This paper produced a long-term SIF dataset (LHSIF) from 1995 to 2023. This topic is of significant importance for global-scale carbon simulation and vegetation studies. I think the datasets preprocessing procedures are solid. However, there are some questions I am concerned, especially about the reliability of the inter-annual trend in the SIF data from 1995 to 2000. There are many spelling errors in sentences throughout the paper. I listed as much as I could found. Please double check to avoid such errors and also try to avoid using ‘Figure.... illustrated...’ as the begining of a paragraph in the result section. Following a logical writing orders is necessary. I rcommend minor revision and here are some comments I wish the authors could give explanations before accepting this paper for publication.

Minor Reviews:

1) You didn’t mention how do you do quality control for GPP provides by FLUXNET. Did you use the quality flag data of “GPP_DT_VUT_REF” during SIF production?

• I am concerned whether the interannual trend in SIF before 2000 is true or reliable. For example, in Figure 7, we did not see much difference in yearly maximum global-averaged SIF based on the SIFu or SIFc from 2000 to 2023. The large discrepency mainly occurs in year 1995 to 2000, when SIFu is significant higher than the SIFc with a significnatly decreasing trend (mainly provided by GOME?). However, there exists larege discrenpency between SIFu and SIFc before 2000. How could you convience the readers that such inter-annual trend in SIF in year 1995 to 2000 is reliable? There is no validation of this trend from 1995 to 2000 as far as I can see in the paper.

• Line 21: You should better change a word to substitute ‘tool’ in line 21. How could a dataset be a tool? How about “Therefore, the long-term harmonized SIF dataset with a fine 0.05° resolution is valuable for estimating global photosynthesis over extended periods.”
• Missing space in multiple places in sentences.

Line 22: Missing space: “The LHSIF dataset is available at https://doi.org/10.5281/zenodo.14854185 (Zou et al., 2025).”

Line 138: There should be a comma in“0.5°×0.5°, and 1°×1°”.

Line 144: GPP (Berry et al)

Line 153: Missing space : 2 m.

Line 154: Missing comma: f(NIRv), f(VPD), and f(AT),

Line 159: “L-BFGS-B” algorithm (Byrd et al., 1995)

Line 167: Two balnk spaces here. “0.05̊  ，were”

Line 445: Missing space before the reference.

• Incomplete sentence in Line 267: “The temporal and spatial distributions of the spatial downscaling residuals were analyzed (Fig. 4). The residual was calculated as the difference between ? As shown...”
• L353: Please do grammer check and change this sentence to: “To highlight the overall correlation, the data in June for these two sites were removed from the scatter plots.”
• Line434: Upper foot label for m-2 in “residual less than 0.05 mW m^-2 sr^-1 nm^-1”.
• Line 303: please add an ‘and’ in “0.05; and significant increase”.

The author provides a systematic work for the development of the Long-term Harmonized multi-satellite SIF (LHSIF) dataset by coordinating SIF satellite observations from GOME, SCIAMACHY, GOME-2, and OCO-2. The TCSIF dataset, also developed by the same author, provides strong support for this study. The overall technical process is complete and feasible, and some interesting results are obtained. This study fits well within the scope of ESSD journal. However, I have some comments about the development process of this dataset. Therefore, my recommendation is major revision. Specific concerns and suggestions are outlined as follows:

Major Comments

1. I have concerns regarding the innovation and validation of the dataset. The authors mention two sets of long-term harmonized SIF products (Wang and Wen) and briefly describe their processing steps. However, neither in the Introduction nor in the Results/Discussion sections do I clearly see what specific methodological innovations the authors have introduced or where their data demonstrate superior performance. For instance, while they argue that the CDF method has limitations, there appears to be insufficient validation and justification—such as a comprehensive comparison of overall errors across the three datasets or scenario-specific analyses. I recommend that the authors conduct additional analyses to provide more quantitative evidence supporting the advantages of their dataset.
2. Several aspects of the validation remain unclear

Fig. 3: Why was only a single year (1998) selected for comparison to evaluate the data performance? This limited temporal scope may not sufficiently represent the overall dataset characteristics.

Long-term trend validation: The analysis appears to rely solely on the global average SIF shown in Fig. 5. Furthermore, the description in Lines 275-280 is largely qualitative. More quantitative metrics (e.g., statistical significance tests or error metrics) would strengthen the validation.

Fig. 8: What is the rationale for using annual maximum values rather than annual means? Maximum values are generally more susceptible to noise interference. Additionally, how was the claimed 'reduction in uncertainty' quantified?

Temporal inconsistency: Why does Fig. 8 cover 1996–2023 while Fig. 9 starts from 1995? This discrepancy in time ranges should be explicitly justified.

To clarify, these observations are not meant to imply that the authors' approaches are inherently flawed. However, clearer explanations for these methodological choices are necessary to ensure robust interpretation of the results.

3. In Section 4.1, the authors show that the overlap between GOME and SCIAMACHY SIF records is limited to only six months. However, the manuscript lacks a detailed explanation for addressing this issue within the current study. In fact, the challenge of short-term overlap is also present in this study. The limited temporal overlap constrains the representativeness of the GOME SIF data and may introduce uncertainty and potential biases in the mean-standard deviation-based matching approach. This issue is evident in Figures 7 and 8. For instance, in Figure 7, noticeable discrepancies between the corrected and uncorrected datasets are observed prior to 2003. It is worth noting that these early-period differences reflect the extent to which the correction improved the quality of this dataset. Therefore, I suggest that the authors further clarify the applicability and effectiveness of this matching method in mitigating biases arising from short-term overlaps, and possibly, supplement the analysis with a quantitative assessment of associated uncertainties.

4. The authors' descriptions in multiple sections of the manuscript are overly qualitative and subjective, lacking sufficient experimental support. It is recommended that the authors revise the relevant language to provide more detailed explanations or supplement with additional experiments.

Line 86; 268-269; 391; 394; 398;

5. To more comprehensively evaluate the quality of the LHSIF dataset, it is recommended that the authors incorporate the LCSIF dataset (Fang et al., 2025) for comparative experiments in their study. Comparing these two datasets can reveal the potential strengths and limitations of LHSIF, particularly in terms of long-term trend analysis and responses across different ecosystems. Furthermore, the long-term record of LCSIF can serve as a reference to validate the consistency and accuracy of LHSIF data during the earlier years.

Detailed Comments

1. Line 17: How did you define the reduction in overall error? Compared to what?
2. Line 27: garnered significant attention(s)
3. The second and third paragraphs can be merged to one paragraph after reducing some sentences about SIF-GPP relationship which is not the focus in this study.
4. Line 65: delete “itself”
5. Line 70-74: Could you provide a direct explanation of where Wang's research may have fallen short or failed to consider comprehensively?
6. Line 84-85: Wen's research has significant overlap with yours, so I'm unclear why it was only mentioned so briefly and at such a late stage.
7. Line 87: ‘A precise, reliable, harmonized, and global high-resolution SIF dataset is not yet available for long-term vegetation monitoring.’ Based on the preceding sections of the introduction, the authors do not appear to have clearly explained why the prior data were neither precise nor reliable.
8. Line 131: Oco-2/Oco-3 → OCO-2/OCO-3
9. Line 207: Is this spatial resolution (0.072727° × 0.072727°) commonly referred to?
10. Line 221: There appear to be numerous NDVI products based on AVHRR data. Is this a newly developed set? Could you please provide some additional introduction?
11. Line 238-243: Could be more quantitative
12. Figure 5: the order of GOME and SCIAMACHY in the legend is opposite
13. Line 276: as the difference between ? some words missed
14. Line 340-345: These belong to Methods. Please check throughout the manuscript.
15. Line 343: please add the citation for this factor
16. Lines 26, 144, 158, etc.: please add the space between the text and the brackets
17. Section 4.1: Could you provide more concrete examples demonstrating the advantages of using TCSIF compared to the uncorrected GOME-2A data employed in the other two studies?
18. Line 374: Missing word.
19. Line 385-387: Providing concrete examples rather than textual descriptions would enable readers to better understand.
20. Line 391: Please see above.