

lgriggs

3OO

varialote

Fig. 9. Plume trajectories at Drax against Br(Fb, x, u) in m. Occasions when cooling towers were upwind of the stack have been indicated with dashed lines.

Vertical vs I

I

I

t

Lateral t

plume spread I

I

at I

Drax I

I

I

O8

N

06 sl

~N 0 4

02"

'

0

t

02

'

I

'

04 NormaLized

I

'

I

06 sigma

08

'

I I 0

' 2

y

Fig. 10. Trajectories of vertical against lateral plume standard deviation, both normalized by Br(Fb, x, u), at Drax. Occasions when the southern cooling towers were upwind of the stack have been indicated with dashed lines.

from the value at Fawley. Overall we find r/z=0.43+0.04 and t/y=0.51 +0.05, while the aspect ratio is trz/ar=0.85+O.07. While this difference between the values of C1 for the power stations with and without cooling towers may appear modest, it should be recalled that peak ground-level concentrations are extremely sensitive to the plume height, with typically an inverse square or

inverse cube dependence. The apparent differences in plume rise may thus make a significant difference to the environmental impacts of the power stations. It was therefore important to attempt to understand whence the 20% difference in plume rise may have arisen. Several possible causes of the discrepancy were considered. Firstly, it could have been an artefact due

3260

M. BENNETTet al. occupies a substantial volume as it leaves the top of the tower; simple modelling suggests that this should reduce the effective emission height by about 30 m. The buoyancy fluxes from the various sources may not thus simply be added to predict the plume rise from the complete plant. As already discussed, however, were the cooling towers merely to change the buoyancy flux in the stack plume, then plume spread as well as plume rise would be enhanced. A more coherent model is that for certain wind directions the stack plume finds itself rising through the rising plume from one set of cooling towers; the buoyant rise is thus to some degree additive while the spread is largely unaffected. The interaction of cooling tower and stack plumes has been modelled in a wind tunnel by Lohmeyer and Nester (1987). They found that, for a passive emission near a cooling tower with a sensible heat emission of 300 MW, the stack plume height could be reduced by of order 10 m when the stack plume was 200-300 m crosswind of the cooling tower plume: the point being that the atmosphere surrounding a buoyant plume must descend to replace the rising effluent. Conversely, elements of the stack plume closer than this shared in the full rise of the cooling tower plume. This model is apparently supported by observations such as in Fig. 11, where the bifurcated stack plume passes over the plumes from the northern set of cooling towers. After correction for the right-hand

to the variables entering Equation (1) having been consistently in error for one or more of the power stations. Given the cube-root dependence on buoyancy flux in Equation (1), it" would seem highly implausible that the 10% uncertainty in the estimated Fb should have had a significant effect. A more likely source of error might be in the estimation of the wind speed. The mechanics of the wind speed measurement were changed after the Fawley survey and it is conceivable that some artefact was thereby introduced, though every effort was made to check for this. Were such an artefact present, however, it would affect not just the estimates of plume rise but also the estimates of normalized plume spread. We should therefore expect r/= and t/y to be 20% greater for the indirectcooled stations. In fact, r/Z is identical while ~/y is, if anything, 10% smaller. The discrepancy is probably therefore a real effect. The most obvious physical explanation is that the rise of the stack plume can be enhanced by the rise of the plumes from the cooling towers. At Didcot or Drax the buoyancy flux from each set of cooling towers is comparable with or larger than the buoyancy flux of the stack emissions (and is larger again if there is substantial condensation in the plumes). There is thus scope for a significant addition to the buoyant rise from the stack emission alone. The buoyancy flux from the cooling towers is not as effective in causing plume rise as that from the stack since it already

260691 5

sume05 15 !

25

35 f

i

45

55

65

i

600

4OO

2OO

I

I000

2'

I OO

'

1400

Ronge (rn)

Fig. ll. Plume cross-sections measured at Drax between 1517 and 1539 GMT on 26/6/91 at 539 m from the stack. The lower plume is from the northern set of cooling towers; note that the plume of warm, humid air from the towers may b¢ much more extensive than the plume of visible condensation.

Analysis of Lidar measurements vortex being further downwind, it appears that it has been displaced upwards by about 34 m relative to the left-hand one, possibly as a result of the cooling tower plumes. A complication on this occasion, however, was that the wind direction veered with height. Such veering results in the net shear on the right-hand vortex being less than that on the left-hand one: this should lead to a reduced spread and enhanced rise. In order to study the effects of the cooling towers more methodically, the plume rise data were divided according to the position of the height determination (i.e. horizontal distance and direction) relative to the cooling towers. This is shown in Table 6. The first row denotes plume directions such that the line of the stack plume has not passed over a cooling tower. The second row denotes directions such that a set of cooling towers is upwind of the stack. (These trajectories have been marked as dashed lines in Fig. 9.) The third row denotes height determinations which have been made with a set of cooling towers downwind of the stack. Given the layout of Drax (Fig. 8), there were no wind directions for which cooling towers were both upwind and downwind of the stack. It may immediately be seen that the stack plume could only have been directly affected by cooling tower plumes on about a quarter of the measurements. To enhance the mean rise by 20%, therefore, would require an enhancement of the rise on those occasions by more than a factor of 2. Comparing rows 1 and 2 of the table, however, shows that such an enhancement, if it exists, can only amount to a few per cent. The apparent enhancement in the third row is not statistically significant. If a cooling-tower effect exists, therefore, it must affect all wind directions equally, implying the existence of some large-scale convective cell enveloping the entire power station: it is difficult to see how such an effect could be demonstrated. A second obvious difference between the direct and indirect-cooled power stations is that the former are on the coast and therefore in principle more susceptible to wind shear. Changes in wind direction with height were of course measured automatically as part of the survey since they appeared as changes in the plume direction between the stack and the various downwind cross-sections. At Fawley, it was found that the median wind shear was less than 4 ° per 100 m and this seems inadequate to explain an overall 20%

3261

reduction in plume rise. There was also no discernible dependence of C1 on wind shear on individual occasions. It is clear that one should expect differences in plume rise for onshore and offshore winds at a coastal site. Subsidence will occur for offshore winds or for winds blowing with the land on the left-hand side (Brettle, 1989). Some direction dependence of plume rise was indeed noted at Fawley in I; but averaged around all wind directions the mean effect should have been quite small. The net effect may, however, have been significant at Littlebrook, where only plumes travelling over the Estuary were sampled. Such direct coastal effects are experimentally analogous to direct cooling-tower effects in that they should show themselves as a direction dependence of the plume rise and would have to be very large to cause the difference in mean rise between Fawley on the one hand and Didcot or Drax on the other. It is probable that at inland power stations, the stack plume tends to engulf ambient thermals which assist its rise. At a coastal site, such thermals would be less well developed. The presence of cooling towers will, of course, also tend to attract and trigger ambient instability. At Didcot or Drax such convective cells may be enhanced by convection from the coal stock, which could contribute a sensible heat flux of up to 100 MW depending on meteorological conditions. It is difficult to see how such a complicated theoretical picture could be directly tested by experiment. To a degree, the effects of coast and cooling towers may be distinguished by repeating the measurements at an inland, direct-cooled power station: unfortunately none such exist in the U.K.

6. VALIDATION O F P L U M E RISE M O D E L

A useful application of these data is to test the plume rise formula, Equation (1). As it stands, it is not straightforward to perform linear regression on this equation to verify the dependence on u and x since the momentum and buoyancy terms are supposed to have different functional forms. If we rewrite the equation, however, as

Z=Cl (Fb/IZ)a(l q-U---~)#x3#+r U- 3#

(11)

= Ct (Fb/~)~ (t2(t + z)) # x ~, Table 6. Plume rise parameters in relation to cooling towers at Drax Cooling towers

n

Crosswind

80

Upwind

23

Downwind

7

C1 +95%

RSD (%)

r

1.64 +0.05 1.66 _+0.09 1.81 _ 0.20

14.5

0.894

12.7

0.915

17.0

0.657

where t = x/u is the advection time, it may be seen that the momentum/buoyancy ratio has been included in a single term. In addition, the various terms correspond to largely independent measurements (recalling that an advection time rather than the wind speed is the primary measurement). By comparison with Equation (1), we see that ct=fl= - ~ = 1/3. Setting up, then, a linear regression of the form

log(z)-~1 log(t2(t + z)/x)=~log(Fb)+ A,

(12)

3262

M. BENNETTet al.

where A is a constant, we obtain, using the Drax data in neutral conditions, ~t= 0.15 + 0.18 (95% confidence limits). Rearranging Equation (12) to give equivalent regressions for the other variables gives fl=0.326 +0.019 and ) , = - 0 . 4 0 + 0 . 0 6 . The inverse relation of plume rise and wind speed has thus been demonstrated fairly accurately with 3fl = 0.98 ___0.06. The rate of plume rise with distance, however, is rather less than the x 2/3 of Equation (1), with the best estimate of the index being found to be 0.58 +0.08. It is possible that this is due to increasing wind speed at greater heights reducing the rate of rise at the further distances. For this data set, the inclusion of the momentum term does not improve the agreement between theory and measurement. It should, nevertheless, be included so that the results can be directly extrapolated to high-momentum sources (e.g. gas turbine plant). For the reasons discussed in Section 2, the value of ct should be treated with caution. The buoyancy flux is not strictly proportional to the electrical load on the plant and over the period of the survey the range of loads sampled was relatively small (mostly between 2000 and 2700 MW). In practice, the data are equally consistent with a Q1/3 or Q~/4 power law (Briggs, 1975b).

ation is %=0,51; these ensemble values have both been estimated with 95% confidence limits of about 10%. Similarly, the normalized mean radius of the instantaneous plume is 0.85. Using these values, it is straightforward to estimate instantaneous plume concentrations out to where the plume breaks up. Regression analysis of the measurements at Drax has also improved confidence in the accuracy of the simple plume rise model, Equation (1), for broadly neutral conditions. The inverse dependence on wind speed has been verified to within a few per cent while the dependence on distance downwind has been shown to be slightly less than predicted, probably as a result of wind shear. The effect of the initial momentum of the plume could not be studied; further measurements at gas turbine plants would be useful. Acknowledgements--We are grateful to the managers and staff of all the power stations for their help in making these measurements; to Plant Engineering Branch at Harrogate for providing temperature and excess oxygen data from Drax; to Dr P. M. Hamilton for making available the original data from Northfleet; to Mr A. R. Fuller and Mr M. Chown for operating the model aircraft; and to Dr A. G. Robins for useful discussions. M. B. is grateful to Mr W. Thorp for the opportunity to experience the Drax plume at first hand in his glider. This work was carried out under the joint environmental programme between National Power and PowerGen.

7. CONCLUSIONS This analysis has demonstrated the power of modern Lidar techniques in characterizing plume rise and dispersion and has enabled plume rise coefficients to be estimated to accuracies of a few per cent. This accuracy has allowed anomalies between different sources to become apparent. In particular, there appears to be a modest but real difference in normalized plume rise between coastal direct-cooled and inland indirect-cooled power stations. We find C1 = 1.35 for the former and C1 = 1.64 for the latter: this could lead to a significant difference in ground-level pollutant concentrations. The reason for the anomaly remains obscure but is most probably due to a large-scale convective cell associated with the cooling towers and possibly the coal stock. The presence of the coast may substantially affect the development of such a cell. Surveys on further power stations, with differing fuels and cooling arrangements, would obviously be valuable in order to see if the anomaly persists. The survey at Drax estimated C1 to an accuracy of 3%. It should thus be possible to detect the estimated 12% change in plume rise when the F G D plant is commissioned. Further measurements at that time would be of value. Some of the most useful data arising from these surveys have been the measurements of plume spread. Thus if instantaneous plume standard deviations are normalized by Br(Fb, x, u), we find that on average the instantaneous vertical plume standard deviation is r/: = 0.43 and the instantaneous lateral standard devi-

REFERENCES

Bennett M. (1989) Meteorological logging on a 389 m TV mast. Weather 44, 116-120. Bennett M. (1992) Wind statistics from the Belmont TV mast, 1983-89. Weather 47, 114-123. Bennett M., Sutton S. and Gardiner D. R. C. (1992) Measurements of wind speed and plume rise with a rapid-scanning Lidar. Atmospheric Environment 26A, 1675-1688. Brettle M. J. (1989) Sea temperatures and coastal winds. Weather 414, 249-256. Briggs G. A. (1975a) Plume rise predictions. In Lectures on Air Pollution and Environmental Impact Analyses, pp. 59-111. American Meteorological Society, Boston, MA. Briggs G. A. (1975b) Discussion: a comparison of the trajectories of rising buoyant plumes with theoretical/empirical models. Atmospheric Environment 9, 455-457. Hamilton P. M. (1967) Plume height measurements at Northfleet and Tilbury power stations. Atmospheric Environment 1, 379-387. Hoult D. P. and Weil J. C. (1972) Turbulent plume in a laminar cross flow. Atmospheric Environment 6, 513-531. Leabey D. M. and Davies M. J. E. (1984) Observations of plume rise from sour gas flares. Atmospheric Environment 18, 917-922. Lenschow D. H. and Stepbens P. L. (1982) Mean vertical velocity and turbulence intensity inside and outside thermals. Atmospheric Environment 16, 761-764. Lohmeyer A. and Nester K. (1987) Einfluss einer Kuehlturmfahne auf die Ausbreitung einer Schornsteinfahne. Met. Rund. 40, 130-141. Moore D. J. (1974) A comparison of the trajectories of rising buoyant plumes with theoretical/empirical models. Atmospheric Environment g, 441-457. Moore D. J. (1975) Discussion:a comparison of the trajectories of rising buoyant plumes with theoretical/empirical models. Atmospheric Environment 9, 457-459.

Analysis of Lidar measurements Snyder W. H. and Lawson R. E. (1991) Fluid modelling simulation of stack-tip downwash for neutrally buoyant plumes. Atmospheric Environment 25A, 2837-2850. Sutton S. and Bennett M. (1992) Measurement of wind speed using a rapid-scanning Lidar. In Proc. 6th Int. Symp.

Acoustic Remote Sensin 0 and Associated Techniques of the Atmosphere and oceans (edited by Asimakopoulos D. N.), Athens, 26-29 May, pp. 143-148.

3263

where the lag correlation for height variations at distances x 1 and x2 from the source is given by r12(t) -



Expanding Equation (AI) using Equation (A2), the variance terms cancel. Exchanging the integration and the ensemble average, the integral may then be found using the stochastic relation

Z'l(t')z'2(t' +t)=z'l(t')z'2(t' +ts(t'))tl~(ts(t')--t), APPENDIX ENSEMBLE FORMULA FOR MEAN ADVECTION TIME

The mean advection time is calculated from the centre of gravity of the lag correlation, i.e.

t=S r12(t) tdt/ ~ rlz(t) dt,

(A1)

(A2)

,/

(A3)

where t^(t') is the advection time for a parcel of effluent leaving x ~ at t' to arrive at x2, and t, is the turbulent integral time scale; this cancels between the numerator and denominator in Equation (A1). We thus obtain f --





(A4)