Mapping Cropland Abandonment in Mountainous Areas in China Using the Google Earth Engine Platform

Round 1

Reviewer 1 Report

Dear authors, in general, this manuscript has a qualified, didactic and organized structure.

Understanding the behavior and pattern of distribution of crop areas, in space and time on perspectives of the present, past and future, becomes fundamental for adjustments in the planning of public policies and management in making effective and sustainable decisions, especially regarding to water resources.

The present study has the potential to attract both the interest of national and international readers, being extremely suitable for the present journal. Therefore, the contribution and differential of the research is evident.

The need for simple observations for this manuscript is highlighted:

1. I emphasize that all highlights in yellow deserve attention;

2. In the introduction, it is good to highlight the NDVI, mainly its importance for this type of monitoring, and also to describe in the methodology the main wavelengths of the bands used to calculate the NDVI;

3. Carefully observe the numbering of the figures, which are confused in their current order. The sequential numbering must be corrected;

4. In the conclusions, it is worth highlighting the efficiency of the algorithm and its importance for future studies.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The article focuses on mapping abandoned cropland in China and proposes an algorithm based on vegetation succession with different scenarios. The method is based on detection and classification (CCDC) with Landsat time series, and NDVI-based harmonic analysis. Overall, the paper is well written and structured, but I recommend a review on English usage (see example of issues below). I also mention a few other issues that need to be clarified and edited from the methods, data, and figures (see below). The validation procedure is somewhat disappointing in the sense that it was not based on independent data (preferably from the field).

Introduction:

Line 33: “Cropland, a primary food source, is the foundation of the sustainability and development of human society.” I am not sure what “sustainability” means here please clarify. There are many types of agriculture and not all are sustainable.

Line 36: is it expected that the population increases at the same rate as cropland? This sentence does not mention the fact that productive has increased as well through intensification and other forms of food. It is not only area that matters, but also production, so this argument is incomplete.

Line 44: There is also less cropland per capita due to the aggregation of cropland fields and large corporations, so again, I do not fully agree with this argument.

Line 53: Please add reference year for the FAO citation.

Line 57: are authors using the term large-scale as in a large extension or are you using the term scale in the geographical sense? Please clearly and use the term “scale” as is geographical definition. Large scale in this case would be high resolution/detail.

Line 80: explain why MODIS is less suitable to map smallholder agriculture.

Data

Line 150 please explain how this issue with the metadata has been addressed.

Table 1. please explain what you mean by “union of cropland”.

 Methods

Line 214 , please fix “area weighed” the sentence seems incomplete.

Please add reference for the NDVI procedure.

Please fix all the instances of “Error! Reference source not found” it is not professional to submit the manuscript like this.

Line 296, nice figure, I do not see much difference between 1 and 2. Please explain how you accommodate for transitions that did not start between the years 85 to 95. In other words, explain how you are (or not) accommodating for transitions occurring later in time.

Fig. 3. I do not fully understand the difference between scenario 1 and 4. Please clarify.

Also please clarify here again that areas that transition to urban are removed and refer to the explanation of why. Those lands are also important to keep track of as abandoned and “lost” due to development.

English Usage:

The following are examples of issues that need to be addressed in the text. Please revise the text thoroughly and edit as needed.

 Line 12: Though providing high accuracy in local study, it requires quality training sample for land cover classification at each epoch (add S to sample)

Line 55: change the verb tense: After abandonment, the original cropland gradually changed to bare land, grassland, shrubland, or forest, depending on the duration of abandonment and local climatic conditions [7].

Line 59: add “be” before “divided”

Line 120: add “s” to “comprise” so it is “comprises”

 Figure 2

Remove “VALUE” from legend.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

General comments

In this manuscript, authors sought to build an automatic mapping algorithm that identify crop abandoning in a mountain region of China. The algorithm is formed by four main steps: 1) The acquisition of the cropland extent by the union of existing cropland maps. Apparently, authors selected some of these areas as training areas to identify crop abandoning using this cropland extent. 2) Croplands are segmented using the CCDC algorithm. 3) Overall trends and changes are retrieved applying the CCDC algorithm using time series of NDVI. 4) Validation of results are estimated on a stratified random sampling by referencing the NDVI series and satellite images. I have some main concerns in this manuscript:

1) The first step of the algorithm is not right. Authors stacked six different land cover maps (which were generated using different methods, imagery, time, spatial resolution, and even legends) to identify cropland extent and areas of training where crop abandonment is identify. Authors therefore assumed that all these maps are equally perfect, this is a main mistake. These maps are not comparable; therefore, these maps cannot merge to identify training data of crop abandoning. These maps have different errors that authors cannot control (commission and omission); thus, authors could not identify correctly the potential historical cropland extent (which is the first step of the proposed algorithm). Even, authors criticize the methods used to build these maps in the introduction; authors wrote these lines (L71-72. Classification errors in individual land cover map are compounded in the formed land cover change trajectories, which has a low spatial-temporal consistency).

2) The exclusive use of NDVI to classify crop abandoned. Authors must clarify why they only used NDVI to run the CCDC algorithm. Authors had many other bands and spectral indices, such as EVI or SAVI, which can reduce the likelihood of saturation of the NDVI).

3) The writing of the methods is difficult to understand, authors repeat or lack important information. For instance, authors must identify clearly the obtention of the variables (response and predictors) of the Random Forest modeling. Also, they must explain the application of the Random Forest modeling (selection of predictors and tuning of parameters).

4) Although I am not native in English, I detected problems related to the use of non-academic writing or English mistakes.

Though the research objective is relevant in the field of Remote Sensing, this manuscript cannot be accepted for publication at the journal Remote Sensing due to the previous problems and others minor issues that authors can see below. My advice to the authors is to review their analyses, specially the first step of the algorithm, and carefully rewriting a new manuscript.

Other comments.

Title

Author should reduce the title; I suggest deleting the next part: based on post-abandonment. Thus, the tittle can be Mapping cropland abandonment in mountainous areas of China using Google Earth Engine.

Abstract

In general, the abstract is difficult to understand. Although I am not native in English, I recommend the review of the writing. For instance, the next lines are difficult to read and understand (many commas that not fit well with the conjunction Though):  providing high accuracy in local study, it requires quality training sample for land cover classification at each epoch, which is labor- and time-consuming, especially in regions of smallholder agriculture.

Introduction

Although I am not native in English, I recommend the review of the writing. For instance, it is recommended to avoid the use of the word so (line 41); author can write other options that are more used in academic writing, such as thus, therefore, etc.

L43. Million should be written in plural: millions.

L53-54. Authors need a citation; information of the parenthesis is not enough.

L62-72. All the information of these lines must form one paragraph. Authors also need to improve writing. I suggested some changes, but I found more problems after the word Especially (many commas): See below.

Multi-date classification approaches use machine learning algorithms to obtain a series of land cover maps, which are stacked to form a land cover change trajectory [2,9– 63]. These approaches have two main limitations: 1) Collecting quality training samples for training supervised machine learning model is time and labor consuming and requires local and expert knowledge. Especially for mapping historic abandoned croplands in smallholder agricultural region, in most cases, Landsat imagery is the only source of reference to collect samples, where abandoned croplands are challenging to interpret due to small field size relative to image’s spatial resolution. 2) Classification errors in individual land cover map are compounded in the formed land cover change trajectories, which has a low spatial-temporal consistency.

L93-96. It is difficult to understand the main idea of these lines. I believe the problem is the use of the word “obscuring”. Authors must re-write these lines looking for the better understanding.

L96-97. It is difficult to understand the main idea of these lines. I believe the problem is the use of the word “attractive”. Authors must re-write these lines looking for the better understanding.

Methodology

Main issues of the methodology

I have a strong criticism about the use of the Cropland products (maps) to obtain the potential historical cropland extent (it means the training points). Authors stacked six different land cover products (which were generates using different methods, imagery, time, spatial resolution, and even legend) to identify the areas of crop abandoning. Authors therefore are assuming that all these maps are equally perfect, this is a main mistake. These maps are not comparable; therefore, these maps cannot merge to identify training points of crop abandoning. These maps have different errors that authors cannot control; thus, authors could not identify correctly the potential historical cropland extent (which is the first step of the proposed algorithm). Even, authors criticize the methods used to build these maps in the introduction; Authors wrote these lines (L71-72. Classification errors in individual land cover map are compounded in the formed land cover change trajectories, which has a low spatial-temporal consistency).

A second issue of the applied methodology is the only use of NDVI; authors must clarify why they only used NDVI (they had many other bands and spectral indices, such as EVI or SAVI, which reduce the likelihood of saturation problems of the NDVI) to run the CCDC algorithm.

Other problems of the methodology.

L148-150. The use of the phrase “Error! Reference source not found” is not understandable. I suggest 1) a better explanation and 2) the use of an acronym (“Error! Reference source not found” seems like the text that software produced, that is, this type of phrase is not appropriate for a research paper. The phrase “Error! Reference source not found” is used in several parts of the manuscripts, so authors must change for the acronym.

L153. This acronym, FROM_GLC10, has not been explained in the manuscript. I assume FROM_GLC10 is a landcover product but authors need to explain when an acronym is written by first time in the manuscript. I found similar problems when other acronyms were used, I highly recommend the review of all acronyms and their explanations in the manuscript.

L223-225. Authors stated that each landcover map has its own omission error for the crop classification; thus, authors reduced the omission (false negatives or error generated when crop areas are classified as another landcover) by unifying the croplands of the multiple landcover products. I agree that omissions can be reduced applying this method if the maps are comparable; the problem is that these maps are not comparable. Authors cannot control this map errors. What about the commission (false positives; when areas are classified as crops when they should have been excluded)? Again, authors cannot control the map errors because these maps are not comparable to the point of identifying training points.

L352-354. The Google Earth Engine tool developed by Arevalo allows the estimation of the CCDC statistics for the bands and some vegetation indices. Why authors only used NDVI?

Results

L374-380. This result does not have support. Again, Potapov map and CLCD can be not merged to determined areas where crops were abandoned.

Discussion

The discussion is poor. As I mentioned in this review, authors should further explain the validity of these assumptions in the methodology and expand the possible effects of these assumptions in here, the discussion.

Author Response

Please see the attachment.

Author Response File: Author Response.docx