Risk Assessment of Potential Black and Odorous Water Body Based on Satellite and UAV Multispectral Remote Sensing

Response to Reviewer 1 Comments

1. Summary

Thank you very much for taking the time to review this manuscript. In response to your valuable comments, we have made substantial revisions to the manuscript. These include: (1) discussing the optical physical mechanisms of water quality parameters, (2) reorganizing the article structure, (3) enhancing the quality of figures and tables, and (4) re-optimizing the models. Please find the detailed responses below and the corresponding revisions/corrections highlighted/in track changes in the re-submitted file.

2. Questions for General Evaluation

Reviewer’s Evaluation

Response and Revisions

Does the introduction provide sufficient background and include all relevant references?

Yes/Can be improved/Must be improved/Not applicable

[Please give your response if necessary. Or you can also give your corresponding response in the point-by-point response letter. The same as below]

Are all the cited references relevant to the research?

Yes

Is the research design appropriate?

Yes

Are the methods adequately described?

Yes

Are the results clearly presented?

Yes

Are the conclusions supported by the results?

Yes

3. Point-by-point response to Comments and Suggestions for Authors

Comments 1: First, the manuscript focuses on the retrieval of four non-optically active parameters. These parameters typically lack direct spectral response characteristics. The authors merely selected band combinations through correlation analysis without discussing the underlying mechanisms of the retrieval process. It is generally accepted that these parameters are retrieved by establishing indirect relationships with optically active substances (such as Chlorophyll-a, suspended particulate matter, or CDOM). The lack of discussion on this mechanism results in weak physical interpretability of the model. It is recommended to supplement the analysis with the relationships between these parameters and optically active components. Regarding the retrieval mechanisms for non-optically active parameters, you may refer to the following papers: Li, L., Gu, M., Gong, C., et al. An advanced remote sensing retrieval method for urban non-optically active water quality parameters: An example from Shanghai[J]. Science of the Total Environment, 2023, 880: 163389. Zhao, Y., Yu, T., Hu, B., et al. Retrieval of Water Quality Parameters Based on Near-Surface Remote Sensing and Machine Learning Algorithm[J]. Remote Sensing, 2022, 14(21): 5305. Chen, J., Zhu, W., Tian, Y. Q., et al. Estimation of Colored Dissolved Organic Matter from Landsat-8 Imagery for Complex Inland Water: Case Study of Lake Huron[J]. IEEE Transactions on Geoscience and Remote Sensing, 2017, 55(4): 2201-2212.

Response 1: In the introduction section, a discussion on the inversion mechanism of non-optical active parameters has been added. The correlation between such parameters and optical active components has been further analyzed, and relevant literature has been cited. (Page 2, lines 52-59)

Comments 2: Secondly, and more importantly, there is significant temporal inconsistency in the data used for the satellite model. The authors point out that the 40 samples used to develop the satellite remote sensing model (MBOWI threshold) were collected in July 2023, yet this model was applied to identify potential black and odorous water bodies using satellite imagery from August 2024. The spectral characteristics of water bodies and atmospheric conditions can differ significantly between different years. Applying thresholds defined in 2023 directly to 2024 data without calibration or validation casts serious doubt on the accuracy of the identification results. This is a serious issue that affects the reliability of the "Satellite-Air-Ground" framework proposed in this paper.

Response 2: In this study, the satellite remote sensing model was established in July 2023. The threshold for the MBOWI model was determined through field water sampling and satellite spectral analysis conducted at that time. In August 2024, we employed the 2023-established satellite model to screen for high-risk water bodies, while simultaneously conducting UAV-based water quality inversion. Due to the limited number of sampling points collected in 2024, the thresholds derived in 2023 were not recalibrated. We acknowledge that this may affect the accuracy of the satellite remote sensing model to some extent. However, the primary focus of this paper is to demonstrate a workflow that utilizes satellite remote sensing for the preliminary and rapid identification of suspected black-odor risk water bodies, followed by UAV remote sensing for detailed water quality inversion. This approach aims to verify the feasibility of using UAV inversion results for the quantitative assessment of black-odor risks. Therefore, the results from the satellite model serve primarily as a reference. We will address this limitation in future studies by collecting sufficient data to recalibrate and re-validate the model thresholds.

Comments 3: Abstract Page 1 Line 24: It is recommended to rewrite the abstract to highlight the physical mechanism or the novelty of the method, rather than just listing values. The current abstract lacks a description of the importance of combining satellite and UAV (compared to using them individually).

Response 3: Thank you for pointing this out. We agree with this comment. Therefore, we have substantially revised the abstract in the modified manuscript. The revision emphasizes the significance of the integrated method—combining satellite remote sensing, UAV remote sensing, and ground monitoring—proposed in this study for the quantitative risk assessment of malodorous water bodies. This approach effectively addresses the limitation of traditional satellite remote sensing in quantitatively evaluating such water bodies. Furthermore, the abstract briefly presents the model's inversion results and accuracy, highlighting the innovation of our research in terms of physical mechanisms and methodology. [Lines 12–35 of the Abstract]

Comments 4: Introduction Page 2 Lines 42-46: The explanation regarding the formation mechanism of black and odorous water bodies is too basic. It is recommended to cite recent literature on the optical characteristics of black and odorous water bodies to support the spectral basis of the research.

Response 4: Agree. The darkening of water is caused by pollutants altering the optical absorption characteristics of the water body, which constitutes the theoretical basis for satellite remote sensing monitoring of black-odor water. The manuscript cites recent relevant literature to elaborate on this mechanism, thereby providing strong support for the spectral analysis in this study. [Page 2, Lines 49–56]

Comments 5: Thirdly, regarding the machine learning models used for UAV retrieval, the sample size (64 samples) is relatively small for training neural network models, which can easily lead to overfitting. More importantly, the Random Forest (RF) model performed extremely poorly, with the for even being negative (-0.250). RF is typically considered an algorithm that is robust to small samples and high-dimensional data. The failure of RF suggests there may be unresolved issues with parameter tuning or data distribution. The authors merely concluded that RF performed poorly without analyzing the reasons. I strongly recommend that the authors check the model configuration and consider comparing more algorithms or using ensemble learning strategies to improve robustness.

Response 5: In the revised manuscript, we have retrained the models. To enhance model robustness under small-sample conditions, we implemented 5-fold cross-validation, which has effectively mitigated the issue of negative   values observed in the Random Forest model. Following your valuable suggestion, we introduced the SA-SVR (Simulated Annealing-Support Vector Regression) model, which is known for its superior performance in nonlinear fitting with limited data. We retrained the models using this approach, and the results demonstrate that the SA-SVR model outperforms the original Neural Network model in inverting water quality parameters. Consequently, all inversion results have been recalculated based on the optimized SA-SVR model.

Comments 6: Page 2 Lines 77-80: The authors mentioned the limitations of satellite remote sensing. However, in the conclusion section (Page 15 Line 492), the authors point out that "high-resolution satellites... lack atmospheric correction models for urban water bodies." If this is the case, how do the authors ensure the accuracy of the MBOWI calculated from JL-1 satellite data in this study? Please provide a more detailed explanation and modify this part with appropriate references.

Response 6: Agree. In this study, we constructed the Malodorous Black Water Index (MBOWI) model using Jilin‑1 satellite data to realize the rapid monitoring of potential black and odorous water bodies. For Level 1 radiometric calibration products, we adopted the FLAASH atmospheric correction model for preprocessing, converting radiance to surface reflectance, thereby improving the inversion accuracy of the MBOWI model (Original manuscript, Page 4, Line 150). In addition, we have revised the inaccurate expressions in the conclusion section of the original manuscript (Page 15, Line 492): This study only conducted atmospheric correction on L1 products, and has not yet established a dedicated water-leaving radiance correction model for Jilin‑1 satellites. Therefore, it failed to effectively eliminate the signals from water surface specular reflection and bottom reflection, which affected the final accuracy of the model to a certain extent. (Page 18, Lines 516–519)

Comments 7: Study Area and Materials Page 4 Figure 1: The map of the study area is not aesthetically pleasing and does not reflect the characteristics of "crisscrossing canals and typical plain river networks."

Response 7: Agree.The study area map has been revised for improved aesthetics and clarity. We have integrated a river network layer to highlight the characteristics of the plain river network topography. Additionally, as suggested by Reviewer 2, a new inset map showing the distribution of sampling points has been included.

Comments 8: Methodology Page 8 Section 3.2: The reference cited for the "Multispectral Black-Odorous Water Index (MBOWI)" is [39] (Wang Z, 2024). Is this index proposed by the authors in a previous paper, or is it citing another scholar's index? If it is an existing index, the contribution of this paper in terms of "constructing" the model needs to be clarified.

Response 8: The MBOWI index was initially proposed by Fu Li (2024), who developed it using GF-2 satellite data for the identification of black-odor water bodies in regions such as Jilin, Yunnan, and Guangxi.In this study, experiments on the equivalent reflectance of black-odor water revealed that the reflectance characteristics in the North Zhejiang Plain follow a similar trend. However, considering that the spatial resolution of the GF-2 satellite is insufficient for monitoring the fine river networks in this area, we adapted the model framework established by Fu Li (2024) to the higher-resolution Jilin-1 (JL-1) satellite data. Furthermore, the model's threshold was recalibrated based on the field sampling results from the North Zhejiang region.The relevant discussion has been added to the revised manuscript. [Page 8, Lines 285–290]

Comments 9: Page 10 Lines 301-304: 1446 band combinations were generated from 18 bands. With only 64 samples, screening from such a large number of features poses a huge risk of overfitting. Besides simple Pearson correlation, did the authors use any feature selection methods? It is recommended to discuss this limitation or apply a stricter feature selection process.

Response 9: In this study, we employed a band combination correlation analysis method. By exhaustively enumerating linear combinations of two to three bands (generating 1446 combinations in total), we analyzed the correlation between these combinations and the water quality parameters to select the optimal set. This method is suitable for small-sample datasets, featuring simple and efficient computation with strong physical interpretability. However, this approach has certain limitations: it only considers linear relationships, exhibiting poor performance in capturing the non-linear response of water quality; it fails to eliminate multicollinearity among bands, resulting in average generalization performance and a tendency to overfit. In our future work, we will adopt methods such as analysis of variance (ANOVA) or LASSO regression to optimize the feature selection process. A discussion regarding these limitations has been added to the revised manuscript. [Page 18, Lines 531–541]

Comments 10: Page 10 Line 340: Please provide the specific hyperparameters of the machine learning models; reproducibility is crucial for academic papers.

Response 10：All hyperparameters used in the machine learning models have been detailed in Table 4 (page 11) of the revised manuscript.

Comments 11: Discussion Page 12 Section 4.2: The discussion regarding the spatial distribution of water quality parameters (Figure 9) is purely descriptive. The authors attribute spatial variation to "poor water mobility". Are there visible point-source pollutions (such as outfalls) in the UAV images that correlate with high values of or depletion of DO? Combining visual interpretation of UAV RGB images with retrieval results would make the discussion more insightful.

Response 11：The discussion regarding the inversion results of water quality parameters has been revised in the manuscript. By integrating the inversion values with UAV imagery, we conducted a visual interpretation of four high-risk water bodies (inferior to Class V) and two low-risk water bodies to analyze the potential causes of pollution. The analysis revealed that the low DO or high NH₃-N zones in JS-1, XZ-2, and WX-2 correspond to areas with visible sewage discharge. Additionally, HN-1 and WX-2 exhibit blocked river channels and poor water flow, leading to reduced dissolved oxygen levels.By validating the inversion results against high-resolution UAV imagery, the persuasiveness of our discussion has been strengthened. [Page 16, Lines 471–493]

Comments 12: Page 13: Risk assessment is conducted based on retrieved values. Since there are errors in the retrieval (e.g., TP =0.58), how is this uncertainty propagated into the risk classification? The manuscript lacks a sensitivity analysis or confusion matrix based on ground-truth data (Risk vs. No Risk).

Response 12：In the revised methodology section (Page 133, Line 399), we have added the confusion matrix for the water quality classification results, along with the precision and recall for each category. These metrics serve to validate the applicability of the risk assessment method based on water quality parameters. In terms of classification accuracy, the Overall Accuracy (OA) reached approximately 0.70.

Comments 13: Conclusion Page 15 Line 496: The authors admit that "this study did not involve an analysis of the spectral response characteristics of non-optically active water quality parameters".

Response 13：This study emphasizes the methodology for black-odor water risk assessment. Therefore, the spectral analysis was limited to distinguishing between polluted and clean water based on equivalent reflectance derived from satellite imagery. For the UAV-based inversion model, we identified optimal band combinations via exploratory data analysis and built a machine learning model. Nevertheless, spectral response analysis for non-optically active parameters (e.g., DO, COD) was not performed. Future research will employ ASD spectrometers and hyperspectral imagers to investigate the spectral features of these parameters, with the goal of improving inversion accuracy.

Comments 14: I suggest adding a section in the discussion to theoretically speculate why the selected bands are effective. For instance, do b10 (625nm) or b14 (703nm) relate to the specific absorption characteristics of water constituents?

Response 14：In the revised manuscript, Section 4.21 was added to the Discussion. This section analyzes the potential indirect relationships between non-optically sensitive parameters such as DO and COD and band combinations, strengthening the justification for the optimal band combination in the methodology.

4. Response to Comments on the Quality of English Language

Point 1:

Thank you for pointing out the language issues. We have taken this comment very seriously and conducted a comprehensive line-by-line revision of the entire manuscript. We focused on correcting grammatical errors, improving sentence structure, and enhancing the overall flow and clarity of the text. Additionally, the revised manuscript was reviewed by a colleague who is a native English speaker [or: has extensive experience in academic writing] to ensure the language is natural and precise. We hope the current version meets the required standards.

5. Additional clarifications

No.

Comments 1: Figure 1: Another data frame can be added to show the sample sites zoomed, also the color can be changed to make it more prominent.

Response 1: Thank you for pointing this out. I have revised Figure 1 by adding a data frame to show the sample sites zoomed, optimizing the color scheme for improved aesthetics, and incorporating a water network layer.

Comments 2: Line 204: Adjusted R² can be used to check the robustness of the model. Also can add slope and bias.

Response 2: Agree. In the revised manuscript, I have updated the accuracy assessment of the water quality inversion model to the Adjusted R², which better reflects the model’s fitting performance and explanatory capability.

Comments 3: Figure 6: the dimensions of the figure can be changed for better visualization.

Response 3: Agree. Therefore, I have changed the dimensions of Figure 6 for better visualization.

Comments 4: Figure 9: The size can be increased for better readability. The legend and numbers are not readable in current form.

Response 4: Thank you for pointing this out. I have enlarged icons and legend in Figure 9 for better readability.

Response to Reviewer 3 Comments

1. Summary

Thank you very much for taking the time to review this manuscript. In response to your valuable comments, we have made substantial revisions to the manuscript.Please find the detailed responses below and the corresponding revisions/corrections highlighted/in track changes in the re-submitted file.

2. Questions for General Evaluation

Reviewer’s Evaluation

Response and Revisions

Does the introduction provide sufficient background and include all relevant references?

Yes/Can be improved/Must be improved/Not applicable

[Please give your response if necessary. Or you can also give your corresponding response in the point-by-point response letter. The same as below]

Are all the cited references relevant to the research?

Yes

Is the research design appropriate?

Yes

Are the methods adequately described?

Yes

Are the results clearly presented?

Yes

Are the conclusions supported by the results?

Yes

3. Point-by-point response to Comments and Suggestions for Authors

Comments 1: Lines 46-47: Abbreviations such as "(FeS)", "(MnS)", and "(H₂S)" should only be marked when they first appear and will be reused later. The same applies to subsequent instances.

Response 1: Thank you for pointing this out. In the revised manuscript, abbreviations such as "(FeS)", "(MnS)", and "(H₂S)" are only marked when they first appear and have be reused later.

Comments 2: Line 60: Is there an extra "." before or after the citation? The same issue exists in Line 273.

Response 2: Thank you for pointing this out. I have removed the extra "." in Line 60 and Line 273.

Comments 3: Line 92: For the abbreviations "COD/COD_Mn", should consistency be maintained?

Response 3: Agree. Therefore, in the revised manuscript, I have standardized the abbreviation for chemical oxygen demand to COD.

Comments 4: Line 125: Since "NH₃-N" first appears in the text and will be reused later, its full name should be supplemented.

Response 4: Agree. Therefore, I have provided the full name of "NH₃-N" in the abstract and used the abbreviation in the rest of the manuscript.

Comments 5: Line 137: Should the "JL-01KF01C" satellite data used be clearly specified?

Response 5: Agree. As stated in Line 145, Page 4 of the original manuscript, JL‑01KF is introduced. JL‑01KF01C belongs to the same constellation satellite series and shares consistent main parameters with it. To avoid ambiguity, the notation JL‑1KF has been used consistently throughout the revised manuscript.

Comments 6: Line 148: "JL-1KF" should be consistent with the terminology mentioned earlier. The same issue exists in Lines 234 and 367.

Response 6: Thank you for pointing this out. JL‑01KF01C belongs to the same constellation satellite series and shares consistent main parameters with JL‑01KF. To avoid ambiguity, the notation JL‑1KF has been used consistently throughout the revised manuscript.

Comments 7: Lines 179-181: Since "DO" and "NH₃-N" have been marked as abbreviations in the previous text, should the abbreviations be used preferentially? The same issue exists in Lines 188-191.

Response 7: Agree. I have, accordingly, used the abbreviations "DO" and "NH₃-N" in Lines 179-181 and Lines 188-191.

Comments 8: Lines 197-198: Is it the same document as "Environmental quality standards for surface water (GB 3838-2002)" in Lines 187-188? The name should be consistent.

Response 8: Thank you for pointing this out. The Environmental Quality Standards for Surface Water (GB 3838-2002) cited in Lines 197–198 and Lines 187–188 refers to the same document. The name has been revised to be consistent throughout the text.