GOMOS is an instrument based on the stellar occultation technique and provides atmospheric measurements in the UV-visible-IR range (248–690, 755–774 and 926–954 nm). The use of stellar occultation results in a high rate of occultation measurements, and, consequently, a very good spatial coverage compared to solar occultation. As a drawback, the signal-to-noise ratio of each measurement is much lower than in the solar case, and varies with the star characteristics (especially its magnitude and its temperature). The operational retrieval, IPF (Instrument Processing Facility), provides density profiles for trace gases such as ozone (O3), nitrogen dioxide (NO2) and nitrogen trioxide (NO3) , as well as aerosol extinction. However, the extinction shows a poor quality for the reference wavelength at 500 nm. For this reason an alternative inverse algorithm called AerGOM was specifically developed to optimize the aerosol retrieval . AerGOM provides vertical profiles of the same gas species, and the total extinction coefficient for the nongaseous species and its spectral dependence, currently over the range 250–750 nm. The nature of the total extinction fraction for nongaseous species is then inferred using simple criteria based on the geolocation, associated temperature value and extinction value, and each point of the vertical extinction profile is attributed to aerosols, cirrus clouds, polar stratospheric clouds or meteoritic dust.

From the AerGOM extinction, climate data records (CDRs) were developed in the framework of the ESA Aerosol CCI project for different quantities including the aerosol extinction and the related aerosol optical depth at several wavelengths 355, 440, 470, 550 and 750 nm;. Particular attention was paid to the grid choice, which should optimally render the information contained in the GOMOS measurement set. The most important conclusions of this optimization were that grid resolutions should be chosen to ensure a reasonable statistical sampling in most of the grid cells, and that it should optimally reflect the typical transport of volcanic plumes after an eruption reaching the upper troposphere or the lower stratosphere (UTLS). Therefore, the grid should represent, in a coherent way, the longitudinal and latitudinal air mass transport, and the time needed for this transport. Also, the temporal resolution should be short enough to enable the detection of volcanic signatures, taking into account the typical lifetime of the plume. In this respect, we could verify that time intervals of about 5 days are able to represent the signature of most of the eruptions injecting sulfuric gases in the UTLS, while such a signature is often diluted, underestimated or even disappears in the case of coarser grid cells. This is the case, for instance, for monthly zonal means, even though this representation is very commonly used in the field. The ability of the grid to reproduce the signature of volcanic plume in a satisfactory way is of particularly great importance when the CDRs are used to constrain climate models. More detail about the investigations of the optimal grid choice and all other aspects of the implementation of the CDRs can be found in .

In their current version (version 3.0), these CDRs are defined on a grid with a resolution of 5∘ in latitude, 60∘ in longitude, 1 km in altitude and 5-day time period. The records cover the whole ENVISAT period (March 2002–April 2012) and include the total extinction of nongaseous species, but also the polar stratospheric cloud (PSC) fraction and the cloud-free aerosol fraction which is dominated by sulfate aerosols below an altitude of 32 km. It is important to mention that cloud detection is not yet optimal, and that cloud contamination of the aerosol fraction is possible in the UTLS region. This issue is still under investigation.

The IASI dust dataset of the Université Libre de Bruxelles (ULB) was generated in the context of ESA CCI's project . It is based on a statistical regression technique and the use of a neural network trained on synthetic IASI data. A similar scheme has already been applied for the retrieval of NH3 (ammonia; ). As input variables it uses the IASI L2 pressure, humidity and temperature information, as well as spectral information and a CALIPSO (Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation) derived dust altitude climatology. The main output variables are dust optical depth at 10 and 11 µm (and 550 nm). Initial results and validation performance are provided in .

The ATSR (SU) algorithm has been developed at Swansea University for estimation of atmospheric aerosol and surface reflectance for the ATSR-2, AATSR sensors and SLSTR (Sea and Land Surface Temperature Radiometer) on Sentinel-3. Over land, the algorithm employs a parameterized model of the surface angular anisotropy , and uses the dual-view capability of the instrument to allow aerosol property estimation without a priori assumptions on surface spectral reflectance. Over ocean, the algorithm uses a simple a priori model of ocean surface reflectance at both nadir and along-track view angles. A climatology is used to constrain chemical composition of the aerosol components at 1∘×1∘ latitude–longitude grid, while the method retrieves aerosol size and optical thickness on a 10 km grid. Both optical thickness and size are retrieved as vertical column values. Size is not resolved vertically, but is represented by fraction of fine and coarse mode aerosol in total. The algorithm has been developed from initial prototype under the Aerosol CCI program, and results and validation performance for version 4.21 are provided in . The version used here (v 4.3) differs from that summarized in by improvements in retrieval of coarse/fine mode fraction, and improved cloud screening over ocean in the region of dense plumes, resulting in approximately 10 % greater coverage, with small improvement in correlation against AERONET (AErosol RObotic NETwork) values. AERONET is recognized as a reference dataset for validation of satellite data products .

OSIRIS was launched on board the Odin satellite, and has provided vertical profiles of limb scattered radiance between 280 and 810 nm since 2001 . The radiance profiles are inverted to provide aerosol extinction measurements at 750 nm at altitudes between 10 and 35 km with a vertical resolution of approximately 2 km . This technique provides high sampling rates with hundreds of measurements per day over the sunlit portion of the globe, enabling excellent spatial and temporal sampling of short-lived events. OSIRIS aerosol extinction retrievals agree well with coincident occultation measurements from Stratospheric Aerosol and Gas Experiments II and III during background periods but have known low biases above approximately 25 km, and will have some cloud contamination near and below the tropopause . Additionally, seasonal biases are possible due to the orbital geometry and changes in aerosol optical properties such as after volcanic eruptions may also bias the retrievals. These effects are described in more detail by . This work uses the OSIRIS version 5.10 aerosol retrieval averaged into daily, 5∘ latitude by 30∘ longitude bins for comparisons.

Volcanic emissions have a large impact on the stratospheric aerosol burden. Even small and moderate eruptions contribute to the stratospheric aerosol load due to convective transport of SO2 and its gradual uplift to the upper troposphere and the lower stratosphere, and resulting accumulation of sulfate aerosol. Volcanic SO2 injections explain most of the interannual variability of stratospheric aerosol extinction (decadal logarithm) observed by GOMOS, as depicted in Fig.  at three wavelengths. For each wavelength (350 nm in Fig. 1a, b; 550 nm in Fig. 1c, d and 750 nm in Fig. 1e, f), the GOMOS time series (Fig. a, c, e) showing the altitude dependence in the tropics, is compared with the EMAC simulation in resolution T63L90 including the dust contribution (Fig. b, d, f; see Sect. 4.2 for more detail). Figure  shows, at all three wavelengths, that an enhancement of the extinction value is observed around 16–18 km, corresponding to the aerosol load resulting from a succession of volcanic eruptions during the whole period 2002–2012. The eruptions of Nabro in June 2011 and the successive eruptions of Soufriere Hills and Rabaul in 2006 have the largest effects on extinction in the lower stratosphere in the observations and the simulation. The best agreement between GOMOS and EMAC is found in the case of the extinction at 550 nm (Fig. 1c, d), where the quality of the GOMOS retrieval is the best. At 750 nm (Fig. 1e, f) also, GOMOS measurements agree well with EMAC for the aerosol layer (16–22 km) where measured extinction values exceed ≈2×10-4 km-1. At lower altitudes (14–16 km), rather unstructured patterns of enhanced extinction are found by GOMOS, probably corresponding to cloud contamination. At 350 nm, where a decrease in the GOMOS quality is expected due to a loss in signal-to-noise ratio obtained in the UV spectral region while using cold stars, still the volcanic events stick out. More details over these aspects can be found in references . also present the latitude dependence of 550 nm aerosol extinction at 17 km altitude as observed by GOMOS and simulated by EMAC in the coarse resolution T42L90 in their Fig. 10.

Extinction in the lowermost stratosphere and upper troposphere is to a large fraction due to desert dust and organic carbon aerosol. These contributions were strongly underestimated in due to a crude parameterization in the used model version based on , but overestimated in . Both simulations were performed in the relatively coarse resolution T42L90. Dust reaching the UTLS is sensitive to model resolution, mostly via the convection parameterization . In Fig.  the simulated extinction at resolution T63L90 fits well to the GOMOS observations which appear to have a seasonal contribution from the Asian summer monsoon. For more detailed analysis, Fig.  shows observed and simulated extinction in the Asian sector at 17 km in the visible and the near-IR. The largest extinction values are indeed found at the location and time of the Asian summer monsoon at the altitude of outflow. This feature is clearest in years not perturbed by medium strength volcanic eruptions, for example 2010. For a clear separation, the contributions of wet dust and wet sulfate to extinction are displayed separately (Fig. 2e–h). The wet dust particles in the monsoon region have a larger median wet radius than the volcanic sulfate particles (e.g., from Sarychev in 2009, Fig. 2i) which is consistent with a relatively larger extinction in the infrared compared to the visible in the monsoon region in observations and simulations. Figure 2a–d demonstrates that dust is essential to reproduce the observations. Total extinction without wet dust in T63L90 is shown in the Supplement. Comparing Fig. S1b with Fig. 2g shows a small contribution of organics from biomass burning in northern spring (for volume mixing ratios see Fig. S2). Figure S1 also contains results from the T42L90 simulation of , showing that for this resolution the contribution of wet dust to extinction has to be downscaled (i.e., divided) by a factor of 2 to get agreement (Fig. S1d, factor of 3 if only dry dust is considered).

Observations by IASI and ATSR indicate a maximum in dust aerosol optical depth (DAOD) in early Northern Hemisphere summer over the Asian deserts located in the inflow regions of the monsoon (see Sect. 4.4). A similar feature is found in the simulations by EMAC. This supports our findings that desert dust is also important for the UTLS.

Desert dust transported to the UTLS mostly via the Asian summer monsoon contributes significantly to the seasonal cycle of total stratospheric aerosol optical depth (SAOD) in satellite observations and the EMAC simulations shown in Fig. 3b for the tropics (vertical integral of extinction above about 16 km) and in Fig. c for midlatitudes (above about 14 km). Global radiative forcing at the tropopause is depicted in Fig. a. The figure contains in black results from the T42L90 simulation of and in blue the T63L90 simulation with the high volcanic sulfur input derived for the coarse resolution. Green lines and symbols show estimates derived from satellite observations SAGE II, OSIRIS and CALIPSO;. Red shows results of the current model version in T63L90 with the dust scheme and corrected SO2 input (see Sect. 3 and Supplement). Concerning global radiative forcing, the volcanoes are the dominating effect with up to 0.13 W m-2 for Rabaul and Nabro compared to the volcanically quiet period in 2002. Here the use of the SO2 inventory for T42L90 in the T63L90 simulation (blue) causes an overestimate of up to 50 % in 2006 and 2007 due to accumulation effects of eruptions following in short sequence. This is visible in the overestimate of tropical SAOD depicted by the blue curve in Fig. 3b.

Especially in northern midlatitude summer SAOD in T42L90 appears to be high because at that resolution the convective transport of dust to the UTLS in the Asian monsoon region is overestimated (Fig. c). This is clearly seen in Fig.  which shows in black the T42L90 simulation, in green the observations of 550 and 750 nm SAOD by GOMOS, and in light blue (Fig. 4c, d only) by OSIRIS in different latitude bands, including the monsoon region. For the narrow latitude bands in Fig. 4c and d, inclusion of OSIRIS data is important because GOMOS coverage is often too low. Nevertheless, for a lot of features the two satellite datasets agree well. Using the higher resolution T63L90, for which the convection parameterization was developed, the agreement with the satellite observations is much better (Figs. 3 and 4, red) than with T42L90, especially at midlatitudes and in the subtropics. In the subtropics (Fig. 4d), the simulation with low resolution (black) always overestimates the monsoon peaks in August compared to the ones seen in the observations. Comparing the model results with OSIRIS in the northern tropics (Fig. 4c) indicates that some volcanic events are still missing in the inventory, for example in spring 2007 and 2010. This would also explain the differences in radiative forcing (indicated by crosses in Fig. 3a) in these years.

The simulated aerosol radiative heating, derived from radiation calls with and without aerosol, reflects the medium volcanic eruptions with the largest effects near 18 km (Fig. ). There, desert dust causes additional heating at the time of the Asian summer monsoon. In the UTLS, below, every year in September, a clear signal from biomass burning organic aerosol – its volume mixing ratio is shown in Fig. S2 of the Supplement – is visible. Above, around 22 km, the dust below in Northern Hemisphere summer causes a reduction of absorption of terrestrial radiation by ozone.

The first comparisons are carried out for EMAC in T63L90, the standard resolution used in the previous sections. Here AOD refers to the troposphere and stratosphere. The DAOD (dust AOD) in terrestrial infrared is most sensitive to the coarse mode of tropospheric dust. Figure a, b shows that the model reproduces most of the IASI features. DAOD in the visible spectral region (Fig. 6c, d) is too high over central Asia, pointing to an overestimate of dust in the accumulation mode near the Taklamakan Desert. The patterns in the IR and visible spectral range are different despite considering the factor 2 often applied by the AEROCOM/AEROSAT (Aerosol Comparison between Observations and Models) community for conversion in the color scales of Fig. 6a, b and c, d. This holds for model and observations. The fine mode AOD fraction, which is dominated by the accumulation mode, is slightly overestimated over Europe and underestimated in the biomass burning regions in Africa (Fig. 6e, f). In the model this is sensitive to the way the extinction of aerosol water is attributed to the soluble aerosol species, especially sea salt. Absorbing AOD, i.e., AOD × (1-ω) with ω representing single scattering albedo, agrees surprisingly well (Fig. 6g, h). In the total AOD (Fig. 6i, j) there appears to be too much sea salt in the model, or still suboptimal parameters for the sea salt composition which controls water uptake (see Sect. 3).

Figure  compares the annual average for 2011 of the 10 µm DAOD observed by IASI and simulated by EMAC in the low top version with high horizontal resolution (T106L31, about 1.1∘). The satellite retrievals are taken from version 8 of the ULB dataset. The simulation uses the dust emission scheme of which calculates the emissions online considering the meteorological conditions. To extract the DAOD from the total EMAC AOD at 10 µm, we apply a filter nullifying sea-salt-dominated AOD values. To identify the latter, we compare the AOD weighted with the volume of sea salt and dust.

The observed and modeled global DAOD distributions shown in Fig. 7 agree remarkably well. The pixel values of each map are strongly correlated with a correlation coefficient of 0.91. The overall AOD level is consistent as well, so that a similar variance in the pixel values is obtained for the observed (0.00038) and the modeled (0.00041) DAOD distribution. Interestingly, the DAOD from the older version 7 of the ULB dataset yields a pixel by pixel correlation coefficient of only 0.89 and a pixel value variance of only 0.00029. We conclude that the agreement of EMAC and IASI has improved with the update from version 7 to version 8 of the IASI ULB dataset.

The main disagreement of the two maps in Fig. 7 is the less pronounced maximum over the Taklamakan Desert in central Asia in the model result. This underestimation is related to the model surface friction velocity in mountainous regions like the surroundings of the Taklamakan Desert, which tends to be lower in simulations at higher horizontal resolution (e.g., T106) than at lower resolution (e.g., T63), possibly resulting in an underestimation of the dust emissions.

Figure  compares results from the T106L31 EMAC simulation for the annual average of the total AOD at visible and near-infrared wavelengths with AASTR retrievals using the ATSR (SU) algorithm version 4.3. Generally good agreement is obtained at 550 nm which is consistent with the good agreement between the 550 nm MODIS (Moderate-resolution Imaging Spectroradiometer) AOD and model results based on the same EMAC version . As for the T63L90 simulation, the model yields higher sea-salt-related AOD levels over the oceans. In contrast, the model AOD over the Sahara is lower than the satellite retrieved values. This becomes even more evident at larger wavelengths (Fig. 8c, f): the model AOD over the Sahara, in contrast to most other regions, has a stronger wavelength dependence than the observed AOD, corresponding to a larger Ångström exponent. This discrepancy might be resolved by adjusting the dust particle size distribution in the model under the constraint of not sacrificing the good agreement of model and observed AOD at 550 nm and at 10 µm. This could involve modifying the parameters of the log-normal modes, i.e., their widths and boundaries, but also reassessing the parameterization of relevant processes such as emission, deposition, coagulation and hygroscopic growth, or even adding an extra mode for extremely coarse particles which can be relevant close to dust sources. Over South America, the biomass burning regions of Africa, and India and China the wavelength dependence of model and observed AOD is largely consistent.