High-Resolution Bioclimatic Surfaces for Southern Peru: An Approach to Climate Reality for Biological Conservation

Round 1

Reviewer 1 Report (Previous Reviewer 3)

The reviewer would like to thank the authors for revising the manuscript.

Introduction 

The authors are requested to elaborate more on the application of bioclamatic models with regards to natural disasters and cite the following article that reported a major disaster over the Himalayas. The event largely impacted the local human livelihoods as well as manmade infrastructure. How can such modeling help in pre and post-disaster management?

-Shugar et al, A massive rock and ice avalanche caused the 2021 disaster at Chamoli, Indian Himalaya, Science, 2021.

Cook et al., Science 374, 87–92 (2021)
Detection and potential early warning of catastrophic flow events with regional seismic networks

Li et al., Nature Geoscience, VOL 15, July 2022
High Mountain Asia hydropower systems threatened by climate-driven landscape instability

Yao et al., Nature Reviews Earth & Environment  3, 618–632 (2022)
The imbalance of the Asian water tower

Author Response

Dear reviewer, thank you for your constructive comments. We have responded to your recommendations below.

Point 1: The authors are requested to elaborate more on the application of bioclimatic models with regards to natural disasters and cite the following article that reported a major disaster over the Himalayas. The event largely impacted the local human livelihoods as well as manmade infrastructure. How can such modeling help in pre and post-disaster management?

Response 1: It included an introduction about natural disasters and how they can be managed with climate surfaces. In fact, it is very feasible to use them to prevent or foresee plans for a future event. Our surfaces, containing high resolution monthly precipitation information, can be used precisely to model (together with other variables such as slope, orientation, altitude and some others) the places with high risk of landslides. On the other hand, our surfaces (which cover the present), together with other models (which cover the future scenario), could also be used to locate high-risk zones with climate change scenarios, which is also beneficial to avoid this type of problems in the future, and even more so in mountainous areas, such as the Andes.

Finally, we are grateful for the suggestions of the manuscripts, which were included in our article.

Reviewer 2 Report (Previous Reviewer 1)

Dear Authors,

Your paper aims to produce new climatic and bio-climatic data layers for southern Peru, which are expected to be more precise, and more realistic than the existing ones. Thus, these new layers can be more applicable in conservation and distribution mapping purposes in the future.

The reviewed manuscript improved a lot. However, it is not clearly defined at the end of the chapter of Introduction what is the explicit objective of your paper. Otherwise, the manuscript follows the expected general structure of research papers and the Introduction provides sufficient overview of the literature. I did not find a very relevant reference with close connection to the core of the applied method I found considerable in the case of this article (10.1080/01431160010007033).

The materials and method section contains a satisfactory description of applied methods. It would be beneficial to provide a reference in order to underpin the applied equation obtaining missing precipitation data from remotely sensed NDVI values (Not only in section 4.2). I feel it necessary as the equation express a direct functional relationship. Some comments may be helpful regarding the applicability conditions for such an equation.

The results section contains a detailed description of modelling outputs. This section is proportional, and informative.

The discussion section is bit long in my opinion as it contains not only the discussion of authors' results in the spotlight of related studies but it turns back to the characteristics of the region's climatic regime. Otherwise the newly added parts of discussion are very welcome.

The section of Conclusions is renewed and relevant.

Overall opinion: The paper provides new climatic models for a region of southern Peru using innovative methods providing a better approximation of the climatic conditions of the given region than the earlier approaches did. It is suitable to serve as basic data provider for later biological studies in he region. Thus, he novelty of the work is justified. I recommend to publish his paper after minor improvements I indicated.

Specific notes:

Line 253 - per year.

Line 259 - Thirty-six thousand?

The two paragraphs at the beginning of section 3.2 are almost identical between lines 253-265.

Table 3 - It is not referred in the text, or I did not find it.

Kind regards

Author Response

Dear reviewer, thank you for your constructive comments. We have responded to your recommendations below.

Point 1: The reviewed manuscript improved a lot. However, it is not clearly defined at the end of the chapter of Introduction what is the explicit objective of your paper.

Response 1: We have better specified our objective. This is to make it clear what we intend to do with our study.

Point 2: I did not find a very relevant reference with close connection to the core of the applied method I found considerable in the case of this article (10.1080/01431160010007033).

Response 2: Thank you for the article recommendation. This has been included in our manuscript and gives more support for the relationship between NDVI and precipitation.

Point 3: The reviewed manuscript improved a lot. However, it is not clearly defined at the end of the chapter of Introduction what is the explicit objective of your paper.

Response 3: We define in a better way the objective of the study

Point 4: It would be beneficial to provide a reference in order to underpin the applied equation obtaining missing precipitation data from remotely sensed NDVI values (Not only in section 4.2). I feel it necessary as the equation express a direct functional relationship. Some comments may be helpful regarding the applicability conditions for such an equation.

Response 4: Some bibliographic references that help to demonstrate the correlation between NDVI and precipitation were indicated in methods. On the other hand, some comments were made indicating that this application was only exclusive for these small areas. Its application would only be for these fog-dependent ecosystems.

Point 5: The discussion section is bit long in my opinion as it contains not only the discussion of authors' results in the spotlight of related studies but it turns back to the characteristics of the region's climatic regime. Otherwise, the newly added parts of discussion are very welcome.

Response 5: We believe that it would be appropriate to explain the climatic patterns. This in the sense that readers (especially local readers) understand a little about the climatic characteristics that govern these areas.

Point 6: Specific notes

Response 6: All of them were corrected.

Author Response File: Author Response.docx

Reviewer 3 Report (New Reviewer)

The authors present an interesting study, developing local climate rasters for Southern Peru at high resolution from a disparate source of climate data. The results could be of interest to others, especially those who work with SDM. But there are some fundamental errors which will cause issues for any modelling, these need to be re-examined. Additionally, I think the authors have presented a draft version of the paper as there is much-highlighted text, repeated paragraphs and text in the incorrect sections.

The greatest issue is with the use of NDVI as a proxy for precipitation. They are correct that for normal ecosystems this can be a method to fill missing gaps. But it there are many short comings of these methods and these at least need to be discussed (see https://www.researchgate.net/publication/281320755_Predicting_Extreme_Droughts_in_Savannah_Africa_A_Comparison_of_Proxy_and_Direct_Measures_in_Detecting_Biomass_Fluctuations_Trends_and_Their_Causes). Three of the ecosystems that are identified are not driven by direct precipitation and modelling them with NDVI is highly problematic:

1. Lomas is a fog-driven ecosystem and whilst it could be perceived to be “precipitation” it will not directly translate into precipitation per unit area as would be seen with rainfall. This fog is unlikely to be recorded in traditional rainfall meters. The authors may be able to look at others who have at least measured the fog intercepted at certain periods. Additionally, there are only two (and I am not sure are truly in the Lomas) locations within this vegetation type, it is a great stretch to extrapolate using only two-point. Additionally, these stations are likely recording rainfall and not fog captured.
2. The valleys and outwash for valleys are not driven by direct rainfall, most of the rainfall falls in the Andes and flows to these locations, for which the vegetation then flushes. NDVI values in these areas will give false precipitation.
3. Another highly vegetation areas in the hyper-arid zone i.e. agriculture areas will be driven by pumped groundwater and not rainfall. Which again will give false precipitation.

All in all, of the above, will give a very false indication of precipitation for many regions. If this is used to drive SDM the results are likely to be additionally erroneous. Additionally, I would suggest that if you want to do SDM, then using a non-NDVI derived precipitation map (i.e., more traditional) and using NDVI as an additional predictor would give any modellers a better idea of the influence of each layer and model.

To compound the above the authors suggest that only global datasets exist, yet Fernandez-Palomino et al 2022 (https://journals.ametsoc.org/view/journals/hydr/23/3/JHM-D-20-0285.1.xml), have produced a high-resolution precipitation data set for Peru and Ecuador. As did Aybar 2020 (https://www.tandfonline.com/doi/full/10.1080/02626667.2019.1649411). They should at least compare their results to Fernandez-Palomino et al 2022. Why are your results better than Fernandez-Palomino et al 2022.

Validation of the models is highly problematic (and very circular). It is not at all certain what was used for validation (methods very muddled). I would expect at least some type of bootstrapping with the existing station data and error testing here. Additionally, many of the stations will have very limited time periods of data, authors do not give any indication of how limited and details of how they filled any gaps (or if they did), it would be highly erroneous to just treat this data as is (the authors mention homogenized data, but do no supply any detail for this).

Some more specific detailed comments are below, with reference to line numbers.

I have not commented in detail on the text, as some of the text is difficult to understand and I think some translation (from Spanish to English) issues have occurred. I suggest the paper is proofread in detail.

10 SDM do not give a greater understanding of evolutionary process, at least not without other information.

18 Spell out RMSEvs and MAD in the abstract when first used

70-71 “..topographic covariates..” this does not seem connected to the start of the sentence, I think this is methods or conclusion

104-105 fog should be mentioned here as it’s the biggest driver for native vegetation

120 the additional stations outside of the study area should be shown on fig 2

122-150 citation is needed for the NDVI methods, these methods also need an introduction.

165 details are need on the homogenization and autocompletion. Also details should be shown to indicate how much was real data and how much was extrapolated for each station.

200 “compare to [20]..” though out the paper, when referring to a statement to an other paper please spell out ie “compare to Vega et al [20]”

259 why 36000 surfaces? Best to say how you go to this figure (ie variables x months x ???)

270 what is Min and Max V? I assume variance? CI? Confidence interval? Spell these out when first used.

326-343 it seem these two paragraphs are repeated also the citation do not look correct in the first paragraph

345 It would be much better to compare the surfaces (see Fernandez-Palomino et al 2022) to see where the anomalies are and this discuss this.

416-441 a repeat of the prevision paragraph

480-482 of course the virtual station will have a agreement they are modelled, you can’t compare them. This is a non-sense statement.

624 PhD =  Dr.

Author Response

Dear reviewer, thank you for your constructive comments. We have responded to your recommendations below.

Point 1: The authors present an interesting study, developing local climate rasters for Southern Peru at high resolution from a disparate source of climate data. The results could be of interest to others, especially those who work with SDM. But there are some fundamental errors which will cause issues for any modelling, these need to be re-examined. Additionally, I think the authors have presented a draft version of the paper as there is much-highlighted text, repeated paragraphs and text in the incorrect sections.

Response 1: Disculpas por el texto resaltado, es que este fue realizado en razón de lo indicado por la revista científica. Por otro lado, verificamos el texto y los parrafos, y todos fueron corregidos.

Point 2: The greatest issue is with the use of NDVI as a proxy for precipitation. They are correct that for normal ecosystems this can be a method to fill missing gaps. But it there are many short comings of these methods and these at least need to be discussed (see https://www.researchgate.net/publication/281320755_Predicting_Extreme_Droughts_in_Savannah_Africa_A_Comparison_of_Proxy_and_Direct_Measures_in_Detecting_Biomass_Fluctuations_Trends_and_Their_Causes).

Response 2: We understand that NDVI can be somewhat difficult, but we believe that this depends on several factors. There are other articles where the relationship between precipitation and NDVI is evidenced (some are cited in the manuscript), on the other hand, in others (like the one given here as an example), a direct relationship is not observed. But analyzing the article given as an example, it is indicated that NDVI is initially related to rainfall periods (during and after), but that difficulties are observed for long periods of drought. So, two factors are involved here: first the type of vegetation and second the grazing. In this case we are analyzing plots with grass vegetation, and although we do not know their biology in Africa, if we think that they are similar to those found in the Andean areas, these are perennial plants, so their biomass remains dry after the rainy season, and they sprout again during the rains, drying up later (after the rainy season). Then, over a long period of drought (this does not mean that it does not rain even a mm), with the scarce rainfall the plants sprout again growing a certain size, and then dry up. This accumulation of vegetation will eventually interfere with NDVI readings (which we believe is what is actually happening) if successive droughts occur. On the other hand, grazing pressure further aggravates the above, as domestic or wild animals eat the grass, reducing the vigor of the plants. So, perhaps this is happening specifically for these types of plant communities, which may also be a pattern in the Andean grasslands (although we don't really know, many of the Andean grasslands are not under intensive grazing). On the other hand, in lomas ecosystems there are no perennial grassland communities (Poaceae), but rather annual herbaceous plants (diversity of botanical families).

Point 3: Three of the ecosystems that are identified are not driven by direct precipitation and modelling them with NDVI is highly problematic:

1. Lomas is a fog-driven ecosystem and whilst it could be perceived to be “precipitation” it will not directly translate into precipitation per unit area as would be seen with rainfall. This fog is unlikely to be recorded in traditional rainfall meters. The authors may be able to look at others who have at least measured the fog intercepted at certain periods. Additionally, there are only two (and I am not sure are truly in the Lomas) locations within this vegetation type, it is a great stretch to extrapolate using only two-point. Additionally, these stations are likely recording rainfall and not fog captured.
2. The valleys and outwash for valleys are not driven by direct rainfall, most of the rainfall falls in the Andes and flows to these locations, for which the vegetation then flushes. NDVI values in these areas will give false precipitation.
3. Another highly vegetation areas in the hyper-arid zone i.e. agriculture areas will be driven by pumped groundwater and not rainfall. Which again will give false precipitation.

Response 3: Aquí dejamos respueta a cada punto.

Regarding point 1

Although hilly ecosystems are also known as cloud ecosystems, this is due to the high frequency of low altitude stratocumulus clouds (between 600 and 1200 m asl). Our experience working in hilly ecosystems in southern Peru shows that there is precipitation in these ecosystems. Moreover, there is a misconception about the mist in the lomas, i.e. that it translates into moisture in the soil and that thanks to it the vegetation endures. The truth is that there can be years without lomas, and not precisely because there are no fogs, but because even though they enter the soil they do not generate vegetation, i.e. they do not precipitate in the form of rain. In our experience working several years already in lomas, and taking climatic data in these areas thanks to several projects, evaluating the vegetation, is that we came to these conclusions. Fogs are always present, but if they do not contain enough precipitation, there are simply no lomas (no vegetation develops), on the other hand, when it rains (the so-called drizzles or garuas) a rich vegetation develops. On the other hand, the use of fog by means of fog catchers is a technology that we have been working on in lomas ecosystems, so we are adapting to climate change.

As for the two stations used (Atiquipa and Morro Sama), we verified that they are located within the lomas formations.

Regarding point 1 and 2: Here we only use NDVI values to predict rainfall values strictly only in the lomas communities (this thanks to the latest ecosystem maps of Peru, elaborated by MINAM, as well as information from Moat et al., 2021). That is, we did not use an NDVI surface as a covariate to generate the temperature and precipitation surfaces in places other than the lomas communities. We did not use NDVI for valleys or irrigated areas within the desert tablazo, much less in Andean areas.

Point 4: To compound the above the authors suggest that only global datasets exist, yet Fernandez-Palomino et al 2022 (https://journals.ametsoc.org/view/journals/hydr/23/3/JHM-D-20-0285.1.xml), have produced a high-resolution precipitation data set for Peru and Ecuador. As did Aybar 2020 (https://www.tandfonline.com/doi/full/10.1080/02626667.2019.1649411). They should at least compare their results to Fernandez-Palomino et al 2022. Why are your results better than Fernandez-Palomino et al 2022.

Response 4: We include the suggested articles, however, we consider that a comparison would not be possible given the technical differences. Both Fernandes-Palomino et al. (2022) and Aybar et al. (2020), present only precipitation surfaces (they were not performed for temperature), on the other hand, both present lower resolutions to the surfaces proposed here. Fernandes-Palomino et al. (2022) and Aybar et al. (2020), present surfaces of 0.1° (~11 km2), while in our paper we present surfaces of 0.008333° (~0.9 km2) resolution. We believe that the comparison could lead to false interpretations.

On the other hand, we here want to highlight the differences in terms of the data, ideally by performing niche models and comparing the results of those models. Evaluations would have to be made in terms of the performance of the niche models, in that sense bioclimatic surfaces would have to be used. The bioclimatic surfaces are the result of mathematical processes between the monthly surfaces produced (minimum temperature, maximum temperature and presipitation), thus 19 bioclimatic surfaces are known. With the above, we believe that it is difficult to create the bioclimatic surfaces from the data of Fernandes-Palomino et al. (2022) and Aybar et al. (2020), in the sense that they do not present temperature. However, in the interim, we tested the modeling of some species and compared the results with other surfaces. Here is an example.

Here we did tests to model the ecological niche of various species using the MaxEnt algorithm, producing good results with our surfaces. For example, when performing ecological niche modeling with an endemic tree from Arequipa (Neltuma calderensis), WorldClim surfaces returned only areas within the department of Arequipa. However, when using our surfaces, the model indicated areas of high suitability in the department of Moquegua and Tacna. After a trip to Moquegua and conducting herbarium reviews, we were able to verify that indeed the areas where our model predicted high habitat suitability were also where the tree was found. We also did tests for species distributed in the coastal zone (lomas) and satisfactory results were also given for our surfaces. We believe that a comparison in terms of the performance of our surfaces with others carried out for the area requires an additional article. Showing the results obtained and the modeling of the ecological niche of species, would make this work very extensive.

Fig. Ecological niche model of N. calderensis made with WorldClim surfaces. The pink dots were those used for the model and the yellow dots are those collected on a recent trip. Note how the area of the yellow dots is not predicted as a suitable area.

Fig. Ecological niche of N. calderensis made with our surfaces. Note how the area where the yellow dots are predicted as suitable areas, despite not having been integrated into the niche model.

Point 5: Validation of the models is highly problematic (and very circular). It is not at all certain what was used for validation (methods very muddled). I would expect at least some type of bootstrapping with the existing station data and error testing here. Additionally, many of the stations will have very limited time periods of data, authors do not give any indication of how limited and details of how they filled any gaps (or if they did), it would be highly erroneous to just treat this data as is (the authors mention homogenized data, but do no supply any detail for this).

Response 5: As for the validation of the surfaces, this was carried out by means of a cross validation with 10 K-folds, i.e. 10 modelings were performed (extracting in each modeling 20% of the data used for the test) with 10 replications. That is why it is said that 3600 models were made, because for example for the maximum temperature, for each month 10 models were made by subtracting in each model different data sets corresponding to 20%, and each set was made with 10 replications. So 10 data sets, for 10 replications, for 12 months, is a total of 1200 surfaces created, that times 3 (maximum minimum temperature and precipitation), is 3600. From each of the monthly sets, the surface with the lowest residuals was selected. We chose to use cross-validation because in works where modeling is performed to generate bioclimatic surfaces (not only climatic), this process is used. On the other hand, it was decided to perform 10 folds with 80% of the data for training because of the small number for the study area. Finally, the surfaces were evaluated using the processes indicated in the article RMSEcv and MAD (where basically the model performance is evaluated using the test data on the surfaces).

Point 6: Some more specific detailed comments are below, with reference to line numbers.

10 SDM do not give a greater understanding of evolutionary process, at least not without other information.

18 Spell out RMSEvs and MAD in the abstract when first used

70-71 “..topographic covariates..” this does not seem connected to the start of the sentence, I think this is methods or conclusion

104-105 fog should be mentioned here as it’s the biggest driver for native vegetation

120 the additional stations outside of the study area should be shown on fig 2

122-150 citation is needed for the NDVI methods, these methods also need an introduction.

165 details are need on the homogenization and autocompletion. Also details should be shown to indicate how much was real data and how much was extrapolated for each station.

200 “compare to [20]..” though out the paper, when referring to a statement to an other paper please spell out ie “compare to Vega et al [20]”

259 why 36000 surfaces? Best to say how you go to this figure (ie variables x months x ???)

270 what is Min and Max V? I assume variance? CI? Confidence interval? Spell these out when first used.

326-343 it seem these two paragraphs are repeated also the citation do not look correct in the first paragraph

345 It would be much better to compare the surfaces (see Fernandez-Palomino et al 2022) to see where the anomalies are and this discuss this.

416-441 a repeat of the prevision paragraph

480-482 of course the virtual station will have a agreement they are modelled, you can’t compare them. This is a non-sense statement.

624 PhD =  Dr.

Response 6: All of them have been corrected, however, we leave some remarks about some observations.

• SDM do not give a greater understanding of evolutionary process, at least not without other information.
• Here we believe that we can use the term niche conservatism, which gives light or hypotheses about the evolution of species from a climatological-geographical point of view. We change the term distribution for ecological niche.

• fog should be mentioned here as it’s the biggest driver for native vegetation
• Here we refer to the full desert, where the mists no longer reach and have no influence.

• Citation is needed for the NDVI methods, these methods also need an introduction.
• It is included in the methods, on the other hand, since NDVI is not an objective in general, we believe that it is not necessary to mention it in the introduction. A good explanation is given in the discussion.

• Details are need on the homogenization and autocompletion. Also details should be shown to indicate how much was real data and how much was extrapolated for each station.
• It is included in methods on the homogenization performed. As for the autocompletion, it was not performed. Only the data presented at each station were used.

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report (New Reviewer)

Many thanks for your revised manuscript, changes made, and clarifications.

I have not been convinced of the argument that NDVI for Lomas equates to mm of precipitation. They are quite different entities and should be treated as such, plant species have evolved to capture fog in Lomas (ie, trichomes in many species), and if this moisture was delivered as precipitation, this adaptation would not be needed. For SDM's this can be problematic for specific fog-adapted species. I would suggest the author make this clear in the discussion and that more work is needed here (especially as it's dependent on two stations only). 

The authors do give some examples in their rebuttal of species where this works. I would suggest this is added to the discussion. 

One other point : Fig 4. is missing A-D on the map images

Author Response

Dear reviewer, thank you for your constructive comments. We have responded to your recommendations below.

Point 1: I have not been convinced of the argument that NDVI for Lomas equates to mm of precipitation. They are quite different entities and should be treated as such, plant species have evolved to capture fog in Lomas (ie, trichomes in many species), and if this moisture was delivered as precipitation, this adaptation would not be needed. For SDM's this can be problematic for specific fog-adapted species. I would suggest the author make this clear in the discussion and that more work is needed here (especially as it's dependent on two stations only).

Response 1: In our article, we want to address the problem that the bioclimatic surfaces created so far do not reflect the climatic reality of the southern zone of Peru, especially the coastal zone. Although we tried to find as much climatic information as possible from land-based stations, the difficulty was in the hilly ecosystems, where only two stations were recorded for our area. Then, observing the behavior of the vegetation in relation to humidity events, we decided to use NDVI to try to fill these gaps, as an alternative only applicable to these ecosystems. As we explained before, the experience gained working in these ecosystems brought us to this, and that is that there can be years with fog but without "lomas" vegetation, which is reflected precisely in the NDVI values.

On the other hand, as for vegetation adaptations such as hairs or scales, it is true that these allow condensing the surrounding humidity, however, this only works when the plant is already adult and in the open field. During the germination and growth phases, a good dosage of water is needed, which is undoubtedly not provided by the fog (and this is even more important because the vegetation of the lomas is annual, developing in open fields, without any nursery). However, the "lomas" vegetation is very dense, so the fog, coming from the ocean, would only impact with the edges of the vegetation patch, condensing water only in that place (leaving the plants in the central zones without fog to condense). Although in this last point due to this difficulty, plants have developed a growth pattern which is to grow in bands (especially on steep slopes), it is true that many others are found forming large clumps in flat areas. That is why we think that NDVI is a variable closely related to precipitation, and this only for “loma” ecosystems, since these depend only on stratocumulus clouds, and this direct relationship can be observed. Finally, we can observe data from the Lomas de Lachay climatic station, which records precipitation, with July, August and September being the wettest months (between 20 and 30 mm).

Regarding the ENM and SDM, it is true that surfaces that determine the Grinnellian or Eltonian niche are required, however, for general modeling, using the 19 bioclimatic surfaces is common (Grinnellian niche) in general in modeling, although if the biology of the species is known, other variables can be used, for example for lomas ecosystems, cloud cover, however, as mentioned before, precipitation in lomas ecosystems is essential for the emergence of the vegetation.

In any case, in the discussion section we emphasize that these are only an approximation and for exclusive use for these areas, on the other hand, we also indicate that more research is necessary to clarify this relationship.

Point 2: The authors do give some examples in their rebuttal of species where this works. I would suggest this is added to the discussion.

Response 2: We believe that for this manuscript presenting species modeling results, comparing the performance of our surfaces with those of other authors, is still not convenient. This is indicated because we would need to increase more concepts, methods, results and discussions, which is beyond the objectives of this article. However, we plan to prepare a new manuscript precisely comparing the performances for the ecological niche models, taking into account the necessary methods to highlight these differences (statistical comparisons, validation, map generation, etc.).

Point 3: One other point : Fig 4. is missing A-D on the map images

Response 3: The image was corrected.

Author Response File: Author Response.docx

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

Dear Authors,

I found your manuscript interesting especially the objective to develop new and more realistic maps of climate parameters for the study site. Unfortunately, I lost my interest rapidly when I saw that you applied the workflow of an interpolation process, and that is all. If this manuscript was mine I would try to reshape it in the following way:

- First you have to know and set as visibly as possible what is the novelty of your work. If you invented the new maps then you have to show in a better way that they represent a higher quality than the previous ones. In the current version the reader is not convinced that this is really the case. The reasons for that are several: you do not have (or did not present) test dataset of real meteorological stations, which were not used for the interpolation. Thus, we cannot see the real error statistics. You presented statistical measures of reference data and modelled data, too without presenting their distribution properties. You introduced virtual meteorological stations to fill data gaps. In my opinion it causes more trouble than the gaps it fills. On the one hand you applied satellite data what you criticized in connection with the other datasets, on the other hand the conversion method from NDVI to temperature or precipitation is not described in a reproducible way. It is important also in that case if the main novelty of your work is the method how to derive climate maps for areas suffering from lack of data. In my opinion the manuscript leaves the sound scientific baseline at the point where the virtual stations appear.

Last but not least your model seems to be a static version. It does not take into account the ongoing climate change and does not offer any option to estimate the impact of changes for the next decades.

So, my recommendation would be before resubmission to clarify the main point of the paper, elaborate the method section more properly, and try to handle the issue of climate change.

With kind regards

Author Response

Response to Reviewer 1 Comments

Dear reviewer, thank you for your constructive comments. We have responded to your recommendations below.

Point 1: First you have to know and set as visibly as possible what is the novelty of your work. If you invented the new maps then you have to show in a better way that they represent a higher quality than the previous ones.

Response 1: In this article we want to present our surfaces, as an alternative for the modeling of the ecological niche and species distribution. Produce surfaces that provide a better representativeness of the evaluated species. Here we only show that our surfaces are statistically different from others commonly used in these studies. However, using as many ground weather stations as possible for modeling and producing surfaces is somehow reflecting the station data on the produced surfaces. In this sense, we believe that the statistical differences occur precisely because of the quantity, the type of climatic information used and covariates, for the modeling and production of the surfaces. On the other hand, we did tests to model the ecological niche of various species using the MaxEnt algorithm, producing good results with our surfaces. For example, when performing ecological niche modeling with an endemic tree from Arequipa (Neltuma calderensis), WorldClim surfaces returned only areas within the department of Arequipa. However, when using our surfaces, the model indicated areas of high suitability in the department of Moquegua and Tacna. After a trip to Moquegua and conducting herbarium reviews, we were able to verify that indeed the areas where our model predicted high habitat suitability were also where the tree was found. We also did tests for species distributed in the coastal zone (hills) and satisfactory results were also given for our surfaces. We believe that a comparison in terms of the performance of our surfaces with others carried out for the area requires an additional article. Showing the results obtained and the modeling of the ecological niche of species, would make this work very extensive.

(please check the imagen in the word archive)

Fig. Ecological niche model of N. calderensis made with WorldClim surfaces. The pink dots were those used for the model and the yellow dots are those collected on a recent trip. Note how the area of the yellow dots is not predicted as a suitable area.

 (please check the imagen in the word archive)

Fig. Ecological niche of N. calderensis made with our surfaces. Note how the area where the yellow dots are predicted as suitable areas, despite not having been integrated into the niche model.

Point 2: You do not have (or did not present) test dataset of real meteorological stations, which were not used for the interpolation. Thus, we cannot see the real error statistics.

Response 2: We follow the suggested recommendation. In that sense, we performed the modeling again, taking into account a cross-validation (which is commonly used) of 10 folds, in order to obtain a more adequate measure of how well the results are produced taking into account the training data and calculating the bias in relation to the values predicted by the model. How this procedure was performed is described in methods.

Point 3: You presented statistical measures of reference data and modelled data, too without presenting their distribution properties.

Response 3: We do not quite understand this point. As indicated, we present a summary of the stadistical data of the data used and generated. In similar works we have not seen any distributional properties presented about the data used or produced. Could you please recommend some references to guide us in preparing something similar??

Point 4: You introduced virtual meteorological stations to fill data gaps. In my opinion it causes more trouble than the gaps it fills. On the one hand you applied satellite data what you criticized in connection with the other datasets, on the other hand the conversion method from NDVI to temperature or precipitation is not described in a reproducible way. It is important also in that case if the main novelty of your work is the method how to derive climate maps for areas suffering from lack of data.

Response 4: It was not our intention to criticize the climate data taken from satellites (we have corrected this so as not to give that impression). What we want to show is that the scarce climatic data used for the study area does not adequately reflect the climatic surfaces. In addition, the topographic complexity also intervenes on the modeling, so that unlike other models, we include (apart from altitude) slope, orientation, ITH, cloud cover and distance from the coastline to inland. In the case of precipitation values from virtual stations, we perform a direct proportional relationship between the real stations and the NDVI. That is, we took the NDVI values according to the locations of the real stations, averaged these values, extracted the NDVI values from the random virtual stations, and then by means of a proportional and direct relationship, the precipitation values were found (in methods we detail everything we did). However, this method would only work for hilly ecosystems, where the vegetation is almost annual and is closely linked to the light precipitation that occurs there. On the other hand, the intention of this is that the coastal zone shows as close as possible the humidity conditions of reality, and with this we have used, the models produced delimit very well the areas where precipitation actually occurs (the hilly ecosystems are characteristic because only the vegetation occurs in certain orientations, slopes, altitudes, the rest is surrounded by the driest desert in the world). Although we do not have the possibility of validating our assumptions with real data, since there are only two stations in these ecosystems (the rest of the stations located on the coast are very close to the ocean and do not reflect the conditions of the hills, even in ENSO events they have not recorded much precipitation), we believe that this can better delimit the ecological niche and distribution of the species, most of which are endemic. In other models made for the area, the coastal zone only presents flat precipitation values, or there are points of high precipitation in some localities. Finally, in methods and discussion we have added text related to this point.

Point 5: Last but not least your model seems to be a static version. It does not take into account the ongoing climate change and does not offer any option to estimate the impact of changes for the next decades.

Response 5: With a static model, are you trying to say that we have only produced a few surfaces in a certain time? If so, we had the idea of making various surfaces taking into account a time interval of 10 years, that is, surfaces for 1960-1970, 1970-1980, 1980, 1990... However, we decided to only produce them in the interval 1964- 2018, because the temporality in terms of the data collected is irregular (in some cases there are stations with all the temporality indicated above, however, there are cases in which only a certain moment was recorded, with stations appearing and disappearing throughout of the period). That is why we choose to take all the information possible (and that has at least one year of records) to produce the surfaces (which is also recommended by some authors, especially if the data is scarce or limited). Regarding climate change, we do not have the objective of producing models to observe changes in different contexts. However, given the comment, we will plan the way to carry them out and publish them later. What we also intend is that our produced data may be used by others to produce models and assess climate change, either projecting the ecological niche or distribution of species.

Author Response File: Author Response.docx

Reviewer 2 Report

This paper has several turning points. The sentences in lines 50, 51, 60 and 61 are not true since there are bioclimatic models for southern Peru and the whole country. Some of them are even cited in the book by Reynel et al., whose reference is numbered 32.

The Rivas-Martinez bioclimatic belts model has been widely used in South America and other parts of the world, but has not been cited. See the following references:

Amigo J, Ramírez C. 1998. A bioclimatic classification of Chile: Woodland communities in the temperate zone. Plant Ecology 136: 9-26.

Cress JJ et al. 2009. Terrestrial Ecosystems-Isobioclimates of the conterminous United States. USGS.

García Fuentes A et al. 2015. A study of the dry forest communities in the Dominican Republic. Anais da Academia Brasileira de Ciencias 87(1): 249-274.

Macías Rodríguez MA et al. (2014): Clasificación bioclimática de la vertiente del Pacífico mexicano y su relación con la vegetación potencial. Acta Botanica Mexicana 109: 133-165.

Rivas-Martínez S et al. 2011. Biogeographic Map of South America. A preliminary survey. International Journal of Geobotanical Research 1: 21-40 + Map.

Web: globalbioclimatics.org

This predictive methodology for vegetation has also been used in southern Peru, including by one of the authors, thus contradicting the title of the paper itself:

Galán de Mera A et al. 2010. Termoclima y humedad en el sur del Perú. Bioclimatología y bioindicadores en el departamento de Arequipa. Zonas Áridas 14: 71-83.

Galán de Mera A et al. 2014. Plant communities linked with cryogenic processes in the Peruvian Andes. Phytocoenologia 44: 121-161.

Galán de Mera A et al. 2018. Un ensayo sobre bioclimatología, vegetación y antropología en el Perú. Chloris chilensis 20(2).

Montesinos DB et al. 2012. Andean shrublands of Moquegua, South Peru: Prepuna plant communities. Phytocoenologia 42: 29-55.

Pauca-Tanco GA et al. 2020. Distribución y caracterización de las comunidades de Tillandsia (Bromeliaceae) en el sur de Perú y su relación con la altitud, pendiente y orientación. Ecosistemas 29(3): 1-11.

Vlillasante-Benavides F, Pauca-Tanco GA et al. 2021. Distribution patterns, ecological niche and conservation status of endemic Tillandsia purpurea along the Peruvian coast. Plant Systematics and Evolution 307(4).

-In area of study there should be some reference. The authors write about geomorphology but there is no reference. Line 82 emphasises climate in relation to humidity, but what about temperature? It also fluctuates quite a lot.

Results

The maps in figure 4 provide similar results to those of Galan de Mera in Arequipa, but the authors do not relate bioclimatology and species, nor plant communities, nor use any biogeographical reference. Bioclimatology by definition implies the correspondence between climate and vegetation. All this is contrary to the title of the paper when it indicates 'biological conservation'. My question is, what do the authors conserve with this model? They don't say because their model does not predict vegetation types. Conceptually this research is not designed correctly.

Discussion

In the discussion there are many long-known data, which do not relate bioclimatology to vegetation or to any biogeographical distribution at all (e.g. see lines 312 to 337).

Other comments

Is there no fieldwork to verify the results?

The title of the manuscript states 'an approach to climate reality' but all that is used in the manuscript are simulated computer models. Worldclim and Chelsa give values that often do not match reality because they extrapolate, which is why the use of plants and plant communities as bio-indicators is so important. This requires extensive field work. But all this is nothing new, it has already been done.

On line 380, reference 44 is the National Ecosystem Map of Peru. It is not a scientific work, nor does it explain any criteria for making the map. It does not really represent ecosystems, but plant formations mixed with biogeographical units. It does not really represent plant communities that can be paralleled with a bioclimatic model indicating which vegetation to conserve with greater preference, as is being done in Europe, USA, and other countries bordering Peru (Ecuador, Bolivia), where, by the way, the model of bioclimatic belts is being used.

This work ignores everything that has been studied so far in southern Peru, even if it is known to the main author. Therefore, I have no choice but to reject the manuscript.

Author Response

Response to Reviewer 2 Comments

Dear reviewer, we have responded to your recommendations below.

Point 1: In reference to all the written text.

Response 1: Dear reviewer, we think there is a mix-up going on. What we intend to do with our research is to produce surfaces, that is, raster layers, where each pixel indicates a modeled value (temperature or precipitation), based on data from weather stations taken on the ground. We do not intend to create belts or bioclimatic zones, much less provide a classification according to any criteria given by any author. Although both share the use of climatic data, on the one hand, bioclimatic belts are based on processing climatic data (for example, the thermal index) and classifying it according to some threshold (given by some author). This is then captured on a map and a characteristic bioindicator is attributed to each belt (usually vegetation, although something more complete should also use fauna). Weather surfaces, on the other hand, are raster with calculated pixel values on the one hand, with ground weather stations, covariates (which are carefully selected according to the variable - for example, for temperature it is mandatory to use altitude because of obvious reasons), and some algorithm (powered regression trees (BRT), neural networks (NN), generalized additive model (GAM), multivariable adaptive regression splines (MARS), support vector machines (SVM) and random forests (RF )), which uses interpolation of climate data and covariates (these do not extrapolate).

In terms of modeling, climate surfaces as such are the product of using data with a mathematical algorithm, which tries to interpolate known values and assign values to places without information. Being a model as it is, it is necessary to carry out a validation (for which there are certain processes, such as cross validation, cross validation with K-folds, subsampling, Kappa index, etc.), all this to evaluate how effective the model was product obtained. Regarding the bioclimatic zones, these are the result of a characterization and classification of the information from climatic stations and bioindicator agents, therefore, the bioclimatic belts do not require validation.

However, from the climatic surfaces it is possible to calculate the so-called bioclimatic surfaces, which are results of mathematical processes from the climatic surfaces. These bioclimatic surfaces are then used for modeling the ecological niche of the species (which is based on habitat suitability). All this theory about the suitability of the habitat has been studied by Joseph Grinnell (1917), Charles Elton (1927), G. Evelyn Hutchinson (1944-1958), their concepts being treated by later ones. Fundamentally, the niche of a species is given by an n-dimensional hypervolume, where the species finds its adequate conditions for survival. In this sense, temperature conditions (max, min, average), precipitation, wind speed, evapotranspiration, competition conditions, trophic niche, productivity, soil type, and many other variables condition the niche.

Although the ecological niche models should consider all the variables that intervene in the niche of a species, this is difficult, so it is decided to carry out the models based on variables which inevitably intervene in the species (such as climatic variables).

Now what a niche model does is indicate the suitability of the habitat from the environmental point of view, that is, it gathers the variables used and they are expressed in a geographical area. Although similar maps may be presented for some species, if this is analyzed from a three-dimensional image, using point clouds, we will see that the niches are different, given the variables of greatest importance.

So, the meaning of this work is to produce climatic surfaces and with these bioclimatic surfaces, which can be used for ecological niche modeling. And it is that, the existing surfaces do not adequately reflect the climate of this part of Peru (given the difficulties already explained in the work). Bioclimatic belts are an excellent way to characterize the climatic environment and designate bioindicators, but unfortunately, they cannot be used to determine the ecological niche or potential distribution of a species (although the categorical area of the bioclimatic belt can be added as complementary to make the niche model).

This work seeks to improve the quality of the climatic and bioclimatic surfaces, and therefore improve the ecological niche models, in order to have a better estimate of the distribution and thus propose some conservation measures for the treated species (which will logically depend interest of the researcher).

It is logical that our coastal surfaces are similar to some of its belts (hill ecosystems). This only indicates that there we were successful in representing these singular and special ecosystems, which in other authors are poorly represented (they are simply seen as flat surfaces or 0 precipitation).

Author Response File: Author Response.docx

Reviewer 3 Report

The reviewer would like to thank the authors for this thoughtful manuscript. This work has good potential. The authors are requested to put in some additional efforts to improve the quality of this manuscript. 

Introduction

The introduction is not satisfactory. The authors are requested to include more literature and present a strong case for their work.

Evaluation Metrics

With regards to model evaluation, the authors are requested to discuss the effectiveness of performance metrics (like RMSD etc.) demonstrated in the following interdisciplinary articles and discuss the use of these metrics for comparing modelled and observed data. 

i) Hastie et al., 2009. The elements of statistical learning: Data Mining, Inference, and Prediction

ii) Muhuri et al., 2021. Performance Assessment of Optical Satellite-Based Operational Snow Cover Monitoring Algorithms in Forested Landscapes, IEEE JSTARS.

iii) Valipour and Dietrich, 2022. Developing ensemble mean models of satellite remote sensing, climate reanalysis, and land surface models.

Fig. 1

The quality of the figures presented in the manuscript is good but they can be represented in a better manner. Please provide a colour coding to the altitude overlay in Fig. 1. It also appears like, except for the western part of Peru, the rest of the country is quite flat. Is this true?

Fig. 4

The authors are requested to plot some profiles in the North-East South West direction to observe the change in various variables. Please also include a way to show the altitudinal variation in these graphs.

Tables

The authors are requested to illustrate the data in Table 1, 2, and 3 in a graphical manner. Please provide the units in the tables’ captions.

Conclusion

The authors are requested to rewrite the conclusion. At the moment this section 

Author Response

Response to Reviewer 3 Comments

Dear reviewer, thank you for your constructive comments. We have responded to your recommendations below.

Point 1: Introduction

The introduction is not satisfactory. The authors are requested to include more literature and present a strong case for their work.

Response 1: We have expanded the introduction, including new bibliographic citations to reinforce the idea of our research. We have further elaborated on the ecological niche models, as well as on ground-based and satellite climate data.

Point 2: Evaluation Metrics

With regards to model evaluation, the authors are requested to discuss the effectiveness of performance metrics (like RMSD etc.) demonstrated in the following interdisciplinary articles and discuss the use of these metrics for comparing modelled and observed data.

Response 2: According to your suggestion and the suggested bibliography, we have performed our analysis again and used cross-validation with 10 folds. This is to get a more solid idea as to the measure of bias and how well the model predicts the new values relative to the training data.

Point 3: Fig. 1

The quality of the figures presented in the manuscript is good but they can be represented in a better manner. Please provide a colour coding to the altitude overlay in Fig. 1. It also appears like, except for the western part of Peru, the rest of the country is quite flat. Is this true?

Response 3: We have improved the images according to your suggestion. Actually the Peruvian Andean territory is very irregular in its topography, only the intermediate zone of the desert and the Amazon jungle are more or less flat.

Point 4: Fig. 4

The authors are requested to plot some profiles in the North-East South West direction to observe the change in various variables. Please also include a way to show the altitudinal variation in these graphs.

Response 4: Graphs were constructed showing the changes that occur with minimum temperature, maximum temperature and precipitation. In these graphs, the altitude of the terrain, the NE direction and the distance from the ocean to the interior of the study area were related.

Point 5: Tables

The authors are requested to illustrate the data in Table 1, 2, and 3 in a graphical manner. Please provide the units in the tables’ captions.

Response 5: The data in the tables indicated were illustrated. The recommendation to indicate the units at the bottom of the tables was also taken into account.

Point 6: Conclusion

The authors are requested to rewrite the conclusion. At the moment this section

Response 6: The conclusions were rewritten.

Author Response File: Author Response.docx