The Atmospheric Infrared Sounder (AIRS) was launched on board the Aqua satellite in 2002 with the purpose to improve weather predictions and provide measurements of gases important for our understanding of the climate (Chahine et al., 2006). AIRS measures infrared light from the atmosphere in 2378 channels between 650 and 2665 cm-1 with a high spectral resolution (Chahine et al., 2006). The horizontal resolution of AIRS pixels in a product is 15 × 15 km2 at nadir and 18 × 40 km2 at the edges of a swath (Prata and Bernardo, 2007).

Using the channels sensitive to SO2 in the spectrum covered by AIRS, SO2 column densities are retrieved (Carn, 2005; Prata and Bernardo, 2007). In this study we used SO2 data provided by Fred Prata, with methods described in Prata and Bernardo (2007). Their dataset is based on a least-squares fit between results from radiative transfer simulations and the observed spectrum from AIRS. They reported that the accuracy of the retrieval scheme was ± 6 DU (1 DU = 2.9×10-5 kg SO2 m-2), but in a case with little water vapour interference and good background conditions, the accuracy has been estimated to be as good as ± 3 DU (Eckhardt et al., 2008). A collection of AIRS swaths covering the Sarychev SO2 cloud is shown in Fig. 2. Using this collection of AIRS swaths, the complete SO2 emissions are studied in this paper.

CALIOP is a polarization-sensitive lidar instrument, and was launched on board the CALIPSO satellite in 2006 (Winker et al., 2007) . CALIOP has been used extensively to track the height of volcanic aerosols (Kristiansen et al., 2010; Vernier et al., 2011; Andersson et al., 2015; Friberg et al., 2018; Sandvik et al., 2019). In Fig. 3, we show a CALIOP swath containing volcanic aerosol from the Sarychev 2009 eruption. In this study we have used the Level 1B product from version 4–10 (Kar et al., 2018; Getzewich et al., 2018), which has a horizontal resolution of 300 m and is the raw product.

While SO2 data from passive satellite sensors can have an estimated altitude resolution of 1–2 km (Carboni et al., 2012), CALIOP has a higher vertical resolution. The vertical resolution is highest at lower altitudes: 30 m between the surface and 8.2 km altitude, 60 m between 8.2 and 20.2 km altitude, 180 m between 20.2 and 30.1 km altitude, and 300 m above 30.1 km (Winker et al., 2010), i.e. more than an order of magnitude better than satellite-borne SO2 sensors in the lower stratosphere.

Since CALIPSO and AIRS at the time of the 2009 Sarychev eruption were in the A-train satellite constellation, CALIOP passes over the same air masses as AIRS. However, due to the narrow footprint of the CALIOP laser beam, it measures a thin slice of the AIRS footprint.

In this study we have used the attenuated backscattering from CALIOP to calculate the scattering from volcanic aerosol particles. From the total attenuated backscattering measured by CALIOP, βtot, we define aerosol backscattering as 1βaer=βtot-βmol, where we calculated the molecular backscattering, βmol, from the meteorological parameters from the MERRA-2 (Modern-Era Retrospective analysis for Research and Application) model provided with version 4–10 of the CALIOP data (Kar et al., 2018). In order to have a more height-invariant variable for aerosol scattering, we also define the scattering ratio as 2SR=βtotβmol. In this study, we use individual CALIOP swaths to observe the early transport of volcanic aerosol layers that clearly contrast the background aerosol. Since CALIOP did not produce any data on 13 June, our selection of swaths starts on 14 June, when the first aerosol layers are observed. The aerosol layers become less dense in individual CALIOP swaths as time progresses, and the last day for our selection of swaths is 22 June. Further into our method, this time interval has the added benefit of avoiding long transport times with the FLEXPART model. Due to the rapid removal of aerosols in the troposphere, we have solely focused on stratospheric aerosol layers in this study.

To make sure that not a large portion of the SO2 mass in the AIRS swath collection is located in the upper troposphere, we performed manual inspection of co-located CALIOP swaths to determine the height of the volcanic clouds. In half of the AIRS swaths in our time period, the SO2 was solely located in the stratosphere. Within the other half of the AIRS swaths, the volcanic aerosol layers were mostly located in the stratosphere, with minor aerosol layers in the underlying troposphere. Thus, the AIRS swaths shown in Fig. 2 contain volcanic aerosol in the stratosphere and very little in the troposphere. In the swaths passing over the region close to the volcano, there is also volcanic aerosol further down in the troposphere, close to the ground. However, the AIRS SO2 measurements are less sensitive at these altitudes due to water vapour interference.

In the chosen time interval, we found 75 CALIOP swaths (see Table S1 in the Supplement for full list). These swaths contained stratospheric aerosol layers that were clearly separated from the background aerosol and not in contact with ice clouds. The CALIOP swaths identified in this way contain most of the stratospheric aerosol layers from the Sarychev eruption in this interval.

To check if there was ash or ice altering the measured height profiles within these stratospheric aerosol layers, we used the ratio between perpendicularly polarized and total backscattering, called the volume depolarization ratio, and the ratio between the total backscattering of the 1064 nm and 532 nm channels, called the colour ratio. Previous CALIOP classification algorithms in version 3 would misclassify fresh stratospheric volcanic aerosol as ice clouds when the stratospheric aerosol had depolarization ratios over 0.03 (Liu et al., 2019). For ice-containing pixels, the depolarization ratios would mostly be above 0.3. This threshold was used by Khaykin et al. (2018). The threshold is corroborated by Fig. 4 in Liu et al. (2019), where aerosol layers become increasingly scarce when the depolarization ratio exceeds 0.3. These pixels were instead classified as clouds in their study. In the aerosol layers of the 75 CALIOP swaths, the particle depolarization ratios were 0.1–0.3. The depolarization ratios are lower in the later swaths, indicating formation and condensation of sulfate. Sedimentation of ash particles could alter the vertical profile of the aerosol, whereas the submicron sulfate particles have a negligible sedimentation rate during the short time period in our study (Martinsson et al., 2005). This may not be the case for more ash-rich eruptions. In such a case, the ash would be unevenly distributed vertically inside the layers since the large particles would settle more quickly than smaller ones, inducing a vertical gradient in the depolarization ratio (Vernier et al., 2016). We investigated the vertical distribution of the depolarization ratio and colour ratio in relation to the scattering ratio for the 75 CALIOP swaths. We found no evidence for vertical inhomogeneity in this study of the Sarychev 2009 eruption. This indicates that SO2 and aerosol particles have the same vertical distribution. This co-location means that we have detailed height profiles of how the SO2 is distributed, albeit in thin slices in each CALIOP swath. These thin slices of height profiles from all 75 CALIOP swaths were transported using the dispersion model FLEXPART (see next section) to the times of individual AIRS swaths so that the AIRS swaths get matching height profiles.

FLEXPART, full name FLEXible PARTicle dispersion model, is a trajectory and dispersion model with Lagrangian dynamics (Pisso et al., 2019). By being Lagrangian, FLEXPART tracks each individual air tracer particle's position instead of calculating box quantities. This study uses FLEXPART version 10.4. The model was originally developed to track radioactive particles (Pisso et al., 2019) but has since been used for many other types of studies, for example, to track volcanic clouds (Eckhardt et al., 2008; Kristiansen et al., 2010). Transport with FLEXPART produces results in good agreement with both transport simulated by the Norwegian Earth System Model (Cassiani et al., 2016) and observational data (Groot Zwaaftink et al., 2018; Langford et al., 2018).

In this study we released air tracer particles in FLEXPART from each aerosol layer observed by CALIOP, with each FLEXPART release corresponding to a single pixel in the CALIOP data's aerosol layers. In order to manage this and to simulate FLEXPART both forwards and backwards, starting from each CALIOP swath, we created one “RELEASES” file, specifying where and how many particles should be released into the model, and two “COMMAND” files, specifying how the model is run and for how long, for each CALIOP swath. Each RELEASES file contained around 95 000 air tracer particles.

We used ERA5 meteorological data in the FLEXPART simulations. ERA5 is the latest reanalysis product from ECMWF, replacing ERA-Interim (Hersbach et al., 2020). The preparation of ERA5 data for input into FLEXPART was done with flex_extract_v7.1 (Tipka et al., 2020) with an hourly 1∘ × 1∘ resolution on 137 vertical levels. In the CALIOP data, the tropopause and potential temperature levels come from the MERRA-2 model (Kar et al., 2018). Using the supporting meteorological dataset for the CALIOP swaths, our final vertical profiles have a resolution of 1 K in potential temperature. Converted to geometric altitude and pressure, this corresponds to 61 ± 56 m and 1.8 ± 2.9 mbar (mean ± 1 standard deviation) for the altitude range of the volcanic sub-clouds in the AIRS swaths in Fig. 1.

Starting from a section of a CALIOP swath that contains volcanic aerosol layers (see Fig. 5a and the solid blocks in Fig. 1), information that can be turned into a FLEXPART release is extracted and prepared. Smoothing is applied to the CALIOP data to reduce noise. This is done by applying a moving mean on the data; see Fig. 5b. A scattering ratio threshold (10 for night-time swaths and 15 for daytime swaths due to more noise in daytime data) is used to separate the volcanic cloud from the background. With the background filtered away, all that remains in the CALIOP data is separated volcanic sub-clouds; see Fig. 5c. In order to keep track of these sub-clouds throughout our analysis, we grouped the grid cells belonging to each sub-cloud and put ID labels on them; see Fig. 5d. These ID labels were used in the later analysis to keep track of which FLEXPART particles come from which sub-cloud. During manual inspection of the selected sub-clouds, sub-clouds with a high likelihood of containing ice and ash were removed. These ice and ash sub-clouds contained only 2.1 % of the total light scattering observed by CALIOP from the clouds classified as volcanic. Hence, their removal did not affect our final results.

With the sub-clouds now clearly identified, we want to convert the backscattering into FLEXPART air tracer particles. This is done by swath-wise scaling the number of released FLEXPART air tracer particles by the aerosol backscattering (βaer) in a pixel, making βaer the relative strength of a FLEXPART release. Thus, FLEXPART air tracer particles are released for each pixel in a sub-cloud in the CALIOP data and at the time, latitude, longitude, and geometric altitude of the pixel. Around 95 000 air tracer particles were released for each FLEXPART simulation, i.e. for each CALIOP swath (except for the two swaths that were divided into two separate simulations).

For the released FLEXPART air tracer particles in each CALIOP pixel, the potential temperature is stored. Potential temperature is a robust height coordinate in the stratosphere since the air transport normally follows isentropes, which are usually not aligned with geometric altitudes. Therefore, the potential temperature is used later as the vertical coordinate for making combined height profiles out of the FLEXPART simulations that cover the locations and times of the AIRS observations. Dense super-micron aerosol layers, such as those from Mt. Pinatubo 1991, can be warmed through absorption of radiation causing cross-isentropic transport, so-called self-lofting. It is unknown to what extent self-lofting is important for moderate-sized eruptions, but a study of Nabro (Fairlie et al., 2013), another moderate-sized eruption, found the effect to be minor (max 0.3 K d-1). Self-lofting is not captured by our method, but its impact can be assumed negligible for Sarychev given the low rates in Fairlie et al. (2013) and the short time span in our study.

The released particles from CALIOP were transported with FLEXPART forwards or backwards in time until the time of the AIRS swaths in Fig. 2 (see the black arrows in Fig. 1). Hence, we have collected and transported all the vertical information from the 75 CALIOP swaths to the times of each AIRS swath. An advantage of making one FLEXPART simulation for each CALIOP swath is that it increases the possible number of air tracer particles from every CALIOP swath. One swath contained aerosol layers at two locations separated by a large distance and could not be enclosed by boundaries without also including a large chunk of noise. The aerosol at the two locations was therefore prepared and run separately. Our approach enabled us to do all FLEXPART simulations on an ordinary personal computer. The alternative would have been to make a single FLEXPART simulation containing all CALIOP swaths, where program restrictions would limit the number of air tracer particles per CALIOP swath.

From a technical point of view, the procedures in the paragraph above were implemented by creating two FLEXPART COMMAND files for each RELEASES file: one COMMAND file tells FLEXPART to simulate in the forward time direction and one in the backwards direction (see the Supplement for an example COMMAND file). For forward (backward) simulations, FLEXPART starts at the closest hour before (after) the first (last) release in the corresponding RELEASES file and stops after the last (before the first) AIRS swath in Fig. 2. All in all, 152 FLEXPART simulations were made for a total of 836 simulated days.

While FLEXPART's internal calculations have no gridded resolution, the produced output data were placed on a grid. In our study, we chose 0.5∘ E × 0.5∘ N resolution for the Northern Hemisphere. The Southern Hemisphere was ignored since the cloud from Sarychev stayed far from the Equator in the first weeks after the eruption. The high horizontal resolution of the FLEXPART output was used because both CALIOP and AIRS have high horizontal resolutions. The chosen FLEXPART output time intervals were set to 30 min and are therefore within 15 min before or after an AIRS swath; i.e. there is only minor possible misalignment in time between the satellite and model data.

We used FLEXPART to transport the data in the CALIOP swaths horizontally through time. Since CALIOP already has the highly resolved height information, we rely on these observational data for vertical coordinates, while FLEXPART produces the horizontal transport. We therefore set only one single vertical coordinate for the FLEXPART output. This approach has the advantages that our method becomes independent of possible errors in vertical transport in FLEXPART and allow us to increase the resolution of the other output coordinates.

After FLEXPART has simulated every CALIOP swath, the numerous output files were combined. Vertical profiles were created for each AIRS pixel by spreading the SO2 mass in each AIRS pixel over a range of potential temperature bins. The amount of SO2 in each bin was computed by weighting the FLEXPART data with the aerosol backscattering data from CALIOP.

No chemistry is used during the transport or for the preparation of data before FLEXPART. Therefore, one unit of βaer represents the same amount of SO2, regardless of CALIOP swath time. The time it takes for the aerosol to form could thus affect the representation of the cloud. The 9 d used here is shorter than previous estimates of volcanic SO2 conversion in the stratosphere (Andersson et al., 2013), but ongoing SO2 conversion adds uncertainty to our estimation. The short time span covered in the present study can thus be assumed to have a small effect on βaer per SO2. This can be seen in Fig. 8 in Friberg et al. (2018), where the stratospheric aerosol load from Sarychev peaks in September, i.e. months after the last swath used in this study.

In Fig. 6, we show a snapshot of the FLEXPART-transported aerosol scattering from all 75 CALIOP swaths at the times of the AIRS swaths in Fig. 2 (similar to the grey areas shown in the concept sketch in Fig. 1). The geographical extents have been chosen to focus on the regions of Fig. 2 with the most SO2 for each of the five orbits. Starting from the rightmost subfigure and earliest time, the transported aerosol scattering outlines the largest SO2 cloud seen by AIRS. At 00:00 UTC on 19 June, more southern clouds are outlined, while the northern string of SO2 is sparsely covered by time-adjacent CALIOP swaths. At 01:30 UTC, the transported aerosol scattering fails to outline the faint SO2 cloud. In the two final comparisons, the transported aerosol scattering is centred on the SO2, but there are also difficulties in contouring the more peripheral SO2. While the transported aerosol scattering generally outlines the SO2 seen with AIRS, there are a few mismatches and false positives of where the FLEXPART-transported aerosol scattering indicates SO2 presence where there is none detected by AIRS.

The overlap between the transported aerosol and the SO2 from Fig. 2 varies for the five orbits of the AIRS instrument, which is shown in Fig. 6, where the rightmost subfigure contains the first orbit. While the aerosol generally outlines the horizontal location of SO2 during the first, fourth, and fifth orbits, the transported aerosol does not contain the SO2 seen during the second and third orbits. For the second orbit at 00:00 UTC shown in Fig. 6, the northern part of the SO2 is scarcely covered by the transported aerosol with 0.0086 Tg (0.8 % of all SO2 in the AIRS swaths) outside the 10-5 contour. The reason for this could be that the SO2 is located below the tropopause, but the aerosol scattering in the closest CALIOP swath is located only in the stratosphere. Also, CALIOP does not observe all parts of the volcanic cloud, and at some locations the height information can come from a relatively weak CALIOP profile. Therefore, we chose to horizontally average the combined transported aerosol within each individual AIRS swath. This horizontal averaging was done by first interpolating the combined transported CALIOP data from FLEXPART to the coordinates of the AIRS pixels. Then a horizontal averaging was made over the whole AIRS footprint to get a representative height profile. This averaging accounts for when FLEXPART and the observed SO2 are slightly mismatched or when there is limited coverage from the transported CALIOP swaths. The averaging created a single height distribution for each AIRS swath containing observations of the volcanic clouds. The horizontal averaging also compensates for the fact that CALIOP has limited spatial coverage, which means that not all parts of the AIRS cloud was completely covered by CALIOP's height information transported by FLEXPART. Of the 75 CALIOP swaths, 69 have trajectories that enter the main SO2 AIRS observations. The six CALIOP swaths whose trajectories do not enter the AIRS observations in Fig. 2 contain little aerosol scattering. There is a list of which CALIOP swaths are transported into which AIRS swaths in the Supplement.

The collection of AIRS swaths from Fig. 2 was found to cover almost all of the volcanic clouds from Sarychev during several hours of orbit; see Fig. 7a. One AIRS swath takes 6 min to scan. Thus, there are 240 swaths for each day. In this study, AIRS swaths will be referenced as “< day of June 2009 >. < swath number of day >”; for example, 18.223 means the 223rd swath on 18 June. The height profiles for the AIRS swaths in this collection are shown in Fig. 7b–d. Height distributions using geometric altitude and pressure as height coordinates were calculated from the potential temperature using ERA5 temperature data for the AIRS footprints. The height profiles in Fig. 7b–d clearly show that the clouds over the eastern Pacific Ocean (130∘ W) are located at higher altitudes than the clouds over eastern Siberia (130∘ E).

From this collection of AIRS swaths, a 3D global dataset was compiled for implementation in models. The dataset has 1∘ latitude × 1∘ longitude horizontal resolution and comes in the three different versions depending on which vertical coordinates are preferred (potential temperature, geometric altitude, and pressure). The vertical resolutions are 1 K for the potential temperature grid and 200 m for geometric altitude, and for the pressure grid, the pressure levels correspond to geometric altitude steps of 200 m. The dataset is available at https://www.snd.gu.se (last access: 12 November 2021).