Wildland Fire Tree Mortality Mapping from Hyperspatial Imagery Using Machine Learning
Round 1
Reviewer 1 Report
The structure of the manuscript in the first version was inappropriate. However, from the technical, methodological, and scientific side the work is quite advanced and interesting for the scientific community and readers. Authors in the resubmitted version improved the drawbacks and now it has a more sound appearance. The only remark is that the concluding remarks from line 492 onwards, 5.1. Future work is more appropriate for the Discussion section.
Author Response
See attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The article (remote sensing 1043498) entitled “Wildland Fire Tree Mortality Mapping from Hyperspatial Imagery using Machine Learning” is an interesting piece of research about the use of machine learning to identify burn severity as a measure of canopy reduction in forested systems
The article represents a very high application for managers.
My recommendations are:
1.- Rewrite the introduction section focusing in the background and the gaps that the article cover. At the end of the introduction explain the objectives and the hypothesis of the work.
Add references to explain that fire severity could evaluated through the canopy reduction.
2.- The material and methods section must be revised trying to simplify and clarify the content of the article.
3.- Apparently, part of the results section continues as method, so it must be revised to focus exclusively on the results.
4.- Expanding the discussion
Author Response
See attachments
Author Response File: Author Response.pdf
Reviewer 3 Report
Dear Authors,
This manuscript is a very good addition to the literature. At first I made several notes on clarifications I thought were needed. But all were provided later in the paper. The paper really contains two stories; the comparison of UAS and Landsat derived data, and the utility of UAS data in detecting post-fire changes. Maybe put more emphasis on the unburned comparison early in the paper. This manuscript is very complete, providing a lot of information on the thought processes of the authors.
A few thoughts on potential improvements:
One addition useful to readers is a section with some more information on the platforms, sensors, and details of the UAS operations.
The terms active and passive crown fire, while adequately explained in the background section, might not give the best idea of post fire conditions as ‘observed’ by the UAS sensor. High-, moderate-, and low severity are more common in the literature and better describe the post-fire conditions. And it relates to the canopy cover changes.
What is the optimal time post-fire to collect data for your objectives? Time-since-fire matters (delayed tree mortality is common).
68 moderate severity is more accurate (instead of medium)
93 – 95 Not sure what this means. Please be a little more specific on what the observers saw and how it relates to the imagery.
106 Mortality can also be caused by prolonged exposure to heat on or below the surface, where the roots or cambium sustain damage not observed immediately.
373 Be more descriptive of the fire area and the size of the collect. Use table instead of bullet points with somewhat vague descriptions. Maybe add a map. Maybe expand table 1 with specifics.
Overall it is a great paper. Each section is thorough and well referenced. The tables and figures are of good quality.
Author Response
See attachment
Author Response File: Author Response.pdf
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
See attached.
Comments for author File: Comments.pdf
Reviewer 2 Report
32-83 Introduction: Please consider including an overview of the previous research related to the estimation of the canopy cover density by UAS hyperspatial imagery. Provide an overview of the approaches that have been used for segmenting the canopy layer; which statistical and machine learning models are state of the art, what are usual features (satellite indices, textural features…) that show the best results in disclosing the canopy for the purpose of the canopy mask production. These are requirements required by the MDPI journal concerning this section; The current state of the research field should be reviewed carefully and key publications cited.
84-385 Materials and methods: Please give the name and version of any software used and make clear whether the computer code used is available. Please consider providing the flow chart of your processing steps for better visualization and understanding for the reader audience.
129-130 The sentence: Canopy cover describes the vertical projection of the tree canopy onto a horizontal plane in forested ecosystems. Is repetitive and redundant (85-86)
158-161 Question about: “What level of accuracy can one expect for the value of canopy cover calculated from the 600 × 600 pixel region of the hyperspatial orthomosaic corresponding to a Landsat pixel? Precisely, what is the standard error of this value? Is this significantly smaller than what one might hope to obtain from LANDFIRE data?” present one of the key issues in this research. It would be more suitable to place this sentence at the end of the introduction section where the aims of the research and related questions are provided.
162-215 Could these calculations be presented in a more tidy manner? Consider placing this in the more suitable results section instead of in the materials and methods.
214-215 This presents one of the conclusions of the work, suitable for the concluding remarks of your paper at the end.
216-237 Please consider including basic information on the LANDFIRE project in the introduction. What is the state of the art of fire monitoring in the US provided by LANDFIRE? What are the observed shortcomings that you noticed as the motivation for your research? You provide in this chapter comparison between LANDFIRE pixel of 30m with your results from UAS and it would be practical to provide additional information to readers also in the introduction section.
459-478 Discussion: Please take notice in the MDPI guide for authors: Authors should discuss the results and how they can be interpreted in perspective of previous studies and the working hypotheses. The findings and their implications should be discussed in the broadest context possible and limitations of the work highlighted.
Please consider providing a broader overview of the previous results of canopy density estimation from UAS. Is there any quantitative metric that could be used to make a comparison of the published research with your results? Please comment on the advantage of your relatively complex and time-consuming approach (using MR-CNN) in relation to, for example, much simpler approaches such as simple thresholding of indices derived from the UAS image, some unsupervised image segmentation technique, or simpler statistical or machine learning model used for this purpose. What are the utmost advantages and eventual drawbacks of using your approach?
Is your approach scalable and to which extent? LANDFIRE results are relatively course (30m) but they have coverage of the whole of the US. However, UAS daily coverage is approx. 400 ha what you mentioned in your paper. For what size or extent of the area is your approach mostly suitable; 1000, 5000, 10000 hectares, maybe for some local forest administrative unit?
