Next Article in Journal
Assessing Interannual Urbanization of China’s Six Megacities Since 2000
Next Article in Special Issue
Climatic Drivers of Greening Trends in the Alps
Previous Article in Journal
New ECOSTRESS and MODIS Land Surface Temperature Data Reveal Fine-Scale Heat Vulnerability in Cities: A Case Study for Los Angeles County, California
Previous Article in Special Issue
Coupled Spatiotemporal Characterization of Monsoon Cloud Cover and Vegetation Phenology
 
 
Article
Peer-Review Record

How Does Scale Effect Influence Spring Vegetation Phenology Estimated from Satellite-Derived Vegetation Indexes?

by Licong Liu 1, Ruyin Cao 2,*, Miaogen Shen 3, Jin Chen 1, Jianmin Wang 4 and Xiaoyang Zhang 4
Reviewer 1: Anonymous
Reviewer 2: Anonymous
Reviewer 3: Anonymous
Submission received: 7 August 2019 / Revised: 8 September 2019 / Accepted: 10 September 2019 / Published: 13 September 2019
(This article belongs to the Special Issue Remote Sensing of Vegetation Phenology)

Round 1

Reviewer 1 Report

In this submitted manuscript, the authors presented a novel article about scale effect influence spring vegetation phenology estimated from satellite-derived vegetation indexes. However, the reviewer recommended that the manuscript need a minor revision and some of the points need well address. In particular:

1) abstract section, please compress and concise the “background and aim” (lines 15-22) as two or three sentence, this part is too long.

 

2) the manuscript need carefully English proof-reading, and please avoid using too many long sentences, in particular, introduction section.

 

3) I recommend that the discussion section need a more related publication to compare and then could spark the authors’ finding and argument. This is critically important.

 

 

and the reviewer still have some SPECIFIC COMMENTS (line numbering referring to the submitted manuscript)

Line 18: this sentence is not accurate, for example, land surface phenology metrics derived from time series from sentinel-2 to AVHRR is not from “hundreds of meters” …

 

Line 19: does the “green-up” suitable for the landscape scale or species/community scale? Please use landscape and land surface terms.

 

Line 20: “scale effect” is not new in remote sensing, vegetation phenology is a part of remote sensing. It is better to replace “remains unknown” with “needs further …”

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The research into the variation of the GUD is very useful and adds value to use in agriculture and environmental studies. I think this is a well written article and presents sound research. I think it would have been really useful to incorporate environmental factors as they directly influence the GUD.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

This manuscript seeks to address the “scale effect” on phenological metrics – i.e., why do estimates of parameters such as green-up date (GUD) vary with the spatial resolution of the sensor? The authors approach this question primarily using forward modeling. The results indicate that the scale effect is greatly affected by the heterogeneity of vegetation growth speed at fine pixel resolution. The GUD of a coarse pixel is found to be closer to that of the fine subpixels with earlier green-up, as opposed to the straight spatial average. Covariation between GUD and growth speed of fine pixels was also observed. A straightforward model was proposed that was able to explain about 60% of the scale effect.

 

The question addressed by this manuscript is applicable to a wide range of work in the remote sensing community and it is clearly suited for this journal. The manuscript is well-written and appears to address the question in a novel way, with clear implications for multiscale phenology studies. I am happy to recommend that this manuscript be published with only minor revisions, conducted at the authors’ discretion.

 

The most significant points I would suggest the authors focus on are what I consider to be the two primarily limitations of the study: 1) the assumption of a logistic functional form, and 2) the use of a vegetation index as the phenological metric. As the authors briefly acknowledge at the end of the Discussion, and is evident in Figure 3, actual time series may deviate significantly from the idealized logistic shape. Additionally, vegetation indices have several known problems, especially with sensitivity to soil (and soil moisture) background, saturation at moderate to high values, and nonlinear scaling. While fully addressing these problems is clearly beyond the scope of this manuscript, it would be appropriate to at least acknowledge and comment on them in the Discussion section.

 

Detailed comments below:

 

Lines 36-91: The introduction is clear, informative, and concise. It accurately summarizes the literature on the subject.

 

Lines 95-108: While I understand it would be beyond the scope of this manuscript, it would be interesting to compare dates of phenological transitions estimated by the logistic method with other methods. While the logistic method is indeed widely used, it carries with it an implicit assumption of the functional form of the phenology signal which is not necessarily accurate and may have an impact that propagates into what is inferred about the scale effect. The same is true about the use of curvature maximum as the green-up date. It could be worthwhile to investigate if using other ways of estimating the GUD results in different conclusions about the scale effect.

 

Line 108: “As a result, GUD is mainly determined by parameters a, b and c.” Following the logic of this paragraph, is it not true that GUD is actually entirely determined by these parameters?

 

Line 112: Is this a straight unweighted average of the GUD of all the fine pixels within the coarse pixel? Or something more complex? A number of approaches are used in the Peng at al. reference and it is not clear which one is being referred to here.

 

Lines 121-123: The three references given here do provide evidence for linear mixing of VI. However, a significant body of other work shows that at least some vegetation indices do not scale linearly. While EVI is certainly better behaved than NDVI, I suggest additional caveats about the assumption of linear scaling of vegetation indices. For instance, see:

 

Friedl, M. A., Davis, F. W., Michaelsen, J., & Moritz, M. A. (1995). Scaling and uncertainty in the relationship between the NDVI and land surface biophysical variables: An analysis using a scene simulation model and data from FIFE. Remote Sensing of Environment, 54(3), 233-246.

 

Fisher, J. I., Mustard, J. F., & Vadeboncoeur, M. A. (2006). Green leaf phenology at Landsat resolution: Scaling from the field to the satellite. Remote sensing of environment, 100(2), 265-279.

 

Lines 130-135: I do not think I understand the purpose of this paragraph. It seems excessive to go to the trouble to fit a logistic model, then replace the coefficients of that model with other parameters. Is the purpose to show that the parameters a, b, c of the logistic model can be written in terms of GUD, MP and GC? Or is the explicit quantitative relationship among them relevant? It might be useful to reassess this section of the text and see if it can be made clearer.

 

Line 153: While it is clearly beneficial to assume that the effects of delta_MP_fine and delta_GC_fine are independent, it is not clear that this is the case, especially given the functional form of the logistic model. It might be worth adding a sentence or two here explaining possible implications if they are not independent.

 

Lines 150-155: There are several cases here where it is not clear if there is a comma after the equation or if there is a ‘ mark, indicating a derivative. I suggest clarifying this in revision.

 

Figure 2d: What are the units on the axes here? Days for both x and y? Or is the bias unitless?

 

Lines 171-172 and Figure 2b, 2e: It would be helpful to add a clause clarifying if negative vs positive delta_MP_fine corresponds to faster or slower greenup of pixel 1 vs pixel 2. Once understood, the curves in figure 2e are very informative, but in the current form it is not trivial for the reader to understand how to interpret them.

 

Figures 2e and 2f: Are the y-axes of these plots normalized by delta_GUD? Does this imply, for instance, that for the -20 delta_MP, -20 delta_GUD case for figure 2e, the bias in delta_MP alone is actually 80 days (80 = -4 * -20)? Or is this a “given” symbol instead of a division symbol? The axis labels should be clarified and/or explained in more detail in the text.

 

Lines 185-190: All of this can be inferred from the results – with the caveats that 1) a logistic model is assumed, and 2) vegetation indices are used. I suggest that this caveat be stated explicitly and discussed in more depth.

 

Figures 8 and 9. It would make an even more convincing case to show a few examples of VI time series for fine and coarse pixels so that the reader can explicitly see what is changing during the averaging process, and how these changes are represented by the phenological parameters.

 

Figure 11. This figure makes a very important point, and effectively summarizes the message of the paper. However, as the figure is currently composed, the similarities between the blue and gold curves are much more obvious than the differences. I suggest that the figure be modified to more clearly show the differences. This could be done in a number of ways: panel C could be enlarged, additional background horizontal and vertical lines could be added, the axis boundaries could be changed to more clearly emphasize the fine-scale differences, or an additional panel could be added showing the the time series of the difference (pixel b – pixel a). It also might help if the black curves of the individual pixels in panels A and B were made partially transparent, gray, or thinner so that they are less prominent in the figure and allow the colored curve to be more visible. It is not clear what the dashed line in the inset in panel C represents. Also, the positions of the “b” and “c” labels are too far to the left.

 

Figure 12. I do not understand why both greater “growth speed” implies changes in both GC and MP, rather than just MP. I would think it would be easier to understand the example if only one parameter were changed at a time. Also, it is difficult to compare the mean (black) curves across panels. It might be easier to see the difference if both were plotted on the same axes.

 

Lines 382-387: Again, it would be helpful if the sign of the bias were also explained in terms of earlier or later green-up for coarse vs fine pixel.

 

Lines 409-410: This statement is not true for all biomes and vegetation types (e.g. water-limited rather than temperature limited vegetation). A qualifier should be added.

 

Lines 427-430: This is an excellent point. It may be useful to expand upon it further. It may also be worth adding an additional caveat about the use of vegetation index as the phenological metric (e.g. as opposed to subpixel vegetation fraction derived from spectral mixture analysis).

 

Equation A2: Does the “X” denote multiplication, or a variable? It is not clear as written.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Back to TopTop