An Empirical Bayesian Approach to Quantify Multi-Scale Spatial Structural Diversity in Remote Sensing Data

Round 1

Reviewer 1 Report

I find the manuscript is written in a way not very accessible for readers/researchers in remote sensing or statistics. It would be very hard for readers to understand the motivation, data, and methods in the manuscript unless they have a strong background in ecology, in particular landscape ecology. I have to admit that I myself had difficulty understanding the manuscript, though it seems that the work is trying to use a statistical model to understand the landscape structure. 

Author Response

Dear Reviewer,

Thank you for your helpful feedback!

We rewrote large parts of the manuscript to better embed our research in an ecological context, especially to reason the importance of landscape structure for biome changes in the northern high latitudes. We paid attention to motivate the choice of study region, of data, and of methods. We also added more explanation about the usefulness of empirical Bayesian methods and the used approach to quantify spatial structural diversity.

To improve our research design, we added two new analyses targeting pixel resolution and extent. We quantified structural diversity in coarser resolution data, and we chose 3 aggregation factors (2, 6 and 10) to demonstrate the loss of information about landscape heterogeneity, as pixel size increases. We also quantified structural diversity in the whole of northern Eurasia, using 1 km² resolution Modis NDVI data, and we cropped the study region from this data to demonstrate that the detection of structural diversity features is not compromised.

We hope the manuscript is now better accessible for readers and researchers in remote sensing or statistics.

All major changes are marked in blue color in the manuscript.

Thank you very much for your time and effort helping us to improve this manuscript!

Kind regards,

Leila Schuh, on behalf of all co-authors

Author Response File: Author Response.pdf

Reviewer 2 Report

Although I think this manuscript is really soundly constructed, well structured, and overall also well written, hereafter I try to briefly sum up some conceptual points that actually can limit the application of this detailed work (for some minor detailed comments you can look at the attached file).

1) the pixel dimension for your analysis is 1km... now if this work wants to intercept the real need for environmental planning the parametrization of this approach should use the most commonly diffused Copernicus Sentinel data (NDVI at 10 meters of ground resolution). It's not only a technical issue: the problem is that, especially  where environmental policies are needed (transition zones between rural/natural and urban) the subarticulation of land uses and ecosystems are much more mosaicked and scattered across some limited space. I wonder if this approach can guarantee the same positive results even using a finer scale assessment.

2) I don't have any chance to comment on the statistical application which sounds really robust, but I am wondering why you didn't link more the results with your study area... the discussion misses any relationship with the real transition of ecosystems in the area of investigation and at the end, the research is somehow missing of any practical test... normally this methodologies can be practically used to define the landscape structure, of map the vulnerabilities, or define core, edge and link areas in a catchment to design ecological networks... i guess you should think if it's better to emphasize in the discussion a tangible application of your method to create a supporting decision-making system for environmental planning.

Finally, but that's really minor, pay attention to repetitions in the text... there are many.

Good luck

Comments for author File: Comments.pdf

Author Response

Dear Reviewer,

Thank you for your kind and very helpful feedback! We addressed the comments in the text. Please find below our detailed response, following the structure of your feedback.

• We fully agree that finer resolution data is most important for landscape planning. We chose 1 km² resolution because, based on the literature, this resolution is adequate to assess large-scale biome changes in tundra taiga habitat, which is the geographical context of our study. If the methods are applied to data from a different sensor, differences in structural diversity maps may also be caused by sensor differences. To address the highly relevant points you raised, we added two new analyses targeting pixel resolution and situations where the study region comprises many hierarchical levels. We quantified structural diversity in coarser resolution data, and we chose 3 aggregation factors (2, 6 and 10) to demonstrate the loss of information about landscape heterogeneity, as pixel size increases. This design ensures that differences in structural diversity maps are not due to sensor differences, which would be an additional source of disagreement if sentinel data were used. However, since higher resolution data in our study region would mean that more hierarchical levels of landscape structure are present, we assessed our methods in such a context. We quantified structural diversity in the whole of northern Eurasia, using 1 km² resolution Modis NDVI data, and we cropped the study region from this data to demonstrate that the detection of structural diversity features is not compromised. We paid attention to better motivate our choice of resolution and we added to the discussion of pixel size to make the relationship between resolution and the methods we present clearer. We hope our additions are considered adequate, and we look forward to receive feedback on whether we were able to convince.

• We rewrote large parts of the manuscript, particularly of the introduction and of the discussion, to better embed our research in an ecological context, and particularly to emphasize the importance of landscape structure for biome changes in the northern high latitudes. We paid attention to motivate the choice of study region, of NDVI data, and of 1 km² resolution for the study of multi-scale heterogeneity in tundra taiga habitat.

We also added more explanation about empirical Bayesian methods, about the used approach to quantify spatial structure, and about how this is embedded in the entropy metric.

• We reduced repetitions in the text. Due to the topic of this manuscript, the term “scale” appears quite often, and since we introduce a new method the manuscript is quite technical. We believe that for the sake of clarity, keeping the term “scale” if sometimes better that removing it. We hope our revisions are acceptable

All major changes are marked in blue color in the manuscript.

Thank you very much for your time and effort helping us to improve this manuscript!

Kind regards,

Leila Schuh, on behalf of all co-authors

Author Response File: Author Response.pdf

Reviewer 3 Report

Overall comment

This study explores how scale affects the landscape patterns using remote sensing data. It is generally an interesting topic. Please find below for detailed comments for the authors to consider.

Detailed comments:

1.     Please highlight the specific gap this study is trying to address.

2.     Please specify how the approach proposed in this study can be applied to other similar empirical studies?

3.     Line 205. This study employs a specific metric (structural diversity entropy) to test the hypothesis. Would the conclusions be applied to other similar metrics?

4.     Line 235. Please provide more detailed reasons for why using the specified three ways to combine information from different.

Author Response

Dear Reviewer,

Thank you for your helpful feedback! We rewrote large parts of the manuscript to better embed our research in an ecological context, and especially to emphasize the importance of landscape structure for biome changes in the northern high latitudes. We paid attention to motivate the choice of study region, of NDVI data, and of 1 km² resolution.

We also added more explanation about empirical Bayesian methods, about the used approach to quantify spatial structure, and about how this is embedded in the entropy metric. Please find below our detailed response, following the structure of your feedback.

1. Rewriting the introduction, we paid attention to make a clearer argument about which gap we are trying to close.

2. We included a paragraph at the end of the discussion to point the reader to the roadmap on how to specify these methods to use them in other applications. Following another reviewers’ request, we moved this roadmap to the appendix. We also included a more detailed discussion of pixel resolution, and we added two new analyses targeting pixel resolution and situations where more hierarchical levels are present in the landscape. We hope with the discussion of these additions, the requirements for successfully applying these methods are clearer.

3. We added a paragraph about other second-order texture metrics, which we have previously tested and which can detect similar features.

4. We added a paragraph in the indicated methods section and a sentence in the background to make clear why we used these three nesting schemes.

All major changes are marked in blue color in the manuscript.

Thank you very much for your time and effort helping us to improve this manuscript!

Kind regards,

Leila Schuh, on behalf of all co-authors

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

I think the authors well-addressed all the comments of the first review.

The manuscript can be suitable for publication.

Author Response

Dear Reviewer,

Thank you for your time and effort helping us to improve this manuscript! We paid attention to spelling and grammar of the text, which was checked by native English speakers.

Additionally, we undertook the following revisions, as requested by the Academic Reviewer:

We moved the whole Appendix to supplementary Materials. We re-numbered the content accordingly and adjusted references in the main text. We also removed several references to figures in the Supplementary Materials, and instead wrote “(for example Figure xyz)” in the main text.

We structured all sections of the text into paragraphs that cover one idea and start with a topic sentence. We also removed repetitions to construct shorter sentences, and we worked on all parts of the text to improve overall readability. We added no new content, but we marked in blue color where we changed the structure of the text more substantially. In one marked section in the Discussion, we used the term ‘scale of variation’, which was previously not used, because we believe it adds clarity. For this reason, we also added another header in the Discussion. In the Results, we moved the Section ‘Multi-scale structural diversity in simulated data’ as the next section after ‘Multi-scale structural diversity features in NDVI data’, because we find this to follow a better logical order.

We hope the manuscript is now better readable and will be useful to the remote sensing community.

Kind regards,

Leila Schuh, on behalf of all co-authors

Author Response File: Author Response.pdf