Long Time Series High-Quality and High-Consistency Land Cover Map** Based on Machine Learning Method at Heihe River Basin

Round 1

Reviewer 1 Report

Basic and laborious article. Congrats.

Author Response

Thank you very much for the time you spent reviewing our manuscript and for approving our job.

Thank you for your recognition of our work!

Reviewer 2 Report

The manuscript, entitled "Long time series high-quality and high-consistency land cover map** based on machine learning method at Heihe River Basin" (by Zhong, B. et al.) proposes a methodology for producing long-term land cover (LC) data time series (1986-2015) with high accuracy and consistency. Applied to to Heihe River Basin (China) and for the time period of 1986-2015, the proposed methodology takes the advantages of time series of different Landsat images and the available high-quality LC datasets from LCMM method. The LCMM-derived LC datasets were used for the accurate determination of training samples and the random forest (RF) classifier was trained with the collected samples of each examined year leading eventually to the production of the targeted LC maps. These maps were then validated in terms of accuracy metrics (kappa coefficient and overall accuracy) and their cross-comparison with GlobeLand30 LC datasets.

General remarks

The manuscript deals with a topic of international interest. The applied methodology and its output results are certainly valuable. In general, I find the structure and content of the manuscript being acceptable. Just a few comments and suggestions as minor revisions are provided below so that the authors take them under consideration.

Minor suggestions

Line 40: “a” instead of “an”.

Line 72: I would suggest you to rephrase this sentence as “In order to make long time series of land cover datasets of high spatial resolution”.

Line 77: “provides” instead of “provide”.

Line 78: What 16 refers to? Days? Clarify it.

Line 94: Delete “a”.

Table 2: The heading of 3rd column would be more appropriate as “Number of scenes”, and not as “Description”.

Line 163: Relation confusion due to different order numbering of Tables. The Tables in the manuscript are ordered by 1, 2, 3 etc., but in the manuscript’s text they are cited by I, II, III etc. Correct it and consider about this correction in the entire extent of text.

Line 177: Delete “but”.

Line 254: What do you mean using the term of “object”? Land cover feature/unit? Clarify it.

Lines 324-326: Rephrase this sentence.

Author Response

Thank you very much for the time you spent reviewing our manuscript and for approving our job. Thank you very much for your constructive suggestions.

Reviewer 3 Report

Dear Authors,

The manuscript is very interesting, especially in the part of the model that refers to the sampling strategy. However, you should pay more attention in the Introduction section to the references overview related to the land cover map**, time series, and sampling strategy. In the Introduction section, you refer to 15 papers, of which only one has been published in the last five years (in 2019). Considering that the presented topic has been very popular in the last five years, as well as the number of published papers on this topic, an up-to-date overview of the existing published research should be presented. Without a quality review of the literature, it is difficult to conclude what exactly is the contribution of this paper.

The whole area is presented using 400,000 pixels, while for the validation only approximately 1,000 pixels were used. It should be discussed how this disproportion influenced the realistic assessment of land cover map**.

Line 270: Shouldn't it be in section 2.4.2.?

Lines 306 - 309: Is there any particular reason why those values of RF parameters are used? This should be explained to readers.

A few minor issues:

• The font in the figures should be adjusted to the text font. The text in the figures is not clear.
• Line 119: “Table II” correct to “Table 2.”.
• Lines 227 and 228: “Table III” correct to “Table 3.”.
• Line 321: “Table IV” correct to “Table 4.”.

I hope that you find this review helpful, as it is meant to be.

Author Response

Thank you very much for the time you spent reviewing our manuscript and for approving our job.

The Introduction section has been updated and some new references were added.

Thank you very much for your constructive suggestions.

Round 2

Reviewer 3 Report

Although the comments and suggestions in round 1. were very easy to correct or give answers to, the authors failed to pay enough attention to address them. For example, two reviewers suggested correcting the table references, but there are still several omissions:

• Line 137 (round 1. line 119): “Table II” correct to “Table 2.”.
• Lines 245 and 246 (round 1. lines 227 and 228): “Table III” correct to “Table 3.”.
• Line 338 (round 1. line 321): “Table IV” correct to “Table 4.”.

As indicated in the first round of reviewing process “The font in the figures should be adjusted to the text font.” However, the text in figures 4,6, 7, and 8 is still unreadable.

For some comments and questions, the authors didn’t provide any answers:

• The whole area is presented using 400,000 pixels, while for the validation only approximately 1,000 pixels were used. It should be discussed how this disproportion influenced the realistic assessment of land cover map**.
• Line 288 (round 1. line 270): Shouldn't it be in section 2.4.2.?
• Lines 324 – 327 (round 1. lines 306 – 309): Is there any particular reason why those values of RF parameters are used? This should be explained to readers.

 

Comments and suggestions for added text in Introduction:

• Line 90: “cases [22, 23]. subsequently” correct to “cases [22, 23]. Subsequently”.
• The authors should match terms in the newly written text with the previously written text. Correct “landcover” to “land cover”.

 

Author Response

We are so sorry for the wrong copy in round 1. Therefore, the response to these questions is omitted. Thank you very much for the time you spent reviewing our manuscript. Here is our new reply.

(1) Inconsistent case of table serial number has been corrected.

(2) These figures are updated. The font in the figures has been changed to be the same as that in the text.

(3) We selected about 1,000 pixels for the following reasons:

Firstly, although the whole area is presented using 400,000 pixels, about 85% of these pixels is pure bare land. That is, only about 60,000 pixels are other types. We selected about 200 pixels to validate bare land type and about 800 pixels to validate other types. Without considering the bare land type, the selection proportion of validation samples of other types is about 1.3%.

Secondly, in this research, although the validation sample are chosen randomly, each pixel type is determined by manual inspection, which is a labor-intensive job.

Last but not least, the distribution of land coverage types is patchy in this area. Each pixel can represent a class object. All the 1,000 pixels are representative.

(4) It has been changed to “section 2.4.2”.

(5) These parameters (the number of trees, the number of variables per split, the min leaf population, and the bag fraction) are the necessary input of the RF method. Among the four parameters, the number of trees is set to 100 because it proved can achieve better result when considering the classification accuracy and efficiency. The other three parameters are default.

(6) “subsequently” has been changed to “Subsequently”. And, “landcover” has been changed to “land cover”.