The overall design of the level 3 stratospheric aerosol product is shown in Fig. 2. To begin with, three input files are required for each granule under consideration. A CALIOP granule comprises half an orbit of data either from the daytime or the nighttime part of the orbit and divided by the day–night terminator. As noted in Sect. 1, the primary input files used for the present product are the lidar level 1B file, with the corresponding level 2 5 km merged layer and PSC mask files (Pitts et al., 2009) used for filtering. While the levels 1B and 2 merged layer files are based on V4, the currently available level 2 PSC files are based on V3. The latter is only available as a daily file and not for each granule separately. The 5 km merged layer file is a new product in V4 that reports the locations of all aerosol and cloud layers detected at both 5 km (also 20 and 80 km) and single shot (333 m) resolution (Vaughan et al., 2016).

In the background mode, clearing the features detected in the level 2 analyses is done by removing all the level 1B (L1B) attenuated backscatter values (for 15 consecutive L1B profiles) beginning at the top of the uppermost cloud or aerosol layers detected above the local tropopause using the layer heights reported in the 5 km merged layer file. Not only are signals from within the boundary of the layers removed, but the backscatter values at all altitudes below the layers are also removed to avoid issues in correcting for signal attenuation from overlying layers. While the attenuated backscattered signals within and below these layers are removed, this step will retain values that fall below the minimum detectable attenuated backscatter threshold of the CALIPSO layer detection algorithm (McGill et al., 2007). In this sense, the retrieved extinction in this mode will reflect only the aerosol loading below this threshold. Similarly, the signals below the uppermost PSC layers are also removed using the PSC mask file for the PSC-active months in the two hemispheres (December through March in the Arctic and May through October in the Antarctic). The PSC mask files report the occurrence of PSCs in both hemispheres (Pitts et al., 2007, 2009) and are reported for a single day on a 5 km horizontal and 180 m vertical grid for nighttime conditions only.

After clearing all level 2 and PSC layers detected above the local tropopause, all L1B attenuated backscatter values below the tropopause are removed. Further, all L1B profiles within the South Atlantic Anomaly (SAA) region are also removed. In this region, approximately between the Equator and 50∘ S in latitude and 20∘ E to 80∘ W in longitude (in the operational algorithm a polygon is used), the Van Allen belts come down to their lowest altitude (<200 km), thus exposing the satellite sensors to high fluxes of energetic charged particles that are trapped within the belts (Hunt et al., 2009; Noel et al., 2014). Large-amplitude noise excursions are often observed in attenuated backscatter profiles within this area, thus degrading the already low SNR in the stratosphere. Consequently, data over the SAA are not included when calculating the level 3 stratospheric aerosol product.

When creating the all-aerosol mode of the stratospheric aerosol product, it is necessary to remove any clouds and PSCs, much the same way as for the background case, but retain the detected layers classified as aerosols by the CALIPSO cloud–aerosol discrimination (CAD) algorithm (Liu et al., 2009, 2019). It should be mentioned that the CAD algorithm was also modified in V4 in order to be compatible with the new V4 532 nm calibration (Liu et al., 2019). In fact, the CAD algorithm was extended to the stratosphere for the first time in V4. Up until V3, any layer in the stratosphere was simply classified as a “stratospheric feature” and no distinction was made between clouds and aerosols, which is no longer the case in V4. However, even the V4 CAD algorithm may not perform very well at high altitudes because of low SNR, leading to generally lower absolute values of CAD scores (Liu et al., 2019). In any case, in the stratospheric altitudes above ∼20 km, clouds are seldom observed (except in the polar regions) and uncertainties in the CAD algorithm are not likely to affect the stratospheric aerosol product. For this mode, only aerosol layers with acceptable CAD scores (-100 to -20) are retained within the stratosphere. Layers identified as aerosols but with unacceptably low CAD scores (between 0 and -20) are removed as if they were clouds or PSCs.

In the next step, a nominal 5 km resolution profile is constructed by taking the average of these 15 filtered L1B attenuated backscatter profiles. Subsequently, a noise filter is used to screen out strong outliers from these 5 km profiles that might otherwise lead to biases in high-latitude and/or high-altitude regions. The noise filter used for the current version of the product is a reconfigured version of the same filter that is used in the CALIPSO range-dependent automated level 2 layer detection algorithm. Essentially a range-dependent threshold array of attenuated scattering ratios is constructed, which incorporates noise from two types of sources. The first category is the range invariant noise and includes detector dark noise and noise from the solar background light. The second category is the range-dependent noise from single shot measurements and is calculated from the molecular models. Using this range-dependent threshold, outliers are removed (for details see Vaughan et al., 2009, Sect. 2c). After removing the outliers, the 5 km profile is assigned to the appropriate spatial grid. This process is then repeated for all the profiles in the level 1B file. The resulting filtered 5 km profiles are then averaged to create a single mean attenuated backscatter profile for each grid cell.

In the final processing step for each granule, another quality screening is employed to identify and remove any lingering tenuous cirrus cloud in the lower stratosphere that might have escaped the layer detection mechanism due to low backscatter values. For the background mode, we can safely assume that the background aerosols are uniformly spherical and thus have a near-zero depolarization ratio. Since ice crystals in even the most tenuous cirrus violate this assumption, we use a threshold of 5 % in the volume depolarization ratio (ratio of the attenuated backscatter measured in the perpendicular and parallel channels at 532 nm; Hunt et al., 2009) to detect weakly scattering residual clouds. However, for the all-aerosol mode, this strategy will not work. This is because volcanic ash is typically nonspherical, has high volume depolarization values (∼25 %–30 %), and would thus be removed along with the cirrus clouds. On the other hand, attenuated color ratio values (i.e., the ratio of the total attenuated backscatter coefficients at 1064 and 532 nm) are generally larger for clouds compared to volcanic ash and may thus be used to filter out clouds while still retaining volcanic ash (Winker et al., 2012; Vernier et al., 2013). We have used a threshold value of 0.5 in the attenuated color ratio in the all-aerosol mode in an attempt to retain the volcanic ash rather than the filter on the volume depolarization ratio. The effects of these filters in both modes are illustrated in Fig. 3 using height–latitude cross sections of attenuated scattering ratios for the month of June 2011.

During this month two strong volcanic eruptions took place, Nabro in the Northern Hemisphere (13 June; 13∘ N, 41∘ E) and Puyehue–Cordón Caulle in the Southern Hemisphere (4 June; 40∘ S, 72∘ W). The composition of the Nabro plume was mostly sulfate, while the composition of Puyehue–Cordón Caulle was mostly ash, at least initially (Penning de Vries et al., 2014; Vernier et al., 2013). In the background mode (Fig. 3a), the removal of all detected layers combined with the application of the volume depolarization filter ensures that stratospheric perturbations from these two volcanoes are mostly excluded. Figure 3b shows the effect of including aerosol layers in the stratosphere with acceptable CAD scores (|CAD|>20) while still using the volume depolarization filter. Now the Nabro plume can be clearly seen but not that of Puyehue–Cordón Caulle. This is because the sulfates in the Nabro plume have low volume depolarization ratios that fall below the threshold and are thus retained, while the ash layers with a high volume depolarization ratio from Puyehue–Cordón Caulle are removed. On the other hand, including the aerosol layers but substituting an attenuated color ratio threshold of 0.5 in place of the volume depolarization ratio filter, as shown in Fig. 3c (all-aerosol mode), reveals both the Nabro (near 30∘ N) and Puyehue–Cordón Caulle plumes (near 50∘ S) quite clearly. Note the high scattering ratio values in the Antarctic latitudes between 15 and 25 km. Because the PSC mask algorithm is optimized specifically for PSC detection, its increased sensitivity allows it to detect a considerably larger fraction of faint PSCs relative to the more generalized and generic level 2 feature detection algorithm. Since all PSC layers detected by the dedicated PSC detection algorithm were removed, what remains are the signatures of only those particles below the detectability threshold of the PSC mask data product. Note that the enhanced scattering ratios near 25–30 km represent the tropical reservoir of stratospheric aerosols (Trepte and Hitchman, 1992; Kremser et al., 2016). Further, the high scattering ratios near 50∘ N are likely due to the Grímsvötn volcano, which erupted in May 2011.

Using a constant threshold to discriminate between different classes of inherently noisy measurements can entail significant risk of misclassification. For example, using a higher attenuated color ratio acceptance threshold to ensure the identification of strong ash plumes (e.g., for Puyehue–Cordón Caulle above) may result in a significant amount of cloud contamination. Similarly, an acceptance threshold set too low will likely exclude all clouds while simultaneously discarding much of the ash signal.

The impact of using the attenuated color ratio and volume depolarization ratio filters on removing thin cirrus clouds in the all-aerosol and background-only components, respectively, is illustrated in Fig. 4 using the attenuated scattering ratios measured at 17 km during December 2011. In Fig. 4a, all the aerosol layers are retained, much like the all-aerosol component, except that neither the volume depolarization ratio filter nor the attenuated color ratio filter is used. The high scattering ratios between about 30 and 55∘ N are due to the Nabro plumes that have spread around the Northern Hemisphere by December 2011. Apart from this band, high scattering ratios are also seen in a tropical band (between 25∘ S and 25∘ N) over the western Pacific as well as over parts of Africa. These reflect the thin cirrus clouds occurring in the upper troposphere and lower stratosphere near the tropopause (Sassen et al., 2009). Figure 4b shows the distribution in the all-aerosol mode wherein the attenuated color ratio filter is used. Clearly a significant number of pixels with high scattering ratio (thin cirrus clouds) in the tropics has been removed while still retaining the volcanic aerosol signature. In Fig. 4c we see the impact of the volume depolarization filter in the background mode. Now most of the cirrus clouds have been removed. Note that the aerosol signature has remained much the same in all three distributions. That is because by December 2011, there may not be many volcanic layers left; as such, enhanced scattering from the volcanic material is still present. This figure shows that for quiescent conditions or when the aerosol load is not very high (so that not many plumes are detectable as layers in the CALIPSO L2 algorithm), the volume depolarization filter will do a better job of clearing the thin cirrus clouds, thus making the background component the mode of choice. In any case, note that the thin cirrus clouds mostly affect the tropical latitudes and near the tropopause at ∼16–18 km (Sassen et al., 2009). Note that both the volume depolarization ratio and attenuated color ratio filters are applied only below 25 km, as no cirrus is expected above this altitude.

Figure 5 shows the profiles of the attenuated scattering ratio for the background and all-aerosol modes in July 2009 for the grid cell centered at 47.5∘ N and 130∘ E. The enhanced scattering ratio in the lower stratosphere between 10 and 17 km is due to the inclusion of detected aerosol layers from the Sarychev volcano (48.1∘ N, 153.2∘ E), which erupted in June 2009. Note that backscatter from some of the Sarychev aerosols that fall below the minimum detectable backscatter threshold of the level 2 layer detection algorithm will contribute to the background profile.

After deriving the granule-averaged data, we create monthly averaged gridded profiles of attenuated backscatter by aggregating all profiles during each month of the mission. In addition to the attenuated backscatter coefficient profiles, profiles of molecular and ozone number densities, temperatures, and pressures reported in the L1B files are also averaged and gridded for use in the subsequent retrieval procedures.

Figure 6 depicts the spatial distribution of the number of samples that contributed to the two components at 17 km during July 2009. The grey grid cells over South America and parts of South Atlantic Ocean correspond to the SAA, over which all data samples are rejected. The higher number of samples for the all-aerosol mode over the Asian summer monsoon region reflects the signature of the aerosol in the Asian tropopause aerosol layer (ATAL; Vernier et al., 2011b). A higher number of samples in the all-aerosol mode, albeit to a lesser degree, can also be seen over North America, which is likely related to the Sarychev volcano as mentioned above. Also note the high number of samples over parts of Antarctica, which is partly from oversampling due to orbital configuration and partly related to small particles below the detectability of PSCs by the PSC mask algorithm.

The monthly mean profiles of the gridded 532 nm attenuated backscatter coefficient (β′), constructed using the procedure described in the preceding section, along with gridded profiles of molecular backscatter coefficients (βm), molecular extinction coefficients (αm), and ozone absorption coefficients (αO3), are used to retrieve the particulate backscatter coefficient (βp) using 2βp(z)=β′(z)/Tm2zTO32zTp2z-βm(z), where 3aTm2z=exp⁡-2∫0zαmr′dr′, 3bTO32z=exp⁡-2∫0zαO3r′dr′,and 3cTp2z=exp⁡-2ηpSp∫0zβpr′dr′. In these expressions, ηp is the particulate multiple scattering factor, Sp is the particulate lidar ratio (i.e., the extinction-to-backscatter coefficient ratio), and Tm2z, TO32z, and Tp2z are, as previously defined, the molecular, ozone, and particulate two-way transmittances. The molecular backscatter coefficients and molecular and ozone two-way transmittances can be calculated from molecular model data (e.g., as described in Kar et al., 2018a). The molecular model used exclusively throughout the CALIPSO V4 data products is MERRA-2 provided by NASA's Global Modeling and Assimilation Office (Gelaro et al., 2017). For the CALIPSO stratospheric aerosol product, the particulate multiple scattering factor is taken as 1 for all species of stratospheric aerosols, consistent with the approach taken in the CALIPSO level 2 aerosol retrievals (Young et al., 2013, 2016, 2018).

Given an appropriate value of the lidar ratio, Eqs. (2) and (3c) can be solved iteratively to obtain estimates of βp(z) (Young and Vaughan, 2009). Estimates of particulate extinction coefficients are subsequently obtained using σp(z)=Sp×βp(z). The V1.00 release of the level 3 stratospheric aerosol product uses a value of Sp=50 sr for the stratospheric aerosol lidar ratio, which is a typical value used for stratospheric aerosols for background conditions and in the absence of significant ash and sulfate injections from volcanoes (Trickl et al., 2013; Ridley et al., 2014; Sakai et al., 2016; Kremser et al., 2016; Khaykin et al., 2017). Note that the lidar ratios could also be significantly different for stratospheric perturbations resulting from smoke intrusion from pyroCb events (Peterson et al., 2018; Khaykin et al., 2018). For this first version of the CALIPSO stratospheric aerosol product we have used only a single lidar ratio. We also assume Sp to be constant at all latitudes and over the entire altitude range. The retrievals are carried out beginning at 36 km and extending downward to either 8.3 or 1 km below the tropopause; processing stops when the higher of these two altitudes is reached. Extending the range below the tropopause, when possible, is intended to help in studies of the upper troposphere and lower stratosphere (UTLS). To guarantee uniform results across multiple CALIPSO data product levels, the level 2 CALIPSO extinction retrieval module is used to calculate the level 3 profiles of stratospheric aerosol extinction and backscatter coefficients and their uncertainties. The details of the retrieval process and uncertainty estimates are given in Young and Vaughan (2009) and Young et al. (2013, 2016, 2018) and are not repeated here.

Volcanoes are one of the primary sources of stratospheric aerosols (e.g., Kremser et al., 2016). Ground-based lidar studies have indicated a positive trend in stratospheric sulfate aerosol loading since the turn of the century, which was initially attributed to anthropogenic emissions of SO2 from coal burning in southeast Asia (Hofmann et al., 2009). However, closer scrutiny suggests that the increase is instead related to emissions of SO2 from a large number of moderate volcanic eruptions, as was subsequently suggested by Vernier et al. (2011a) based on analyses of CALIPSO data. Several volcanoes with stratospheric impacts have been recorded since the study by Vernier et al. (2011a). Volcanic signatures in CALIPSO data were examined more recently by Friberg et al. (2018).

Figure 7 shows the time–altitude cross section of the zonally averaged extinction in the middle to high northern latitudes (40–60∘ N), the tropics (25∘ N–25∘ S), and the middle to high southern latitudes (40–60∘ S) from the CALIPSO level 3 stratospheric aerosol product between January 2007 and December 2017. The signatures of many volcanoes are clearly evident in this figure in all three latitude bands. The strongest extinctions are seen for Sarychev and Nabro in the Northern Hemisphere and Calbuco in the Southern Hemisphere. Note that the effects of some of the volcanoes can last for several months as they spread to other latitudes by isentropic transport.

Apart from volcanic material, smoke from strong biomass burning events can also reach the stratosphere during so-called pyrocumulonimbus events (Fromm et al., 2010; Peterson et al., 2018). During the “Black Saturday” event, smoke from strong bushfires in Victoria, Australia, on 7 February 2009 is known to have impacted the stratosphere. Plumes from this blaze eventually reached altitudes of 16–20 km and were readily visible in satellite imagery (de Laat et al., 2012; Glatthor et al., 2013). The signature of this event can also be identified in Fig. 7b, reaching up to nearly 22 km. The signature of another strong pyroCb event can be seen at northern middle to high latitudes (top panel) in August–September, 2017. This event is discussed in detail below. Note the seasonal pattern of high extinction near 12–15 km in the northern middle- to high-latitude summer, seen most clearly between 2012 and 2017. The reason for this is not entirely clear at this time but could again be due to fire events. Cirrus cloud contamination could be another factor. However, the same pattern is also seen in the background mode (not shown) to a slightly lesser degree, which suggests that cloud contamination may not be very significant.

The extreme pyroCb event that occurred in August 2017 over British Columbia in Canada has been extensively studied recently and has been likened to volcanic perturbations in the stratosphere in terms of intensity and duration (Khaykin et al., 2018; Ansmann et al., 2018; Haarig et al., 2018; Peterson et al., 2018). Figure 8 shows an example of the CALIPSO measurements of this pyroCb event. The signature of the smoke plume is seen as extremely high attenuated backscatter (opaque at 532 nm) between 60 and 65∘ N. The very high attenuated color ratio (∼1.6) seen at the base of the plume (Fig. 8c) is a tell-tale signature of smoke (e.g., Liu et al., 2008). The high volume depolarization ratio (≥0.1) seen in Fig. 8b is somewhat unusual for smoke and suggests the presence of irregular soot particles, mineral dust, and possibly some ice particles, with fast adiabatic lifting possibly retaining the initial irregular shapes (Haarig et al., 2018; Khaykin et al., 2018). This high color ratio combined with the unusually high depolarization ratio results in the plume being identified as a mixture of smoke and “volcanic ash” (Fig. 8d), the latter being misclassifications by the CALIOP V4 level 2 scene classifier.

Figure 9 shows the height–latitude cross sections of CALIOP attenuated scattering ratios from the stratospheric aerosol product between August 2017 and November 2017; it captures the evolution of the aforementioned pyroCb event. After the original injection of smoke in August 2017 at midlatitudes, the smoke spreads to lower latitudes as can be seen in these monthly mean spatial distributions from the level 3 stratospheric aerosol product. As in Fig. 3, the feature with a high attenuated scattering ratio near 25–30 km seen in all four panels is the signature of the tropical reservoir of stratospheric aerosols, maintained by a complex interplay of transport from the troposphere and stratospheric dynamics as well as microphysical processes including the Brewer–Dobson circulation, the quasi-biennial oscillation (QBO), evaporation, and sedimentation (Trepte and Hitchman, 1992; Kremser et al., 2016).

Figure 10 shows the height–latitude cross section of the retrieved 532 nm extinction coefficients for the all-aerosol mode from March to December 2014, which captures the evolution of the Kelud eruption (February 2014; 7.9∘ S, 112.3∘ E) in altitude. The gradual lofting of the plume, with its top rising from ∼21 km over the tropics in March to ∼24 km in the same general location several months later, shows the signature of stratospheric dynamics in the CALIPSO stratospheric aerosol product. The persistence of the stratospheric perturbation for several months is consistent with the results of Vernier et al. (2016), who found the presence of ash in the lower stratosphere 3 months after the Kelud eruption from balloon observations. Note the high extinction values near 50–60∘ N in the lower stratosphere (∼10–15 km). These are similar to the summer rise in extinctions at these latitudes as discussed earlier (Fig. 7) and are possibly due to biomass burning effects but could also be related to possible cloud clearance issues. As also mentioned above, the high extinctions at high southern latitudes could be related to scattering from particles below the PSC detection threshold as well as to transported volcanic material from Kelud.

Figure 11 shows the spatial distribution of the retrieved extinction coefficients at 17 km for the month of August 2015 for the all-aerosol mode. Two strong perturbations of the lower stratosphere can be seen in this plot. The first is the plume from the Calbuco volcano in Chile, which erupted in April 2015. The initial plumes would be missed in the level 3 stratospheric aerosol product because data over the SAA region were not included. However, the plumes quickly spread around the Southern Hemisphere in a belt between 60 to 30∘ S (Lopes et al., 2019) and can be seen in the level 3 stratospheric aerosol product from May 2015 onwards for several months. The other is the plume of high extinction over southeast Asia and extending to the Arabian peninsula to the west. This is the location of the Asian summer monsoon anticyclone, which has been known to be a reservoir of pollution during the monsoon months and results from the deep convective outflow of pollutants, including both gases and aerosols as well as their precursors from the surface layers (Kar et al., 2004; Vernier et al., 2011b).