the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 04 Sep 2026
Propagation of CAMS Aerosol Uncertainty to Clear-sky Solar Irradiance Estimates
Abstract. Atmospheric aerosols strongly influence surface solar radiation (SSR) through scattering and absorption processes, introducing substantial uncertainties into SSR estimates. Despite the increasing use of atmospheric reanalysis products for solar resource assessment, the propagation of aerosol-related uncertainty into SSR calculations remains insufficiently quantified. In this study, we derive aerosol optical depth (AOD) uncertainties for the fourth-generation ECMWF Atmospheric Composition Reanalysis 4 (EAC4) and investigate their propagation into clear-sky estimates of global horizontal irradiance (GHI) and direct normal irradiance (DNI) over the 60° S–60° N domain for the 2003–2024 period. Prognostic AOD uncertainties are calculated using collocated EAC4 and Aerosol Robotic Network (AERONET) observations, while the resulting global mean relative AOD uncertainty is approximately 47 %. Radiative kernels are subsequently used to quantify the sensitivity of clear-sky SSR to AOD perturbations, enabling the propagation of AOD uncertainty into GHI and DNI estimates. The results reveal pronounced spatial and seasonal variability, with the largest propagated uncertainties occurring over regions affected by persistent anthropogenic pollution, biomass burning, and desert dust. Relative GHI uncertainties generally remain below 5–6 %, whereas relative DNI uncertainties frequently exceed 18 % over major aerosol source regions and reach 25–35 % seasonally under elevated aerosol loading. Overall, the proposed framework provides a computationally efficient methodology for generating uncertainty-aware solar radiation products and supports future investigations of aerosol-driven dimming and brightening, as well as long-term solar energy resource assessments under changing atmospheric conditions.
- Preprint
(2294 KB) - Metadata XML
-
Supplement
(1129 KB) - BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2026-4413', Anonymous Referee #2, 25 Sep 2026
-
RC2: 'Comment on egusphere-2026-4413', Anonymous Referee #1, 08 Oct 2026
General comments
This manuscript presents a global assessment of the uncertainty in clear-sky surface solar radiation associated with uncertainties in aerosol optical depth (AOD) from the ECMWF Atmospheric Composition Reanalysis 4 (EAC4). The authors derive an empirical AOD expected-error relationship from collocated EAC4 and AERONET observations, propagate the resulting AOD uncertainty to global horizontal irradiance (GHI) and direct normal irradiance (DNI) using radiative kernels derived from radiative transfer simulations, and evaluate the resulting clear-sky irradiance estimates against BSRN observations. The analysis covers a relatively long period (2003–2024) and provides global and seasonal distributions, as well as an analysis of selected high-AOD AERONET sites.
The topic is relevant to both atmospheric radiation and solar-resource assessment, and the manuscript addresses an aspect that has received comparatively less attention than either the validation of surface solar radiation products or the quantification of aerosol radiative effects. In particular, the distinction between the sensitivity of GHI and DNI to AOD uncertainty is potentially valuable. The use of a computationally efficient radiative kernel/LUT approach also provides a potentially useful framework for extending uncertainty calculations to global datasets.
The manuscript is generally well structured, and I consider the scientific question worthwhile and the proposed framework potentially useful. Some aspects of the methodology and interpretation, however, would benefit from clarification and revision to further strengthen the work. The main points are detailed below.
Specific comments
- The manuscript derives an expected-error relationship from the differences between EAC4 and AERONET AOD and subsequently refers to this quantity as “EAC4 AOD uncertainty”. The authors should discuss more explicitly the distinction between an empirical expected error relative to AERONET and the intrinsic uncertainty of the EAC4 reanalysis AOD. The EAC4–AERONET differences can contain contributions from AERONET measurement uncertainty, representativeness differences, spatial and temporal collocation errors, systematic model errors, and errors in aerosol speciation and vertical distribution. Therefore, the terminology should be carefully defined throughout the manuscript.
The manuscript should also clarify whether AERONET observations are assimilated, directly or indirectly, into EAC4. If there is any dependence between the two datasets, the degree of independence of the comparison used to derive the expected error should be discussed.
- Equation (4) in the manuscript removes a fixed AERONET uncertainty of 0.01 in quadrature. This assumes, among other things, that the quoted AERONET uncertainty can be treated as a one-standard-deviation random uncertainty and that it is independent of the EAC4 error. The use of a constant value should also be justified given the different wavelengths, aerosol regimes, stations, and loading conditions involved. Furthermore, the AERONET AOD is not directly available at 550 nm and is obtained through spectral interpolation. The uncertainty introduced by this interpolation should either be quantified or explicitly justified as negligible relative to the EAC4–AERONET differences.
- The temporal and spatial collocation procedure between EAC4 and AERONET should be described in sufficient detail to make the analysis reproducible. In particular, the authors should specify the temporal matching window and whether the EAC4 value is taken from the nearest grid cell or obtained by spatial interpolation. This is especially important given the 0.75° EAC4 grid and the point-like nature of AERONET observations. Spatial representativeness errors may be significant in regions with strong aerosol gradients.
- The choice of 500 observations per AOD bin and the use of a linear relationship between expected error and EAC4 AOD should be justified. The authors should assess whether the resulting coefficients (a) and (b) are sensitive to the bin size and whether alternative functional forms provide significantly different uncertainty estimates, particularly at high AOD. This is important because the subsequent global uncertainty field and all propagated GHI/DNI uncertainties are directly dependent on this parameterization.
- The radiative-transfer calculations combine 3-hourly EAC4 AOD with monthly climatological SSA and Ångström exponent, monthly EAC4 TCWV and TCO, monthly surface albedo, and hourly solar geometry. The authors should discuss the implications of this temporal inconsistency. In particular, aerosol optical properties such as SSA and Ångström exponent can exhibit substantial short-term and seasonal variability and may be correlated with aerosol type and AOD. Using monthly climatologies may therefore affect both the baseline irradiance and the calculated AOD sensitivity, especially for DNI. The consistency between the AeroCom SSA/AE climatology and the aerosol representation in EAC4 should also be discussed.
- Equation (5) expresses the propagated uncertainty as the absolute value of the sensitivity multiplied by the AOD uncertainty. However, in Eq. (7), the absolute value is not used. Since increasing AOD will generally produce a negative perturbation in both GHI and DNI, the kernel is expected to be negative. Please check the consistency of the equation.
- The BSRN comparison is useful, but it primarily evaluates the clear-sky irradiance estimates generated by the LUT/RTM framework. It does not directly validate the magnitude of the propagated uncertainty estimates. A particularly useful additional analysis would be to compare the observed model–measurement residuals with the predicted propagated uncertainties.
Technical corrections
- Replace “AOD retrieval uncertainty” with “EAC4 AOD uncertainty” or equivalent terminology where appropriate, since EAC4 is a reanalysis rather than a retrieval product.
- Provide information of the ten high-AOD AERONET stations discussed in Section 3.3.
- Standardize capitalization of geographical terms such as “Southern Hemisphere” and “Northern Hemisphere”.
- Check line 320 “indicate” -> indicates.
- Ensure consistent use of “clear-sky” throughout the manuscript.
- Check consistency in the reporting of units, significant figures, and decimal places for AOD, uncertainty, and irradiance statistics.
- Figure 3 – Please consider if the colour scale used for the AOD uncertainty map should be changed. Since most uncertainty values are substantially below 1, using the same 0–1 scale as the AOD panel, greatly reduces the visual contrast of the uncertainty field. A separate, different scaled colour bar for Fig. 3b would allow the spatial variability of the estimated uncertainty to be assessed more clearly.
Citation: https://doi.org/10.5194/egusphere-2026-4413-RC2
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 258 | 153 | 58 | 469 | 61 | 60 | 62 |
- HTML: 258
- PDF: 153
- XML: 58
- Total: 469
- Supplement: 61
- BibTeX: 60
- EndNote: 62
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This work presents an approach for estimating the uncertainty of the aerosol optical depth (AOD) provided by the ECMWF Atmospheric Composition Reanalysis 4 (EAC4) and for assessing its impact on clear-sky solar irradiance. The approach relies on a linear parameterization of the AOD estimation uncertainty as a function of the EAC4 AOD, followed by the propagation of this uncertainty into solar irradiance estimates using a radiative-kernel approach.
The topic is highly relevant for a wide range of applications in which the accuracy of clear-sky solar irradiance estimates is important, particularly for solar resource assessment. The proposed framework is interesting and provides a potentially useful contribution to the quantification of aerosol-related uncertainties in surface solar radiation estimates. Overall, I find the approach promising and the topic relevant to the scope of the journal. However, several aspects of the methodology and their underlying assumptions would benefit from further clarification and discussion, as detailed in the comments below.
General comments:
The proposed approach relies on several assumptions regarding the spatial and temporal “stationarity” of the AOD uncertainty. In particular, the linear relationship between AOD and its estimation error is assumed to be spatially homogeneous and to remain valid for periods other than the one used to derive the regression.
This assumption may be particularly important under extreme or unusual aerosol conditions, for which the actual AOD uncertainty could differ substantially from that predicted by the model. It would therefore be useful to clarify more explicitly what type of uncertainty is being quantified by the proposed approach. As currently formulated, the resulting uncertainty appears to represent a statistical or climatological characterization of the AOD estimation error, rather than an uncertainty that dynamically responds to the specific atmospheric conditions at a given time and location.
This distinction is important for interpreting the resulting uncertainty in the propagated solar irradiance estimates and should be discussed more explicitly in the manuscript.
The evaluation of the proposed uncertainty model remains limited. In particular, the evaluation using BSRN appears to focus on the accuracy of the EAC4 clear-sky irradiance itself, rather than on the ability of the proposed uncertainty model to characterize the actual errors.
A comparison between the observed EAC4 clear-sky irradiance errors and the uncertainty estimated by the proposed approach would therefore be highly relevant. Such an analysis would help assess whether the estimated uncertainties are consistent with the actual distribution of errors and, in particular, whether they provide an adequate characterization of the uncertainty under different atmospheric conditions.
Specific comments :
L11-13 : “Despite the increasing use of atmospheric reanalysis products for solar resource assessment, the propagation of aerosol-related uncertainty into SSR calculations remains insufficiently quantified”
The logical connection introduced by “Despite” is not entirely clear. The increasing use of atmospheric reanalysis products for solar resource assessment does not, by itself, seem to contrast with the insufficient quantification of aerosol-related uncertainty in SSR calculations. Please clarify the intended link between these two statements, or consider rephrasing the sentence.
L15-17 : « Prognostic AOD uncertainties are calculated using collocated EAC4 and Aerosol Robotic Network (AERONET) observations, while the resulting global mean relative AOD uncertainty is approximately 47%. »
The use of “while” is unclear here, as the two clauses do not seem to express a contrast. If the second clause is intended to report the result of the uncertainty calculation described in the first clause, “resulting in” or “yielding” might be more appropriate.
L110: “automatedcloud” a space is missing
Fig1: you represents the surface albedo at 550nm which is not consistent with the text that mention the use of a broadband albedo (line 150).
L148-149: you decided to use monthly average values for TCWV and TCO, which is an important methodological choice. Knowing that these quantities are available in EAC4, can you explain the motivations and the consequences of this choice?
Eq.4 This equation presumes that a+b AOD is less than 0.01. This is guaranteed by the value of a and the fact that AOD>0. Maybe add a sentence?