Transport of Canadian forest ﬁre smoke over the UK as observed by lidar

. Layers of aerosol at heights between 2 and 11 km were observed with Raman lidars in the UK between 23 and 31 May 2016. A network of such lidars, supported by ceilometer observations, is used to map the extent of the aerosol and its optical properties. Spaceborne lidar proﬁles show that the aerosol originated from forest ﬁres over Western Canada around 17 May, and indeed the aerosol properties – weak depolarisation and a lidar ratio at 355 nm in the range 35–65 sr – were consistent with long-range transport of forest ﬁre smoke. The event was unusual in its persistence – the smoke plume was drawn into an 5 atmospheric block that kept it above North-west Europe for nine days. Lidar observations show how the smoke layers became optically thinner during this period, but the lidar ratio and aerosol depolarisation showed little change. The study shows the value of combining different kinds of lidars in following the evolution of long-range smoke transport events.

. We show here how a dense network of lidars and ceilometers tracked the evolution of a smoke episode from 23 to 31 May 2016 as an atmospheric block trapped the air over Western Europe. Spaceborne lidar data from the CALIOP and CATS instruments, supported by SEVERI images from Meteosat-10, enables the smoke to be tracked back unambiguously to the fires over Alberta on 17 May. A brief description of these facilities is now given.

Capel Dewi Raman Lidar
The Capel Dewi Raman lidar is a biaxial ultraviolet lidar based on that used for the EARLINET project in 1999-2002(Wandinger et al., 2004. Since then, it has been updated and now contains a Continuum 8030 Nd-YAG laser emitting pulses 15 at 354.7 nm at 30 Hz with pulse energy 300 mJ. A tenfold beam-expanding telescope directs the light vertically into the atmosphere. The receiver is based on a 1 m diameter mirror used in a Nasmyth-Cassegrain configuration, which directs the backscattered radiation through a collimator on to a dichroic beamsplitter (Fig. 2). This beamsplitter reflects and transmits radiation with wavelength greater or less than 397 nm, respectively, into the receiver channels. Interference filters centred around 387 and 408 nm isolate Raman scattering from nitrogen and water vapour respectively (Table 1). A third channel measures 20 elastic backscattered radiation reflected from the other two filters. The receiver is mainly sensitive to the polarisation component parallel to the laser, which reduces background noise in the Raman channels but does not permit measurements of the aerosol backscatter when there is a significant cross-polarised component.
The lidar is designed for free tropospheric measurements and so the receiver field-of-view does not fully overlap the laser beam below 2 km. Measurements below 2 km are therefore not used here. 25 The signals are measured using EMI 9124 photomultipliers and a photon-counting electronics system (ORTEC PCI-MCS) with range resolution 15 m (100 ns time bins). A dead time of 10 ns is applied in the counting system, and corrected according to the non-paralysable equation: where S 0 is the measured count rate, S is the corrected count rate and τ the dead time. nm is 10 −8 . Neutral density filters are used in the elastic channel to avoid saturating the signal at low altitude.
Although elastic measurements can be made in daytime, the Raman signals are too noisy in daytime and are only collected at night. The system is normally operated alongside a second lidar which measures both polarisation components of the elastic signal, but this system was inoperative during the period of interest here.

Met Office Ceilometers
The Met Office operate an extensive network of ceilometers around the UK, of various types. The 12 Lufft CHM 15k ceilometers are of particular interest to this study, because of their greater sensitivity to thin aerosol layers with low backscatter. These ceilometers emit infrared radiation (1064 nm) and use photon-counting detectors. While they cannot provide the quantitative detail of the Raman lidars, they operate continuously and can provide more complete coverage in space and time of the 5 free-tropospheric aerosol.

Retrieval of aerosol optical depth and lidar ratio
The power received by a lidar, P (z), obeys the lidar equation (Wandinger, 2005), which for elastic (P e ) and Raman (P R ) scattering takes the form: 10 and respectively. Here, β aer , β ray and β ram are the backscatter coefficients for aerosol, elastic molecular (Rayleigh) and Raman scattering respectively, n(z) is the number density of air molecules, z is the height above the lidar, and σ ram and σ ray are the scattering cross-sections for Raman and Rayleigh scattering by air molecules, which are taken to be 1.929×10 −30 m 2 and 15 2.76×10 −30 m 2 respectively (Bates, 1984). The extinction coefficient of aerosol, α, is assumed to be the same at the elastic and Raman wavelengths.
Retrieval of the aerosol optical depth uses the N 2 Raman signals and a nearby radiosonde profile. From the latter, a synthetic molecular-only scattering profile may be constructed, and fitted to the measured profile in a region of the atmosphere free from aerosol, here taken to be 13-16 km. The ratio Ram(z) of the measured to the synthetic signals then leads to curves such as that shown in blue in Fig. 3. In principle, this ratio can be inverted to give a profile of α(z), but this approach tends to lead to large random errors. We take advantage here of the fact that in the episode under discussion the aerosol was distributed in very distinct layers, so a method was devised to calculate only layer-average or layer-total quantities. Figure 3 also shows R(z), the ratio of the elastic channel to the Raman channel, normalised to 1 between 13 and 16 km. Aerosols show up in this curve as 5 departures from 1 (the molecular background), clearly showing the layered structure. (A small correction has been applied to R(z) to account for the difference between σ ram and σ ray , using the radiosonde profile). Such structure was observed at all sites during the course of this event. We therefore calculate the integrated aerosol optical depth (AOD) across each layer: where the overbars indicate that Ram(z) has been averaged in the aerosol-free regions above and below each layer. Each lidar 10 profile used here was examined separately to determine the layer altitudes and the width of the aerosol-free regions, which were chosen as far as possible to be at least 1 km deep.
As the photon-counting signals can be assumed to follow Poisson statistics, the precision error in AOD is readily calculated from the number of photon counts in the regions above and below each layer. A further source of error comes from the choice of radiosonde profile used to normalise P R (z). For stations like Camborne and Watnall where co-located radiosondes were 15 released, this error is small, but for the other stations it is not negligible. As an example, Fig. 4 shows the same lidar data as Fig. 3, but using a radiosonde from Camborne rather than Castor Bay (Fig. 1). The difference in AOD for the five layers is greater than the statistical uncertainty, showing that this source of error is important for free tropospheric aerosol measurements by Raman lidar.
Retrieval of the integrated aerosol backscatter, IAB, and hence the mean lidar ratio LR (LR = AOD/IAB) for each layer 20 requires measurement of both polarisation components, and this was only possible for the Met Office Raymetrics lidars.
Signals from the two elastic channels were added (after calibration of the cross-polarised channel as described by Buxmann et al. (2017)), and used to generate an R(z) profile as before. The integral of B[R(z) -1]n(z) across each layer then gave IAB (where B = 3.31×10 −31 m 2 sr −1 is the molecular differential backscatter cross-section, after Bates, 1984). Errors in IAB come from the Poisson statistics in both the signals in the layer and the background noise subtracted from the measured lidar 25 signals, which are treated differently under the integral (variances being added for the signals and standard deviations for the background). Finally, the two errors are added in quadrature to give the error in IAB, and LR calculated for each layer in the usual way.

Results
To gain an appreciation of the extent and persistence of the aerosol, we first examine the ceilometer measurements since  To examine the duration of the event, the total number of ceilometers which observed definite aerosol layers, trace amounts or no aerosol, or were restricted by cloud cover, was plotted for each day from 22 to 27 May (Fig. 7). This plot shows that the event seems to have peaked in terms of coverage on 24 May, although the increasing cloud cover thereafter means that some aerosol is likely to have been missed. None of the ceilometers detected aerosol after the 27th, although continuing low cloud cover restricted observations to around half the stations until clear skies returned on 5 June.

15
The ceilometers only provide consistent measurements up to 8 km, and their infra-red wavelength means they can only give qualitative information on the presence or not of aerosols. To extend the measurements to the tropopause and obtain quantitative information about the aerosol, we now turn to the Raman lidars. A qualitative inspection of the Met Office elastic channels (parallel and perpendicular polarisation) showed that there was extensive aerosol between 8 km and the tropopause which was not captured by the ceilometers. This was observed at Camborne during 24-28 May, East Malling from 25-27 May, Exeter 20 from 25-31 May and Watnall from 26-31 May. The Capel Dewi lidar also observed aerosol between 8 and 12 km up to the end of May. This shows that the event persisted from 22 to 31 May with aerosol layers found at all altitudes in the troposphere.
We now concentrate on the night-time measurements from the Raman lidars to obtain quantitative information on these layers.
A similar analysis to that described in section 3 was conducted for all the continuous night-time data collected by the Raman lidars between 23 and 31 May, after eliminating periods affected by cloud. Figure 8 shows the total aerosol optical depth above 25 2 km measured by the Raman lidars during this period, aggregated into whole-night averages. The greater sensitivity of these lidars means that aerosol was measured up to the night of 30-31 May at Capel Dewi with traces evident up to the same time at some of the other stations. However, the coverage is patchy due to the combination of intermittent sampling and low cloud cover.
The highest AOD measured was that at Capel Dewi on the night of 23-24 May, from the profile shown in Fig. 4. Although 30 the value is sensitive to the choice of radiosonde profile, the total AOD of ∼0.13 above 2 km clearly results from aerosol observed throughout the free troposphere. All other AOD measurements at all the stations were below 0.1, with the values decreasing with time to below 0.05 after the 28th.
The evolution of lidar ratio as a function of optical depth is shown for the Met Office lidars in Figs. 9 and 10, for aerosol layers above and below 7 km respectively. Here, hourly average data are presented, as there could on occasion be considerable to the south-west and south-east.) As the block moved and distorted, the UK lay first under the eastern trough (0600 UTC on the 22nd to 1200 UTC on the 23rd), then the anticyclone (1800 UTC on the 23rd to 1200 UTC on 24th, Fig. 12c), the western cut-off low (1800 UTC on the 24th to 0000 UTC on the 29th, Fig. 12d) and finally a broad area of almost no flow which persisted until a second, weaker omega block was established on the 30th as another depression developed in the western Atlantic and moved eastwards (not shown). From 23-31 May therefore the flow over the UK was slack and variable, which 5 meant that smoke transported in the zonal jet up to the 22nd was able to remain in the vicinity of the UK.

Air parcel trajectories
We now examine air parcel trajectories for evidence that the aerosol-laden air crossed the Atlantic from Canada. To be useful for this purpose, trajectories need to be non-dispersive -i.e. trajectories from nearby starting points need to follow a similar path. Unfortunately, this did not prove to be the case for most of this event, precluding any meaningful conclusion on air-mass 10 origin. We concentrate therefore on the period leading up to the start of the event when coherent sets of trajectories were found.
Trajectories were calculated using NOAA's HYSPLIT trajectory model (Draxler & Hess, 1998;Stein et al., 2015), both backward in time from the locations of the lidars and forward in time from locations in Western Canada. A matrix of 9 starting points was defined, spaced by 0.5 • in latitude and longitude; low dispersion of these 9 trajectories is required if the calculations are to be considered reliable. However, examples like this proved rare. At other heights on 24 May, the back-trajectories from Capel Dewi were too dispersive to reveal an air mass origin -by then the block was well set up with slack, incoherent flow. We therefore turn to 25 satellite observations for evidence that the smoke crossed the Atlantic.

Satellite data
Several sources of data were used to track the smoke plume from Canada to the UK: In these images, smoke appeared as a faint blue-grey colour and was most distinct just after dawn and just before dusk, when the scattering of sunlight towards the satellite from the small smoke particles was more prevalent.
The eastward transport of aerosol across North America and over the Atlantic Ocean in Figs. 13 and 14 is shown by the OMPS-AI measurements (Fig. 15). The broad shape of the smoke plume heading eastward from Alberta is consistent with the HYSPLIT trajectories shown in Fig. 13, and shows aerosol reaching the western Atlantic on the 20th. Thereafter, the smoke 20 tends to disperse and progress eastward towards Europe, with a strip of elevated aerosol index lying west of Ireland by 0500 UTC on 22 May.
The passage of smoke eastward can also be followed in night-time CALIOP data. Figure 16  The presence of smoke across the Atlantic Ocean is best shown by the total and perpendicular backscatter measurements by the CATS lidar in the early hours of 22 May (Fig. 19). Optically thin aerosol is identified as the light blue layers in the total backscatter plot, and further classified as smoke by the lack of such layers in the perpendicularly polarised signal. Figure 19 shows that by 22 May the smoke plume extended from 55 • W to 15 • W and was present from the top of the boundary layer to above 10 km.
The EUMETSAT Natural Colour RGB analysis of SEVIRI data shows the arrival of smoke over the UK (Fig. 20) This study has presented observations of free-tropospheric aerosols by ceilometers and Raman lidars over the UK from 23-31 May 2016, and examined the origin of the aerosol. The principal conclusions are as follows.
-Ceilometer measurements showed that much of the United Kingdom was covered by free-tropospheric smoke layers on 23 and 24 May.
-Raman lidar observations showed that the smoke was found throughout the troposphere, but with the greatest optical 20 depth above 7 km.
-The maximum optical depth measured was ∼0.15 with most values between 0.1 and 0.05: these values diminished with time through the event.
-The properties of the aerosol as determined from Raman lidar were consistent with those of smoke from forest fires: low volume depolarisation (<6% and a lidar ratio in the range 35-65 sr).

25
-The smoke lingered over western Europe for nine days due to an atmospheric block which prevented eastward advection.          in the bottom right image is the shadow of a smoke streak on some low-lying water clouds. Images taken from EUMETSAT web site.