| 14 Jan 2026
Employing smoothness of the time series of sky radiances measured in the solar aureole for cloud screening
Abstract. Cloud screening algorithms have always been a critical component of Aerosol Robotic Network (AERONET) aerosol optical depth (AOD) Level 1.5 and 2.0 product. The initial cloud screening algorithm in the Version 1 and 2 database was semi-automatic and required involvement of human analyst to finalize the results. It became fully automatic in Version 3 (V3) due to employing information on the angular shape of sky radiances measured in aureole (curvature algorithm). Although efficient, the curvature algorithm is threshold based and fails to detect clouds when its parameters are beyond the corresponding pre-determined thresholds. This is especially noticeable at high latitudes where the size of ice crystals in cirrus clouds are sometimes relatively small and therefore comparable in size to aerosols. It is shown that additional information can be extracted from analysis of the smoothness of diurnal variability of sky radiances measured at the 3.3-degree scattering angle. This measurement is a part of so-called curvature scan (CCS), which takes measurements from 3 to 7.5 degrees scattering angle with 0.3-degree steps after each measurement of AOD. The analysis of the diurnal variability of CCS (3.3) for cloud-free conditions shows relatively smooth temporal dependencies, which can be fitted by polynomials with high correlation coefficients while in conditions almost completely dominated by clouds, the temporal variability is completely random. For partially cloudy days, the two main features are observed: relatively smooth aerosol signature and irregular spikes due to clouds. The new technique is proposed that employs the smoothness of the diurnal variability of CCS(3.3) scan as a criterion of the cloud free conditions. In the case when both features are present, the idea of the new algorithm is to remove irregular spikes due to clouds while keeping smooth part due to aerosols intact. The new algorithm detects spikes associated with clouds by comparing magnitudes of CCS(3.3) at neighboring time stamps through calculating their first differences (FD). This algorithm was applied to the CCS(3.3) measurements taken at several AERONET sites. The results were analyzed in terms of net change in Angstrom exponent (AE) as well as number of AOD measurements. The analysis showed the algorithm performs satisfactorily at AERONET sites dominated by fine mode aerosols, however at sites dominated by dust, the algorithm removes a big fraction of cloud-free observations. The issue was corrected by introducing an additional cloud screening parameter. It is based on observation of the different rate in changing of AE with iterations for cloud-free and cloudy conditions with much higher rate in the former case. The new parameter was selected as a slope of the linear regression between integration number and the value of AE after the corresponding iteration. Algorithm disregards FD algorithm results if the slope is smaller than certain threshold value. Finalizing the FD algorithm threshold setting as well as evaluation of the algorithm performance is done by using independent cloud detection information available from Micro-Pulse Lidar Network (MPLNET) data. The AERONET and MPLNET data were time and space collocated with additional averaging over one hour period. The comparison showed that, on average, the FD algorithm outperformed V3 L1.5 by about 0.02 in Mathews Correlation Coefficient (MCC), suggesting consistent improvement in overall cloud detection accuracy. Additional analysis performed in terms of MCC metrics also showed that the FD algorithm achieves a more balanced and accurate classification of clouds vs clear.
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.
This manuscript presents a supplemental cloud-screening algorithm for AERONET that identifies thin cloud contamination by evaluating the temporal smoothness of sky radiances at 3.3° in scattering angle. The method utilizes a First Difference (FD) threshold to iteratively filter high-frequency radiance spikes in the time series of sky radiances at 3.3°, alongside an Angstrom Exponent regression check to preserve coarse-mode dust. Validation against MPLNET lidar observations demonstrates a modest improvement in classification accuracy over the operational Version 3 curvature-based mask.
This manuscript documents significant updates to the AERONET cloud screening methodology. However, given the importance of this dataset, the presentation needs tightening and the algorithm requires a more complete explanation. My recommendation is that the following key points, as well as specific comments below, be addressed before publication:
1. The new screening algorithm revolves primarily around First Differences (FD) of the 3.3° measurement time series but it is unclear exactly what is being differenced. Most crucially, it is not clear if the data is detrended via a polynomial fit before FD is calculated.
2. The FD part of the algorithm is described in Section 4 "Algorithm Description" but the AE threshold aspect of the procedure is not mentioned until the following section entitled "Algorithm applications at selected AERONET sites". So that a clear, concise summary of the method is easily accessible, I would recommend describing the algorithm fully within a single section.
3. The iterative scheme (remove largest magnitude in each FD pair, recompute STD, repeat) could be sensitive to the order of removal and to gaps it creates in the time series. Once you remove a point, the next FD pair connects what were previously non-adjacent measurements. If FD is calculated in a way to avoid these introduce artifacts the approach should be described. If not, the possible influence of these artifacts needs to be further explored.
4. The writing needs significant editing. There are numerous grammatical issues ("themself" for "themselves" and "loaden" for "laden"). "Zenobo.org" should be "Zenodo.org" in the Yang et al. reference. The paper is also quite long for the amount of new content — the review of the V3 curvature algorithm (Section 2) could be substantially trimmed since it's already published in Giles et al. 2019.
5. The paper has 34 figures but in many cases they seem to convey redundant information. Many of these could be merged into single figures making the manuscript much easier to follow.
-----------------
Specific Comments
-----------------
Abstract (and ln 352): "Mathews" -> "Matthews"
ln 115: The explanation of curvature is helpful here but could be made clearer and more concise. I'm particular struggling to follow the parentheticals in this and the following sentence.
ln 141: The exact variables should be stated explicitly here (the magnitude of the curvature decreases?), rather than describing the peak as "flatter" which is ambiguous.
ln 142/Figure 5: Is "change" here meant to refer to change with respect to cloud fraction? The seemingly redundant "as [a] function of cloud fraction" later in the sentence confuses the meaning, especially in the context of so many other derivatives and slopes (e.g., W.R.T. scattering angle) discussed in this section.
ln 143/Figure 5: Technically this is as a function of case number which is not monotonic with cloud fraction.
ln ~175: This is very difficult to follow across four separate figures. Given that the x-axes are identical, could they all be merged into a single figure?
Figures 12, 13, 15, 16: The label the y-axis as "CCS, 3.3°", but lack physical units. The text indicates these are sky radiances at 1020 nm but the appropriate radiance units need to be provided.
Figure 12: The prior plots used time while this plot and the ones after it use day fraction. It would be easier on the reader to stick with one x-axis scheme throughout the manuscript.
ln 207: This sentence is a bit contradictory. Is the smooth signature coming from SZA variation's impact on aerosol scattering?
ln 208: The caption of the corresponding figure 13 says July 18th, not the 19th.
Section 4: The manuscript introduces TRS as a threshold for both the standard deviation of the FD and the magnitude of the individual FD elements themselves. Are these FD values calculated from the raw 3.3° radiances or a detrended version of the data using polynomial fits like those shown in Figure 12? If it is the former, which is what the text seems to imply, variability in geometry and aerosol conditions could cause large values of FD without any clouds present. For example, Figure 12(b) shows FD much larger than 3.0 at the beginning and end of the day that are likely driven predominantly by SZA change.
Ln 225/Figure 14: The text states that if the daily standard deviation (STD) exceeds the clear-sky threshold (TRS), the algorithm marks individual first differences above the static TRS as cloud-contaminated. However, Figure 14 instructs to "Mark all the FD above STD as cloud contaminated," implying the use of the dynamic daily STD. Please clarify in the text and figure which value is actively used to cull the data points.
Ln 225: The manuscript introduces TRS as a threshold for both the standard deviation of the First Differences (FD) and the magnitude of the individual FD elements themselves. Applying a dispersion threshold directly to a raw magnitude mathematically assumes the FD distribution is centered exactly at zero. While this is a reasonable assumption for perfectly stable, flat conditions, the mean of the FD will naturally shift away from zero throughout the day due to changing aerosol loading or the continuous variation of sky radiances with the solar zenith angle. Could the authors elaborate on the motivation for using a single absolute threshold to clip both std(FD) and the FD magnitudes themselves? It would be helpful to clarify why a standard outlier detection method centered around the mean (e.g., removing points outside the mean +/- TRS) was not implemented.
Figures 13 and 16: I think figure 16 has all the information of 13 but just with an additional blue line. My feeling is two figures are excess here and just figure 16 would suffice.
Ln 275: Regarding the conclusion that the FD algorithm removes cloud-free observations at Capo Verde due to a lack of AE correlation, could this simply be an artifact of the very low baseline AE of coarse-mode dust? Since both dust and cirrus clouds have very low Angstrom exponents, removing cloud contamination at this site might inherently produce negligible shifts in AE, rather than indicating a failure of the algorithm.
Section 5: Building on my Ln 275 comment, the proposed AE regression slope threshold evaluates the absolute rate of change in AE to distinguish between cloud and aerosol removal. However, this absolute slope is highly sensitive to the initial baseline AE. At fine-mode dominated sites like Thule, removing large ice crystals yields a steep absolute change in AE. Conversely, at dust-dominated sites like Capo Verde, the initial AE is already low ; thus, the maximum possible absolute change in AE upon removing cirrus is inherently constrained. A fixed, universal threshold of 0.01 biases the algorithm to systematically reject cloud screening in coarse-mode environments. That said, throttling the cloud mask in coarse-mode environments is practically understandable, as the optical similarity between large dust particles and cirrus ice crystals makes definitive separation physically difficult. If this is the goal, the authors should explicitly state this radiative transfer limitation as the justification for a less aggressive screening approach at these sites, rather than framing the AE threshold solely as an empirical fix. Furthermore, the authors should discuss whether a relative threshold (e.g., normalized by the initial AE) was considered.
Figure 30: What does the blue track represent on panel (b)?
Ln 433: The slope is stated to be 0.1 here but generally described as an order of magnitude lower in Section 5. Please clarify.