The 2019 Raikoke volcanic eruption part 2: Particle phase dispersion and concurrent wildfire smoke emissions

Between 27 June and 14 July 2019 aerosol layers were observed by the United Kingdom (UK) Raman lidar network in the upper troposphere and lower stratosphere. The arrival of these aerosol layers in late June caused some concern within the London Volcanic Ash Advisory Centre (VAAC) as according to dispersion simulations the volcanic plume from the 21 June 2019 eruption of Raikoke was not expected over the UK until early July. Using dispersion simulations from the Met Office Numerical Atmospheric-dispersion Modelling Environment (NAME), and supporting evidence from satellite and in5 situ aircraft observations, we show that the early arrival of the stratospheric layers was not due to aerosols from the explosive eruption of the Raikoke volcano, but due to biomass burning smoke aerosols associated with intense forest fires in Alberta, Canada that occurred four days prior to the Raikoke eruption. We use the observations and model simulations to describe the dispersion of both the volcanic and forest fire aerosol clouds, and estimate that the initial Raikoke ash aerosol cloud contained around 15 Tg of volcanic ash, and that the forest fires produced around 0.2 Tg of biomass burning aerosol. The operational 10 monitoring of volcanic aerosol clouds is a vital capability in terms of aviation safety and the synergy of NAME dispersion simulations and lidar data with depolarising capabilities allowed scientists at the Met Office to interpret the various aerosol layers over the UK, and attribute the material to their sources. The use of NAME allowed the identification of the observed stratospheric layers that reached the UK on 27 June as biomass burning aerosol, characterised by a particle linear depolarisation ratio of 9%, whereas with the lidar alone the latter could have been identified as the early arrival of a volcanic ash / sulphate 15 mixed aerosol cloud. In the case under study, given the low concentration estimates, the exact identification of the aerosol layers would have made little substantive difference to the decision making process within the London VAAC. However, our work shows how the use of dispersion modelling together with multiple observation sources enabled us to create a more complete description of atmospheric aerosol loading. 1 https://doi.org/10.5194/acp-2021-448 Preprint. Discussion started: 7 July 2021 c © Author(s) 2021. CC BY 4.0 License.

, and Nabro in June 2011 . Satellite imagery of the plume indicates that the eruption was ash-rich in nature, estimates of the emission are derived in section 3.1.
The 2019 wildfire season in the Canadian province of Alberta was unusually severe, with more than 883,000 hectares 90 burned across the province, compared to the five yearly average of 245,000 hectares (Short, 2019a;Jenner, 2019;Gabbert, 2019). The Mackenzie County area in the north of Alberta suffered a particularly large number of fires in May and June 2019, resulting in the evacuation of thousands of people and the destruction of property (Short, 2019b). Continued hot and dry conditions along with high winds caused intensification of the wildfires; the heat released fuelled pyrocumulus clouds, which lofted smoke into the UTLS and allowed high-level winds to transport it long distances. On 17 June 2019, four days prior to 95 the eruption of Raikoke, wildfires burning in Chuckegg Creek in Mackenzie County produced biomass burning aerosols and formed pyrocumulus clouds which breached the tropopause and entered the lower stratosphere. Figure 1a shows a true colour reflectance MODIS image from the 17 June 2019 overlaid with thermal anomalies associated with forest fires; an aerosol cloud can clearly be seen to have moved north-eastwards from an area with several active fires. We will show that the aerosol cloud continued eastwards over the following days, with the lower tropospheric layers moving south over the United States and the 100 stratospheric layers continuing eastwards to reach Europe by 26 June.

NAME dispersion model simulations
The dispersion of gas and aerosols is simulated within NAME by tracking notional 'particles' as they move within air parcels advected by the wind vectors supplied by a driving meteorological model. Operationally the Met Office Unified Model (UM) 105 version 7.1 (Walters et al., 2019) provides the global driving meteorology with a three-hourly temporal resolution and approximately 10 km horizontal resolution at mid-latitudes, and on 59 model levels between the surface and 30 km altitude. Turbulence and sub-grid scale diffusion are represented within NAME by the addition of random perturbations to the wind vectors at each time step (Webster et al., 2012). de Leeuw et al. (2020) make a detailed quantitative investigation of the effect of this diffusion parameter on the spatial extent and concentration of the diffused volcanic SO 2 cloud. For the Raikoke eruption, de Leeuw 110 et al. (2020) obtained better agreement between the simulated and observed SO 2 clouds by using a reduced parametrisation of sub-grid scale diffusion, although the improvements were confined to the first five days after the eruption. The BBA and ash simulations described here were run using both the default and the reduced diffusion parameter. However, the satellite observations of the relatively low aerosol concentrations emitted by the Raikoke eruption and the Alberta forest fire were generally of poorer quality than the TROPOMI SO 2 observations, and we found no substantive improvement between the aerosol 115 simulations with the default and reduced diffusion parameter, when compared with the available observations. Therefore, only the results using the default values for diffusion as detailed in Webster et al. (2012) are shown here. NAME simulations are initialised by defining source terms that describe the location, vertical profile and mass emission rate of the event producing the gas or aerosol being simulated. Where relevant, there is the option to define further characteristics such as chemical species, size distribution and particle shape. The emitted material is then subject to chemical and/or physical 120 transformations and can be removed by gravitational settling, wet and dry deposition. Relevant to this study is the chemical transformation of sulfur dioxide (SO 2 ) into particulate sulfate aerosol  ). This process is represented within NAME by gaseous and aqueous phase chemistry schemes. The species involved in these reactions are simulated explicitly within Table 1. Cumulative aerosol size distributions for BBA and Volcanic ash according to Hobbs et al. (1991); Maryon et al. ((1999);Petzold et al. (2007),used in the NAME simulations. schemes to laboratory measurements and found that the fall scheme described by Ganser (1993) was the most accurate. We 135 have therefore adopted this scheme in our simulations. As shown in Beckett et al. (2015) the aerosol particle size distribution (PSD) used to initialise NAME aerosol simulations has a significant impact on the number of particles removed by gravitational settling, and therefore on the number that survive in the distal aerosol cloud. Without in-situ sampling, or samples collected near the fires or volcano, it is not possible to know the size distribution or particle shape of the emissions from either the fires or the Raikoke eruption and this distribution must be assumed. For operational ash model simulations, the London VAAC uses 140 a default PSD that is based on in-situ measurements of the 1990 Mount Redoubt ash cloud (Hobbs et al., 1991;Maryon et al., (1999), and in-lieu of measurements of the Raikoke ash, we have used this PSD in our simulations. For the BBA cloud we have used a size distribution from in-situ measurements for Canadian forest fire emissions from Petzold et al. (2007). The distribution of mass over these size distributions is shown in table 1. Saxby et al. (2018) describe how for real particles for a given volume and density, non-spherical particles fall more slowly than spheres owing to their larger projected area, and that 145 therefore the particle shape used when initialising NAME also has an impact on the number of particles surviving in the distal field. NAME uses a "sphericity" parameter to describe the degree of non-sphericity of a particle, taking a value between 0 and 1 (with the sphericity of a sphere being equal to 1). The sphericity parameter is the ratio of the surface area of a sphere with equivalent volume to the actual surface area of the particle. To try to best capture the long-range transport of large grains, we have used a sphericity parameter of 0.5 for our simulations of volcanic ash. For BBA, which we expect to be nearer spherical 150 (e.g. Taylor et al., 2020), the sphericity parameter was set to 1.

Mass emissions and vertical profiles
Volcanic ash -For volcanic ash simulations, the approach used by the London VAAC is to assume that most of the larger grains of the total mass ejected by an eruption fall out close to the volcano, leaving only a fine mass fraction (FMF) available to be dispersed. Operationally this FMF is set somewhere between 1% and 5% of the total ejected mass, with 95% to 99% of the total 155 mass assumed to have fallen out close to the volcano. To estimate the amount of ash ejected during the 2019 Raikokeeruption, we have used the relationship between plume top height and mass eruption rate presented in Mastin et al. (2009). Using a plume top height of 15 km as reported by the Volcano Response group (VolRes -https://wiki.earthdata.nasa.gov/display/volres -an international climate initiative from the WMO-SPARCSSiRC working group (http://www.sparc-ssirc.org)), and eruption start and end times of 21:00 (UTC) on the 21 June to 03:00 (UTC) on the 22 June (taken from HIMAWARI-8 imagery) we 160 estimated a total ejected amount of ash of around 300 Tg. We use this estimate as a representative figure. We note that the eruption was not continuous during this time, but rather progressed in several pulsed phases, each of which possibly reached different altitudes -and as such, this is an upper estimate. Using this value gives a FMF of between 3 and 15 Tg. Again taking the upper limit, we released a total of 15 Tg of ash over a vertical profile taken from the VolRes report ( fig. 2 in red). This vertical profile was formulated from a best estimate for the Raikoke SO 2 emission profile using IASI measurements. In lieu of 165 a specific profile for ash, we have adopted this SO 2 vertical profile as the best available estimate for ash. Note however that ash is not necessarily released in the same profile as the gas emissions.
SO 2 -For our volcanic SO 2 simulations we use the source term and dispersion parameters from de Leeuw et al. (2020) that have been optimised against TROPOMI observations. This source term uses the height profile (shown in fig. 2 in orange) referred to as "Stratprofile" in de Leeuw et al. (2020) using a total emission of 1.57 Tg of SO 2 . In this profile, a larger fraction of the 170 total mass is released into the stratosphere compared to the un-optimised profile reported in the VolRes report (and as described above is used for the volcanic ash source term). As described in de Leeuw et al (2021), differences between the vertical profiles do not necessarily make them inconsistent because the VolRes profile is derived closer to the time of the eruption while those derived in de Leeuw et al. (2020) are subsequent to potential additional lofting caused by the absorption of solar radiation (Muser et al., 2020).

175
Sulphate aerosols -The sulphate aerosol results presented below are the result of the conversion of the simulated SO 2 into sulfate by the chemistry scheme within NAME, and no primary sulphate was emitted in the Raikoke simulation source term.
Biomass burning aerosol -The location and vertical profile for the source term used to initialise the BBA dispersion simulation was derived from a MODIS image from the 17 June 2019 (shown in Fig. 1a). In this image a smoke cloud has already been advected in a north easterly direction from an area of active forest fires. While there are no measurements available to show 180 the height of the aerosol cloud, we assume that it is a hot and buoyant aerosol cloud that has continued to rise into the lower stratosphere as it was advected away from the source fires (Fromm et al., 2005(Fromm et al., , 2006. To represent this in NAME we have used a series of stacked and offset cylinders. A conceptual diagram of this structure is shown in fig. 1b. An arbitrary mass of 0.1 Tg of material was released in each of the cylinders and the results later scaled to match observations (see section 4). Height profiles of mass of ash and SO2 released in NAME dispersion simulations of Raikoke. The ash profile (in red) was taken from the VolRes group report which used IASI satellite observations to arrive at an emission profile for SO2. In lieu of a profile for ash only this profile has been adopted for the ash simulations. Plotted in orange (and hatched) is the 'StratProfile' from the part 1 paper -this is the altered profile for the release of SO2 arrived at following a detailed comparison between the SO2 product from TROPOMI and NAME simulations. In this emission profile a larger fraction of the total mass of SO2 is released in the stratosphere than in the original VolRes profile.

185
Having initialised the simulations as described, NAME was configured to provide output files for a) column loading (two dimensional -latitude x longitude), and b) air concentration (three dimensional -latitude x longitude x altitude -with altitude measured from sea level). The data in each hourly file represents the mean of the simulated values calculated at ten-minute intervals in the preceding hour.

190
In this section we briefly describe the observations used to track the various aerosol clouds. The observations, platforms and applications in this study are summarised in Table 2.   . Additionally, we use the procedure described in Freudenthaler (2016) to correct for the polarisation effects of the various optical elements in the lidar. For each lidar we have also measured the angular offset between the plane of polarisation of the emitted laser beam, and that of the optical elements used to split the incoming light by polarisation state, using a method similar to that described in Alvarez et al. (2006), and any measured offset has been included in the polarisation corrections. To confirm the robustness of the calibration, the volume depolarisation ratio is measured in pristine air and compared against a 205 theoretical value calculated as in Behrendt and Nakamura (2002) and checked to see that it is within 1% of expected values.
NASA-MPLNET lidars -The NASA Micro Pulse Lidar (MPLNET) network currently has 17 active sites around the globe, each with a depolarisation MPL and a co-located Cimel sun-photometer (Welton et al., 2001;Welton and Campbell, 2002;Welton et al., 2018). The instruments operate at 532 nm and also have co-polar and cross-polar depolarisation detection. To study the vertical structure of the Raikoke and BBA clouds we make use of data from the MPLNET sites at Fairbanks in Alaska 210 (64.86N, 147.85E) and the Goddard Space Flight Center in Washington DC (38.99N,76.84E). We make use of the products from the new version 3 processing system which provides normalized relative backscatter (NRB), volume depolarisation ratio (VDR) as well as particle linear depolarisation ratio. which Mie calculations are unsuitable, we have used a literature value. We do not take into account factors such as hygroscopic growth or particle cohesion etc. that will have an effect on scattering efficiency and these values are representative only. A more 220 complex treatment of K ext is beyond the scope of this study. Errors in the mass estimates have been calculated by extending the Monte Carlo technique used to take into account the statistical errors in the raw lidar signals to include errors in the assumed depolarisation ratios and K ext . In total, the error in the mass estimates is of the order of ±50%.
Dust RGB images -In section 4 we present dust RGB images made using infrared bands B11 (8.6 µm), B13 (10.4 µm) and B15 (12.4 µm) from the advanced HIMAWARI imager carried by the HIMAWARI 8 satellite and compare these with the 225 NAME model simulations for volcanic ash (Francis et al., 2012). The RGB images were created by assigning red to the difference between B15 and B13, green to the difference between B13 and B11, and blue to B13. In this scheme, aerosol clouds containing volcanic ash should appear bright red, or bright yellow if there is also significant SO 2 (Francis et al., 2012).
While not a quantitative measure, the RGB imagery is useful for assessing the horizontal spatial distribution of volcanic ash clouds during the initial dispersion.

230
UV Aerosol index -The Ozone Mapping Profiler Suite (OMPS) carried in the Suomi NPP satellite measures backscattered ultraviolet radiance. The UV aerosol index (AI) is derived from radiances at 340 nm and 378.5 nm, with a temporal resolution of 24 hours (Torres et al., 1998. A positive AI value indicates the presence of UV absorbing aerosols such as dust (Christopher et al., 2008), volcanic ash (Carn and Krotkov, 2016) or smoke (Torres et al., 2020). The parameter is unitless and takes a value between 0.0 and 5.0, with 5.0 indicating heavy aerosol loads. The AI is available from https://snpp-235 omps.gesdisc.eosdis.nasa.govdataSNPP_OMPS_Level2 (date last accessed 21/06/2020). The product is semi-quantitative, and is used in section 4 to assess the horizontal spatial distribution of the BBA cloud from the Alberta fires.
CALIOP backscatter and aerosol vertical feature mask -The CALIOP lidar on board the CALIPSO satellite measures elastic backscatter and depolarisation at 532 and 1064 nm (Winker et al., 2009). The level 2 data downloaded, available from http:www-calipso.larc.nasa.gov (last accessed 10/07/2020) also includes a vertical feature mask (VFM), which identifies 240 aerosol layers and assigns an aerosol subtype -for example dust, volcanic ash or elevated smoke (Omar et al., 2009). In section 4 we use CALIPSO data to assist with tracking both the volcanic and BBA aerosol clouds as they travel to the UK.

IAGOS aircraft-based observations
The European research infrastructure In-service Aircraft for a Global Observing System (IAGOS; see www.iagos.org for details) performs routine measurements of basic atmospheric state variables, atmospheric chemistry and aerosols using their 245 standard payload on a fleet of passenger aircraft equipped with IAGOS-CORE instruments (Petzold et al., 2015).  5.411. The optical particle counter used here measures particle concentration (as number of particles per cm3) for particle diameters between 0.25 µm and 2.5 µm.
produced by the Raikoke eruption and the Alberta forest fires as they were transported to the UK, before using the simulations to assist in interpreting the UK lidar data. We first look at the initial five to six days subsequent to the emission of the ash and BBA, and then use the NAME simulations to track the transport to the UK.

Volcanic aerosols: transport
The left-hand column of fig. 3 shows the HIMAWARI 8 dust RGB images for the first four days after the 21 June Raikoke 260 eruption. The central column shows the NAME simulated column loadings for volcanic ash, and the left-hand column shows the NAME simulated column loadings for sulfate aerosol. In the simulations both the ash and sulfate aerosol clouds initially move east from the volcano while also dispersing to the north and south. The leading edges of the aerosol clouds then begin to be entrained by a low pressure cyclonic system centred at around 55 o N and 175 o W (seen as a green and brownish cloud in the RGB image) and turn northwards over the Aleutian Islands. The eastern ends of both the ash and sulfate clouds split 265 into two distinct structures. Figure 3 shows that the model, the aerosol clouds continue to disperse and reach Russia by the 24 June and Alaska by the 25 June. As expected, the sulfate aerosol concentration is initially relatively low, and increases as the oxidation of SO 2 progresses. The RGB images reflect the general structure of the most intense parts of the model-simulated aerosol clouds and show the aerosol cloud moving east and being entrained in the warm conveyor of the cyclonic system before encircling it and dispersing to the east and west. Figure 3 shows that on the 22 June the northern part of the aerosol cloud is 270 bright yellow -indicating the presence of SO2 as well as ash, whereas the south-westerly part is bright red, indicating that the infrared emission in this region was dominated by ash (Francis et al., 2012). Several of the model structures are absent however, and this may in part be due to an overlying cloud obscuring the aerosol cloud from the sensor. For example, on the 23 June the most southerly structures in the model aerosol clouds are co-located with an area of cloud in the RGB images, which appear as green and brownish colours. After 25 June the aerosol cloud can no longer be discerned in the RGB product. is occluded by the cloud at around 10 km. In areas where the model places ash in the troposphere, the VFM has identified this as dust or polluted dust (2 = yellow and 5 = brown respectively), and sulfate has sometimes been identified as elevated smoke (6 = black). Almost all of the stratospheric layers identified in the VFM have been given the "volcanic ash" classification (9 = light grey), with very few "sulphate/other" classifications. According to Kim et al. (2018) the distinction between the two classifications for layers that are optically thicker than 0.001 sr −1 (vertically integrated attenuated backscatter) is based on the 295 PDR: if it is above 15% the layer will be classified as "volcanic ash". Depending on the relative concentrations, a mixture of sulfate aerosols (PDR ≈ 1% (e.g. Ansmann et al., 2012)) and strongly depolarising volcanic ash (PDR ≈ 35% (e.g. Ansmann et al., 2011)) may well result in an overall PDR of more than 15% (Groß et al., 2013) -and the designation of these layers as ash in the VFM is not inconsistent with an ash-rich mix of ash and sulfate aerosols.
After having analysed the volcanic plume dispersion over the ocean using CALIPSO, we can keep following the ash and the 300 sulphate aerosol over the American continent from ground-based remote sensing instruments. Figure  Late on 24 June a banked structure (red outline 1) can be seen starting at 10 km and descending to around 5 km by 26 June.
While this structure is mostly dominated by meteorological cloud, it appears to contain some aerosol, noticeable in the areas with no cloud. The data does not allow the retrieval of a PDR or mass concentration in this structure. Above this structure, 305 (red outline 2) a vertically thin aerosol layer can be seen at around 12 km, persisting from early on 25 June until around midday on 27 June. The whole day mean PDR within this layer on 26 June is 25%, consistent with a mixture of ash and sulfate particles (Ansmann et al., 2011;Groß et al., 2013). Using assumed values for PDR for ash (35%) and sulfate (0.01%) together with an ash-like specific extinction of 0.6m −2 g −1 (Ansmann et al., 2011;Marenco et al., 2011), and a Mie calculated sulfate specific extinction of 4.2m −2 g −1 to convert the particle extinction measurements into separate mass concentration estimates 310 for ash and sulfate, the mean estimates within this layer are 250 µgm −3 and 10 µgm −3 for ash and sulfate respectively. Given the uncertainty of the assumptions for PDR and specific extinction, the error on these mass estimates is on the order of 50% vertical shear in the atmosphere with faster transport at higher altitudes (e.g. Dacre et al., 2015). Both of these structures are somewhat representative of the outlined structures 1 and 2 in the lidar data. However, there is a large structure between 5 km and 10 km in the simulations that is not present in the lidar observations. Attempts at modifying the vertical structure of the ash source term by changing the mass released at each altitude could not remove this structure. The maximum concentrations in the higher layers between the times indicated by the vertical red lines is 400 µgm −3 and 25 µgm −3 for the ash and sulfate 325 simulations respectively. While these values are higher than those estimated from the lidar data, they are of the same order of magnitude (and their ratio is more or less the same), giving some indication that the mass of ash allowed to remain in the distal field, and the amount of sulfate produced in the simulations by the chemistry scheme are somewhat representative of the real values.
15 https://doi.org/10.5194/acp-2021-448 Preprint. Discussion started: 7 July 2021 c Author(s) 2021. CC BY 4.0 License. In the right hand panels a selection of CALIPSO overpasses are marked in magenta, and four CARIBIC flight tracks are marked in cyan.  Figure 6 shows the OMPS AI, together with the NAME simulated BBA column loadings for 18 June to the 24 June. In both the simulations and the AI the BBA cloud initially moves fairly uniformly north east towards Hudson Bay for around 24 hours, before the leading edge is caught in an anticyclonic system and part of it begins to turn south before looping back and turning north and then east once again by 19 June. On 20 June the easterly edge of the aerosol cloud has crossed Hudson Bay and, in both the measurements and the simulation, the BBA aerosol cloud is also starting to extend westwards towards the Pacific Iceland by the 23 June. By 24 June, this more northerly part has now turned south and moved over the Atlantic towards the West of Ireland. The southerly part of the aerosol cloud has continued to be entrained by the cyclonic system and has itself moved north to Greenland. Missing entirely from the AI is the large structure that extends out to the Pacific coast. This feature is also not present in any CALIPSO profiles. This part of the simulated aerosol cloud originated almost entirely from material 345 released at the very top of the model source, between 16 km and 18 km, and this suggests therefore that the BBA cloud did not extend above 16 km. In terms of spatial extent and horizontal structure, many of the features of the simulated BBA cloud are reflected in the AI, particularly in the first three to four days. However, the simulated aerosol cloud looks to be more diffuse than is suggested by the AI observations. This may in part be due to the detection limits of the AI, which is unlikely to be sensitive to some of the very low column loadings suggested by the model simulations. For example, as is shown in section  6), showing that the BBA cloud can be present even where no AI index is observed. A quantitative comparison between the AI index and NAME simulations is difficult not only because the lower concentrations can be missed by the AI index, but also because the AI index is sensitive to other common aerosols (for example dust -see Met Office operational dust forecast plotted in fig. 6), and so will always detect significant amounts of 355 background aerosol. However, to give some quantification within the bounds of the simulated BBA cloud we have re-gridded the AI onto the same latitude / longitude grid as the NAME simulations and compared the number of grid boxes within the simulated aerosol cloud that also have a positive AI index. The results of this comparison are shown in table 4. For the first four days following emission over 90% of the grid boxes within the simulated BBA cloud also contain an AI observation. This then falls to as low as 60% over the next four days. While there have been efforts to determine the relationship between Aerosol Index and aerosol optical depth and hence column loading, such relationships are frequently spatially variable owing to the sensitivity of the Aerosol Index to both the aerosol absorption optical depth and the altitude of the aerosol cloud (e.g. Christopher et al., 2008). BBA in the stratosphere will likely be associated with a stronger AI than a similar column loading of BBA in the troposphere. A more quantitative assessment of the aerosol column loading utilising the AI is beyond the scope of this paper. June there is good agreement between the two in terms of structure and position. The panel (c) shows the NAME BBA concentration for the grid box over Washington DC for the same time. Although it is initially partly obscured by clouds, a well defined aerosol layer arrives over the site at around midday on the 22nd. Beginning as a single layer at an altitude of 5 km, the layer then rises slightly to 8 km, before descending to around 4 km and splitting into upper and lower layers, separated by around 500 m, by mid-morning on the 23

Biomass burning aerosol: transport
June. The layers then persist until midnight on the 23 June. The mean PDR with these layers (at 532 nm) is 8±2%, consistent 390 with the layer being comprised of BBA (Groß et al., 2015a;Adam et al., 2020). Using a specific extinction value of 5.6 m 2 g −1 (calculated using Mie scattering theory and assuming a refractive index of 1.5 -0.07i at 532 nm, a density of 1.9 g cm −3 , and a size distribution taken from (Petzold et al., 2007) to transform the extinction coefficient product to a mass concentration estimate, the mean backscatter weighted concentration within the layers is 15±5 µgm −3 . Figure 9. NAME air concentration data [µgm −3 ] along the CARIBIC tracks shown in fig. 6. Overlaid are the CARIBIC flight paths coloured to show the particle count [number per cm3] recorded by the Grimm OPC. The NAME data is a mean of the one hour output files corresponding to the time the aircraft was transiting the aerosol cloud (times shown in panels).
So far we have presented the observation data and NAME simulations together, charting the dispersion of the volcanic and 415 BBA clouds over the first five to six days. The observations indicate that the NAME simulations reflect the general pattern of the dispersion seen in the various observations, and as such can be used to gain some insight into the dispersion of the aerosol clouds in areas where there are no observations. The availability of observations as the aerosol clouds crossed the Atlantic towards the UK is sparse, as the concentrations were mostly below the detection limits of passive satellite observation, and so we now rely upon only the simulations as the aerosol clouds move towards the UK and mainland Europe. Figure 10 shows the 420 column loadings for all three simulated aerosols on the 27 June, the 1 July, and the 11 July. The Met Office operational dust forecast is also plotted with the BBA column loadings. The NAME simulations show the BBA cloud reaching the UK on the 27 of June, along with an aerosol cloud of Saharan dust. The Saharan dust moves away from the UK over the next four days, however the BBA cloud persists over the UK and is joined on the 1 July by the ash aerosol cloud and the first parts of the sulfate aerosol cloud.  Both simulated aerosol clouds continue to impact the UK over the next week to ten days, while also dispersing over much of the northern hemisphere. By 11 July the more substantial parts of the simulated sulfate cloud also reached the UK.
4.6 Regional analysis and Interpretation of material over the UK Figure 11 shows examples of the data from the network lidars recorded on the dates displayed in fig. 10. The upper panels in each pair of axes show the log10 of the range corrected signal, and the lower panels show the log10 of the volume depolarisation 430 ratio. Table 5 gives a summary of the aerosol layers measured across the network. To be consistent with the simulations the heights quoted are above sea level. As described below we have used long averaging windows to calculate backscatter and PDR. However, even using this large averaging time it was not possible to measure the lidar ratio (which involves dividing a noisy signal by another noisy signal) as the results were so noisy as to be unusable.
The network first detected stratospheric layers over the Glasgow site on the evening of 27 June, with an upper layer at 14 435 km, and a lower layer at 11 km (axes A in fig. 11). There is also a much thicker layer in the troposphere between 2 km and 6 km after around 04:00 on the 28th. In this lower layer the mean PDR (at 355nm) was 26±3%. This is consistent with the layer being comprised of Saharan dust (Groß et al., 2013;Osborne et al., 2019), and the Met Office operational dust forecast confirms that this is the source of this lower altitude layer, and that it is unrelated to either the forest fires or the Raikoke eruption. The measurements in the stratospheric layers are near the limit of the useful range of the lidars (≈ 15km) and the 440 layers were only detectable at night when the daylight background was absent. To calculate PDR values and mass estimates at this range we have used whole night averages (6 hours of continuous photon counting) to improve the signal to noise ratio as far as possible. We note that this assumes that the aerosol particle type and properties remain constant during that time. Using the whole night average, using the Raman and elastic channels, we calculate that the PDR in the upper and lower layers was 9±1% and 7±1% respectively. There are a very limited number of reported measurements of the PDR of stratospheric BBA 445 at 355 nm.  report values of up to 26%, and (Haarig et al., 2018) report values around 20%. These values are considerably higher than our measurement. It is possible that the variation is due to differences in the production and ageing process of the aerosols in this layer compared to those measured in the previous studies. Tropospheric measurements of PDR of BBA are much more numerous, and are often reported to be below 5% (e.g. Groß et al., 2015b;Illingworth et al., 2015;Adam et al., 2020), but the range in reported values is quite large, and (Burton et al., 2015) and  report 450 tropospheric values of 17% and more. There is discussion in the literature as to the cause of the variation in the tropospheric measurements. Possible explanations being that some smoke layers contain irregularly shaped soil and dust particles lofted at the same time by the strong convection associated with wildfires, or alternatively that the aggregation of small soot particles as part of the ageing process can produce significantly non-spherical agglomerations (e.g. Jahl et al., 2021). It is possible that a similar variation exists in the stratosphere. The concentration in the layers was small, with the backscatter weighted mean 455 concentration in the upper layer being 14±8 µgm −3 . Rain and low cloud prevented any further measurements in the stratosphere until the evening of 1 July, when several UK network lidars detected aerosol layers between 11.5 km and 14 km. Axes B shows the data from Loftus for the 1 July and into the 2 July. Again using whole night averages, the mean PDR in these layers was 28±4% on 1 July into 2 July, and 32±7% on 2 July into 3 July ( fig. 11(c)). This is a marked change from the PDR measured on the 27th/28th, and while it is not possible 460 to unambiguously identify the aerosols using the PDR values alone, taken together with the arrival times of simulated BBA and ash clouds, we interpret this change to significantly higher PDR values as the arrival of the ash cloud, and the transition of the dominant aerosol type from BBA to volcanic ash. The maximum concentration was observed at Loftus and was 135±78 µgm −3 .
Low cloud and rain again prevented usable lidar measurements until 11 July when the lidars at Camborne and Glasgow 465 detected an aerosol layer at 9.5 km, and an optically very thin layer (AOD ≈ 0.005) at 11 km (axis pair D in fig. 11). The PDR in the lower layer was 0.9±0.2%, and 13±5% in the higher layer. On 13 July the lidars detected a layer at 13 km with a mean PDR of 1.6±0.4%. Again, this is a marked change from the previous PDR measurements on 4 July, and when viewed together with the arrival times suggested by the NAME simulations, we interpret this as the arrival of the sulfate cloud, and that by 12 July the scattering in the stratospheric layers was dominated by spherical sulfate droplets.

Conclusions
The early arrival of aerosols layers over the UK in late June 2019 has been established to be due to the emissions of biomass burning aerosols from intense fires in Alberta that penetrated the tropopause on 17 June 2019. Volcanic ash and sulfur species able to make measurements of the particle linear depolarisation ratios and estimate the concentrations within these layers over the UK.
We have presented results from NAME model simulations alongside observations charting the evolution of the aerosol clouds produced by the Raikoke eruption and the Alberta forest fires. The results of the simulations were used to aid in the interpretation of the UK lidar data. A mostly qualitative comparison between the observations and the simulation over the first 480 five to six days after emission showed that the simulations were representative of the evolution of the aerosol clouds as resolved by the observation, and as such could give some insight into the dispersion of the aerosol clouds where no observations were available. This was particularly true of the simulations of the forest fire aerosol cloud. After around five to six days following emission, the aerosol clouds were no longer visible in the satellite observations. However, the dispersion simulations showed that the BBA arrived over the UK by 27 June, the ash by 1 July and the main bulk of the sulfate aerosol cloud by 11 July and 485 this was confirmed by the UK lidar network.
By scaling the total mass of BBA released in the simulations so that the air concentration in the distal aerosol cloud over Washington DC matches the concentration estimated from the Washington DC MPLNET lidar observations, we estimate that the Alberta fires released around 0.2 Tg of fine mode BBA. The mass of volcanic ash released in the simulations (estimated using the Mastin relationship with plume top height) was not scaled to match observations. The simulated ash and sulfate 490 concentrations above Fairbanks, Alaska (400 µgm −3 and 25 µgm −3 respectively) were on the same order of magnitude as concentrations estimated from the Fairbanks MPLNET lidar observations (250 µgm −3 and 10 µgm −3 for ash and sulfate respectively), suggesting that the mass of ash and SO 2 released in the simulations was reasonably representative of the real value .
The PDRs measured by the UK lidars were 1% in the sulfate-dominated aerosols, 9% in the BBA aerosol cloud, and 30% 495 in the volcanic ash. These values did not allow an unambiguous identification of the material within the aerosol clouds, and it was not possible to measure lidar ratios within the layers over the UK. In this respect, NAME simulations were critical in interpreting the observations; they explained the change in the measured PDRs, from around 9% in the BBA cloud, over 30% in the ash-containing aerosol cloud, and around 1% in the sulfate dominated aerosol cloud. Importantly in terms of scientific advice given to the London VAAC the dispersion simulations showed that the layers detected on the 27/28 June 2019 were in