Mapping Relict Charcoal Hearths in New England Using Deep Convolutional Neural Networks and LiDAR Data

Round 1

Reviewer 1 Report

Overall, this is an excellent paper, mild issues are:

1. There are different implementations of U-NET using Keras + tensorflow in github as well as other web pages (https://keras.io/examples/vision/oxford_pets_image_segmentation/), please provide the link of the U-NET implementation you used.
2. Lines 214-216 state that to prevent overfitting a bacth normalization layer and a dropout layer were added to the convolutional blocks. There is concern that the dropout layer should not be used in
   convolutional blocks, they should used only for fully connected layers (https://arxiv.org/ftp/arxiv/papers/1512/1512.00242.pdf, 
   https://www.kdnuggets.com/2018/09/dropout-convolutional-networks.html), so batch normalization is used on convolutional layers or with other dropout techniques such as blockdropout (https://arxiv.org/pdf/1502.03167.pdf, https://arxiv.org/abs/1810.12890). Please try using only
   batchnormalization to fit the model and compare when using also dropout in this case. On the other hand, given the small batch size used (due to memory issues) is possible that batch normalization does not help much in this case, so please discuss this issue in the paper. 

minor issues:

lines 108-109: "DEM tiles were retrieved from ground LiDAR point cloud that an average point spacing is between 0.7m to 1m" should be changed to "DEM tiles were retrieved from ground LiDAR point cloud, where an average point spacing is between 0.7m to 1m"
line 114: CT ECO should have a reference (https://cteco.uconn.edu/data/lidar/index.htm).
Figure 9 is missing the (C) label

Author Response

Thank you for your letter of November 3 inviting revisions to our manuscript “Mapping Relict Charcoal Hearths in New England using Deep Convolutional Neural Networks and LiDAR Data.” We have taken the reviewers’ recommendations very seriously, substantially clarifying our methodology. We have also made several corrections considering reviewers’ suggestions. A detailed response to reviewer comments is provided with the revision.

Thank you for considering this revised paper for publication in Remote Sensing. If I may provide any further clarifications about the revised work or its significance, please do not hesitate to contact me at [email protected].

Author Response File: Author Response.docx

Reviewer 2 Report

This is an interesting and well-structured paper with clear objectives that are followed by clearly presented analysis and convincing results. Thus, I only have a few minor comments:

Page 1, line 44-45: I am not sure about the use of the term (charcoal)mound here. If it refers to ref. 11 this study is mainly about grave barrows/mounds (and Celtic fields).

Page 3, line 109: Check the wording in the sentence beginning with: DEM tiles were.....

Page 4, line 123: Please add [33] after (Verbovsek et al. 2019).

Figure 2, caption: Consider to split up A, B, C on the one hand and D and E on the other since they are different kinds of images. It will make the figure and caption more readable.

Table 2: The scenario-numbers should preferably refer to 1. Preparing data set in Figure 4.

Page 7, line 226: Please define towns: in many places town is used for a build-up area, but here it is perhaps more like a municipality (administrative entity)?

Page 7, line 231-233: Please elaborate this step (Next, noisy and.....), i.e. explain better.

Page 7, line 238: The terms true positives, false negatives etc. should be explained shortly. 

 Page 9, line 273: Could be added after LiDAR quality: ...caused by dense vegetation

Page 9, line 276-277: The comparison between coniferous and deciduous forest cover should also shortly explain the impact of season variations. I.e., that it is a precondition to obtain better quality in deciduous forest that the scanning is done in the part of the year where the trees are without leaves. Consider to mention this on page 3, line 110-112 instead. 

Table 7: add the full word first time avg. is used, or explain in caption.

Table 7: charcoal kilns (10-20m) are mentioned twice under the study with ref. [9]?

Page 15, line 375: I will suggest that you add that this is not a general conclusion but valid only in those cases where land use has left traces identifiable on lidar data.

Page 15, line 378: Reference 48 is not found in the reference list.

Author Response

Thank you for your letter of November 3 inviting revisions to our manuscript “Mapping Relict Charcoal Hearths in New England using Deep Convolutional Neural Networks and LiDAR Data.” We have taken the reviewers’ recommendations very seriously, substantially clarifying our methodology. We have also made several corrections considering reviewers’ suggestions. A detailed response to reviewer comments is provided with the revision.

Thank you for considering this revised paper for publication in Remote Sensing. If I may provide any further clarifications about the revised work or its significance, please do not hesitate to contact me at [email protected].

Author Response File: Author Response.docx