This study proposed a method to correct the angular effects, spatial mismatch, and small holes in daily NTL data and generated a wonderful product of high-quality daily NTL data covering more than 650 cities across the world. This a very valuable dataset, and the generation procedures and validation assessments are reasonable and solid. I think it should be published after a revision.

Comments:

1. In Line 97-98, the authors mentioned that the knowledge of the local building environment is important for identifying the sources of light. I am not so sure why the authors paid attention to this? It looks like there are no associations with the HDNTL generation.
2. In Line 174, what does the "fixed area" mean? An area with stable light emissions?
3. Adding a scale bar in Fig. 4 and 5 can help us tell if the current cases are solid or not. For example, in Fig. 5, if a 500x500m pixel is much larger than a flyover, the flyover’s surrounding light emissions could affect the validation.
4. In Fig. 10, what do (a) and (b) refer to?
5. I found that all pixels in the 2021-6-9 NTL data of Fuzhou, Fujian, China are 0, as I known Fuzhou is rainy that day. So I am wondering if the method proposed in this study can correct the daily NTL data on a cloudy or rainy day, especially during a continues cloudy/rainy days?
6. What is the maximum/optimal size of holes that the spatiotemporal interpolation can deal with?

This paper presents a valuable contribution to the field of nighttime light remote sensing by addressing the limitations of the VNP46A2 dataset and producing a high-quality daily nighttime light (HDNTL) dataset for 653 cities. The proposed framework for error correction, interpolation, and validation demonstrates significant improvements in data accuracy and reliability. The results are well-supported by comparisons with other datasets and case studies.

1. I suggest the authors should provide a more detailed explanation for choosing 3 times the standard deviation as the threshold for identifying NTL variation (𝑁𝑇𝐿𝑣𝑎𝑟𝑖𝑎𝑡𝑖𝑜𝑛). This threshold lacks a clear justification based on the data characteristics or statistical principles.
2. The use of mean NTL radiance for angular effect correction may introduces some uncertainty. Maybe the authors can consider exploring methods such as fitting a model to the relationship between NTL pixels for each VZA. This could provide a more accurate and robust correction, particularly for complex urban environments.
3. The paper only uses one flyover site to validate the effectiveness of spatial mismatch correction. It is recommended to include additional examples to strengthen the evidence of the correction’s generalizability.
4. The sudden increase in RMSE for the temporal window size of 4 in Figure 7 requires further investigation and explanation.
5. The paper mentions the “mutation signal recognition threshold” without providing a clear definition.

This paper utilizes an improved SFAC method to correct the VNP46A2 NTL product, specifically addressing the VZA effect and mismatches. These corrections enhance the stability of the NTL time series while preserving critical anomalies caused by events such as natural disasters or holidays. The overall structure of the manuscript is clear; however, it lacks explanation of some key procedures and a deeper discussion of results. Therefore, my recommendation is a major revision. My specific comments are as follows:

1. On page 5, line 125, the phrase “Grow a square with... is less than 1” is unclear. What exactly does “less than 1” refer to in this context
2. On page 5, line 127, the authors mention that in highly urbanized areas, overlapping convex hulls may emerge. To address this, a manual adjustment method is employed. However, the manuscript does not specify the criteria for this manual adjustment, nor does it describe the resulting delineation of urban areas after resolution.
3. On page 6, line 149, the data used to validate the correction results is from Tianjin City, covering the period from January 25 to February 21, 2022. Clearly, relying on a single region and a one-month span is not robust. Is there a way to increase the volume of validation data? If acquiring more data from SDGSAT is not feasible, could alternative validation methods be considered?
4. On page 8, line 182, the authors claim that “it is reasonable to assume that the annual lowest light intensity is close to the NTL_fix.” However, the rationale behind this assumption is not provided. Why wouldn’t the annual mean light intensity be a better approximation to NTL_fix?
5. On page 9, line 204, the authors divide the VZA into 16 groups. How exactly was this division made? Was the 0–60° range evenly partitioned into 16 intervals? Although the grouping is said to be based on the 16-day revisit cycle, in practice, very similar VZA values can occur within the same cycle. For example, in a single cycle, if one day has a VZA of 54° and another day 58°, would these be grouped together?
6. On page 9, line 227, the authors apply an inverse distance weighting to non-central pixels in a 3×3 window. However, in such a window, the distances to the center pixel are identical for all non-central pixels, resulting in identical weights. What then is the significance of applying this weight?
7. On page 11, line 270, the authors select the Liuyang Firework Festival as a case study to test whether their method suppresses genuine NTL anomalies. However, aerosols from fireworks can significantly affect NTL intensity, especially during windless nights. Have the authors considered this potential confounding factor?
8. In Figure 5, the selected flyover region seems unsuitable for validation purposes, as it is surrounded by many tall buildings within a 500-meter radius, which could introduce substantial VZA effects.
9. In Figure 5, the IR values derived from the original VNP46A2 data and from the HDNTL product are almost identical. Does this imply that the proposed method is ineffective in this case? This may support the previous concern regarding the selection of this flyover site.
10. In the Beirut port explosion case shown in Figure 9, why does the HDNTL product not show the NTL intensity on the day of the explosion as the lowest of the year? The authors should provide an explanation for this discrepancy.

This study generated daily nighttime light (NTL) data for 600 more global large cities. Since NTL is the only satellite remote sensing data to show the nighttime activities on Earth, this dataset is highly desirable for urban studies. Spatial mismatch and angular effect are two major problems which lead to big uncertainties in the daily NTL time series, so this study improves the data quality of daily NTL by correcting these two effects. In addition, it fixes the small holes in cloud-free images to further improve the data availability. Generally speaking, the methodology used by this study is correct and the produced data has high value in applications. I have some specific comments for the authors to further improve this study.

1. Abstract: I would suggest using r instead of r2 to report the similarity between interpolated and reference data.
2. Introduction: The authors mentioned that the daily NTL is affected by external factors such as measurement errors and retrieval uncertainties. It is not clear what measurement errors are. In addition, NTL is easily affected by clouds, which is a big problem in applications. I understand that cloud removal is not done by this study, but the authors should clarify this at the very beginning to avoid misleading readers.
3. The authors argued that the spatial mismatch, angular effect, and missing data holes introduce significant uncertainties in daily NTL. May I know how much of the uncertainties were caused by these three issues? Some examples can be used to explain this. It can help readers understand the necessity of correcting these issues.
4. Study area: The authors talked about the situation of 2030, but they used the 2025 population to select cities as their study areas. It is a little bit confusing.
5. Page 5 line 125: Most cities’ built-up areas are within the ROI. How many cities are not completely covered by the ROI? If a city has a large area not covered by the ROI, was it manually adjusted?
6. Page 9 line 205: Please clarify how to handle the situation if the near-nadir observations were contaminated by clouds.
7. Page 9 line 215: These spatiotemporal methods have been developed. Why not directly use them for interpolating holes?
8. Figure 5: Only one example of a flyover case was shown. It is better to show more cases. I am not sure whether flyover has no angular effect.