A series of field campaigns has been made at British airports using a rapid-scanning lidar and other instrumentation in order to measure the dispersion of exhaust plumes from commercial aircraft. The lidar operated at a wavelength of 355 nm and was thus effectively eye safe. Analysis software for the lidar signals has been elaborated to enable the rather weak signals (typically a few tens of percent of ambient backscatter) from aircraft exhaust to be distinguished and to facilitate automatic processing of the measurements obtained. Such processing can deliver images, animations, and numerical parameterizations of the dispersing plumes.
Overall, 1353 air traffic movements were monitored over two campaigns at Manchester and 439 in a single campaign at Heathrow. All modes were observed: taxiing, takeoff, rotation, climb-out, approach, and landing. Of these, the most complete dataset was that obtained for the start of the takeoff run: in this mode, the source is on full power but is still moving relatively slowly. Emissions thus remain at their most concentrated. For the same reason, this is the most important mode in respect to local air quality. Tire smoke on landing was likewise easily detected. Conversely, the lidar could only see the engine emissions from about 30% of the aircraft on approach. These data have been archived in an accessible form and are currently being used to develop improved regulatory dispersion models for airports.
Local air quality is a significant issue at many large airports. We note, for example, that NO2 concentrations are already above the permitted annual limit value of 40 μg m−3 around Heathrow and Frankfurt Airports, largely as a result of motor vehicle emissions (Department for Transport 2006; Frankfurt International Airport 2007); and Heathrow will be seeking planning consent for a third runway, while Frankfurt has started construction of a fourth. Robust modeling of the aircraft exhaust emissions will therefore be essential. The dynamics of such emissions, however, make dispersion modeling rather difficult. The plume is emitted as a horizontal jet, which is subject to aerodynamic forces from the airframe of the aircraft and drag from the ground and, being hot, tends to rise buoyantly. On takeoff, it is also emitted by an accelerating source.
To permit the validation and further development of exhaust plume models, we have therefore undertaken a series of field trials at Heathrow and Manchester airports using a rapid-scanning lidar in conjunction with various other observations. This paper describes the field work undertaken and how we have elaborated the hardware and software of the lidar system so that it should be capable of monitoring aviation emissions. We illustrate this capability with images of dispersing aircraft plumes under a range of operational modes. Subsequent papers (Bennett and Christie 2010; A. Graham et al. 2010, unpublished manuscript) will report detailed quantitative analysis of these measurements and some resultant theoretical modeling. Partial and preliminary results have already been reported as an appendix to the Project for the Sustainable Development of Heathrow (Graham et al. 2005) and in various conference papers (Christie et al. 2006; Bennett et al. 2008; Graham et al. 2008).
The complete dataset has now been included in a wider dataset of airport air quality measurements (Schäfer et al. 2009) so as to be more generally accessible within the research community.
2. Rapid-scanning lidar system
The lidar used for these surveys was originally used for measuring the dispersion of power station emissions (Bennett et al. 1992). As first constructed, it employed a frequency-doubled neodymium-doped yttrium aluminum garnet (Nd:YAG) laser (HyperYag 750, Innolas United Kingdom) emitting visible green light at λ = 532 nm. This gave a nominal ocular hazard distance of 3.1 km on a single shot and as such was wholly unsuited for operation at a commercial airport. Following the lead of researchers at the National Oceanic and Atmospheric Administration (NOAA) and the U.S. Department of Transportation (Wayson et al. 2002; Eberhard et al. 2005), however, we converted the system to frequency-tripled Nd:YAG, emitting in the UV-A at λ = 355 nm. Technical specifications of the system are given in Table 1. Within the relevant British Standard [British Standard European Norm (BSEN) 60825-1:1994 with International Electrotechnical Commission (IEC) 60825-1], it may now be considered eye safe; detailed calculations are given by Bennett et al. (2006). The energy density of a single pulse may be calculated to be 38 J m−2 at the system output, while the permissible energy density for an 8-ns pulse is 53 J m−2. Because the beam is invisible, there is likewise no distraction hazard for pilots.
An observer sitting on top of the system and staring directly into the outgoing beam would have accumulated an impermissible dose after 8.2 s. We do not consider this to be a plausible scenario. It may be feasible, however, that an observer may stare at the vehicle through binoculars from a few hundred meters range. The magnification of the binoculars would then compensate for the beam divergence, leading to an impermissible dose being received after 10–20 s. (At this wavelength, normal optical glass has only a modest absorption.) Our only operational constraint, therefore, is that the system should not be deployed in staring mode below the horizon.
Two digitizers are installed in the system. In this application, the slower one was employed, alternating between the two detection channels, to give 30 megasamples per second per channel. This delivered an effective range resolution of a relatively coarse 5 m; given the tenuous nature of aircraft plumes, range bins of this width proved to be necessary to give an adequate signal above the background noise.
The system is mounted in a 6 metric ton commercial vehicle (Mercedes 709D) and has onboard power generation. It can thus easily be sited at any reasonable location and operated autonomously. The beam is steered in elevation or azimuth using a 10″ × 20″ plane mirror (MgF2 coated), permitting horizontal or vertical cross sections of atmospheric aerosol to be obtained. In practice, a typical scan takes 2 s and scans may be repeated every 4–5 s for up to several hours at a session. The steering mirror is mounted at 3 m above ground level and is thus high enough to see over, for example, an airport boundary fence.
The lidar vehicle incorporates an automatic weather station comprising the instruments listed in Table 2. Wind speed and direction, temperature, and humidity are measured at mast height (nominally 10 m), while shortwave radiation is measured on the roof of the vehicle. All variables are logged at 10-s intervals. At some locations, considerations of air traffic safety precluded the mast from being fully extended.
3. Lidar analysis
Previous experience with the system had mostly involved monitoring dispersion from industrial plants (cf. Carruthers et al. 1996). The monitoring of aircraft exhaust emissions presented several new challenges:
Aircraft emissions are transient: the ground run of a commercial aircraft accelerating from rest typically only lasts for 30–40 s. In addition, while all aircraft start from and land at much the same location, permitting good results from a fixed scanning plane, the aircraft show much more variation in the point of rotation or lift-off, because heavily laden aircraft are much slower to leave the ground than are light ones. Data capture in these modes is thus much poorer.
Modern aircraft engines burn fuel extremely efficiently. Even at full thrust, therefore, the smoke emission from a modern commercial airliner is (usually) barely visible to the eye. This has severe implications for its ease of detection with a backscatter lidar.
It was therefore necessary to develop more sensitive algorithms for distinguishing the backscatter from exhaust emissions from that of the ambient air. This was achieved both at the level of individual laser shots and at that of scanned cross sections.
a. Analysis by shot
The signal V(r) received by a lidar from a single shot is determined by the lidar equation (Svanberg 1994). This is given in simplified form as
The signal returned from a radial segment of length Δr arises from the backscatter β per unit pathlength and unit solid angle, but falls off with the square of the distance r from the source. It also decays exponentially through the extinction σ per unit pathlength. Within this equation, W is the power per pulse and A is a system constant (which depends on the optical efficiency, etc.). This signal is superimposed on a background arising from the dark current in the photomultiplier tubes and (by day) from scattered sunlight within the pass band of the input filters.
For spherical particles, β and σ both depend upon the ratio of the diameter of the particles to the wavelength of the scattered radiation and on their complex refractive index at that wavelength. Neither of these is known in general. At any given range, the lidar equation thus relates a single measured variable to two unknowns. Conventionally, the equation is solved using the “Klett inversion” (Klett 1981), which assumes that β = σκ, with κ being a numerical constant. This approach may have difficulties, however, when applied to a heterogeneous distribution of aerosol, for example, the plume from a combustion source superimposed on ambient aerosol. We have therefore preferred to treat the additional extinction arising from the plume as being small (as it certainly is in this case) and the additional backscatter as purely a relative measurement. Our approach is thus to subtract the background light and then the ambient backscatter from the received signal; the remainder must be the combustion plume in which we are interested.
For the purposes of detecting tenuous aircraft plumes as part of this project we devised an algorithm for fitting a straight line to the ambient backscatter arising from the clear air.
First, the zero level is calculated from 40 bins accumulated before the laser is fired. For statistical robustness, the 5 largest and smallest values are rejected and the mean of the remaining 30 are taken to be the background zero.
Second, we note that in a uniform atmosphere, we should have L0(r) = ln[r 2V(r)] = C − 2σr, where C is a constant encompassing the pulse energy, the optical efficiency of the system, the ambient backscatter, and any calibrations involved. If we can fit such a straight line to the background signal within the measured curve of L(r) = ln[r 2V(r)], then L(r) − L0(r) must represent the combustion plume. This is most robustly done by eye (cf., e.g., Eberhard et al. 1987), but this would clearly be impractical given the tens of thousands of shots involved. We have therefore devised a numerical algorithm in which we use conventional linear regression iteratively to estimate σ and C.
We start by calculating the standard deviation of the measured values of L(r) for a given shot around their median value over many range bins. We then calculate a best-fit linear regression for L(r) versus r, but using the original median and standard deviation to apply a Gaussian weighting to each point in the fit. In turn, we then calculate the residual standard deviation away from this best-fit line for use in a subsequent Gaussian weighting for a second linear regression. This continues for up to 10 iterations. Points furthest removed from the background signal (i.e., combustion plumes, etc.) are thus rapidly weighted out of the fitting, leaving us with a good fit to L0(r).
A further issue is that aircraft exhaust is emitted close to the ground and is drawn still closer to the surface by the Coanda effect (Tritton 1988, p. 150ff). Measurements at the lowest angles of elevation may thus also include returns from hard targets. The detectors saturate at this point and may subsequently ring. Our algorithm treats a large positive excursion followed quickly by a large negative one as a hard target and rejects these data from the fit. These points are illustrated in Fig. 1, which shows a backscatter profile L(r) from a single shot through the exhaust plume from a BAe146 aircraft. The fitted ambient background is shown by the sloping dashed line, while the exhaust plume is the small peak at 190-m range. The shot direction is slightly below horizontal and thus impinges on surface-based obstacles (i.e., trees) at 830 and 940 m. Interestingly, the obstacles were not so solid as to totally obscure the signal from more distant ambient scattering. Returns from the first hard return and outward were automatically excluded from the fitting routine.
Finally, we exponentiate the profile again as V ′(r) = exp[L(r) − L0(r)] − 1. This gives a profile scaled relative to the backscatter from the ambient aerosol. This profile is linear in the backscatter from isolated plumes, it has been corrected for the 1/r 2 decay of signal with distance, it has been corrected for ambient extinction, and it is independent of the shot energy. Typical sensitivities are up to 3% of the ambient backscatter at a range of 200 m. (We estimate this from the observed noise in the clear-air signal at this distance.)
The slope of the fitted line L0(r) is twice the ambient extinction σ per unit pathlength. Because ambient extinction-to-backscatter ratios are well known, at least to within a factor of 2 or so (Ackermann 1998), an additional product of this fitting procedure is thus an absolute, if imprecise, calibration of the lidar system.
b. Analysis by scan
The above procedure for processing shots individually relies on the system efficiency being independent of range, and hence L0(r) varying linearly with r. This is not the case for our system in the near field. Optically, the system is rather efficient, with an essentially complete capture of the return optical path being possible for ranges greater than ∼50 m. This is not necessarily desirable, however, because the return signal from this range could easily be hundreds of times greater than that at the ranges where we might wish to observe an exhaust plume. This saturates the detectors, which then take a few microseconds to return to full gain. As a consequence, L0 curves noticeably with range (cf. Fig. 1). Some compromise is thus required at modest ranges (i.e., ∼200–300 m), whereby we reduce the aperture of the system in order to increase the minimum range of the full optical efficiency to (say) 175 m. The sum of the nonlinearities arising from the increasing optical efficiency and the recovering detector gain is then minimized. (If working exclusively in the very near field, on the other hand, the detector aperture would be fully opened and the laser power reduced).
Any residual curvature is largely compensated for by adjusting the estimated ambient base level as a function of range. This is achieved by taking all of the shots in a scan, above the ground and below cloud base, and taking the median of V ′(r) at a given range as being the base level of ambient scattering at that range. In the near field, some miscalibration of the backscatter distribution within the plume may arise, because we continue to use the original shot-by-shot values for the scaling relative to the ambient base as a function of range. The point may be appreciated by looking at Fig. 1. The automatic procedure scales the near-field plume relative to the local ambient backscatter indicated by the dashed line: inspection shows that this is slightly larger than the real backscatter at this distance. The envelope of the plume above the background, however, is defined by where it deviates from the observed base level on either side: this should be approximately correct.
It should be borne in mind that the classical Gaussian plume arises as a time average over many times the time scale of the turbulence. Our backscatter lidar measurements, on the other hand, are essentially instantaneous: on this very short time scale, dispersing plumes tend to have much sharper edges. This assists us in our identification of the plume envelope.
It will be appreciated that this procedure will only be successful if the curvature of the return does not change significantly between shots. This may not be the case for the early shots of a scan if the laser power is changing rapidly. Artifacts may then arise in the apparent shape of the plume if a slightly too low base level is chosen.
Artifacts may also arise if the ambient backscatter changes rapidly with height. By default, in case low cloud might be present, we exclude returns from above 400 m from the calculation of the zero level. From visual analysis of the data from a particular field trial, it is immediately obvious if 400 m is too high and a lower value may then be used. We can also censor, if necessary, returns from close to the surface, thereby removing ground clutter; multiple scattering may also be an issue for shots very close to the ground. It would be very difficult, however, to extract a plume from surface-based mist.
We may note that in earlier applications of the lidar system (Bennett et al. 1992), we had only used this scan-by-scan background-fitting technique, with the only shot-by-shot normalization being that for the (rather noisy) energy monitor. Though adequate for power stations, this technique was quite inadequate for aircraft plumes.
c. Display and parameterization
Having calculated a polar array of V ′(r) within a scan, it is then straightforward to display the results either as a density contour plot (cf. the figures in section 5), or, indeed, as an animation of a sequence of such cross sections. It is also possible to calculate first and second moments of the density distribution, thereby obtaining plume statistics in respect of total backscatter, total area, centroid location, centroid height, and vertical and horizontal spread. The software also generates third moments of the cross sections, but in practice these have only been used to flag up anomalous scans, for example, if the laser has misfired.
In practice, some degree of human intervention (and hence subjectivity) may be required when calculating such dispersion parameters. The calculation takes place within a frame defined by the operator, thereby excluding (say) low cloud, ground clutter, or far-field noise. There is also the issue that higher moments are extremely sensitive to noise outside the real envelope of the plume. For this reason, the operator must define a noise threshold below which signals will be ignored. This is painlessly achieved through the operator adjusting the contour levels used in the backscatter density plots so that, where possible, a clear plume is visible above background noise. The software then takes the significance threshold as being the lowest contour level.
We should note that such a threshold is not used in the calculation of total backscatter, because it would give us a positively biased result.
4. Field campaigns
Following its conversion to UV operation, our first field trials with the lidar were made at Heathrow airport over the period of 10–20 May 2005.
Heathrow has two runways, with lengths of 3902 m (north) and 3658 m (south), oriented 90°–270°T (true). We had the use of the following two sites for our lidar:
Northern runway: The lidar was set up in a deicing bay at the end of terminal 1. This was 330 m south of the centerline of the northern runway, 800 m southwest of the threshold (cf. Fig. 2a). In westerly operations (i.e., when aircraft are taking off and landing into a westerly headwind), this was a good site for observations of the start of the takeoff run, or of the approach to landing. In easterly operations, there was little observable activity.
Southern runway: The lidar was set up beside a cargo warehouse at the southwest corner of the airfield. This was 220 m south of the centerline of the southern runway, 550 m southeast of the displaced threshold (cf. Fig. 2b). (The displaced threshold is 300 m from the end of the metaled runway.) In easterly operations, this was a good site to observe both takeoffs and landings. In westerly operations, only the takeoffs of the most heavily laden aircraft were observable.
For most of the period, the weather was dominated by cool east or north winds. This changed to a milder southwest flow on 18 May. The lidar was thus installed at the southern site for the period of 13–17 May, to take advantage of easterly operations, and at the northern site for the rest of the campaign. On 17 May 2005 observations of takeoffs continued until dark, that is, at 2100 UTC; otherwise, measurements stopped at 1600 UTC, or earlier if the switch of runways at 1500 UTC led to air traffic movements (ATMs) being unobservable.
The lidar anemometer had good exposure at both sites for the wind directions experienced. If necessary, the meteorological data acquired could be supplemented by those from the synoptic meteorological station near the northern runway (Fig. 2a).
The overall lidar data capture from the northern site was somewhat disappointing. There were only westerly operations on 10 May 2005 and 18–20 May 2005, but half of the latter period was lost through rain. Overall, 67 ATMs were monitored. Data capture from the southern site was much more satisfactory. The lidar started operations there from midafternoon on 13 July 2005 (Friday). Takeoffs in easterly operations were monitored on Friday, Saturday, and Tuesday; takeoffs in westerly operations were monitored on Sunday and Monday. Overall, 372 ATMs were monitored.
The lidar data were supplemented by operational ATM and ground-radar data. These were provided by the British Airports Authority and National Air Traffic Services, respectively. The ATM data include scheduled and actual times of arrival (at threshold) and departure (at takeoff), flight number, airframe type, aircraft registration number, and runway number. Times were logged to the nearest minute. Using Web resources, full details of any individual aircraft can be found from its registration number. The radar data give times of all ATMs together with airframe type. Speeds and block positions along the runways are also given. Timings are logged to the nearest second.
With care, movements in the lidar, ATM, and radar datasets can be aligned to provide dispersion data from identified aircraft. The lidar beam would occasionally strike an aircraft; this would then provide a precise relative timestamp for the radar and lidar clocks.
b. Manchester I
The first field trial at Manchester Airport took place over the period of 12–23 September 2005.
Manchester has two runways, with lengths of 3048 m (06L/24R) and 3047 m (24L/06R), oriented 51°–231°T, with runway 06L/24R being closer to the terminal buildings. Both runways can be used in mixed mode (i.e., alternating landings and takeoffs as necessary). There is a bias to westerly operations (which are used 80% overall), so as to reduce noise over the nearby suburbs of Manchester. Outside peak hours, the airport does not operate at full capacity; between 1200 and 1500 UTC, only 06L/24R is used.
The footprint of the airport is much smaller than that of Heathrow and we were able to operate from a landside site (i.e., outside the perimeter fence), 480 m from the displaced threshold along the extended centerline of 06L/24R and 191 m to the southeast side of it (cf. Fig. 3). (The displaced threshold is 250 m from the end of the metaled runway.) In westerly operations, this is an excellent site for observing plumes from the start of the takeoff run or from aircraft on approach. In easterly operations, heavy aircraft may be observed on climb-out.
Several supplementary measurements were added to the basic lidar data obtained at Heathrow.
First, ground-radar data were not available from this airport. We instead made video recordings of ATMs from a fixed point on the roof of one of the terminal buildings (Olympic House). These recordings are precise to the nearest second and, with the simultaneous use of the audio channel, it is possible to time engine power up accurately (due allowance being made for the speed of sound). After the trials, theodolite readings were made of markers along the runway, thereby permitting the video recordings to be interpreted in terms of aircrafts’ position, velocity, and acceleration.
As at Heathrow, we were provided with operational ATM data, logged to the nearest minute. Given aircraft registration numbers, it is thus possible again to identify particular aircraft precisely. Where possible, the lidar operators also took notes of the types, registration numbers, and timings of the aircraft whose emissions were being measured.
Second, a commercial Fourier transform ultraviolet (FTUV) long-path spectroscopic monitoring system (Siemens, UV-Falcon) was set up across the northeast extension of the runway centerline (Fig. 3) at 475 m from threshold. This was set to monitor SO2, NO, NO2, O3, and H2S (Padgett and Harvey 1995; Cruz-Jimate 2007), with a pathlength of 50 m and a time resolution of 20–30 s (depending on signal strength). Previous comparison with routine monitoring in central Manchester had indicated that the system gave robust results for SO2, O3, and NO; manufacturer’s data implied detection limits of 5, 40, and 10 ppb, respectively, for these gases. Measurements of NO2 with the UV-Falcon, however, were quite unreliable (Cruz-Jimate and Bennett 2006). In an attempt to monitor this gas (and hence NO2/NOx ratios) in the aircraft plumes, prototype passive UV-Differential Optical Absorption Spectroscopy (DOAS) systems (Galle et al. 2002; Oppenheimer et al. 2005) were also deployed both in parallel with the FTUV system and close to the lidar.
As at Heathrow, meteorological measurements were made with the lidar’s onboard weather station. At this site, however, for air-safety reasons, it was not permissible to raise the mast to more than 4 m, and thus there may be issues with flow distortion around the body of the vehicle. The exposure at this site was otherwise good. As a cross-check, the lidar operators also regularly logged the meteorological transmissions of the Automatic Terminal Information Service; these data are available to pilots in real time and include temperature, dewpoint, cloud cover, and wind speed and direction.
Lidar measurements were made on 10 days. The majority of these (6 days) were with westerly operations in light to moderate southwest to northwest winds. There were 5 days of easterly operations (9–11 and 15–16 September) associated with east to northeast winds. Persistent heavy rain on 15 September precluded any lidar measurements. Overall, 702 ATMs were monitored (with 302 arrivals and 400 departures).
c. Manchester II
The second field trial at Manchester Airport took place between 27 March and 7 April 2009. In this case the lidar was shifted 60 m northeast to lie within a gated compound. This was 540 m from the displaced threshold along the extended centerline of 06L/24R and 196 m to the southeast side thereof.
The meteorological instruments of the previous trials were supplemented by a Scintec phased-array sodar (model SFAS64, i.e., comprising 8 × 8 piezoelectric transducers), located as shown in Fig. 4. This provided 15-min averages of vertical profiles of wind and turbulence up to a height of 100 m, with a 5-m vertical resolution. Given the challenging acoustic environment of an airport, data capture was poor at greater heights.
Overall, meteorological conditions were much less favorable than in the previous trial. The flow was mostly southwest with intermittent rain; on 27–28 March and 2 April, heavy rain precluded any measurements. Winds were generally moderate and at times strong; on 1 April wind speeds were around 10 m s−1 for most of the day, gusting occasionally to 15 m s−1. An unexpected casualty of this poor weather was the UV-Falcon; the strong winds and soft ground prevented its being reliably aligned. (These problems did not affect the passive DOAS system.) Temperatures were low, typically 7°–8°C at midday. On a single day (4 April), the wind was light northeast and the airport was on easterly operations. Overall, lidar measurements were made on 9 days, with 651 ATMs being monitored (with 395 arrivals and 256 departures).
For one day (3 April), the lidar was moved to a different location, 4.77 km east and 3.09 km north of its standard location. This corresponds to 6.2 km from the displaced threshold along the approach path and 800 m to the southeast side of it. With the standard 3° approach path, landing aircraft should have been at a height above ground of ∼314 m at this point, and thus at an angle of elevation of 21.4° relative to the lidar. From this location we monitored the passage of 35 aircraft on approach in an attempt to measure their plume vortices well away from ground effect.
Given that a commercial airliner has a typical length of 70 m and approach speed of 75 m s−1, and that the lidar scans were repeated at 4-s intervals, we should have expected to strike [70 × 35/(4 × 75)] = 8.2 aircraft in the course of these measurements (or rather more, given the head wind). In fact, nine aircraft strikes were identified, at horizontal distances of 784.2 ± 12.5 m and heights above the lidar of 343.9 ± 8.9 m. This therefore corresponds well to our topographic calculation; the precision with which aircraft follow the prescribed approach path is also impressive.
Overall, we can divide the plume measurements made during these campaigns into four observational categories distributed around the landing and takeoff (LTO) cycle (Penner et al. 1999, p. 247ff). These categories comprise the wall jet arising from the start of the takeoff run, wing vortices in the latter part of the takeoff run and in climb-out, wing vortices on approach, and tire smoke from landing. The exhaust plumes from a few aircraft were also visible while idling or taxiing.
a. Takeoff run
Aircraft exhaust plumes are seen most clearly by the lidar at the start of the takeoff run; at this stage, the engines are on full power and the aircraft is moving slowly. For the same reasons, it is this part of the LTO cycle that is most critical for local air quality.
The dynamics of the starting plume are very different from those of conventional industrial sources. Relevant factors are as follows:
The presence of the ground: The plume thus forms a wall jet; the Coanda effect brings the exhaust rapidly down to the ground and initially keeps it there. This effect also enhances the mutual entrainment of the exhaust from the individual engines into a single plume.
Its transient nature: The starting plume forms a head, into which later plume elements are advected.
The acceleration of the source: An aircraft typically accelerates at about 0.25g. After a few seconds, therefore, the speed of the source is much larger than that of the ambient wind. The latter thus serves to advect the starting plume, but does not dominate its formation.
The buoyancy of the source: The thrust in the jet is, of course, almost horizontal, while its buoyancy gives it a gradually increasing vertical momentum. For a modern high-bypass engine burning kerosene, the two forces are equal after a travel time of about 2.3 c/g, where c is the speed of sound (Graham et al. 2008). This works out as about 80 s. Note that this is about twice the duration of the aircraft’s ground run: the plume is thus unlikely to leave the ground before the aircraft does.
These aspects have been clearly seen in our lidar measurements. Examples may be seen in Figs. 5 and 6, which show longitudinal and lateral sections through starting plumes. Because of its importance for local air quality, most of our theoretical analysis has gone into understanding the physics of this developing wall jet (A. Graham and M. Bennett 2010, unpublished manuscript). We note that only a single plume is visible in these two figures.
Fuselage-mounted engines are often mounted with the axis of thrust slightly inclined relative to the axis of the fuselage; this helps balance the trim in cruise. This would not be helpful for the wing-mounted engines, which generated the plume shown in Fig. 6. At low airspeeds, the flow over the wings will also have only a small effect on the direction of thrust. We are thus confident that the ground-hugging nature of the plume arises through the Coanda effect. Given that the Coanda effect also forces the jets from individual engines to merge rapidly into a single plume, it was only very occasionally that a lidar scan could be obtained sufficiently close to the aircraft to distinguish individual engine plumes; out of the thousands of scans taken, on only a single occasion could we distinguish all four plumes from a Boeing 747 (B747).
We see in Fig. 5 that the exhaust jet is very shallow for most of its travel but rises to a terminating head. The image shown is one of a sequence at 4-s intervals. Within this sequence it may be seen that, because of the motion of the source, the shallow jet rapidly approaches the lidar, while the head slowly propagates away from it. All of the section shown in Fig. 5 is to the south of the runway. The advection of the emissions into the lidar scanning plane has been assisted by the ambient wind, but mostly arises from the forced lateral spread of the wall jet. (Typically, such a jet is nearly as wide as it is long.) It is interesting that the peak of backscatter in the head is no longer at ground level; this may be an indication that buoyant rise is already starting to take effect.
Figure 7, conversely, shows a starting plume after a travel time of ∼80 s, by which time it had lost all contact with the ground (i.e., “lofted”). It was in fact somewhat unusual for a plume still to be detectable so far from the source. Generally, plumes were seen more easily in cool, humid conditions. On this occasion, however, the 10-m temperature was 16°C and the relative humidity was only 66%. It was fortunate that the wind (u4 = 5.3 m s−1) happened to carry the starting plume directly over the lidar, so that it could be viewed at short range, even after a travel distance of >550 m. Scanning took place using this geometry for 88 min on the afternoon of 14 September 2005. Over this period, 50 ATMs were monitored, of which 11 gave a detectable plume at a range from the lidar of 200 m. Of these 11 plumes, 7 were already aloft at 200 m, 2 were apparently lifting at this distance, 1 had fully lofted by 400 m, and 1 could not be followed. In addition, the sky was overcast, limiting any ambient convection. We are thus confident that the plume rise illustrated in Fig. 7 is representative of the effects of source buoyancy, rather than of the plume simply being lofted through being caught in a rising thermal.
We may note that it is the buoyancy of the emissions that apparently moderates local air quality in the vicinity of Heathrow. The highest concentrations (e.g., of NO2) are experienced here for wind speeds of ∼8 m s−1. The dilution arising from longitudinal dispersion is then counterbalanced by the suppression of buoyant plume rise. As illustrated by Fig. 7, aircraft emissions may have only a modest local impact in light winds (Department for Transport 2006).
The lift experienced by the airframe in the ground run depends on the square of the airspeed of the aircraft, its angle of attack, and the amount of flap deployed. (On takeoff, there is a trade-off between lift and drag, so flaps are only partially deployed. On approach, however, flaps are fully deployed, thus maximizing both lift and drag.) Lift is thus minimal in the early stages of the run, but it causes increasing surface flow divergence as the aircraft accelerates. When the aircraft approaches takeoff speed, the pilot depresses the tail, thereby increasing the angle of attack, and hence the lift. This is known as “rotation.” At that point, the aircraft takes off. Figure 8 illustrates the effect of these factors on pollutant emissions. In this case, the aircraft had completed nearly 1 km of ground run and attained a ground speed of 75 m s−1 (this being very close to its to airspeed) before passing through the scanning plane. It took off 4 s later. Prior to rotation the lift coefficient would typically be only 0.4 (Mair and Birdsall (1992), p. 139), giving a lift of ∼⅕ of the maximum takeoff weight of the aircraft. (The maximum thrust, by contrast, is ∼¼ of the maximum takeoff weight.) The surface outflow associated with the circulation around the wings has entrained most of the emission into a pair of plumes on either side of the runway centerline; by the time that the scan shown in Fig. 8 was obtained, these had separated by 106 m. In still air, this is equivalent to a lateral propagation velocity of ±9% of the forward velocity of the aircraft.
In still air, the lift on the aircraft is proportional to the length of its run, so that at some distance <1000 m, the emissions distribution must switch from that shown in Fig. 6 to that shown in Fig. 8. The LHR2 monitoring station at Heathrow (Fig. 2) is at a distance corresponding to a takeoff run of 500 m. Regular series of spikes in NOx concentrations have been observed at this point as aircraft take off past it (Carslaw et al. 2008), but these tended to be single rather than double peaks, implying that lift is not yet dominant at that point [though double peaks may have been difficult to observe, given the time response of the instrument (D. Carslaw 2009, personal communication)].
As with the start-off run, the engines are on near maximum duty for lift-off and climb-out. Despite the greater speed, this mode thus is also clearly visible to the lidar. There are nevertheless practical difficulties with observing climb-out because, as noted, the length of the takeoff run depends on the weight of the aircraft. In practice, we were obliged to choose a scanning azimuth and simply accept such aircraft that happened to take off through it.
The dynamics of plume dispersion on climb-out differ from those early in the takeoff run in that the airframe lift is now dominant. The engine emissions are very rapidly entrained into the two wing vortices. These descend to the ground at ∼1 m s−1, gradually separating, and then move apart more rapidly, under the influence of the image vortices reflected in the surface, as they approach the ground. Given that the lidar scans are separated by 4–5 s, one may expect that the measured height of the vortices would depend on precisely when the first scan fell after passage of the aircraft. Due allowance must be made, however, for the headwind into which the aircraft would usually take off. Because the source is rising into a headwind, successive scans sample emissions released at a greater height, which have then descended before being scanned. Overall, the observed height of the vortices thus typically changes little between successive scans.
With a slight crosswind, it also proved possible to take longitudinal sections of the plume from the aircraft in climb-out. To achieve this, we made vertical scans at an acute angle to the runway, with the scanning plane crossing the runway close to a typical point of lift-off, and waited for a trail to advect through the scanning plane. Again, a descent velocity of ∼1 m s−1 could be estimated, once the effect of the headwind had been subtracted.
Particulate matter (PM) concentrations in the plume tend to be low on approach, because the aircraft is moving rapidly and the engine duty is close to the minimum for smoke production. We note also that typical diameters of the particulate are smaller on low power settings: this strongly reduces the backscatter per unit mass. Thus, even where the lidar was relatively close (∼200 m) to the approach path, most exhaust plumes were invisible to it. A single cross section would be visible on about 30% of approaches at this range, with perhaps half of these persisting long enough (up to 20 s) for the descent of the vortex pair to be observed (Fig. 9). In this case, the headwind works with the descent, so the observed height of the vortices decreases rapidly between scans.
Measurements 6 km farther out on the approach path were somewhat disappointing. The lidar system functioned well, with the detection threshold being around 10% of ambient backscatter, even at a range of 800 m. Nevertheless, only two smoke trails were tentatively identified out of the 35 ATMs monitored. Interestingly, neither showed the classic dual vortex structure illustrated in Fig. 9. Somewhat tenuous single plumes were instead visible, with an additional faint wisp of backscatter being drawn across to one side. We speculate that such plumes may arise from a slightly increased consumption of lubrication oil in one engine; an asymmetric smoke plume is sometimes visible to the naked eye during climb-out.
Although the International Civil Aviation Organization (ICAO) standard LTO cycle specifies the engine thrust on approach as 30% of the rated thrust (or 27% of the maximum fuel flow), the actual fuel usage can be much less than this for most of the approach, because the aircraft is decelerating. Thrust tends to be applied in bursts as the aircraft deviates below its intended path.
d. Tire smoke
On landing, most aircraft (particularly the heavier ones) release a visible puff of smoke from each wheel as it makes contact with the runway. (In a crosswind, the upwind wheels usually make contact with the ground first, and more heavily.) These emissions were monitored at Manchester both with the lidar and with the video recordings.
Figure 10 shows a typical cross section through such a puff. The wind direction was slightly veered relative to the runway (u4 = 6.5 m s−1); on occasions when it was veered by a few more degrees, it was possible to follow the tire smoke as far as the minimum range from the lidar; this would then be followed after ∼40 s by the smell of burning rubber (Bennett and Christie 2010). The video record showed that touchdown occurred 250 m upwind of the position of the puff in Fig. 10 (i.e., at ∼30 m to the left of the scanning plane), with a ground speed at runway threshold of 58 m s−1.
While it is believed that tire smoke emissions make a significant contribution to total aerosol emissions at airports (BAA 2006), and mass balance estimates have been made of the rubber lost per landing (Morris 2006), rather little is known about the composition and size spectrum of such aerosol. While we expect tire smoke to be a combination of soot from burning rubber and coarse mechanically generated dust from the runway [more recent measurements support this (Bennett et al. 2010, manuscript submitted to Environ. Sci. Technol.)], it is not so coarse as to preclude significant and persistent lifting as the puff advects downwind. An important consideration here is that the airframe is still developing sufficient lift at the point of touchdown to more than balance the weight of the aircraft. The tire smoke may thus be entrained in the resultant wing vortices; this will both enhance the vertical spread of the smoke, to about a wing span, and encourage the puffs from port and starboard wheels to separate further laterally. Both effects may be seen in Fig. 10. It may also be seen that the peak backscatter is now well separated from the ground. This is probably a consequence of the thermal buoyancy of the emission. There is no indication of coarse aerosol sedimenting out. The maximum sedimentation velocity implied by Fig. 10 is thus less than ∼0.5 m s−1; it follows that aerodynamic particle diameters must be <80 μm (for concrete spheres).
A second point may be noted from Fig. 11. Both engine smoke and tire smoke appear here in the same frame. Despite the tire smoke having already traveled 500 m, and despite the lidar operator having noted visible smoke in the engine plume, both emissions give a similar peak backscatter. Indeed, a 767-300 landing 3 min later gave a puff of tire smoke with a peak density 4 times that shown in Fig. 11 at the same location. Close to the point of emission, tire smoke typically gives a signal an order of magnitude greater than that of engine smoke.
According to the conventional theory of Mie scattering, the maximum backscatter efficiency per unit mass of fine particulate occurs when the particle circumference is comparable to the wavelength of the radiation being scattered. This corresponds to a particle diameter of ∼100 nm in this case. For particles smaller than this, the backscatter per unit mass is proportional to the individual particle volume; for particles larger than this, it is inversely proportional to the particle diameter (neglecting oscillatory terms). Detailed calculations by Eberhard et al. (2005) suggest a ratio of backscatter to gravimetric concentration of ∼700 cm2 g−1 sr−1 for 40-nm soot aerosol. Conversely, a simple geometric calculation for spherical soot particles of 10-μm diameter and a UV albedo of 0.04 gives a ratio of only 6 cm2 g−1 sr−1. (Given their greater albedo, concrete spheres of 50-μm diameter would give a similar ratio.) Thus, the similar backscatters shown in Fig. 11 for the two plumes could conceal a gravimetric concentration 100 times greater in the tire smoke (though such smoke may well be too coarse to be respirable).
Careful measurements on a CFM-56-2C1 engine (Wey et al. 2006), developing 98 kN of thrust, suggested that this should emit ∼0.214 g s−1 of soot with a median diameter of 40 nm. Scaling up to a Boeing 777 (B777), suggests a soot mass emission rate of ∼1.8 g s−1 on takeoff. (Obviously, this will vary greatly with the engines employed and their state of maintenance.) By comparison, Morris (2006) suggests that the mass of rubber loss per landing is a fraction ∼2 × 10−6 of the aircraft takeoff weight. For a B757-200, this amounts to 230 g. These values are thus broadly consistent with the lidar signal that we observe.
The lidar did not detect any PM emissions from braking. Such emissions would, of course, be spread out along a much greater distance of runway. In addition, because the airframe by this stage is generating much less lift, flow around it will be much less effective at lifting the dust away from the ground and into the path of the lidar.
6. Discussion and conclusions
The lidar is a safe and effective method for monitoring the dispersion of aircraft exhaust plumes. With the conversion of our system to frequency-tripled YAG, it has become eye safe, while improvements in the optics and data processing have given it sufficient sensitivity to see exhaust plumes from modern aircraft within a range window of 150–800 m and with a range resolution of 5 m.
Our previous experience with the lidar had been in applications to continuous stack emissions. Aircraft plumes, however, are transient and much more tenuous. In our previous applications, we had automatically distinguished plume from background on a scan-by-scan basis. For aircraft, this was no longer adequate and we were obliged to develop a technique to parameterize the ambient background on a shot-by-shot basis. Because aircraft emissions are often very close to the ground, it was also necessary to develop techniques to identify and exclude returns from hard targets. Measurements of such tenuous plumes also require close attention to the optimization of the lidar optics and detection systems for the anticipated range of the target.
Overall, nearly 1800 air traffic movements (ATMs) have been monitored: 439 at Heathrow and 1353 at Manchester. While these ATMs included all stages of the LTO cycle, the most successful observations were of the start of the takeoff run. PM concentrations in the exhaust plume were then almost always dense enough, and particle diameters large enough, to be observed by the lidar.
These observations were extremely valuable in supporting the theoretical expectation that the starting plume is initially dominated by the Coanda effect, forming a ground-hugging wall jet. This effect also forces the emissions from the individual engines to merge into a single plume. As a wall jet, this plume deepens to a head, which may ultimately (after about 80 s) part contact from the ground, apparently under the influence of buoyancy. This supports the analysis of Carslaw (Department for Transport 2006), who invoked buoyancy to explain the low observed concentrations at Heathrow in light winds. Using these measurements, we have developed a theoretical framework of such behavior (Graham et al. 2010, manuscript submitted to J. Appl. Meteor. Climatol.), which should be parameterized into regulatory models in the near future.
As the aircraft accelerates down the runway, lift around the airframe increases. This manifests itself through counterrotating wing vortices, which then trap the engine emissions, dividing them into port and starboard plumes. These may already be seen before the end of the takeoff run, but become clearly apparent after takeoff, where they match the expectations of models of wing vortices developed by Graham and Raper (2006). On ∼30% of occasions, they may also be detected in the more tenuous plumes emitted by aircraft on approach. The vortices descend at ∼1 m s−1, carrying the engine emissions with them to the ground.
The lidar can easily detect smoke released from the tires on landing. Scattering calculations suggest that the resultant gravimetric concentrations of such aerosol are likely to be substantially greater than those of engine smoke. The lidar alone, however, cannot determine the particle size spectrum of the smoke, or distinguish between soot from burning rubber and mechanically generated dust from the runway surface. In either case, the aerosol may well be too coarse to be respirable.
All of these data (lidar, ATM, and meteorological) have now been amalgamated into an Aviation Air Quality database (Schäfer et al. 2009) and can be made available to bona fide researchers.
The work at Heathrow was funded by the U.K. Department for Transport and that at Manchester Airport by the Engineering and Physical Sciences Research Council. We are grateful to the management and staff at both airports for their generous and willing help in carrying out these measurements. We are grateful to Landis and Gyr (Stockport) for permission to site the lidar in their staff car park. Funding for archiving the data was provided by the EU Network of Excellence “Environmentally Compatible Air Transport Systems.”
DOAS measurements were made by Dr. Clive Oppenheimer and Dr. Vitchko Tsanev (Cambridge) and Dr. Andrew McGonigle (Sheffield). UV-Falcon measurements were made by Dr. Ivonne Cruz-Jimate (Manchester). We are also grateful to Dr. Adrian Hayes, Dr. Des Doocey, Ms. Georgina Sawyer, and Ms. Tsvetina Evgenieva for technical assistance.
We are grateful to Wynn Eberhard of NOAA and Kevin Morris, formerly of British Airways, for helpful discussions.
Corresponding author address: Michael Bennett, Chester St., Centre for Air Transport and the Environment, Manchester Metropolitan University, Manchester M1 5GD, United Kingdom.Email: firstname.lastname@example.org