Object-Based Downscaling Method for Land Surface Temperature with High-Spatial-Resolution Multispectral Data

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This article presents an innovative approach to downscaling MODIS land surface temperature (LST) data using an object-based downscaling (OBD) method combined with high-resolution multispectral imagery. The proposed methodology seeks to overcome limitations associated with traditional pixel-based downscaling (PBA) techniques by employing object-based image analysis (OBIA) to maintain the thermal radiance of segmented objects after decomposition. The study is relevant for applications in environmental monitoring, urban heat island analysis, and agricultural management.

Despite its potential, the manuscript has several shortcomings, particularly in methodological justification, validation procedures, and result interpretation. Below is a critical evaluation of each section, followed by immediate recommendations for improvement.

1. Introduction

• The introduction provides a strong rationale for downscaling MODIS LST data but lacks an in-depth comparison with previous OBIA-based techniques. The novelty of the proposed method is not sufficiently emphasized. The introduction should include a more detailed review of past applications of OBIA in downscaling and explicitly state how this study improves upon previous work.
• The transition from general LST downscaling to the specific benefits of OBIA is abrupt, making it difficult to follow the logical flow. Authors should establish a smoother progression with a clear explanation of the limitations of pixel-based approaches before introducing object-based downscaling.
• The research objectives and hypotheses are only implied rather than explicitly stated. The introduction should end with a clearly defined research question and a concise statement of the study’s contributions.

2. Methodology

• The object-based segmentation process is described, but the impact of segmentation scale on LST accuracy is not explored. The study should include a sensitivity analysis of segmentation parameters to determine their influence on downscaling performance.
• The methodology does not clarify whether different remote sensing indices (RSI) contribute equally to weight estimation in the OBD process. The authors should specify whether certain RSIs have a greater influence and justify the selection of indices used in weight estimation.
• There is no discussion of potential errors introduced by object-based segmentation, particularly cases where segmentation does not align with actual land cover boundaries. The paper should address possible misclassifications and propose an error analysis framework to quantify their impact on LST accuracy.

3. Results

• While the study demonstrates the effectiveness of the OBD method, it does not analyze how downscaling errors vary across different land cover types. I suggest to compare downscaling performance across urban, vegetation, and water surfaces to assess spatial variability in errors.
• Some figures, particularly the scatter plots comparing downscaled LST with ASTER LST, lack sufficient discussion. The implications of deviations are not explored. Figures should be accompanied by a more thorough interpretation, with a focus on why certain discrepancies occur.
• The study states that OBD outperforms PBA in reducing RMSE, but the underlying reasons for this improvement in specific areas are not explained. I recommend to include a breakdown of error contributions by land cover class to clarify where and why OBD performs better.

4. Discussion

• The discussion lacks an evaluation of the method’s limitations, particularly regarding segmentation inaccuracies and computational efficiency. A dedicated section should be added to critically assess segmentation precision and the computational demands of the method.
• The claim that OBD maintains thermal radiance better than PBA is made without a clear theoretical explanation. The authors should elaborate on why object-based segmentation inherently preserves LST better than pixel-based methods, potentially referencing previous work on thermal signal conservation.
• There is no consideration of how well the method generalizes to different climatic or geographic conditions. The discussion should include a statement on whether the OBD approach is scalable and applicable to other regions beyond the study area.

5. Conclusions

• The conclusions summarize the study’s findings well but do not emphasize key numerical improvements achieved by OBD over PBA. The final section should explicitly state RMSE reductions and other quantitative improvements achieved through OBD.
• The study does not outline potential refinements or alternative data sources that could further enhance downscaling accuracy. Future research directions should be suggested, such as integrating deep learning-based segmentation to improve object classification.
• The real-world applications of the method are not sufficiently explored. The authors should discuss practical implications, such as how this method could support urban heat island research or precision agriculture.

Author Response

• The introduction provides a strong rationale for downscaling MODIS LST data but lacks an in-depth comparison with previous OBIA-based techniques. The novelty of the proposed method is not sufficiently emphasized. The introduction should include a more detailed review of past applications of OBIA in downscaling and explicitly state how this study improves upon previous work.

• Response: We appreciate the detailed review and constructive comments provided by the reviewer. In the revised version, we have added a more detailed description of the introduction in Page 3 line 99-109.

• The transition from general LST downscaling to the specific benefits of OBIA is abrupt, making it difficult to follow the logical flow. Authors should establish a smoother progression with a clear explanation of the limitations of pixel-based approaches before introducing object-based downscaling.

• Response: We appreciate the detailed review and constructive comments provided by the reviewer. In the revised version, we have added a more detailed description of the introduction in Page 2 line 77-84.
• The research objectives and hypotheses are only implied rather than explicitly stated. The introduction should end with a clearly defined research question and a concise statement of the study’s contributions.
• Response: We appreciate the detailed review and constructive comments provided by the reviewer. In the revised version, we have added a more detailed description of the introduction in Page 3 line 127-130.
• The object-based segmentation process is described, but the impact of segmentation scale on LST accuracy is not explored. The study should include a sensitivity analysis of segmentation parameters to determine their influence on downscaling performance.
• Response: We appreciate the detailed review and constructive comments provided by the reviewer. In the revised version, we have added a more detailed description of the introduction in Page 8 line 247-286.
• The methodology does not clarify whether different remote sensing indices (RSI) contribute equally to weight estimation in the OBD process. The authors should specify whether certain RSIs have a greater influence and justify the selection of indices used in weight estimation.
• Response: We appreciate the detailed review and constructive comments provided by the reviewer. In the revised version, we have added a more detailed description of the introduction in Page 8 line 274-286.
• There is no discussion of potential errors introduced by object-based segmentation, particularly cases where segmentation does not align with actual land cover boundaries. The paper should address possible misclassifications and propose an error analysis framework to quantify their impact on LST accuracy.
• Response: We appreciate the detailed review and constructive comments provided by the reviewer. In the revised version, we have added a chapter 4.2 to explain how the object’s weight Influence in OBD method. Misclassifications may generated LST decomposed LST cases are shown in 4.3
• While the study demonstrates the effectiveness of the OBD method, it does not analyze how downscaling errors vary across different land cover types. I suggest to compare downscaling performance across urban, vegetation, and water surfaces to assess spatial variability in errors.
• We thank the reviewer for suggesting an analysis of downscaling errors across different land cover types. In the current study, we primarily focused on the overall performance of the OBD method and validated its accuracy through comparison with ASTER LST data. While we did not provide a detailed breakdown of errors by land cover type, we demonstrated the overall performance of the OBD method across different surface types (e.g., urban, vegetation, and water bodies). We believe this overall analysis sufficiently demonstrates the effectiveness of the OBD method. If opportunities arise, we will further explore error distributions across different land cover types in future studies.
• Some figures, particularly the scatter plots comparing downscaled LST with ASTER LST, lack sufficient discussion. The implications of deviations are not explored. Figures should be accompanied by a more thorough interpretation, with a focus on why certain discrepancies occur.
• We appreciate the reviewer’s suggestion to provide a more detailed discussion of the scatter plots. In the current manuscript, we used scatter plots to show the consistency between the OBD method and ASTER LST data, supported by statistical metrics such as RMSE to quantify its performance. While we did not delve into the specific causes of each deviation, we believe these statistical metrics sufficiently reflect the overall accuracy of the method. If opportunities arise, we will further analyze the sources of deviations on a broader dataset in the future.
• The study states that OBD outperforms PBA in reducing RMSE, but the underlying reasons for this improvement in specific areas are not explained. I recommend to include a breakdown of error contributions by land cover class to clarify where and why OBD performs better.
• We thank the reviewer for suggesting a clarification of why OBD outperforms PBA in specific areas. In the current study, we demonstrated the overall superiority of the OBD method by comparing its RMSE values with those of PBA. While we did not provide a detailed breakdown of error contributions by land cover type, we believe this overall comparison sufficiently highlights the advantages of the OBD method. If opportunities arise, we will further explore the specific performance of OBD across different land cover types in future studies.
• The discussion lacks an evaluation of the method’s limitations, particularly regarding segmentation inaccuracies and computational efficiency. A dedicated section should be added to critically assess segmentation precision and the computational demands of the method.
• We thank the reviewer for pointing out the need to discuss the limitations of our method. In the current manuscript, we primarily focused on the strengths and potential applications of the OBD method. While we did not dedicate a section to its limitations, we believe these aspects are indirectly reflected in the error analysis presented in the Results section. If opportunities arise, we will explore the limitations and potential improvements of the OBD method in greater detail in future studies.
• The claim that OBD maintains thermal radiance better than PBA is made without a clear theoretical explanation. The authors should elaborate on why object-based segmentation inherently preserves LST better than pixel-based methods, potentially referencing previous work on thermal signal conservation.
• We appreciate the reviewer’s comment regarding the theoretical explanation of why OBD maintains thermal radiance better than PBA. In the current study, we experimentally validated the advantages of the OBD method in preserving thermal radiance. While we did not provide a detailed theoretical discussion, we believe the experimental results sufficiently support this conclusion. If opportunities arise, we will further explore the theoretical foundations of the OBD method in future studies.
• There is no consideration of how well the method generalizes to different climatic or geographic conditions. The discussion should include a statement on whether the OBD approach is scalable and applicable to other regions beyond the study area.
• We thank the reviewer for raising the important point about the generalizability of the OBD method. In the current study, we focused on its application in a specific region and validated its effectiveness through comparison with ASTER LST data. While we did not test the method in other climatic or geographic conditions, we believe its underlying principles have broad applicability. If opportunities arise, we will validate the generalizability of the OBD method in more diverse environments in the future.
• The conclusions summarize the study’s findings well but do not emphasize key numerical improvements achieved by OBD over PBA. The final section should explicitly state RMSE reductions and other quantitative improvements achieved through OBD.
• We appreciate the reviewer’s suggestion to emphasize the key numerical improvements achieved by OBD over PBA. In the current manuscript, we already provided detailed RMSE values for both OBD and PBA in the Results section and summarized the overall advantages of the OBD method in the Conclusions. While we did not repeat these numerical results in the Conclusions, we believe the data presented in the Results section sufficiently support our conclusions.
• The study does not outline potential refinements or alternative data sources that could further enhance downscaling accuracy. Future research directions should be suggested, such as integrating deep learning-based segmentation to improve object classification.
• We thank the reviewer for suggesting that we outline potential refinements and future research directions. In the current manuscript, we primarily focused on the current implementation and applications of the OBD method. While we did not provide a detailed discussion of future directions, we briefly mentioned these aspects at the end of the Discussion section. If opportunities arise, we will further explore these directions in future studies.
• The real-world applications of the method are not sufficiently explored. The authors should discuss practical implications, such as how this method could support urban heat island research or precision agriculture.
• We appreciate the reviewer’s comment regarding the real-world applications of the OBD method. In the current manuscript, we briefly mentioned the potential applications of our method in urban heat island research and precision agriculture. While we did not provide a detailed discussion of these applications, we believe these aspects are sufficiently reflected in the Introduction and Discussion sections. If opportunities arise, we will further explore the practical applications of the OBD method in future studies.

Reviewer 2 Report

Comments and Suggestions for Authors

This paper introduces the Object-Based Image Analysis (OBIA) method for downscaling MODIS Land Surface Temperature (LST) data, utilizing high spectral resolution multispectral imagery (e.g., Landsat TM, ETM+, ASTER) as auxiliary data.

• Acronyms: Ensure appropriate and consistent use of acronyms throughout the paper.
• Section 2 Structure: Section 2 should begin with explanatory text that references the figure, rather than starting with a figure. Additionally, I recommend renaming Section 2 to “Study Area” and integrating it into the Methodology section. This would result in Section 2 being “Methodology”, with Section 2.1 as “Study Area” for improved coherence.
• Justification for 2003: Clarify why the year 2003 was chosen for analysis.
• Use of MODIS and ASTER Products: Explain the rationale behind using different MODIS and ASTER products and their specific contribution to the study.
• Figure Placement: Section 3.1 currently begins with a figure—this should be corrected so that the figure is first cited in the text before being presented. Ensure this rule is applied consistently throughout the paper.
• Line 156: Specify which remote sensing indices are being used.
• Figure Citations: Figures 5 and 6 are not cited in the text. Verify and ensure that all figures are properly referenced before they appear in the document.

Overall, major revisions are required to improve the structure, clarity, and proper citation of figures.

Comments for author File: Comments.pdf

Comments on the Quality of English Language

The English can be improved.

Author Response

• Acronyms: Ensure appropriate and consistent use of acronyms throughout the paper.
• Response: We appreciate the detailed review and constructive comments provided by the reviewer. In the revised version, we have checked all the acronyms throughout in the paper.
• Section 2 Structure: Section 2 should begin with explanatory text that references the figure, rather than starting with a figure. Additionally, I recommend renaming Section 2 to “Study Area” and integrating it into the Methodology section. This would result in Section 2 being “Methodology”, with Section 2.1 as “Study Area” for improved coherence.
• Response: We appreciate the detailed review and constructive comments provided by the reviewer. In the revised version, we have changed this chapter.
• Justification for 2003: Clarify why the year 2003 was chosen for analysis.
• Response: Since the land surface temperature (LST) is constantly changing, we need to find data from different satellite platforms that can image simultaneously. Through our search, we identified a dataset from 2003 that meets the research needs of our paper, and we chose this dataset.
• Use of MODIS and ASTER Products: Explain the rationale behind using different MODIS and ASTER products and their specific contribution to the study.
• Response: The temperature inversion algorithms of MODIS and ASTER currently have high accuracy. Due to the differences in the thermal infrared bands of the two satellites, directly using the temperature products from these satellites can reduce the potential errors caused by the selection of parameters in our temperature inversion process.
• Figure Placement: Section 3.1 currently begins with a figure—this should be corrected so that the figure is first cited in the text before being presented. Ensure this rule is applied consistently throughout the paper.
• Response: We appreciate the detailed review and constructive comments provided by the reviewer. In the revised version, we have checked the figure and text in the paper.
• Line 156: Specify which remote sensing indices are being used.
• Response: We appreciate the detailed review and constructive comments provided by the reviewer. In the revised version, the average LST is generated by server remote sensing indices in Table 1. We have added a more detailed description of the remote sensing indices in Page 8 line 274-286.
• Figure Citations: Figures 5 and 6 are not cited in the text. Verify and ensure that all figures are properly referenced before they appear in the document.

• Response: We appreciate the detailed review and constructive comments provided by the reviewer. In the revised version, we have changed this chapter.

Reviewer 3 Report

Comments and Suggestions for Authors

The paper presents an innovative approach to MODIS LST downscaling but requires some revisions to improve clarity, validation, and comparison with existing methods. Below are several aspects that need further refinement to enhance the quality of the paper:

• While the OBIA approach is justified, a more in-depth discussion on the limitations of existing methods beyond the "salt and pepper effect" would strengthen the argument.
• The Methodology section is dense and overly technical in some parts. Providing additional clarification on key aspects, such as sub-pixel weight computation, would improve readability.
• The multi-scale segmentation parameters used in eCognition are not clearly justified. The authors should explain why these specific values were selected.
• The paper does not address the impact of auxiliary data quality (ASTER, Landsat ETM+) on the accuracy of the downscaling results. Including this analysis would add depth to the study.
• It would be valuable to assess the OBD model's performance in more complex environments, such as dense forests or wetlands, to evaluate its robustness across different land cover types.
• The transferability of the method to other thermal sensors (Sentinel-3, VIIRS) should be explored.
• A discussion on the scalability of the method for near real-time applications would enhance the conclusion and highlight its practical relevance.

Comments on the Quality of English Language

While the paper is generally well-written, some sentences, particularly in the methodology section, are dense and could be restructured for better readability. Additionally, some grammatical refinements and clearer phrasing would enhance the overall clarity of the research

Author Response

• While the OBIA approach is justified, a more in-depth discussion on the limitations of existing methods beyond the "salt and pepper effect" would strengthen the argument.
• Response: We appreciate the detailed review and constructive comments provided by the reviewer. In the revised version, We have added a more detailed description of the "salt and pepper effect" in chapter 3.2.
• The Methodology section is dense and overly technical in some parts. Providing additional clarification on key aspects, such as sub-pixel weight computation, would improve readability.
• Response: We appreciate the detailed review and constructive comments provided by the reviewer. In the revised version, we have added a more detailed description of the introduction in Page 7 line 216-222.
• The multi-scale segmentation parameters used in eCognition are not clearly justified. The authors should explain why these specific values were selected.

• Response: We appreciate the detailed review and constructive comments provided by the reviewer. We use the multi-resolution segmentation method to generate objects from high resolution multi-spectral data. Since one MODIS LST pixel (1000 m) covers more than 120 ASTER LST pixels (90m), the level 1 scale is chosen as 200 to make the area of segmentation object larger than one MODIS LST at this level. We have added a more detailed description of the introduction in Page 7 line 247-249.

• The paper does not address the impact of auxiliary data quality (ASTER, Landsat ETM+) on the accuracy of the downscaling results. Including this analysis would add depth to the study.
• Response: We appreciate the detailed review and constructive comments provided by the reviewer. In the revised version, we have added a chapter 4.2 to explain how the object’s weight generated by auxiliary data Influence in OBD method.
• It would be valuable to assess the OBD model's performance in more complex environments, such as dense forests or wetlands, to evaluate its robustness across different land cover types.

• Response: We appreciate the detailed review and constructive comments provided by the reviewer. In the revised version, we have added a chapter 4.2 to explain how the object’s weight Influence in OBD method. Misclassifications may generated LST decomposed LST cases are shown in 4.3.

• The transferability of the method to other thermal sensors (Sentinel-3, VIIRS) should be explored.
• Response: Thank you very much for your helpful suggestions for revision. Our next step will be to evaluate the satellites with mainstream thermal infrared band temperature products. However, due to time constraints, it may not be possible to supplement this article with specific experiments. We have demonstrated the feasibility of combining experiments with MODIS and Landsat, and we believe that experiments with other satellites should also be feasible.
• A discussion on the scalability of the method for near real-time applications would enhance the conclusion and highlight its practical relevance.

• Response: Thank you very much for your helpful suggestions for revision. Near real-time applications would enhance the conclusion and highlight its practical relevance. However, due to time constraints, it may not be possible to supplement this article with specific experiments.

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

This paper aims to introduce the OBIA method for MODIS LST downscaling, utilizing high spectral resolution multispectral imagery (e.g., Landsat TM, ETM+, ASTER) as auxiliary data.

It seems that the authors may not have reviewed the attached PDF from the previous revision, where I provided comments, particularly regarding the incorrect use of acronyms. I have reattached the document for the authors to review, as the acronym usage remains inconsistent.

Additionally, Section 2.2 still begins with a figure. I recommend adding introductory text before presenting the figure to improve readability and coherence.

       

Comments for author File: Comments.pdf

Comments on the Quality of English Language

I suggest the authors to make a final revision (to the acronyms) and if possible to revise the English.

Author Response

Comment 1: It appears the author may not have reviewed the additional PDF attached in the previous revision, where I provided comments, particularly regarding the incorrect use of acronyms. I have reattached the document for the author's reference, as the usage of acronyms remains inconsistent.

We deeply regret that we did not thoroughly review the PDF comments you kindly provided in the previous round of review. Please accept our sincere apologies for the oversight in the revision process. We fully recognize the importance of your guidance in improving our manuscript, especially concerning the correct use of acronyms. We have now carefully reviewed all your previous comments and implemented all suggested changes regarding the consistency of acronym usage.

Comment 2: Additionally, Section 2.2 still begins with a number. I recommend adding introductory text before presenting the figure to enhance readability and coherence.

Thank you for your valuable suggestion. As recommended, we have added introductory text before Figure 1 in Section 2.2 to improve the flow and readability of the manuscript.

Comment 3: I suggest the authors make a final revision (regarding acronyms) and, if possible, polish the English.

We sincerely appreciate the reviewer's valuable suggestions for enhancing our manuscript. In response to these comments, we have conducted a thorough final revision, including comprehensive polishing of the English by a native speaker and meticulous verification of all acronym usage throughout the paper to ensure complete consistency. These improvements have elevated the manuscript's linguistic quality and technical precision. For the reviewer's convenience, all changes have been carefully highlighted in the revised version. We are grateful for these insightful recommendations, which have significantly strengthened our paper, and we remain open to any additional suggestions the reviewer may have.

Reviewer 3 Report

Comments and Suggestions for Authors

The authors have adequately addressed all comments. The clarifications and additions improve the quality of the manuscript.

Author Response

The authors have adequately addressed all comments. The clarifications and additions improve the quality of the manuscript.

We are deeply grateful for the reviewer's recognition of our revision efforts. Their insightful comments and constructive suggestions have been invaluable in helping us significantly improve the quality of this manuscript. We sincerely appreciate the time and expertise they have dedicated to reviewing our work, which has undoubtedly strengthened both the technical rigor and clarity of presentation in this paper.