Extent, Severity, and Temporal Patterns of Damage to Cuba’s Ecosystems following Hurricane Irma: MODIS and Sentinel-2 Hurricane Disturbance Vegetation Anomaly (HDVA)

Round 1

Reviewer 1 Report

It is an interesting study based on the title. However, no much innovation can be seen.

1. The major innovation is HDVA, but it lacks strong proof/evidence to use this metric. Besides doing basic analyzing of the images, you need to provide more important innovations in this manuscript.

2. (Table 2) The severity category is based on HDVA's quartiles. What if a different distribution is not like Fig. 6 at all? Are you still using the same classification standard?

3. Most "compare to" need to be changed into "compare with".

Author Response

Reviewer1.1It is an interesting study based on the title. However, no much innovation can be seen.

We appreciate the reviewer’s time and feedback. We have updated the title to focus more on the application to the study area based on the feedback from other reviewers. We have included some new analysis, including calculating the HDVA classification by land use type and performing additional statistical analysis to compare data sets and methods. We’ve updated the period of study through 2022 to analyze changes or recovery over several years following Irma. Other literature suggests a 9 month post-hurricane study period but we find that the extent of damages continues to change until 17-18 months post-hurricane. We’ve now accounted for other hurricanes in this region and included more discussion of the ecological impacts found in the study area as well as potential use of this method, especially in comparison to other methods and studies. These edits strengthen the manuscript and make it more compelling.

 

Reviewer1.2 The major innovation is HDVA, but it lacks strong proof/evidence to use this metric. Besides doing basic analyzing of the images, you need to provide more important innovations in this manuscript.

We appreciate this feedback. Based on this feedback and that from other reviewers, we have focused on the documentation of the cast study location, Cuba and Caguanes National Park. We emphasize the unique nature of this study area, including relatively undisturbed ecosystems (no deforestation, for example), and the impacts of Hurricane Irma. Unlike other recent remote sensing studies, it appears this region was affected by fresh water inundation (not saline sea water) for over a year that may have exacerbated the damages that we assess. We emphasize that this work was done at the request of local ecologists who monitor the area. Due to the limited access to petrol, road infrastructure, and other research obligations, the CNP authors can not do a full ground based assessment of ecological damages so this work is very important to local scientists and land managers. Additionally, it adds evidence of fresh water inundation as a potential new type of hurricane damage to mangroves.

 

Reviewer1.3. (Table 2) The severity category is based on HDVA's quartiles. What if a different distribution is not like Fig. 6 at all? Are you still using the same classification standard?

We have updated the quartile analysis so that our figures show the full distribution of HDVA data, including observations with no loss. It is a normally distributed curve. In other study areas, we would expect most observations to be close to 1 and decline to the tails. We did consider using quartiles by land cover type (e.g., different class boundaries for mangrove, etc.) but this would effectively decrease the representation of mangrove damages and increase the representation of minor dry forest damages to appear more extreme. Statistically, we found the quartiles were nearly identical for MODIS and Sentinel-2 data so these break points may be more wildly applicable but could use further investigation through other case studies.

 

Reviewer1.4. Most "compare to" need to be changed into "compare with".

We found one instance of the term “compare to” and have changed it to “compare with.”

Reviewer 2 Report

The study estimated the relative extent and severity of damage caused by Hurricane Irma (category 5) to the mangroves in the north of the island of Cuba. I think that title is inconsistent with respect to the objective of the study, since the title refers to Hurricane Disturbance Vegetation Anomaly, while the questions proposed were What is the potential for detection of mangrove damages and damage severity with Sentinel-2 data? 2) How does the analysis process and information from Sentinel-2 compare to MODIS? And 3) What is the extent and duration of ecological disturbance of Cuban mangroves from Hurricane Irma, particularly in Caguanes National Park and the Buena Vista reserve? I believe that the title could be modified to reflect the questions that the authors posed. The title is the first approximation that readers have about the article and should reflect the results obtained by the authors.

 Observations

Supplementary 1 was no available. "Files are currently under embargo but will be publicly accessible after November 28, 2023."

 Introduction

 Line 2. The word anomaly, presented in the title, methods, and results, is absent in the introduction. Therefore, I suggest that the authors present the theoretical basis relating to the concept of anomaly in hurricane disturbance on vegetation.

 

Line 87-88. The statement presented is inaccurate, since there are studies that used Sentinel 2 satellite scenes to evaluate the damage to mangroves caused by hurricanes. I attach some examples:

 https://www.mdpi.com/2072-4292/13/24/5042

https://www.mdpi.com/2072-4292/13/13/2565

https://www.mdpi.com/2072-4292/12/11/1740

https://www.sciencedirect.com/science/article/pii/S0303243420300398

https://besjournals.onlinelibrary.wiley.com/doi/full/10.1111/2041-210X.12923

https://www.cifor.org/publications/pdf_files/articles/ARoman-Cuesta2001.pdf

https://www.sciencedirect.com/science/article/pii/S0034425719302421

 Line 107. The study analyzed 17 years prior to the impact of the hurricane (2000-2017) and one year after (2018), so it is inaccurate to assert that “quantified the long-term damages of a high intensity storm to a mangrove”.

 

Methods

 I suggest that authors estimate magnitude change of EVI in 10 points in Caguanes National Park among hurricanes that impacted from 2000 to 2017, for example:

 Michelle (2001)

Isidore (2002)

Lili (2002)

Ivan (2004)

Dennys (2005)

Rita (2005)

Wilma (2005)

Ernesto (2006)

Gustav (2008)

Ike (2008)

Paloma (2008)

Matthew (2016)

 I suggest including a table with a summary of the Sentinel and MODIS scenes used in the study by year, for example, number of scenes or date.

 Line 131. The section suggests that study area was the path of the hurricane (figure 2), but in Figure 3 authors presented all protected areas of Cuba. Therefore, I suggest including in figure 2 a rectangle of study area and exclude protected areas of west coast.

 Line 239. I consider that use of the term "long-term" could be changed to short-term, since a one year has elapsed after impact of hurricane, while "long-term" could be considered decades or centuries.

 Lines 264-266. Question. Does the division of data into quartiles have a biological correspondence? Such a proposal could be supported by including the years 2019 to 2022. The HDVA classification was generated for all vegetation types, but each vegetation type has a different photosynthetic activity, which can be seen in Figure 12, where the mangrove and wetland vegetation type have the highest differences between before and after hurricane. However, both types of vegetation are adapted to this disturbs. Therefore, I suggest that HDVA classification be generated for each vegetation type.

 Since they performed a classification with two different sensors and the result may be different, I suggest applying a concordance test between classification obtained with Sentinel and MODIS (Cohen Kappa).

 Results

 I consider that subsections 3.1 and 3.2 could be moved to methods section, since they present the description of the inputs used.

 Line 347. Table 2 presents five categories, I consider that “no loss” category in Figure 6 should also be included, since it implies that there were areas that hurricane had a null or positive effect on vegetation. As shown in Figure 8.

 Line 364. The activities that led to the realization of figure 9 was not reported in methods. In subsection 2.2 authors refers that using 5m resolution scenes from 2017 and 2018, but not from 2019 and 2020.

 Line 368. Subsection 3.4 should be moved to methods section, as well as the criteria used to choose the 10 points in Caguanes National Park. Include in the maps the 10 points chosen.

 Line 387. Suggest that authors change “Significant damage” to “HDVA values were higher in the northeast of the island”. The word significant implies a statistical test, which they did not use.

 Line 394. I suggest that authors can apply a statistical test (e.g. X2), due observed variations may be due to chance.

 Discussion

 I suggest that author can increase their discuss on sources of variation that may compromise their results. As well as the studies that need to be carried out in the future to support and expand their results. Also, I suggest that author can contrast their results with other studies that analyzed the effect of hurricanes on mangroves. Include some examples:

 https://www.mdpi.com/2072-4292/13/8/1427

https://www.tandfonline.com/doi/pdf/10.1080/15481603.2018.1533679

https://www.jstor.org/stable/40663559

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8238932/

https://www.scirp.org/journal/paperinformation.aspx?paperid=110545

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

I suggest considering the following:

·       - Line 201, last word

·      -  In the legend of Figure 1, as well in other parts of the paper you refer to “Land use map”. I suggest using the term “land cover map” instead.

·      -  The part of Conclusions could be enriched with the benefits of the used methodology as well the future work.

Author Response

Reviewer 3.1 I suggest considering the following: Line 201, last word

We have changed “choses” to “selects” in this location.

 

Reviewer 3.2 In the legend of Figure 1, as well in other parts of the paper you refer to “Land use map”. I suggest using the term “land cover map” instead.

We have changed the term “land use map” to “land cover map” in the two instances that occurred in the original manuscript to clarify our message.

 

Reviewer 3.3 The part of Conclusions could be enriched with the benefits of the used methodology as well the future work.

We thank the reviewer for this suggestion. We have edited the Conclusions to include reflection on the benefits of these methods for future work (1 paragraph). We believe the additional edits suggested by Reviewer 2 in the discussion section and introduction also address the concern over broader contextualization of this work. In the Discussion (4.3 Sentinel-2 vs MODIS and 4.4 HDVA approach) we have added more detail to compare our findings to the literature for additional context on how the methods used here are working.

Round 2

Reviewer 1 Report

1. Show your major innovations in terms of methods/results in the Introduction or Abstract, so that reviews can understand if this study meet the high standard of this journal.

2. It is still lacking strong proof/evidence to use the HDVA, which is the base metric for most analysis in this study.

3. The severity category is based on HDVA's quartiles. What if a different distribution is not like Fig. 6 at all? What are your solutions?

4. Please check the different usage or definition between "compare to" and "compare with", after you only modify one one "compare to"!

Author Response

Reviewer1.2.1 Show your major innovations in terms of methods/results in the Introduction or Abstract, so that reviews can understand if this study meet the high standard of this journal.

We appreciate the Reviewer’s attention to the high impact of these sections on the interpretation of the study. We reviewed these sections and agreed that the findings of the study could be better portrayed. We have re-written the Abstract and amended the last paragraph of the Introduction, which are copied below in this document. Of particular significance is the vast spatial coverage of this event (>200,000 ha of mangrove and coastal wetlands were disturbed) that have not previously been documented elsewhere. When we say that there are limitations on field work, it is both limitations by local scientists who do not have sufficient budget to survey Caguanes National Park as they receive no funds for petrol and the mangroves in Cuba are largely undissected by roads or other human infrastructure, making access very difficult. Further, the lead authors visited the region in 2018 as part of an educational exchange but government relations make it very difficult to obtain research permissions for Americans in Cuba. Hence, there is a need for a severity class-based mapping approach to inform local scientists and this is how this study developed.

 

Abstract

Mangrove forests provide a range of ecosystem services but may be increasingly threatened by climate change in the North Atlantic with high intensity storms. Hurricane Irma (Category 5) hit the northern coast of Cuba (Sept. 2017), causing wide-spread damage to mangroves that has not previously been extensively documented due to financial and logistical constraints for local scientists. Our team estimated Irma’s impacts on Cuban ecosystems in a coastal and upland study area spanning over 1.7 million ha. We developed a multiresolution timeseries “vegetation anomaly” approach, where post-disturbance observations in photosynthetically active vegetation (Enhanced Vegetation Index, EVI) were normalized to the reference period (dry season mean over a historical timeseries). The Hurricane Disturbance Vegetation Anomaly (HDVA), was used to estimate the extent, severity and temporal patterns of ecological changes with Sentinel-2 and MODIS data and used vicarious validation with microsatellite interpretation (Planet). HDVA values were classed to convey qualitative labels useful for local scientists: 1) Catastrophic [0-.25], 2) Severe [.26-.50], 3) Moderate [.51-.76], 4) Mild [.76-1.0], 5) No Loss [>1.0]). Sentinel-2 had a limited reference period (2015-2017) compared to MODIS (2000-2017) yet the HDVA patterns were similar. Mangrove and wetlands (>265,000 ha) sustained widespread damages, with a staggering 78% damaged largely as severe to catastrophic (0-0.5 HDVA; >207,000 ha). The damaged area is 24x greater than documented impacts from Irma elsewhere. Caguanes National Park (>8,400 ha, excluding marine zones), experienced concentrated, severe mangrove and wetland damages (nearly 4,000 ha). The phenological declines from Irma’s impacts took up to 17 months to fully actualize, a much longer period than previously suggested. In contrast, dry forests saw rapid green flushes post-hurricane. With the increase of high intensity storm events and other threats to ecosystems, the HDVA methods outlined in here can be used to assess damages-- intense or gradual.

 

Revised end of Introduction

The ecological damages of Hurricane Irma were extensive, as witnessed first-hand by this team, but have not previously been extensively documented in the literature to follow the spatial extent, duration, and severity. Here, we quantified the short-term (years) damages of a high intensity storm to a mangrove ecosystem while testing new methods to provide rapid evaluation and locally relevant scale information. We apply our new method, Hurricane Disturbance Vegetation Anomaly (HDVA), which utilizes historical phenology represented by the Enhanced Vegetation Index (EVI) to normalize the magnitude of hurricane disturbances on land cover or so-called anomaly (HDVA >=1 when there is no damage and less than one when damages occur). Land covers like forests or mangroves have some interannual variations (EVI +/-~5-10%) but large variations may suggest more extreme impacts to the ecosystems, including mortality. We compared analyses with Sentinel-2 and MODIS derived damage detections (e.g., [25]). We addressed the following questions: 1) What is the potential for detection of mangrove damages and damage severity with Sentinel-2 data? 2) How do the analysis process and information vary when MODIS and Sentinel-2 data are compared? And 3) What is the extent and duration of ecological disturbance of Cuban mangroves from Hurricane Irma, particularly in CNP National Park?

 

 

Reviewer1.2.2 It is still lacking strong proof/evidence to use the HDVA, which is the base metric for most analysis in this study.

The motivation for this work was to provide a qualitative assessment of ecological impacts of a large disturbance in a region where extensive field work is not possible. We acknowledge that the review may be seeking further information, such as field-based estimates of canopy cover, which are not feasible for this case study due to the extreme limitation of access and resources to do that type of work. We do include these limitations in the manuscript. To further acknowledge the reviewer’s concern, we added a paragraph in section 4.4. (Results: HDVA results) on the future work that would be needed to improve the application of HDVA outside this study area, including additional work in well-studied regions that provide pre- and post-disturbance metrics to correlate with HDVA.

 

Reviewer1.2.3 The severity category is based on HDVA's quartiles. What if a different distribution is not like Fig. 6 at all? What are your solutions?

We have updated the quartile analysis so that our figures show the full distribution of HDVA data, including observations with no loss. It is a normally distributed curve. In other study areas, we would expect most observations to be close to 1 and decline to the tails similar to this case study. We did consider using quartiles by land cover type (e.g., different class boundaries for mangrove, etc.) but this would effectively decrease the representation of mangrove damages and increase the representation of minor dry forest damages to appear more extreme. Statistically, we found the quartiles were nearly identical for MODIS and Sentinel-2 data so these break points may be more wildly applicable but could use further investigation through other case studies. We have acknowledged this in the Discussion (4.4), which now concludes “Further refinement may be needed to determine the suitability of the quality labels (e.g., HDVA category breaks), as we found applying a quartile approach worked well for our data but might need refinement for other applications. For example, if we applied quartiles by land cover type instead of the entire study region then we would have marginally damaged dry forests classed as “catastrophic” the same as mangroves with severe mortality. In the future, there may be a more robust break in categories that would suit many different ecosystems.”

 

Reviewer1.2.4 Please check the different usage or definition between "compare to" and "compare with", after you only modify one one "compare to"!

Thank you for this note. We should use “compared with” when noting differences and similarities between data sets. Lines 26, 270, 283, 366, 505, 546, 547, 553, 567, 574: Compared to was changed to compared with. We changed Line 114 to read “2) How do the analysis process and information vary when MODIS and Sentinel-2 data are compared?”

Reviewer 2 Report

The changes the authors made to the munuscript substantially improved its quality. The authors used a z-test to compare the HDVA between evaluation periods, but this statistical test is absent in the methods section. For example, did they use a sample of paixels to compare before and after the hurricane? If so, I think they would have to use a paired z-test to compare the same pixel before and after the hurricane impact.

Author Response

Reviewer2.2.1 The changes the authors made to the munuscript substantially improved its quality. The authors used a z-test to compare the HDVA between evaluation periods, but this statistical test is absent in the methods section. For example, did they use a sample of paixels to compare before and after the hurricane? If so, I think they would have to use a paired z-test to compare the same pixel before and after the hurricane impact.

 

A one sample Z-test was used to determine if just the MODIS derived mean EVI value of the 2017-2018 dry season (year after Irma) varied significantly from the MODIS derived mean EVI of the 2000-2017 dry season (reference period). The mean EVI values we used for the Z-test are shown in figure 10. With more time, we were able to conduct a matched pairs t-test using 500 random points and comparing the pixel EVI value of the reference period to the same pixels EVI value of the disturbance year. The test was run on both MODIS and Sentinel-2 outputs and run for the Mangrove and Wetland as well as the Dry Forest land cover types. The addition of the t-test allowed us to test the significance of the reference period mean EVI composite raster to the mean EVI composite raster of the disturbance year as opposed to the Z-test which was testing if the year after Irma varied significantly from the mean EVI of each year before.

 

In the results section, we have added “We found that the EVI post-hurricane was significantly different than the reference period. Results for the matched pairs t-test showed the Mangrove and Wetland land cover type had a significantly lower dry season mean EVI in the disturbance year than the reference period (MODIS: t(453)=-17.68,P<0.001, Sentinel-2: t(477)=-18.29, P<0.0001). In Dry Forests, the disturbance year dry season mean EVI was significantly greater than the reference period (MODIS: t(499)=12.09, P<0.0001, Sentinel-2 t(499)=8.50, P<0.001). Considering all land cover types, both sensors agreed that the disturbance year mean EVI was significantly greater than the reference period although at a lower significance with Sentinel-2 (MODIS: t(486)=6.96,P<0.0001, Sentinel-2: t(491)=2.94, P=0.0017).”

 

In the Discussion section, we have added to section 4.2 (second paragraph) “While there was a significant decrease in EVI within Mangroves and Wetlands, a significant increase in EVI was detected across the entire study area as well as within the dry forest land cover type (see 3.2 Data Processing). Increases (HDVA>1) can be explained by ecological responses such as increased light availability and new plant growth with high cellulose content relative to non-photosynthetic vegetation. Additionally, there were several years of below normal rainfall or drought prior to Hurricane Irma that could affect a post-hurricane green-up.” We also added to section 4.2 (last paragraph) “The EVI values (Figure 10)  from 2020-2022 shows what a near a complete recovery within the Mangrove and Wetland land cover type. Based on field observations in CNP, we know these results are not indicative of the health of the entire mangrove population within the park. While the trends of these 10 points look promising, the catastrophically damaged mangroves may take over a decade to recover based on the local knowledge of this author team.”