Stratospheric pollution from Canadian forest fires

Abstract. Forest fires in British Columbia in August 2017 caused a pyrocumulonimbus event that injected a polluted airmass into the lower stratosphere. The Microwave Limb Sounder (MLS) on the Aura satellite first observed the polluted airmass on 14 August 2017 and continued to observe it for 60 days (100 days in water vapour). We estimate the mass of CO injected into the stratosphere to be 2 – 3 Tg. Events such as this are rare: this is the third of four such events in the 16 years since the launch of Aura, the second-largest of the four events, and the only one in the Northern Hemisphere. Unlike the preceding two events, 5 but like the most recent event, the polluted airmass described here had an unusually high water vapour content.


Data from MLS
The MLS instrument (Waters et al., 2006) was launched in July 2004 on NASA's Aura satellite and has operated almost continuously since then. Aura is in a polar orbit with a 98 • inclination angle. The MLS instrument views in the direction of travel; as a result the measurement locations lie close to the orbit track and cover the latitude range 82 • S to 82 • N. Once processed, the data consist of 3495 vertical profiles per day, spaced at intervals of 167 km along the orbit track. Estimated 25 quantities include temperature, geopotential height, and the mixing ratios of 16 chemical species, including H 2 O (Lambert et al., 2007), CO (Livesey et al., 2008;Pumphrey et al., 2007) and HCN (Pumphrey et al., 2018). Data are reported on pressure levels; for most products these are spaced at 6 levels per pressure decade, giving a representation with a vertical resolution of about 2.7 km. For most products the true vertical resolution is similar to this spacing, but for species with low mixing ratios and/or weak spectral lines it may be considerably poorer than this. For a few species (H 2 O, O 3 ) the data are reported at a 30 resolution of 12 levels per pressure decade. Full details on use of the data may be found in Livesey et al. (2018).

Carbon Monoxide
MLS CO data are recommended for use at pressure levels from 215 hPa (∼ 11 km) up to 0.0046 hPa (∼80 km). Data are reported at the 316 hPa level (∼ 8 km), but these values are thought to mostly represent the real atmosphere at 215 hPa and 35 are not recommended for any quantitative work. They can, to some extent, be a useful qualitative indicator of excess CO somewhere in the 8-10 km altitude range; we show the 316 hPa data in some of the figures in this paper, but do not use them to derive any quantitative results.
Enhanced values of CO from the Pacific Northwest Event were first observed on 14 August 2017. Figure 1 shows MLS CO over North America at this time, at 215 hPa. The maximum values seen here are about 1.2 ppmv; this is about 24 times the 40 typical zonal mean value of 50 ppbv and about 6 to 8 times the typical daily maximum value. Figure 2 shows a time series of the maximum CO values observed at several pressure levels. It is clear that the polluted airmass reached 215 hPa almost immediately. From there, it ascended fairly rapidly to 100 hPa and then more slowly from 100 hPa to 46 hPa.
To summarise the whole event, we applied the strategy used by Pumphrey et al. (2011) to detect values that exceed the usual spread expected for a given latitude band. We found the mean value and standard deviation for all measurements in a given 45 latitude band and at a given pressure level. We then describe the CO at a point as "enhanced" if its mixing ratio was more than a chosen number, α, of standard deviations above the mean. The value chosen for α is somewhat arbitrary and was different for each level: α = 5.7 at 215 hPa, α = 4.6 at 147 and 100 hPa and α = 3.9 at all other pressure levels. These choices are made in order to show the interesting events while rejecting almost all of the background points. An iterative approach is used to ensure that the mean and standard deviation are calculated from the unaffected points; there are few enough affected points that this 50 calculation converges very quickly.   Figure 3 shows the locations where enhanced CO was detected. The plume is first observed over northwestern Canada, travelling southeastwards and then eastwards across northern Canada and then across the Atlantic to Europe. The behaviour of the plume while over Eurasia is more complex. After 30 days, one part of it reaches northwestern China, where it divides into two parts, one heading east across the Pacific at about 47 • N and the other returning westwards across the Atlantic at about 55 33 • N.
As the polluted airmass moves far more rapidly in the zonal than the meridional direction, it is helpful to plot it against time and longitude, as in Fig. 4. The ascent of the polluted airmass (shown by the changing colour and shape of the points) is   Longitude Time / days after 2017−08−12 q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q qq q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q  clearly visible in this figure. It can also be seen in Fig. 4 that the polluted airmass is divided into two parts on three occasions: at approximately 9 days, 16 days and 41 days after 12 August 2017. After the first division, one branch remains at 147 to 60 100 hPa, travels more rapidly than the other branch and is last seen on day 16 near 80 • E, 70 • N. The higher branch is at 100 to 68 hPa at day 9 and continues to ascend. It appears to divide into two parts somewhere around day 16; these two parts appear in Fig. 4 to cross over at around day 30, after which one part travels east and one west. The more northern part travels eastward from northern China across the Pacific Ocean, while the more southern part rises to 31.6 hPa and moves westward across northern Africa and the Atlantic Ocean. Around day 41, the eastward-travelling part appears to divide into further parts.

65
The faster-travelling part is last seen on day 60 at 90 • E; the slower part is last seen on day 63 at 60 • W. We can obtain an estimate of the total injected mass of CO by integrating the MLS data into partial column (215 hPa to 31 hPa) values and summing over all latitudes affected by the event. The seasonal behaviour of CO is fitted with a mean value and annual and semiannual sinusoids as shown in Fig. 5(a); this fit ignores the data for a period of 99 days after the PNE.
Subtracting this first fit from the data leaves a fairly featureless time series apart from a sudden increase after the fire and a 70 decay after that; we fit the decay with an exponential: The exponential curve is shown in Fig. 5(b). The injected mass, M 0 , is 2500 ± 150 Gg, with a decay time t d of 27 ± 2 days; the errors here are those given by the nlm() function in R (R Core Team, 2018). The CO data from the first three days after the event are not well fitted by the exponential used to fit the subsequent days, presumably because the plume has a small horizontal

Water vapour
The event described in this paper differs from the Black Saturday event (Pumphrey et al., 2011) in that a large amount of water 85 vapour is found in the polluted airmass. Figure 6 shows the longitudes of unusually high water vapour in the same manner as  After about 50 days the water vapour in the polluted airmass exceeds the background value to a greater extent than is the 95 case for CO. After 63 days, the polluted airmass can only be identified in the water vapour data; this is presumably due to the short chemical lifetime of CO in the stratosphere. A small parcel of enhanced H 2 O may be seen travelling westwards from 150 • E to 80 • W, between 65 and 90 days after the event, before ceasing to travel westwards, and fading into the background.

Other species
All chemical species observed by MLS were examined for the few months after the PNE using the same thresholding tech- for about 60 days (100 days for H 2 O).
-Species for which anomalous measurements are believed to be artifacts arising from spectral interference due to the presence of another molecule: HOCl, HO 2 , HCl, ClO (640 GHz).

110
A summary of the ranges of time and altitude over which each species has unusual values is shown in Table 1. Pumphrey et al. (2011) demonstrate that an airmass containing biomass-burning products may appear from the MLS level 2 data to contain enhanced amounts of the 640 GHz ClO product, but that this is actually caused by enhanced amounts of methanol (CH 3 OH), which has a similar spectral signature to ClO in the passband of the MLS 640 GHz radiometer.
As is the case for many MLS data products, CH 3 OH is produced from a dedicated retrieval "phase", in which a ClO product

115
(not the standard one) is also retrieved (see Livesey et al. (2018) for details). The ClO data from the methanol phase show far less correlation with CO than the standard 640 GHz ClO product. This also suggests that the enhanced values seen in the standard 640 GHz ClO product are an artefact caused by enhanced methanol. HOCl, HO 2 and CH 3 Cl are all detected via spectral lines that are very close to the CH 3 OH lines. It seems reasonable to suppose that HOCl and HO 2 appear to be enhanced in the MLS data due to spectral contamination from CH 3 OH, especially as they are both observed only for a very 120 short period. CH 3 Cl, on the other hand, is known to be a biomass-burning product (Santee et al. (2013) and references therein); its spectral signature is sufficiently different from that of ClO (or CH 3 OH) that the enhancement observed is almost certainly genuine. HCl, which shows the smallest enhancement in the plume, is observed at a different frequency from the ClO, CH 3 OH and CH 3 Cl lines, but one very close to the main CH 3 CN lines in the MLS 640 GHz radiometer. Although HCl is a minor fire product (mainly produced from the burning of garbage ), it seems likely that the enhanced values observed 125 are spurious and are caused by the large amounts of CH 3 CN in the plume.
The relationships between species in the plume were investigated further by calculating a linear fit between enhanced CO and a given species at the same retrieval level and location. Data from profiles immediately adjacent to the plume are included to provide an appropriate baseline. We show an example in Fig. 7 in which the other species is CH 3 CN. Plotting the data in this manner reveals that in some cases (O 3 , N 2 O) there is a correlation between the two species, even when the non-CO species 130 does not have values outside its usual range. For other species (SO 2 , HNO 3 , BrO), no such correlation is observed, or the correlation is small enough to suggest that it is an instrumental rather than a geophysical effect. We note values of the linear fit slope in Table 2 and show a selection of the data from Table 2 in Fig. 8. For some species (including the CH 3 CN shown in Fig. 7) we have used data outside the vertical range recommended by the MLS team. This was done where the correlation with CO was clear; we note that the linear fit slopes will not be affected by a constant bias in the mixing ratio of either CO or the 135 other species. For species produced by biomass burning and which have similar lifetimes to CO in the lower stratosphere, we would expect the ratios observed in Table 2 to be constant with altitude and to be in reasonable agreement with the emission ratios measured previously. The relationships with CO of CH 3 CN and CH 3 Cl show relatively little variation with height. The slopes we obtain are generally within the published range; for CH 3 Cl this range is rather large. The methanol/CO ratio is within the rather wide range of published values and decreases with height. This might appear to suggest that methanol undergoes chemical loss in the plume more rapidly than CO, but the MLS methanol data are potentially useful only at 147 hPa (and at low latitudes at 100 hPa), so any such conclusion must be regarded as speculative. The agreement between our value at 147 hPa and the published values is reasonable given the large variability in the latter.
The H 2 O/CO ratio varies considerably with height between 215 and 68 hPa, but is approximately constant between 68 and 31 hPa. The standard 640 GHz ClO data behave in a similar manner to the biomass-burning products. This is compatible with    Fig. 7. Triangles are the data from the "other measurements" column of Table 2, with their error bars (where available). All the triangles refer to measurements made in the troposphere; their position in the vertical has no meaning. q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q q GSL Fires  from an orbit at a matching time.
reasonably close to the Pacific Northwest Event; we show examples in Fig. 9. Although MLS observes enhanced CO at the 160 147 hPa level at this time, trajectories launched at this level pass far to the west of the PNE. This suggests that over these first few days the polluted airmass is closer to 215 hPa than to 147 hPa. There were other forest fires burning at the same time as the PNE, notably in the region around Lake Athabasca and the Great Slave Lake (GSL); the location of these fires is marked on Fig. 9. The trajectories make it reasonably clear that the source of the polluted airmass observed by MLS is the Pacific Northwest Event and not the fires near the Great Slave Lake.

165
By launching trajectories at many points spread between the MLS observations of high CO, both vertically and along the measurement track, we obtain sufficient times of closest pass to the PNE to construct a histogram, as shown in Fig. 10. The mean and median closest pass times are 01:20 UTC and 00:57 UTC respectively on 2018-08-13. The standard deviation of the closest-pass times is 12 hours. The closest pass times are in good agreement (given their own standard deviation) with the times reported by Peterson et al. (2018), who state that pyrocumulonimbus clouds (pyroCb) are observed to develop between 170 23:00 UTC on 12 August 2017 and 00:30 UTC on 13 August 2017, and that they remain active for 1-4 hours.

Plume trajectory modelling
The FLEXPART model (Stohl et al., 2005) was used to model the dispersion of the plume from the Pacific Northwest Event.
FLEXPART allows the user to release a mass of a pollutant at a chosen place and time and to observe how it disperses. We released the pollutant in a small cuboid region; the release is defined by the location and size of this region, by a start and 175 end time, and by the mass released. We show in Fig. 11 an example of FLEXPART output for a time approximately 66 hours after the PNE. An MLS orbit that passes through the polluted airmass at that time is also shown; points where the MLS CO is enhanced are marked and lie close to the modelled plume. We adjust the properties of the mass released to best match the MLS observations using the Markov Chain Monte Carlo (MCMC) inverse modelling technique (van Ravenzwaaij et al., 2018). As FLEXPART is not able to model the self-lofting of 180 the polluted airmass, we used only the first few days of data after the fire. Some experimentation demonstrated that the data do not provide sufficient information to determine the horizontal extent of the release region, nor its extent in time. We therefore constrained the region to be 0.5 • wide in longitude, 0.375 • wide in latitude and 1.5 hours in duration. (These choices were somewhat arbitrary and were intended to ensure that the box was smaller than the observed pyroCb clouds.) The estimated quantities were then the location and time of the release cuboid, its vertical extent, and the total mass of CO released. 185 We show the horizontal release locations found by the MCMC approach in Fig. 12. There were several large anvil clouds in the area at the time; these are clearly visible in AVHRR images and are outlined in green in Fig. 12. The MCMC solutions have a standard deviation of about 6 km perpendicular to the wind direction and 30 km in the wind direction. The injection altitude is determined rather precisely: 11.5 ± 0.1 km; as expected from the back trajectories, this is closer to 215 hPa than to 147 hPa. The injected mass is 620 ± 80 Gg, considerably smaller than the value of 2400 ± 300 Gg that we obtained by the 190 simple method discussed in section 3.1. This is qualitatively reasonable, given that the MCMC approach only considers CO that is observable by MLS near the start of the event, while the simple method of section 3.1 includes CO that was injected at too low an altitude for MLS to observe it and subsequently ascended to observable altitudes. release time is about 5:00 UTC on FLEXTRA trajectories, but is well within the 12 hour standard deviation of the FLEXTRA results. As noted above, Peterson hours. The release times we determined using FLEXPART and the MLS data are at the late end of this range.

Discussion
Significant injections of pyroconvective plumes into the stratosphere generally require both extremely intense fires and particular meteorological conditions, such as those preceding the passage of a cold front. The meteorology surrounding the Pacific 200 Northwest Event is described in some detail by Peterson et al. (2018); the fire indeed occurred while a cold front was approaching.
Most forest fires do not produce a pyroCb cloud. Moreover, most pyroCb clouds do not extend to a great enough altitude to loft the products high enough for MLS to observe them. The only events in the MLS record that are in any way comparable to the PNE are the Black Saturday fire, which occurred in Australia in February 2009 (Pumphrey et al., 2011), and the Australian 205 fires of December 2019 to January 2020 (paper in preparation, also Kablick et al. (2020), who refer to this as the Australian New Year (ANY) event). The mass of CO injected into the lower stratosphere by the Black Saturday fire was, when estimated by the technique of section 3.1, 1.3 Tg -just over half of the mass injected by the PNE. Like the PNE, the Black Saturday fire occurred at a time when a cold front was approaching (Dowdy et al., 2017). The 2019/20 ANY fire injected a mass of CO considerably larger even than the PNE: initial esimates, again using the method of section 3.1, place this mass in the range 210 8-10 Tg. MLS also observed enhanced CO from the Great Divide fires in Australia in December 2006. These fires burned for two months, but MLS only observed enhanced CO from them for a short time, beginning on 2006-12-12. The injected mass of CO from the Great Divide fires was between 1.5 and 1.8 Tg, similar to that from the PNE, but the polluted airmass did not ascend in the manner observed with the PNE, ANY and Black Saturday events, and was observed by MLS for only 10 days (Pumphrey et al., 2011).

215
The MLS data shown in this paper are for most species the first data reported from the PNE.  Khaykin et al. (2018) also report aerosol data from the CALIOP lidar instrument on the CALIPSO satellite. The maps of the CALIOP data are similar to Fig. 3, indicating that the aerosol and CO travelled together; presumably the aerosol particles were too small to settle out 220 on the timescale over which the CO was observed. Khaykin et al. (2018) suggest that the pollution is from fires near Lake Athabasca; we show clearly that it is from British Columbia and from the vicinity of the Pacific Northwest Event as described by Peterson et al. (2018). Ansmann et al. (2018) ascribe the source of the pollution to British Columbia in agreement with our conclusions. In addition to the MCMC results, Fig. 12 shows the areas burned in 2017 marked in purple (British Columbia Wildfire Service, 2018). Figure 12 also shows the AVHRR thermal image; three large anvils of high cloud are clearly visible.
pyroCb cloud at the southwest edge. It is not clear from the visible images that an active pyroCb is associated with Anvil 2; images taken during the previous two hours (Peterson et al., 2018) show that Anvil 2 formed near 00:00 UTC. Anvil 3 also formed around 00:00 UTC; Anvil 4 formed half an hour later and had merged with Anvil 3 by 01:45 UTC.
The possible solutions for the source of the polluted airmass that emerge from the MCMC inverse method are shown in 230 Fig. 12 as a cloud of small points. The colours of these dots indicate the time, varying from 2:00 GMT (red) to 5:00 GMT (blue). The location of these dots suggests that the polluted airmass observed by MLS is probably associated with Anvils 2, 3 and 4 and is less likely to be associated with Anvil 5. This in turn suggests that the emissions come from the plateau complex of fires (west of Nazko) and not from the Elephant Hill fires (north of Ashcroft).
The polluted airmasses from the PNE and ANY contained large amounts of water -this distinguishes them from the Black 235 Saturday event, in which the water content of the polluted airmass was similar to that of the surrounding air, except at 100 hPa and lower altitudes during the first few days of the event.
The rapid ascent of the polluted airmass from the PNE is due to its containing a large amount of black carbon aerosol (Yu et al., 2019;Kablick et al., 2020). The aerosol absorbs incident sunlight, heating the polluted airmass and causing it to rise.
The event can in some ways be considered a small-scale demonstration of what would happen in the aftermath of a nuclear 240 war: such a war would cause widespread fires that would put a great deal more black carbon into the middle atmosphere than the PNE did. Self-lofting of this carbon (as demonstrated by the PNE) would raise it to an altitude where it would remain for a number of years, giving rise to the so-called nuclear winter [Robock et al. (2007) and references therein]. Simulations (Reisner et al., 2018;Robock et al., 2007) of the response of the climate to a regional-scale nuclear war disagree on the severity of this effect. However, much of the disagreement is caused by widely different estimates of how much black carbon would reach the 245 stratosphere from a given nuclear explosion. Once it did reach the stratosphere, the black carbon aerosol from a nuclear war would ascend in the manner we observe in the PNE.

Conclusions
The Pacific Northwest Event was a forest fire accompanied by unusually strong convection, injecting both aerosol and gasphase pollutants into the stratosphere. Observations of gas-phase pollutants from the MLS instrument show that the polluted 250 airmass reached all longitudes on a timescale of 60 days and ascended to an altitude of 24 km. The polluted airmass was much wetter than the surrounding air in the stratosphere; in this respect it was different from the otherwise similar Black Saturday event and similar to the recent ANY event. The polluted airmass contained biomass-burning products HCN, CH 3 CN, CH 3 Cl and CH 3 OH in broadly similar proportions to CO as observed by other techniques in other biomass-burning events.
Events of the type we describe here are rare; the PNE, ANY, Black Saturday and Great Divide fires are the only such events 255 in the MLS dataset at the time of writing. Whether they will become rarer or more frequent as the climate changes over the coming decades remains to be seen.