The manuscript addresses an important topic in multi-angle polarimetric aerosol retrieval and makes use of valuable DPC observations. However, in its current form, the study exhibits substantial shortcomings in the justification of novelty, the description and reproducibility of the retrieval methodology, and the physical interpretation of several key results. In particular, the uncertainty characterization and its consistency with the assumed measurement accuracy are insufficiently addressed, and some systematic features in the retrieval products remain unexplained. These issues are considered fundamental and would require major methodological restructuring rather than incremental revision. I therefore recommend rejection of the manuscript in its present form. Specific comments are provided below to clarify the main concerns and to explain in more detail the basis for this recommendation.

1. L47: “However, existing aerosol retrievals based on DPC observations have primarily focused on AOD.” This statement is not valid. A large number of studies have already used DPC observations to retrieve aerosol optical properties beyond AOD.
2. L55: “while numerical methods have so far been limited to regional and oceanic applications.” This statement is not correct. DPC observations have already been used to retrieve aerosol parameters over land and at the global scale. Please add recent references related to aerosol retrievals using DPC observations. In addition, “This highlights the need for further exploration of global SSA retrieval using DPC observations” is also not valid, as global SSA retrievals based on physical methods and DPC observations have already been reported.
3. L63: “This study uses measurements from the first DPC sensor since data is not well calibrated for the following two sensors.” This statement is unclear. Which sensor is referred to as the “first DPC sensor”?
4. L68: What do you mean by “instrumental drift” in this context?
5. L70: What does “Rayleigh scattering-based” refer to?
6. L76–77: “the screening of cloud/ice/snow pixels additionally employs the original measurements from the other bands.” Why are uncalibrated measurements used for cloud/ice/snow screening? You mentioned that the uncertainty of DPC observations may exceed 20%. Please quantify the retrieval errors introduced by not using calibration coefficients. You did not specify which calibration coefficients were used for the polarization measurements. First, please clearly state whether calibration coefficients were applied and provide their values. Second, if no calibration coefficients were used, please quantitatively assess the retrieval errors caused by ignoring calibration.
7. L92: “The AOD measurements can be obtained at several wavelengths ranging from 340 nm to 1640 nm.” AERONET does not appear to provide AOD measurements at 1640 nm.
8. L96–97: “we utilize the all-point Version 3 Level 2.0 direct AOD measurements.” This statement is unclear. Does “direct AOD” mean that the AOD is not quality-controlled? What does “all-point” refer to?
9. In Eq. (4), the cost function of the retrieval algorithm is given, but the state vector to be retrieved and the prior information are not specified.
10. Section 3.1: What SSA value is used in the simulations? Under different SSA conditions, a ±0.03 perturbation may lead to completely different conclusions. Moreover, the apparent reflectance change induced by SSA perturbations does not seem equivalent to the accuracy required for satellite aerosol retrievals.
11. In Fig. 5, please explain why AOD and DHR are simultaneously underestimated under high aerosol loading conditions. In addition, why do the numbers of retrieved AOD values differ among different wavelengths?
12. In Fig. 8, the map of AOD contains missing values, while SSA does not. Why?
13. Under biomass burning conditions, why do high SSA values overlap with possible aerosol plumes? Biomass burning is generally expected to emit strongly absorbing aerosols. Please explain.
14. 9 shows similar issues to Fig. 8. The missing values of AOD and SSA do not overlap. In addition, irregular blank areas appear in coastal regions. Please explain the reasons.
15. Please explain the anomalously high AOD values at high latitudes in Fig. 10(c) and over northwestern China in Fig. 10(d).
16. Please explain the anomalously low SSA values over South America in Fig. 11(b) and over northern Asia in Fig. 11(c).
17. Please add comparisons with other satellite aerosol products and provide global difference maps.
18. Please explain the reason why AOD and DHR are simultaneously underestimated under high aerosol loading conditions.
19. Please correct the grammatical errors throughout the manuscript.

This article develops a numerical method for simultaneously retrieving AOD, SSA and DHR from DPC multi-angle polarimetry (MAP) observations. The authors analyze the sensitivities of both scalar and polarimetric reflectance to SSA, and get reliable results of global AOD, SSA, and DHR based on the retrieval algorithm. As one of the few operational spaceborne MAP sensors, inversion algorithms based on DPC observations, particularly those for SSA retrieval, remain quite limited, with few global SSA products currently available. This study provides a reference for the design of future MAP sensors and the development of aerosol retrieval algorithms. The global SSA maps presented in this article also provide valuable insights into aerosol scattering and absorption characteristics in recent years. However, several aspects of the data, methods, and results require further clarification and improvement. 

General comments:

1. Data used in the article: The authors retrieve aerosol and surface parameters from DPC measurements, and validate the retrieved results according to comparing with AERONET and MODIS products. However, there are several points need to be clarified. (1) DPC data: The description of DPC data is not very clear. The authors introduce that DPC provide observations across 8 spectral bands, but only 5 bands are mentioned in Section 2.1. Moreover, the use of polarized observations is also not specified. (2) AERONET data: The authors use all the Level 2.0 quality control criteria except AOD threshold to screen the AERONET Level 1.5 SSA. But the 0.4 AOD threshold are still utilized to screen the SSA in the validation progress. Why not directly utilize the Level 2.0 SSA data? (3) The periods of DPC and AERONET data used in the article are not specified.
2. Retrieval algorithm: The authors have provided a detailed description of the retrieval algorithm, but the introduction to the flowchart could be strengthened by providing more details. The authors can consider the following aspects. (1) The DPC observations have “up to 12 viewing angles” and typically “exceeding 9 angles”. Are observations from all the angles used for inversion? What about those pixels with viewing angles less than 9? (2) Which method is employed to minimize the cost function? (3) What criteria are utilized for the matching of DPC and AERONET data?
3. Validation results: The authors analyzed several statistical indicators (e.g., correlation coefficient, RMSE, bias) to validate retrieval results. For SSA validation, the ratio of points falling within the Error Envelope (EE) is also a useful indicator to assess the result. It would be helpful to add EE lines to the figures and annotate the corresponding point ratios.
4. Global maps: Section5 shows global retrieval results. These maps provide valuable insights into aerosol and surface properties. I have two comments about these results. (1) The authors explain that the zonal discontinuities in the maps are primarily due to defects in the original data. However, the discontinuities observed over oceanic areas also seems to be related to the incompletely filtered clouds. Would be better to clarify in the manuscript. (2) Heavy AOD is observed over Southern Oceans in January 2020. The authors attributes it to the strong Australian wild fires. However, the AOD over Southern Oceans almost reaches 0.5 in Fig. 10(d). Is this magnitude reasonable? Was it also observed by other sensors?

Minor comments:

1. L68: The calculation of DOLP is not described in the article.
2. Fig. 2-4: Please explain the x label “scat_ang” in the caption.
3. Fig. 7: DHR data are not AERONET observations. Please revise the caption.
4. Section 3.2: Please keep the number of significant digits consistent.
5. Fig. 8: The “(a)/(b)/(c)” labels are not marked in the figures.
6. Figs. 8-12: The wavelength labeled in the figures (440 nm) does not match the wavelength mentioned in the text (443 nm). Please update either the figures or the relevant text for consistency.

General comments:

This manuscript presents the simultaneous retrieval of aerosol optical depth (AOD), single scattering albedo (SSA) and directional-hemispheric reflectance (DHR) from the Directional Polarimetric Camera (DPC) aboard China’s Gaofen-5 satellite. The sensitivity analysis, performance evaluation by synthetic data, and application to actual DPC data are discussed. The sensitivity analysis demonstrates the challenges in retrieving SSA values from the DPC data, while the algorithm based on optimal estimation is designed to overcome the challenges. The performance of the developed algorithm is evaluated both based on synthetic data and Aerosol Robotic Network (AERONET) data. A case study and global maps are also presented. While this manuscript covers a very challenging topic and takes a conservative approach, the restrained reproducibility and incomplete analysis limit the value of this manuscript to the research community. The manuscript would require a substantial revision for final publication in AMT if authors are willing to do so. The following points should be addressed to increase the impact of this manuscript.

1. Identifying the original dataset
   
   This study uses DPC, MODIS, ERA5, HYCOM and AERONET data. However, some descriptions are missing regarding product type, time period (start date and end date), dataset size (number of pixels and number of AERONET sites). Please detail as much as possible, and if not sufficient, use Appendix to describe the data so that readers can collect necessary data.
2. Describing applied corrections and filters
   
   It appears that authors have applied some corrections to DPC data, but the details are not presented. Some references are provided, but it would be beneficial to briefly describe the outline of the applied corrections. In addition, the “fine-mode filtering” of AERONET data (and potentially retrieval results) remains unclear. Documenting the number of points (pixels) before and after filtering would help readers understand the extent of filtering.
3. Clarifying the measurement vector, a priori state vector, model parameters, and variance-covariance matrix in the retrieval algorithm
   
   The developed algorithm is based on optimal estimation. Optimal estimation is a blending of measurements and a priori information based on variance, and therefore the information on the a priori state vector, model parameters, and variance-covariance matrices should be documented and the reasons behind the choice should be described. This shortcoming makes the study non-reproducible and the interpretation of results very challenging. In addition, the measurement vector elements remain ambiguous. It is requested to document the used spectral channel, polarization or not, and the filtering applied to the input data.
4. Improving the compatibility with other studies
   
   The validation sections discuss mainly the results at 443 nm, but a number of previous research use 550 nm for comparison. The contribution of fine-mode particles is substantially different between two wavelengths and presented results are not directly comparable. Even if the results at 550 nm may not be very encouraging, it serves the community to understand what to improve in the next instrument and algorithm development.
5. Deepening the analysis of the case study and global statistics
   
   Although authors mention in the conclusions that the analysis remains qualitative, there are unexplained patterns and characteristics in the case study and the global map. Figure 9 seems to be influenced by the number of view directions (bands-like patterns running in satellite cross-track direction) and Figures 10 and 11 show the difficulties probably related to cloud mask (Southern Ocean AOT values) and view geometry (sharp contrast of SSA that stretches along the orbit).

Specific comments:

Line 44:

It is somewhat misleading to claim that DPC follows up the data gap after the retirement of the POLDER sensors because there are no Chinese investments in international public data dissemination to my knowledge. If there is any news about this, please comment or refer to an article. This is important for the European community that is the main target of this journal.

Line 76: “theoretical correction errors”

This expression is confusing. Is this the magnitude of correction, or the remaining bias after applying the correction? Please clarify.

Lines 84-86: “Considering relatively …”

Two referenced papers are not providing the ground to use a single aerosol model to analyze polarimetric measurements. Both studies focused on MODIS retrievals.

Lines 121-122: “accurately”, “high-precision”

The relevance to this paper is the accuracy and precision needed for this research. Unless there are strict requirements that limits the choice of radiative transfer models, I suggest removing these non-quantitative expressions.

Line 157: “fixed constants”

Optimal estimation is a blending of a priori and measurements, and therefore the assumption used in the a priori can easily bias the estimation when the measurement information content is not sufficient. The precise values used in this study should be presented with reasons of choice.

Line 211:

This section presents the results of numerical simulation, but the simulation geometry is not mentioned anywhere, limiting the reproducibility of the obtained results.

Lines 215-216: “not sensitive”

This sentence is an overstatement as only one aerosol model and two land surface models are employed. It also contradicts to the previous studies, notably to the study of critical surface albedo by Seidel and Popp (2012).

Figure 2, 3, and 4:

Fluctuating results from the radiative transfer simulation of such smooth phase function is surprising. Are small fluctuations of these curves significant? What is the precision of the radiative transfer simulation in this study? Adding error bound and explanation in the main text deems necessary.

Line 251:

In this section (Section 3.2) and in the following section (Section 3.3), I consider that it is necessary to perform the performance comparison at 550 nm (or 565 nm) rather than at 443 nm. Otherwise, this manuscript is not comparable to other previous publications that authors themselves referenced. As authors mention near Line 285, the retrieval performance depends on wavelength and available information. This is why the comparison at an inconsistent wavelength is not adequate.

Line 254: “443 nm AOD”

This has to be kept as AOT at 443 nm, even when comparing the performance of AOT and SSA at 550 nm, in order to be consistent with AERONET processing algorithm that applies the “fine mode filter” by AOT 440 nm.

Line 288:

Same comment as Line 251 for this section.

Line 292: “similar performance”

This reads contradictory to Line 266, “relatively weaker performance”. The correlation coefficient of SSA drops from 0.688 at 443 nm to 0.347 at 565 nm, and it is difficult to claim that the performance at 565 nm is “similar” to that at 443 nm.

Line 299: “comparable to the operational MODIS AOD product (Levy et al., 2010)”

This claim is an overstatement. The referenced paper reports the correlation coefficient of 0.896 for AOT at 470 nm and 0.882 for AOT at 550 nm. This sentence claims that correlation coefficient of 0.75 for AOT at 443 nm is comparable.

Line 339: “strong absorption properties”

In Levy et al. 2007a, heavy smoke by biomass burning is recommended to be modeled by their “absorbing” model, instead of “moderately absorbing” model, while this study uses their “moderately absorbing” model and employ the heavy smoke case as a case study. It is worth mentioning the reasons behind the inconsistency.

Technical comments:

Line 67: “normalized radiation”

normalized radiance

Lines 140-141, 145: x^a

Using x^a as a-posteriori state vector is rather confusing as S_a is used for a-priori variance-covariance matrix. I suggest replacing by x^ (x-hat, circumflex above x) , following the common convention.

This study develops a numerical inversion algorithm to simultaneously retrieve AOD, SSA, and land surface reflectance using multi-angle polarimetric data from the Gaofen-5 DPC. By employing an optimization-based approach and validating against both simulated data and AERONET observations, the authors demonstrate the potential of DPC for global aerosol monitoring. While the topic is relevant and the retrieval framework shows promise, significant concerns remain regarding the methodology's novelty, the representativeness of the simulated data, and the rigor of the validation process. My specific comments are as follows:

1. The English expression should be refined to meet academic writing standards.

2. Lines 54-56: please clearly justify what novel contribution this work makes beyond RemoTAP, GRASP, and ML-based approaches. 

3. Section 2.8: there is no evidence that the simulated data realistically represent actual DPC observations, and the manuscript fails to quantify potential biases arising from the integration of multiple data sources.

4. Lines 157-159: how the multi-angle information from DPC is utilized in the proposed retrieval method?

4. Lines 200-202: it is unclear whether the actual DPC observation scattering angle distribution has been considered.

5. Section 3.1: the authors should clarify its relevance to the study.

6. Figure 5: the same AERONET-MODIS matched dataset both to generate the simulated observations and to validate the retrievals? Please clearly clarify this point. 

7. Figure 7: The validation accuracy against AERONET is markedly lower than that obtained with simulated data; how do the authors justify the reliability of the simulation-based validation? Are the AERONET datasets used for simulation-based and real-observation validation independent, or is there potential overlap that could bias the results?

8. Lines 318-320: this explanation is not convincing, as such pronounced striping is not evident in the corresponding true-color imagery.

9. Figure 10: The cloud masking appears to be problematic, resulting in extensive missing retrievals and apparent errors in the global 440 nm AOD product.

General comments

This study focuses on digging the potential of space-borne multi-angular polarimetric sensors to simultaneously retrieve wavelength-dependent optical properties of atmospheric aerosols and the Earth surface, namely the aerosol optical depth (AOD), single scattering albedo (SSA) and directional hemispherical reflectance (DHR). The sensitivity study highlights sensitivities of multi-angle intensity and polarimetric measurements to SSA at different wavelengths and under different AOD loadings, leading to the demand of an accuracy better than 5% for intensity and 0.01 for DOLP measurements, respectively to retrieve SSA with 0.03 uncertainty. The study can contribute to the community by providing aerosol and surface properties retrieved from DPC data, while I have some major concerns related to the motivation, method, and validation of the study:

1. The motivations of this study are not well demonstrated. The potential and limitations of multi-angular polarimetric observations for aerosol and surface property retrievals has already been well demonstrated by numbers of previous studies, for both POLDER and DPC measurements. The concept of the retrieval method (optimal estimation) has been well established and the retrieval accuracies of AOD, SSA and DHR don’t show significant improvement compared with existing algorithms, such as GRASP/PARASOL (Dubovik et al., AMT, 2011; Chen et al., ESSD, 2020) and machine learning methods (Dong et al., TGRS, 2024). Thus, the necessity of developing the retrieval algorithm is not convincing enough. While the motivation of efficiently processing DPC data to generate global aerosol and surface products may stand, it is somewhat regrettable that the retrieval performance over the ocean is not strictly validated in this work.
2. The retrieval algorithm is not clearly illustrated. The retrieval algorithm is based on the optimal estimation which is a well-established inversion concept, while many aspects are still ambiguous: what are the exact definitions of the utilized measurements and the parameters to retrieve? Are all the spectral, angular, intensity and polarimetric measurements from DPC used? How are the a priori values and covariance matrix determined? How is the optimization procedure realised (iteration step, convergent criteria, initial points, etc.)?
3. The analysis of the algorithmic accuracy and comparison with other studies are not adequate neither based on simulated data nor based on real measurements. The retrieval from simulated data assumes absolutely accurate measurements fully represented by the radiative transfer model, while it does not assess the influence of measurement noise on the retrieval results. There is also lack of comparison with other multi-angle polarimetric inversion algorithms based on simulated data (e.g., the work by Dubovik et al., (AMT, 2011)) to evaluate the advantages and disadvantages of the retrieval algorithm proposed in this work. Furthermore, when applying to real DPC measurements, there is only a small part (L312-313) mentioning the comparability of SSA correlation coefficient with one study based on DPC data (Fang et al., RS, 2022). If the development of DPC retrieval products is one of the main motivations of this study, more comprehensive comparison of the DPC retrieval accuracy with others studies, such as those you mentioned in Introduction (L47-54), is probably necessary.

According to these concerns, more adequate illustration of the motivation and a more comprehensive discussion section might be needed in order to be qualified in the publication in AMT. In addition, there are several statements not true or at least not rigorous enough in the main text. Please see the following comments for more details.

Specific comments

L54-56: What do you mean by “lack physical interpretability”? A method is considered as effective as long as its accuracy is sufficiently validated, isn’t it?

L57-58: It is recommended to insert a short paragraph at the end of the Introduction to briefly describe the structure of the manuscript as well as the main contents of each section.

L84-87: The phase matrix is an important microphysical property which depends on aerosol size, shape and refractive index, and which in turn influences aerosol SSA which is the core retrieval parameter throughout this study. Thus, rather than simply referring to other studies, you need further sensitivity studies to justify the statement of “relatively minor influence of aerosol phase matrices on the inversion”.

L136-137: What are the exact retrieval parameters that compose the state vector? The AOD, SSA, and three RTLS parameters? At which wavelengths?

L156-158: Do you mean the x0 and Sa in Eq. (4) and Eq. (5) are derived from the statistics of multi-year datasets? If so, then:

• Which datasets (i.e., which climate models, or observations) do you use? By “multi-year”, what are the exact time periods? And what are the rules for data screening?
• On a yearly and global scale, how do you take into account the variability of aerosol and surface properties and ensure no systematic bias is introduced by the a priori constraints?
• Please specify the “fixed constants” of x0 and Sa.

And if not, please explain the way of determining the a priori constraints in more details.

Section 3.1: The authors are encouraged to reduce the discussions on some already-known phenomena and focus on key features related to the retrieval accuracy, so that the section can be more refined. For example, the authors may simplify the narrative about the relationship between SSA and TOA radiance, the influence of AOD loading on SSA sensitivity… instead, incline more to the angular and wavelength variations of the SSA sensitivity and how do they influence the selections of viewing geometry and measurement bands.

L219: As we can see from Figure 2, ΔI for different wavelengths does not approach zero at 180° scattering angle.

Figures 2-4: What is the SSA value used for producing these figures? Is the value wavelength-dependent? Why is this value chosen and considered representative?

L226-228: Why does this sign reversal happen?

Figure 8 (a) and Figure 9 (a):

• What are the grids of latitude and longitude of these panels? Does the geolocation exactly match with (b) and (c)?
• The figures are not labeled with lowercase letters.

L325-328: It would be interesting to compare the retrieved AOD with the corresponding MODIS product, which adds more insights on the retrieval performance when the algorithm is applied to DPC data. The same comment for the following dust event.

L334: Please refer to the last comment.

L340-341: Looking at Figure 9, there is an obvious contrast in retrieved SSA between the region (21-23°N, 16-17°W) and the region (19-20°N, 17-19°W), while from the true color picture, the aerosols in these regions seem to be in the same type (mineral dust). What’s the reason for such a contrast in SSA? Is it an artifact of the algorithm?

L352-355: The global patterns retrieved here may not be quite comparable with the MODIS, OMI and POLDER products from the cited studies due to the differences in measurement period and wavelength. The comparison with the study of Dong et al. (TGRS, 2024) seems to bring more insights since the retrievals are from the same DPC measurements. Why don’t the authors present the retrievals at 670 nm, just the same wavelength showed by Dong et al. (2024), for a more straightforward comparison?

L325: “… composed of…” → be composed of