A Semi-Automated Method for Estimating Adélie Penguin Colony Abundance from a Fusion of Multispectral and Thermal Imagery Collected with Unoccupied Aircraft Systems

Round 1

Reviewer 1 Report

Good work. Please proof read it for any possible errors or omissions

 

Author Response

We would like to thank Reviewer 1 for their time and consideration of this manuscript. We have conducted thorough proofreading to double-check all content.

Reviewer 2 Report

Some of my initial concerns were addressed in the revision and there are other suggested changes the authors have decided not to follow up on and there is no use re-stating them here.

I think the workflow diagram in Figure 2 could be moved to a Supplementary Materials.

The fonts in Figure 5 does not look correct to me, and the word spacing is irregular and difficult to read. I cannot tell whether this is an artifact of something in the manuscript upload process or if these problems are actually in the figure itself, but its something that needs to be addressed.

Author Response

We would like to thank Reviewer 2 for their time and consideration of this manuscript.

Point 1. I think the workflow diagram in Figure 2 could be moved to a Supplementary Materials.

Response 1. Thank you for this suggestion, Figure 2 is now in the supplementary materials and referenced in the manuscript as Figure S1 (L177).

Point 2. The fonts in Figure 5 does not look correct to me, and the word spacing is irregular and difficult to read. I cannot tell whether this is an artifact of something in the manuscript upload process or if these problems are actually in the figure itself, but its something that needs to be addressed.

Response 2. Thank you for catching this. It appears that this was an artifact of the conversion of the manuscript to PDF format. We have changed the font of Figure 5 (now Figure 4 in the manuscript) to a style that should not be affected by compression.

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

I wish to thank the authors for submitting this manuscript, because it represents a very worthwhile body of work to help defining the size of remote penguin populations and the guano covered areas. I really enjoyed reading your manuscript since I’m working on similar methods quite a while and I feel that UAVs are much underrepresented in the Antarctic science community despite there obvious advantages. 

Page (P) 3, Line (L) 118, and caption of Figure 1: Please provide more precise coordinates in case someone is looking for it in GIS. Furthermore, the statement "approximately" does not fit with the fact, that you give the coordinates in geographic seconds and even milliseconds. So I would suggest to delete “approximately” and give the coordinates for a point that is within the island/colony and skip milliseconds: e.g. Torgeson: 64°46'20”S, 64° 4'30”W

At this point I noticed that you used a rather uncommon definition of colony although you never defined “colony” in the manuscript. You say that every single of these blue polygons is an own colony? In the consequence, this would mean that on Avian island are 100 colonies and 8 on Torgeson Island. I found this very confusing since a more common definition of a colony includes all breeding penguins in a much greater area e.g. 1 km Ainley 1995 (doi 10.2307/4088776), or 500 m in Jablonski 1984 (Distribution and numbers of penguins in the region of King George Island (South Shetland Islands) in the breeding season 1980/1981) or slightly modified by Pfeifer 2019 (https://doi.org/10.3390/drones3020039). Woehler 1993 (The Distribution and Abundance of Antarctic and Subantarctic Penguins on page 22 and 25), also just listed one colony at Torgerson and Avian Is. I would suggest you either define colony in the method section the way you used it in the paper or you could use e.g. the expression “breeding groups” instead of colony when refereeing to the breeding groups as Jablonski 1984 defined it.

P4, Figure 1: It looks like the north arrow in the overview map is pointing not towards the north. If you are using Arcgis for creating the maps you could: right click on the north arrow, go to properties, and then set “Align To:” to “True North”. But even than the north arrow would only point to the north for a specific Longitude. I would suggest deleting the north arrow and insert a grid instead (in ArcGIS: View - Date frame properties - Grids).

On the detailed maps A and B I would suggest changing the scale bar from km to m to avoid fractional numbers. 100 m is much more intuitive than 0.1km in my opinion. This applies to all figures.

P4 L125: What are the basic parameters of the eBee you used, like wingspan and take off-weight?

P4 L131: Where did you start and land the eBee? Was it in the colonies, close to the breeding penguins or out of the sight of the penguins, did you use a parachute for landing (the area there looks quite rough)?

P4 L133: The flight altitude is defined as the height above the sea level. More relevant would be the flight height which is defined as the height above the ground or the height above ground level (AGL). What was the flight speed (air or ground speed in m/s)?

Most of the time you used the SI unites symbol like m for meter after a number but here you wrote the unite out after a number.

How many flights were carried out? Was there a new flight with each sensor or was there only one flight with all 3 cameras onboard? If there were multiple flights, how much time was in between the single flights?

How was the weather during the flights in particular the thermal IR flights; sunny, cloudy, windy? What time of the day were the flight? All these parameters are necessary to be able to judge the quality of the thermal images.

P4, Table 1: The numbers in the table are not the ground sampling distance but the pixel area. The pixel area is a very uncommon measure for the ground resolution of an image. The GSD is the distance between the center of two pixels on the ground. If the camera sensor has square pixels (like for the Canon S110 and most cameras I know) it is completely sufficient to give just one value for the GSD like 16 mm or 1.6 cm. If the sensor has rectangular pixels, specify the GSD for x and y.

The SI unit symbol for square meter is m², you also used m² in the rest of the paper.

P4 – L143: It’s not clear to me how you created the orthomosaics. First you write you used Pix4D to create the orthomosaics (which are by definition georectified). So why did you use ArcGIS to georectify them again? I suppose the single orthomosaics were not align correctly (there was a shift of 1-8 meter between the absolute position of them)? So, you manually set TIE Points in Arcgis using the georeferenced-tool to align them more precisely as the GNSS accuracy allows?

P4 – L149: Your flights were quite late in the breeding season and chicks should be already very big. So how did you distinguish them?

P4 – L159: Are the reflectance values derived out of the box from the Parrot Sequoia or do you had to process the raw images with a reflectance calibration target? I suppose the RGB images were in DN only? I suggest you specify this in the methods.

P5  - L162: The NDVI is calculated with NIR and red spectral bands, not with the red edge (see your citation [52]). You calculated the Normalized Difference Red Edge Index (NDRE).

P5 – L174: There are both versions in the paper: “square meter” and “square-meter”.

P5 – L177: Defining the threshold values for the multispectral and thermal images is a key point in your method. Could you describe more in detail how you derived them? How many reference points/ selection points were necessary, how did you choose the position of the reference points? Did you also take reference points for non-penguins/guano? How many pixels were included in one point?...

P5 – L181: I would suggest adding a space character between number and unit (except for %), like you did most of the time in the rest of the paper.

P6 – L185: Why did you choose a buffer of 0.6 m here and on line 188 0.5 m? How did you define the buffer size anyway?

P6 – L195: How did you deal with sheathbills that were in the colony area? Is there a way to distinguish them from penguins in thermal imagery?

P6 Figure 3: The legend says “Value”: please specify the unite there. Also, the number of decimal places in the legend are very high and indicating a wrong accuracy of the measurement. I would suggest using at most one decimal places if at all. Please check this too in the other figures.

There is a scale bar in panel B but nor in A.

P6 - L201: “...full island image...” isn’t it just the northern part of the island?

P6 - L205: Can you quantify how much the temperate varies?

P6 - L207: How far must single penguins be apart so they could be separated in the thermal images?

P7 – L209: What’s that cell size in GSD e.g. mm?   

P7 – L210: What’s the final threshold exactly for both colonies? I know they will be different from site to site, but it would be good to get an impression of what you can expect.

P7 – L216: Can you quantify how many hotspots you had (absolute and or relative)? You could then discuss this uncertainty in the discussion section when talking about the limits of the method.

P7 – L224: What are these “accepted ratios”? By whom are they accepted? Can you give us a reference or if they are unpublished how exactly they were conducted? As you can see in Southwell 2013 [10] in figure 2 the ratio of individuals per nest (and therefore also the ratio of individuals per m²) changes dramatically in one season. So the timing when these ratios are recorded is decisive.

P7 – L238:  The term “error rate” is not correct here or at least it has to be specified since there is no general definition of the error rate. So please give us the formula for the calculation if you like to stick with the term "error rate". I think what you meant and calculated is the percentage error = ((True - Observed) / True) *100?

P7 – Table 2:  On decimal place would be more than enough for the mean and SD. Always adjust the number of decimal places with the accuracy of your measurements. 

P10 Figure 5:

For all axes annotations, you didn’t give us the unit of measurement. e.g. Density [individuals/m²], Workflow counts [individuals]. It might be obvious for you but for someone how only sees the graphs without reading the whole paper it could not.

Panel A and B: There is no annotation of the x-axis. I also can not interpret these graphs. Are these histograms? In the text, you say they show “the distribution of colony densities”. So I expected they would show the number of colonies/nest groups :-) in the y-axis and the density on the x-axis.

Panel E and F: I have never seen a graph like this before and it's not intuitive how to read it. Could you give a short explanation of how to read it?

Panel G and H: What does colony number mean? What do you mean with "small" and what with "large"? Isn’t there a way to write the absolute number of individuals in the x-axis or at least describe at which colony size (number of indiv.) the differences start to increase (colony number 80)?

P11 – L323: Do you know if there is vegetation on the islands? Vegetation has a high NDVI/NDRE or NIR reflectance and therefore could interfere with the guano area classification.  See Frettwell 2011, https://doi.org/10.1007/s00300-010-0880-2

P13 – L394: some references are in caps lock e.g. 18, 36. 43...

P13 – L419: Curly brackets in the name of the paper that are not part of the real name

Optional: In the discussion, I missed, at least, a brief comparison with other automated methods for counting individuals in RGB imaged like in Borowicz 2018 page 6 (10.1038/s41598-018-22313-w) or Zmarz 2015 (10.5194/isprsarchives-XL-1-W4-189-2015). Thermal and multispectral cameras are much more expansive and have a much lower spatial resolution than standard RGB cameras. Also if the UAV has not enough room to carry multiple sensors at the same time you have to fly at least twice over the same colony to get thermal and multispectral images. So in which cases do you think your method is worth the effort in comparison to methods with a standard RGB camera? The one big advantage I see, without thinking too much about it, is that the proposed method is very easy to be carried out, you only need a GIS and no special hardware (CUDA GPU) or software as for the Borowicz 2018 method. Also, the method of Zmarz 2015 only works when the ground is homogeneous and flat without shadows or bigger stones in the colony area. But again this is completely optional from my point of view.

Reviewer 2 Report

This manuscript is exceptionally well written and I have few detailed concerns. My overarching concerns are:

1) This manuscript presents a method for incorporating thermal imagery in counting penguins, but there is no comparison to other methods of counting penguins in UAV imagery, and a relatively simple blob detector would likely have performed just as well as the much more complicated workflow presented in this paper. More sophisticated CNNs perform much better than what is presented here and do not require either the complicated workflow or the need for thermal imagery. This paper would have been much stronger if the authors had compared their workflow with the thermal imagery to one that did not use the thermal imagery. In this way, they might have convinced a skeptic like myself that the thermal imagery was actually necessary. Since thermal cameras are heavy, expensive and often difficult to carry across foreign borders due to import controls, it would be incumbent on the authors to make the case that the thermal imagery was in fact beneficial.

2) The workflow described is complex and lacks any novelty, and is likely to be of relatively minimal interest outside the community of penguin researchers. I think the audience for this paper is limited, especially because the workflow itself is so complex and requires, as noted above, a thermal camera. The paper would have been more compelling if there has been some more general statements about the use of integrated optical+thermal imagery that might be of interest in other applications, or a more general analysis of the pros and cons of different analysis approaches.

I have a few additional minor comments:

1) Line 73: The reference included here (12,15-20) are fine but ignore a lot of the work that has already been done on UAVs on the Peninsula, including the work of the German research program. A lot of what is presented here was essentially worked out by the Germans a decade ago.

2) The Introduction proceeds in a non-traditional order, and I would suggest re-arranging it to start with more general applications and then narrowing it down to the specific application at hand. The paragraph starting on Line 86 is a non-sequitor because it "zooms out" suddenly to applications in agriculture (etc.) after having focused on the survey of penguins, and it makes it difficult to follow the thread of the narrative.

3) Figure 5, Panels E and F: What are the diagonal lines for? I can't figure it out from the caption.

Reviewer 3 Report

Excellent research and well written manuscript.

I have added few suggestions in the attached file.

Comments for author File: Comments.pdf