In situ measurements mainly provide information about the stratospheric aerosol size distribution, from which moments (such as surface area density and volume density) can be derived. These instruments directly sample stratospheric air, using a balloon-borne or aircraft platform. Examples include the optical particle counters (OPCs; Hofmann and Deshler, 1991; Jaeger and Deshler, 2002), the airborne focused cavity aerosol spectrometer (FCAS; Jonsson et al., 1995), and the nucleation-mode aerosol size spectrometer (NMASS; Brock et al., 2000). Size information can be obtained for particles in the size range 0.05–10 µm, depending on the particular instrument. In situ measurements can provide higher temporal and spatial sampling than other techniques during a measurement campaign. But the cost and effort associated with such a campaign limit this method to intensive case studies, rather than providing broad temporal and spatial coverage.

Ground-based lidars also make measurements of stratospheric aerosols (Poole and McCormick, 1988; Chouza et al., 2020). In this case, the signal consists of photons backscattered by stratospheric aerosols, and therefore the measurements provide the backscattering coefficient of the particles. This is typically converted into an aerosol extinction coefficient, using an assumed “lidar ratio” as the conversion factor. The signal-to-noise ratio of a single lidar measurement is generally low, which requires a combination of many individual shots to produce useful retrievals. Ground-based lidars are typically stationary, so they provide information at just one location, but they can provide high temporal sampling and good vertical resolution.

The primary space-based method for characterizing stratospheric aerosol is the occultation method, in which photons emitted by a bright source of radiation are transmitted through the atmosphere as the source rises or sets through the atmosphere (as viewed by the instrument). These line-of-sight transmission profiles are then converted into vertical profiles of extinction coefficient for the various atmospheric constituents, including aerosol extinction coefficient. The transmission measurement is essentially “self-calibrating,” and this fact (combined with the high signal-to-noise ratio provided by a bright radiation source) leads to retrievals with high precision and accuracy. However, the locations and frequency of occultation events are entirely determined by the orbit of the instrument, which limits the coverage that can be achieved.

Most occultation measurements for stratospheric aerosol studies use solar occultation (with the sun as the source of photons). Examples of this technique include the Stratospheric Aerosol Measurement (SAM II; McCormick et al., 1982), the Stratospheric Aerosol and Gas Experiment (SAGE II; Chu et al., 1989) and SAGE III (Thomason and Taha, 2003), Polar Ozone and Aerosol Measurement (POAM II and III; Lumpe et al., 2002), the Halogen Occultation Experiment (HALOE; Hervig et al., 1996), the Improved Limb Atmospheric Spectrometer (ILAS I and II; Hayashida et al., 2000; Saitoh et al., 2006), and the Measurement of Aerosol Extinction in the Stratosphere and Troposphere Retrieved by Occultation (MAESTRO; McElroy et al., 2007; Sioris et al., 2010). Aerosol extinction profiles were also measured by the stellar occultation instrument Global Ozone Monitoring by Occultation of Stars (GOMOS; Vanhellemont et al., 2010). The particular solar occultation instrument used in this study (the SAGE III instrument on the ISS) is described further in Sect. 3.1.

In recent years, several limb scattering (LS) measurements have become available, including SAGE III limb (Rault and Taha, 2007), Optical Spectrograph and InfraRed Imager System (OSIRIS; Llewellyn et al., 2004; Bourassa et al., 2007), SpectroMeter for Atmospheric CHartographY (SCIAMACHY; Taha et al., 2011; von Savigny et al., 2015; Malinina et al., 2018) and OMPS LP (Gorkavyi et al., 2013; Rault and Loughman, 2013). These instruments measure the radiance scattered by the limb of the atmosphere. The strength of the stratospheric aerosol signal in these radiances depends on several aerosol properties, including the aerosol refractive index, shape, size and number density. These properties affect both the aerosol phase function and aerosol extinction coefficient, both of which influence the aerosol signal present in the measured limb radiance.

This study evaluates the OMPS LP V2.0 aerosol extinction retrieval algorithm, which estimates the aerosol extinction coefficient based on the measured LS radiance as described in Sect. 2.2. The resulting extinction coefficients are compared to the OSIRIS aerosol extinction product, which is described further in Sect. 3.2.

Spaceborne backscatter lidar measurements by Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) are also used to explore the stratospheric aerosol layer (Thomason et al., 2007a; Vernier et al., 2009; Kar et al., 2019). The description given in Sect. 1.2.1 for ground-based lidars generally applies to the spaceborne lidars as well. The main difference is that an orbiting lidar provides the additional advantage of mobility (near global coverage in the case of CALIPSO). The spaceborne CALIPSO lidar instrument is used in this study and is described further in Sect. 3.3.

In 2017, accurate solar occultation measurements of stratospheric aerosols resumed after the deployment of SAGE III on the International Space Station (ISS; Cisewski et al., 2014). Combined with the ongoing OSIRIS and CALIPSO missions, we now have coincident stratospheric aerosol measurements from several space-based platforms. The structure of this paper is as follows: in Sect. 2 we provide a brief description of the OMPS LP instrument and V2.0 algorithm changes. The correlative satellite aerosol measurements are described further in Sect. 3. Section 4 describes the validation methodology. The comparison results are shown in Sect. 5, followed by conclusions in Sect. 6.

Unlike solar occultation, limb scattering retrievals require complex forward model calculations, and the aerosol retrieval in particular requires assumptions of aerosol refractive index and size parameters. Previous versions of the OMPS LP aerosol retrieval algorithm were mainly designed for minimizing the errors of ozone profiles and level-1 radiance diagnostics. The V1.0 and V1.5 algorithms use OMPS LP radiance measurements, Im(λ,h), for a range of tangent heights, h, at a single wavelength (λ=675 nm) to estimate the aerosol extinction profile. This wavelength was selected to be near the Chappuis band that is used to retrieve the ozone profile in the visible region, for which the OMPS LP radiances are best characterized. The algorithm assesses the measurements by comparison to two analogous sets of calculated radiances, Ic(λ,h) and Ic0(λ,h). These calculated radiance profiles are generated by the Gauss–Seidel limb scattering (GSLS) radiative transfer model (RTM) (Loughman et al., 2004) for the same viewing and solar illumination conditions that existed when Im(λ,h) was measured. The model atmospheres used (described further below) are identical for these two calculations, with one exception: in the case of Ic0(λ,h), the model atmosphere contains no aerosols, while the Ic(λ,h) model atmosphere contains the first-guess aerosol profile.

The model atmosphere consists of static atmospheric temperature and pressure profiles derived from the operational geopotential height product provided by the NASA Global Modeling and Assimilation Office (GMAO). The algorithm uses the McPeters and Labow (2012) ozone climatology and the PRATMO photochemical box model NO2 climatology (McLinden et al., 2000) to define the model atmosphere. The first-guess aerosol extinction profile, x0, is defined based on a single SAGE climatological profile. Aerosols are assumed to consist of spherical liquid sulfate particles (75 % H2SO4) with index of refraction m=1.448+0i (Yue and Deepak, 1983; Wang et al., 1996). In the V1.0 algorithm, the aerosol size distribution (ASD) is assumed to be a bi-modal log-normal distribution (Loughman et al., 2018); this was updated to a gamma distribution in V1.5 (Chen et al., 2018). Mie scattering theory is used to calculate the aerosol phase function based on the assumed ASD and optical properties.

The Earth's surface is modeled as a Lambertian surface (for which a fraction, R, of the incident downward radiation is reflected as isotropic, unpolarized upward radiance field at each point). The value of R is determined by requiring that Ic0(λ,h)=Im(λ,h) at h=40.5 km (Loughman et al., 2018). An approximate ozone correction is also applied to the model radiances to correct for possible ozone error, as described in Sect. 4.3 of Loughman et al. (2018). To reduce the sensitivity of the algorithm to a variety of interfering factors, the radiances are normalized with respect to tangent height h. The measured altitude-normalized radiance (ANR) is defined as ρm(λ,h)=Im(λ,h)/Im(λ,hn), with hn, namely the normalization tangent height, set to a value of 40.5 km in the V1.0 and V1.5 algorithms. The value of hn is generally selected as a compromise between two competing interests. It should be as high as possible to minimize the atmospheric aerosol extinction at hn, but not so high that the radiance at hn is poorly characterized (due to residual stray light contamination, low signal-to-noise ratio, etc.). Analogous expressions define ρc(λ,h) and ρc0(λ,h) based on the calculated radiance profiles.

As a final step, the ANR values are combined to produce the aerosol scattering index (ASI), which serves as the measurement vector y in the retrieval. The measured ASI is defined as ym(λ,h)=[ρm(λ,h)-ρc0(λ,h)]/ρm(λ,h), with similar definitions for yc(λ,h) and yc0(λ,h). Since the V1.0 and V1.5 algorithms use a single wavelength, this notation can be abbreviated to ym(λ,hi)=yim, with a similar abbreviation yc(λ,hi)=yin used to represent the ASI calculated based on the model atmosphere after n iterations of the retrieval algorithm.

The V1.0 and V1.5 algorithms use the Chahine nonlinear relaxation method (Chahine, 1970) to derive the aerosol extinction coefficient (which represents the state vector, x) based on the measurement vector y defined above. The state vector is updated iteratively as shown in Eq. (1): 1xin+1=xinyimyin=xinfin. In this expression, xin+1 is the state vector at altitude zi, and n+1 is the number of iterations. The measurement vector yim is the ASI at tangent height h=zi, while yin is the calculated ASI based on the extinction profile corresponding to the nth iteration. As shown in Eq. (1), the retrieval creates the updated aerosol extinction coefficient estimate xin+1 by multiplying the previous estimate, xin, by the factor fin. The V1.0 algorithm constrains the value of fin to lie between 0.2 and 2.0 and sets the number of iterations to N=3. These constraints were primarily motivated by caution in the early stages of developing the aerosol extinction retrieval algorithm, when the stability of the retrieval was relatively untested. The V1.5 algorithm relaxed these constraints somewhat, using N=4 and allowing fin values between 0.2 and 3.0.

Since the limb scattering radiances at visible and near-infrared wavelengths are very sensitive to aerosol properties, the V2.0 OMPS LP aerosol algorithm is modified to include multiple wavelengths in this spectral region, similar to the SAGE III aerosol channels (Thomason and Taha, 2003). The V2.0 algorithm uses OMPS LP measurements at wavelengths 510, 600, 675, 745, 869 and 997 nm, selected to minimize the effect of gaseous absorption, with the exception of 600 nm, which will be used primarily for diagnostics. Each wavelength is retrieved independently, as described in the preceding section leading to Eq. (1). Taha et al. (2011) showed that, because of its strong weighting function or Jacobian matrix, retrieving aerosol profiles at longer wavelengths can improve the quality of the profile in the Southern Hemisphere, where OMPS LP observes backscattered radiation, and extend the retrieval further down in altitude. The Jacobian matrix quantifies the changes in the radiance with respect to the aerosol extinction. Multiple wavelength aerosol measurements can also provide limited information about aerosol particle size and can be used to identify cloud presence. Notice that the 997 nm radiance measurements are only available after 26 November 2013.

The assumed ASD is the same in V2.0 as in V1.5, but the single first-guess aerosol extinction profile has been replaced by a first-guess climatology that varies with wavelength, latitude and season, again based on the SAGE aerosol data record. The V2.0 algorithm further relaxes the constraints that were previously applied to the Chahine iteration results: N=5 and fin has an upper bound of 10.0 and no lower bound. The V2.0 algorithm also checks for convergence after each iteration, rather than always performing the stated number of iterations: iterations end when the retrieved aerosol extinction changes by<2 % at 20 km or when it reaches the maximum number of iterations. The planned V2.1 release next year will use modified convergence criteria that check for multiple altitudes.

Limb scattering instruments such as OSIRIS, SCIAMACHY and OMPS LP suffer from increased stray light at increasing wavelength and altitude due in part to diminishing scattered signal (Jaross et al., 2014; Rieger et al., 2019). To reduce the stray light effect on the retrieval at longer wavelengths, hn was lowered to 38.5 km in V2.0 (from the 40.5 km value used in previous versions). The GSLS radiative transfer model used in the V2.0 algorithm was also updated as described by Loughman et al. (2015). The main improvement associated with this change involves the use of several zeniths to calculate the multiple-scattering source function along the limb line of sight, which improves the radiance calculations near the terminator. Unlike the V1.0 and V1.5 algorithms, the V2.0 GSLS model also includes refraction in the line-of-sight calculation. The V2.0 algorithm also excludes polarization (which had been included in the V1.5 radiance calculations). The exclusion of polarization is primarily done for speed purposes: scalar (unpolarized) radiance calculations are considerably faster than their vector (polarized) counterparts, and the resulting change in ρc(λ,h)is very small. Recent calculations performed for a RTM comparison project (Zawada et al., 2020) allow the ρc(λ,h) values computed by the scalar and vector versions of GSLS to be compared, for a variety of atmospheres and illumination conditions. For the relevant wavelengths (500 nm and greater), these values agree to within 1 % or better at 20 km and within 2 % or better at 10 km.

Figures 1 and 2 illustrate the contrasting effect of the scattering angle on the measurement vector and subsequent retrieved aerosol extinction profiles at different wavelengths. Figure 1 shows that at scattering angle 154∘, wavelengths shorter than 745 nm have poor sensitivity to aerosol, which limits the accuracy and altitude range of the OMPS LP SH aerosol retrieval. In contrast, the longer wavelengths show high sensitivity to aerosol, thus improving the retrieval significantly at lower altitudes. Notice that the cloud at 10.5 km is only observed by the longer wavelengths. Figure 2 illustrates the strong sensitivity of all six wavelengths to aerosol when the scattering angle is small. In the NH, the OMPS LP measurement vector for all wavelengths is positive for all altitudes, and the aerosol retrieval quality does not vary significantly with wavelength. Notice that all retrieved aerosol wavelengths can detect the cloud layer evident as enhanced extinction near 10.5 km.

To evaluate the quality of OMPS LP aerosol retrieval with SAGE III, we use a coincidence criterion of same calendar day measurements, Δlat=±3∘ and Δlong=±10∘, which is selected to minimize the effect of spatial and temporal differences between the two instruments. Coincidence pairs are averaged over 10 (Figs. 5, 7 and 8) or 40∘ (Fig. 6) latitudes bins covering the first 2 years of SAGE III/ISS measurements.

Figure 5 depicts the mean OMPS LP and SAGE III/ISS aerosol extinction profiles for the set of coincident measurements, binned in 10∘ latitude bins, shown at selected altitudes for four different wavelengths. In general, OMPS LP and SAGE III extinction values show similar latitudinal distribution and are well within 20 % of each other for most altitudes, with 869 nm showing the best agreement of better than 10 %. At 18.5 km in the SH tropics, OMPS aerosol is systemically higher than SAGE III, mostly influenced by cloud presence at lower altitudes. The OMPS 675 nm extinction shows a negative bias at 18.5 km that increases with increasing latitudes in the SH, where OMPS LP observes the backscatter solar radiation, and the attenuation of Rayleigh radiation is substantial below 20 km at 675 nm.

Figure 6 is a summary plot of the mean difference between OMPS and SAGE III coincidences for wavelengths 510, 600, 675, 745, 869 and 997 nm. In general, 869 nm is the best OMPS-retrieved wavelength relative to SAGE III with differences of 5 % or less for most altitudes and latitudes. Other wavelengths agree with SAGE III to within 10 %. Exceptions to this occur at high altitudes (above ≈28 km) where the aerosol loading is minimal, and near the tropopause, which is affected by cloud contamination. The 510 and 600 nm OMPS extinction values have a slightly larger bias of 20 % in the tropics. This is due to the ozone interference in both OMPS and SAGE III 600 nm aerosol retrievals. The 997 nm OMPS extinction values have a systematic bias of -10 % between 60∘ S and 20∘ N, caused by stray light contamination in the OMPS measurements. Unlike the other wavelengths, the 997 nm laboratory characterization is poor, and its stray light correction, therefore, has lower quality (Jaross et al., 2014). In the SH, 510 nm shows a large positive bias relative to SAGE III below 18 km. This is an artifact in the OMPS retrieval algorithm, which often results in noisy and large extinction values when the measurement vector is too small relative to gaseous absorption and Rayleigh scattering (see Fig. 12).

It is worth noting that the best agreement between OMPS and SAGE is found in the NH, where OMPS observes in forward scattering and the weighting function is strong for all wavelengths. In that region, the agreement is mostly within 5 % for an altitude between 14 and 22 km. Above 24 km, the observed biases for 510, 600, 675 and, to some extent, 745, 869 and 997 nm gradually increase with altitude, mainly caused by instrument noise and errors under low-aerosol conditions, although OMPS assumed aerosol size model uncertainty also contributes to the larger differences.

Figure 7 summarizes the quality of OMPS LP aerosol extinction at six retrieved wavelengths, showing the zonally averaged mean differences between OMPS LP and SAGE III aerosol (in percent). Below 25 km, the differences between OMPS LP and SAGE III are largely driven by OMPS weighting functions or Jacobians. The weaker Jacobians for short wavelengths under backscatter conditions in the SH and below 20 km lead to limited accuracy, while stronger Jacobians at the longer wavelengths improve its accuracy significantly (Taha et al., 2011; Rieger et al., 2019). Overall, the shorter wavelengths (510, 600 and 675 nm) are biased low against SAGE III with a difference greater than 25 % below 20 km in the SH. In addition, these short wavelengths exhibit pronounced large aerosol in the tropics below 20 km caused by the algorithm's reduced accuracy when the measurement vector is very small. The agreement is well within 25 % at the altitude range 20–25 km and better in the NH. Above 25 km, the comparison between the two instruments is poor, caused by either SAGE III ozone correction errors and/or OMPS reduced sensitivity to aerosol at these short wavelengths. The best agreement between OMPS and SAGE III can be seen at 869 and 997 nm, where they are mostly within 10 % of each other for all altitudes and latitudes. The 745 nm OMPS extinction agrees with SAGE to within 15 % everywhere except for the SH tropics below 18 km.

The standard deviation shown in Fig. 8 is influenced by several factors: OMPS LP uncertainties such as measurement noise, forward model errors and retrieval algorithm sensitivities, in addition to SAGE III/ISS's own uncertainty and atmospheric variability. In general, the standard deviation is 15 %–20 % for altitudes that show good agreement with SAGE III (Fig. 7). The large standard deviation of ≈50 % at high altitude is due to instrument noise and low aerosol loading. Below 20 km, the standard deviations for the shorter wavelengths increase to 50 %, caused by the OMPS LP reduced accuracy. In the UT–LS, the standard deviation is >50 % due to larger dynamical variability, especially during periods when dispersal of plumes due to volcanic eruptions and other events causes longitudinal variations, as well as cloud interference.

Based on the comparison with SAGE III, we can estimate the OMPS aerosol retrieval relative accuracy to be ≈10 % for 745, 869 and 997 nm in the stratosphere and 20 % for 675 nm above 20 km and in the NH. The 510 and 600 nm retrievals have limited accuracy in the SH and 25 % relative accuracy at altitudes between 20–26 km and in the NH. Furthermore, the standard deviation can be used to determine the retrieval relative precision, which can be estimated to be better than 15 % for the longer wavelengths and close to 20 % for wavelengths less than 745 nm. The real precision is probably better than the quoted values since the calculated standard deviation includes atmospheric variability and both instruments' biases, none of which was removed (Rault and Taha, 2007; Wang et al., 2020).

In order to investigate the OMPS LP retrieval performance under different seasonal or geographical conditions, we compare the OMPS LP monthly zonal mean time series with the SAGE III/ISS monthly zonal mean time series for four wavelengths at three different altitudes. The comparison is also divided into three different regions, the SH (Fig. 9), tropics (Fig. 10) and NH (Fig. 11). In general, the agreement between the two instruments in the SH is mostly within 10 %–20 %. The 675 nm extinction at 20.5 km is a notable exception, as the OMPS LP aerosol extinction values drop significantly when the SSA is greater than 145∘, and the attenuation of Rayleigh scattering below 20 km becomes significant. This behavior appears as an apparent seasonal pattern, in which the OMPS LP/SAGE III difference becomes much larger during SH winter months. It is therefore recommended that OMPS aerosol measurements at λ≤675 nm should be excluded when SSA is greater than 145∘ below 21 km.

A similar agreement is found in the tropics, at or above 20.5 km, with the exception of the first few months following the Aoba volcanic eruption in July 2018, where OMPS LP initially reported higher aerosol extinction than SAGE III. This might be caused by the different coverage and frequency of measurements for each instrument. Still, the difference between the two measurements was mostly within 20 % in the aftermath of this volcanic eruption. At 18.5 km, the difference is often greater than 20 %, reaching more than 60 % following the subsidence of the volcanic plume. The reason for such large differences is unclear as OMPS LP still shows elevated aerosol levels when SAGE III measurements indicate that the aerosol values are back to pre-eruption levels, although spatial variability and spatial resolutions can contribute to such large differences. SAGE III aerosol extinction profiles are produced on a 0.5 km grid with an estimated vertical resolution of 0.7 km (SAGE III ATBD, 2002; Thomason et al., 2010) while OMPS LP vertical sampling is 1.0 km with an instantaneous resolution of 1.5 km (Jaross et al., 2014). Bourassa et al. (2019) compared nearby OMPS LP and SAGE III/ISS aerosol profiles following the aftermath of the British Columbia fires in 2017. They showed that both instruments have very similar layered vertical structure and magnitude. However, they noted that some differences in layer height and magnitude can be expected from differing vertical resolutions. Another possible explanation is that OMPS LP cloud clearing can be incomplete, and residual cloud contamination can contribute to the large differences near the tropopause. The best agreement between the two instruments can be found at 25.5 km, well within 10 %.

In the NH, all OMPS LP wavelengths show similar robust agreement to SAGE III, mostly within 10 %, since OMPS LP observes in the forward scattering and all wavelengths are strongly sensitive to aerosol (Fig. 11). A notable exception is the first couple of months of the August 2017 Canadian PyroCb period and the June 2019 Raikoke eruption, when the aerosol loading was very high and spatially inhomogeneous. Spatial inhomogeneity also caused a large bias after 2018 following the sharp drop in aerosol extinction at 25.5 km. While the OMPS assumed ASD model may contribute to the larger differences at 25.5 km, instrument noise and calibration errors are also more significant under low-aerosol conditions. OMPS LP 997 nm is affected by stray light contamination at the normalization altitudes in the NH high latitudes, which might explain the negative bias during 2019. On the other hand, SAGE III ozone correction uncertainty near the Chappuis band can cause a dip in SAGE III aerosol extinction measurements at 676 nm. In particular, the SAGE III 676 nm values at 25.5 km are either zero or negative during 2019 when the measured aerosol is at its lowest levels in the NH during the short lifetime of ISS SAGE III.