General comment

The authors develop the identification method of clear- and cloudy-sky conditions from the zenith radiances measured by the sky radiometer for reliable retrievals of aerosols and clouds. The characteristics of zenith radiances are investigated in detail. The process of developing the new method is easy to understand, and it is a reasonable technique. However, the results for aerosol retrieval are similar to those of the previous method. Although the authors suggest that the new method is easy-to-use and effective, I believe it does not contribute significantly to scientific progress and, in its current form, is insufficient for journal publication. Most of the cloud retrieval methods using zenith radiances are based on one-dimensional radiative transfer model. For this reason, it is necessary to select cases of overcast skies where the parallel-plate approximation holds. Since the new method distinguishes overcast conditions, I am more interested in its usefulness for cloud analysis than for aerosol analysis. I recommend adding an assessment of its impact on cloud retrieval.

Specific comments

• L30: Is the standard cloud-screening procedure of the SKYNET?
• L62: The definition of clear-sky is ambiguous. The clear-sky is used to describe both completely cloudless skies and skies with just a few clouds.
• L87-89: The authors mention that the zenith radiance measurements are more useful than the irradiance-based screening methods or purely statistical approaches. However, there is insufficient evidence. Why don’t you compare those with the new method in this study?
• L101: If it is possible to compare the new method with that of AERONET, the authors would be able to demonstrate the advantages of this study.
• L127: The more detailed information or references regarding sky radiometer is required. The sky radiometer measures not only the zenith irradiances but also the direct irradiance, and the distribution of sky irradiances.
• L173-174: The nine MODIS pixels are used for the references of cloudy sky. Can this be interpreted as an extraction of overcast conditions?
• L185-186: Since the radiometric calibration constants vary by instruments, the ratios of radiance spectrum are different from instruments. Is it necessary to determine a threshold for each instrument? The sky radiometer has a self-calibration method. Shouldn’t the calibrated radiances be used? The authors determined the threshold using the long-term observation data, but did the changes in the calibration constants over time have any influences to the threshold?

• L205: DNI is sensitive to only the presence of clouds near the sun.
• L273-274: Isn’t the wavelength pair of 0.500 and .0675 um the most distinct pair?
• Figure 3 and 4: The authors discuss the characteristics of spectral slope using only their mean values. Can we assume that the variability is small?
• L499: The more information of the standard SR-CEReS package is required. How does the standard method discriminate the clear- and cloud-sky conditions?

The proposed cloud-screening approach appears physically consistent with the different spectral behavior of aerosols and clouds. Author found that the spectral slope derived from the logarithmic contrast between zenith radiances at 0.400 and 0.675 µm has strong sensitivity to cloudy-sky conditions but limited capability for detecting clear-sky conditions, whereas slopes based on the 0.500 and 0.675 µm and 0.400 and 0.675 µm wavelength pairs exhibits more balanced detection performance for both clear and cloudy-sky conditions. The proposed method cannot be considered a fundamentally new concept, but represents a useful and practical cloud-screening technique to apply to sun-sky radiometers.

There are some points that the authors should address:

1) In line 138, it is stated that the SKYNET archive is available at atmos3.cr.chiba-u.jp/skynet. However, this is not the official SKYNET website, which is www.skynet-isdc.org. The latter provides two processing modes, namely CERES and ESR-MRI. Are the products downloaded from the former website exactly the same as those available from the official SKYNET portal? If not, it would not be accurate to state that the database is obtained from SKYNET without further clarification.

2) To what extent does the value of the epsilon parameter (line 406) depend on the measurement site? Please discuss its site dependence and whether a site-specific tuning is required.

3) Instead of the qualitative comparison presented in Figure 9, I recommend providing a quantitative assessment of the different cloud-screening methods. For example, the authors could report the percentage or number of data points rejected by each method.

4) The proposed method should be evaluated under a wider range of atmospheric conditions. In particular, it would be useful to compare the performance of the different cloud-screening approaches during representative case studies, such as desert dust outbreaks, wildfire smoke events, and cirrus cloud conditions, in order to demonstrate any improvement achieved by the proposed method.

5) The title should clearly specify that the proposed cloud-screening method applies only to direct-sun measurements.