Vegetation Dynamics of Sub-Mediterranean Low-Mountain Landscapes under Climate Change (on the Example of Southeastern Crimea)
Round 1
Reviewer 1 Report
The team of 4 authors submitted a manuscript that is interesting and is linked to the objectives of the journal, however, there are some issues that have to be reconsidered.
The objective of the manuscript is to provide original insight into vegetation dynamics based on remote sensing data and climatic variables, including annual air temperature, annual precipitation, and annual solar radiation.
The subject area is rather interesting, and, possibly, not enough approached by other scholars, so there is potential room for this manuscript to bring new information, once it reaches the expected level of quality.
The Abstract has to be reconsidered, providing more information about the representative results (mainly the measurable ones). Also, please do not use any abbreviations on the Abstract.
For better visibility on databases, the authors are asked not to repeat among keywords the words/concepts included in the title of the article. Entering different words in the title and in the keywords can improve the search for the paper in metasearch engines and internet databases.
In the introduction, the objective of the study is detailed and defined, but the structure of the manuscript should be mentioned. The part of the Literature Review is part of the Introduction and looks a bit superficial, so the literature gap it could not be better pointed out.
The methodology part is well-constricted and offers a reliable base for the results.
The Results. The results are interesting and well-constructed and, also, The discussion. Part is consistent.
The conclusion. Limits of the research must be mentioned.
Author Response
Response to Reviewer 1 Comments
Point 1. The team of 4 authors submitted a manuscript that is interesting and is linked to the objectives of the journal, however, there are some issues that have to be reconsidered. The objective of the manuscript is to provide original insight into vegetation dynamics based on remote sensing data and climatic variables, including annual air temperature, annual precipitation, and annual solar radiation. The subject area is rather interesting, and, possibly, not enough approached by other scholars, so there is potential room for this manuscript to bring new information, once it reaches the expected level of quality.
Response 1: We appreciate the thoughtful feedback from the reviewer regarding our submitted manuscript. We are pleased to hear that the manuscript aligns with the objectives of the journal and addresses an interesting subject area that potentially offers unique insights. We acknowledge the reviewer's observations that certain aspects need further consideration to ensure the manuscript attains the desired level of quality. We will diligently address the outlined issues and concerns highlighted by the reviewer. Our aim is to meet the journal's expectations and contribute valuable new information to the field. We greatly value the reviewer's time and expertise in evaluating our work. We are committed to delivering a manuscript that meets the high standards of the journal and advances the understanding of the interplay between vegetation and climate dynamics. Thank you again for your valuable feedback and your consideration of our work.
Point 2. The Abstract has to be reconsidered, providing more information about the representative results (mainly the measurable ones).
Response 2: Done. We have revised the abstract to incorporate additional information regarding representative results, with a particular focus on measurable outcomes.
Point 3. Also, please do not use any abbreviations on the Abstract.
Response 3: Done. We have revised the abstract to eliminate all abbreviations, as per your request
Point 4. For better visibility on databases, the authors are asked not to repeat among keywords the words/concepts included in the title of the article. Entering different words in the title and in the keywords can improve the search for the paper in metasearch engines and internet databases.
Response 4: Done. We have addressed the concern by ensuring that the keywords in the abstract do not replicate words or concepts already present in the article's title, thereby enhancing visibility on databases.
Point 5. In the introduction, the objective of the study is detailed and defined, but the structure of the manuscript should be mentioned.
Response 5: We have made all the necessary corrections.
Point 6. The part of the Literature Review is part of the Introduction and looks a bit superficial, so the literature gap it could not be better pointed out.
Response 6: Done.
Point 7. The methodology part is well-constricted and offers a reliable base for the results.
Response 7: We sincerely appreciate the reviewer for their valuable feedback regarding the methodology section. We are pleased that our efforts to establish a comprehensive and reliable methodology have been recognized.
Point 8. The Results. The results are interesting and well-constructed and, also, The discussion. Part is consistent.
Response 8: We extend our gratitude to the reviewer for their thoughtful assessment. Your insightful comments provide valuable encouragement as we continue to refine our work.
Point 9. The conclusion. Limits of the research must be mentioned.
Response 9: The limitations of the study are delineated at the conclusion of Section 4, "Discussion." In Section 5, "Conclusion," we outline the prospects for potential future research endeavors.
Reviewer 2 Report
The manuscript, entitled “Vegetation Dynamics of Sub-Mediterranean Low-Mountain Landscapes under Climate Change (on the Example of Southeastern Crimea” is to assess changes in NDVI values and climatic changes, and analyze relationships between climatic changes and vegetation dynamics and trends in vegetation change in the Southeastern Crimea from 2001 to 2022. The topic falls within the scope of the journal.
I recommend a following major revision.
1. Previous studies have pointed out that NDVI is insensitive to high vegetation cover and inaccurate in estimating low vegetation coverage areas. Why not use other vegetation indices to analyze forest dynamics?
2. Why choose the annual NDVI average value instead of other factors that can reflect vegetation changes such as the minimum and maximum values during the growing season as factors for evaluating vegetation dynamics in the study area? Suggest supplement the relevant results in Section 3.Resutls.
3. Please modify the texts in Figure 1 to ensure that they can be seen clearly in the printed version.
4. Suggest supplement the meteorological conditions and size of the study area in Section 2.1 Study Area.
5. Suggest supplement the mapping time in Line 84 in Page 2.
6. Suggest supplement the spatial scales in Line 108 in Page 4 and Line 171 in Page 5.
Author Response
Response to Reviewer 2 Comments
Point 1. I recommend a following major revision.
Response 1: We extend our gratitude to the reviewer of the scientific article for their valuable input and their recommendation for a significant revision. Your insights are highly appreciated and will undoubtedly contribute to enhancing the quality and rigor of our work.
Point 2. Previous studies have pointed out that NDVI is insensitive to high vegetation cover and inaccurate in estimating low vegetation coverage areas. Why not use other vegetation indices to analyze forest dynamics?
Response 2: Done. We have outlined our considerations in lines 547-619.
Point 3. Why choose the annual NDVI average value instead of other factors that can reflect vegetation changes such as the minimum and maximum values during the growing season as factors for evaluating vegetation dynamics in the study area? Suggest supplement the relevant results in Section 3.Resutls.
Response 3: Done. We have outlined our considerations in lines 312-357.
Point 4. Please modify the texts in Figure 1 to ensure that they can be seen clearly in the printed version.
Response 4: Done. We have incorporated the necessary revisions into the manuscript. Your guidance has been invaluable in refining the content, and we appreciate your thorough review.
Point 5. Suggest supplement the meteorological conditions and size of the study area in Section 2.1 Study Area.
Response 5: Done. We have incorporated the necessary revisions into the manuscript. Your guidance has been invaluable in refining the content, and we appreciate your thorough review.
Point 6. Suggest supplement the mapping time in Line 84 in Page 2.
Response 6: Done. We have incorporated the necessary revisions into the manuscript. Your guidance has been invaluable in refining the content, and we appreciate your thorough review.
Point 7. Suggest supplement the spatial scales in Line 108 in Page 4 and Line 171 in Page 5.
Response 7: Done. We have incorporated the necessary revisions into the manuscript. Your guidance has been invaluable in refining the content, and we appreciate your thorough review.
Reviewer 3 Report
This paper discusses the analysis of MODIS NDVI data to study vegetation dynamics with the final goal of establishing strategies for the future development of different vegetation cover types. To this end, the authors estimate trends in vegetation cover photosynthetic activity and the fractal index H (Hurst exponent) to evaluate long-range memory properties of NDVI time series.
The authors do not say what method they used to estimate the exponent H. I suppose they used the very popular R/S analysis, which is not appropriate in this context. It is well known that R/S can be applied to series that are stationary in mean (J. W. Kantelhardt, in Encyclopedia of Complexity and Systems Science (Springer, Berlin, 2009), pp. 3754–3779) whereas the analysis proposed in this paper focuses on the estimation of H for time series with drift. “In the natural world, the real-life data usually contains some trends, making the series nonstationary and invalidating the R/S analysis…”. (Holl et al., Physical Review E, 2016, doi: 10.1103/PhysRevE.94.042201).
I may recommend the use of more sophisticate methods (e.g. Detrended Fluctuation Analysis; C.-K. Peng, et al., Phys. Rev. E 49 (1994) 1685) to solve this problem, but I think this is not sufficient because the proposed methodology focuses specifically on the values of H for different values of trends. The criteria stated to evaluate future change on the basis of the contextual analysis of slopes and H values (table 2) are groundless. Slopes refer to a deterministic representation of change in “mean” whereas long-range memory refers to the stochastic properties of the “departure from this mean”.
A similar incorrect approach characterizes the estimation of correlation with climate variables. The basic assumption of the correlation analysis is the stationarity of the time series. Any whatever textbook recommends the removal of trends before estimating cross correlation. Rigorously speaking, serial auto-correlation too should be removed. Unfortunately, many non-expert authors and referees do not pay attention to this delicate matter thereby leading to wrong conclusions.
In addition, another serious problem is linked to the time series length. In any scaling analysis, such as the analysis of the Hurst effect, the original sample is successively divided in ever finer subsamples to investigate the scale property of the data. The authors look at annual values and then they have only 22 data per pixel. In practice, there is not sufficient sampling to perform any reasonable scaling analysis. The availability of free software platforms often favours the use of concept and tools of the statistical mechanics without any control.
Finally, I do not agree with the use of the word “degradation”. Land degradation has a well-defined meaning concerning the ability of ecosystems to provide goods and services. Natural not significant (line 292) fluctuations cannot be discussed in terms of degradation and recovering, which is a more serious problem.
I
Author Response
Response to Reviewer 3 Comments
Point 1. This paper discusses the analysis of MODIS NDVI data to study vegetation dynamics with the final goal of establishing strategies for the future development of different vegetation cover types. To this end, the authors estimate trends in vegetation cover photosynthetic activity and the fractal index H (Hurst exponent) to evaluate long-range memory properties of NDVI time series.
1. Dear Reviewer. I would like to express my sincere gratitude for your valuable review of our article. Your insights and recommendations will assist us in making our article more comprehensive and accurate. We truly value the time and effort you invested in evaluating our work. Your comments will be incorporated into the further development of the article, aiming to make it even more valuable and relevant to our readers.
Point 2. The authors do not say what method they used to estimate the exponent H. I suppose they used the very popular R/S analysis, which is not appropriate in this context. It is well known that R/S can be applied to series that are stationary in mean (J. W. Kantelhardt, in Encyclopedia of Complexity and Systems Science (Springer, Berlin, 2009), pp. 3754–3779) whereas the analysis proposed in this paper focuses on the estimation of H for time series with drift. “In the natural world, the real-life data usually contains some trends, making the series nonstationary and invalidating the R/S analysis…”. (Holl et al., Physical Review E, 2016, doi: 10.1103/PhysRevE.94.042201).
Response 2. We concur with the reviewers' feedback and, consequently, have revisited the calculations within the manuscript. We have implemented the recommended changes and amendments to the manuscript's content. Please review the following sections for a more detailed examination of the revisions made.
Data values were obtained from the raster using the Quantum GIS software package. The calculation of the Hurst index was conducted using the R Studio programming environment, facilitated by the «pracma» library. This method involves the computation of the Hurst index through R/S analysis. To execute the R/S analysis, temporal series spanning the years 2001 to 2022 were individually subjected to a comprehensive examination for stationarity or non-stationarity. This evaluation was undertaken for each spatial cell. The Augmented Dickey-Fuller test, available within the R Studio environment and facilitated by the «tseries» library, was employed for this purpose. In instances where non-stationary time series were encountered, a sequence of transformations was applied to render them stationary. The R Studio program was employed for this transformation process, utilizing techniques such as logarithmization and differencing. These methods were strategically utilized to normalize variance and mitigate the presence of trends within the data. Upon the attainment of stationary time se-ries, the computation of the Hurst index was undertaken through the utilization of the 'pracma' library within R Studio. This index is derived from the R/S analysis and serves as an indicator of long-range dependence or persistence within the data. The calculated Hurst index values offer insights into the underlying temporal dynamics of the analyzed variables across different spatial cells. It is worth noting that this methodological approach aligns with a comprehensive workflow involving data preprocessing, statistical analysis, and computational procedures, all orchestrated within the R Studio environment.
Point 3. I may recommend the use of more sophisticate methods (e.g. Detrended Fluctuation Analysis; C.-K. Peng, et al., Phys. Rev. E 49 (1994) 1685) to solve this problem, but I think this is not sufficient because the proposed methodology focuses specifically on the values of H for different values of trends. The criteria stated to evaluate future change on the basis of the contextual analysis of slopes and H values (table 2) are groundless. Slopes refer to a deterministic representation of change in “mean” whereas long-range memory refers to the stochastic properties of the “departure from this mean”.
Response 3. We did not seek to show the Detrended Fluctuation Analysis study in the work, but based on your recommendations, we added this in section 5 "Conclusion" to the prospects for future research on the topic. Regarding Table 2, we used its values based on previously published works. However, after making a deeper analysis of the publications, we agree with the reviewer that this issue requires further study, in this regard, we have removed Table 2 and the map based on it from the manuscript.
Point 4. A similar incorrect approach characterizes the estimation of correlation with climate variables. The basic assumption of the correlation analysis is the stationarity of the time series. Any whatever textbook recommends the removal of trends before estimating cross correlation. Rigorously speaking, serial auto-correlation too should be removed. Unfortunately, many non-expert authors and referees do not pay attention to this delicate matter thereby leading to wrong conclusions.
Response 4. To assess the correlation, we employed methodologies as detailed in references [38, 68, 69]. However, it has become apparent that these methods are not exhaustive in nature. We express our gratitude to the reviewer for their insightful observations, which prompted us to address this limitation. Consequently, we have made the necessary revisions to the manuscript's content and have also undertaken a recalibration of the maps based on the updated calculations.
The correlation assessment was performed using the R Studio software package. For the purpose of conducting correlation analysis, NDVI and climatic datasets spanning the years 2001 to 2022 were subjected to rigorous examination in terms of their stationarity or non-stationarity. This examination was undertaken utilizing the Augmented Dickey-Fuller test, a statistical method, within the computational environment of R Studio, specifically leveraging the «tseries» library. Time series exhibiting non-stationarity underwent a process of transformation into stationary series, accomplished through the application of logarithmic and differencing techniques. These techniques were instrumental in homogenizing variance across the temporal dimension and effecting the elimination of underlying trends. Subsequent to the attainment of stationary time series, correlation coefficients were computed individually for each temporal data point, a crucial step in elucidating the relationships between the NDVI and climatic variables. Upon generation of these correlation datasets, a seamless transition was effected into the ArcGIS software package. Within this geospatial environment, an interpolation procedure, employing the well-regarded «Spline» method, was executed. This interpolation operation facilitated the derivation of intermediate values between discrete data points. Subsequently, cartographic representations were generated to visually articulate the spatial patterns emerging from the interpolated data, providing valuable insights into the dynamics of the studied variables across the geographical extent.
Point 5. In addition, another serious problem is linked to the time series length. In any scaling analysis, such as the analysis of the Hurst effect, the original sample is successively divided in ever finer subsamples to investigate the scale property of the data. The authors look at annual values and then they have only 22 data per pixel. In practice, there is not sufficient sampling to perform any reasonable scaling analysis. The availability of free software platforms often favours the use of concept and tools of the statistical mechanics without any control.
Response 5. The study of NDVI dynamics, as an indicator for global and regional monitoring of vegetation and forest ecosystems, currently lacks extensive historical measurements due to the commencement of satellite missions. The dataset utilized in this study is constrained by the start of 2001, coinciding with the inception of scientific operations involving the MODIS satellite. This limitation is a direct result of the launch and operation timelines associated with satellite missions.
Point 6. Finally, I do not agree with the use of the word “degradation”. Land degradation has a well-defined meaning concerning the ability of ecosystems to provide goods and services. Natural not significant (line 292) fluctuations cannot be discussed in terms of degradation and recovering, which is a more serious problem.
Response 6. In our research, we have adopted the terminology proposed in the publication [37], as well as in numerous works by other authors. We intentionally refrain from inventing and introducing new terms into the scientific discourse and instead rely on previously published studies by other researchers in this regard. At the same time, taking into account your comments made in paragraph 3, we decided to remove table 2 and this term from the text of the article.
Round 2
Reviewer 3 Report
I would like to acknowledge the efforts made by the authors to improve their manuscript. Nonetheless, in my opinion there are still basic weak points in this study. The time series are too short (22 yearly data) to perform any reliable scaling analysis thereby attributing a realistic meaning to the estimated Hurst exponents. A lower limit for the size of the sample is usually N=28=256 (e.g., Crevecoeur et al, 2010) while the convergence of the estimation to the real value is obtained asymptotically with increasing N. In addition, using these estimated values to infer persistence properties of trends in mean (see comments on figure 9) is not correct, as already stressed in the first round. I understand that there is recent literature that uses the same approach, however, as an expert in the field, I cannot endorse it.
Crevecoeur F, Bollens B, Detrembleur C, Lejeune TM. Towards a "gold-standard" approach to address the presence of long-range auto-correlation in physiological time series. J Neurosci Methods. 2010 Sep 30;192(1):163-72. doi: 10.1016/j.jneumeth.2010.07.017.
The paper is easy to read.
Author Response
Dear Esteemed Reviewer 3,
We extend our sincere gratitude for your invaluable comments and insights, which have greatly contributed to the refinement of our manuscript. We wish to address your concerns and provide further elucidation regarding the length of the time series and its relevance to the estimation of the Hurst Index.
Firstly, we acknowledge your point about the length of our time series, which spans 22 years, encompassing data acquired by the MODIS satellite from 2001 to 2022. It is important to clarify that the MODIS satellite does not provide direct annual averages. To facilitate the presentation of our findings, we computed these annual averages from the satellite's 8-day data acquisition cycle. This results in an average of three measurements per month and a total of 36 measurements per year. Over the 22-year duration, this accumulates to a substantial 792 observations or measurements for each pixel in our dataset, significantly surpassing the minimum threshold of 256 data points as suggested.
Hence, we assert that the dataset utilized in our study, spanning 22 years and featuring non-annual average values, is indeed ample for conducting the requisite calculations and estimating the Hurst Index.
We greatly value your expertise in this field, and we remain committed to addressing your concerns comprehensively. Your feedback continues to be a pivotal part of our manuscript improvement process.
Additionally, we removed the unsuccessful evaluation characteristics after Figure 9, as indicated by Reviewer 3.
Respectfully, Authors.
