Does Hyperspectral Imagery Improve Satellite-Derived Bathymetry? A Case Study from a Posidonia oceanica-Dominated Mediterranean Region

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The article “Does Hyperspectral Imagery Improve Satellite-Derived Bathymetry?” addresses the scientific question of whether SDB estimation can be improved by incorporating additional hyperspectral bands compared with conventional multispectral satellite imagery. The study focuses particularly on shallow coastal areas covered with seagrass, as seagrass exhibits a distinct reflectance signature. To estimate bathymetry from PRISMA and LANDSAT data, the authors use a Random Forest machine-learning model, applying the same training dataset for both image sources. Overall, the article is well structured and accessible, allowing the reader to follow the methodology and findings with ease.

Points for consideration:

• The title is too broad; it should emphasize the study region (Mediterranean) and the specific environment investigated (shallow-water seagrass meadows).
• Keywords should not repeat or mirror the title.
• In the abstract (lines 10–12), the statement regarding SDB performance is inaccurate, as several limiting factors influence SDB prediction.
• In the introduction (lines 48–50), the limiting factors for SDB estimation should be addressed more clearly.
• The introduction highlights the focus on Posidonia-rich areas in the Mediterranean; this should also be reflected in the title.
• For example, line 81 refers to “clear waters” — is there a quantitative metric to support this claim?
• Compared with Sentinel-2, PRISMA has lower spatial resolution; this should be acknowledged, even though LANDSAT imagery was used.
• Lines 126–134: It is unclear which data points were used for model training and which for validation. Distinct datasets are needed to avoid overfitting and ensure evaluation.
• Lines 295–298: This conclusion should be more explicitly linked to the study’s results; the current wording appears poorly aligned with the findings.
• The in situ bathymetric data originate from 1972 and 1973. The authors should provide justification for assuming that seafloor topography has remained unchanged.
• The conclusion should be more concise and direct.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript provides a systematic evaluation of the potential benefits of PRISMA hyperspectral data compared with Landsat-8 multispectral imagery for Satellite-Derived Bathymetry. The topic is scientifically meaningful and relevant to the community, and the study is generally well structured with a clear workflow. The results offer useful insights into how different spectral sources contribute to shallow-water depth retrieval. At the same time, several aspects of the manuscript would benefit from further clarification, strengthening, or additional analysis. I recommend that the authors address the following issues before the manuscript can be considered for publication.

1. The manuscript uses SHOM single-beam bathymetry collected in 1972–1973 as reference data, while the satellite imagery is from 2021. This ~50-year temporal gap raises concerns because seabed morphology, sedimentation, and Posidonia distribution may have changed substantially during this period. The authors should provide a stronger justification for why this dataset remains suitable as ground truth, and discuss potential biases introduced by such temporal mismatches.

2. In the data preprocessing section, several PRISMA bands are excluded due to being “longer wavelengths” or being “too noisy,” yet no quantitative metric (e.g., SNR, ENL, reflectance distribution) is provided to support this decision. The authors should clarify how these bands were identified and the threshold or criterion used to determine that they were unsuitable.

3. The study relies exclusively on Random Forest (RF) regression. Given that hyperspectral data are high-dimensional and RF is prone to overfitting noisy features, the manuscript should explain why RF was chosen over other models. Alternatively, providing comparisons with at least one additional method (e.g., linear regression, SVM, or a simple physics-based/semianalytical model) would greatly strengthen the methodological justification.

4. The iterative band-addition strategy is performed using a single 80/20 train-test split and MAE as the only evaluation metric. This may lead to band-ranking instability due to data partition randomness. The authors are encouraged to include k-fold cross-validation or repeated random subsampling to assess whether the “optimal” band combination remains consistent across different splits.

5. The manuscript does not provide exact numbers of training/testing points, nor the distribution of samples among shallow/deep water, seagrass, and non-seagrass categories. Without this information, it is difficult to assess whether the hyperspectral model suffers from data sparsity, class imbalance, or sample representativeness issues. A summary table of sample statistics would greatly improve transparency.

6. Figures 3 and 8 rely on visual inspection to describe spectral differences between classes, but no quantitative measures (e.g., Jeffries–Matusita distance, Bhattacharyya distance, or statistical significance tests) are used. To support claims about certain wavelengths being discriminative, quantitative separability metrics should be incorporated.

7. The study focuses on depths from 0–30 m, but PRISMA shows pronounced improvement specifically in deeper waters (>25 m). A more detailed depth-stratified analysis (e.g., 0–5 m, 5–15 m, 15–25 m, 25–30 m) would help clarify where hyperspectral data provide the most benefit, and whether this advantage extends toward the optically deep boundary (>30 m).

8. Even though PRISMA performs slightly worse than Landsat-8 overall, it shows clear potential in deeper regions. The manuscript would be strengthened by explicitly discussing real-world scenarios in which hyperspectral data may still be preferable (e.g., areas lacking multispectral coverage, deeper coastal zones, high water clarity, or research focused on substrate discrimination).

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

All raised issues have been resolved.

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript has been revised as required and meets publication standards. Thank you.