HY1C/D-CZI Noctiluca scintillans Bloom Recognition Network Based on Hybrid Convolution and Self-Attention

Round 1

Reviewer 1 Report

An interesting work I only advise the authors to review their references as some are referred on the text by name and not by number (example line 84).  

 

Author Response

Comments and Suggestions for Authors.

An interesting work I only advise the authors to review their references as some are referred on the text by name and not by number (example line 84).  

Response:

Thank you. We carefully checked the references and citations in the manuscript and the incorrect citations were corrected. We also added the missing DOIs, abbreviate names of journals and dates and places of the conferences in the revision.

Reviewer 2 Report

The article describes a neural network model for identifying the presence of Noctiluca scintillans bloom (NSB), a type of bioluminescent algae that can cause harmful algal blooms. The model uses a combination of convolutional neural network (CNN) and residual network (ResNet) architectures to analyze images of HY1C/D-CZI and accurately identify the presence of NSB. Overall, this article presents a valuable contribution to the field of marine biology and environmental monitoring, as accurate and timely identification of harmful algal blooms is crucial for protecting both marine ecosystems and public health. The use of neural network models, such as the NSBRNet presented in this article, has the potential to greatly improve the efficiency and accuracy of NSB detection and monitoring efforts. It is suggested to publish after modification. But I have some comments

 

From the perspective of spectral shape, NSB and the other algal blooms seem to have no other difference. How to avoid the impact of other algae when extract the area of NSB (such as Sargassum, Diatom)?

 

The article does not seem to compare other algae extraction methods, for example, FAI?

Author Response

1. From the perspective of spectral shape, NSB and the other algal blooms seem to have no other difference. How to avoid the impact of other algae when extract the area of NSB (such as Sargassum, Diatom)?

Repones: Thank you for this valuable comment. RNS does have unique spectral features making it possible to be identified from space. Figure 4(a) showed the distinct spectral feature between RNS and macroalgae.  Specifically, RNS has strong absorption in the blue and green bands and strong scattering in other bands, which leads to a sharp enhancement of reflectance from 520 nm to 600 nm and the enhanced reflectance in the near-infrared band. This unique spectral feature was used by Qi et al.(2019,2022) to identify and extract RNS in the East China Sea from MODIS data. The same method proposed by Qi et al was used in this manuscript to avoid the impacts of other macroalgae such as Sargassum.

• Qi, L.; Tsai, S.F.; Chen, Y.; Le, C.; Hu, C. In Search of Red Noctiluca scintillans Blooms in the East China Sea. Res. Lett. 2019, 46, 5997-6004, doi:10.1029/2019gl082667.
• Qi, L.; Hu, C.; Liu, J.; Ma, R.; Zhang, Y.; Zhang, S. Noctiluca blooms in the East China Sea bounded by ocean fronts. Harmful Algae 2022, 112, doi:10.1016/j.hal.2022.102172.

 

2. The article does not seem to compare other algae extraction methods, for example, FAI?

Response: Thank you. Threshold-based index such as FAI or NDVI utilize the spectral feature to identify and extract algae. As the observing conditions change images by images (e.g., occurrence of thin clouds or sun glint), the thresholds need to be manually adjusted by visual examination. In this manuscript, we used the powerful feature extraction and nonlinear approximation capabilities of the deep learning technique to automatically identify and extract RNS from remote sensing data. Actually, the method proposed by Qi et al, a threshold-based index like FAI, was used in this manuscript to extract RNS from satellite data by carefully visualizing each image and to label the dataset as the ‘ground truth’. We compared the RNS identified by NSBRNet and the Qi’s method and they agreed very well with the precision of 92% and recall of 88% (Figure 11, Figure 12 and Table 3).

Reviewer 3 Report

1. What is the main question addressed by the research?

 

The paper is dedicated to automatically recognize Noctiluca scintillans blooms from remote sensing data and improve the recognition accuracy. Authors proposes a Noctiluca Scintillans Bloom Recognition Network incorporating convolution and self-attention to achieve automatic recognition of Noctiluca scintillans blooms using high spatial resolution satellite remote sensing data. Their model was applied to the Coastal Zone Imager data onboard Chinese ocean color satellites. The results show that the Noctiluca Scintillans Bloom Recognition Network can automatically identify Noctiluca scintillans blooms using the Coastal Zone Imager data. Compared with other common semantic segmentation models, the Noctiluca Scintillans Bloom Recognition Network showed its better performance.

 

 

2. Do you consider the topic original or relevant in the field, and if so, why?

 

The topic is original and relevant in the field. The proposed Noctiluca Scintillans Bloom Recognition Network has good compatibility with other remote sensing data and therefore can be extended to other satellite sensors, such as MODIS, VIIRS, or OLCI.

 

 

3. What does it add to the subject area compared with other published material?

 

Noctiluca Scintillans Bloom Recognition Network combines the advantages of convolution and self-attention through the channel partitioning mechanism, which extracts the global spatial information while preserving the channel and local detail information, making Noctiluca Scintillans Bloom Recognition Network flexible in constructing and fitting most of the information compared with other models. The hybrid feature of Noctiluca Scintillans Bloom Recognition Network enables the model to delineate Noctiluca scintillans blooms features at different spatial scales and different background conditions. The Noctiluca Scintillans Bloom Recognition Network can be extended to other satellite sensors, such as MODIS, VIIRS, or OLCI.

 

 

4. What specific improvements could the authors consider regarding the methodology?

 

There is no need to make any improvements or something else. The authors outlined the current achievements in the field in the introduction, and provided detailed description of the materials and method, as well as the obtained results, and conclusions. The results obtained are based on the Coastal Zone Imager data.

 

 

5. Are the conclusions consistent with the evidence and arguments presented and do they address the main question posed?

 

The conclusions are consistent with the evidence and arguments presented in the manuscript and address the main questions of their study. The future research perspective should be outlined in Conclusions.

 

 

6. Are the references appropriate?

 

The references are appropriate. Nevertheless, authors should prepare the References section exactly in accordance with the journal template, abbreviate names of journals, provide missing DOIs, date of access to the internet sources, and dates and places of the conferences and similar events.

 

 

7. Please include any additional comments on the tables and figures.

 

All the tables and figures are appropriate. They show well the research and experiment details and results.

Nevertheless, an empty line should be inserted between the table and following text. Please, increase resolution in Figure 1, 2, 5, 6, 11, 12, 13, and 14.

 

 

8. Other comments.

 

Please, remove empty section 6. Patents.

Author Contributions and Conflicts of Interest paragraphs are missing. Please, provide them in accordance with the journal template.

 

 

 

After detailed consideration of the manuscript, I have found that the results obtained are new and significant for the field. The manuscript is written well but needs some corrections before its publication in the journal.

 

 

So, the paper needs at least minor revision.

Author Response

Please see the attachment