| 06 Aug 2026
Correction for above-clouds aerosols in geostationary satellite cloud property retrievals
Abstract. The high-temporal-resolution observations from Spinning Enhanced Visible and InfraRed Imagers (SEVIRI) and its successor Flexible Combined Imager (FCI) of cloud optical thickness (COT) and cloud particle effective radius (CER), suffer from biases due to absorption by aerosols, when smoke is overlying the cloud scene. This is often the case over the south-east Atlantic Ocean (SEAO), which is an important part of the field of view of these instruments, which are onboard the geostationary Meteosat satellites series, stationed around 0° latitude and longitude. The SEAO is an important area to study the impact of smoke on clouds, because the semi-permanent stratocumulus cloud decks over the dark ocean effectively reflect solar radiation and have a strong cooling effect on the Earth climate system. During the annual dry seasons in Africa, smoke from vegetation fires is advected over the clouds over the SEAO and mixes with them, making this region a natural laboratory to study aerosol-cloud-radiation interactions. In this paper, the biases in the COT and CER are quantified for typical smoke over cloud scenes, using an adaptation of the traditional cloud retrieval algorithms, including an extra channel in the usual bispectral minimisation procedure. This allows the simultaneous retrieval of COT, CER and above-cloud aerosol optical thickness (ACA AOT), which has been shown in several papers. The impact of smoke on the SEVIRI retrievals was investigated during July – October 2017 for a small region of the SEAO in the satellite field of view. The SEVIRI COT and CER were increased by a factor of 1.9 and 1.4, respectively, as a result of accounting for overlying smoke during two days with an average ACA AOT (550 nm) of 0.82. A similar event in 2025 during one day saw an increase in FCI COT and CER of 2.2 and 1.3, respectively, for an average ACA AOT of 0.95. Furthermore, the distribution of COT changed significantly when accounting for the overlying aerosols, showing a much wider distribution of COT values.
egusphere-2026-4481
Manuscript title: Correction for above-clouds aerosols in geostationary satellite cloud property retrievals
Authors: Martin de Graaf et al.
General Comments:
Authors, in the submitted paper, quantifies the biases in the retrieved cloud optical thickness (COT) and cloud effective radius (CER) produced by the carbonaceous smoke aerosols overlaying the low-level stratocumulus cloud deck in the southeastern Atlantic Ocean—a natural laboratory for studying effects of absorbing aerosols on clouds. Based on Peers et al (2019) above-cloud aerosol (ACA) algorithm and aerosol model, ACA aerosol optical thickness (AOT), aerosol-corrected COT and CRE were derived using the time-resolved, geostationary METEOSAT second generation (MSG) SEVIRI observations over the region during July-October 2017. When compared against the standard CLASS-3 COT retrieval dataset not corrected for aerosol effects, the ACA COT and CER retrievals demonstrated by the authors were found to be higher by factors of 1.9 and 1.4, respectively. An extension of the same technique to recent third generation sensor FCI for an event in 2025 for one day resulted an increase in FCI COT and CER of 2.2 and 1.3, respectively.
While the authors demonstrated the effects of aerosol attenuation, primarily which are primarily driven by the absorption effects, on the cloud retrievals, the submitted work miss citing and discussing similar analyses of aerosol effects on cloud retrievals in the UV (Jethva et al., 2018, 2024) and visible (Meyer et al. 2015) domains.
Authors are suggested to also present theoretical calculations, based upon the RT simulations/LUT, of biases in COT retrievals as a bifunction of ACA AOT and COT.
The OMI-OMAERUV Collection 4 and NASA TROPOMAER aerosol datasets also contain ACA retrievals globally. Bringing one of these datasets for the SEAO region in the present work for relative comparison would further elaborate the findings discussed in the paper.
The Absorbing Aerosol Index (AAI) datasets synchronized from TOMS, GOME-1, SCIAMACHY, OMI, GOME-2A, GOME-2B, and GOME-2C sensors has been shown in the paper (Figure 11). However, the details, at least at the preliminary level, of such synchronization that also accounts for wavelength and footprint differences, is missing.
Detailed comments on specific sections and text are included below with this review report.
Overall, the paper is well-written and organized, although the language can be improved. The theme of the presented work very well fit into the scope of the journal and also important for cloud remote sensing in the presence of absorbing aerosols. Given some additional analysis suggested here, the paper is recommended as “major revision”.
Thanks for this review opportunity.
Specific Comments:
Abstract:
Page 1
line 3: Not just smoke but dust over clouds too.
Line 4-5: “which are onboard the geostationary Meteosat satellites series, stationed around 0-degree latitude and longitude”. This sentence looks out of place.
Introduction:
Page 2, line 40: “During northern hemisphere monsoonal and dry season in southern Africa…”
Page 3, line 59: There are several papers quantified the effects of absorbing smoke layer over clouds in SEAO, such as, Jethva et al. (2013, 2018), Meyer et al. (2015), and few others. Citing a couple of these papers here would further strengthen this statement, since this is the center topic of the presented work.
Section 2.2
Page 4, line 94-95: this statement needs a reference.
Section 2.3
Page 4, lines 101-103: CALIOP-CALIPSO vertical curtain plots, over the southeastern Atlantic Ocean, often show aerosol layers residing between 2-4 km (in some instances, the aerosol layer also touch the cloud-top) and cloud layer between ~800-1500 meters.
Page 5, lines 129-131: This is the best way to quantify the errors in the cloud retrievals eliminating remaining biases between the CLASS-3 and CPP.
Table 1: Aerosol particle size distribution (PSD) parameters look reasonable, based on the in-land AERONET stations (Mongu, Mongu_Inn). However, assuming wavelength-independent imaginary part (0.029) of the refractive index, representative of black carbon dominated particles, seems to be inappropriate for smoke aerosols. AERONET ground inversions show spectral dependence of the imaginary part from 440 nm to 675 nm, with minimal spectral effects at longer wavelengths. What value of Extinction Angstrom Exponent, describing spectral dependence of AOT, was assumed in the simulations?
So, from Figure 2, it is interpreted that first ACA AOT is derived using 0.824 and 0.64 microns (left panel). In the next step, the derived ACA AOT was used to estimate the errors in the retrieved COT and CER caused by smoke aerosols. Please confirm.
Section 3.1
Figure 3. It is worthwhile to combine Figure 1 and Figure 3 to facilitate a direct comparison between the original CLASS-3 COT-CER retrievals with that retrieved after accounting for aerosols over clouds.
Section 3.2
Figure 4. Can author plot percent different plot of COT, i.e., (ACA-CPP)/CPP.
Figure 5. What could be the reason for a sharp increase in ACA AOT at higher solar zenith angle past 15:00 UTC? Is it due to the limitations in the RT simulations/look-up tables at extreme geometries? Does RT simulation correctly account for the sphericity of the atmosphere. This should be discussed here.
Figure 6: It is understood that ACA COT is retrieved at 0.84 microns and assumed to be spectrally neutral, which is a fair assumption in the VIS-NIR spectral domain. This assumption allows us to directly compare 0.84 microns ACA COT retrievals with CLASS-3 standard (no aerosol correction) COT dataset. The “apparent” COT retrieved without aerosols being accounted for, is a function of wavelength, since the aerosol attenuation led by absorption over clouds exhibits spectral dependence, i.e., higher absorption (hence lower COT) at shorter wavelength and lower absorption (hence relatively higher COT) at longer wavelength. I wonder if author can directly compare their “apparent” COT using retrieval domain presented in Figure 2 assuming AOT=0.0 and compare it with the true COT retrieved after aerosol correction. It is expected that the difference between ACA COT and non-corrected COT retrieved at 0.84 microns would be lower than that calculated at 0.64 microns. It is worth to add the suggested analysis in Figure 6.
Section 3.2.1
Page 10, lines 188-189: While the assumed imaginary part of the refractive index a representative value, it exhibits variability with the same month as well as seasonally (Eck et al., 2013).
Figure 7: the y-axis metric (CACA-CCPP)/(CACA+CCPP) is a bit hard to interpret. Instead, a simple percent change (CACA-CCPP)/CCPP) * 100 would be a straightforward way to understand the effects of AOD on COT derivation. Furthermore, the dependence of the COT correction as a function of aerosol absorption optical thickness (AAOT) can also be examined. AAOT can be calculated as the retrieved AOT times (1-assumed SSA).
Figure 7 demonstrates that the effects of absorbing smoke aerosols on COT retrievals do not strictly follow a linear relation. Instead, both are related in, what appear to be, quadratic way. Authors are encouraged to refer to similar analysis published in Figure 11 of Jethva et al. (2018) and Figure 11 Jethva et al. (2024), where the percent difference in COT, caused by aerosol absorption, was parameterized as a bifunction of abs. AOT and COT. For instance, for a fixed value of AAOT over clouds, the difference between corrected and non-corrected COT depends on the true COT underneath the aerosol layer. Authors can carry out similar analysis here to show the dual dependence of COT differences on ACA AOT and ACA COT.
Section 4
Line 234-235: AAI is also a strongly dependent on the spectral dependence of absorption.
Figure 10: Alternatively, these data can also be represented in running-mean fashion. Just a suggestion. A good agreement between ACA AOT and CAMS AOT dataset indicates that most part of the transported aerosol load in vertical column resided over the clouds. The continental biomass burning source region is elevated at about 1 km terrain height. Under the prevailing circulation, smoke particles are further carried over in westward direction and over semi-permanent cloud deck over the southeastern Atlantic Ocean.
OMI’s latest Collection 4 OMAERUV aerosol product now includes the pixel-level direct radiative effects for above-cloud aerosol scenes. The DRE is calculated by contrasting the TOA fluxes estimated for non-corrected and aerosol-corrected COT. Authors are encouraged to bring this new dataset in Figure 10 to see how it compares with the estimates from the OMI-MODIS technique that the first author has pioneered. The OMAERUV product can be freely accessible from NASA’s EarthData at https://www.earthdata.nasa.gov/data/catalog/ges-disc-omaeruv-004.
Section 4.2
The AAI has been calculated using a pair of wavelengths in the near-UV region. Different sensors mentioned here carry different pair of wavelengths, rendering the same AAI quantity but tailored to distinct wavelength pair. The caption of Figure 11 states that AAI from this series of UV-capable sensors was synchronized to account for wavelength and footprint differences. Does it mean that AAI dataset from these sensors were converted to a common, reference wavelength pair in the UV? A brief description on the AAI synchronization method is needed here.
Line 269-270: I would suggest rewording this statement, such as “In very bright scenes on the other hand, like over clouds or snow- or ice-covered surfaces, the scattering by aerosols is negligible compared to the reflected light from the underlying background surface. Under such scenario, the absorption of the background-reflected radiation by aerosols becomes very pronounced”.
Section 5
Figure 13: The results derived from FCI are mostly consistent to those obtained from SEVIRI. Stark differences in ACA COT between the near-noon and early morning-late afternoon retrievals are concerning. A 4- to 8-fold increase in COT seems unrealistic. The diurnal variation of cloud cover is well-captured from time-resolved SEVIRI and DSCOVR-EPIC (and now with drifting Terra-Aqua MODIS too) over the southeastern Atlantic Ocean. These observations show increased regional cloud fraction during the morning and evening hours relative to the near-noon timeframes. However, this may not translate into multifold increase in COT. Such unexpected COT behavior is likely caused by the limitation of plane-parallel RT simulations, which is noted earlier in the paper.
Meyer et al. (2015) noted small differences in the MODIS-based CER retrievals for above-cloud aerosols scenes in the same region. However, the MODIS CER retrievals used 2.1 microns instead of 1.6 microns of SEVIRI/FCI employed in this study. Author should include a discussion on these differences and the results of Meyer et al. (2015).
Section 6
Line 354: While it is true that 0.44 microns has stronger absorption effects than at 0.64 microns, the uncertainty in ACA retrievals arising from the aerosol model (SSA, AE, AAE, PSD) still remain.
Last paragraph: Similar to the OMI-OMAERUV Collection 4 retrieval dataset, NASA TROPOMAER UV aerosol algorithm also produces above-cloud aerosol retrievals, including ACA AOT, ACA COT, and apparent (non-aerosol corrected) COT at 388 nm. Authors are encouraged to look at and use this dataset for their future intended study.
One last comment, which should have been added earlier in this review report: In addition to comparing the ACA COT retrievals against CLASS-3 non-corrected COT dataset, authors should also conduct theoretical estimates of the difference in COT retrievals (i.e., aerosol corrected minus non-corrected) using RT calculations or LUT shown in Figure 2. The suggested analysis can be carried out in two ways by treating 1) 0.84 microns on x-axis to retrieve COT, and 2) 0.64 microns observations on x-axis to retrieve COT. Both exercises should result in different errors in COT, with 0.64 microns COT bias is expected to be higher than that at 0.84 microns due to stronger absorption at former wavelength than at longer one. By performing such analysis, authors can establish the theoretical framework for estimating biases in COT as a function of AOT and underlying true COT, thereby adding further value to the paper.
Although the SEAO is a regional hotspot for observing and studying absorbing smoke aerosols over clouds, other regions, such as tropical Atlantic Ocean (Saharan dust outflow over clouds), Arabian Sea, Southeast Asia (agricultural burning smoke over thick clouds in southern China and adjacent ocean), and North American wildfire smoke transport in North Atlantic, also frequently encounter aerosol-cloud overlap situation on seasonal scale. The cloud remote sensing, and resultant aerosol-cloud interactions, are also important in these regions. Including a brief discussion of these regions will further increase the scientific value of the presented work.