A Machine Learning Method for Predicting Vegetation Indices in China

Round 1

Reviewer 1 Report

This paper used XGBoost to predict NDVI variances (both spatially and temporally) in China. The paper is really well written, with a good flow. I enjoyed reading it. The methods section is quite clear, and the results and conclusions well justified. After reading the main texts, I am left with a few minor suggestions.

 

1. In the last paragraph of introduction, the authors stated that they focused on China as their research area, without explaining the reason. China has diverse ecosystem types. I suggest the authors to mention this point to justify their selection of research area.
2. NDVI was used as an indicator of vegetation growth. This assumption is reasonable, but I suggest the authors to refer to existing studies to support their point. The use of NDVI in China has a history. A potentially good reference is: Fang et al. 2001 Science. 5536,1723-1723.
3. Similar, I would appreciate it if the authors could refer to more research using XGBoost over China’s ecosystems. I suggest the following: Zhang et al. 2019 Remote Sens. 11(12), 1505.
4. Why the models were built using monthly predicting variables, but the results presented into yearly data? The authors might want to explain this point in the methods section.
5. Another minor concern is, some vegetated areas in China were disturbed by human activities. Does the model capture this effect?
6. I didn’t find related texts how the authors defined “vegetated areas”. For sure not all 1km- resolution pixels were vegetated. Were all pixels included into the models? Will non-vegetated pixels influence your results? The authors might also want to discuss the influence of pixels with mixed landcover types.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper entitled “A machine learning method for predicting vegetation growth in China” developed a machine learning model using the extreme gradient boosting method to predict vegetation growth (in terms of NDVI) in China. Results show the excellent performance of the model: the overall bias between predicted and observed annual average NDVI values was less than 5 %, and the mean RMSE was 0.05, which was less than 0.1 % for 98.4 % of pixels.

I think the study is very interesting. I would develop more the potential and importance of a forecasting tool of this kind. Why is it important to predict the growth of vegetation and who would use this tool.

The below aspects should be addressed by the authors before publication.

SPECIFIC COMMENTS:

- Line 44: Could you briefly mention what you mean when you refer to the "ecosystem process mechanism"? I think it is important to specify this in order to highlight the difference between the two methods.

-Line 53: Why only in China? I mean, I understand the application in this study, but I think the model can be extended and used in other areas too, right? Or are there particular limitations?

- Could you improve figures resolution’s?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The authors used XGBoost method to estimate the well-known vegetation index, NDVI, in China. Results suggest that annual NDVI for different climates can be predicted using their model.

The uploaded manuscript is 10 pages long, which is not enough for an article (Please read Information for Authors). Thus, my comments from now on are for Communication type of publication.

If the authors want to change the manuscript to article, they need to add further analysis and explanations, thus it would need a new submission.

Few comments on the submitted manuscript

-The title should read “A machine learning method for predicting vegetation index in China”.

In line 57- I am not sure you investigated environmental factor that influence the vegetation growth. You have done a feature selection analysis, which is highly dependent on the machine learning (ML) method. I do not believe that the feature selection is an investigation of the physical factors that explain the vegetation growth, it is a simple analysis of the variable´s weight within the methodology structure (which were provided arbitrarily by the authors)

In line 61- why XGBoost?

In line 139 –. Why the SPEI index? It is a fairly new index … it has not been world-wide adopted as a standard. However, I do think it is a good drought index, but would like to ready why “this and not that”.

In Results –

---I would like to have an analysis by vegetation type, not only by climate type.

In a short communication, pages are limited and not everything can be shown, but at least should be clearly mentioned what you understand by season.

In Discussion –

 ---In line 239, I do not understand the "the seasonal analysis" with annual NDVI modelling. Would “the season” be related to the ENSO index?  Considering that the phenology, with monthly NDVI, is not studied here, I would like to read a discussion about the plant season modelling.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

The authors have improved the communication paper. The analysis by land cover class was incorporated and my questions have been answers.