Retrieving Inland Water Quality Parameters via Satellite Remote Sensing: Sensor Evaluation, Atmospheric Correction, and Machine Learning Approaches

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The authors conducted a systematic review of satellite remote sensing methods for monitoring inland water quality parameters (WQPs), critically evaluating three pivotal components: sensor capabilities, atmospheric correction (AC) algorithms (including adjacency effect mitigation), and bio-optical modeling techniques encompassing both physics-based and machine learning (ML) approaches. The study systematically reviews advances and challenges in retrieving both optically active (e.g., Chl-a, CDOM, NAP) and non-optically active water quality parameters (WQPs), addressing critical technical gaps such as sensor limitations, atmospheric correction uncertainties, and machine learning (ML) generalizability. The work aligns well with the scope of Remote Sensing journal. The essay is well structured, logical, with clear diagrams and language. The number of references is sufficient and of high quality, and most of the years are newer, and the study will be very helpful to future scholars in related fields in selecting appropriate remote sensing sensors, data, and methods. However, there are some issues that need to be addressed before the paper can be accepted for publication.

Some specific comments with respect to various section:

1. This study has a high number of abbreviations for terms, and it is recommended that the full name be used on the first occurrence (e.g., "SBG").

2. lines 138~139: “only those deemed suitable (N =164) were included in the review”.Is the "N" a filter or the number of articles? We generally think of this as the number of articles, but it's easy to misunderstand here.

3. "However, due to operational limitations, no single sensor currently meets all these 82 requirements.". That being the case, can different sensor data be combined to complement each other to meet a wider range of research needs? More discussion could be added in this regard.

4. The Data Record Period of SBG in Table 2 is 2028, which is highly recommended in the conclusion, but whether its lack of research on the validation and analysis of its practical application scenarios is misleading.

5. Should spatial resolution be sacrificed for more reliable flare correction (e.g. spatial aggregation to 30m) for small water bodies?

6. Which is more robust, SWIR or NIR methods, in regions with complex aerosol types (e.g., industrially polluted areas)?

Author Response

Thank you for your comments. The responses to your comments are provided below, and the corresponding revisions have been highlighted in yellow in the manuscript.

 

Comment 1: This study has a high number of abbreviations for terms, and it is recommended that the full name be used on the first occurrence (e.g., "SBG").

Response 1: Thank you for your comment. The full names of the following terms have been added to the paper:

• Sea-viewing Wide Field-of-view Sensor (SeaWiFS)
• Medium Resolution Imaging Spectrometer (MERIS)
• Geostationary Ocean Color Imager (GOCI)
• Surface Biology and Geology (SBG)
• Operational Land Imager (OLI)
• Multispectral Imager (MSI)

Comment 2: lines 138~139: “only those deemed suitable (N =164) were included in the review”.Is the "N" a filter or the number of articles? We generally think of this as the number of articles, but it's easy to misunderstand here.

Response 2: Thank you for your comment. The text has been updated as follows:

The chosen papers were then read in detail, and only those deemed suitable (number of articles =164) were included in the review.

Comment 3: "However, due to operational limitations, no single sensor currently meets all these 82 requirements.". That being the case, can different sensor data be combined to complement each other to meet a wider range of research needs? More discussion could be added in this regard.

Response 3: Thank you for your comment. The text has been updated as follows:

However, due to operational limitations, no single sensor currently meets all these requirements. Reviewing past, present, and upcoming sensors will help determine which sensors meet these requirements and to what extent. Such an evaluation also highlights the potential for combining complementary sensors, enabling the integration of multi-sensor data. This multi-sensor integration approach can enhance inland water monitoring by providing higher temporal [15] or spatial resolution [16] observations compared to those obtained from a single sensor.

Comment 4: The Data Record Period of SBG in Table 2 is 2028, which is highly recommended in the conclusion, but whether its lack of research on the validation and analysis of its practical application scenarios is misleading.

Response 4: Thank you for your comment. The text has been updated as follows:

The Surface Biology and Geology (SBG) sensor, set to launch in 2028, is highly suitable for modeling Chl-a, NAP, and CDOM (Table 2). It has spatial and temporal resolutions similar to Landsat OLI but offers higher spectral resolution, improved SNR, and a tilting mechanism to minimize sun glint, making it highly suitable for inland WQP monitoring. However, its performance remains to be validated once operational data become available.

Comment 5: Should spatial resolution be sacrificed for more reliable flare correction (e.g. spatial aggregation to 30m) for small water bodies?

Response 5: We assume that by “flare”, the reviewer refers to sunglint and skyglint –reflection of sunlight off the water surface. We have added the following to page 11: “Alternatively, in situations where high spatial resolution is not required, spatial aggregation could be used to facilitate statistical glint correction models to be applied to the data.”

Comment 6: Which is more robust, SWIR or NIR methods, in regions with complex aerosol types (e.g., industrially polluted areas)?

Response 6: Thank you for your comment. As discussed in section 4.4 of the paper, the first option is to use NIR, as SWIR bands have a relatively lower signal-to-noise ratio. However, in cases where water-leaving radiance in the NIR region is non-negligible (as noted in [87]), it can lead to potential overestimation of aerosol effects. In such cases, we are therefore compelled to use alternative options, such as SWIR.

Reviewer 2 Report

Comments and Suggestions for Authors

This paper reviews three key components for effective water quality parameter inversion using optical satellite remote sensing: (1) sensors that are sensitive to changes in water quality, (2) precise atmospheric correction to eliminate the effects of absorption and scattering in the atmosphere and recover the emissivity/reflectance of leaving water, and (3) a bio optical model for estimating water quality from optical signals.
Suggestions are as follows:

Line 25: The result description should use the past tense
The results show that no atmospheric correction algorithm performs consistently across all conditions.show Should be changed to 'shown'

Line 154-157 “If all the pixels in the 3×3 or4×4  grid appear to be within the water body, then at least the center pixel(s) can be con-sidered to represent pure water, minimizing the risk of mixed signals from surrounding land or nearby features” This viewpoint lacks relevant theoretical basis or references to other studies to support it, and lacks authority and credibility.

One of the viewpoints on Line 275-277 regarding near-infrared interference "potentially leading to excessive correction of Rrs in visible wavelengths" requires research results to support

Line 295-297 “Because image - based methods offer partial correction, the original TOA radiance captured in the images is partially changed, preserving its initial optical properties.” The meaning of "preserving its initial optical properties" in the text is ambiguous, and it is not clear what the specific "initial optical properties" to be retained refer to, as well as the relationship between it and partial correction

Line 310 "Sentinel 2" should be changed to "Sentinel-2"

Line 446-449 should introduce that the OLI sensor is an important sensor mounted on the Landsat 8 satellite, and the MSI sensor is an important sensor mounted on the Sentinel-2 satellite, before establishing a connection with Figure 1

512 lines "[36]" repeated citation of references
Numerous studies have compared ML model performance for retrieving inland 512 WQPs [30,36,36,36,36,119,121–138].

Author Response

Thank you for your comments. The responses to your comments are provided below, and the corresponding revisions have been highlighted in green in the manuscript.

Comment 1: Line 25: The result description should use the past tense
The results show that no atmospheric correction algorithm performs consistently across all conditions. show Should be changed to 'shown'

Response 2: Thank you for your comment. The text has been updated as follows:

The results showed that no atmospheric correction algorithm performed consistently across all conditions.

Comment 2: Line 154-157 “If all the pixels in the 3×3 or4×4  grid appear to be within the water body, then at least the center pixel(s) can be con-sidered to represent pure water, minimizing the risk of mixed signals from surrounding land or nearby features” This viewpoint lacks relevant theoretical basis or references to other studies to support it, and lacks authority and credibility.

Response 2: Thank you for your comment. The text has been updated as follows:

First, to obtain a valid pixel over a water body, it is generally recommended to use a square configuration of approximately 3×3 or 4×4 pixels [18]. If all the pixels in the 3×3 or 4×4 grid appear to be within the water body, then at least the center pixel(s) can be considered to represent pure water, minimizing the risk of mixed signals from surrounding land or nearby features. This criterion makes sensor selection dependent on the size of the study area.

Comment 3: One of the viewpoints on Line 275-277 regarding near-infrared interference "potentially leading to excessive correction of Rrs in visible wavelengths" requires research results to support.

Response 3: Thank you for your comment. The following references has been added:

Mishra, D.R.; Ogashawara, I.; Gitelson, A.A. Bio-Optical Modeling and Remote Sensing of Inland Waters; Elsevier, 2017; ISBN 978-0-12-804654-8.

Hu, C.; Carder, K.L.; Muller-Karger, F.E. Atmospheric Correction of SeaWiFS Imagery over Turbid Coastal Waters: A Practical Method. Remote Sens. Environ. 2000, 74, 195–206, doi:10.1016/S0034-4257(00)00080-8.

Ruddick, K.G.; Ovidio, F.; Rijkeboer, M. Atmospheric Correction of SeaWiFS Imagery for Turbid Coastal and Inland Waters. Appl. Opt. 2000, 39, 897–912, doi:10.1364/AO.39.000897.

Comment 4: Line 295-297 “Because image - based methods offer partial correction, the original TOA radiance captured in the images is partially changed, preserving its initial optical properties.” The meaning of "preserving its initial optical properties" in the text is ambiguous, and it is not clear what the specific "initial optical properties" to be retained refer to, as well as the relationship between it and partial correction

Response 4: Thank you for your comment. The text has been simplified and updated as follows (along with a minor update to Table 3):

Image-based methods are thus simple to implement, but their performance is often limited in inland waters [64] potentially due to atmospheric heterogeneity.

Comment 5: Line 310 "Sentinel 2" should be changed to "Sentinel-2"

Response 5: Thank you for your comment. The naming of 'Sentinel 2' has been corrected to 'Sentinel-2' throughout the manuscript, including Figure 3. Similarly, 'Sentinel 3' has been revised to 'Sentinel-3'.

Comment 6: Line 446-449 should introduce that the OLI sensor is an important sensor mounted on the Landsat 8 satellite, and the MSI sensor is an important sensor mounted on the Sentinel-2 satellite, before establishing a connection with Figure 1

Response 6: Thank you for your comment. The text has been updated as follows:

Recent studies have compared AC algorithms across different sensors to evaluate their effectiveness in retrieving R_rs and WQPs. The OLI onboard Landsat 8 and 9 and the MSI onboard Sentinel-2 have been frequently used in inland water studies. As shown in Figure 1, the number of publications related to these sensors has notably increased since their launch.

Comment 7: 512 lines "[36]" repeated citation of references
Numerous studies have compared ML model performance for retrieving inland 512 WQPs [30,36,36,36,36,119,121–138].

Response 7: Thank you for your comment. The text has been updated as follows:

Numerous studies have compared ML model performance for retrieving inland WQPs [32,37,120,122–139].

Reviewer 3 Report

Comments and Suggestions for Authors

In general, it's a well-structured and useful study. The research seems solid and highly relevant, especially given the growing interest in efficient, low-cost environmental monitoring. It addresses a complex issue —assessing water quality in inland water bodies— from a comprehensive perspective, analyzing satellite sensors, atmospheric correction algorithms, and bio-optical models, including machine learning approaches.

What stands out most is that it not only reviews the current state of the field but also provides a framework for evaluating technologies and offers practical recommendations. Moreover, it openly acknowledges limitations, particularly in machine learning models, which adds to its credibility.

The figures are clear and effectively aid in the understanding of what is included in the text that refers to them.

In the conclusion, key findings are numbered and thematically structured, which makes them easy to read and understand. They address sensors, atmospheric correction (AC) algorithms, and machine learning (ML) approaches, reflecting a comprehensive understanding of the problem. Some points offer practical advice (e.g., assessing sensor suitability, avoiding information leakage in ML), making the conclusions actionable.

While the conclusions list findings and challenges, they tend to be descriptive. They would benefit from more critical insights. The conclusions appear as isolated observations. Including a brief synthesis at the end or transitions between findings could help present a more integrated view of the future of WQP monitoring. All conclusions are presented at the same level. It would be helpful to identify which findings are most critical to advancing the field and why. It detects overgeneralization in point 9. It is suggested that increasing data dimensionality improves model performance, without acknowledging the risk of overfitting or the need for proper regularization.

Comments on the Quality of English Language

It is technically sound and well-written but could benefit from slight revisions for clarity, sentence flow, and readability.

Author Response

Thank you for your comments. The responses to your comments are provided below, and the corresponding revisions have been highlighted in blue in the manuscript.

Comment 1: In general, it's a well-structured and useful study. The research seems solid and highly relevant, especially given the growing interest in efficient, low-cost environmental monitoring. It addresses a complex issue —assessing water quality in inland water bodies— from a comprehensive perspective, analyzing satellite sensors, atmospheric correction algorithms, and bio-optical models, including machine learning approaches.

What stands out most is that it not only reviews the current state of the field but also provides a framework for evaluating technologies and offers practical recommendations. Moreover, it openly acknowledges limitations, particularly in machine learning models, which adds to its credibility.

The figures are clear and effectively aid in the understanding of what is included in the text that refers to them.

In the conclusion, key findings are numbered and thematically structured, which makes them easy to read and understand. They address sensors, atmospheric correction (AC) algorithms, and machine learning (ML) approaches, reflecting a comprehensive understanding of the problem. Some points offer practical advice (e.g., assessing sensor suitability, avoiding information leakage in ML), making the conclusions actionable.

While the conclusions list findings and challenges, they tend to be descriptive. They would benefit from more critical insights. The conclusions appear as isolated observations. Including a brief synthesis at the end or transitions between findings could help present a more integrated view of the future of WQP monitoring. All conclusions are presented at the same level. It would be helpful to identify which findings are most critical to advancing the field and why. It detects overgeneralization in point 9. It is suggested that increasing data dimensionality improves model performance, without acknowledging the risk of overfitting or the need for proper regularization.

Response 1: Thank you for the review and comments.

First, we have added what we hope to be additional critical insights to the numbered list of conclusions – please see all the blue text in the numbered list of conclusions.  We have retained the list of conclusions as a numbered list, because to some extent we believe they are in fact isolated observations.

Secondly, we have also added a synthesis at the end of the conclusion. It is our view that improvements in WQP monitoring are likely to result from gradual improvements in sensor characteristics, atmospheric correction and other pre-processing, as well as the improvements in machine learning models and an understanding of how to apply them effectively to the available data (as listed in the numbered points). But improvements can come from these components in isolation – for example, improved ML models, e.g. from deep learning, may on their own produce better WQP monitoring, independently of other factors. Which factors, or combinations of factors, will turn out to be the most important? We will only know the answer to that once data from new sensors are processed and tested, thoroughly, in a wide range of scenarios. In that spirit, we have added the following text at the end of the conclusion section: “Improvements in WQP monitoring over the coming years are likely to result from gradual improvements in sensor characteristics and data quality, AC and other pre-processing, as well as improvements in ML models and an understanding of how to apply such models effectively to the available data. Improvements may come from these components in isolation, for example improved ML models, e.g. from deep learning, may on their own produce better WQP monitoring, independently of other factors. Or improvement may come from combinations of factors, such as ML models increasingly able to extract information from the additional dimensionality of hyperspectral data or the varied information provided through multi-sensor integration. In this study we have identified some of the most promising avenues for future improvement in WQP monitoring using remote sensing technology, real-world testing will demonstrate which of these will provide the greatest value over the coming years.”