This study demonstrates the application of two methods for integrating MODIS and MISR satellite data with ground-based observation. The content of the study may be solid, but it is not well-organized and the meaning is not sufficiently emphasized. Before proceeding to the next step, this study/manuscript requires at least an essential revision.

General comments:

• In the abstract section, it seems that the study employed two methods, namely the UK and RK-ML. These two methods have both been suggested to be effective for fusing AOD data in the Indian region in this study. However, the relationship between these two methods is not clear. They seem to be completely separate from each other. Furthermore, do these two methods have distinct advantages in addressing the AOD fusion issue? What are the main contributions of this study? Is the proposed method an existing one or has it been specially modified/adjusted for the India region? These contents should also be reflected in the abstract section.
• The introduction section also should be revised and improved, mainly in two aspects: (1) Some of the current advanced research findings have not been mentioned or introduced. For example, regarding synergistic inversion (10.1029/2024GL113448; 10.5194/amt-18-7679-2025), data filling (10.5194/essd-16-2425-2024; 10.5194/amt-17-4317-2024), and some physical-based ML method (10.1016/j.rse.2023.113763). These three aspects, although not the main focus of the study, are still closely related to the topic. (2) In the second paragraph of the introduction, the author lists a large number of methods (Line 67-81). But these methods were basically not discussed at all. Although I do not object to the author's highlighting the advantages of the Kriging method, a brief description of the advantages, disadvantages and applicable scope of various methods should also be provided. Otherwise, the advancement/necessity of the research methods will be difficult to be shown.
• I noticed that the grid used in this study is 0.5°, which is commonly employed in some atmospheric models. But for a regional study, this resolution might be rather coarse, especially since the satellite data used has a higher spatial resolution (~ 10 km and 4.4 km). Is this step of reducing the resolution necessary? What studies does the merged data in this study serve for?
• The author has provided a detailed explanation of the calculation process of methods such as Variogram Analysis and UK, which is good. However, there seems a lack of description regarding the application region of these methods. For instance, the author states that the ground AOD is regarded as the response variable, while the satellite AOD is considered as the regressors (Line 302) in the final spatial fusion. Does this above process also apply to the gap filling step? Were the MODIS and MISR data processed for gap filling separately? Were MODIS and MISR data used simultaneously during the fusion process? Have these two sets of data taken into account the differences in their accuracy (or EE%)?
• The discrepancy between the ground data and the satellite data is what I am concerned about. MODIS observations (Figure 3) have revealed the familiar AOD distribution pattern in South Asia. However, it seems to have a significant difference from the ARFINET results obtained through ground-based observations. Especially in the high-value areas in the north, these data seem to have been completely unobserved from ARFINET. Without considering the issue of the instrument itself, this phenomenon might be caused by the algorithm mistakenly classifying haze (high aerosol loading) as clouds (10.1109/TGRS.2023.3252264) and this caused all the high values to be blocked. However, in any case, if there is a significant discrepancy or difference between the ground data and the satellite data in the overall trend distribution, this will have an impact that cannot be ignored on the subsequent results.
• Is there a way to simply evaluate the interpolated results? Otherwise, the information conveyed in 3.3.2 section is rather limited.
• Is the fused AOD (Figures 4, 5, 6) calculated using UK method? Please Clarify. Furthermore, I can observe some high-value points/regions in the figure, such as Figure 5k and Figure 6f. Its value is higher than that of MODIS and MISR simultaneously, but there is no supporting data from ground stations. Therefore, it is difficult to explain. Some recent Publications should be discussed, for example, Significant uncertainties from overlooking aerosol-cloud coexistence in surface solar radiation estimates using passive satellite observations；Effects of Different Types of Aerosols on Diffuse Radiation Based on Global AERONET；First high temporal resolution retrievals of AOD over shallow and turbid coastal waters for Himawari-8； Synergistic Estimation of Surface Particulate Matter and Ozone Pollutants to enhance accuracy and interpretability by a Deep Learning Approach； A Physics-Guided Neural Network Model to Estimate All-Sky Diffuse Solar Radiation Using Himawari-8 Data。
• Can we increase the contrast/validation between the fusion results and the ground, as shown in Tables 4 and 5?
• The Limitation section (3.3.6) is not good here. Discussions on minor technical details will increase doubts about your method. Such as many of the contents in this section have not been discussed or elaborated upon in the previous text. If one wants to discuss the limitations of the research, it should be explained from the perspective of scientific methodology.

Specific comments:

• Line 102: The full name of the ARFINET site should be provided, and a brief introduction to the development history of this network should also be given.
• Line 106: what is the frequency of data observation?
• Line 140: When estimating the AOD at 550nm using the wavelength-dependent relationship of AOD, which two wavelength bands were used for the calculation?
• Line 183-186: Specifically, what methods and concepts were introduced?
• Section 2.3.4: The first paragraph of this section seems more appropriate to be placed in the introduction section.
• Line 381: What does "LOOCV" mean? Is it for a certain range or a specific site? Furthermore, in some machine learning scenarios, cross-validation based on the site, time, and total sample is also considered to test the overall stability of the model (10.1109/TGRS.2026.3657522).
• Line 383: Compared with the unbiased estimation model performance, LOOCV is more focused on evaluating the generalization ability of the model, such as in different time and spatial regions.
• Line 398: Apart from training and testing, are there any uses of individual sample set for validation?
• Line 400-411: This part of analysis is not closely related to the content of this section. It is recommended to remove it or reorganize it into the introduction part.
• Figure 2: Is there any display of data from AERONET (It was also used in this study)? Furthermore, it would be better to use different symbols to represent different sites/
• Line 465: why only using data from 2012, 2016, and 2021?
• Line 501: Multi-angle measurement makes it easier to describe the anisotropic scattering of the surface (BRDF). However, under high AOD conditions, since the surface reflection is smoothed out by thickness aerosol, it may instead lead to an underestimation of AOD (Then, the contribution of some aerosols in total signal will be wrongly attributed to the surface).
• Figure10: Please clarify the color indications in the column chart.
• It is clear that RK-ML is better than UK, rather than being similar to UK and RK-ML as indicated in the Abstract.

Major Comments:

Authors employ two-stage Universal Kriging (UK) method for the synergistic fusion of aerosol optical depth dataset from MODIS and MISR, and ground-based ARFINET data over the Indian subcontinent. The seasonal fused maps of AOD integrates more than 80% of the ground-based observations, highlighting the impacts of number of ground measurements in the fusion process. Integrating Machine Learning (ML) approach with the Residual Kriging is found to capturing stable spatial patterns under limited or sparse coverage of ground observations.

The paper feels lengthy, but informative and presented with sufficient details (i.e., text, maps, and plots). One primary concern with this paper is that the fusion approach has been applied to the monthly mean MODIS/MISR aerosol products, which are already providing spatially extensive distribution of AOD in the first place. The improvements demonstrated by authors add AOD data over very northern parts of India, around Kashmir region, where AOD is very low.

However, an attractive and novel result of the presented work is the integration of ground-based ARFINET AOD observations in the fused maps, which improves the aerosol representation over and around the ground sites.

Another concern is about the choice of satellite sensors. MISR’s repeat overpass occurs every 4 days, limiting the spatial coverage and its ability to capture daily changes in aerosol loading. MODIS flies on Terra (morning overpass) and Aqua (early afternoon overpass). The Dark Target and Deep Blue combined AOD product used in the present work is available from both satellites. Utilizing these two MODIS sensors in the fusion process would have offered more complete spatial coverage on daily basis, in addition to the diurnal variations in aerosol patterns.

Additionally, an application of the UK-ML approach on daily AOD fields would have added more value than the presented monthly scale, at which the satellite aerosol products fill the gaps and provide more extensive spatial coverage.

The paper is well-written with academic standard usage of English. The paper is lengthy, but it is understood that authors wanted to provide all details of the adopted UK-ML approach and its application to the satellite-ground aerosol dataset.

Authors are expected provide explanations on the concerns listed above. Currently, the paper is ranked between the major and minor revision. Specific comments on the paper are provided below.

Specific comments:

Abstract: Satellite and ground-based datasets should be introduced first in the abstract, followed by Universal Kriging and ML approaches. The abstract doesn’t give an impression of how UK-ML approaches improve spatial distribution of AOD. Have authors validated the fused AOD dataset? Although, ground-based ARFINET data is used in the fusion, the AOD still can be validated against available AERONET sites in India. Also, it is unclear how the suggested fusion approach optimally uses either MODIS, MISR, and ARFINET AOD datasets. Abstract should address these concerns.

Introduction:

The use of AERONET AOD measurements is mentioned here (2nd paragraph), but not in the abstract.

2.1 Ground-based AOD:

Which agency operates ARFINET? ISRO? This should be cited here.

Are the measurements from MWR made simultaneously across all ten narrow wavelength filters?

Does the variance of the Langley intercept cause an uncertainty of ~ 5% in AOD derivation?

A map of ARFINET operating sites would be desirable.

2.2 Satellite retrieved AOD

Note that Dark Target and Deep Blue algorithm retrieve AOD using two different respective algorithms. The combined product is derived by selecting the best retrieval from either of the algorithms (i.e., DT over darker surfaces and DB over semi-arid and arid surfaces).

2.3.2 Variogram Analysis

Fig S1: the detrending process is not well understood. Did author use long-term MODIS AOD datasets over India? Does this include all four seasons?

Figure 3: Similar long-term seasonal climatology maps from MISR are wroth showing here to examine consistency or lack thereof.

3.2 Inter-comparison of satellite-ground AOD

Lines 466-473: This description appears belonging to Figures S7-S9, not S4-S6. Please check and verify. Also, add a brief discussion on satellite vs. ground AOD comparison (MISR underestimates ground AOD at moderate to high aerosol loadings).

Lines 495-503: This paragraph should go to the beginning of the section 3.2 (also change Figure numbers S7-S9 to S4-S6).

Section 3.3.3: Up to this point, it was unclear whether daily or monthly MODIS/MISR dataset is used in the fusion process. It looks like monthly datasets are used. Please confirm.