Thermal Infrared UAV Applications for Spatially Explicit Wildlife Occupancy Modeling

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The article employs geospatial techniques to model occupancy of different species. However, it lacks the visual features  needed to tackle such a topic. All visuals refer to the hotspot models, plus a couple box and whiskers plots.

In that regard, I'm presenting several major points that need to be addressed, but in my opinion the article needs major revision to be more in line with a journal like Land.

First, and foremost, the materials and methods section lacks, essentially, "materials" and "methods". Which UAV was used? What are its specs? What camera was used? What was the resolution of the imagery? How many images were taken (or was it video)?

Also, how were those images classified to determine the spatial distribution and occupancy patterns? Which specific programs or algorithms were used? 

The materials and methods section also needs a well explained methodology, with a diagram showing all steps of the process - from the acquisition of imagery to the creation of hotspot models and hypothesis tests. This diagram should be accompanied by a list explaining every step and how they were conducted.

The results section ends up being very short and devoid of proper maps, graphs or explanations on the procedures that were taken, so it is very difficult to infer any constructive criticism. 

On top of that, as a general rule, the discussion section should be based on the results but also related to current bibliography so that results that were obtained may either reinforce what has been analyzed qualitatively or contrast what was thought to be the reality. Despite the long bibliography, in the results section, there are few citations in that regard. The ones there, whilst appropriate, are much more related to complementary matters on the use of UAVs or the spatial distributions of species. What would be interesting would be to find other studies, based on different technologies, that come to the same conclusions.

Finally, the conclusions section is very short, containing not more than one paragraph. 

Kindly revise those points and resubmit the article after they have been addressed, for a more in-depth discussion. 

--------------------------------------------------------------------------------

 

This revised version of the article was greatly improved. The materials section now provides enough information about the sensors and the methodology makes sense.

Please provide a diagram showing the steps, which will enhance the readability of the article.

Also, it wasn’t clear to me if the location of animals was estimated through fully manual or semi-automatic techniques. If the aggregation of mammals’ locations was performed through purely visual methods, the limitations of such an approach should be addressed in the results and discussion sections.

The results section was greatly enhanced, but the conclusions are still somewhat short. I suggest, like in my previous report, to relate the results to current bibliography so that results that were obtained may either reinforce what has been analyzed qualitatively or contrast what was thought to be the reality. Despite the long bibliography, in the results section, there are few citations in that regard. The ones there, whilst appropriate, are much more related to complementary matters on the use of UAVs or the spatial distributions of species. What would be interesting would be to find other studies, based on different technologies, that come to the same conclusions.

Kindly revise those points and resubmit the article after they have been addressed.

Author Response

Thank you for your thoughtful and detailed feedback on my manuscript. I have carefully reviewed your comments and made substantial revisions to address the concerns raised. 

Please provide a diagram showing the steps, which will enhance the readability of the article.

Thank you for your feedback. I have added a list of steps in Table 1.

Also, it wasn’t clear to me if the location of animals was estimated through fully manual or semi-automatic techniques. If the aggregation of mammals’ locations was performed through purely visual methods, the limitations of such an approach should be addressed in the results and discussion sections.

Thank you for your thoughtful comment.

In the Results and Discussion sections, I have clarified how the manual approach was applied and explicitly mention the limitations associated with this methodology in the discussion. These include the potential for observer bias and increased subjectivity, which can sometimes limit the precision of location estimates. I have added a section about study design considerations.

We hope this explanation addresses your concern. Thank you again for bringing this up.

The results section was greatly enhanced, but the conclusions are still somewhat short. I suggest, like in my previous report, relating the results to current bibliography so that results that were obtained may either reinforce what has been analyzed qualitatively or contrast what was thought to be the reality. Despite the long bibliography, in the results section, there are few citations in that regard. The ones there, whilst appropriate, are much more related to complementary matters on the use of UAVs or the spatial distributions of species. What would be interesting would be to find other studies, based on different technologies, that come to the same conclusions.

Thank you for your thoughtful comment. I appreciate the suggestion to strengthen the conclusions by relating the results to the current body of literature. In the revised manuscript, I have rewritten parts of the methods and results to add citations. I have also expanded and reorganized the discussion to incorporate relevant studies that utilize alternative methods, such camera trapping, manned aerial surveys, and RGB UAV.

Reviewer 2 Report

Comments and Suggestions for Authors

The study carried out by Bohnett et al. “Thermal infrared UAV applications for spatially explicit wildlife occupancy modeling“ appears very interesting for the methodological approach used. However, the manuscript leaves something to be desired because in some parts it gets lost in unnecessary details (e.g. what an analysis of variance is for) and in others it lacks fundamental information such as e.g. the extension of the study areas. In this regard, the authors on lines 213-215 state that the flights were carried out based on two grids of 151 and 144 cells of 150 m². I believe there is a mistake or that the matter was not explained well. In fact, if this were the case, an area of ​​2.26 ha in the CNP and 2.16 ha in the KCF would have been covered. It is impossible to formulate occupancy models for wild ungulates and large carnivores based on such small areas.

I believe that the authors should make a considerable effort to improve the readability of the text and to provide all the necessary information.

Other comments directly on the manuscript

 

Comments for author File: Comments.pdf

Author Response

Kindly revise those points and resubmit the article after they have been addressed.

Peer Reviewer 1: The study carried out by Bohnett et al. “Thermal infrared UAV applications for spatially explicit wildlife occupancy modeling“ appears very interesting for the methodological approach used. However, the manuscript leaves something to be desired because in some parts it gets lost in unnecessary details (e.g. what an analysis of variance is for) and in others it lacks fundamental information such as e.g. the extension of the study areas.

Thank you for your valuable feedback. We appreciate your recognition of the study’s methodological approach. I have made revisions throughout the manuscript to address this and have incorporated your line by line comments.

In this regard, the authors on lines 213-215 state that the flights were carried out based on two grids of 151 and 144 cells of 150 m². I believe there is a mistake or that the matter was not explained well. In fact, if this were the case, an area of ​​2.26 ha in the CNP and 2.16 ha in the KCF would have been covered. It is impossible to formulate occupancy models for wild ungulates and large carnivores based on such small areas.

Thank you for your valuable and insightful feedback. We appreciate your concerns regarding the limited area covered by the UAV grids and its implications for the applicability of occupancy models for large carnivores and wild ungulates. We have carefully reviewed your comments and made significant revisions to clarify the rationale behind our study design and to address these concerns directly.

First, we acknowledge the limitations of applying occupancy models in small-scale grid-based designs, which we had discussed in the introduction:

“Since grid-based sampling designs are not geographically independent, the geographic closure assumption may lead to biases, where the models represent 'habitat suitability' rather than true 'occupancy' [50,51].”

Our estimates, therefore, are based on habitat use rather than true occupancy, a distinction we have emphasized throughout the revised manuscript. Numerous studies have used occupancy models for habitat suitability assessments, and we similarly applied this framework to evaluate habitat suitability and biodiversity differences between the national park and community forest.

Given the current limitations of UAV-based methods—particularly line-of-sight restrictions and the small survey area feasible for flight operations—we designed this research along the lines of simulation studies that have demonstrated that surveying blocks within a larger study area is effective. To address your comments, we have added a new discussion section on study design considerations for grid-based sampling. We cite relevant simulation studies that demonstrate how grid-based block designs can yield comparable density estimates to transect-based designs, even in constrained study areas. These references reinforce the validity of our approach and provide a broader context for interpreting our findings. I have also added a section that discusses the effective area sampled of the grid cell may improve detectability compared to camera traps. Please check out the new expanded discussion on these topics.

Also, while large-scale surveys would indeed be valuable, the community forest area within the grassland is limited in this area, and no other suitable replicates exist nearby in the grasslands. Expanding the study would have required incorporating a suite of nature reserves throughout India and Nepal, which was beyond the scope of this project of demonstrating the methods and analysis techniques. Despite the dataset’s limitations, the models successfully generated valid results confirmed by evaluation metrics.

We have significantly expanded the discussion and conclusions to clarify our study’s objectives, methods, and findings. This manuscript serves as a proof of concept, offering valuable insights for future research in larger and more diverse landscapes, and I have made that clear throughout the various sections to improve clarity of our intent. We hope that the revisions address your concerns and demonstrate the relevance and validity of our approach.

Thank you again for your thoughtful and constructive review.

I believe that the authors should make a considerable effort to improve the readability of the text and to provide all the necessary information.

Other comments directly on the manuscript

Thank you for your valuable feedback. We have made substantial revisions to improve the readability of the manuscript and ensure all necessary information is clearly presented. Your inline comments were carefully addressed, and we appreciate your insights in strengthening the work.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The authors construct occupancy models to test whether there are differences in biodiversity between a protected area and an adjacent community forest. The species observation data used as input for the occupancy models come from manual observations of UAS-TIR imagery. Species occupancy models are difficult to carry out due to the sampling biases inherent in most observation datasets, and the authors pilot a new way to capture abundance data in a small study area. The authors make it clear that the study is intended as a pilot, but there are some areas where additional detail and considerations will strengthen the manuscript.

• One issue that would benefit from additional clarification is whether the two areas can indeed be studied as independent study areas. While the study is described as a pilot for occupancy data collection, the authors do go on to draw conclusions about community-based forest management vs. protected areas. If the river between the two study areas poses a barrier to movement, then the two areas can likely be treated independently. However, if it does not, an explanation of how the authors can be sure that the differences they are finding in the models (or lack thereof) are due to how the two areas are managed and not due to other factors, particularly given the limited co-variates that were included in the model.
• Relatedly, grasslands featured prominently in the framing of the study and site importance, and this land cover is clearly limited from Fig. 1. Therefore, I was surprised that the models did not include a covariate to capture the proportion of grassland within some distance of each grid cell (with the distance based on typical dispersal distances of the animals in question). From the map of the two study areas, it is clear that edge grid cells have a different surrounding matrix than core grid cells.
• Additional details on how the flights were conducted are needed, including how the pilots/observers knew which grid cell to assign an individual they were observing at any given time since identification was done in the field. For the post-flight image analysis, were the images stitched into an orthomosaic? If so, how was that carried out and how did the authors ensure that detected animals were assigned to the appropriate grid? The sampling design should include sufficient details that someone else could replicate the same data collection protocol if desired.
• Were all grid cells ‘flown’ at some point or were there grid cells that were never ‘flown’? If the latter, how were the zero/NA grid cells handled in the occupancy model? Figure 2 does not actually show the flown and not flown grid cells as is suggested in the text where the table is referenced. It would be helpful if this context was added to the figure.
• According to the description in the text, there were 12 hours of AM flight time and only 5 total hours of PM flight time. It was not clear from the text whether this difference created more ‘flown’ grid cells during the AM hours and thus potentially biased the findings, or whether an equal number of grid cells were ‘flown’ during each day. Please clarify these details in the text.

Equation 2: I did not see where y is defined in the text

Figure 1: It would be helpful for readers to see a key map of where in the world this area is located. It would also be helpful to show the actual flight areas on the lower map so the reader has a sense of where they are located and how much area they cover. The legend text is not legible.

Table 1 is difficult to follow with the text center justified. I recommend left justification of the text or creating a more visual depiction as a figure.

Appendix: Several of the maps in the Appendix are missing legends to interpret the color schemes. Most of the maps would also benefit from an improved numbering/lettering system.

Author Response

We sincerely appreciate the reviewer’s thoughtful and detailed feedback, please see the attached word doc. 

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

I really appreciated the effort made by the authors to revise the manuscript which has certainly improved a lot. Unfortunately, however, the main problem has not been solved. That is, the sampling scale is too small for the target species of the study. I was convinced that this was a mistake and that the 150 square meter cells were actually 150 ha. If this had been the case, the two areas covered by the sampling would have been 22650 and 21600 ha respectively, sufficient surfaces to formulate habitat occupancy models and to evaluate the habitat selection of large carnivore and large herbivore species. Unfortunately, this is not the case and, incredibly, the work was based on a sampling of two areas of 2.26 and 2.16 ha, surfaces that could be useful for small rodents but certainly not for large carnivores and large wild ungulates. I regret having to say that the work was poorly set up from the point of view of the sampling scale and that thus nothing can be inferred about the habitat occupancy or even about the selection.

Author Response

We sincerely appreciate the reviewer’s thoughtful and detailed feedback, as well as the recognition of our revisions. We fully acknowledge the concern regarding the spatial scale of the sampling design and its relevance to interpreting large carnivore and ungulate occupancy.

To clarify, this study was designed as a pilot to demonstrate a methodology that could be scaled to broader spatial extents. We have revised the manuscript to more clearly communicate that the objective was to evaluate how UAVs can be used to detect fine-scale patterns of site use and vegetation structure within distinct land management zones.

We agree that the spatial extent of our two sampling zones (approximately 226 ha and 216 ha, respectively) is not sufficient to infer full home range occupancy for wide-ranging species. Multiple replicates across larger areas would be required for that purpose. However, our analysis was not intended to make landscape-scale inferences, but rather to serve as a proof of concept for evaluating relative site use and habitat structure using UAV-based methods. We further discuss the opportunities and limitations of UAVs in terms of detectability, and highlight outstanding questions related to integrating UAVs into wildlife monitoring workflows.

Additionally, we outline the potential for incorporating finer-scale covariates and offer guidance for researchers aiming to refine or expand upon this approach. While our results are not conclusive at broader scales, we believe they provide a useful foundation for others designing studies that compare land management zones or apply fine-scale habitat covariates.

In response to this concern, we have revised the manuscript to clarify our study’s objectives and explicitly state that we are not modeling species occupancy at the landscape or home range level. We hope these revisions address the reviewer’s concern and better situate the study within its intended scope.