Terracing is one of the most important soil and water conservation (SWC) measures in China, playing a critical role in mitigating soil erosion. This study proposes a two-stage mapping framework to classify different types of terrace measures and develops a new dataset—the Soil and Water Conservation Terrace Measures Dataset (SWCTMD)—based on time-series Landsat imagery and digital elevation model (DEM) data from 2000 to 2020. The framework incorporates a refined classification system that provides detailed information on both terrace distribution and associated SWC measure factors, offering significant value for understanding and managing soil erosion dynamics.

(1) It is recommended to revise the title to: A 30 m resolution dataset of soil and water conservation terraces across China (2000–2020).

(2) What specific subcategories of cropland are included in the study?

(3) What information is contained in the 30 m grid dataset? Does it include whether the area is terraced or not, the type of terrace, and the associated conservation measure factor?

(4) It is recommended to retain numerical values to one decimal place for better clarity and consistency.

(5) In Table 2, the four types of terrace measures are shown with remote sensing images. To enhance clarity and visual recognition, it is recommended to replace these with high-resolution photographs.

(6) In the accuracy assessment section, the evaluation metrics should be further explained, such as the possible value ranges of each indicator and whether higher or lower values indicate better accuracy.

(7) The methodology section lacks details about the estimation of soil erosion and the other influencing factors used in the analysis related to terrace responses in China. Please specify the estimation methods and data sources for these additional factors.

(8) In Figure 3, should the legend indicate "cropland with terraces" and "cropland without terraces," rather than "cropland" and "terraces"? Additionally, does the map on the left represent the distribution of cropland? Please clarify it.

(9) To highlight the novelty of this dataset, it is recommended to include a comparison table in addition to Figure 3. This table should provide quantitative details of the differences between this new dataset and existing ones, such as the extent of area change in specific regions and the types of terraces contributing to those changes.

(10) In Figure 6, it is suggested to label the numerical values within each grid cell to improve readability and interpretability.

(11) Many quantitative descriptions in the manuscript only provide absolute values; it is recommended to also include relative percentages to help readers better interpret the significance of the results.

(12) In Figure 7, are fixed values assigned to each type of terrace measure? Please clarify the value assignment approach.

(13) The sentence: “According to our estimation, the soil erosion of the Loess Plateau accounts for only 10.95% of the total cropland erosion in China, indicating that the SWC measures previously implemented have achieved good governance…” is better rephrased with the emphasis placed on the comparative effects of having terraces vs. not having terraces, or the differences among this dataset and previpus datasets, in order to reflect the value of this dataset in soil erosion estimation. The relatively low share of cropland erosion in the Loess Plateau does not necessarily indicate the effectiveness of conservation measures alone—it may also be influenced by factors such as total cropland area, topography, vegetation cover, climate and so on.

This paper aims to provide a dataset of multi-type terraces in China from 2000 to 2020. However, this study primarily focuses on the practical applications of the dataset (many contents related to spatial-temporal analysis and variable importance evaluation), while current methods have obvious shortcomings in terms of data applicability, model robustness, and validation rigor, which affect the credibility of the datasets. This paper may be more appropriate for agricultural journals than for dataset publication. Please see specific comments: 

1. The author primarily utilized 30-meter resolution imagery and DEM. However, many terrace surfaces are smaller than 30 meters. It is uncertain whether these data can accurately represent the objects and extent of multi-type terraces. This implies that there are numerous areas with mixed pixel issues. Additionally, especially for the Zig terrace, which have relatively steep slopes (as stated in Table 2), 30-meter data is insufficient to distinguish such objects. I have the same concerns regarding Slope-separated Terrace. 

2. While SRTM DEM (2002 version) was utilized for classification of multi-type terraces, its applicability for decadal-scale change detection (2010-2020) is problematic. The key limitation lies in the dataset's inability to capture terrain modifications occurring after its acquisition date, particularly for newly constructed terraces during the study period. 

3. The author used cropland from GlobeLand30 as the range restriction and then entered it into a random forest for classification. However, this approach may introduce bias since terraced fields are frequently misclassified as forest or grassland in global land cover products due to resolution limitations and spectral confusion. The methodological justification for using cropland as an exclusionary mask requires further justification. 

4. The author should provide the specific number of non-terrace and terrace samples. As shown in Figure 2, the number of non-terrace sample points far exceeds that of terrace sample points. These non-terrace areas are generally easier to distinguish (e.g., many points are located on the Himalayan Mountain range, and some are in desert regions). Even without machine learning methods, it is entirely possible to determine this, as terraced fields simply cannot exist in these regions. This suggests that the subsequent accuracy validation may be inflated. The author should consider balancing the number of sample points across different types.

5. The authors primarily focus on analyzing the spatiotemporal patterns within the dataset and evaluating variable importance, while paying comparatively less attention to methodological validation and uncertainty quantification. The authors lack a comprehensive assessment of the data quality. In addition, the authors did not provide visual results for the reference true values, making it difficult for readers to judge the validity of the results. 

6. Table 4 shows notably low F1-scores for Level terrace, Zig terrace and slope-separated terrace, suggesting limited model performance in discriminating these specific terrace types. The classification methodology requires optimization to achieve satisfactory accuracy levels that meet the scientific objectives. 

7. The authors compare their results with the 30-meter resolution terrace dataset published in ESSD (2021). However, as shown in Figure 3, the validation areas are geographically proximate. This limited spatial scope reduces the robustness of the comparative analysis. To more convincingly demonstrate the advantages of their approach, the authors should expand the comparison to include diverse geographic regions representing different environmental conditions and terrace types. 

8. The phrase “from 2000 to 2020” in the author's title may mislead readers into assuming that the dataset is on an annual scale, but in fact the author only produced three years.

9. After constructing feature factors, authors should perform correlation analysis between factors, which is a necessary step in feature construction. In addition, authors should also provide more detailed formulas or steps for calculating each factor.

This study presents a 30-meter resolution dataset of Soil and Water Conservation Terrace Measures in China. The manuscript is generally well-structured, and the dataset holds significant value for accurate soil erosion assessment and for supporting soil and water conservation planning. However, I have several suggestions and comments that may help improve the clarity and rigor of the paper:

1. Lines 61–63:

The authors mention that ignoring terrace types could lead to inaccuracies in soil erosion assessments. I recommend expanding on this point to clarify the mechanisms through which different terrace types influence erosion rates. This would help to better justify the necessity of developing a terrace-type-specific dataset.

2. Spatial Resolution Clarification:

Please explicitly state the spatial resolution of the final product in the Data and Methods section. Additionally, if input datasets used in the production process have differing spatial resolutions, please clarify how this issue was handled (e.g., resampling methods or resolution harmonization techniques).

3. Validation with Statistical Data:

Has the dataset been validated or cross-referenced with any official statistical records on terrace areas, either at the national or regional level? If feasible, a comparison with statistical data, including breakdowns by terrace type, would strengthen the credibility and applicability of the dataset.

4. Comparison with Existing Datasets:

To further highlight the value and potential advantages of the newly developed dataset, I suggest comparing it with existing terrace-related datasets (if available). This could include spatial consistency, classification accuracy, or coverage of different terrace types.