Clouds play an important role in atmospheric radiation (thus Earth energy budget), weather and climate. Among various cloud properties, cloud optical thickness (COT) and effective radius (CER), which can be revealed by reflected solar radiation, have been widely used to infer cloud radiative effects. Thus, COT and CER have become almost standard cloud products of satellite imagers, and, among them, MODIS provided one of the most popular ones. This manuscript by Sun et al. presented numerical study on the influences of cloud vertical inhomogeneity on the retrieval of cloud COT and CER using AGRI observations. The manuscript is overall clearly organized and presented, while there are some major issues that should be considered before it could be considered for publication.

Major Comments:

1. The focus of this study is suggested to be clarified. This study mainly investigated the influences of vertical structures perturbation on the retrieval, while the use of a particular kind of satellite observations, i.e., FY4A/AGRI, for COT and CER is not new at all. The COT and CER algorithm was the classic Nakajima and King (1989) algorithm, and Liu et al. (2023) (referred in the manuscript) presented the operational FY4A AGRI COT and CER retrieval. Thus, the key contributions of this study are the investigation on vertical structure.

2. Actually, the used of FY4A introduces more “uncertainties”. A recently study indicated that the operational AGRI L1 radiance data themselves may be less reliable due to the calibration degradation (https://ieeexplore.ieee.org/document/11071868). Such uncertainties should be considered in the evaluation of the AGRI results.

3. Excellent consistency was noticed in Figure 4 between the results of MODIS and AGRI, while the direct comparisons in Figures 5 and 6 shows clear differences. Such significant differences should be carefully checked. Again, considering the uncertainties on AGRI calibration as well as data collocation, the agreement in Figure 4, which is much better than results in Figures 5 and 6, should be exampled.

4. Some key previous studies on the influences of vertical structure on COT and CER retrievals should be mentioned and discussed. For example, Wang et al., did a systematic study using MODIS observations (https://doi.org/10.1029/2018JD029681).

5. Actually, Teng et al. (2020) did not try to show that ice-over-water system give more consistent results with observations. They developed a much advanced algorithm to retrieval COTs and CERs of ice-over-water clouds, which is almost a new retrieval algorithm. Meanwhile, their algorithm has been improved to infer cloud top heights as well (https://doi.org/10.1016/j.rse.2022.113425). The ideas of Teng et al. (2020 and 2023) are quite different from the presented study, and should be clarified.

6. What’s the physical reasons for the differences on reflectance of L and M water clouds (i.e., Exp. 1 and Exp. 2) if the only differences was the cloud location.

7. The oscillations in Figures 9c, 9d, 9f, 10a and 10b seems problematic. Especially, the peak in Figure 9f is not natural, and should be carefully checked.

This study presents a bispectral retrieval algorithm for determining cloud optical thickness (COT) and cloud effective radius (CER) using data from the Advanced Geostationary Radiation Imager (AGRI) on the FengYun-4A satellite. The algorithm was validated against coincident MODIS cloud products, demonstrating its reliability. Furthermore, ten idealized multi-layer cloud scenarios were employed to investigate the sensitivity of visible and shortwave-infrared (SWIR) reflectance to vertical cloud structure. This study is interesting. I think this paper is publishable after several major revisions before I could recommend it with enthusiasm.

Major comments:

1. This study comprises two primary components: i) the development of a CER and COT retrieval algorithm utilizing AGRI data, and ii) an assessment of the sensitivity of simulated visible/SWIR reflectance to vertical cloud configurations through ten idealized multi-layer cloud scenarios. That said, the linkage between these sections currently lacks immediacy. I suggest that improving the second part to detail how vertical cloud structures influence real-world retrieval outcomes would greatly enhance the paper's coherence and narrative progression.
2. In agreement with Reviewer #1, although Figure 4 demonstrates excellent consistency between AGRI-derived results and MODIS cloud products, the findings presented in Figures 5 and 6 reveal persistent limitations within the current retrieval algorithm. It is advised that the authors conduct a thorough reassessment of their results, ensuring that the representation in Figure 4 provides an unbiased depiction of the algorithm's capabilities.

Minor comments:

1. Line 132: As COT and CER were previously defined, repeating their definitions here is unnecessary.
2.  Figure 1: Is the logical relationship depicted between “cloud detection” and “cloud” accurate?
3. Section 3.1: How are the 4-km AGRI observations/retrievals spatially matched with the 1-km MODIS cloud products?
4. Line 349: “10b” should be corrected to “11b”.
5. Figure 12: Lacks clarity in showing how visible/SWIR reflectance responds to vertical cloud structure.