Assessing Typhoon-Induced Canopy Damage Using Vegetation Indices in the Fushan Experimental Forest, Taiwan

Round 1

Reviewer 1 Report

In the article entitled " Assessing typhoon-induced canopy damage using vegetation indices in the Fushan Experimental Forest, Taiwan” the authors used remote sensing measurements and vegetation indices (VIs) to assess the cyclone impacts on Fushan forest ecosystems. The variation between pre and post typhoon vegetation indices values were analysed using Landsat imagery. After studying the Nari, Herb, Aere, Soudelor, and Dujuan typhoons, the authors suggest the use of NDII and EVI together when comparing cyclone disturbance-induced changes in vegetation cover. The article is providing useful information for the journal readers. I recommend the authors to reduce the abstract to 250 words. The paper can be accepted for publication with minor revisions.

Author Response

Reply: We appreciate the positive comment and have shortened the abstract to 250 words.

Abstract: Cyclonic windstorms profoundly affect forest structure and function throughout the tropics and subtropics. Remote sensing techniques and vegetation indices (VIs) have improved our ability to characterize cyclone impacts over broad spatial scales. Although VIs are useful for understanding changes in forest cover, their consistency on detecting changes in vegetation cover is not well understood. A better understanding of the similarities and differences in commonly used VIs across disturbance events and forest types is needed to reconcile the results from different studies. Using Landsat imagery, we analyzed the change between pre- and post-typhoon VI values (ΔVIs) of four VIs for five typhoons (local name of cyclones in the North Pacific) that affected the Fushan Experimental Forest of Taiwan. We found that typhoons varied in their effect on forest canopy cover even when they had comparable trajectories, wind speeds, and rainfall. Most VIs measured a decrease in forest cover following typhoons, ranging from -1.18% to -19.87%, however the direction of ΔVI-topography relationships varied among events. All typhoons significantly increased vegetation heterogeneity, and ΔVI was negatively related to pre-typhoon VI across all typhoons. Four of the five typhoons showed that more frequently affected sites had greater VI decreases. VIs ranged in their sensitivity to detect typhoon-induced changes in canopy coverage, and no single VI was most sensitive across all typhoons. Therefore, we recommend using VIs in combination, for example Normalized Difference Infrared Index (NDII) and Enhance Vegetation Index (EVI), when comparing cyclone disturbance-induced changes in vegetation cover among disturbances and across forests.

Author Response File: Author Response.docx

Reviewer 2 Report

This manuscript assesses the capability of the variation between pre- and post-typhoon VIs values for characterizing cyclone impacts on forest ecosystems. Overall, this paper is in good shape. And then the advantages of their strategies were appealed. I think some revisions would be necessary before acceptance.

L.82-84
Although I agree that there are a lot of VIs based on reflectance values over near-infrared (NIR) domain, some literatures showed that shortwave infrared (SWIR) band of Landsat was effective for evaluating the photosynthetic capacities. You should have mentioned the relationships between landcover and SWIR band.

2.3. Pre-processing
Could you clarify the methods related to the radiometric and atmospheric corrections? Which software did you use?
Did you conduct any pan-sharpening?

LL.223-224
Why did you apply ordinary leas squares regression? I think there are a lot of more effective methods whose abilities have been shown in some previous studies.

3. Results and 4. Discussion
There are some Chinese (?) characters. You must modify.

Author Response

Reviewer 2

This manuscript assesses the capability of the variation between pre- and post-typhoon VIs values for characterizing cyclone impacts on forest ecosystems. Overall, this paper is in good shape. And then the advantages of their strategies were appealed. I think some revisions would be necessary before acceptance.

Reply: We appreciate the overall positive comment and have carefully addressed the specific comments below.

L.82-84

Although I agree that there are a lot of VIs based on reflectance values over near-infrared (NIR) domain, some literatures showed that shortwave infrared (SWIR) band of Landsat was effective for evaluating the photosynthetic capacities. You should have mentioned the relationships between landcover and SWIR band.

Reply: We modified the sentence to acknowledge that SWIR band has been shown to be effective for evaluating forest canopy cover.

“VIs based on measurements using near-infrared (NIR) spectral bands, such as the Enhanced Vegetation Index (EVI), the Normalized Difference Vegetation Index (NDVI), and the Soil-adjusted Vegetation Index (SAVI), have been widely used in vegetation assessments [48,50-52] although shortwave infrared band of Landsat has also been shown to be effective for evaluating photosynthesis and forest canopy cover [53,54]” (L75-79)

Added references:

53. Lobell, D. B., Asner, G. P., Law, B. E., & Treuhaft, R. N. (2001). Subpixel canopy cover estimation of coniferous forests in Oregon using SWIR imaging spectrometry. Journal of Geophysical Research: Atmospheres, 106(D6), 5151-5160.
54. Tian, Y., Zhu, Y., & Cao, W. (2005). Monitoring leaf photosynthesis with canopy spectral reflectance in rice. Photosynthetica, 43(4), 481-489.

2.3. Pre-processing

Could you clarify the methods related to the radiometric and atmospheric corrections? Which software did you use?

Did you conduct any pan-sharpening?

Reply: We used images with radiometric and atmospheric corrections provided by by the USGS using the LEDAPS (Typhoons Aere, Herb, and Nari) and LaSRC (Typhoons Dujuan and Soudelor) algorithms. Topographic correction was done through R, with the topcor package as specified by:

“Rasters were topographically corrected using the C correction method with the topcor function of the ‘RStoolbox’ package [80] in R 3.6.1 [81]” (L.171-172)

Hence, the following sentence was modified to clarify it.

“Data from satellites Landsat 5, 7, and 8 were downloaded from the EarthExplorer website [79] as surface reflectance at 30 m resolution, with atmospheric and radiometric corrections performed with LEDAPS (Typhoons Aere, Herb and Nari) or LaSRC (Typhoons Dujuan and Soudelor) algorithms.” (L160-163)

LL.223-224

Why did you apply ordinary leas squares regression? I think there are a lot of more effective methods whose abilities have been shown in some previous studies.

Reply: We applied the ordinary least square (OLS) regression because it is a commonly used method used in forest ecology (e.g., Turner et al., 1999; Fisher et al., 2008; Saatchi et al., 2015). It is one simple way to evaluate the relationship between two variables of interest, and although there are more sophisticated analyses, usually if a relationship is found with OLS regression, it holds with the more sophisticated methods.  Following this comment, we explored other regression techniques and will take them into consideration for future studies. We added this to the Methods (L223-224).

Added references:

Fisher, J. I., Hurtt, G. C., Thomas, R. Q., & Chambers, J. Q. (2008). Clustered disturbances lead to bias in large‐scale estimates based on forest sample plots. Ecology Letters, 11(6), 554-563.

Saatchi, S., Mascaro, J., Xu, L., Keller, M., Yang, Y., Duffy, P., ... & Schimel, D. (2015). Seeing the forest beyond the trees. Global Ecology and Biogeography, 24(5), 606-610.

Turner, D. P., Cohen, W. B., Kennedy, R. E., Fassnacht, K. S., & Briggs, J. M. (1999). Relationships between leaf area index and Landsat TM spectral vegetation indices across three temperate zone sites. Remote sensing of environment, 70(1), 52-68.

3. Results and 4. Discussion

There are some Chinese (?) characters. You must modify.

Reply: We have carefully checked through the Results and Discussion and removed the Chinese characters. These Chinese words were not in our WORD file but was generated when the online submission system created the PDF file.

Author Response File: Author Response.docx

Reviewer 3 Report

Summary of study

Authors assessed the typhoon-induced forest canopy damages using some commonly used vegetation indices (VIs) in the Fushan experimental forest (FEF) in Taiwan. Authors chose FEF for their study, as this area had received the multiple typhoons that caused substantial forest canopy damages. Authors used four different VIs, which were derived from spectral reflection measurements, to assess the canopy damages caused by typhoons, and they are the enhanced VI (EVI), normalized difference VI (NDVI), soil-adjusted VI (SAVI), and normalized difference infrared index (NDII). Each of them was derived from the Landsat images taken before and after the typhoon events. Damages caused by five subsequent typhoons passing through FEF in were assessed. Authors showed that typhoons varied in their damages on the forest canopy even if they had comparable wind trajectories, wind speeds, and rainfall. Most VIs measured a decrease in forest cover following the typhoons, ranging from -1.18% to -19.87% (negative sign indicates damages). VIs varied among and within typhoon events in their sensitivities to detect typhoon-induced changes in the forest canopy coverage, and none of the single VIs that was the most sensitive to all five typhoon events. Topography appeared the most effective predictor for changes in VIs in the linear regression model.  Even though each of the VIs appeared almost identical on reflection of damages across the typhoons and forests, using a combination of NDII and EVI together could provide better results.

General comments

I enjoyed reading the manuscript, as this is written generally well and subject under consideration is interesting and problems were appropriately addressed. Objectives and hypotheses are appropriate and timely and according to which, methods employed are also appropriate. The current study is not completely new one, as there have been several studies of almost similar types already conducted to assess the damages caused by different typhoons in the different countries (authors have cited most of them in the manuscript). However, authors intended to evaluate four different VIs derived from images of multiple typhoons passing through the same forest area. Authors’ idea of comparing the canopy damages shown by different VIs can be appropriate, as any single VI may or may not be adequate enough to assess the damages. Thus, this study, to some extent, is relevant for the given problems. However, no one can guarantee that any of the VIs evaluated using remote sensing images could have provided the best results until and unless the field-based investigation (ground-truthing) or any other validating and verification approach was used. Authors did not consider this or any other suitable validation and verification approaches for confirming their results (canopy damages shown by VIs). Given the difficulties of this study, authors considered four VIs, which I considered somehow reasonable, to show the better results. This study has some novelty, and therefore this manuscript is deserved for publication in RS. However, I found following main issues that need to be properly addressed through major revision.

• Authors used the satellite images taken before and after the typhoons events to derive the VIs information. It would be more informative and better understandable to the potential audiences if authors present the image of FEF for each typhoon event (image of before and after typhoon, side by side). This needs presentation of only 10 small images of FEF by typhoon events, which were used to derive VIs information, in the data material sections.

• Authors considered studying the damages caused by five typhoons, which occurred in different years and passed through FEF and far away from FEF (Figure 1). Only two typhoon (Herb and Nari) were passing through FEF, and all other were passing far away from FEF. This means that the formers could have more damaging effects than other three typhoons. However, canopy damages due to typhoon events may depend on the typhoon speed, location, direction, forest structural development, time of derived VIs from imaged of the post-damages, etc. Authors, however, did not find more substantial effects of the herb and Nari on the forest canopy in the FEF. This gives me a little doubt over the reliability of VIs used to evaluate the damages. Why any of the VIs did not show more substantial effects caused by these two typhoon events, whose speed were also higher than other three (Table 2)? Please consider focusing more or investigating finely on these two typhoons and their damaging effects shown by VIs, and show the results accordingly.

• Authors used the correlation statistics of any of the two VIs for comparison. I don’t know how it could be beneficial, because none of them are not considered as a “standard” or “benchmark” VI. Authors are supposed to assume or make one of the VIs as benchmark and make other to be correlated to this. This only gives a better sense. Authors are suggested to find the way of making one of them as a standard or benchmark VI, however, it will be certainly a difficult task, I guess. As mentioned earlier, some kinds of ground-truthing or field-based investigation is necessary to make any parameter or index derived from remote sensing image as a standard or benchmark. If possible, authors are suggested to consider this approach to apply or other alternative approach and increase the reliability of VIs. If authors were able to do this, this study would become much more interesting and innovative, attract the international readerships overwhelmingly.

• Authors investigated the effects of topographic features on the damages caused by typhoons using multiple linear regression, which I consider a quite good approach. However, I wonder how it was possible to estimate the model parameters (model coefficients) of the model, as there may not be enough number of observations. Authors are suggested to formulate a general model form containing all the predictor variables and a response variable. How many observations were involved in estimating the parameters of the regression model? I am not sure, authors could have adequate enough observations to run the model and estimate the parameters. When there were not adequate observations (more than number of parameters in the model assumed), estimated values of the parameters of the predictor variables may not provide reliable information as expected because of biased estimation. One should be cautious while interpreting such a biased model’s results.

• Authors are suggested to present the results graphically rather than in Tables, where applicable (to be specified in minor comments). This study is based on the number of limitations and assumptions, which need to be briefly discuss in the manuscript (in discussion section) and possible ways of increasing the reliability of VI or making standardization of VI or possible improvement of the methods for future investigation on the similar problems.

Minor issues

Please be consistent while using the term: typhoon or cyclone throughout the manuscript, if not any difference between them.

Line 24, 35: Please define each acronym when it first appears in the manuscript. Most importantly, abstract should be made independently readable and understandable, meaning that each of uncommon acronym used in this section and symbol should be defined.  

Line 108: whether patterns….

Line 217-218: as mentioned above in my general comments, why correlations between individual VI were evaluated? What are the advantages of doing this? Please give reasons, if you have any.

Line 222-225: I have major concern here, whether you had adequate number of observations to build the regression model here and estimate the unbiased parameter estimates of the predictor variables involved in the model. I also suggest to present the scattered graphs of change of VIs and topographic variables (elevation, slope, aspect) in the data material section. This can show how these predictor variables are related to response variable (or change of VI). Alternatively, you may present these plots when you present the regression results. You may overlay the fitted line on the observed data there. Here also formulate general model form containing all the response and predictor variables.

Line 248-254: Please condense this part substantially, as most of the numerical results are presented in Table 4, which should not be repeatedly presented in the texts. Only most important results should be briefly described in the text, rather than presenting minor results in details.

Line 261-271: Please condense this part substantially, briefly describing only most important results with citation of Table 5. However, I am not well convinced with your idea here that these results (Table 5). See my comments elsewhere in details for this.

Line 272: As mentioned above, correlations of any two VIs may not provide useful information until and unless one of the VIs is standardized using your own data or any other reliable approach. What message does this table intend to deliver? The third column has remarkably higher correlation values, manning that EVI and SAVI are more significantly related to each other’s, rather than any other two. What does this mean? Is any of EVI and SAVI was assumed as a standard VI or benchmark VI? Your probable answer is no. I don’t see any advantages of carrying out this analysis, and presenting results here.

Line 282-288: This analysis is important, and therefore I suggest author to present the results graphically. I have (certainly readers also) difficulty of understanding these results. Why CI is necessary here in addition to CV, as usually CI comes along with mean. Presenting mean values and their CV or mean and CI may be suitable. Please consider simplifying the results here, and present them graphically for making them better understandable.

Line 306: As mentioned earlier, I have main concern with the regression model you had assumed and estimated the model coefficients (parameters). It seems all estimated parameters of the predictors in the model are significant. I wonder how it was possible, if there were not adequately enough observations available to this analysis. Please consider presenting a general model form that should contain all the predictor variables and mention the number of observations used for this modeling. Please add zero before decimal point in this table. Also, my concern here is that why you considered using Nari and Soudelor, and what basis they were considered for putting them side by side, why not Nari and Herb according to the reasons I have mentioned in general comments. Thus better to focus more on Nari and Herb rather than other.

Line 320-321: if adjusted R2 is too small, i.e. 0.001, relationship of response variable and predictors are so much weak, almost no existence of relationship. The numerical statistical indicators alone cannot be adequate enough to tell you much, so it is necessary to show the relationships of variables graphically. Please plot the response variable against predictor variables used in the model and overlay the fitted line of the regression on each of them.

Line 338-345: This is also important result. So please consider presenting results graphically for making them better understandable. Please see comments line 282-288, for making more improved presentation, here also.

Please define uncommon acronyms used in conclusion. This section should be made readable independently to other section of the manuscript.

Author Response

Reply: We appreciate the overall positive comment about our manuscript and the insightful comments that really help to improve the quality of the study. We carefully addressed all comments. We add two paragraphs in subsection 4.4. L.452 describing the limitations of our study due to the lack of ground-truthing and the difficulty of finding high quality images. Please see the attached file for a point-by-point response.

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

Thanks, authors for remarkable efforts you made to address my all concerns, and improve manuscript as per suggestion. I am satisfied with your responses and works done to improve manuscript. One very minor correction you forgot in your revision is adding of zero before decimal point in Table 6 (revised version). So, please do for this during proof-read correction phase. The revised version is good enough to be published in RS. Congratulations!