Next Article in Journal
Temporal Variation and Spatial Structure of the Kuroshio-Induced Submesoscale Island Vortices Observed from GCOM-C and Himawari-8 Data
Next Article in Special Issue
HTD-Net: A Deep Convolutional Neural Network for Target Detection in Hyperspectral Imagery
Previous Article in Journal
Estimating Ground-Level Particulate Matter in Five Regions of China Using Aerosol Optical Depth
Previous Article in Special Issue
Global and Local Tensor Sparse Approximation Models for Hyperspectral Image Destriping
Article
Peer-Review Record

Fusing China GF-5 Hyperspectral Data with GF-1, GF-2 and Sentinel-2A Multispectral Data: Which Methods Should Be Used?

Reviewer 1: Anonymous
Reviewer 2: Anonymous
Received: 7 February 2020 / Revised: 4 March 2020 / Accepted: 5 March 2020 / Published: 9 March 2020
(This article belongs to the Special Issue Advanced Techniques for Spaceborne Hyperspectral Remote Sensing)

Round 1

Reviewer 1 Report

The topic of this manuscript is of interest and well written. I liked reading it. However, I have three main comments:

  • L86-87: The authors claim that: Compared with others, our work favors two main contributions. The proposed evaluation framework for fusing hyperspectral images is more comprehensive and thoughtful than others!!  This claim needs to be explained well by providing a literature review!  Now, as a reader, I ended up a bit disappointed that I cannot net see any justification for this claim! I think you may move this part to the discussion section or you can keep it here and give a comprehensive literature review.
  • I think it would be a good idea to merge both Data and methods in one section (Data and methodology)
  • The authors need to add more reference comparison to the discussion section! In other words, the findings of this study need to be compared with earlier studies. I cannot see a single reference in this section.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

General comments

The manuscript deals with the interesting topic of increasing the spatial resolution of satellite hyperspectral data (generally @30 m). The authors evaluate different fusion approaches for the Chinese GaoFen-5 sensor with high-resolution satellite multispectral data (@10, 8, 4 m).

The interest of RS readers is very high considering the large number of Hyperspectral Satellite Missions that will soon be in orbit and those that have recently been launched.

 

The paper is quite long, some time redundant and difficult to follow. Despite the length, in some parts information is missing, even in vital sections of the manuscript.

As an example, to evaluate the spectral distortions of tested fusion methods, the authors used resampled images of the same GF-5 as reference images because no high spatial resolution hyperspectral data were available. This is certainly an excellent solution; but it must be taken into account that the selection of the re-sampling algorithm can strongly influence the assessment of spectral distortion metrics by modifying the reference images.

Similarly, no details are provided on the reference classifications used to evaluate the “application performances”.

These represent very critical points to be better addressed to support the results of the study.

 

A rearrangement of the manuscript is necessary to be more clear and fluent. Specific comments are listed below.

 

A revision of English is suggested to eliminate some mistakes and make sentences more fluid. Additionally, you can eliminate many redundant introductory sentences for each section.

 

 

Specific comments

 

Introduction: A slightly more in-depth overview of spaceborne HS sensors in orbit and scheduled would be useful for RS readers.

 

Lines 37-39: “Its ..(GF-5)..spatial resolution surpasses or equals to those of most on-orbit or planned spaceborne HS imagers, e.g., EO-1 Hyperion and HICO of USA [1,2], DESIS of Germany [3], HysIS of India [4]…”.

Actually, EO-1 was decommissioned in March 2017, and therefore the Hyperion sensor as well. Similarly, HICO operations ended in September 2014.

Please delete these sensors that are no more on-orbit and replace them with the following sensors with lower or equal spatial resolution:

PRISMA of Italy [Pignatti et al.] and HISUI of Japan [Matsunaga et al.] recently on-orbit (launch in 2019), and EnMAP of Germany (scheduled in 2020).

 

Pignatti, S., Acito, N., Amato, U., Casa, R., Castaldi, F., Coluzzi, R., ... & Matteoli, S. (2015, July). Environmental products overview of the Italian hyperspectral Prisma mission: The SAP4PRISMA project. In 2015 IEEE International Geoscience and Remote Sensing Symposium (IGARSS) (pp. 3997-4000). IEEE. DOI: 10.1109/IGARSS.2015.7326701

 

Matsunaga, T., Iwasaki, A., Tsuchida, S., Iwao, K., Tanii, J., Kashimura, O., ... & Mouri, K. (2017, July). Current status of hyperspectral imager suite (HISUI) onboard International Space Station (ISS). In 2017 IEEE International Geoscience and Remote Sensing Symposium (IGARSS) (pp. 443-446). IEEE. DOI: 10.1109/IGARSS.2017.8126989

 

 

 

https://www.usgs.gov/centers/eros/science/usgs-eros-archive-earth-observing-one-eo-1-hyperion?qt-science_center_objects=0#qt-science_center_objects

 

https://eo1.gsfc.nasa.gov/new/EO1EndofMissionPlan/index.html

 

https://www.nasa.gov/mission_pages/station/research/news/HREP_HICO_Legacy

 

http://hico.coas.oregonstate.edu/

 

 

 

Lines 39-40: “… both the number of bands and the swath width are advantaged to other spaceborne HS sensors.” Please explain such a sentence.

 

Lines 54-67: The detailed description of the fusion categories and algorithms should be moved in section 3.1 Fusion Methods. Moreover, please add some details on Color Mapping-based methods. Compared to the other methods it is no well explained.

 

Table 1: To correctly associate the different spatial resolutions with the proper band of MS sensors, please add a column for each MS sensor and list close to the bands the respective resolution. Alternatively, you can indicate only the bands used for the fusion methods.

 

Section 2.1 (GF-5): No information about the preprocessing steps (shown in figure 1) are present throughout the manuscript. Please add in this section details on data calibration, atmospheric correction, cloud detection, image registration, etc. As the authors highlight, these are essential steps (lines 282-284), and can greatly affect the tests on fusion methods.

 

Moreover, please add information on where the hyperspectral data were gathered (e.g., website, distribution database) and the level of the input product (e.g. L0, L1).

 

Section 2.2 (MS sensors): Similarly to the GF-5 section, please add information on the source database, level of data, and preprocessing phases.

 

Section 3.2.1 (Quantitative Evaluation Measures): Due to the unavailability of high spatial resolution hyperspectral data, you use the “resampled” GF-5 images as reference images to evaluate spectral distortions (lines 239-240). No information on the adopted resampling method is present.

This represents a very critical point because the evaluation of the spectral distortion measures strongly depends on the reference images. Therefore, a more detailed description of the adopted resampling method is mandatory. In addition, an assessment of how different resampling algorithms affect the spectral metrics would be very helpful. Tests can be run on at least one of the nine subzones to support your selection (to be added as a Supplementary Material file).

 

Moreover, please separate the spectral and spatial measures into two sections.

 

Section 3.2.2 (Classification Evaluation Measures): Please add information on how the reference classifications to evaluate the classification accuracy of fused images are obtained.

 

 

Figure 1: The cloud detection step is not included. As the author highlight in line 73 “...even slight cloud contaminations, will definitely complicate the image fusion procedure”.

The reference image for the assessment of the spatial distortions is not generated from GF-5 data, please add the correct source.

 

Line 342: Please specify the step of cross-validation, on what it is implemented.

 

Table 4: Such a table can be moved to a Supplementary Material file.

 

Lines 352-357: Introductory sentences can be removed.

 

Lines 375-376: Please add a reference for this sentence “The number of endmembers in MAP-SMM affects the behaviors of SMM and accordingly limits the enhancement of spatial resolution of HS data”.

 

 

Table 5: To show the best and second-best result it is more intuitive to use single and double underlines respectively, instead of using bold and single underline. Moreover, the two rounded decimals of the first set of SAM//Two-CNN-FU have to be standardized.

The same comments are valid for the other similar tables (Table 6, and 7).

 

 

Figure 4: To facilitate the comparison for readers it is useful to replace the letters of the insets with the algorithm acronyms. Moreover, uniform the dimension of hyperspectral image to the other images (the hypercube drawing does not provide information).

Finally, the resolution of the images has to be increased.

 

The same comments are valid for the other similar figures (Figure 5, 7, 8, 10, and 11)

 

Lines 405-415: Table 6, Figures 7, 8, and 9 have to be moved in the respective GF-5//GF-2 section (4.3.2). 

 

Lines 466-467: Please, explain the approach used to generate the complete evaluation results shown in Table 8.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

I am satisfied with the responses to my review and the associated revisions. I recommend publication.

Author Response

Thanks for your affirmation of our work

Reviewer 2 Report

General comments

I appreciated the efforts to improve the readability of the manuscript, but due to the length of the text and the presence of many 4th level subsections, also the revised version remains difficult to follow. Moreover, many of the previous comments are replayed as implemented in the cover letter instead they are not present in the text.

A further revision is, therefore, necessary to make the structure of the document more fluid, to improve readability and provide useful details to highlight the peculiarities and novelty of the study.

 

To avoid misunderstandings and speed up the review process, many tips have been added directly to the manuscript word file (revision mode to allow acceptance or refuse).

 

Manuscript Structure

Although some paragraphs have already been divided, due to the length of the manuscript and the presence of many nested subsections, the revised version remains difficult to follow. To reduce the subsection levels, I suggest splitting Data and Methods by including the data processing in the first one and the parameter setting into the methods.

To support this process, here there are two hypotheses of a possible relocation of the sections (the first one, more linear, is already implemented in the attached word file. Attention !!!, the number of figures, tables, and formulae have to be rearranged).  

 

I hypothesis of structure

 

  1. Introduction
  2. Data

  2.1. GF-5 Spaceborne Hyperspectral Sensor

  2.2. GF-1, GF-2, and S2A Spaceborne Multispectral Sensors

  2.3. Data Preprocessing             

  2.4. Study Area and Fusion Datasets

  1. Methods

  3.1. The Study Framework

  3.2. Fusion Methods

    3.2.1. CS-based Methods: GSA

    3.2.2. MRA-based Methods: MTF-GLP and SFIM

    3.2.3. Subspace-based Methods: CNMF, LANARAS, FUSE, MAP-SMM and TWO-CNN-FU

    3.2.4. Color Mapping-based Methods: HCM

    3.2.5. Parameter Settings

  3.3. Comprehensive Evaluation Measures

    3.3.1. Spectral Evaluation Measures

    3.3.2. Spatial Evaluation Measures

    3.3.3. Classification Evaluation Measures

    3.3.4. Computational Efficiency Measures

  1. Results

  4.1. Fusion Results of GF-5 and GF-1

  4.2. Fusion Results of GF-5 and GF-2

  4.3. Fusion Results of GF-5 and S2A

  1. Discussion
  2. Conclusions

  Appendix

  References

 

II hypothesis of structure

 

  1. Introduction
  2. Data

  2.1. GF-5 Spaceborne Hyperspectral Sensor

  2.2. GF-1, GF-2, and S2A Spaceborne Multispectral Sensors

  2.3. Data Preprocessing             

  1. Methods

  3.1. Fusion Methods

    3.1.1. CS-based Methods: GSA

    3.1.2. MRA-based Methods: MTF-GLP and SFIM

    3.1.3. Subspace-based Methods: CNMF, LANARAS, FUSE, MAP-SMM and TWO-CNN-FU

    3.1.4. Color Mapping-based Methods: HCM

  3.2. Comprehensive Evaluation Measures

    3.2.1. Spectral Evaluation Measures

    3.2.2. Spatial Evaluation Measures

    3.2.3. Classification Evaluation Measures

    3.2.4. Computational Efficiency Measures

  3.3. Test Implementation

    3.3.1. The Study Framework

    3.3.2. Study Area and Fusion Datasets

    3.3.3. Parameter Settings

  1. Results

  4.1. Fusion Results of GF-5 and GF-1

  4.2. Fusion Results of GF-5 and GF-2

  4.3. Fusion Results of GF-5 and S2A

  1. Discussion
  2. Conclusions

  Appendix

  References

 

Specific comments

 

Line 16: “… and the acquired images are advantageous to those of other on-orbit or planned sensors.” Please detail such a cryptic sentence (see word file for a hypothesis of details)

 

Lines 45-46: (old Lines 37-39) – The comment on the previous manuscript version as well as the attached links on “no more on-orbit sensors” referred only to EO-1 Hyperion and HICO.

DESIS of Germany [3], HysIS of India [4]…” launched in 2018 are still operating and should remain in the list for a complete overview of currently available Hyperspectral spaceborne sensors.

Obviously, Table 1 has to be integrated also with DESIS and HysIS you previously mentioned.

 

Table 1: As the authors correctly report at line 37 “Germany plans to launch EnMAP hyperspectral satellite in 2020”; therefore, add scheduled in the corresponding launch time of the table and modify the caption accordingly.

 

Figure 2. Study areas: Please add letters for the images of the three study sites and describe them in the figure caption.

 

Quantitative Evaluation Measures: Old comment …“In addition, an assessment of how different resampling algorithms affect the spectral metrics would be very helpful. Tests can be run on at least one of the nine subzones to support your selection (to be added as a Supplementary Material file).”

The authors did not perform the suggested tests on how different resampling algorithms affect the spectral metrics. They added a description of the resampling methods (Discussion section) by enhancing different pros vs cons of each resampling.

Unfortunately, although such a discussion can be considered sufficient, they have not clearly indicated what the method adopted is. In particular, the selection must be declared when the authors describe the reference images for the spectral evaluation metrics (Method section).

 

Classification Evaluation Measures: Old comment …“Please add information on how the reference classifications to evaluate the classification accuracy of fused images are obtained.

 

Please add the information on training samples and verification samples provided in the reply also into the text.

Moreover, information on the used classification algorithm/approach is still missing into both the manuscript and reply.

 

We can consider the classification maps and the validation samples (cited in the reply) as part of the fusion datasets, which are used as input for the Classification Evaluation Measure. Therefore, more info is required on their acquisition/elaboration in the Dataset section.

 

Computational Efficiency Measures: The unit of time is missing.

 

Figure 1 - Flowchart: Old comment “…The reference image for the assessment of the spatial distortions is not generated from GF-5 data, please add the correct source.”

The info on reference images for the spatial distortion assessment has been added into the text but the correct box is not present the Flowchart figure.

 

Figure 4: Old comment …”To facilitate the comparison for readers it is useful to replace the letters of the insets with the algorithm acronyms. …”

The authors improved the resolution of the images, but the acronyms of fusion methods were not added to the respective images.

 

Figures 5, 8, and 11: They are not cited in the text.

 

Figures and Tables: often they are placed before they are cited in the text or far from the citation.

 

Captions: In general, the captions of both Tables and Figures are quite inadequate/skinny (they are not self-consistent).

 

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Back to TopTop