A Colourimetric Approach to Ecological Remote Sensing: Case Study for the Rainforests of South-Eastern Australia
Round 1
Reviewer 1 Report
The authors provided adequate clarifications to my initial comments. I have no major concern with this draft. There are many instances where the authors have used too many citations to back a point. See examples on lines 135 and 256. Probably best to revise and limit to seminal and most recent research work.
Author Response
Thank you for taking the time to provide your appraisal and feedback. We have edited the manuscript accordingly to the feedback you provided.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
A Colourimetric approach to Ecological Remote Sensing: Case study for the rainforests of south-eastern Australia.
In general, the manuscript is interesting and also well done. The only problem is that it is too heavy to read. Perhaps to extensive and descriptive. There are also many figures.
Section 2. line 122: there are 5 subsections;
point 4. Methods, line 346: there are 7 subsections (4.1-4.7).
Authors should try to summarize the above points to improve the readability of the manuscript.
Author Response
Thank you for taking the time to provide your appraisal and feedback. We have edited the manuscript accordingly to the feedback you provided.
Reviewer 3 Report
I found this research very well designed and validated. The proposed hue angle approach is innovative. I thus suggest the acceptance of this paper with one minor suggestion:
The title uses the phrase "ecological remote sensing", but this paper is dedicated to "image classification". Ecological remote sensing is a much more inclusive and broad term. I therefore suggest the authors to consider this point.
Author Response
Thank you for taking the time to provide your appraisal and feedback. We have edited the manuscript accordingly to the feedback you provided.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
In the revised version, the manuscript has been sufficiently improved by the authors to warrant publication in Remote Sensing
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
This is a very interesting study, and very well written. I only have two minor suggestions:
Throughout the paper, there are instances where an in-text citation is missing and it says "Reference missing!". I'm not sure if this is simply an artifact of the online PDF conversion, but I would recommend double checking just to make sure that the citations have not been lost from the authors' original document.
Page 3, Line 10: Perhaps I am just reading it wrong, but I think there might be a typo in the phrase "the most commonly referenced perceptual cue, co-occuring with - association, site, shadow and tonality, shape, size, pattern, and texture - which can be interpreted..." Is the hyphen in the wrong place? Should it instead be "...cue - co-occuring with association..." ?
Reviewer 2 Report
The authors presented a new idea to perform satellite imagery classification—a colorimetric approach and demonstrate its application in mapping rainforest of south-eastern Australia. Although the idea sounds interesting, I have several concerns about the proposed method and writing of this ms.
- Introduction:
- It is too long (5 pages) and choppy (5 subsections but lacks connection or a broad picture for those subsections). I would limit the intro in 2-2.5 pages.
- The authors fail to persuade me why the proposed method matter. The authors tried to justify the proposed methods focusing on three main points: simplicity/scalability, and easier communication to non-expert users (subsections 1.2 – 1.4, section 5).
In term of simplicity/scalability, I do agree that computation of the proposed method is easy. This method also does not require as much ground truth as many other classification methods. However, there are several points to be aware. First, the proposed method is an expert-based (knowledge-based) classification, thus it may not be quite comfortable for non-expert people to perform (while other classification methods rely on training data to bypass this requirement). Second, supervised classification often requires more ground data but also often yield much better accuracy—a critical aspect to consider when mapping land cover. Third, with more computational power are available (for example, GEE), computationally expensive classification methods become more and more feasible. Fourth, scalability can be understood flexibly (computation or parameter transfer). In term of computation, many other classification methods can be easily used at regional/global scale (as many regional/global land cover products are available). Perhaps, they may be more computationally expensive but can be done quite easily with current advance I computer science. In term of transferring model’s parameters, existing classification approaches do face the same issue that they cannot transfer directly to other regions (if far away) or time (different years) due to variation in remote sensing observations due to time and geographic locations. That limitation still holds true for the proposed method as it still based on observation of a certain area over a certain time and manipulating spectral bands cannot mitigate differences in observations).
About communication to non-expert users. First, I believe most of users only care about land cover map (final product) rather than how to create one. Second, although idea behind the proposed method is straight forward, I don’t think it is necessary easy for non-expert to visualize/understand and apply the method because how to choose proper RGB bands (and interpret different colors) require good knowledge about study area and good eyes (I myself cannot differentiate many colors).
I would provide a bit more background about interpreting color (RGB => HSV, etc..). Note that remote sensing is not quite the same with traditional digital image interpretation (although the core concept behind them is somewhat similar). Visualization is not as important in remote sensing as it would be in digital image interpretation. Would be helpful to provide some background about color and visualization at the beginning.
Bottom line, I suggest the authors to provide a good review of state-of-art classifications and consider the proposed method in that context to give strong reasoning why the proposed method matters.
- Methods: the hues were compared with some commonly used vegetation indexes (NDVI, EVI2). Yes, the hues may perform better than VIs but it still does not mean much because the way Vis used in this study is naïve. In recent classification study (especially from main remote sensing journals), no one would use VI to perform classification as shown in this study. Again, the authors need to revisit current classification studies to have a better view of state-of-art classification.
Reviewer 3 Report
A colourimetric approach to ecological remote sensing: case study for the rainforests of south-eastern Australia.
This is a well written paper with a clear message for the readers of the Remote Sensing Journal. It is also methodologically sound and the authors demonstrated ingenuity by advancing the application of colour metric in image processing by blending spectral bands for better feature visualisation. I recommend that the paper be published after addressing the following minor comments.
The authors propose that the one-class, one-index approach as an analytically simple technique, but I wonder if GEE and its code-heavy data sourcing and image processing model supports this message. Although, the use of GEE is timely and demonstrates the importance of cloud computing in remote sensing it may be worth mentioning the fact that anyone seeking to follow the steps propose here needs to be conversant with basic coding skills.
From a methodologically point of view, it is not clear how the performance of the colour metric was compared with vegetation indices that include non-forest areas. Did the authors differentiate forest areas and the range of index values before accuracy assessment? I think that would be a more logical and conceptually rigorous way to compare between the indices.
The authors should consider including in the discussing of their approach the implication of colour metric for monitoring seasonal changes in forest areas and whether it is able to inform local knowledge.
Abstract:
I wouldn’t go as far as suggesting that this approach has any implication for forest landscape productivity – at least not in its present form.
Introduction:
1st paragraph
Vegetation “typing” reads awkward.
Ecological observations or ecological interpretations? Considering that you immediately say “for environmental consultations, negotiations and management….”
2nd paragraph
Replace “Rainforests” with “rainforests”
The authors state that “but it has been observed that false colour RGB composites can distinguish land cover features uniquely and have the potential to serve as indicators of physiognomic traits for vegetation.” Readers would want to know who made these observations? Citing relevant articles that back this claim is the easiest way to address this potential question.
1.2. Scalability and reducing sampling cost with visual imagery interpretation and ecological-colourimetric deductions
Authors should consider simplifying/shortening this sentence“Statistical replication and control can attain reliable site or stand level inferences by isolating patterns and processes at a scale from patterns and processes at other scales, enabling inferences at that scale, but disabling them from being extended to other scales ”
“vegetation typing data”: I think it is simpler to just say vegetation data here.
1.3. Colourimetric ontologies and multidimensional colour blending
Who is the “we” in this sentence after saying humans- “and while humans can only distinguish about 30 shades of grey, we can distinguish about 10 million unique colours….”
Also, the authors need to explain that effective distinction of 30 or more shades of grey is dependent on viewing conditions such as lighting effects.
2nd paragraph in this section
It seems to me that the authors need to give more justification how their approach of combining spectral bands in a HSV colour space is useful for easy distinction of vegetation state or absence of vegetation stress. In its current form, this section does little to justify why we need a colour metric as an alternative for mapping forest areas states.
Methods
Table 1: It appears that the existing 100m resolution classification overestimates forest areas. Or is this a reflection of the difference in dates between existing data and sentinel-2 imagery. If so, it is worth stating the years of the different image scenes.
3.3. Two stage classification design
2nd paragraph
What indices are you referring to? IBest to give an example here just to keep your reader engaged.
3.6. Accuracy assessment design
How did you differentiate forest areas from classifications of other vegetation indices (i.e. NDVI, EVI2 and NDRDI)?
Results
I find this part of this sentence a bit strange – “The Colourimetric hue-based indices and the new Aravena ratio subtraction index consistently produced more accurate maps of Rainforest than the conventional indices in all the ecoregions…” This is because common vegetation indices are not designed for identification of forests alone but for all vegetation types and it doesn’t seem like the authors carried out similar density slicing of vegetation index values to identify forest areas before undertaking accuracy assessment. At least, this is not clear from the methods.
Discussion
I wonder if the authors could discuss the implication or lack of any for phenological monitoring of forest areas of the colour metric approach.
Could the authors also discuss in more detail how the colour metric approach complements conventional vegetation indices.
