The smoke was observed using lidar data from the NASA satellite-borne CALIOP (Cloud-Aerosol Lidar with Orthogonal Polarization) instrument. We used the latest version of the rawest product available (v4.10, level 1B). Backscattering coefficients were computed by correcting for light attenuation by air molecules (including ozone absorption) and for particles in the stratosphere. The molecular part was estimated using modeling data of the ozone and air densities provided in the CALIOP files compiled by the NASA Langley Research Center (Friberg et al., 2018). The wildfire smoke is optically dense and strongly attenuated the lidar signals. The particle light attenuation was computed from the lidar signals themselves in an iterative approach explained in Martinsson et al. (2022). This procedure retrieves also the extinction-to-backscattering ratio, the so-called lidar ratio, used to compute aerosol extinction coefficients from the CALIOP lidar backscattering data. CALIOP has a polarization filter, separating backscattered light into parallel and perpendicular polarization. The ratio of the two forms the volume depolarization ratios used for ice cloud screening of the entire dataset. The volume depolarization ratio describes the properties of the complete volume of air, i.e., the aerosol particles together with the air. To study the temporal evolution of the smoke particles, we also compiled the particle depolarization ratios for individual smoke layers (Martinsson et al., 2022), which describe the properties of the particles themselves.

Ice clouds were removed in the lowest 3 km of the stratosphere. CALIOP data were averaged to 8 km horizontal resolution, and volume depolarization ratios above 0.20 were classified as clouds. The process is described in more detail in Martinsson et al. (2022).

Further aerosol data were compiled from the limb-scatter-observing instrument OMPS-LP (Ozone Mapping and Profiler Suite Limb Profiler). We used level 2.0 version 5.10 light extinction data (Suomi-NPP OMPS LP L2 AER Daily Product, version 2.0, Taha et al., 2021) of two wavelengths (745 and 997 nm). The OMPS-LP wavelength (510 nm), most similar to the CALIOP wavelength (532 nm), and used in Martinsson et al. (2022), is unfortunately not reliable in the Southern Hemisphere (Taha et al., 2021). Data were filtered to minimize influence of ice-clouds and polar stratospheric clouds, and flags were used to prevent influence from erroneous data. A detailed description on this approach can be found in Martinsson et al. (2022).

Stratospheric AODs were computed by integration of the aerosol extinction coefficients from the tropopause to 35 km altitude, as well as in selected layers of the stratosphere, where the LMS was defined as the layer between the tropopause and the 380 K isentrope. Tropopause heights from the MERRA-2 (Modern-Era Retrospective analysis for Research and Applications) reanalysis were retrieved from the CALIOP and OMPS-LP files provided by NASA.

The UV aerosol index (UVAI) level 3 data (v2.1) from OMPS-NM (Ozone Mapping and Profiler Suite Nadir Mapper) were used to track horizontal smoke transport. The data product is compiled from observations at 379 and 340 nm (Torres, 2019). It indicates presence of UV-absorbing aerosol particles and increases with altitude, making it well suited for tracking BC-containing wildfire smoke in the UT and stratosphere. Data were screened based on NASA recommendations on data usage (https://ozoneaq.gsfc.nasa.gov/docs/NMTO3-L3_Product_Descriptions.pdf, last access: 6 October 2022). Horizontal UVAI distributions were combined with vertical information from CALIOP to identify smoke transport to the upper troposphere and stratosphere from different fire events.

Smoke from the different fire events was identified based on daily maps over the UVAI, stratospheric wind directions, and altitude distributions from CALIOP curtains. We tracked the motion day to day of the central parts of the fire events (p. 1 in the Supplement), and used the information to classify the individual smoke layers described above. OMPS-NM shows two separate major events of increased ultra-violet (UV) aerosol index from the 2019–2020 Australian wildfires, first observed on 29 December and 4 January as described in Peterson et al. (2021). Some days the two events overlapped horizontally. In such case additional information on altitude from CALIOP was used, because the smoke from the two main events on any given day differed markedly in altitude (Supplement). We will elaborate more on this in the Results section.

The depolarization ratios for smoke from the second fire were clearly lower than those for smoke from the first fire, as seen in Figs. 1 and 2, as well as in the Supplement (Figs. S2–S49. This difference remains for more than 1 month, i.e., smoke layers from the second fire continues to have lower depolarization ratios than smoke from the first fire. This particle optical property verifies that we have classified the smoke successfully.

CALIOP provides vertical distributions of aerosol and clouds. Besides the attenuated backscattering provided as curtain plots, information on particle morphology (depolarization ratio, where zero corresponds to spherical particles) and particle size (color ratio, i.e., ratio in attenuated backscattering of wavelength 1064 to 532 nm). Figures 1 and 2 show these three CALIOP features over the regions with increased UVAI from the first days after the first and second fire event, respectively. Non-cloud features can be identified by strong backscattering signal in connection with depolarization ratios less than approximately 0.2 and color ratios well below 1. For an example see the observation on 3 January 2020 at 10:07 UTC (Fig. 1n–p), where a thin smoke layer in the tropopause region resides over deep cloud layers. In optically thick smoke layers, there is a shift in color ratio from a low value at the top to significantly higher values lower down in the smoke layer. These increased values deep into the layers are artifacts caused by stronger attenuation for the shorter wavelength. The signal from the layers closest to the satellite (at the layer top) is less affected by attenuation, whereas deeper into the layer the shortwave signal (532 nm) becomes attenuated more than the longwave one (1064 nm) (Martinsson et al., 2022).

Satellite observations of water vapor from the Microwave Limb Sounder (MLS) aboard the Aura satellite was used together with aerosol data. We use the level 2 nighttime data version 5.0-1.0a data of individual smoke layers. Data are provided at 12 levels per decade change as pressures of 1000–1 hPa. Low-altitude data were excluded to reduce impact of the strong gradient in the H2O mixing ratio across the tropopause. The highest peak pressure was 73 hPa (average of 38 hPa), which is lower than reported in our recent study (Martinsson et al., 2022) due to the higher altitude of the tropopause caused by the lower latitudes in the present study.

MLS data were used in comparison with CALIOP data. A shift down of CALIPSO orbit in September 2018 to the CloudSat level caused a variable horizontal distance between CALIOP aerosol and MLS H2O observations. Measurements with horizontal distances less than 330 km (average 180 km) were used, which led to periodical loss of data. The data were screened using recommendations by the MLS team (Livesey et al., 2022).

The CALIOP satellite instrument observed a dense smoke layer at 11–16 km altitude located around the tropopause over the Tasman Sea (Fig. 1a–d), causing a clear increase in the aerosol load in the LMS and shallow BD branch already on New Year's Eve (Fig. 4). Large amounts of smoke were observed in the following days. Strong smoke signals were seen spread within the stratosphere of the southern mid-latitudes in the beginning of January (Figs. 3c and 4b–d) and continued to be strong during the rest of the month.

The aerosol load was low in the southern extratropical stratosphere before the smoke injection by the Australian fires (Fig. 3a). Volcanic perturbations were present at 20 km altitude in the tropics and in the northern extratropics. These stem from eruptions in June and August 2019 by the volcanic eruptions of Ulawun and Raikoke (Kloss et al., 2021), which had a low impact on the southern extratropics. In fact, the stratospheric aerosol load was lower in the southern extratropics than anywhere else.

The smoke spread latitudinally, almost exclusively to the south (Fig. 3). Figure 4 illustrates the AOD in three stratospheric layers. Most of the smoke stayed below the 470 K isentrope (the two lowest layers, Fig. 4b–c), but a minor part of the smoke rose by radiation heating to the layer with the deep BD branch (Fig. 4a), where it continued to rise. A clear AOD increase was evident in the southern mid-latitudes and high latitudes persisting throughout 2020.

We find that the injected wildfire smoke increased the stratospheric aerosol load by volcanic proportions (Fig. 5a). The fires induced an AOD increase of more than 3 times higher in its first year than the North American fires in 2017 did and slightly higher than the Calbuco eruption in 2015 (Table 1). That eruption yielded the largest volcanic impact in the Southern Hemisphere since the Mount Pinatubo eruption in 1991 (Friberg et al., 2018; Martinsson et al., 2022; Rieger et al., 2015; Thomason et al., 2018). The impact of the Australian fires was only matched in size by the large eruptions of Sarychev (2009) and Raikoke (2019), which both occurred in the Northern Hemisphere (Table 1).

Stratospheric background aerosol consists mainly of sulfurous and carbonaceous compounds (Martinsson et al., 2019). Volcanic aerosol contains large amounts of sulfurous, carbonaceous and crustal components (Andersson et al., 2013; Friberg et al., 2014; Martinsson et al., 2009), whereas smoke mainly consists of organic compounds and BC (Garofalo et al., 2019). The smoke had a rather different impact on the stratospheric aerosol load than the volcanic particles from the Calbuco eruption (Fig. 5a). The rise in AOD for several months after the volcanic injection stems from prolonged particle formation from volcanic SO2 and particle growth. Conversely, wildfire smoke particles are mixtures of primary BC particles and organics that form within hours or days, explaining the initial rapid rise in the AOD after the fires. In the CALIOP data, a second peak in AOD arose a few weeks after the first. We will elaborate more on this unexpected feature in the following sections.

The AOD evolution showed similar patterns in the two lower layers (Fig. 5b). The first peak in AOD is seen early both in the mid-layer and lowest layer similar to our observations of smoke from the North American fires (Martinsson et al., 2022). The upper layer shows a slow rise in AOD, due to the time required for smoke to rise from the mid-layer. This rise was also observed by the limb-scattering instrument OMPS-LP (Khaykin et al., 2020). However, that AOD increase constitutes only a small portion of the total stratospheric AOD from the fires (Fig. 5b).

The rapid rise in AOD after the fires is not seen in the OMPS-LP data (Fig. 6). CALIOP data reveal that the smoke had twice the peak increase over the background in AOD as the eruption of Calbuco did. Studies based on OMPS-LP data (Khaykin et al., 2020) report almost indistinguishable peak impact on the AOD from these two events. The OMPS-LP AODs after the fires increased much more slowly than for CALIOP and did not capture the first peak in the AOD in Fig. 5a. This discrepancy is explained by differences in observation systems. OMPS-LP suffers from event termination already at low light extinction, which inhibits quantification of dense aerosol layers such as fresh wildfire smoke (Fromm et al., 2014; Lurton et al., 2018; Martinsson et al., 2022). The line of sight is orders of magnitude longer for the limb viewer OMPS-LP than for the nadir viewer CALIOP, causing the difficulty of observing dense aerosol layers (Martinsson et al., 2022). Hence, CALIOP suffers less from light attenuation, and the data can be corrected for light attenuation from the smoke particles, enabling us to compute the AOD of also the densest smoke layers (Martinsson et al., 2022). After 1–2 months the limb viewer problem with event termination is reduced, making a comparison of the different instruments feasible. The evolution in stratospheric AOD for the two instruments are compared in Fig. 6, illustrating the slower rise for OMPS-LP. As pointed out in the Methods section, the OMPS-LP wavelength closest to CALIOP is not useful in the Southern Hemisphere (Taha et al., 2021). The light at longer wavelengths (OMPS-LP) is scattered less than at shorter wavelengths (CALIOP), resulting in lower AODs for OMPS-LP. Similarly slow rise in the AOD from OMPS-LP was shown for smoke from the 2017 North American fires (Martinsson et al., 2022). Our present study shows another example of when space-borne lidar is required for quantification of the stratospheric AOD when dense aerosol layers are present.

We find that smoke was transported to the stratosphere from two fire events, by tracking smoke back to fires in eastern Australia, combining CALIOP with OMPS-NM (Fig. S1). The first-event pyroCbs started on 29 December (Peterson et al., 2021), and the first CALIOP observations were 2 d later, on New Year's Eve. Those injections positioned smoke directly around the tropopause, i.e., partly in the stratosphere (Fig. 1a–d). One large, dense smoke layer was transported east from the fire region and was stuck for weeks over the southeastern Pacific (Figs. 7a and S1), where it formed a vortex, isolating it from mixing with surrounding air (Kablick et al., 2020; Khaykin et al., 2020). The isolation made it easier to track this smoke. We find that the large, dense smoke layer rose by a mean velocity of 260 m d-1 for the first 50 d after the pyroCb injections (Fig. 7b), similar to previously reported figures (Khaykin et al., 2020).

We also identified several other smoke layers from the first event located at lower-stratospheric altitude and not connected to the large, dense smoke layer (Fig. S1). Horizontally, these were transported more rapidly and could not be tracked during as many days due to mixing with the surrounding air.

The second fire event occurred on 4 January, but smoke from this event showed only little immediate stratospheric influence (Figs. 2, S8–S16). Also Peterson et al. (2021) reported much larger stratospheric impact from the first fire, based on studies of the fires' immediate impact (2021). Ten days after the pyroCb formations we start to see more stratospheric influence (Fig. 7). CALIOP images reveal the addition of large, dense smoke layers to the upper troposphere after 4 January (Fig. 2). We studied the temporal evolution of the smoke layers' position relative to the tropopause (Fig. 11). The smoke layers are clearly located below the tropopause in the first days after the second fire with minor overshooting parts (e.g., Fig. 2f and j). Over time, more and more smoke appears in the stratosphere. Hence, the smoke was transported gradually across the tropopause in the weeks following the fire injections to the upper troposphere. We kept following this smoke in the stratosphere for 20 d. Interestingly, it rose at approximately the same rate as the large, isolated smoke layer from the first event, 250 m d-1 (Fig. 7b).

By studying individual smoke layers, we find evidence of morphological transformation of the smoke particles during the first month after pyroCb injections. The CALIOP instrument is depolarization sensitive. Non-spherical particles depolarize the scattered light, increasing the depolarized signal retrieved by the sensor. We find a steady increase in the particle depolarization ratio in stratospheric smoke from both the first and second event (Fig. 7c). The trend lasts more than 30 d in the isolated layer from the first event, after which the particle depolarization ratio becomes stable at a value of 0.15. A similar trend was observed in the weeks following the August 2017 fires in western North America (Martinsson et al., 2022), whereas the opposite trend was observed when comparing fresh smoke with aged smoke well mixed with the background aerosol (Baars et al., 2019). The depolarization ratio of the aerosol from the second event deviates clearly from that of the first event by being much lower. We will discuss this difference further in a section below.

To study the individual stratospheric impact of the two events, we need to separate data into two groups. Peterson et al. (2021) reported that pyroCbs reached the stratosphere mainly during the first-event fires and to a lesser extent during the second event. Figure 1 shows that smoke from the first event reached the stratosphere shortly after the fires, whereas large amounts of smoke from the second event reached the upper troposphere (Fig. 2). We do not see evidence of large direct smoke injection to the stratosphere from the second-event fires in the CALIOP data, neither in the nighttime nor in the daytime data. Hence, most of the immediate stratospheric impact stems from the first event.

Smoke from the first event rose markedly in the stratosphere before smoke from the second event entered the stratosphere (Figs. 7b and 8a–c). Also their depolarization ratios differed markedly (Fig. 7c and Supplement). Clear differences between the first and second injection events are evident in the time–altitude distributions (Fig. 8) of the extinction coefficients, scattering ratios, and depolarization ratios, remaining over the course of more than 2 months. These parameters all show rising smoke in the weeks after the first event with particle depolarization ratios that increase over time (Fig. 8c). In mid-January, smoke with lower depolarization ratios started to ascend into the stratosphere (Fig. 7c), connecting the smoke below the minimum in Fig. 8c to the second event that ascended later into the stratosphere. The minimum in Fig. 8, occurring in between the smoke occurrences from the two injection events, illustrates the impact from a rapid stratospheric injection from mainly the first fire event and slow transport of smoke to the stratosphere from the second event. We use this minimum to separate smoke data from the two fires (dashed lines in Fig. 8) to form the AOD of the two events and investigate their individual impact on the stratospheric AOD.

The smoke AOD from the first event decreased rapidly over the first weeks, followed by a slow decrease until spring (Fig. 9a). A similarly rapid decline in smoke AOD and increasing particle depolarization ratio was seen for stratospheric smoke in our earlier study on the western North American wildfires in August 2017 (Martinsson et al., 2022). We have considered transport out of the stratosphere, sedimentation, cloud formation, and hygroscopic growth/shrinkage as explanations for the decline and found that the loss of material from the particles by photolysis is a plausible explanation for the decline (Martinsson et al., 2022). The long residence time due to the practically absent wet deposition in the stratosphere makes the effects of photolysis simpler to study compared with the troposphere. The importance of photolysis as a removal mechanism of organic aerosol is also supported by studies of photolysis in numerous laboratory experiments (Molina et al., 2004; Sareen et al., 2013) and by modeling (Hodzic et al., 2015; Zawadowicz et al., 2020).

The trend of decreasing stratospheric AOD after the first fire event together with increasing particle depolarization ratio over time suggests that photolytic loss depletes organic aerosol in the smoke. Thus, the BC fraction in the smoke will increase over time and may eventually constitute most of the smoke particles mass. We therefore interpret the morphological transformation (depolarization ratio) and AOD decrease after the first event as decay of organic aerosol in the stratosphere.

The particle depolarization ratio is much lower for smoke from the second event than from the first (Figs. 1, 2, 7c, and 8c). This difference may be found in the history of these smoke layers. Depolarization ratios for tropospheric smoke is lower than stratospheric (Haarig et al., 2018). The low depolarization ratios for smoke from the second event indicate (chemical) processing of the smoke in the troposphere. The presence of water in the smoke particles can cause a collapse of the BC agglomerates to a more spherical shape (Fan et al., 2016), which should result in lowering of the depolarization ratios. This explains the low depolarization ratios for smoke from the second-event fires, where smoke particles were exposed to the more humid tropospheric conditions for 10 d or more before entering the stratosphere. Different aging processes in the troposphere and stratosphere could thus be the cause of differing particle depolarization ratios of smoke from the two events, although differences in fire conditions cannot be ruled out (Haarig et al., 2018; Zhang et al., 2008).

We present two estimates on the depletion rate of organics for smoke from the first event. Our first estimate is computed directly from the zonal mean smoke AOD (Fig. 9b), suggesting a smoke half-life of 10±2 d.

Our second estimate on the decay rate of the smoke from the first event is based on the CALIOP observations of individual smoke layers marked as circles in Fig. 7a. We normalized the smoke signal with the local water vapor concentrations to investigate the evolution of the smoke layer composition. Water data were derived from the satellite-borne microwave limb sounder (MLS). MLS and CALIOP ran in different orbits during the smoke observations, limiting the amount of collocated data to 10 occasions for comparison, which were spread out in three groups over the first 50 d. We used an exponential fit and computed a corresponding half-life of 10±3 d (Fig. 9c).

The two estimates of the decay rate (10±2 and 10±3 d) presented here are identical to the half-life observed for the stratospheric smoke after the 2017 North American fires (10±3 d; Martinsson et al., 2022).

Stratospheric smoke from the first event is shown on New Year's Eve (Fig. 1), 2 d after the first pyroCb formations from the event. Peterson et al. (2021) coupled this transport to pyroCbs. Hirsch and Koren (2021) argued that smoke injections to the stratosphere may have occurred in the first week of January via cross-tropopause transport by convective clouds south of the fire region (38∘ S), where the tropopause height is lower. From the first event we do not see evidence of extensive cross-tropopause transport beyond the initial pyroCb-caused smoke injections in the CALIOP data (Fig. 8 and Supplement). Furthermore, the temporal evolution in the UVAI (Fig. 10a) indicates that most of the smoke remained north of 40∘ S in the days following each fire event, when most of the UVAI was generated.

Magaritz-Ronen and Raveh-Rubin (2021) suggested that cyclones and isentropic cross-tropopause transport caused smoke transport to the stratosphere over the South Pacific Ocean in the first few days of January. This is to some extent in agreement with the UVAI (Fig. 10), which increased on 2–3 January at 140–180∘ W, 35–40∘ S, indicating upwards transport of smoke. However, the depolarization ratios during the first week of January do not indicate cloud formation connected to the smoke layers. Furthermore, CALIOP observations show that large amounts of smoke were present in the stratosphere several days before the suggested cyclonic transport, indicating that pyroCbs were the primary cause of the direct smoke transport to the stratosphere.

The second-event fires (4 January) positioned dense smoke layers in the mid-troposphere and upper troposphere (Fig. 2). Figure 11 illustrates CALIOP observations of an individual smoke layer's vertical position relative to the tropopause, including layer tops, mid-points, and bases. These three parameters all show a gradual transport of smoke from the troposphere to the stratosphere, occurring over the course of 1–2 weeks. Such transport could be caused either by self-lofting by radiation heating, by isentropic cross-tropopause transport, or by a combination of the two phenomena. We investigated potential self-lofting of smoke from the second event by studying the potential temperature of the smoke layers at their top, mid-point, and base (Fig. 11). The increasing potential temperature over time indicates that they were subject to self-lofting by radiation heating, thus following the rising trend in the stratosphere, as demonstrated in Fig. 7b, also in the upper troposphere. The continued addition of smoke, transported from the upper troposphere, most likely resulted in the second peak in the AOD (Fig. 5). Hence, we explain the bimodality in AOD as the combined effect of rapid decay of pyroCb-injected smoke, mostly from the first event, and slower self-lofting of tropospheric smoke from the second event.

The second AOD peak does not show as rapid of a decay as the first one (Fig. 9a), likely due to depletion of organics during its long residence in the troposphere before the smoke from the second event entered the stratosphere. The AOD evolution (Fig. 9a) suggests cross-tropopause transport over the course of several weeks.

Our study indicates that smoke from the second event had a larger long-term impact on the stratosphere compared with the first event. It constituted 80 %–90 % of the smoke signal 6 months after injection to the stratosphere (Fig. 9a). Peterson et al. (2021) reported the opposite, namely that the first event injected 2–8 times more smoke than the second event did. Their study focused entirely on injections by pyroCbs. Our study indicates that additional processes, acting on smoke layers deposited in the upper troposphere by pyroCbs, were more important for the long-term stratospheric aerosol load than the direct smoke injection by pyroCbs. This is in part supported by Peterson et al. (2021), who reported more blowups and a larger area burnt for the second event. Smoke from the second event was likely already depleted by photolysis before entering the stratosphere and therefore more resistant to depletion by photochemical processing. Hence, the smoke from the second-event fires led to a more long-term impact on the stratospheric AOD and more climate cooling.