Mapping Cropland Abandonment in the Cloudy Hilly Regions Surrounding the Southwest Basin of China

Round 1

Reviewer 1 Report

The paper explored the spatiotemporal pattern of the cropland abandonment in the study area, and analyzed its driving factors. It is an interesting topic. However, there are still some problems in the research methods and results. Therefore, at present it cannot be accepted in the present form.

However, here are some suggestions for the authors.

1.      Why use Landsat 8 dataset? It has less temporal and spatial resolution than Sentinel-2. In this paper, the spatial resolution of Landsat 8 was resampled from 30 m to 10 m. What is the scale conversion method? How to validate the sampling accuracy?

2.      Line 227-230：” two distinct multi-year cropland abandonment trajectories are depicted: ‘Abandonment to Shrubland’ (transitioning from farmland to shrubland) and ‘Abandonment to Forestland’ (transitioning from farmland to forest)” In this paper, the conversion of farmland to forestland is also defined as cropland abandonment. However, the national policy of "returning farmland to forest" has reduced the area of farmland, which cannot be counted as cropland abandonment.

3.      Line 363-367:” As the distance from the river increases, the abandoned cropland area gradually decreases (Figure 7c). The most concentrated abandoned cropland lies within the range of 0 to 250 meters from the river” Crop cultivation requires water resources, so it is generally believed that the further away from the river, the abandonment rate may increase. But the results in the paper are the opposite, which is interesting, but the paper doesn't explain in detail why.

4.      Figure 7 is difficult to understand. For example, Figure 7c shows that the abandoned area of 0-250 m is larger than that of 250-500m. Therefore, the abandonment rate of the latter should be smaller than that of the former, but the line chart in the figure shows an upward trend. The same problem appears in Figure 7d.

Moderate editing of English language required

Author Response

Response to reviewers

Dear Reviewer:

Thank you for your letter and for the reviewer’s comments concerning our manuscript (ID: land-2974236). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our research. We have studied the comments carefully and made corrections, which we hope meet with approval. The reviewer comments are laid out below in italicized font and specific concerns have been numbered. Our response is given in normal font and changes/additions to the manuscript are given in the green text. The corrections in the paper and the responses to the reviewer’s comments are as follows.

(1) Why use Landsat 8 dataset? It has less temporal and spatial resolution than Sentinel-2. In this paper, the spatial resolution of Landsat 8 was resampled from 30 m to 10 m. What is the scale conversion method? How to validate the sampling accuracy?

Response 1: Thank you very much for your suggestions. The detailed explanation is as follows.

Firstly, the fusion of Sentinel-2 and Landsat data is feasible and increasingly adopted in research. Existing studies have demonstrated that Sentinel-2 and Landsat share similar wavelengths and utilize the same geographic coordinate system, offering an excellent opportunity to integrate data from these two satellite sensors. Please refer to the paper titled "Fusion of Landsat 8 OLI and Sentinel-2 MSI data" for a comprehensive exploration of the integration of these two satellite datasets.

Secondly, Landsat imagery was selected for this study to address issues related to cloud cover in the study area. The region experiences frequent cloud cover and rainfall throughout the year, making land use classification challenging with single-source remote sensing imagery. It is difficult to obtain cloud-free images consistently. Hence, a multi-source data fusion approach was employed. To match the resolution of Sentinel-2, Landsat images were resampled to 10 meters using the nearest neighbor resampling method in Google Earth Engine (GEE). When multiple optical sensors provide cloud-free pixels at a specific location and time, pixels with the highest values are prioritized. Landsat data is primarily utilized to fill in gaps in cloud cover within the Sentinel-2 time series，see lines 185-192 for details. Additionally, the average contribution of optical satellite products was computed to create satellite image mosaics. Each mosaic incorporates Sentinel-2 images covering 85-95% of the study area, ensuring accuracy in subsequent classification tasks.

Finally, during image classification, feature selection was conducted to include only the top-performing 30 features each year, and Sentinel-1 index were also considered, which enhances the accuracy of the classification as well. For details, see lines 243-247 and Figure 4.

(2) Line 227-230：” two distinct multi-year cropland abandonment trajectories are depicted: ‘Abandonment to Shrubland’ (transitioning from farmland to shrubland) and ‘Abandonment to Forestland’ (transitioning from farmland to forest)” In this paper, the conversion of farmland to forestland is also defined as cropland abandonment. However, the national policy of "returning farmland to forest" has reduced the area of farmland, which cannot be counted as cropland abandonment.

Response 2: Thank you very much for your suggestions. The detailed explanation is as follows.

Firstly, there are ongoing debates surrounding the definition of abandoned cropland and whether policies such as returning cropland to forests should be categorized as abandonment. Certain research classifies policy-driven conversions of cropland to forests, like the Grain for Green Program, as induced abandonment. Please refer to the paper titled " Mapping Cropland Abandonment in Mountainous Areas Using an Annual Land-Use Trajectory ". Currently, the prevailing methods for detecting abandoned cropland rely primarily on land use change detection. Given that returning cropland to forests is a long-term policy, a direct transition from cropland to forest is unrealistic and invariably involves an intermediate shrubland stage. This poses a challenge in distinguishing whether such conversions are driven by human factors or natural succession, becoming a universally present dilemma.

Secondly, the second trajectory proposed in this study refers to a transition from cropland to shrubs, ultimately evolving into a forest, as detailed in lines 257-259. In our five-year monitoring of abandoned cropland, we observed that most abandoned cropland followed the first trajectory, while the second trajectory was relatively rare, accounting for only a small fraction of the monitored data. Additionally, the Grain for Green Program primarily occurred in the past 20 years (1999-2019), whereas this study began in 2018, making the impact of this policy relatively minor on our research.

(3) Line 363-367:” As the distance from the river increases, the abandoned cropland area gradually decreases (Figure 7c). The most concentrated abandoned cropland lies within the range of 0 to 250 meters from the river” Crop cultivation requires water resources, so it is generally believed that the further away from the river, the abandonment rate may increase. But the results in the paper are the opposite, which is interesting, but the paper doesn't explain in detail why.

Response 3: Thank you very much for your suggestions. The detailed explanation is as follows.

Firstly, we believe that it is not entirely reasonable to solely focus on the area of abandoned cropland. This is because, given the large base of cropland, the area of abandoned cropland naturally tends to be relatively high. Most of the cropland in Mingshan County is distributed in areas close to rivers, which explains why the highest concentration of abandoned cropland is found within the 0-250m range. Additionally, the Ya'an area enjoys a humid and rainy climate with an annual average rainfall of 1455.1 millimeters. This may expose cropland too close to rivers to the risk of flood erosion, potentially damaging the soil structure and rendering the land unsuitable for further cultivation, leading to abandonment. A similar conclusion has been drawn by related research, indicating that approximately 48% of abandoned cropland is located in areas with relatively favorable locational characteristics, particularly in terms of access to water and roads. Please refer to the paper titled "The role of harmonized Landsat Sentinel-2 (HLS) products to reveal multiple trajectories and determinants of cropland abandonment in subtropical mountainous areas".

Secondly, the abandonment rate referred to here is the proportion of abandoned cropland area within a certain interval to the cropland area in that interval，see lines 423-426 and Figure 10 for details. We adopt this definition for subsequent analysis. The results indicate that the abandonment rate exhibits a fluctuating trend. Furthermore, the results from the geodetector suggest that the distance from rivers has the weakest explanatory power for cultivated land abandonment, which indirectly validates our findings. Prior research has indicated that irrigation, as a crucial aspect of agricultural production, can significantly enhance grain yields. However, water resources may not solely depend on rivers but also on natural rainfall or channel irrigation. Therefore, being farther away from rivers does not necessarily lead to easier abandonment, as detailed in lines 559-564.

(4) Figure 7 is difficult to understand. For example, Figure 7c shows that the abandoned area of 0-250 m is larger than that of 250-500m. Therefore, the abandonment rate of the latter should be smaller than that of the former, but the line chart in the figure shows an upward trend. The same problem appears in Figure 7d.

Response 4: Modified. We have revised the relevant expressions for better understanding. For details, see lines 423-426 and Figure 10.

The abandonment rate referred to here is the proportion of abandoned cropland area within a certain interval to the cropland area in that interval. Therefore, although the area of abandoned cropland within 0-250m is larger than that within 250-500m, the cropland area within 0-250m is smaller than that within 250-500m. Consequently, the abandonment rate is still higher in the 0-250m range compared to the 250-500m range, exhibiting an upward trend. Hence, it is possible for the area of abandoned cropland to be significant, yet the abandonment rate may not necessarily be high.

In all, we found the reviewer’s comments quite helpful, and we revised our paper point-by-point. We appreciated for reviewers’ warm work earnestly, and hope that the correction will meet with approval. Once again, thank you very much for your comments and suggestions.

Yours sincerely,

Wei yali

April 23, 2024.

Reviewer 2 Report

This paper aims to map crop abandonment in a hilly and cloudy region in southwest China based on multisource remote sensing (Landsat 8, Sentinel-1, and Sentinel-2) data and Random Forest algorithm over the period 2018-2022. Elevation and slope were the main driving factors for crop abandonment due to the dominant small-scale, low-intensity agricultural systems in hilly terrains. I enjoyed reading the submission, it is well-written, easy to read, and quite clever in all aspects. My comments are listed below:

1. In the Introduction section, I suggest defining what the authors are considering crop abandonment, since this concept is region or country dependent. For example, if a crop land is not cultivated in the following year, it is not necessarily abandonment, it can be fallow. Basically, the discussion of crop abandonment found in L223-230 should be moved to the Introduction section.

2. I suggest adding a table to the manuscript showing all optical and SAR indices used in the study. In this table, please identify the short names of the indices, their meanings, equations, and citations. In this way, all seven first equations can be moved to this table.

3. In Figure 2, there is an activity named cart classification that I believe was not used in the methodological approach. Please, check it.

4. How many samples were used to perform accuracy analysis of annual LULC map?

5. In Results, it would be nice if the authors present a table quantifying the area (either in hectares or percentage) occupied by each LULC class in the study area for six years.

Author Response

Response to reviewers

Dear Reviewer:

Thank you for your letter and for the reviewer’s comments concerning our manuscript (ID: land-2974236). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our research. We have studied the comments carefully and made corrections, which we hope meet with approval. The reviewer comments are laid out below in italicized font and specific concerns have been numbered. Our response is given in normal font and changes/additions to the manuscript are given in the blue text. The corrections in the paper and the responses to the reviewer’s comments are as follows.

(1) In the Introduction section, I suggest defining what the authors are considering crop abandonment, since this concept is region or country dependent. For example, if a cropland is not cultivated in the following year, it is not necessarily abandonment, it can be fallow. Basically, the discussion of crop abandonment found in L223-230 should be moved to the Introduction section.

Response 1: Modified. We have incorporated the definition of cropland abandonment into the Introduction section, as shown in lines 107-109.

(2)  I suggest adding a table to the manuscript showing all optical and SAR indices used in the study. In this table, please identify the short names of the indices, their meanings, equations, and citations. In this way, all seven first equations can be moved to this table.

Response 2: Modified. We have added a table showing all optical and SAR indices used in the study. For details, see line 207 and Table 2.

(3) In Figure 2, there is an activity named cart classification that I believe was not used in the methodological approach. Please, check it.

Response 3:  Thank you for your suggestions. Decision tree classification encompasses CART classification. In order to maintain consistency, we have made modifications to the figure, as detailed in lines 154-155 and Figure 2.

(4) How many samples were used to perform accuracy analysis of annual LULC map?

Response 4: Modified. We have included specific details regarding the number of samples utilized for the accuracy analysis of the annual Land Use/Land Cover (LULC) map. For details, see lines 265-269.

(5) In Results, it would be nice if the authors present a table quantifying the area (either in hectares or percentage) occupied by each LULC class in the study area for six years.

Response 5: Modified. We have added a table quantifying the area (in percentage) occupied by each LULC class in the study area for five years. For details, see line 337 and Table 4.

In all, we found the reviewer’s comments quite helpful, and we revised our paper point-by-point. We appreciated for reviewers’ warm work earnestly, and hope that the correction will meet with approval. Once again, thank you very much for your comments and suggestions.

Yours sincerely,

Wei yali

April 23, 2024.

Reviewer 3 Report

Comments can be found in the attached PDF. All the very best!

Comments for author File: Comments.pdf

Author Response

Response to reviewers

Dear Reviewer:

Thank you for your letter and for the reviewer’s comments concerning our manuscript (ID: land-2974236). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our research. We have studied the comments carefully and made corrections, which we hope meet with approval. The reviewer comments are laid out below in italicized font and specific concerns have been numbered. Our response is given in normal font and changes/additions to the manuscript are given in the red text. The corrections in the paper and the responses to the reviewer’s comments are as follows.

(1) In the start of the paragraph, authors already mentioned the factors contributing to cropland abandonment in hilly and mountainous terrain. Why do they want to do that again? For suppose, authors found the patterns and driving factors. What is its importance, why is it important especially for this region? Why international researchers/readers, land budget managers or local governments should be interested in this study. How or what will study contribute to the current knowledge on cropland abandonment or its driving factors? All these important topics are missing in the introduction. Instead, authors just wrote that cropland abandonment is increasing in various reasons, so we want to identify the factors using remote sensing. It is not convincing and so, the entire introduction needs revision, rewriting and restructuring focusing on the aspects mentioned above.

Response 1: Modified. We have revised the introduction focusing on the aspects mentioned above, as detailed in lines 30-79.

(2) The authors just mapped it for 4 years. It cannot be called as long-time series.

Response 2: Modified. We have revised the term " long-time series " to simply " time series ", see line 116 for details.

(3) What do authors mean by auxiliary data? What samples were collected and how many were collected? How authors identified these samples from the imagery?

Response 3: Thank you for your suggestions. Auxiliary data includes Google Earth images and various land use products, as outlined in lines 145-152. These images are solely used to provide reference for selecting training and validation samples and are not directly utilized for sampling purposes.

(4) In figure, stable samples were written twice in the same box. please check and fix accordingly. Also, all short forms of bands or indices needs to be elaborated as a footnote in the figure caption.

Response 4: Modified. We have revised the figure and added a footnote in the figure caption, see lines 154-155 and Figure 2 for details.

(5) What seasonal or months were used for identifying the cropland abandonment? There will be cropping seasons in peak growing seasons and cover crops in the winter or non-growing seasons. How are these issues handled? How crop residue or cover crops or primary plantation was differentiated while classifying?

Response 5: Thank you for your suggestions. The detailed explanation is as follows.

Regarding the first question, we computed the monthly summary metrics each year, spanning from January to December. These encompassed an array of values, including those for the blue, red, NIR, and SWIR bands, along with NDVI, BSI, VV, VH, CR, DPSVIm, Pol, RVIm, Entropy, and Inertia. Furthermore, to aid in the identification of abandoned cropland, we computed the annual minimum, maximum, amplitude, and standard deviation for both NDVI and VV. For details, see lines 192-206 and Figure 2.

Regarding the second issue, firstly, concerning NDVI. During the peak growing season, the NDVI values of cropland tend to significantly increase. However, compared to shrubs and forests, cropland exhibits relatively lower NDVI values. This is primarily because the vegetation density and hierarchy of cropland are typically not as rich as those of shrubs and forests, resulting in differential NDVI peaks. During the winter or non-growing season, when cropland is covered by crops or lies fallow, NDVI values decrease. However, this decrease is seasonal and usually does not drop to very low levels. Importantly, the NDVI of cropland may exhibit significant changes throughout the year, primarily due to differences in NDVI values between the growing and non-growing seasons. This seasonal pattern of NDVI variation is typical for cropland and can therefore serve as a key characteristic for its identification. Simultaneously, concerning texture features, cropland exhibits a regular, orderly texture; shrubs present a rougher and irregular texture; while forests display a continuous, dense texture.

Regarding the third question, there exist distinct differences in NDVI (Normalized Difference Vegetation Index) among crop residues, cover crops, and primary forests. For crop residues, their NDVI values are relatively low. This is because crop residues primarily consist of harvested or wilted crop parts with low biological activity, resulting in a different proportion of reflected near-infrared and red light compared to active vegetation. Cover crops, on the other hand, exhibit higher NDVI values, especially when they are in their vigorous growth stage. Since cover crops retain biological activity, they can effectively absorb red light and reflect near-infrared light, leading to elevated NDVI values. Primary forests, typically characterized by dense vegetative cover, often have the highest NDVI values. The well-grown trees and other vegetation in primary forests form a complex ecosystem that efficiently absorbs red light and reflects near-infrared light, resulting in high NDVI.

(6) Are these samples per year? If these are for all the years combined, how many are per year? Also, the land cover state might change from year to year. How was this verified?

Response 6: Thank you for your suggestions. These samples encompass data from all years combined. Additionally, we have introduced a table displaying the distribution of training samples across land-cover classes and years, as shown in line 235 and Table 3.

To generate these training samples, we utilized a decision tree-based method to establish a consistent set of land use types. The process involved three steps: first, labeling stable and changing samples of land use types; second, identifying stable land use regions; and third, creating a stable land use sample set by randomly generating a specific number of sample points for different land use categories within stable regions each year, in conjunction with existing land use data. Subsequently, the sample point data undergoes verification and screening, resulting in the final validated dataset. For details, see lines 208-234 and Figure 3.

(7) I suggest authors plot response curves to explain the contribution of variables in classifying the map using RF model.

Response 7: Modified. We have added a figure to explain the contribution of variables in classifying the map using RF model, as shown in line 250 and Figure 4. Additionally, we have included relevant descriptions, as detailed in lines 243-247.

(8) 10m land cover maps were available from ESRI Sentinel. Try validating your results with the already available datasets.

Response 8: Modified. We have included a figure comparing the classification results of this study with the global ESRI 2020 land cover dataset, as well as high-resolution images accessible on Google Earth, across two distinct regions. For details, see lines 332-336,341-143 and Figure 6.

(9) Can you plot them spatially and show, where these points are located.

Response 9: Modified. We have added a figure showing the spatial distribution map along with on-site photographs, both derived from the 2022 field survey data. For details, see lines 351-352 and Figure 7.

(10) 2018-2021 has shades of green. We can’t see or distinguish anything from these colors. Please provide 4 different panels for four years of abandonment. This will help us understand the patterns of abandonment.

Response 10: Modified. We have provided four distinct panels, each representing a year of abandonment, to assist in understanding the patterns of abandonment. For details, see lines 353-355 and Figure 8.

(11) Don’t use shades of green. Please use 4 different colors to help readers distinguish the patterns easily.

Response 11: Modified. We have employed four distinct colors to facilitate readers in easily distinguishing the patterns. For details, see lines 423-426 and Figure 10.

(12) Please use radar charts instead of line graphs to show the intervals for each factor. Its hard to understand from these line charts.

Response 12: Modified. We have substituted the line graphs with radar charts for a more comprehensive and visual representation. For details, see lines 444-445 and Figure 11.

(13) One good module of geodetector is that it can derive the optimal ranges of the considered factors. The authors haven't described the optimal range results from the model. These results can help suggest the thresholds that need to be maintained for mitigating cropland abandonment.

Response 13: Modified. We have described the optimal range results from the model, see lines 428-443 for details.

(14) Add/discuss about the implications, future scope and uncertainties of the study in the discussion.

Response 14: Modified. We have added the implications, see lines 576-603 for details. The limitations and prospects of this study have been pointed out, as detailed in lines 604-635.

(15) I haven’t found any new framework. Its already well established and people have published several papers.

Response 15: Modified. We have revised the term " a new framework " to simply " a framework ", see lines 504 for details.

In all, we found the reviewer’s comments quite helpful, and we revised our paper point-by-point. We appreciated for reviewers’ warm work earnestly, and hope that the correction will meet with approval. Once again, thank you very much for your comments and suggestions.

Yours sincerely,

Wei yali

April 23, 2024.

Round 2

Reviewer 1 Report

Thanks to the authors for carefully addressed all the revisions. The manuscript was improved substantially since its last version. The paper needs some minor edition to polish the language and avoid grammatical mistakes. Other than that, I believe is ready for going to publication.

Minor editing of English language required

Reviewer 3 Report

The authors have incorporated most of my suggestions, and the manuscript is in good shape for publication.