Deep Learning in Forest Tree Species Classification Using Sentinel-2 on Google Earth Engine: A Case Study of Qingyuan County
Round 1
Reviewer 1 Report
Commonly-used deep learning algorithms are applied in the forest tree species classification in this study. Some comparison work has been done, yet the study meaning and value need to be strengthen. Otherwise, the work is a lack of scientific value. In addition, English writing needs to be improved and I think the manuscript is not prepared well. Specifically,
1) The study area name needs to be added in the title because the authors made a test and the methods in the manuscript may be not universal in other region (If it is universal, please strength it).
2) Some abbreviations are unnecessary, e.g., “GEE” in abstract because it does not appear again in abstract. Please check all abbreviations in the manuscript.
3) Section 3.3. Transfer learning. More detailed information needs to be provided here, especially the differences among these methods.
4) English writing needs to be improved and I think the manuscript is not prepared well.
Other mistakes
1) P32 “87.90\%” - “87.90%”
2) P36 “; (3) Too ” - “; (3) too”
3) P46 “However, Remote sensing ” - “However, remote sensing ”
4) P86 “This study aims to answer the question” - “This study aims to answer the questions”
5) P88 “as follows: First, ” - “as follows: first, ”
6) P138 “lowing Equation:” - “lowing equation:”
7) P139 P261 “Where ” - “where ”
8) P261 and Eq.2 to Eq.4 . Please keep the style of variable (e.g., “TN”) with the equations
Please notice that the above-mentioned do not mean all mistake, check all text before submit
Author Response
Point 1: The study area name needs to be added in the title because the authors made a test and the methods in the manuscript may be not universal in other region (If it is universal, please strength it).
Response 1: Because we have not selected other regions’ data, so we have added a subtitle of “a case study of qingyuan county”
Point 2: Some abbreviations are unnecessary, e.g., “GEE” in abstract because it does not appear again in abstract. Please check all abbreviations in the manuscript.
Response 2: Because PCA and NDVI were calculated in GEE, So we retain the abbreviation of “GEE”, but added GEE in the subsequent sentence in the abstract. Also, we have checked all abbreviations in the manuscript carefully.
Point 3: Section 3.3. Transfer learning. More detailed information needs to be provided here, especially the differences among these methods.
Response 3: In comparison with the above-mentioned CNNs, ResNet aims to solve the problem of vanishing gradients that occurs in very deep networks by introducing residual connections, which allows the network to learn identity mappings as well as more complex mappings. This allows for much deeper networks, with over 100 layers, to be trained effectively.
Point 4: English writing needs to be improved and I think the manuscript is not prepared well.
Response 4: We have modified some mistakes, and improved English.
Author Response File: Author Response.pdf
Reviewer 2 Report
Dear Authors,
I have included my comments to the PDF file of the manuscript, which requires further revision. Please check the attached PDF to revise your work.
Comments for author File: Comments.pdf
Author Response
Point 1: Rephrase the sentence ”There is no benchmark dataset in forest tree species, a forest tree species dataset (FTSD) was built in this paper”.
Response 1: There is no benchmark dataset in forest tree species, so a forest tree species dataset (FTSD) was built in this paper to fill the gap.
Point 2: FTSD contained nine kinds of forest tree species with 8,815 images, for which area?
Response 2: we have added the research zone “Qingyuan county”.and change the sentence to ”FTSD contained nine kinds of forest tree species in Qingyuan county with 8,815 images.”
Point 3: The images were produced by combining forest management inventory data and sentinel-2 images preprocessed. which date acquisition about sentinel-2?
Response 3: Sentinel-2 images were acquired with less than 20% clouds from 1st, April to 31st, October, including 2017,2018,2019,2020, and 2021.
Point 4: Capital letters ”normalized difference vegetation index”.
Response 4: We have Capitalized “normalized difference vegetation index” and “principle component analysis”.
Point 5: Use reference number instead of year base on the journal style.
Response 5: We have modified all mistakes([5],[6],[10],[11],[12],[13],[15],[16],[20],[21]).
Point 6: 2.Study Area and Dataset Description. I suggest change this heading to be Material and Methods.
Response 6: Thank you for your suggestion, we have changed this heading name and removed the contents(lines 88-104) from the introduction to methods.
Point 7: The county is 49km from north to south and 67km from east to west, with a total area of 1,898km2. Rephrase this sentence.
Response 7: The county has a total area of 1,898 km2, which spans 49 km from north to south and 67 km from east to west.
Point 8: Change the figure caption to be imformative, please.
Response 8: We are sorry for the mistakes when changing the figures’ caption from Letex version to Word version. Now, we have revised the captions of Figure 1 and Figure 4.
Point 9: Rephrase the sentence “A Digital Elevation Model(DEM) to project the images in cartographic geometry is used in the product. Per-pixel radiometric measurements are provided in top of atmosphere (TOA) reflectance along with the parameters to transform them into radiances.”
Response 9: Level-1C products utilize a Digital Elevation Model (DEM) to project the images into cartographic geometry. In addition to per-pixel radiometric measurements in Top of Atmosphere (TOA) reflectance, the products also include parameters for converting the measurements into radiances.
Point 10: Rephrase the sentence “ For the classification of forest tree species based on Sentinel-2, a dataset was used in our research, which is named Forest Tree Species Dataset (FTSD) in the paper.”
Response 10:, The Forest Tree Species Dataset (FTSD) was used in our research to classify forest tree species using Sentinel-2 data.
Point 11: Please provide the URL from which you download the data “ RESISC-45 is a famous satellite imagery database”
Response 11:, We had given the URL in Data Availability Statement. Now, we have added the URL in the paper as RESISC-45(https://www.tensorflow.org/datasets/catalog/resisc45) is a famous satellite imagery database and is widely used in remote sensing classification.
Point 12: Place the overall workflow at the beginning of the Material and Methods
Response 12: We have done it.
Point 13: Line 191 change the word “paper” to “study”
Response 13:, We have done it.
Point 14: Please add information about Stochastic Gradient Descent (SGD)
Response 14: SGD is an optimization algorithm for finding the minimum of a function. SGD is particularly useful when the function to be minimized is very large or has many local minima, because it is computationally efficient and only requires one training example at a time to update the parameters. SGD is widely used in machine learning and deep learning, particularly for training neural networks.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Thanks for your revision.
The most problems are solved in the new version, and the manuscript is good for published now.
Reviewer 2 Report
Dear Authors,
Thank you for the revised version of your manuscripts.
Regards.