Estimation of Vegetation Leaf-Area-Index Dynamics from Multiple Satellite Products through Deep-Learning Method

Round 1

Reviewer 1 Report

This paper is very awkwardly written with a lot of missing or wrongly stated references. Scientific contribution is not clear mostly because of the lack of exact and relevant observations regarding research gap which the authors are trying to solve, as well as unclear and inconsistent data description used in this paper. Also, methods should be explained in more detail. I suggest rejecting it.

In the Introduction, it is stated that estimates of time series LAI are essential, but afterward that some LAI products provided invaluable information. Is it important or not, and why? By describing similar studies, it is stated „few studies“ and described just last. Please describe them together with their importance.

In the Study area and data section, the text on the top of every image in Figure 1. (c) is not visible. In this section, there are a lot of missing or wrongly stated references. Both name and exact path of data source (e.g. line 146) and every reference should be written in References.

In the Methodology section, it is not clear if there is train and test data for evaluation and what means „theoretical performance based on the training database“? Is data listed in line 187 mentioned in the Data section? I am very confused about the evaluation process in this paper. Why is data mentioned in lines 271-272 chosen for validation? The section is very awkwardly written. Please explain the usage of multiple Fully connected layers.

The Result section starts with the Table and the text on the top of every image in Figure 7. is not visible. Text is also not visible in Figure 5.

In the Discussion section, all graphs should be removed. The entire sections is awkwardly written and represent an extension of the Results without much actual discussion about the achieved scientific contribution in this research. Additionally, there is a lack of exact comparison with the similar previous studies, as the authors largely focus on interpreting solely their own results without stating their implications in the research area.

The Conclusion section should be extended and the contributions of this paper are not clearly stated.

 

Specific comments:

line 16: Why is it important for climate change?

line 41: Why are they essential and better than other metrics?

line 47: Wrongly stated reference, remove brackets.

line 131-132: Wrongly stated reference. The sentence is unclear, rephrase it. Also, Fig.2 does not have (c).

line 133: Missing references for used data.

line 139: What means „good accuracy“?

line 146, 157, 161, 167, 169: Please state reference.

line 157, 166: Wrongly stated reference.

line 161: What is the „main algorithm“?

Author Response

Dear Reviewer:

We would like to express our sincere appreciation for your careful reading and helpful comments to improve the quality of our manuscript. We have addressed all the major and minor issues raised. You can view them in the following PDF.

Author Response File: Author Response.pdf

Reviewer 2 Report

More details should be provided on the validation of fine resolution LAI maps (Line 138-140).

Unless version 6 GLASS LAI was not used in the paper. Do not describe the MODIS LAI Version 6 product (Line 159-165). Instead, describe the MODIS LAI Version 5 product.

Explain what was the basis for adopting such values for the hyperparameters of the LSTM network (Line 244-246).

Line 284 - LSTM VIIRS has a value of R2<0.85

Complete the R2 and bias results in Table 2.

It is worth using statistical test and answer the question whether LAIfusion results are significantly better than the others or not?

Figure 5 The legend should have unique values for DOY 177, 185, 193, ... Similarly, corrections should be made in other figures.

It should be completed in the legend what values are in the box plot for Reference LAI (minimum, mean, maximum)?

Author Response

Dear Reviewer:

We would like to express our sincere appreciation for your careful reading and helpful comments to improve the quality of our manuscript. We have addressed all the major and minor issues raised.

Author Response File: Author Response.pdf

Reviewer 3 Report

The manuscript “Estimation of vegetation leaf area index dynamics from multiple satellite products using a deep learning method” is dealing with leaf area index estimation through deep learning method while using multiple satellite products.

The manuscript is addressing an interesting aspect of deep learning and satellite data conversion into plant parameter representing vegetation growth and such conversion are being done in every field of life through deep learning. The manuscript is quite interesting for the readers and would contribute to machine learning research. Please find below some suggestions to improve the manuscript and increase its readability:

            Title:

1.     Title of the manuscript “Estimation of vegetation leaf area index dynamics from multiple satellite products using a deep learning method” may be changed as “Estimation of vegetation leaf area index dynamics from multiple satellite products through deep learning method”. It’s a suggestion but authors can keep the same title.

Abstract:

2.     Page #1, line # 6, “High-quality leaf area index (LAI) is important for climate changes” may please be changed as “High-quality leaf area index (LAI) is important for climate change”.

Introduction:

3.     Page #1, line # 43-45, sentence meanings are somehow confusing, so the authors are requested please rephrase the sentence.

4.     Page #2, line # 55-58, “Given that different products were generated using various models and algorithms that usually made different physical assumptions, the product-specific approaches easily caused evident inconsistencies among multiple LAI products [12]” sentence is quite long, and it should be rephrased for easy understanding.

5.     Page #2, line # 96, please add “Ma and Liang [27]” according to journal citation format.

6.     Page #3, line # 105-110, “aims to (1) validating the accuracy” correct it as “aims to (1) validate the accuracy”, please remove “ing” from all aims.

Results

7.     The results section is fine.

Discussion

8.     The discussion section is also written well.

Conclusions

9.     Page #13, line # 433-434, “Our proposed method not only accurately estimated LAI values” may be changed as “Our proposed method not only estimated LAI values accurately………”

These are my submissions. All the best for revisions.

Author Response

Dear Reviewer:

We would like to express our sincere appreciation for your careful reading and helpful comments to improve the quality of our manuscript. We have addressed all the major and minor issues raised. You can view them in the following PDF.

Sincerely,
Tian Liu 

Author Response File: Author Response.pdf

Reviewer 4 Report

In this work, the authors investigate how an Long-Short-Term-Memory (LSTM) model can be applied to generate Leaf-Area-Index (LAI) timeseries from remotely-sensed observations by being trained through different LAI products, by using as input the land-surface-reflectance and vegetation indices, and by combining several satellite products, το improve the accuracy of the regional-scale LAI and the smooting between its pixel vaues, as well as to allow for a sensitivity analysis on the input products. The paper is well-written, very interesting, and within the scope of the journal; and it requires only minor changes:

1) Please mathematically define the leaf-area-index (by using an equation), and its units. For example, in Figure 3, what are the units of the LAI, is it in % of the area covered by leaves? Also, please discuss what are is the difference/limitations/advantages between this index and the Normalized Difference Vegetation Index (NDVI).

2) It is mentioned in the text that in situ measurements were performed in order to validate the LAI maps; however, I could not find many information about these records. Please consider giving more details on the in situ observations (for example, when did these measurements occured, with what instrument, what cost, etc., show a Map with the locations of the in situ measurements, include a Table with statistical information, a Figure with a sample timeseries of these records, etc.), since this is considered an important innovation of this study (to validate the LAI maps with true observations).

3) It is mentioned in the text that a deep learning algorithm (such as the LSTM) can deal with the long non-linear dependencies of sequence values and temporal-evolution of the inherent uncertainty of natural processes, such as the vegetation dynamics. Please include in the discussion that this inherent uncertainty is shown to be quantified and analyzed through the so-called Hurst phenomenon (I think the first global-scale study to show that this phenomenon exists in the hydrological-cycle processes is by Dimitriadis et al., 2021), which the vegetation dynamics (and so, the LAI) is shown to exhibit (I think the first study to show this in vegetation dynamics is by Bashir et al., 2020).

Bashir B, Cao C, Naeem S, Zamani Joharestani M, Bo X, Afzal H, Jamal K, Mumtaz F. Spatio-Temporal Vegetation Dynamic and Persistence under Climatic and Anthropogenic Factors. Remote Sensing. 2020; 12(16):2612. https://doi.org/10.3390/rs12162612.

Dimitriadis P, Koutsoyiannis D, Iliopoulou T, Papanicolaou P. A Global-Scale Investigation of Stochastic Similarities in Marginal Distribution and Dependence Structure of Key Hydrological-Cycle Processes. Hydrology. 2021; 8(2):59. https://doi.org/10.3390/hydrology8020059.

4) It is mentioned in the analysis that for each pixel of the LAI, seven parameters were used. However, little information is given on these parameters. Please include in the discussion whether 7 parameters were used per pixel or for the whole area (in the latter case, please mention that one limitation of the LSTM models is the use of high number of parameters). Also, please consder adding a Figure with the values of these parameters for each pixel, the physical meaning of these parameters, etc.

 

5) Please mention the first author of the [27] study, when referring to it in the first person. For example, the sentence:

"In addition, [27] compared the theory performances of four deep learning methods but was not validated the retrieval results of LSTM models with the reference “true” LAI." should be changed to "In addition, Ma et al. ([27]) compared the theory performances of four deep learning methods but was not validated the retrieval results of LSTM models with the reference “true” LAI.".

6) Please consider including in the discussion more details about the variability between LSTMfusion, MODIS, GLASS, VIIRS, and the reference LAI. For example, in Figure 8, please quantify the gray area, so that one may understand how larger is the variation between the above models and the reference LAI.

7) Please consider adding another limitation of the LSTM models, which is related to the spatio-temporal correlation (not just the heterogeneity). It is expected that there is a large spatial (not just temporal) auto-correlation among the LAI pixels, since a high clustering is expected. For example, it is known that vegetation forms clusters, and so, the LAI is expected to also form clusters, and since, the stochastic clustering is linked to high correlation values, it is expected that a strong spatial correlation exists (there is a recent study on the spatial clustering mechanism in the mdpi's encyclopedia that includes a literature review: https://encyclopedia.pub/3294).

Author Response

Dear Reviewer:

We would like to express our sincere appreciation for your careful reading and helpful comments to improve the quality of our manuscript. We have addressed all the major and minor issues raised. You can view them in the following PDF.

Sincerely,
Tian Liu 

Round 2

Reviewer 1 Report

The authors addressed all my previous comments. I have nothing more to add.