Identification of Smoke and Sulfuric Acid Aerosol in SAGE III/ISS Extinction Spectra Following the 2019 Raikoke Eruption

The 2019 eruption of Raikoke was the largest volcanic eruption since 2011 and it was coincident with 2 major wildfires in the northern hemisphere. The impact of these events was manifest in the SAGE III/ISS extinction coefficient measurements. As the volcanic aerosol layers moved southward, a secondary peak emerged at an altitude higher than that which is expected for sulfuric acid aerosol. It was hypothesized that this secondary plume may contain a non-negligible amount of smoke contribution. We developed a technique to classify the composition of enhanced aerosol layers as either smoke or 5 sulfuric acid aerosol. This method takes advantage of the different spectral properties of smoke and sulfuric acid aerosol, which is manifest in distinctly different spectral slopes in the SAGE III/ISS data. Herein we demonstrate the utility of this method using 4 case-study events (2018 Ambae eruption, 2019 Ulawun eruption, 2017 Canadian pyroCb, and 2020 Australian pyroCb) and provide corroborative data from the CALIOP instrument before applying it to the Raikoke plumes. We determined that, in the time period following the Raikoke eruption, smoke and sulfuric acid aerosol were present throughout the atmosphere and 10 the 2 aerosol types were preferentially partitioned to higher (smoke) and lower (sulfuric acid) altitudes. Herein, we present an evaluation of the performance of this classification scheme within the context of the aforementioned case-study events followed by a brief discussion of this method’s applicability to other events as well as its limitations.


CALIOP
The Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) instrument is a space-borne elastic backscatter lidar that has been orbiting the earth in the A-train constellation since 2006 (Winker et al., 2010). In September 2018 the orbit was lowered by 16.5 km to correspond to the orbit of CloudSat which. The onboard Nd:YAG laser emits polarized radiation at 1064 nm 130 and 532 nm. The total backscatter at 1064 nm and both parallel and perpendicular backscatter at 532 nm provides information on the size and shape of the scattering particles. We used data from the version 4.2 product, which has improved calibration particularly suitable for stratospheric studies (Kar et al., 2018;Getzewich et al., 2018;Kim et al., 2018). We also used the level 3 stratospheric aerosol product, which provides aerosol extinction and attenuated scattering ratios in the stratosphere at 5 • (latitude), 20 • (longitude), and 900 m (vertical) resolution (Kar et al., 2019). 135 3 The Raikoke plumes The 22-June, 2019 eruption of Raikoke was rated a VEI-4 that injected SO 2 and ash directly into the stratosphere (between 13 and 19 km). The SAGE instrument observed enhanced extinction layers within one week of the eruption (Fig. 1). The immediate increase in extinction was ≈8-9 times the background conditions, and the stratosphere remained in a non-background state throughout the remainder of 2019 and into 2020. Figure 2 shows the monthly zonal mean extinction coefficient at 1550 nm 140 (k 1550 ) and extinction ratio between the 520 nm and 1550 nm channels (k 520 : k 1550 ) from SAGE as well as the attenuated scattering ratio from CALIOP. The progression of enhanced extinction is seen in panels a-f of Fig. 2. Beginning in July, the extinction coefficient increased between 11 and 13 km and is attributed to Raikoke. No significant enhancement was observed in June because these figures present monthly zonal means and the eruption occurred late in the month, effectively averaging out any enhancement that was detected in the SAGE data. Subsequent months showed significant enhancement as well as how 145 this enhanced layer was transported southward, which is better seen in the extinction ratio plots (panels g-l) and attenuated scattering ratio from CALIOP (panels m-r).
What stood out in Fig. 2,panel (d), was the presence of an enhanced layer at ≈23 km. The initial ascent of this "secondary plume" might be seen as early as August (c) and remained visible in the extinction coefficient plots for the remainder of the year. The extinction ratio plots (panels g-r) as well as the attenuated scattering ratio plots (panels m-r) more readily show 150 the persistence of this layer through November. Historically, extinction ratios have used the 1020 nm channel as reference.
However, here we used the 1550 nm channel as reference because this enhanced the contrast between the enhanced layers and background, resulting in a more prominent contrast. It is because of this heightened contrast that the 1550 nm channel was used for reference throughout the remainder of this analysis.
wildfires (Boers et al., 2010;de Laat et al., 2012;Yu et al., 2019). Per this hypothesis, as the Sun shone through this portion of atmosphere, particles within the secondary plume absorbed the incoming solar radiation, which resulted in the air mass heating, thereby decreasing its density, which resulted in further lofting until equilibrium was reached. Though this scenario is not unreasonable for an absorbing aerosol, this is unexpected behavior for weakly-absorbing species like sulfuric acid aerosol and is generally not observed in association with the mid and high-latitude eruptions of the past. Below 2 µm, the imaginary 160 component of sulfuric acid's refractive index is effectively zero (i.e., ≪1E-5; Palmer and Williams (1975)), precluding the level of absorption required for subsequent heating/lofting. However, this behavior would be consistent with absorbing particles typically found in smoke from biomass burning events that occasionally inject black and brown carbon directly into the stratosphere during pyroCb events. Therefore, we hypothesized that smoke was present in the stratosphere during the Raikoke eruption and that this smoke layer lofted up to ≈25 km as it circulated the globe and migrated southward as demonstrated in  4 Evaluation of smoke and sulfuric acid extinction spectra from Mie theory In order for the ascending air mass to be diabatically heated there must be an absorbing species present and we hypothesized that this absorbing species is black and brown carbon found in smoke from coincident wildfires in Siberia and western Canada.
While the composition and spectral characteristics of smoke are highly variable (Bergstrom et al., 2002;Müller et al., 2005;170 Park et al., 2018;Kozlov et al., 2014;Womack et al., 2021), there is commonality between burning events in that the real component of the refractive index is spectrally flat and the imaginary component is variable (both behaviors being significantly different from sulfuric acid aerosol). Therefore, it is reasonable that the extinction spectra (extinction coefficients or extinction ratios as a function of wavelength) for smoke and sulfuric acid aerosols would differ significantly and that this difference may be useful in distinguishing between the two aerosol types.

175
As an initial test of this hypothesis we used Mie theory to calculate extinction coefficients at SAGE wavelengths for sulfuric acid aerosol and smoke. The primary challenge in carrying out this simulation is the highly-variable nature of smoke's refractive index, which is dependent on fuel source, burn temperature, humidity, age, etc. Further, smoke in the stratosphere is aged and its composition has likely changed during its transport due to ongoing chemistry (Yu et al., 2019). Therefore, the likelihood of this smoke's refractive index being consistent with the refractive index measured from aircraft or laboratory settings is small.

180
To our knowledge there have been no refractive index measurements for stratospheric smoke. Therefore, we used two sets of smoke refractive indices to span the range of reasonable refractive index values. The Bergstrom et al. (2002)  The simulation consisted of two parts: 1. assume a lognormal distribution with constant mode radius (200 nm) and distribution width (1.5) to visualize the expected extinction spectrum; 2. assume a lognormal distribution with constant distribution width (1.5) and variable mode radius (40-500 nm) to visualize how the slopes changed as a function of particle size. The results of this simulation are presented in Fig. 3 wherein it is observed that the two species have different spectral behavior for 190 extinction coefficients (panel a) and that the slopes were consistently different for small particle sizes. We emphasize that this model is very simple, contains multiple assumptions, and presents a general relationship that is in no way intended to be representative of actual conditions. Certainly, changing the smoke refractive indices from BC to BrC values significantly changed the extinction spectrum as well as the spectral slopes with the BrC slopes remaining significantly different from sulfuric acid (≈2x smaller). However, the model remains robust as a general guide for providing a testable hypothesis. While the details 195 of the size distribution, refractive indices, and number densities can modulate the differences in the slope, this has no impact on the subsequent analysis. Finally, we note that, as shown in panel (b) of Fig. 3, when particle sizes become large the slopes become less distinguishable because of the convergence of extinction coefficients at large particle sizes (as demonstrated by Thomason (1992) and Thomason et al. (2008) using extinction ratios). The consequence of this is that the current method is not applicable to large eruptions, such as the 1991 eruption of Mt. Pinatubo, which result in the formation of large sulfuric acid 200 particles. The intent of this simulation is solely to demonstrate that smoke will have flatter spectra than sulfuric acid aerosol that is the product of small to moderate volcanic eruptions. What stood out most in this simulation was the stark contrast between the sulfuric acid and smoke aerosol types; i.e., the difference in how rapidly the extinction coefficients changed with wavelength. Indeed, the sulfuric acid values changed more rapidly than those for smoke, indicating that, when sulfuric acid aerosol is the predominant aerosol type, the overall slope of the 205 extinction spectrum will be much larger (i.e., more negative) than when the atmosphere is laden with smoke. This distinction provides a testable hypothesis to determine, preliminarily, the viability of separating smoke and sulfuric acid aerosol in realworld data. To this end, data collected during the four case-study events listed in Table 2 were used to see, broadly speaking, whether the different events showed consistent spectral differences. Data were selected for each event by truncating the data record to include profiles collected within ±5 • latitude of the event, and included data collected one month prior to, and three 210 months after, the event (a four-month window) from 14-25 km. The extinction ratios (a proxy for spectral slope) for these four events are presented as a function of k 1020 in Fig. 4.
Similar to the theoretical work (Fig. 3), the two volcanic events in the SAGE data (Fig. 4, panels a and b) showed very different behavior from the wildfire events (panels c and d). On one hand, as k 1020 increased for the volcanic events the extinction ratio increased slightly, though it remained mostly unchanged, suggesting that both the composition and mean size 215 of the optically important aerosol has remained unchanged from background (following Thomason et al. (2021)). On the other hand, the extinction ratios for the wildfire events had distinctly different behavior, quickly merging to smaller values (<10) as the extinction coefficient increased. This figure demonstrates that the measured extinction ratios behave as expected from the model and that, at least preliminarily, the two event types can be distinguished.
At this point we must reiterate the caveat that this holds true only for small or moderate eruptions and would not be applicable 220 to larger eruptions such as the 1991 eruption of Mt. Pinatubo. Eruptions that inject large amounts of SO 2 into the stratosphere, like Pinatubo, have been observed to rapidly produce extinction ratios indistinguishable from water clouds and presumably smoke. This process involves the conversion of SO 2 to gaseous sulfuric acid (e-folding time of ≈30 days) which then either deposits on to existing aerosol or nucleates to form many small particles that coagulate to form optically large aerosol. Indeed, the first SAGE II observations of the main Pinatubo plume (when transmission was not saturated) showed a 525 to 1020 nm 225 extinction coefficient ratio of essentially 1 (Thomason, 1992). However, within the framework of the current study we only consider relatively smaller eruptions that inject much less SO 2 into the stratosphere, with Raikoke (VEI-4) being the largest.
Up to now we have only considered the raw extinction spectrum (e.g., Fig. 3 (a)) and a simple combination of extinction coefficients expressed as the extinction ratio (Fig. 4). This was useful for comparing measurements to theory and for providing rudimentary visualizations, though it requires the analysis to be done on a channel-by-channel or extinction ratio-by-extinction 230 ratio basis (hence the three colors in Fig. 4). Indeed, using a single extinction ratio (e.g., k 520 : k 1020 ) yielded results that were similar to the spectral-slope approach. However, all of the information in these four ratios can be efficiently combined into a single number within the spectral slope, thereby eliminating the channel-by-channel approach, streamlining the analysis, mitigating the potential for noise in a single channel to influence the outcome, as well as mitigating the impact of the low bias in the k 520 channel. Therefore, given the consistent behavior between the model and the measured extinction ratios we 235 hypothesize that small-to-moderate volcanic eruptions that inject material into the stratosphere can be distinguished from wildfire events in the SAGE record by looking at the spectral slope.

Detection and classification method
To test the aforementioned hypothesis, we evaluated the change in spectral slope as a function of k 1020 for 4 case-study events (2 pyroCb, 2 volcanic; see §6 for details). To do this, the spectral slope was calculated via linear regression where channel 240 wavelength (nm) acted as the independent variable and log 10 (k) was the dependent variable. The 385 channel was excluded from this analysis because of its rapid attenuation at relatively high altitudes (≈18 km). The 600 and 675 nm channels were excluded from the linear regression due to the impact ozone has on these aerosol channels. Further, to reduce the influence of potentially spurious measurements, a conservative cutoff was applied by excluding all extinction coefficients that had relative error >20% and we only used extinction spectra that had valid values in the 6 remaining channels (450,520,755,870,1020,245 2. Sulfuric acid aerosol: When extinction was enhanced, and the slope was less than (i.e., more negative) or equal to the background slope.
3. Smoke: When extinction was enhanced and the slope was flatter than background conditions.
A shortcoming of this classification scheme is that it uses hard cutoff values to separate the aerosol types while, in reality, particles near the smoke/sulfuric acid cutoff would likely be a mixture of the two and not strictly homogeneous. However, the utility of this method, as described, makes the identification of smoke highly conservative.

Layer identification with CALIOP
Ideally, this characterization scheme would be validated with in situ sampling of the various and disparate aerosol layers, which requires expansive sampling on a global (or at least a hemispherical) scale that is not feasible. However, the CALIOP lidar has polarization sensitivity at 532 nm which can be used to make general composition estimates (e.g., sulfuric acid aerosol, smoke, dust, cloud, and volcanic ash). Smoke injected into the stratosphere due to pyroCb events can be discriminated from sulfuric (2020)). This feature can be used to separate stratospheric smoke from the volcanic sulfate particles which are spherical and 285 thus do not depolarize. In addition to depolarization ratio, the CALIOP data products contain a vertical feature mask (VFM) product that classifies the different types of detected layers as aerosol (tropospheric and stratospheric) and clouds . Both depolarization ratio and the VFM were used herein to corroborate the identification of sulfuric acid aerosol and smoke within the SAGE data.

Application to case studies events 290
In this study we considered five events that had significant impact on the stratosphere as detailed in Table 2. Excluding Raikoke, these events were classified as either primarily volcanic or wildfire related, which provides four test cases for evaluating distinct behaviors for each event class. While the majority of the data collected for these events appears to come from a single source, we add the caveat that some events were close enough in time and geography to experience some carryover (e.g., the two Ulawun eruptions and the Australian pyroCb), which will be briefly discussed below.

295
To better appreciate the finer details of the profile data, and to demonstrate which parts of the atmosphere were most impacted by each event, the data were broken into 1 km bins. Statistics for labeling the different layer types in the 4 case-study events are  Table 3, which contains the total number of valid spectra collected at each altitude, the number of non-background spectra identified using the above cutoff criteria, and the fraction of enhanced spectra identified as either smoke or sulfuric acid aerosol.
Altitude ( Table 3. Layer classification statistics from SAGE data. Total number of valid spectra, total number of identified layers as well as the fraction of spectra identified as smoke or sulfuric acid aerosol for each case-study event. background conditions, which makes sense since the lowermost altitudes are the most impacted by these events. The Canadian wildfire likewise showed a nearly uniform identification of smoke. While the volcanic events showed nearly uniform identification of sulfuric acid aerosol under elevated conditions, the wildfire events showed a significant portion of the spectra identified as sulfuric acid aerosol (≈10,000 out of >58,000; ≈19%). Of 335 these ≈10,000 sulfuric acid classifications, ≈48% of them were in the Canadian wildfire event between 23 and 25 km. The reason for the presence of elevated sulfuric acid aerosol within the wildfire events could be for two reasons. First, within the current identification scheme, it is possible for smoke to be identified as sulfuric acid because of the combination of the SAGE viewing geometry and the optical thinness of parts of the smoke plume (i.e., depending on whether SAGE is sampling through the centroid of the plume or only the outer edge). It is reasonable that, when sampling optically thin smoke layers, the extinction 340 will be elevated above background levels, but the slope may not deviate significantly. This can be achieved when the viewing geometry is such that only optically thin portions of the smoke plume are sampled, thereby raising the extinction coefficients, but the overall, integrated aerosol along the viewing path is not sufficiently different from background conditions to significantly change the spectral slope. This could lead to an ambiguous characterization of aerosol composition at extinctions that are outside background values, but still at the lower end of extinction values for that particular event as seen in Figs. 9 & 10. This 345 scenario is the most likely and is expected from a statistical viewpoint. Secondly, a less likely scenario is there may be elevated levels of sulfuric acid aerosol within the sampling volume due to transport from a nearby volcanic event. Indeed, this may be the case for some of the spectra classified as sulfuric acid in the Australian wildfire case study. In contrast to the Canadian  wildfire, the range of background extinction coefficients for the Australian fire spanned a wider range and extended into higher extinction coefficients (e.g., >5E-4 at 16 km), indicating the apparent background conditions were perturbed, potentially from 350 the 2019 Ulawun eruptions (Kloss et al., 2021). Regardless of why spectra were classified as sulfuric acid, the performance of this identification scheme remains encouraging as the majority (>81%) of non-background values were identified as smoke within these layers as shown in Table 3. Further, the distribution of data for the wildfire events is markedly different from the volcanic events, indicating that we are observing two distinctly different aerosol types. Figures 11 and 12 show examples of the CALIOP and SAGE profile data collected over the two wildfire case-study events.

355
Here, CALIOP showed significant depolarization near 19 km (Canadian pyroCb) and 14 km (Australian pyroCb), which corresponded well with a rapid increase in both aerosol extinction and spectral slope in the SAGE profile data. We note that in Fig. 12 SAGE saw another layer at 19 km that was not manifest within the CALIOP VFM or depolarization ratio profiles.
This altitude is well within the SAGE instrument's operational altitude range and may be reflective of the relatively poor return signal at this altitude for CALIOP and its narrow swath width. Alternatively, SAGE may have sampled a narrow smoke filament 360 that was not within the CALIOP sample volume.
Overall, the CALIOP data products provided good support for the SAGE-based classification of stratospheric aerosol composition. Therefore, this identification scheme will now be applied to a much more heterogeneous event: the 2019 Raikoke eruption.      classified the Siberian wildfire near Lake Bolon as a pyroCb. Similarly, we applied our method to identify an unambiguous smoke signal in the stratosphere prior to, or immediately coincident with, the Raikoke eruption. To this end, we evaluated SAGE and CALIOP data collected to the west of Raikoke (i.e., upwind, see Fig. 13). Unfortunately, SAGE did not begin sampling this latitude band until after the eruption, therefore we limited our analysis to the first two weeks after the eruption and to a region far enough west of Raikoke to not have been impacted by the volcanic ejecta.

375
Both SAGE and CALIOP collected profiles within this region and indicated the presence of smoke at ≈13 km as seen in Fig. 14. These profiles were collected within 8 days of the eruption, while the plume still resided over the North Pacific ocean (Kloss et al., 2021;Vaughan et al., 2021). This enhancement was most noticeable in the CALIOP data as well as the spectral slope profile. Indeed, the spectral slope shows a sharp gradient at this altitude, before returning to background conditions between 15 and 20 km (compare to background conditions during the 2017 Canadian pyroCb event, Fig. 11), while the CALIOP 380 depolarization ratio and VFM showed a distinct layer between 10 and 15 km. Therefore, while the SAGE sampling schedule did not allow us to evaluate profiles collected prior to this time, this smoke layer was persistent over this region, in both the SAGE and CALIOP records, throughout the first ≈2 weeks after the eruption and we concur with previous authors that smoke was present in the stratosphere during this time period and was visible in both the SAGE and CALIOP profiles. Figure 13. Field of regard for sampling air that has not been impacted by the Raikoke event, but may have been impacted by coincident NH wildfires. LB and R indicate location of the Lake Bolon, Siberia fire and Raikoke, respectively.

385
As stated in the introduction, Raikoke erupted on 22-June 2019 and injected SO 2 and ash directly into the stratosphere, at around 15 km altitude, and was observed by SAGE approximately one week later (Fig. 1). Immediately after the eruption, the main Raikoke plume broke into two distinct plumes. One plume moved southward and appeared to be primarily ash as determined by Kloss et al. (2021) and Vaughan et al. (2021). The ash in this plume settled out within a week of the eruption (Kloss et al., 2021). The second plume moved to the north and east and was composed primarily of SO 2 (Kloss et al., 2021),

390
(which was in the process of being converted into sulfuric acid) before getting temporarily trapped within the Aleutian low Based on the results presented in the previous section, we anticipated that the spectral slopes of the SAGE data collected over the Raikoke event would behave similar to those for Ambae and Ulawun. However, as shown in Fig. 15 and Table 3 the Raikoke data presented what appears to be a mixture of sulfuric acid aerosol and smoke, with the predominant composition being smoke (only 10-30% of spectra were identified as sulfuric acid aerosol). Indeed, the majority of lower-altitude spectra 400 were identified as smoke, while the balance seemed to shift at the highest altitude (23 km). While we anticipated observing smoke within the profiles we did not expect the majority of the spectra to be identified as such. Overall, the Raikoke data look more like a wildfire event than the other volcanic events in this study. Given the magnitude of this eruption, the spectra identified as smoke here may be the product of both ash and large particle formation, both of which have short lifetimes in the stratosphere.

Interpretation of the secondary Raikoke plume
As shown in Fig. 2, a secondary layer of elevated aerosol broke off from the main Raikoke plume as it moved southward and continued to loft to higher altitudes. The composition of this secondary plume was speculated to contain smoke from NH wildfires, which would cause it to absorb incoming solar radiation, warm the surrounding air, and diabatically loft. This layer