Predicting Microhabitat Suitability for an Endangered Small Mammal Using Sentinel-2 Data

Round 1

Reviewer 1 Report

The study describes an attempt for mapping the microhabitat suitability for an endangered small mammal at the south of Portugal. Firstly, the authors extracted biophysical and spectral information at fine-scale by using Sentinel-2 data. Authors also used topography and structural landscape information as predictors. They then calibrated a habitat suitability model through by using Random Forest as modelling algorithm, and tested the contribution/importance of each predictive variable to the model performance and identified if satellite-derived variables best explain the species distribution at fine spatial scale.

Overall, the paper is well written and touches a topical subject in the context of remote sensing applied to biodiversity conservation. However, there are quite few issues to be addressed before the manuscript can be suitable for publication. Following are my comments and suggestions.

 

General comments

The introduction is solid, well balanced in length, and clearly introduces the main points the authors are interested in discussing. However, I have doubts about whether it is really the first time, as authors stated, that such an analysis is carried out, since there is a quite extensive body of literature (perhaps more focused on plants) on the application of satellite remote sensing data at fine-scale for habitat modelling. Other thing is the application of these techniques to ecological niches of small mammals, which is of great interest to the scientific community considering the threats to which these species are subjected. Methods and Results sections are in general well described, but please see specific comments. In general, the workflow was well described, but a more self-criticism in terms of replicability of this study is needed. Despite its promising results, the authors have not faced its methodology against other approaches, and neither considering or testing other methods or algorithms of habitat modelling that consider presence/absence data. Overall, current findings in other studies illustrate that higher mapping accuracies are achieved with images possessing high spectral and spatial resolutions, thus capturing information across the visible and reflected infrared solar spectrum. Despite the authors something stated about it, it is not enough and this should be more expanded in the Discussion section since understanding the trade-offs in spectral and spatial resolution can assist land managers in deciding the most appropriate imagery or technique with respect to habitat dynamic. In other words, among other, what are the next steps in its framework and their recommendations to improve the performance of their approach and how they would address it. For instance, it would be interesting to explore if individual models calibrated only with satellite products, topography and landscape variables would report similar (-or lower) performance than the results presented in this study.

 

Specific comments

Line 110. Please check: Regos et al., 2019. Line 113. Please check: Arenas-Castro et al., 2018. Line 148. Based on the Authors´s guidelines, I think the caption should be located below the figure. Please check it out. Line 175. “Évora 2018”. Please check it out. Lines 258-268. This paragraph was repeated. Line 295. “… and associated mean (between sensitivity and specificity) …” Could be missed in Table 3? Line 310. “… that the Cabrera vole habitat is more suitable in areas close to roads …” Despite it is interesting, perhaps the authors should consider rewrite this statement based on the comments in the Discussion section. Line 313. Figure 2 (1 -column fitting image); Figure 3 (1.5 -column fitting image); and so on. This info could be misallocated? Line 328. Please provide references. Line 412. Please check: Arenas-Castro et al., 2019; Andreo et al., 2019. Line 417. “… across time …” Consider to rewrite as Sentinel 2 has not a temporary resolution yet to state this comment. Lines 419-421. “… time that super-resolved and freely available satellite remote sensing data may provide an important tool for fine scale conservation planning and population monitoring, …” Please consider this statement and support with a deeper search on related bibliography. Figure 3. This legend is not complete, or perhaps only the color ramp. Please review. Figure 4. Please increase the axes font and variable titles. Figure 5. Presences are not easily recognised. Please change the color or increase dot size.

 

Additional references provided in this review.

1.- Regos A, Vidal M, Lorenzo M & Domínguez J. 2019. Integrating intraseasonal grassland dynamics in cross-scale distribution modeling to support waterbird recovery plans. Conservation Biology. in press. DOI:10.1111/cobi.13415.

2.- -        Arenas-Castro S, Gonçalves J, Alves P, Alcaraz-Segura D, Honrado JP. 2018. Assessing the multi-scale predictive ability of ecosystem functional attributes for species distribution modelling. PLoS ONE 13(6): e0199292. https://doi.org/10.1371/journal.pone.0199292.

3.- -        Arenas-Castro, S.; Regos, A.; Gonçalves, J.F.; Alcaraz-Segura, D.; Honrado, J. 2019. Remotely Sensed Variables of Ecosystem Functioning Support Robust Predictions of Abundance Patterns for Rare Species. Remote Sens. 11, 2086.

4.- Andreo, V.; Belgiu, M; Brito Hoyos, D.; Osei, F.; Provensal, C.; Stein, A. 2019. Rodents and satellites: Predicting mice abundance and distribution with Sentinel-2 data. Ecological Informatics, 51:157-167. https://doi.org/10.1016/j.ecoinf.2019.03.001.

Author Response

Response to Reviewer 1 Comments

 

Comments and Suggestions for Authors

The study describes an attempt for mapping the microhabitat suitability for an endangered small mammal at the south of Portugal. Firstly, the authors extracted biophysical and spectral information at fine-scale by using Sentinel-2 data. Authors also used topography and structural landscape information as predictors. They then calibrated a habitat suitability model through by using Random Forest as modelling algorithm, and tested the contribution/importance of each predictive variable to the model performance and identified if satellite-derived variables best explain the species distribution at fine spatial scale.

Overall, the paper is well written and touches a topical subject in the context of remote sensing applied to biodiversity conservation. However, there are quite few issues to be addressed before the manuscript can be suitable for publication. Following are my comments and suggestions.

 

General comments

Point 1: The introduction is solid, well balanced in length, and clearly introduces the main points the authors are interested in discussing. However, I have doubts about whether it is really the first time, as authors stated, that such an analysis is carried out, since there is a quite extensive body of literature (perhaps more focused on plants) on the application of satellite remote sensing data at fine-scale for habitat modelling.

Response 1: Thank you for your good point. We agree that the idea of incorporating fine-scale remote sensing data into SDMs is not new. However, applications to animal species are still relatively scarce in the literature, and virtually absent when considering the utilization of very highly detailed remote sensing data from open-access satellites, as it is the case of image enhancement of Sentinel 2 images. As mentioned in the Introduction (lines 84-94), we recognise that remote sensing data has been increasingly used in SDMs, though, to our knowledge, there is a lack of examples showing the utility of Sentinel 2 enhanced imagery in wildlife conservation problems. In order to clarify our point and to better support our claim, we have modified the sentence at line 127-133.

 

Point 2: Other thing is the application of these techniques to ecological niches of small mammals, which is of great interest to the scientific community considering the threats to which these species are subjected.

Response 2: We agree with the reviewer and have added a reference to the work of Vaniscotte et al. 2009 on the use of remote sensing for ecological modeling (SDMs) applied to small mammals of conservation concern  (line 106).

 

Point 3: Methods and Results sections are in general well described, but please see specific comments. In general, the workflow was well described, but a more self-criticism in terms of replicability of this study is needed. Despite its promising results, the authors have not faced its methodology against other approaches, and neither considering or testing other methods or algorithms of habitat modelling that consider presence/absence data.

Response 3: This is an interesting point which we think that could be the subject of further study in a separate paper. Although we fully agree that replicability of the study could be better supported by the selection of more than one algorithm, though this was beyond the scope of our paper. We aimed to introduce Boruta, a flexible algorithm that has been demonstrated to be of special interest in the field of variable selection, in particular when having high dimensional data (e.g. >50 variables; Speiser et al. 2019), as our case. We did not perform multivariate analysis with ensemble modeling (more algorithms) as we were concerned that this was not coherent with the presented framework, being Boruta functioning as a single algorithm. We agreed that what proposed from the reviewer could be an interesting field of research in species distribution models and remote sensing. Therefore, according to the reviewer suggestion, we added in the Discussion section more comments specifically acknowledging the idea that further studies should explicitly test the use of alternative algorithms. Specifically we added the following sentences at lines 466-471.

 

Point 4: Overall, current findings in other studies illustrate that higher mapping accuracies are achieved with images possessing high spectral and spatial resolutions, thus capturing information across the visible and reflected infrared solar spectrum. Despite the authors something stated about it, it is not enough and this should be more expanded in the Discussion section since understanding the trade-offs in spectral and spatial resolution can assist land managers in deciding the most appropriate imagery or technique with respect to habitat dynamic.

Response 4: We agreed with the suggestion of the reviewer and we have expanded our section to include more insights on this important aspect (see line 457-466).

 

Point 5: In other words, among other, what are the next steps in its framework and their recommendations to improve the performance of their approach and how they would address it. For instance, it would be interesting to explore if individual models calibrated only with satellite products, topography and landscape variables would report similar (-or lower) performance than the results presented in this study.

Response 5: We agree with this comment and added this as further recommendation for future studies in the Discussion section (lines 471-474) (see response to comment 3).

Specific comments:

Line 110. Please check: Regos et al., 2019. / Please see comment 1)

Line 113. Please check: Arenas-Castro et al., 2018 / Thank you for the reference. We have added such reference at line 134.

Line 148. Based on the Authors’ guidelines, I think the caption should be located below the figure. Please check it out. / We have located the caption below Figure 1 according to the reviewer suggestion.

Line 175. “Évora 2018”. Please check it out. / We have replaced “[Évora 2018; 41]” to [“IPMA Évora 2018; 42] at line 206-207”

Lines 258-268. This paragraph was repeated / This paragraph is now deleted.

Line 295. “… and associated mean (between sensitivity and specificity) …” Could be missed in Table 3? / Yes, thank you. According to reviewer 3, we moved all the results of table 3 within the main text, and the new text is found at line 334-350.

Line 310. “… that the Cabrera vole habitat is more suitable in areas close to roads …” Despite it is interesting, perhaps the authors should consider rewrite this statement based on the comments in the Discussion section. / We have modified the sentence (line 347-349).

Line 313. Figure 2 (1 -column fitting image); Figure 3 (1.5 -column fitting image); and so on. This info could be misallocated? / We have removed such indications from the main text according to the reviewer suggestion.

Line 328. Please provide references. / We provided a suitable exemplificative reference with the proposed sentence according to the reviewer suggestion (Carrié et al. 2018) at line 375.

Line 412. Please check: Arenas-Castro et al., 2019; Andreo et al., 2019. / We  thank the reviewer for  the proposed of reference of Arenas-Castro et al., 2019, which was added to the main text (line 485). We have checked Andrea et al. (2019) and, though it seems a suitable reference, we argue that the purpose of such article is to identify habitat of species with the aim of lower the risk of vector-borne diseases. We have introduced Andrea et al. (2019) previously when introducing the role of fine-scale remote sensing for ecological modelling in general (line 127), prior to define the purpouse (conservation) of the predictive model.

Line 417. “… across time …” Consider to rewrite as Sentinel 2 has not a temporary resolution yet to state this comment. We have replaced these words at line 376.

Lines 419-421. “… time that super-resolved and freely available satellite remote sensing data may provide an important tool for fine scale conservation planning and population monitoring, …” Please consider this statement and support with a deeper search on related bibliography. / We reformulated the entire sentence (line 493-498).

Figure 3. This legend is not complete, or perhaps only the color ramp. Please review. / We revised the color ramp according to the reviewer suggestion. We clarified the legend in the text (now figure 2; line 351-356).

Figure 4. Please increase the axes font and variable titles. / We have increased the x and y axes fonts, as well the variable titles of the figure (now figure 3; line 356-357).

Figure 5. Presences are not easily recognised. Please change the color or increase dot size. / Checked. We have revised the figure and we increased the dot size; but in a way that the reader still perceives that presences overlapped high suitability areas. Moreover, we reformulated the figure caption (line 364-370).

 

Additional references suggested in this review.

1.- Regos A, Vidal M, Lorenzo M & Domínguez J. 2019. Integrating intraseasonal grassland dynamics in cross-scale distribution modeling to support waterbird recovery plans. Conservation Biology. in press. DOI:10.1111/cobi.13415.

2.- -        Arenas-Castro S, Gonçalves J, Alves P, Alcaraz-Segura D, Honrado JP. 2018. Assessing the multi-scale predictive ability of ecosystem functional attributes for species distribution modelling. PLoS ONE 13(6): e0199292. https://doi.org/10.1371/journal.pone.0199292.

3.- -        Arenas-Castro, S.; Regos, A.; Gonçalves, J.F.; Alcaraz-Segura, D.; Honrado, J. 2019. Remotely Sensed Variables of Ecosystem Functioning Support Robust Predictions of Abundance Patterns for Rare Species. Remote Sens. 11, 2086.

4.- Andreo, V.; Belgiu, M; Brito Hoyos, D.; Osei, F.; Provensal, C.; Stein, A. 2019. Rodents and satellites: Predicting mice abundance and distribution with Sentinel-2 data. Ecological Informatics, 51:157-167. https://doi.org/10.1016/j.ecoinf.2019.03.001.

 

Other additional references:

Carrié, R., Lopes, M., Ouin, A., & Andrieu, E. (2018). Bee diversity in crop fields is influenced by remotely-sensed nesting resources in surrounding permanent grasslands. Ecological Indicators, 90, 606-614.

Pereira, O. J. R., Melfi, A. J., & Montes, C. R. (2017). Image fusion of Sentinel-2 and CBERS-4 satellites for mapping soil cover in the Wetlands of Pantanal. International Journal of Image and Data Fusion, 8(2), 148-172.

Rapinel, S., Mony, C., Lecoq, L., Clement, B., Thomas, A., & Hubert-Moy, L. (2019). Evaluation of Sentinel-2 time-series for mapping floodplain grassland plant communities. Remote sensing of environment, 223, 115-129.

Speiser, J. L., Miller, M. E., Tooze, J., & Ip, E. (2019). A comparison of random forest variable selection methods for classification prediction modeling. Expert Systems with Applications. doi: https://doi.org/10.1016/j.eswa.2019.05.028

Thornton, M. W., Atkinson, P. M., & Holland, D. A. (2007). A linearised pixel-swapping method for mapping rural linear land cover features from fine spatial resolution remotely sensed imagery. Computers & Geosciences, 33(10), 1261-1272.

 

We thank Reviewer 1 which helped us strengthen the manuscript to its current form.

Reviewer 2 Report

Dear Authors,

You can find attached my review (with comment and suggestion in the PDF file). 

The manuscript, relating to the "Predicting microhabitat suitability for an endangered 2 small mammal using Sentinel-2 data" must be improved.

In general the manuscript is well written and pleasant to read.

Best Wishes

Comments for author File: Comments.pdf

Author Response

Response to Reviewer 2 Comments

 

Reviewer 2

Comments and Suggestions for Authors

Dear Authors,

You can find attached my review (with comment and suggestion in the PDF file). 

The manuscript, relating to the "Predicting microhabitat suitability for an endangered 2 small mammal using Sentinel-2 data" must be improved.

In general the manuscript is well written and pleasant to read.

Best Wishes

Specific comments:

Line 57: “2 apex please”. Checked (now at line 65; also at line 112).

 

Line 65: “explain the acronym aka please, also references”. We have deleted that word as was correct by reviewer 3.

Line 84: “System my be better (e.g. https://eospso.nasa.gov/content/nasas-earth-observing-system-project-science-office)”. We respectfully disagree with the referee. We preferred Earth observation Satellites EOS because we believe is the concept normally used (http://faculty.wwu.edu/wallin/envr442/pdf_files/tatem_etal_Amer_Sci_RSreview_2008.pdf), but if the referee insists we are ready to comply such a request.

 

Line 128:  “please add references of Boruta RF”. Checked (now at line 150).

 

Line 135:  “a.s.l. above sea level”. Checked (now at line 158).

 

Line 140:  “please in italics, the same for other species names, also Cabrera vole see Lines 51”. Checked (now at line 164,165,24).

 

Line 155:  “please specify how you have selected the survey plots. Expert knowledge or stratify random selection  based on niche ecology of species? Is to generic in method chapter how you have explained it..."Cabrera vole surveys were conducted by initially identifying in the field suitable and unsuitable habitat patches for the Cabrera vole "”. Checked (now at line 182-183).

 

Line 166:  “please add information of the technical specifications of the GPS (accuracy lon lat and altitude)”. Checked (now at line 194-195).

 

Line 246-256:  “repeated see paragraph below”. Checked (we have deleted the sentences).

 

Line 330:  “spectral images analysis my be better, please specify "Sentinel-2", is to generic. Apply the same suggestion in all the paper please.”. We have introduced more information at line 377, also specified in the methodology section (line 224).We respectfully believe this is sufficient for the designation of Sentinel-2 data.

 

We appreciated and accepted all the suggestions made by the reviewer, which we believe have contributed to strengthen the manuscript.

Reviewer 3 Report

This paper uses Sentinel-2 sensor-derived and other habitat variables to predict the distribution of the Cabrera vole (Microtus cabrerae). Overall the paper is well written and the methodology adequately described. There are a few minor issues which I have added to the attached pdf. More major suggestions are:

The title says endangered, but as the abstract and the paper states the vole is near threatened. Either remove endangered or replace with a more suitable description. 

Although the introduction provides sufficient background, it seems a bit long. The second to last paragraph is especially long and can be at least cut into 3 paragraphs as stated in the attached pdf. 

Table 3 and Figure 2 seem unnecessary and makes more sense written out in the results.

All the figure legends are on top of of the figures, they should be under the figures.

The discussion addresses the findings with relevant citations. However, discussions don't typically have figures and it makes more sense to move Figure 5 to results.  

Comments for author File: Comments.pdf

Author Response

Response to Reviewer 3 Comments

Reviewer 3

Comments and Suggestions for Authors

This paper uses Sentinel-2 sensor-derived and other habitat variables to predict the distribution of the Cabrera vole (Microtus cabrerae). Overall the paper is well written and the methodology adequately described. There are a few minor issues which I have added to the attached pdf. More major suggestions are:

Point 1: The title says endangered, but as the abstract and the paper states the vole is near threatened. Either remove endangered or replace with a more suitable description. 

Response 1: We thank the reviewer comment. In fact, the species is classified as Near threatened by IUCN, but in Portugal and Spain it is classified as Vulnerable , thus included in a threatened category. However, to avoid confusion we placed the IUCN category in second place in the text when appeared the first time (highlighting the denomination in Portugal; Lines 59-60), while we avoided the IUCN denomination at line 328 and 381).

Point 2: Although the introduction provides sufficient background, it seems a bit long. The second to last paragraph is especially long and can be at least cut into 3 paragraphs as stated in the attached pdf. 

Response 2: We agree and are thankful for the reviewer for the improvement of the readability of the manuscript. We have incorporated such suggests.

Point 3: Table 3 and Figure 2 seem unnecessary and makes more sense written out in the results.

Response 3: We incorporated these results within the main text according to the reviewer suggestion (lines 334-350).

Point 4: All the figure legends are on top of the figures, they should be under the figures.

Response 4: We moved figure legends under the figures according to the reviewer suggestion.

Point 5: The discussion addresses the findings with relevant citations. However, discussions don't typically have figures and it makes more sense to move Figure 5 to results.  

Response 5: We moved Figure 5 to the results according to the reviewer suggestion. Moreover, by moving figure 5 (now figure 4 due to the elimination of figure 2) in the result section, we added the following sentence “The habitat suitability map shows that the occurrence fall in high-probability areas (Figure 4).” at line (349-350).

We thank Reviewer 3 which helped us strengthen the manuscript to its current form.