the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 05 Aug 2026
Validation of EarthCARE Surface Irradiance Products Against Ground-Based Observations and Geostationary Satellite Estimates
Abstract. The Earth Cloud Aerosol and Radiation Explorer (EarthCARE) satellite, launched in 2024, is the most complete cloud, aerosol, and precipitation observing satellite to date. Now in its second year of operation, EarthCARE provides a continuous stream of high-resolution data essential for refining weather and climate models. However, due to the advanced technologies and retrieval approaches used in EarthCARE, the credibility of each instrument and of their synergistic products must be verified. In this study, ground-based radiation observations from the Baseline Surface Radiation Network (BSRN) are used to validate surface global horizontal irradiance (GHI) computed using one-dimensional (1D) and three-dimensional (3D) radiative transfer models (RTMs) and reported in the EarthCARE ACM-RT product. To optimize collocation, EarthCARE's 1D fluxes are extended across-track utilizing EarthCARE's scene construction algorithm (SCA). In addition to the BSRN validation, EarthCARE irradiances are compared with gridded solar estimates from the Copernicus Atmospheric Monitoring Service (CAMS) radiation service, which infers high-resolution cloud information from geostationary satellites. Due to limited availability of the CAMS radiation service gridded dataset, this comparison is restricted to September–December 2024.
The results indicate that EarthCARE's 1D RTM systematically underestimates GHI relative to both BSRN and CAMS, with Mean Bias Errors (MBEs) of –9.8 W m-2 (–2.1 %) and –20.1 W m-2 (–3.9 %), respectively. Intercomparison of EarthCARE's 1D and 3D RTMs revealed that the 3D RTM exhibits lower GHI bias against BSRN observations (–4.6 W m-2 compared to –19.5 W m-2). However, the 3D RTM substantially underestimates beam (direct) horizontal irradiance (BHI) (–51.4 W m-2) while overestimating diffuse horizontal irradiance (DHI) (47.1 W m-2), leading to the near-zero GHI bias. Spatial analysis demonstrates that EarthCARE GHI is generally lower than CAMS values across most regions, particularly in Oceania, Central Africa, and Europe, while parts of South America, Northern Africa, and Western Asia are notable exceptions where EarthCARE GHI exceeds CAMS. Approximately 65 % of EarthCARE's GHI bias against CAMS can be attributed to differences in cloud estimation, while the remaining 35 % stems from differences in the clear-sky GHI.
Future data releases from BSRN and CAMS will expand the dataset, enabling a more robust assessment. These findings offer a critical early assessment of EarthCARE's performance and provide valuable benchmarks for the solar energy and atmospheric science communities.
Competing interests: At least one of the (co-)authors is a member of the editorial board of Atmospheric Measurement Techniques.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.- Preprint
(21905 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2026-4519', Anonymous Referee #1, 07 Sep 2026
-
RC2: 'Comment on egusphere-2026-4519', Anonymous Referee #2, 18 Sep 2026
General comments:
This manuscript compares EarthCARE surface downward flux products to BSRN dataset and Copernicus Atmospheric Monitoring Service (CAMS) outputs, with a strong focus on shortwave spectral band. Attentions are paid in data collocation to make the comparison meaningful, for example, by applying the scene construction algorithm (SCA) for the extrapolation of EarthCARE 1D product over the desired domain. The results are stratified by the all-sky-to-clear-sky ratio (CMF: cloud modification factor), distance of the BSRN station to satellite ground track, and geographical locations. In addition, the clear-sky discrepancy is separately addressed by comparing the aerosol optical depth to the values from CAMS and Aerosol Robotic Network (AERONET). While this manuscript handles a large amount of data and offers authors’ cutting-edge analysis, the presentation of results hinders the access to the general perspective of the validation strategy and results. The manuscript would require a substantial revision for final publication in AMT. The following points should be addressed to improve the accessibility of the manuscript.
- Reorganizing the data and methodology section
As this manuscript handles a variety of datasets, Sections 2.1 to 2.6 introduce each of them. However, the detailed information necessary to reproduce the research is not sufficiently provided, for example, dataset names, versions, and variable names in the dataset. In addition to the introduction of products, it is necessary to precisely highlight what are used in this study. To help understanding the use of these products, it may be worthwhile to add a schematic diagram presenting the research strategy (e.g. what variable of which product is compared against what variable of which product). Since Sections 2.7 to 2.10 offers rich details of the methodology, they could form an independent Methodology section. A part of the list in the end of introduction (Lines 102-111) could be move to this new Methodology section to provide the overview of the technical challenges and relative position of them in the entire research framework. - Collecting discussion that is scattered around in Results and Conclusion sections
The Result section presents lots of data, sometimes together with the interpretation and expert’s guess. Conclusion section is relatively long, with some comments that has a value as discussion. Collecting all interpretation, hypothesis, and discussion to form a new Discussion section would make the Results section easier to follow. - Shortening conclusions and associating claims therein to results
The Conclusions section counts nearly 100 lines, exceeding 1/8 of the main manuscript texts. Some repetition of important results is necessary, but the current Conclusions section develops a new discussion with unpresented data and omit a part of presented results (e.g. IR validation). By introducing the Methodology section that describe the overall research architecture, the Conclusions section would become naturally shorter. It is also necessary to highlight each and every claim in the Conclusions section to results.
Specific comments:
Line 4: “However, due to the advanced technologies and retrieval approaches”
Authors reason the need of validation by the novelty of the mission, but the validation is essential for any indirect measurements. Operational meteorological satellite missions invest tremendous efforts in calibration and validation even though the technology is mature. I suggest that authors simply remove this sentence because it is not necessary.Line 33: “averaging around 1361 Wm−2”
When averaging globally and daily (both day and night), the incoming solar flux is about 340 Wm-2, which is the value used to compute the energy imbalance. I request authors to develop the reason why the TOA irradiance is presented here. Otherwise, simply referring to the value by Stephens et al. 2012 or Loeb and Wielicki 2016 would be sufficient.Line 48: “high-temporal-resolution”
Though depending on the swath of the sensor and the latitude of target location, a polar-orbiting satellite imager revisits the same location on the ground only twice a day. Authors should provide the reason why they consider this is high temporal resolution.Line 74: “However, significant discrepancies were identified in the depolarization ratios”
The depolarization ratio is not mentioned anywhere other than this line in this manuscript. I don’t think this sentence is necessary for this paper, though informative.Line 106: “cloud modification factor approach”
This is somewhat confusing because this approach is nowhere defined in the introduction. If the authors consider this approach is their original approach, it should be mentioned as such. Otherwise, it’s preferred to present the previous relevant studies.Line 107-108: “quantify the across-track extension algorithm’s performance”
No reference is mentioned to quantify the performance of the algorithm. Do authors imply, “Evaluate both 1D and 3D EarthCARE products against BSRN data to quantify the across-track extension algorithm’s impact”?Line 110: “differences observed at the TOA”
It would be helpful to specify what measured quantity the authors are focusing on.Line 122: “ACM-CAP and ACM-COM”
This subsection could be easier to understand if it provides the overall relation between datasets (ACM-CAP, ACM-COM) and the required inputs to ACM-RT. Currently, the title “ACM-CAP and ACM-COM” implies that both are prerequisite for the ACM-RT processing, while at the last sentence reads “an alternative set of cloud, aerosol, and precipitation profiles”, implying that ACM-CAP could replace ACM-COM. As this is in the “data and methods” section, it is also necessary to describe which profile is used (ACM-CAP or ACM-COM one) in this study. If this manuscript doesn’t directly use ACM-CAP or ACM-COM products, there is no need to make a section to describe whole details.Line 128: “allowing users”
I suppose this “users” mean the data product provider, not the data product users. The description of CAPTIVATE algorithm is useful to understand the ACM-CAP product, while some sentences are more of introductory nature, making them more appropriate in “introduction” section.Line 148: “scene construction algorithm”
Is this different from “across-track extension algorithm” mentioned in Introduction? If so, please describe, if not, please keep the wording consistent throughout the manuscript.Line 153-154: “cosine of … shoud not exceed a certain amout”
I don’t geometrically understand what “the cosine of solar zenith angles and azimuth angles between sun and satellite tracking direction” means. Perhaps, authors meant as follows: the cosine of solar angles as well as the cosine of azimuth angles between sun and satellite directions should not exceed a certain amount. However, it is still confusing because a cosine exceeding a certain value means that two directions are close. Solar zenith angles close to nadir, or two azimuth angles close to each other are rather favorable condition, isn’t it?Line 167 and 168: “±55° forward”, “±55° backward”
I am confused by the negative in these expressions.Line 232: “AERONET”
This section may surprise readers because AERONET is not mentioned anywhere in abstract and introduction.Line 250: “measured value”, “modelled value”
It’s unclear what measure/modelled value can go into G. Perhaps, “downward shortwave flux value”?Lines 273-279
This paragraph introduces EarthCARE dataset rather than the collocation approach. It could be placed elsewhere.Line 315: “SCA indices”
The subsection fails to describe what SCA indices are, and the difference from the quality index.Line 380: “composite algorithm products (see Sect. 2.1.1)”
This expression doesn’t appear in Section 2.1.1. It could mean ACM-COM, but the explicit mention would be beneficial for the clarity.Line 426: Table 1
The number of samples in each regime should be reported as in Table 2. Since the number of points in cloud-free and overcast conditions are limited, it may be beneficial to show median in addition to exclude the impact of outliers.Lines 459-460: “combined effects of … are smallest when in close proximity”
More description would be beneficial since neither of median nor RMSE (or IQR) approach to zero with decreasing distance. The difference to Fig. 6d could be discussed to emphasize if there is an advantage of using ACMB-3D.Lines 461-485
This paragraph criticizes the radiative transfer model (RTM) used in the production of EarthCARE 1D and 3D products, but the data don’t show the flaw of the RTM itself. I suppose that authors meant the EarthCARE 1D and 3D products overall (including SCA and all relevant processing) rather than RTM. This comment applies to many other places, where RTM is used as the label of the EarthCARE products. It is necessary to reword all confusing usage of the term RTM.Line 462: “This discrepancy is largely attributable to the suboptimal spatial collocation”
I could not find the data supporting this claim.Line 475: “do not produce such significant deviation”
I encourage authors to present these results as it is the backbone of the expert’s guess about the discrepancy.Line 684: “the analysis confirms that …”
As pointed out in a previous comment (Lines 459-460), this statement would become more convincing by providing further description of results. At the current state, it could be considered as an overstatement.Line 685: “our analysis of ACM-RT input data from CAPTIVATE”
I assume that this is based on the results presented in Section 3.3, where CAPTIVATE AOD is shown to be smaller than AERONET AOD. The manuscript would become more convincing to add that the reference is AERONET.Lines 693-695: “while the 12 overpasses …”
This data is not presented anywhere in Result section. I request authors to provide appropriate description of all results that support conclusions.Lines 707-726
This paragraph doesn’t follow the flow of Section 3.2, making it difficult to find the corresponding evidence of authors’ claims in Results section. I strongly encourage authors to reorganize either of this paragraph or Section 3.2 to make their logical flow consistent.Line 707: “quality of the different collocation methodologies”
The second paragraph of Section 3.2 states that authors show the results for different “EarthCARE methodologies”, not really three different collocation methodologies. I suggest removing “collocation” from this line.Lines 724-726: “For the direct and diffuse components …”
I was not able to find the corresponding evidence in Results section.Lines 733 and 741:
Authors claim that the difference in AOD retrievals cannot fully explain the clear-sky irradiance bias in Line 733, while presenting a hope that the clear-sky irradiance discrepancies to be resolved in by switching the retrieval from CAPTIVATE to COM in Line 741. I got an impression that authors are not consistent in the same paragraph. More concise but insightful comments are necessary.Technical comments:
Line 8: “reported in the EarthCARE ACM-RT product”
For the clarity, I suggest moving this clause immediately after the “global horizontal irradiance (GHI)”, and changing “computed using 1D and 3D …” to “that are computed using 1D and 3D …”.Line 52: “CloudSAT”
The capitalization is “CloudSat”.Line 150 and others: “SCA algorithm”
When spelled out, this becomes “Scene Construction Algorithm algorithm”, which is redundant. It is worth considering replacing them by simply “SCA”.Line 220-221 and 230-231: “Copernicus Atmosphere Monitoring Service”
The same section starts with acronym “CAMS” and uses the acronym throughout except these locations. In addition, the years are not consistent (2022, 2021, while 2026 in the second line of the section.)Line 245: “these effects”
Perhaps, “their effects”? It’s difficult to interpret as: “these effects = influence”.Line 398: “extrapolation”
Is this interpolation instead?Line 423-424: “BSRN measured slightly higher irradiance than EarthCARE, …”
As this section is the EarthCARE validation against BSRN, it would be more straightforward to phrase as, “EarthCARE reports slightly lower irradiance than BSRN”. Likewise in Lines 441-442.Line 425: “data sets”
To be consistent to elsewhere, “datasets” would be better.Line 443: “regularly”
I suppose that authors meant “frequently” or “occasionally”.Lines 487-488: “lower values in the performance metrics”
It could be misread as lower performance. I guess that authors meant as follows: Figure 9 demonstrates that EarthCARE 1D downward flux in LW deviates less than in SW from the BSRN measurements.Line 488: “improvement”
“low discrepancy” would be more suitable. This is not an improvement (which implies the gained performance over time), but just the different physics as authors mention later.Line 490: “physical behaviour”
Cloud doesn’t behave differently in SW and LW, but results in different radiative consequences. What about “physics behind the radiative effects”?Citation: https://doi.org/10.5194/egusphere-2026-4519-RC2 - Reorganizing the data and methodology section
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 221 | 96 | 27 | 344 | 22 | 21 |
- HTML: 221
- PDF: 96
- XML: 27
- Total: 344
- BibTeX: 22
- EndNote: 21
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
General
This study compared the GHI from the EarthCare ACM‑RT product against BSRN ground measurements. The EarthCare GHI is also compared with CAMS, which is based on cloud properties from geostationary satellites. Overall, the EarthCare GHI is smaller than its CAMS counterparts. For cloud‑free cases, EarthCare GHI is biased low relative to BSRN, a discrepancy that is not explained by AOT biases.
The study is thoroughly performed, and the manuscript contains relevant information for the clouds, aerosol, radiation, and modeling communities. However, I feel that the manuscript could be improved by presenting clearer messages. I suggest that the authors highlight their most important findings and consider moving several figures to an appendix or supplement, if appropriate.
Major comments:
The reason for the negative GHI biases in the EarthCare product is not well explained in this study. While the analysis shows that aerosol optical depth cannot explain the biases, it would be relevant to discuss other possible causes. Additionally, the reasons for the differences in GHI between EarthCare and CAMS are not explored, except for regions where positive differences appear (e.g., Sahara and Amazon). I do not expect the exact reasons to be fully resolved here, but the manuscript should clearly outline plausible contributing factors.
Specific Comments:
* Section 2.2: I assume there are also uncertainties in the TOA fluxes derived from BBR measurements due to the unfiltering and angular correction processes. Please provide the uncertainty ranges of the BBR fluxes or include relevant references. Are there studies comparing the accuracy of CERES and BBR TOA fluxes?
* Line 197: Figure 1 could potentially be moved to Appendix A, considering the large number of figures in the main manuscript already.
* Line 199: So only daytime EarthCare orbits were compared? This implies that daytime comparison was also considered for LW. Since daytime and nighttime LW biases often differ, it would be relevant to clarify that LW biases were examined only during daytime.
* Line 210: Have you compared cloud fractions and cloud optical depths estimated from EarthCare CAPTIVATE and the McCloud model? These two parameters should be central to explaining differences in GHI for total skies (but not cloud-free regions).
* Line 517: I assume the CAMS radiation dataset is also produced using the 1D RTM. If so, would the use of a 1D model for the EarthCare give better consistency between EarthCare and CAMS, assuming similar cloud properties? However, cloud fractions and optical depths from the two products may differ significantly, likely contributing to the GHI differences (as discussed in Fig. 10).
* Fig. 12b: Could the authors explain the large outliers when EarthCare GHI is 600–800 W/m² and CAMS GHI is 400–600 W/m²? The majority of points lie near the one‑to‑one line, indicating strong correlation. However, after accounting for outliers, the mean negative difference suggests a systematic smaller values in EarthCare GHI relative to CAMS.
* I see the point of including Fig. 13 to motivate the scale used in later figures. However, Fig. 13 could be replaced by simply referencing Line 573: “In contrast, the MBE remains scale‑invariant…”
* In Table 1, when the EarthCARE GHI was compared with BSRN, EarthCARE GHI was biased low for cloud-free cases, and EarthCARE GHI was biased high for overcast scenes. In Fig. 11, EarthCare GHI is smaller than CAMS for cloud-free scenes (CMF ~ 1), and EarthCARE GHI is larger than CAMS for overcast scenes (CMF ~ 0). Fig. 15 shows that BSRN is somewhere between EarthCARE and CAMS for all skies. How about cloud‑free cases specifically—does BSRN lie between the two? What are the main factors deriving all these differences for cloud-free cases? Since AOT was ruled out, could the authors suggest other parameters driving differences under cloud‑free conditions?
* Considering CAMS uses multiple geostationary platforms, are there discrepancies in cloud properties depending on the satellite source?
* Fig. 14c: EarthCare GHI is larger than CAMS GHI over the Sahara and Amazon, likely due to smaller AODs in EarthCare. Other regions show opposite signs. Since AOD does not explain these differences, could the authors suggest additional relevant parameters?
* When comparing RT simulations with TOA BBR or ground measurements, are the biases dependent on solar zenith angles?
* I understand that the comparison in Fig. 18 was performed to ensure consistent sampling between the surface (SFC) and TOA evaluations. However, for the TOA comparison, additional sampling could potentially be included between the ACM‑RT product and the BBR measurements. Have you, or has any previous study, carried out a direct TOA comparison between ACM‑RT and BBR measurements? If so, are the findings of this study consistent with those earlier comparisons?