Optimization of the SKYNET calibration with an iterative Improved Langley method and its validation in the Mediterranean area

Kumar, Gaurav; Momoi, Masahiro; Campanelli, Monica; Garcia-Suñer, Meritxell; Kouremeti, Natalia; Kudo, Rei; Pedrós, Roberto; Estellés, Víctor

doi:10.5194/egusphere-2026-4267

Preprints

https://doi.org/10.5194/egusphere-2026-4267

Preprints

23 Jul 2026

| 23 Jul 2026

Status: this preprint is open for discussion and under review for Atmospheric Measurement Techniques (AMT).

Optimization of the SKYNET calibration with an iterative Improved Langley method and its validation in the Mediterranean area

Gaurav Kumar, Masahiro Momoi, Monica Campanelli, Meritxell Garcia-Suñer, Natalia Kouremeti, Rei Kudo, Roberto Pedrós, and Víctor Estellés

Abstract. The Improved Langley plot (ILP) method was developed for on-site calibration of PREDE-POM radiometers within SKYNET. It eliminated the need to ship instruments to remote high-altitude sites as required by the Standard Langley plot (SLP) method, reducing data gaps, transport-related risks, and operational costs while maintaining field instrument uncertainty below 2.4 %. The study aims to update the ILP method by adopting the latest Skyrad MRI v2 and introducing an iterative procedure. Testing with the data from the QUATRAM (QUAlity and TRaceabiliy of Atmospheric aerosol Measurements) campaigns and the Burjassot site in Spain showed an overall improvement of the ILP method. The median differences in direct aerosol optical depth (AOD) relative to GAW-PFR (Precision Filter Radiometers) decreased to approximately −0.01 at the 500 and 870 nm channels. AOD differences within the World Meteorological Organization (WMO) limits increased by 15-fold and 4-fold at 500 and 870 nm channels, respectively. The PREDE-POM and AERONET Level 2 Version 3 direct AODs at the Burjassot site showed good agreement, with median AOD differences within ±0.01 across all wavelengths, except at 340 nm. However, the fraction of points within WMO limits increases by over 10 % at wavelengths below 500 nm, while remaining nearly unchanged (variations below 6 %) at longer wavelengths. These results confirm the robustness of the ILP currently used in SKYNET and highlight the possibility of an improvement at shorter wavelengths. This limitation is addressed in the present study through the proposed iterative ILP method based on Skyrad MRI v2.

Received: 16 Jul 2026 – Discussion started: 23 Jul 2026

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.

Download & links

Gaurav Kumar, Masahiro Momoi, Monica Campanelli, Meritxell Garcia-Suñer, Natalia Kouremeti, Rei Kudo, Roberto Pedrós, and Víctor Estellés

Status: open (until 01 Sep 2026)

Post a comment Subscribe to comment alert

RC1:
'Comment on egusphere-2026-4267', Anonymous Referee #1, 14 Aug 2026 reply
GENERAL COMMENTS
The manuscript by Kumar et al. describes an update to the Skynet method (ILPU) for the in-situ calibration of POM sun/sky radiometers. In this update, a more recent model (Skyrad MRI v2) is employed, together with an interactive technique. The method is tested using data from the QUATRAM campaigns and a longer dataset from Burjassot. The results show that, in most cases, the updated method performs better than the previous version, while in general it does not lead to any deterioration in the results. As the authors honestly acknowledge, the update to the calibration method does not resolve all discrepancies between the POM measurements and those from the other instruments used as references. Nevertheless, it represents a step towards further progress and, for this reason, I believe that the manuscript is worthy of publication. The manuscript is generally well written in English. I would recommend publication after the following minor issues have been addressed.
SPECIFIC COMMENTS
It is not entirely clear to me how the algorithm converges. After the first iteration (n = 0), the retrieved F0 may still be affected by errors, meaning that the retrieved AOD may also contain errors, which will depend on the airmass. At n = 1, which is the final iteration in the scheme proposed by the authors, AOD is used in addition to normalised radiance as input. This means that the inversion at n = 1, and hence the new estimate of F0, will also be affected by the error in the AOD from the initial iteration. Why, then, should this new estimate be more accurate than the previous one? In other words, what is the mechanism that leads to convergence? This point may also be somewhat confusing for a reader who is not familiar with the details of the algorithm.

So far, the algorithm accuracy appears to have been assessed mainly through comparisons with other reference instruments and with full retrievals. A "synthetic" study could be very useful for investigating the sensitivity of the algorithm and gaining a better understanding of its potential and limitations. For example, the authors could use the forward model to simulate almucantar measurements for known aerosol properties and then apply the ILPU to check if they obtain similar values. By varying the input aerosol properties and the initial conditions assumed by the algorithm, they could gain further insight into its behaviour. I am not suggesting that such a test should be added to the manuscript. However, if the authors already have such material available, it could be included for completeness; otherwise, the test could be mentioned as a possible direction for future work.

Related to the previous point, I find that a sensitivity analysis with respect to both the initial conditions within the iterative scheme and the actual aerosol properties in the atmosphere is somewhat missing from the manuscript. Perhaps a short dedicated paragraph or section, even only based on the relevant literature, could be included in the discussion. Alternatively, if the authors already have suitable data available, the discussion could be supported by those results.

How is the "switch" between the inversion based only on normalised radiance data and the inversion that also uses transmittance data implemented technically?

Regarding the statement, "Moreover, diffuse sky radiance measurements with a scattering angle ranging from 4.5 degrees to 35 degrees is used in this study", I could not find an explanation for this choice. How does it compare with the scattering-angle range usually employed in the ILP?

Are the authors confident that the aerosol (and atmospheric) conditions encountered during the QUATRAM campaigns and at Burjassot are representative of other sites and periods in which the ILPU method might be applied? For example, how does the method perform under conditions of substantially lower or higher AOD?

To ensure that the results are reproducible, could the authors also describe which values of the atmospheric gases were included in the model and how these values were obtained (e.g. from measurements, reanalyses, climatological values, etc.)? How sensitive are the results to this choice?

Why is the sensitivity test performed only by increasing the baseline calculation by 3%, rather than also considering a 3% decrease? Does the algorithm also converge in this case?

Regarding the article structure, the fundamental improvements introduced in Skyrad MRI v2 are described rather late in the manuscript, at line 239: "the iterative approach works better thanks to the algorithm update [...] The retrieval of the aerosol optical parameters is optimised to the transmittance data rather than to the input direct extinction AOD. This allows the retrieved extinction and scattering AOD to adjust freely during the iteration process". Without this explanation, the transition from Skyrad v4.2 to Skyrad MRI v2 appears to be motivated mainly by the fact that the latter is a "newer" version, rather than by specific improvements that make it better suited to the purpose of this study. I would therefore suggest introducing this explanation earlier in the manuscript.

TECHNICAL REMARKS
Line 16: Please also mention the possibility of transferring calibration information from reference instruments. More generally, I would revise the description of the benefits of the Improved Langley Plot (ILP). In my view, the main advantage is not primarily that it eliminates the need to ship the instrument (this latter could be easily replaced by another instrument), but rather that it enables continuous monitoring of the instrument's sensitivity and performance, rather than providing such an assessment only once a year or every two years.

Line 19: Please include information on the locations of the QUATRAM campaigns. Also, specify that Skyrad MRI v2 is a model, as this is not yet clear to the reader at this point in the manuscript.

Lines 26–27: Please clarify whether the reported improvement at wavelengths mainly below 500 nm is primarily due to the higher AOD at shorter wavelengths and to the way the WMO limits are defined (i.e. independently of AOD and wavelength).

Line 32: Please note that the terminology used to describe aerosol effects has evolved since IPCC AR5, with the terms "aerosol–radiation interactions" (ARI) and "aerosol–cloud interactions" (ACI) now being commonly used.

Line 40: "Aerosol measurements" should be qualified here to make clear that it refers to measurements of the entire atmospheric vertical column. I would also replace "measure direct Sun aerosol optical depth (AOD)" with "derive aerosol optical depth (AOD) from measurements of direct-Sun irradiance".

Line 42: Please note that SSA is a derived product, as opposed to the previous ones.

Lines 57–58: Please note that international networks can provide replacement instruments during the calibration period, thereby largely overcoming this limitation.

Line 71: Could you clarify what is meant by the "annual trend" here?

Line 72: What factor changes from 0.5% to 2.4%?

Line 74 (and other occurrences): Please use \citet rather than \citep, i.e. remove "()" around citations.

Lines 76–87: Please clarify whether these studies applied the ILP method to determine F0 using only the period of the campaign or over a longer period (e.g. including measurements before and/or after the campaign).

Line 89: Skyrad is introduced here for the first time; however, it is not yet clear how the models are used to determine the calibration factor. Please provide some clarification at this point.

Line 114: Why is the 940 nm wavelength listed for POM-01 but not for POM-02? Moreover, since the authors have introduced the different wavelength ranges as the main difference between the POM models, do they think that this difference could also have an impact on the ILPU calibration?

Line 118: Are the instruments thermostated or simply heated? How might the temperature dependence affect the results presented in this paper?

Lines 119–121: Please remove one of the two similar sentences (likely a typo).

Sections 2.2 and 2.3: Please clearly state that these instruments are described here because they are used as reference instruments. Also, since the study concerns calibration, please describe how these reference instruments are calibrated. Note also that there are erroneously two sections numbered "2.3".

Section 2.4: Please clearly state the period used for the calibration. Does it coincide with the campaign periods?

Line 145: "Two PREDE-POM": were these POM-01 or POM-02 instruments, based on the description given earlier?

Lines 163–165: This seems to be a fundamental point and, in my view, should be introduced much earlier in the manuscript.

Line 168: "he" → "the".

Section 3.1: Interestingly, the paper does not mention the cross-IL (XIL) method described by Nakajima et al. (2020), which was developed to correct an underestimation by the IL method. Does this mean that the XIL method is not effective for the application considered here?

Line 172: What do you mean by "formalised"?

Line 174: "Similar to...": I guess this is "similar" because the observed signal, as described by the Bouguer–Lambert–Beer equation, does not depend on the calibration method.

Lines 176–177: Strictly speaking, different atmospheric constituents have different airmasses (m). This is particularly relevant at large airmasses.

Line 179: The expression "m = 1/cos(θ)" applies only under the plane-parallel approximation.

Lines 176–181: It appears that τ_g has not been defined or described.

Equation 4: The sign appears to be incorrect: there should be a "+" rather than a "−".

Line 190: Is the ILP, and therefore the ILPU, based on the assumption that SSA remains constant throughout the Langley event (day or half-day)? If so, how might this assumption affect the results?

Equations 6–7: Please define and explain all terms in these equations.

Line 197: Is the "atmospheric" phase function the sum of the aerosol and Rayleigh phase functions?

Line 204: Should this be "the" radiative transfer model or "a" radiative transfer model?

Lines 207–208: Please explain why this is important for the analysis.

Line 218: Why "daily"? Are the morning and afternoon Langley "events" (i.e. two per day) combined?

Figure 1: It is somewhat confusing to see arrows connecting both methods. As currently presented, the figure could suggest that one method depends on the other, rather than that the second method represents an evolution of the first. Also, the text is too small. Please use "(a)" and "(b)" to label the two panels if they are referred to separately in the text (e.g. line 331).

Line 229: Should this be "from" or "with respect to"? Also, what exactly are the "limitations" referred to here?

Line 234: Can the authors explain why including τ_ext^dir improves the retrieval? Is the information contained in the radiance measurements insufficient to estimate F0 accurately?

Lines 256–263: One of the first results that the reader is likely to be interested in is the difference between the F0 values obtained with ILP and ILPU. This is effectively presented later in Table 2, but it could also be mentioned here.

Line 269: The perturbation is described as being chosen "arbitrarily". Is 3% representative of any specific uncertainty or expected variation?

Lines 275–277: "Dashed", not "dotted".

Figure 2: Please use letters to label the different subfigures. Do the authors have an explanation for the slower convergence at shorter wavelengths? The wavelength dependence could point towards an effect related to molecular scattering or to the higher AODs at shorter wavelengths.

Table 2 and Equation 8: Usually, relative differences are calculated as (tested quantity − reference quantity) / reference quantity. Why are the differences here calculated relative to the tested quantity instead?

Lines 318–320: Consider moving these lines earlier, as they explain what is meant by the "full" inversion mentioned above.

Figures 3–4: Is there a more informative quantity than "Measurement index" that could be used on the x-axis? Why not use the date? Also, since the WMO criterion is mentioned, could a plot as a function of airmass be included somewhere? Finally, do percentage differences in AOD have a clear physical meaning, or would absolute differences be more appropriate here, given that AOD is derived from a logarithmic relationship?

Line 350: What are the two "AOD types"?

Tables 4–5: Please also include "%" in the row headers.

Line 386: Is this τ^dir or τ^dif?

Line 399: Please avoid the wording "direct AOD" in the section title.

Lines 402–403: Please revise this sentence; it appears that a verb is missing.

Line 406: Please refer to the SUNRAD "code".

Figures 5–6: Please use colours consistently across the different figures (e.g. red versus green), and explain the meaning of the colours in the captions.

Figure 5, caption: How can the WMO acceptance criteria be represented in this plot, given that the criterion depends on the airmass?

Line 425: Should this be "most" or "all"? Also, please repeat the airmass range during the campaigns in this paragraph, as the results may depend on the range considered.

Table 7: These results are somewhat disappointing compared with the other results presented in the manuscript. Do the authors have an explanation for this, or any ideas on how this aspect could be investigated further?

Figure 7: Please explicitly state which quantity is considered for the bias and RMSD (AOD?). When referring to the bias, clearly specify the order in which the difference is calculated, as the result is not commutative. Also, it appears that the last lines of the caption belong to Figure 8 instead.

Line 470: The order appears to be incorrect: MBD is shown in Figure 7, whereas RMSD is shown in Figure 8.

Lines 470–476: The observed wavelength dependence is also likely related to the fact that AOD is larger at shorter wavelengths. This could be mentioned explicitly.

Section 5: The "Summary and conclusions" section is too long. Please try to summarise the most important findings without repeating all the results presented in the previous sections.

Line 486: Please rewrite this sentence, as it is not clear which algorithm "adopted" refers to.

Lines 487–489: Please explain more explicitly what changes between the model versions, beyond simply changing the model name.

Lines 512–515: Please specify the dates and sites of the campaigns. Also, the results should be critically assessed in this section. For example, is a 1.3% difference considered satisfactory, and on what basis?

Reply
Citation: https://doi.org/10.5194/egusphere-2026-4267-RC1

Gaurav Kumar, Masahiro Momoi, Monica Campanelli, Meritxell Garcia-Suñer, Natalia Kouremeti, Rei Kudo, Roberto Pedrós, and Víctor Estellés

Viewed

Total article views: 99 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
68	23	8	99	10	14

HTML: 68
PDF: 23
XML: 8
Total: 99
BibTeX: 10
EndNote: 14

Views and downloads (calculated since 23 Jul 2026)

Month	HTML	PDF	XML	Total
Jul 2026	47	17	5	69
Aug 2026	21	6	3	30

Cumulative views and downloads (calculated since 23 Jul 2026)

Month	HTML	PDF	XML	Total
Jul 2026	47	17	5	69
Aug 2026	21	6	3	30

Viewed (geographical distribution)

Total article views: 82 (including HTML, PDF, and XML) Thereof 82 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 16 Aug 2026

Short summary

Aerosols are tiny particles suspended in the atmosphere, affecting the air quality and climate. Accurate aerosol measurements are thus important for understanding their effects. Sky radiometers are widely used instruments for the aerosol measurements. We developed a method to improve the calibration coefficients of sky radiometers, using the latest data-processing algorithm and an iterative procedure. The proposed method produced more accurate results, improving long-term aerosol monitoring.


Total:	0
HTML:	0
PDF:	0
XML:	0