Automatic Extraction for Land Parcels Based on Multi-Scale Segmentation
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe article explores the use of multi-scale segmentation and image processing approaches for extracting land parcel units, and identifying the features of land use or land features in Xinghai County of China. The topic itself is interesting and has some scientific values. However, many parts of the paper have to be brushed up, and further technical details have to be provided in different parts of the manuscript.
Major Problem
(1) The goal of the current study (Lines 23-30 of Abstract) is quite vague and unclear, the authors should present the goals and contributions of this manuscript / study in a more specific manner.
(2) Line 49: "leading to analysis results lacking in geographical spatial meaning" - please provide the insufficiency in more details.
(3) Lines 54-57: How about other approaches? The same problem may still persist in other methods for representing topographical elements.
(4) Line 58: "universal connection" - it depends on the type of connections that one wish to establish / investigate, please be more specific.
(5) Lines 65-68: The authors have outlined several ML-based algorithms, like SVM, logistic regression and deep learning, please discuss in more details, especially their usage and application in land use classification, land deformation etc. Some references that the authors should include are as follows:
https://www.mdpi.com/2072-4292/13/16/3337
https://www.sciencedirect.com/science/article/abs/pii/S0341816218300791
https://www.mdpi.com/2072-4292/9/10/1031
https://www.mdpi.com/2072-4292/12/17/2817
(6) Lines 84-85: Here, many important applications are mentioned, but what is the main goal of this study? As the authors have mentioned, the features of land and the methodology will depend on spatial scales, therefore, please focus more on your study.
(7) Lines 105-109: It's very good to point out the uncertainty of different datasets within this study. However, what are the impacts and effects of these uncertainties?
(8) Line 118: "scale threshold" - what is considered to be a reasonable scale threshold? Please discuss here and in different parts of the manuscript.
(9) Section 2.3: Any validation or cross-checking were done for the multi-scale segmentation process?
(10) Line 138: maximum number of iterations - what is the upper bound of maximum number of iterations in this study? Any numerical criterion / stopping criterion?
(11) Lines 137-141: Any constraints in such minimization problem?
(12) Equation (6): Please explain whether the quantities 10.6 and 8.15 are obtained from experiments or analytical means?
(13) Please explain the rationale and implication of the calculations in Section 2.4.4, for example, what does mean and variance symbolize?
(14) Equation (15) Why k variables? What are the k variables? Or it's actually a general setting?
(15) Line 226: What are the grid operations?
(16) Figure 1: Please include more descriptions, especially about the scale spaces within your manuscript.
(17) Section 3.2 - please describe the details of the datasets
(18) Line 288: "After trials" - what are the ranges of those parameters and variables that you wish to conduct trial?
(19) Line 293: "Hartigan-Wong" algorithm - why this algorithm was adopted? From experimental conclusion or there involves theoretical basis?
(20) Lines 324-325: Please elaborate more - what is the use of the statistical law? Why uniformity of such law is important in the study?
(21) Figure 6: Should extract a particular spatial region as an example and include more description and explanation of land features detected in the main text.
(22) Section 4.2: should include some numerical values such as ratio, or other statistical quantities?
(23) Line 356: The threshold was set to be 19 in the study, but what if the threshold here changes? What's the effect towards the results?
(24) Figures 11, 12, 13 and 14: The authors should provide more descriptions and statistical comparison in the main text.
(25) Lines 431-435: Please explain the optimality of these values. What if they are perturbed / slight changes are imposed?
(26) Lines 447-459: Again, the authors should get more focused on the investigation / case study discussed in this research, but not mentioning something in general.
(27) Lines 481-482: To what extent does PCA enhance the retrieval performance? The authors still couldn't explain the effectiveness and advantages of PCA well.
(28) Lines 489-490: The future goals here are quite vague, please be more specific, and associate with sustainable development of a city / county, the urban land management, and government policies etc.
Minor Problem
(1) Equation (1) Is it absolute value instead of bracket? Is it conducted / calculated in 2-norm?
(2) Line 132: mu_i instead of U_i
(3) Line 136: where
(4) Line 156: Why 3*3 instead of 4*4 or 5*5? Please explain the rationale.
(5) Equation (12): underscript x and y, also there should be some problem with the equation.
(6) Line 300: "similar properties" - what kind of properties?
(7) Line 317: "related" - is it positively related or negatively related?
(8) Line 361: "certain directions" - describe and elaborate
(9) Line 372: "wide range" - what is the range?
(10) Lines 475-482: Please provide some statistical values.
(11) Lines 484-485: Please provide a concrete example of the claim that "accuracy of the boundary is greatly influenced by the data structure"
English Grammatical Errors
Refer to the box on "Comments on the Quality of English Language"
The authors should address aforementioned comments in their revised manuscript. A proper round of correction should be conducted.
Comments on the Quality of English Language
Line 23: different spatial scales
Line 35: successive laws
Line 42: methodological framework that is capable of
Line 49: Second
Line 139: meaning that the sum of squared distances
Line 207: In general, each
Line 226: Fourth
Line 277: K-means
Line 302: to be further analyzed
Line 316-317: (Figures 5 and 6)
Line 409: targets at small watersheds
Line 444: In addition
Line 475: variations of indicator datasets
Line 485: algorithm was run in R
Author Response
(1) The goal of the current study (Lines 23-30 of Abstract) is quite vague and unclear, the authors should present the goals and contributions of this manuscript / study in a more specific manner.
Reply and revision: Thank you very much for the suggestion.
We have revised the results section of the abstract in Line 24-36 to show the goal and contribution of this study more specific.
(1) The Land Parcels extracted using the hydrological analysis method are catchment units centered around rivers, including slopes on both sides of the river. In contrast, multi-scale segmentation extracts regions with similar properties of Land Parcels, enabling the segregation of slopes and channels into independent units. (2) At a classification threshold of 19, multi-scale segmentation divides the study area into five different types of Land Parcels, reflecting the heterogeneity of terrain undulations and their hydrological connections. When the classification threshold is set to 31, the study area is divided into 15 types of Land Parcels, primarily highlighting micro-topographic features. (3) Multi-scale segmentation can merge and categorize areas with the least heterogeneity in Land Parcels, facilitating subsequent statistical analysis. Therefore, mesoscale Land Parcels extracted through multi-scale segmentation are invaluable for analyzing regional earth surface process such as soil erosion, sediment distribution and transportation. Microscale Land Parcels are significantly important for identifying high-risk areas of geological disasters like landslides and collapses.
(2) Line 49: "leading to analysis results lacking in geographical spatial meaning" - please provide the insufficiency in more details.
Reply and revision: Thank you very much for the suggestion.
We have made revisions in Line 56-58 to Additional reasons for the lack of geospatial significance.
leading to an incomplete expression of the spatial variability characteristics or surface processes, and may also contain irrelevant information when describing physical geographic features
(3) Lines 54-57: How about other approaches? The same problem may still persist in other methods for representing topographical elements.
Reply and revision: Thank you very much for the suggestion.
Supplemental lines 69-73, describing the comparison of other methods with hydrologic analysis and grid analysis, And add references.
The delineation of land parcels based on machine learning and deep learning can establish connection between land parcels and physical geographic features, thereby addressing the issue of scale variation and its association with the research object, a challenge present in traditional hydrological units and grid units
(4) Line 58: "universal connection" - it depends on the type of connections that one wish to establish / investigate, please be more specific.
Reply and revision: Thank you very much for the suggestion.
Added lines 66-68 to describe the type requirements that need to be established.
it is challenging to find a universal classification method that can simultaneously satisfy the complexity requirements of multi-faceted or multi-scale research.
(5) Lines 65-68: The authors have outlined several ML-based algorithms, like SVM, logistic regression and deep learning, please discuss in more details, especially their usage and application in land use classification, land deformation etc. Some references that the authors should include are as follows:
https://www.mdpi.com/2072-4292/13/16/3337
https://www.sciencedirect.com/science/article/abs/pii/S0341816218300791
https://www.mdpi.com/2072-4292/9/10/1031
https://www.mdpi.com/2072-4292/12/17/2817
Reply and revision: Thank you very much for the suggestion.
The corresponding references have been added in line 80.
(6) Lines 84-85: Here, many important applications are mentioned, but what is the main goal of this study? As the authors have mentioned, the features of land and the methodology will depend on spatial scales, therefore, please focus more on your study.
Reply and revision: Thank you very much for the suggestion.
We have made revisions in Line 95-99 and focus on the goal of this study more specific:
Based on the principles of multi-scale segmentation and employing hydro-geomorphological characteristic factors such as slope, drainage, and undulations across different spatial scales, this paper proposes an automated method for land parcel extraction. It aims to provide a basis for geomorphological or hydrological modeling, landslide susceptibility, and hazard or risk modeling.
(7) Lines 105-109: It's very good to point out the uncertainty of different datasets within this study. However, what are the impacts and effects of these uncertainties?
Reply and revision: Thank you very much for the suggestion.
Descriptions of data uncertainty and its impact on the research have been added in Line 125-127
These uncertainties is very small as comparing with the 30 meter resolution of the raster dataset and the local elevation difference of terrain undulation which is mostly above 500 meters, thereby would impact the conclusions of the research little.
(8) Line 118: "scale threshold" - what is considered to be a reasonable scale threshold? Please discuss here and in different parts of the manuscript.
Reply and revision: Thank you very much for the suggestion.
Explanations regarding the rationality of threshold values have been included in Line 141-145.
Furthermore, the determination of thresholds is based on the research objectives. When there are more land parcel classifications, the threshold is smaller, leading to reduced internal heterogeneity and smaller parcel areas. Conversely, when there are fewer land parcel classifications, the threshold is larger, resulting in relatively greater internal heterogeneity, larger parcel areas, and more continuity.
(9) Section 2.3: Any validation or cross-checking were done for the multi-scale segmentation process?
Reply and revision: Thank you very much for the suggestion.
Section 2.6 has been added in lines 245-269, detailing the methods for validating land parcels.
In Section 3.6, Lines 389-393 and 399-409, the methods and approaches for land parcel validation have been expanded upon.
Additionally, Sections 4.3.3 and 4.3.4 in lines 513-542, have been enhanced to include evaluations and explanations of multi-scale segmented units at both mesoscale and microscale levels.
(10) Line 138: maximum number of iterations - what is the upper bound of maximum number of iterations in this study? Any numerical criterion / stopping criterion?
Reply and revision: Thank you very much for the suggestion.
In Section 2.3, an explanation of the number of iterations has been added in Line162-164 .
Section 3.5 now includes a detailed description of the iterative process (367-375)
(11) Lines 137-141: Any constraints in such minimization problem?
Reply and revision: Thank you very much for the suggestion.
In Section 2.3, an explanation of the number of iterations has been added in Line162-164 .
Section 3.5 now includes a detailed description of the iterative process (374-375)
(12) Equation (6): Please explain whether the quantities 10.6 and 8.15 are obtained from experiments or analytical means?
Reply and revision: Thank you very much for the suggestion.
The equation is built by previous studies about slope steepness effect on soil erosion. We have added
a new reference citation in line 187.
(13) Please explain the rationale and implication of the calculations in Section 2.4.4, for example, what does mean and variance symbolize?
Reply and revision: Thank you very much for the suggestion.
The expression of the formulas has been revised, and their content has been elucidated in Line197-206 .
First, calculate the cosine and sine values of x and y coordinates.
Second, calculate the mean values of x and y within a 3x3 grid surrounding each cell to determine local change patterns.
Third, calculate the variance of these local changes can more accurately highlight topographical variation characteristics.
Fourth, calculate the coefficient of variation for slope aspect.
(14) Equation (15) Why k variables? What are the k variables? Or it's actually a general setting?
Reply and revision: Thank you very much for the suggestion.
We have added explanations for the variable 'K' and the expression of its formula in Line 232-235 .
It represents a weighted combination of K observed variables, offering the maximum variance explanation for the initial set of variables. The second principal component, also a linear combination of the initial variables, ranks second in terms of variance explanation and is uncorrelated with the first principal component.
(15) Line 226: What are the grid operations?
Reply and revision: Thank you very much for the suggestion.
The process of raster calculation has been further explained in Line 281-285 .
By employing factor loadings, different principal component grids are calculated. This involves multiplying the factor loadings of various indices in the first principal component by the index factors, then summing the resultant grids to form the first principal component grid. The second principal component grid is derived in a similar manner
(16) Figure 1: Please include more descriptions, especially about the scale spaces within your manuscript.
Reply and revision: Thank you very much for the suggestion.
We have imporved Figure 1 and included more descriptions of the process.in line 304
(17) Section 3.2 - please describe the details of the datasets
Reply and revision: Thank you very much for the suggestion.
The description of the calculation process has been expanded in Line 309-316.
Firstly, depression filling is conducted to extract flow directions, calculating the flow direction in each pixel. Secondly, flow accumulation is computed to determine the quantity of water passing through each pixel. Thirdly, various thresholds are attempted to extract river networks, calculating thresholds of 50, 500, 1000, and 5000 in relation to the study area to delineate river networks. Fourthly, the river network structure is extracted using river linking functions, forming a hierarchical structure of the river system. Finally, the catchment area function is used to extract catchment areas for different thresholds.
(18) Line 288: "After trials" - what are the ranges of those parameters and variables that you wish to conduct trial?
Reply and revision: Thank you very much for the suggestion.
The procedure and parameters of the experiment have been further detailed in Line 367-375.
Considering the size of the study area and the representation of segmentation, we set the random sample size for multi-scale segmentation to 1000, meaning that 1000 points are randomly selected from the imagery to initiate the segmentation process. The classification threshold is set as a looping value ranging from 3 to 100 to explore the accuracy of different category classifications. The K-means algorithm's random startup count is set to 500, implying 500 runs with different initial centroid positions. Finally, the iteration count is set to 500, based on considerations of both computational resources and optimal clustering. The parameter settings mentioned above aim to balance the accuracy of computational results and the computational capabilities of the computer.
(19) Line 293: "Hartigan-Wong" algorithm - why this algorithm was adopted? From experimental conclusion or there involves theoretical basis?
Reply and revision: Thank you very much for the suggestion.
Modifications were made to this section of the text in lines 375-380 to explain why the results are considered reasonable.
In terms of model selection, we considered the rationality of the optimization structure based on the objective function. We compared the sum of squares for each point assigned to different clusters, examining all three algorithms, 'Lloyd,' 'MacQueen,' and 'Hartigan-Wong,' available in the Rstoolbox package. We found that the 'Hartigan-Wong' algorithm provides a more reasonable combination of results with land parcels in the imagery.
(20) Lines 324-325: Please elaborate more - what is the use of the statistical law? Why uniformity of such law is important in the study?
Reply and revision: Thank you very much for the suggestion.
The manner of expression has been revised in Line421-423 .
Each catchment area is an independent unit, and there is no attribute field that can categorize catchment areas with similar attributes into the same class.
(21) Figure 6: Should extract a particular spatial region as an example and include more description and explanation of land features detected in the main text.
Reply and revision: Thank you very much for the suggestion.
Figure 6 shows mountainous areas, specifically depicted in Figure 4 on line 384.
An additional Table 1 has been added on line 387 to further illustrate the details of typical areas.
(22) Section 4.2: should include some numerical values such as ratio, or other statistical quantities?
Reply and revision: Thank you very much for the suggestion.
The mean, standard deviation, and extreme values for the area of two types of land parcels extracted through multi-scale segmentation at both mesoscale and microscale levels have been added in lines 438-443 and lines 445-447.
(23) Line 356: The threshold was set to be 19 in the study, but what if the threshold here changes? What's the effect towards the results?
Reply and revision: Thank you very much for the suggestion.
In the analysis within Section 4.2, detailed explanations were provided in lines 435-436.
the more sensitive it becomes to changes in terrain slope, Easily identification land parcels at smaller spatial scales
(24) Figures 11, 12, 13 and 14: The authors should provide more descriptions and statistical comparison in the main text.
Reply and revision: Thank you very much for the suggestion.
The statistical data for two of the categories have been included in lines 495-496 and lines 499-502.
For instance, in Classification 9, the mean value is 0.55 with a standard deviation of 2.28. Classification 12 exhibits a mean value of 0.23 with a standard deviation of 1.18.
In the case of Classification 9, the mean value is 0.83 with a standard deviation of 0.89. Conversely, Classification 30 exhibits a mean value of 0.69 with a standard deviation of 0.76. These statistics indicate good internal consistency within Classification 30
(25) Lines 431-435: Please explain the optimality of these values. What if they are perturbed / slight changes are imposed?
Reply and revision: Thank you very much for the suggestion.
The setting of thresholds is explained in lines 141-145 of Part 2.3, where different thresholds are classified for different research objectives.
In part 5.2 the thresholds are adjusted to 5,19 and 31 in order to focus on the research theme in lines 570-571.
(26) Lines 447-459: Again, the authors should get more focused on the investigation / case study discussed in this research, but not mentioning something in general.
Reply and revision: Thank you very much for the suggestion.
We have deleted this content, and make sure this section discusses the causes of error generation in focusing on multiscale segmentation and hydrological analysis.
A new section on the effect of changes in multiscale segmentation metrics has been added on conclusions in lines 574-585.
(27) Lines 481-482: To what extent does PCA enhance the retrieval performance? The authors still couldn't explain the effectiveness and advantages of PCA well.
Reply and revision: Thank you very much for the suggestion.
An explanation highlighting the advantages of principal component analysis was added, along with adjustments to the wording in lines 609-614.
It effectively identifies micro terrain variations when combined with principal component analysis. Validation results confirm that the average overlap area at the mesoscale is 69%, with the mean values of Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) being 1.39 and 0.55, respectively. At the microscale, the average overlap area is 70%, with mean RMSE and MAE values of 0.35 and 0.18, respectively.
(28) Lines 489-490: The future goals here are quite vague, please be more specific, and associate with sustainable development of a city / county, the urban land management, and government policies etc.
Reply and revision: Thank you very much for the suggestion.
The manner of expression has been modified to focus on the research objectives in lines 628-631.
the application of multi-scale segmented land parcels in cold regions requires focusing on aspects such as the hydrological processes of different land parcels, changes in land use, land evaluation and management, and identification of geological disasters. Through scale variation, the universal laws governing these geographic units should be explored.
Minor Problem
(1) Equation (1) Is it absolute value instead of bracket? Is it conducted / calculated in 2-norm?
Reply and revision: Thank you very much for the suggestion.
The parentheses were changed to '|', and the addition of the Euclidean distance was made to enhance the explanation of the formula in line 155.
|·|2 denotes the Euclidean norm
(2) Line 132: mu_i instead of U_i
Reply and revision: Thank you very much for the suggestion.
Change to “ui “ in line 154
(3) Line 136: where
Reply and revision: Thank you very much for the suggestion.
Delete “where” in line 158
(4) Line 156: Why 3*3 instead of 4*4 or 5*5? Please explain the rationale.
Reply and revision: Thank you very much for the suggestion.
No changes were made in the article for the following reasons.
In this study, a 3x3 window was chosen primarily because smaller windows are more sensitive to terrain changes and have less edge effect compared to other window sizes. Additionally, a 3x3 window requires less computational effort, resulting in faster processing speeds.
(5) Equation (12): underscript x and y, also there should be some problem with the equation.
Reply and revision: Thank you very much for the suggestion.
Adjusted the subscript in the equation, where 'focal' represents the focal operation function in R language in lines 214-215.
(6) Line 300: "similar properties" - what kind of properties?
Reply and revision: Thank you very much for the suggestion.
In conjunction with your comments and those of other reviewers, the assessment language formulation and methodology were revised in lines 389-393 and 399-409.
This study's "similar properties" refers to the:Similarity of properties within each type of terrain unit at the level of indicators such as slope direction, surface curvature, elevation, flow direction, flow rate, gradient, CTI, LS, SAR, SAV, TPI, etc.
(7) Line 317: "related" - is it positively related or negatively related?
Reply and revision: Thank you very much for the suggestion.
In lines 414-417 it was clarified that an increase in the threshold value resulted in an increase in the extent of the catchment area, and the two showed a positive correlation.
The variation in catchment area extent is associated with changes in threshold values. As the threshold increases, the catchment area will encompass those of smaller thresholds, increasingly ignoring micro-geomorphological structures during the merging process.
(8) Line 361: "certain directions" - describe and elaborate
Reply and revision: Thank you very much for the suggestion.
Modified the method of expression in line 464
there is a high degree of internal consistency in the flow directions of the five types of land parcels,
(9) Line 372: "wide range" - what is the range?
Reply and revision: Thank you very much for the suggestion.
The previous language was incorrect and is now adjusted in line 464
there is a high degree of internal consistency in the flow directions of the five types of land parcels.
(10) Lines 475-482: Please provide some statistical values.
Reply and revision: Thank you very much for the suggestion.
We have added statistical data in lines 612-617.
It effectively identifies subtle terrain variations when combined with principal component analysis. Validation results confirm that the average overlap area at the mesoscale is 69%, with the mean values of Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) being 1.39 and 0.55, respectively. At the microscale, the average overlap area is 70%, with mean RMSE and MAE values of 0.35 and 0.18, respectively.
(11) Lines 484-485: Please provide a concrete example of the claim that "accuracy of the boundary is greatly influenced by the data structure"
Reply and revision: Thank you very much for the suggestion.
We have improved the statement more specifically in lines 622-624.
firstly, the accuracy of the boundary is influenced by the principal component analysis and the method of weight assignment.
English Grammatical Errors
Line 23: different spatial scales
Reply and revision: Thank you very much for the suggestion. We have made revisions accordingly in Line 23.
Line 35: successive laws
Reply and revision: Thank you very much for the suggestion. We have made revisions accordingly in Line 41.
Line 42: methodological framework that is capable of
Reply and revision: Thank you very much for the suggestion. We have made revisions accordingly in Line 48.
Line 49: Second
Reply and revision: Thank you very much for the suggestion. We have made revisions accordingly in Line 58.
Line 139: meaning that the sum of squared distances
Reply and revision: Thank you very much for the suggestion. We have made revisions accordingly in Line 161.
Line 207: In general, each
Reply and revision: Thank you very much for the suggestion. We have made revisions accordingly in Line 238.
Line 226: Fourth
Reply and revision: Thank you very much for the suggestion. We have made revisions accordingly in Line 285.
Line 277: K-means
Reply and revision: Thank you very much for the suggestion. We have made revisions accordingly in Line 358, 363, 371.
Line 302: to be further analyzed
Reply and revision: Thank you very much for the suggestion.
After modifying 3.6 Methodology for Progress Evaluation, this sentence was deleted to better explain the evaluation process
Line 316-317: (Figures 5 and 6)
Reply and revision: Thank you very much for the suggestion. We have made revisions accordingly in Line 413.
Line 409: targets at small watersheds
Reply and revision: Thank you very much for the suggestion. We have made revisions accordingly in Line 547.
Line 444: In addition
Reply and revision: Thank you very much for the suggestion. We have made revisions accordingly in Line 594.
Line 475: variations of indicator datasets
Reply and revision: Thank you very much for the suggestion. We have made revisions accordingly in Line 612.
Line 485: algorithm was run in R
Reply and revision: Thank you very much for the suggestion. We have made revisions accordingly in Line 624.
Reviewer 2 Report
Comments and Suggestions for AuthorsThis is research that is useful in the field of remote sensing. However, this research needs to be revised before being considered for publication on various issues, including:
1) In the title there is the keyword "Machine Learning", but in this research there is no clear application of machine learning.
2) The author should clearly explain how machine learning is used in the process and should present it in figure 1.
3) The abstract lacks presentation of research results, such as the accuracy of using multi scale segmentation.
4) Figure 4 should be moved to Figure 1.
5) Authors should add a section on Data preparation and processing and include all data used in this research. The information that should be presented includes data resolution, Aquisition date and Data sources, presented in a table format.
6) The author should clearly explain how data with different resolutions are processed before being analyzed. Moreover, are different types of data from the same time period?
7) Figure 1 should be consistent with Sections 3.1 to 3.6 and the author should explain the details clearly.
8) Machine learning has not yet been used. The author should clearly explain which method, such as Supervise or Unsupervise, and which algorithm was used in this research.
9) How does the author know that the proposed method is accurate and consistent with the actual conditions at that time?
10) Figures 7 and 8 lack legend representation, making it difficult to interpret the resulting images.
11) The research results lack an evaluation of accuracy or consistency in comparison with other methods. Therefore, the authors should present accuracy or correlation values.
12) The author should clearly present the contribution of this research in comparison to previous research.
13) The author should mention the limitations of the treshold method and its applicability in other areas with different physical conditions.
Author Response
1) In the title there is the keyword "Machine Learning", but in this research there is no clear application of machine learning.
Reply and revision: Thank you very much for the suggestion.
In this study, multi-scale segmentation was employed using the Rstoolbox package, which utilizes unsupervised classification with the K-means algorithm for image feature extraction. This approach is specifically tailored for extracting meaningful regions from image data.
While machine learning's unsupervised classification can be applied to a broader range of data types with a focus on identifying categories within a dataset, it may not necessarily consider the inherent nature or scale of the data.
To maintain a more focused approach in this research, we have omitted the concept of machine learning and exclusively utilized the concept of multi-scale segmentation.
2) The author should clearly explain how machine learning is used in the process and should present it in figure 1.
Reply and revision: Thank you very much for the suggestion.
In Figure 1, an explanation of the mechanism of scale variation in multi-scale segmentation was added in line 304.
3) The abstract lacks presentation of research results, such as the accuracy of using multi scale segmentation.
Reply and revision: Thank you very much for the suggestion.
We have improved the results of the abstract according to your and other reviewer’s suggestions in lines 24-36.
(1) The Land Parcels extracted using the moisture analysis method are catchment units centered around rivers, including slopes on both sides of the river. In contrast, multi-scale segmentation extracts regions with similar properties of Land Parcels, enabling the segregation of slopes and channels into independent units. (2) At a classification threshold of 19, multi-scale segmentation divides the study area into five different types of Land Parcels, reflecting the heterogeneity of terrain undulations and their hydrological connections. When the classification threshold is set to 31, the study area is divided into 15 types of Land Parcels, primarily highlighting mi-cro-topographic features. (3) Multi-scale segmentation can merge and categorize areas with the least heterogeneity in Land Parcels, facilitating subsequent statistical analysis. Therefore, mesoscale Land Parcels extracted through multi-scale segmentation are invaluable for analyzing regional surface soil erosion, material source distribution, and material migration processes. Microscale Land Parcels are significantly important for identifying high-risk areas of geological disasters like landslides and collapses.
4) Figure 4 should be moved to Figure 1.
Reply and revision: Thank you very much for the suggestion.
Since the content of Figure 1 was enhanced according to the suggestion of other reviewer, the figure would provide an overview of the total methodology. On the other hand, Figure 4 is essentially one part of the analysis procedure. Therefore, the position of Figure 4 remained unchanged to maintain the integrity of the paper's structure.
5) Authors should add a section on Data preparation and processing and include all data used in this research. The information that should be presented includes data resolution, Aquisition date and Data sources, presented in a table format.
Reply and revision: Thank you very much for the suggestion.
Section 2.2 describes the data source in line 120, with the data being titled "GLO-30" and having a resolution of 30 meters.
Additionally, an explanation of data errors relevant to this study was included in lines 125-127.
6) The author should clearly explain how data with different resolutions are processed before being analyzed. Moreover, are different types of data from the same time period?
Reply and revision: Thank you very much for the suggestion.
The metrics used in this article are all available from DME data calculations, so there are no resolution differences or coordinate differences, and they have not been modified in the article.
The elevation data utilized in this study were obtained from the Copernicus
(https://spacedata.copernicus.eu/en/web/guest/collections/copernicus-digital-elevation-model) dataset named '2022_1,' which was last updated on the official website in December 2022. The data were downloaded in August 2023, and they possess a spatial resolution of 30 meters. All other derived indicators in this research were computed using this dataset, ensuring consistent spatial resolution throughout the study.
7) Figure 1 should be consistent with Sections 3.1 to 3.6 and the author should explain the details clearly.
Reply and revision: Thank you very much for the suggestion.
We have improved figure 1 according to other reviewer’s suggestion. Meanwhile, additional details were added in Section 3.2 regarding the extraction of land parcels through hydrological analysis. The content of model construction in Section 3.5 was expanded, including statistical data of typical areas, as well as the precision assessment of land parcels in Section 3.6.
8) Machine learning has not yet been used. The author should clearly explain which method, such as Supervise or Unsupervise, and which algorithm was used in this research.
Reply and revision: Thank you very much for the suggestion.
The concept of machine learning was removed. In Section 2.3, the algorithms used in the software were described as unsupervised classification in line 146.
9) How does the author know that the proposed method is accurate and consistent with the actual conditions at that time?
Reply and revision: Thank you very much for the suggestion.
We have added validation contents according to your and other reviewer’s suggestions.
Section 2.6 has been added in lines 245-269, detailing the methods for validating land parcels.
In Section 3.6, Lines 389-393 and 399-409, the methods and approaches for land parcel validation have been expanded upon.
Additionally, Sections 4.3.3 and 4.3.4 in lines 513-542, have been enhanced to include evaluations and explanations of multi-scale segmented units at both mesoscale and microscale levels.
10) Figures 7 and 8 lack legend representation, making it difficult to interpret the resulting images.
Reply and revision: Thank you very much for the suggestion.
Different colors in Figure 7 and Figure 8 showed different classification. However, since the classification is reflected by different numbers, a legend which only contain numbers is difficult to show the specific type of geographic units. Furthermore, the two figures are sample areas which showed the spatial pattern of the extraction results rather than the total results. Therefore, we have added a statement “different color in the figure represent different types of land parcels” instead of adding a legend.
Add color and terrain unit descriptions to lines 450and 454, respectively
11) The research results lack an evaluation of accuracy or consistency in comparison with other methods. Therefore, the authors should present accuracy or correlation values.
Reply and revision: Thank you very much for the suggestion.
We have added validation contents according your and other reviewers’ comments. Additionally, Sections 4.3.3 and 4.3.4 in lines 513-542, have been enhanced to include evaluations and explanations of multi-scale segmented units at both mesoscale and microscale levels.
12) The author should clearly present the contribution of this research in comparison to previous research.
Reply and revision: Thank you very much for the suggestion.
We address the contributions of this study in lines 24-36 of the abstract, and 628-631 of the conclusions, respectively.
13) The author should mention the limitations of the treshold method and its applicability in other areas with different physical conditions.
Reply and revision: Thank you very much for the suggestion.
A discussion of the impact of changes in the 5.3 multiscale segmentation metrics on the conclusions is added in lines 574-585, and applicability is summarized in lines 617-621.
Reviewer 3 Report
Comments and Suggestions for AuthorsAutomatic Extraction of Land Parcel Units Based on Multi-Scale Segmentation and Machine Learning
Abstract – much as the first part of the abstract is well structured. Authors need to improve the presentation of results. Provide evidence that shows superiority - show values for comparison purposes
Page 1 line 34 – wrong spelling introduction. Generally,
Page 2: study area – the description is not sufficient – e.g. climate, what are the annual rainfall amounts received?
Results
Accuracy of both methodologies should be demonstrated. For now, the comparisons made are insufficient. How were the segments validated in the field? - Fei Liu and Lilei Wu; validation
Discussion
The authors need to take some of the paragraphs to the introduction such as on page 17, line 420. Many statements should be reflected in the introduction
Comments for author File: 
Comments.pdf
Author Response
Abstract – much as the first part of the abstract is well structured. Authors need to improve the presentation of results. Provide evidence that shows superiority - show values for comparison purposes
Reply and revision: Thank you very much for the suggestion.
We have made revisions according to your and other reviewer’s suggestions. The content of the abstract has been modified in Line 24-36 as follows:
(1) The Land Parcels extracted using the moisture analysis method are catchment units centered around rivers, including slopes on both sides of the river. In contrast, multi-scale segmentation extracts regions with similar properties of Land Parcels, enabling the segregation of slopes and channels into independent units. (2) At a classification threshold of 19, multi-scale segmentation divides the study area into five different types of Land Parcels, reflecting the heterogeneity of terrain undulations and their hydrological connections. When the classification threshold is set to 31, the study area is divided into 15 types of Land Parcels, primarily highlighting mi-cro-topographic features. (3) Multi-scale segmentation can merge and categorize areas with the least heterogeneity in Land Parcels, facilitating subsequent statistical analysis. Therefore, mesoscale Land Parcels extracted through multi-scale segmentation are invaluable for analyzing regional surface soil erosion, material source distribution, and material migration processes. Microscale Land Parcels are significantly important for identifying high-risk areas of geological disasters like landslides and collapses.
Page 1 line 34 – wrong spelling introduction. Generally,
Reply and revision: Thank you very much for the suggestion.
We have made revisions accordingly in Line 40.
Page 2: study area – the description is not sufficient – e.g. climate, what are the annual rainfall amounts received?
Reply and revision: Thank you very much for the suggestion.
The description now includes additional information related to climate and precipitation in Line 108-111.
with daily average temperatures ranging from -10.9℃ in January to 13℃ in July, indicating a significant annual temperature variation. Precipitation varies across the year, with higher rainfall typically occurring in June and July, while other months experience less precipitation.
Accuracy of both methodologies should be demonstrated. For now, the comparisons made are insufficient. How were the segments validated in the field? - Fei Liu and Lilei Wu; validation
Reply and revision: Thank you very much for the suggestion.
We have added validation contents according to your and other reviewer’s suggestions. Section 2.6 has been added in lines 245-269, detailing the methods for validating land parcels.
In Section 3.6, Lines 389-393 and 399-409, the methods and approaches for land parcel validation have been expanded upon.
Additionally, Sections 4.3.3 and 4.3.4 in lines 513-542, have been enhanced to include evaluations and explanations of multi-scale segmented units at both mesoscale and microscale levels.
The authors need to take some of the paragraphs to the introduction such as on page 17, line 420. Many statements should be reflected in the introduction
Reply and revision: Thank you very much for the suggestion.
The preamble has not been adjusted due to the detailed description of the land parcels in lines 88-90.
The first sentence of this paragraph has also not been repositioned in order to describe the internal structure of the terrain unit.
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThe content of the manuscript is rich enough now, and after editing, the paper is ready to be published. Good work!
Comments on the Quality of English LanguageA proper round of English editing is advised
