Spatiotemporal Dynamics and Driver Pathways of Soil Erosion in Qilian Mountain National Park (1990–2022) Under Ecological Restoration

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

I have some questions for the authors.

• Can natural erosion be separated from accelerated erosion in the area, deeming an (urgent) need for more management?
• The conversion of bare land to vegetated ecosystems has been conducted using what species, or was it natural revegetation? What is the mechanism for forests going to grassland, deforestation?
• Row 16 we have a figure of 2.77 * 10^-4, row 217 we have 2.77 * 10^-5. Anyway, the figures of erosion would be preferably given in t/ha/year in all situations mentioned, it would be more coherent to follow and to compare for example to the 20-100 t/ha/year cyphers characterizing the erosion hotspots of the world or those mentioned in the Alps or Morrocco.
• Severe problems in what regard references. From some citations which are not needed or others that can be changed (e.g. 1,2 are not the main references for a worldwide assessment of soil erosion, references regarding UNCCD are not given from them, Gupta et al. – ref. 4 cites FAO, Smith et al. – ref. 14 cites FAO).
• Regarding references, the use of at least quilian instead of Quilian repeatedly should raise a question, also the repeated use of authors such as M. A. J. R. L. S. F. e. N. Improper citation of sources (names of journals and paper titles – this makes it tough for a reviewer to check some aspects).
• Row 6 – the study of Ballabio et al. is interesting, although the cyphers given seem exaggerated. What caught my eye is the scale they used (250 m resolution at global scale), while the authors preferred to rescale all data at 1 km resolution. Why?
• Figure 1, please explain the pattern of river network in the left low corner. Secondly, a different color pattern or legend would be needed, since we cannot see any farmland except in the smaller cropped images. Same with figure 3, changes are hard to detect even at the cropped scale.
• What do multi-temporal remote sensing and geospatial data mean? Please describe the way NDVI has been obtained and classified to reach the land cover classes mentioned.  
• RUSLE results obtained statistical explanation insists on rainfall erosivity and LS. What about the other factors? Rows 137-138 – crop cover not mentioned as contributing factor of soil erosion.
• In what regards RUSLE, taking the factors, we can assume that soil erodibility, slope length and steepness and support factor do not change in 33 years, leaving as determinant factors explaining changes rainfall erosivity and land cover.
• Some data on rainfall quantities and rainfall intensities would be suited, and if a multiannual mean was used also.
• In the RUSLE equation, where is the SSi in the explanation fit into the equation?
• What is the purpose / logic in including the GDP in the factors taken into consideration?

Kind regards!

Author Response

To Reviewer #1

We sincerely thank Reviewer for the thorough and detailed evaluation of our manuscript. The comments provided by Reviewer 1 were particularly helpful in clarifying the conceptual framing of soil erosion processes, refining the interpretation of model assumptions, and improving the coherence between methodological choices and result interpretation. In response, we carefully revised the manuscript to better define the scope and limitations of the RUSLE-based assessment, clarify the roles of different driving factors, and ensure consistency in terminology, units, and notation. Below, we address each comment raised by Reviewer in a point-by-point manner, with specific references to the corresponding sections, page numbers, and line numbers in the revised manuscript.

 

 

Comment 1: Can natural erosion be separated from accelerated erosion in the area, deeming an (urgent) need for more management?

Response: Thank you for the reviewer's insightful and academically valuable comment. Natural erosion and human-induced accelerated erosion are fundamentally different processes in terms of their formation mechanisms, and distinguishing between them is important for advancing the understanding of soil erosion dynamics. In this study, the RUSLE model is applied to assess integrated soil erosion risk and actual erosion intensity under given climatic conditions, topographic characteristics, land-cover patterns, and management practices. Conceptually, the model outputs inherently include both natural background erosion processes and human-induced accelerated erosion components. At the regional scale, however, due to limitations in the availability of long-term, high-resolution data on historical land management practices, disturbance intensity, and detailed human activity processes, it remains difficult to achieve a strict and reliable quantitative separation of these two erosion processes within a RUSLE-based assessment framework. This methodological limitation is explicitly stated in Section 2.3.1 (Soil erosion assessment system, Lines 176–188).

To clearly address this methodological constraint, we have implemented targeted revisions at multiple levels of the revised manuscript. First, in the latter part of the Introduction (Lines 66–73), we explicitly clarify that, from the initial research design stage, this study fully recognizes that natural background erosion and human-induced accelerated erosion are often spatially intertwined at the regional scale, and that a process-level separation is difficult under current data conditions. Accordingly, the objective of this study is not to strictly decompose erosion types, but rather to focus on characterizing the spatiotemporal dynamics of integrated soil erosion intensity and its relationships with land-use change and human activity patterns, so as to better support soil and water conservation and ecological management decisions at the regional scale.

Second, in the Methods section (Section 2.3.1, Soil erosion assessment system, Lines 176–188), we further clarify that the soil erosion estimates derived from the RUSLE model represent integrated erosion intensity. Their physical meaning reflects the combined effects of natural environmental factors and anthropogenic influences, rather than a separated quantification of individual erosion processes.

 

Meanwhile, although natural and accelerated erosion are not quantitatively distinguished at the model level, the role of human disturbances is systematically examined in the Discussion section (Section 4.5, Management implications, Lines 464–492). Specifically, we analyze the effects of land-use change, the Human Footprint Index (HF), and the implementation of ecological engineering measures on erosion dynamics. The results reveal a pronounced spatial consistency between variations in local erosion intensity and adjustments in human activities and management interventions. This finding clearly indicates that anthropogenic disturbance remains a key driver of localized soil erosion dynamics and strongly supports the reviewer's judgment that more refined management measures are still required in this region.

 

 

Comment 2: The conversion of bare land to vegetated ecosystems has been conducted using what species, or was it natural revegetation?

Response: Thank you for this insightful and scientifically important comment. We agree that distinguishing vegetation restoration pathways and species composition can enhance the understanding of ecological processes. However, at the regional scale of this study, long-term, high-resolution data on species distribution and detailed records of ecological restoration engineering are not consistently available, which limits our ability to investigate vegetation restoration at the species level.

Accordingly, we have clarified this limitation and methodological choice at both the Methods and Discussion levels in the revised manuscript. Specifically, in Section 2.3.1 (Soil erosion assessment system, Lines 187–191), we explicitly state that vegetation restoration was not analyzed at the species level due to constraints related to study scale and data availability. Instead, vegetation recovery was characterized using NDVI dynamics and land-use/land-cover transitions (e.g., conversion from bare land to grassland, shrubland, or forest), which reflect overall improvements in vegetation cover and their integrated effects on soil erosion. This approach is consistent with the objectives of regional-scale soil erosion assessment based on the RUSLE framework.

Furthermore, in the Discussion section (Section 4.3, Lines 412–420), we further explain that vegetation restoration within Qilian Mountain National Park exhibits characteristics of both natural recovery and assisted restoration. In high-altitude areas and core protected zones, where human disturbance is relatively low, vegetation recovery is mainly driven by natural succession. In contrast, in degraded grasslands, sloping croplands, and areas influenced by ecological engineering measures, restoration has been accelerated through management interventions such as grazing exclusion and enclosure. At the regional scale, this study focuses on changes in overall vegetation cover rather than detailed species composition, thereby ensuring consistency between the data sources, analytical framework, and the conclusions regarding vegetation restoration and soil erosion dynamics.

 

 

Comment 3: What is the mechanism for forests going to grassland, deforestation?

Response: Thank you for the reviewer’s insightful comment with important academic value. We fully agree that distinguishing vegetation recovery pathways and species composition would contribute to a deeper understanding of ecological processes. However, at the regional scale adopted in this study, limitations in the availability of long-term, high-resolution species distribution data and detailed records of ecological restoration projects prevent an analysis of the transition from bare land to vegetated ecosystems at the species level.

In response to this comment, we have strengthened the manuscript from both methodological and discussion perspectives. First, in the Methods section (Section 2.3.1, Soil erosion assessment system, Lines 176–188), we explicitly clarify that vegetation recovery is not analyzed at the species level in this study due to constraints related to research scale and data availability. Instead, vegetation recovery is characterized using NDVI dynamics and land-use/land-cover transitions (e.g., conversion from bare land to grassland, shrubland, or forest), which reflect overall improvements in vegetation cover and their integrated effects on soil erosion. This approach is consistent with the objectives of regional-scale soil erosion assessment based on the RUSLE framework.

In addition, in the Discussion section (Section 4.3, Preliminary effects of ecological restoration over the past 33 years, Lines 420–463), we further elaborate that vegetation recovery within Qilian Mountain National Park reflects the coexistence of natural recovery and assisted restoration processes. In high-altitude areas and core protected zones, where human disturbance is relatively weak, vegetation recovery is mainly driven by natural succession. In contrast, in degraded grasslands, sloping croplands, and areas where ecological engineering measures have been implemented, vegetation recovery has been accelerated through enclosure, grazing exclusion, and related management practices. At the regional scale, this study therefore focuses on overall changes in vegetation cover rather than a fine-scale characterization of species composition.

 

Comment 4: Row 16 we have a figure of 2.77 * 10-4, row 217 we have 2.77 * 10-5. Anyway, the figures of erosion would be preferably given in t/ha/year in all situations mentioned, it would be more coherent to follow and to compare for example to the 20-100 t/ha/year cyphers characterizing the erosion hotspots of the world or those mentioned in the Alps or Morrocco.

Response: Thank you for pointing out this important issue regarding the inconsistency of numerical expressions and units used to report soil erosion.

In response to this comment, we have systematically reviewed all soil erosion values throughout the manuscript and corrected the inconsistent exponential expressions. All soil erosion rates are now consistently expressed in t·ha⁻¹·yr⁻¹, including the Abstract (Lines 16–20), the Results and Discussion sections, as well as all figures and supplementary materials.

Accordingly, the unit expressions in Figure 7 (Manuscript, Lines 381–382) and Figure S2 (Supplementary Materials, Lines 27–31) have been updated to ensure consistency with the revised text. In addition, comparative discussions with erosion rates reported for global erosion hotspots and alpine regions (e.g., 20–100 t·ha⁻¹·yr⁻¹ in the Alps and Morocco) have been revised and consolidated in Section 4.2 (Consistency between soil erosion distribution and alpine mountain ecosystems, Lines 395–405).

Comment 5: Severe problems in what regard references. From some citations which are not needed or others that can be changed (e.g. 1,2 are not the main references for a worldwide assessment of soil erosion, references regarding UNCCD are not given from them, Gupta et al. – ref. 4 cites FAO, Smith et al. – ref. 14 cites FAO).

Response: Thank you for highlighting the concerns regarding the reference selection. In response, we have systematically reviewed the entire manuscript and reference list, and made corresponding revisions to ensure that all citations are necessary, appropriate, and supported by authoritative sources, particularly for global-scale soil erosion assessments and UNCCD-related statements. These revisions have been implemented throughout the manuscript, mainly in the Introduction (Lines 44–63) and consolidated in the updated References section (Lines 553–612).

 

 

Comment 6: Regarding references, the use of at least quilian instead of Quilian repeatedly should raise a question, also the repeated use of authors such as M. A. J. R. L. S. F. e. N. Improper citation of sources (names of journals and paper titles – this makes it tough for a reviewer to check some aspects).

Response: Thank you for your careful and detailed comments regarding the reference list and citation formatting.

In response to this comment, we have conducted a systematic review and correction of all references throughout the manuscript. First, all geographic names have been standardized to “Qilian Mountains” or “Qilian Mountain National Park”, and inconsistent or incorrect spellings (e.g., “quilian”) have been fully corrected in both the main text and the reference list (Manuscript, Lines 31–35; 74–78; 102–105). We carefully checked and corrected all author name entries to eliminate improperly parsed or fragmented author strings (e.g., abnormal abbreviations such as “M. A. J. R. L. S. F. e. N.”). All references now present complete and correctly ordered author lists, as shown in the revised References section (Manuscript, Lines 553–612). These revisions have been applied uniformly throughout the entire reference list (Manuscript, Lines 553–612).

 

Comment 7: Row 6 – the study of Ballabio et al. is interesting, although the cyphers given seem exaggerated. What caught my eye is the scale they used (250 m resolution at global scale), while the authors preferred to rescale all data at 1 km resolution. Why?

Response: We sincerely thank the reviewer for the detailed and constructive comment highlighting that studies by Ballabio, Borrelli, and colleagues conducted soil erosion simulations at the global scale using a spatial resolution of 250 m. This point is methodologically important. However, it should be clarified that our objectives of the present study are fundamentally different from those of the aforementioned global-scale scenario simulations.

Specifically, Borrelli et al. (2020, PNAS) employed the GloSEM framework, supported by high-resolution and globally consistent input datasets, to quantify and project potential global water erosion under different SSP–RCP scenarios. Their work primarily focuses on estimating absolute erosion magnitudes at the global scale and assessing their future changes under alternative scenarios. In contrast, our study aims to investigate the spatiotemporal evolution of land degradation processes at the regional scale and to disentangle the relative contributions and coupled effects of natural factors and human activities on soil erosion dynamics, rather than to produce pixel-level high-precision estimates of absolute erosion rates.

Based on above objectives, all driving factors in this study were harmonized to a spatial resolution of 1 km for the following methodological considerations: (1) the long-term climate data, soil property datasets, and human activity indicators used in this study exhibit higher consistency, temporal continuity, and spatial completeness at the kilometer scale; (2) for regional-scale analyses, a 1 km resolution is sufficient to capture the spatial distribution of erosion hotspots and major environmental gradients, while reducing the propagation of uncertainty associated with multi-source high-resolution inputs; and (3) adopting a unified 1 km resolution facilitates scale consistency among variables within the integrated RUSLE–RF–PLS-SEM analytical framework, thereby improving the robustness and comparability of the results in terms of mechanism interpretation.

Therefore, the use of a 1 km spatial resolution in our study should not be regarded as a simplified substitute for existing high-resolution global assessments, but rather as a scale- and objective-driven methodological choice that appropriately balances research aims, data availability, and multi-source driver compatibility.

Comment 8: Figure 1, please explain the pattern of river network in the left low corner. Secondly, a different color pattern or legend would be needed, since we cannot see any farmland except in the smaller cropped images. Same with figure 3, changes are hard to detect even at the cropped scale.

Response: Thank you for the reviewer's constructive comments on the interpretation of the spatial patterns shown in the figures. With respect to the river network pattern observed in the lower-left corner of Figure 1, this pattern reflects the joint control of regional geomorphology and hydroclimatic conditions rather than artifacts of data processing. The study area is characterized by a semi-arid to arid transitional climate and a mountain–basin transition zone, where river systems and their tributaries are primarily distributed along valleys and low-lying corridors. As a result, the drainage network exhibits a dendritic to quasi-parallel structure that is strongly constrained by topography. This interpretation is consistent with the natural geographical setting of the study region and explains the relatively sparse or discontinuous appearance of river channels in some areas of Figure 1.

Regarding the limited visibility of cropland in Figure 1, this is mainly attributable to its extremely small areal proportion within Qilian Mountain National Park. According to the land-use transition statistics (Table S1, Supplementary Materials, Lines 42–44), the total cropland area during the study period ranges from approximately 200 to 245 km², accounting for less than 0.50% of the total study area (52,456 km²). Consequently, at the regional mapping scale, cropland is highly fragmented and visually indistinct when displayed alongside dominant ecosystem types such as grassland, forest, and bare land. This pattern reflects the actual land-use structure of the study area rather than any omission or bias in figure representation.

For Figure 3, the relatively weak contrast among erosion classes is primarily related to the dominance of low erosion intensity levels (L1–L2) across most of the study area. This results in limited visual differentiation of erosion categories at the regional scale. To address this limitation, the overall magnitude and directional transitions among erosion intensity classes from 1990 to 2022 have been quantitatively summarized in Table S2 (Supplementary Materials), which provides a clearer statistical representation of erosion dynamics that cannot be fully conveyed by maps alone. In the revised manuscript, we explicitly emphasize the complementary roles of maps and tables and clarify in the figure captions and discussion that Tables S1 and S2 should be jointly consulted to correctly interpret the spatial patterns shown in Figures 1 and 3.

 

Comment 9: What do multi-temporal remote sensing and geospatial data mean? Please describe the way NDVI has been obtained and classified to reach the land cover classes mentioned.

Response: Thank you for this important question regarding the meaning of multi-temporal remote sensing and geospatial data, as well as the role of NDVI in relation to land-cover classification. In this study, the term “multi-temporal remote sensing and geospatial data” refers to a suite of multi-source raster datasets acquired and consistently processed over a long-term period (1990–2022), including the Normalized Difference Vegetation Index (NDVI), land-use/land-cover change (LUCC), climatic variables, topographic factors, and socio-economic indicators. These datasets are jointly used to characterize ecosystem patterns and their long-term dynamics at the regional scale, as described in Section 2.2 (Data Sources and Preprocessing, Lines 129–147).

It is important to clarify that NDVI was not used to classify land-cover types in this study. The land-cover categories (e.g., cropland, forest, grassland, shrubland, and bare land) were directly obtained from multi-period LUCC datasets, which were harmonized through remote sensing preprocessing to ensure temporal consistency. This methodological distinction is explicitly stated in Section 2.2 (Lines 129–133) and Section 2.3.1 (Soil erosion assessment system, Lines 187–191).

In contrast, NDVI was primarily used as a continuous vegetation indicator to characterize vegetation dynamics associated with land-cover transitions derived from LUCC data, rather than as a categorical land-cover discriminator. Specifically, NDVI served as a process-based vegetation recovery dynamics, and was further used to derive the vegetation cover factor (C factor) in the RUSLE model, as explained in Section 2.3.1 (Lines 176–183).

Accordingly, vegetation recovery in this study is not represented by NDVI threshold-based land-cover classification, but rather by the combined interpretation of NDVI temporal trends and LUCC transitions (e.g., conversion from bare land to grassland, shrubland, or forest). This integrated approach ensures consistency between the adopted data sources, analytical framework, and the regional-scale objectives of soil erosion assessment.

 

 

Comment 10: In what regards RUSLE, taking the factors, we can assume that soil erodibility, slope length and steepness and support factor do not change in 33 years, leaving as determinant factors explaining changes rainfall erosivity and land cover. Some data on rainfall quantities and rainfall intensities would be suited, and if a multiannual mean was used also?

Response: We sincerely thank the reviewer for the professional and constructive comments regarding the parameterization of the RUSLE factors and the description of rainfall information.

With respect to the reviewer's concerns about precipitation amount, rainfall intensity, and the use of multi-year averages, we would like to clarify that this study does not directly employ a single multi-year mean precipitation dataset or a simple long-term averaged R factor. Instead, rainfall erosivity (R) was calculated on an annual basis using year-by-year precipitation data from 1990 to 2022. Annual rainfall erosivity values (Rᵢ) were first derived and then synthesized to obtain long-term mean characteristics, thereby preserving interannual variability for the analysis of long-term erosion dynamics.

Accordingly, the factor attribution and statistical analyses in this study are based on the temporal variability of the R-factor time series, rather than on a single multi-year mean value. This approach allows precipitation-driven influences on soil erosion to be more realistically represented at the regional scale, while avoiding the excessive amplification of short-term climatic fluctuations or pixel-scale noise.

Following the reviewer's suggestion, we have revised Section 2.2 (Data Sources and Preprocessing, Lines 133–147) of the manuscript to explicitly clarify the sources of precipitation data, the temporal synthesis strategy, and the role of annual aggregation in the calculation of the R factor. These revisions enhance the methodological transparency and reproducibility of the rainfall erosivity estimation.

 

 Comment 11: In the RUSLE equation, where is the SSi in the explanation fit into the equation?

Response: Thank you for pointing out this important and detailed issue. This issue originates from an inadvertent omission in the original manuscript during the description of the symbols used in the RUSLE equation.

In the revised manuscript, we have corrected and standardized the notation of the RUSLE equation and its variable explanations to ensure full consistency between the mathematical expression and the accompanying text. Specifically, this correction has been implemented in Section 2.3.1 (Soil erosion assessment system, Lines 193–197), where the equation symbols are now presented using standard RUSLE terminology, and any ambiguous or incorrect symbols have been removed.

 

Comment 12: What is the purpose / logic in including the GDP in the factors taken into consideration?

Response: Thank you for the reviewer's further attention to this issue. It should be clarified that although the Human Footprint Index (HF) and population density (PD) were introduced to characterize the spatial intensity of human disturbance and population pressure, the role of GDP in the analytical framework is not redundant with these indicators. Specifically, HF and PD primarily reflect the spatial distribution and magnitude of human activities, whereas GDP is used to represent the regional level of socio-economic development, which is associated with land management capacity, investment in ecological governance, and differences in development stages.

In the context of high-altitude protected areas, higher GDP does not necessarily correspond to stronger direct human disturbance. Instead, it may partially reflect differences in the capacity to implement ecological engineering projects and management intensity. Therefore, in this study, GDP is incorporated as a macro-scale socio-economic background variable in the RF and PLS-SEM analyses to explore potential regulatory pathways of soil erosion dynamics under different development stages, rather than being treated as a direct erosion-driving factor or an input variable of the RUSLE model. Relevant explanations have been clarified in the Methods and Discussion sections of the revised manuscript (Section 2.4, Driving factor selection, Lines 210–228; Section 4.4, Socio-economic influences, Lines 440–463).

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Title

The topic (Line 2) “Soil Erosion Dynamics and Driving Mechanisms in an Alpine Agro-Pastoral System under Soil Management and Ecological Restoration” addresses an important and timely topic. However, I found some mistakes that currently limit the paper’s suitability for publication in its current form. Therefore, the author (s) should take major revisions to improve the quality of the paper for publication.  

Abstract

The abstract describes clearly the problem and relevance to alpine ecosystems and SDGs. However, there are some grammar mistakes to be corrected. For example, word breaks (“northwest- ern”, “erosion”). The author also should ensure consistency and readability of erosion units (t·km⁻² yr⁻¹ if annual).

Introduction

This section is well-written, comprehensive, and well-structured. The author (s) clearly describes the research problems, the need for this study, and clearly outlines the research objectives and research questions. Minor grammatical issues and punctuation errors (such as extra spaces before citations, sentence fragments). Some sentences are overly long and can be tightened for readability. Repetition of phrases such as “fragile ecosystem” and “soil erosion” can be reduced.

Methodology

In this section, the author/s should clarify and validate soil erosion estimation methodology; draw a conceptual diagram justifying SEM pathways and clearly distinguish statistical causality from process-based causality.

Results and discussion

This section is well organized though needs some improvements. Some interpretations appear in the results section should be moved to discussion section. Keep results strictly descriptive and quantitative. Authors better to shift interpretation of drivers and relevance of policy in the discussion section.

Conclusion

The conclusion clearly summarizes key quantitative findings. However, still it needs some amendments. There are some grammatical errors such as “During last three-decade” should be corrected as “During the past three decades” and Capitalization errors (“Conversion”).

Comments on the Quality of English Language

The English could be improved to more clearly express the research.

Author Response

To Reviewer #2

We thank Reviewer for the constructive and well-focused comments, which primarily addressed the structure, clarity, and presentation quality of the manuscript. These suggestions were instrumental in improving the readability and logical organization of the paper, particularly with respect to the title, abstract, introduction, and the separation between the Results and Discussion sections. Following these comments, we conducted a comprehensive language and structural revision, corrected grammatical and formatting issues, and strengthened the functional distinction between descriptive results and interpretative discussion. Below, we provide detailed responses to each comment from Reviewer and indicate the corresponding revisions in the manuscript.

 

 

Comment 1: Title    The topic (Line 2) “Soil Erosion Dynamics and Driving Mechanisms in an Alpine Agro-Pastoral System under Soil Management and Ecological Restoration” addresses an important and timely topic. However, I found some mistakes that currently limit the paper’s suitability for publication in its current form. Therefore, the author (s) should take major revisions to improve the quality of the paper for publication.

Response: We sincerely thank the reviewer for recognizing the importance and practical relevance of the research topic. We have carefully considered this comment and conducted a comprehensive review and systematic revision of the manuscript to further improve its quality in terms of research scope definition, accuracy of academic expression, and overall standardization.

During the revision process, we optimized the title to more accurately and clearly reflect the study object, temporal scale, and research focus. The revised title explicitly specifies Qilian Mountain National Park as the study area and indicates the study period (1990–2022), thereby avoiding potential ambiguity in spatial and temporal scope associated with the previous, more generalized wording (Title revision, Lines 1–3). In addition, the term “driver pathways” is adopted in the title to emphasize the analytical focus on the pathways through which multiple drivers influence soil erosion dynamics.

Based on the revised title, we further adjusted the statements of research objectives and research questions in the Introduction to ensure consistency in terms of the study area, time span, and conceptual focus. These revisions place clearer emphasis on the spatiotemporal dynamics of soil erosion and its driving pathways (Introduction, Lines 66–73).

 

 

Comment 2: Abstract   The abstract describes clearly the problem and relevance to alpine ecosystems and SDGs. However, there are some grammar mistakes to be corrected. For example, word breaks (“northwest- ern”, “erosion”). The author also should ensure consistency and readability of erosion units (t·km⁻² yr⁻¹ if annual).

Response: We thank the reviewer for the positive evaluation of the Abstract in terms of problem formulation and research significance. In response to this comment, we have carefully revised the Abstract to improve linguistic accuracy and consistency in technical expression. All grammatical errors and formatting issues have been systematically corrected, including inappropriate word breaks (Abstract, Lines 15–28).

In addition, the units used to report soil erosion rates in the Abstract have been standardized. Soil erosion is now consistently expressed on an annual basis using the unit t·ha⁻¹·yr⁻¹, which is fully aligned with the units adopted throughout the main text (Abstract, Lines 22–24). These revisions ensure clarity, consistency, and conformity with disciplinary conventions.

 

Comment 3: Introduction   This section is well-written, comprehensive, and well-structured. The author (s) clearly describes the research problems, the need for this study, and clearly outlines the research objectives and research questions. Minor grammatical issues and punctuation errors (such as extra spaces before citations, sentence fragments). Some sentences are overly long and can be tightened for readability. Repetition of phrases such as “fragile ecosystem” and “soil erosion” can be reduced.

Response: We sincerely thank the reviewer for the positive evaluation of the overall structure, logical coherence, and presentation of the research questions in the Introduction. In response to these comments, we have carefully revised the Introduction and conducted a comprehensive check of the language and formatting throughout the entire manuscript.

Specifically, we conducted a systematic language and format check throughout the Introduction and corrected grammatical and punctuation issues, including extra spaces before citations, sentence fragments, and other text-processing artifacts. These revisions were implemented across the Introduction section (Lines 31–73).

In addition, sentences that were overly long have been streamlined and reorganized to enhance clarity and readability. We also reduced repetitive use of phrases such as “fragile ecosystem” and “soil erosion” by employing context-appropriate synonyms and rephrased expressions, ensuring smoother narrative flow without altering the intended meaning. These refinements are reflected throughout the Introduction (Lines 31–73).

 

Comment 4: Methodology   In this section, the author/s should clarify and validate soil erosion estimation methodology; draw a conceptual diagram justifying SEM pathways and clearly distinguish statistical causality from process-based causality.

Response: We thank the reviewer for the constructive comments on the Methods section. In response to the reviewer's suggestions, we have systematically refined and strengthened the methodological section.

(1) We have clarified the meaning and applicable scale of the RUSLE-based soil erosion estimation in the Methodology section, emphasizing that the RUSLE results represent integrated soil erosion intensity at the regional scale and over long-term periods, rather than a strict separation of natural background erosion and human-induced accelerated erosion. This clarification has been added to Section 2.3.1 (Soil erosion assessment system, Lines 181–191), strengthening the interpretation of erosion indicators and their suitability for regional comparative analysis.

(2) To improve the overall readability and transparency of the methodological framework, we have added and systematically described a conceptual workflow diagram (Figure 2), which integrates multi-source data preparation, RUSLE-based erosion assessment, random forest–based driving factor screening, and PLS-SEM pathway analysis. This diagram is intended to summarize the analytical workflow and justify the logical structure of the SEM pathways, rather than to represent a direct causal mechanism. The description and interpretation of this conceptual diagram have been clarified in Section 2.4 (Statistical analysis, Lines 221–233).

(3) We have explicitly clarified the distinction between statistical causality and process-based causality in the PLS-SEM analysis. In the revised manuscript, we state that the pathways identified by PLS-SEM reflect statistically inferred latent influence relationships based on observed covariation among variables, rather than direct physical or process-based causal mechanisms of soil erosion. This clarification has been added to the Methodology section (Section 2.4, Lines 233–239) and is further reiterated in the Discussion, ensuring that SEM results are interpreted as explanatory relationships at the regional scale and not as definitive process-level causation.

Figure 2. Technical framework for soil erosion assessment and driving mechanism analysis in Qilian Mountain National Park. The upper panels illustrate the preparation of multi-source data, including the spatial distribution of ecosystem types, climatic and environmental variables (e.g., precipitation, aridity index, vegetation and terrain factors), soil properties (e.g., soil organic matter, sand, silt, and clay content), and indicators of human activity. The middle panels depict the sequential analytical methods applied in this study, including soil erosion intensity assessment using the RUSLE model, key driving factor selection and importance ranking based on the Random Forest model, and   identification of potential impact pathways and relative contributions using Partial Least Squares Structural Equation Modeling (PLS-SEM). The bottom panel summarizes the final objective of the framework, which is to interpret soil erosion driving mechanisms across different ecosystem types.

 

 

Comment 5: Results and discussion   This section is well organized though needs some improvements. Some interpretations appear in the results section should be moved to discussion section. Keep results strictly descriptive and quantitative. Authors better to shift interpretation of drivers and relevance of policy in the discussion section.

Response: We thank the reviewer for the constructive and pertinent suggestions regarding the structure and writing hierarchy of the Results and Discussion sections. We fully agree with the reviewer that the Results section should maintain a strictly descriptive and quantitative focus, whereas interpretations of driving mechanisms and policy-related implications should be concentrated in the Discussion section. Following this recommendation, we conducted targeted revisions to the Results section to further strengthen the functional separation between Results and Discussion. Specifically, explanatory or interpretative expressions were revised or removed, and the presentation was refocused on the objective reporting of spatiotemporal patterns, magnitudes of change, proportional relationships, and model-derived outputs.

The specific revisions in the Results section are as follows: Overall spatiotemporal patterns of soil erosion intensity were revised in Lines 300–312, where the text now strictly describes spatial distribution characteristics and temporal trends of erosion intensity without referring to potential driving factors or mechanisms. Changes in erosion magnitude and proportional relationships across different periods were revised in Lines 313–327, with all results presented using quantitative indicators (e.g., change rates, area proportions, and relative differences), and without interpretative language. Differences in soil erosion among land-use or ecosystem types were revised in Lines 328–340, where the results are now limited to numerical comparisons and ranking patterns. Explanations regarding the causes of these differences have been removed from the Results section. Model-derived outputs (RUSLE, RF, and PLS-SEM results) were revised in Lines 341–358, where the section now presents model results objectively (e.g., variable importance rankings and path coefficients), without discussing causal mechanisms or management implications.

It should be noted that interpretations concerning the mechanisms of soil erosion drivers, explanations of differences among ecosystem types, and implications for ecological management and policy have already been systematically addressed in the Discussion section and remain there in the revised manuscript (Discussion, Lines 400–492).

 

Comment 6: Conclusion   The conclusion clearly summarizes key quantitative findings. However, still it needs some amendments. There are some grammatical errors such as “During last three-decade” should be corrected as “During the past three decades” and Capitalization errors (“Conversion”).

Response: We sincerely thank the reviewer for the careful review of the Conclusions section and for the specific and constructive suggestions provided. We fully agree with these comments and have revised and standardized the Conclusions section accordingly.

Specifically, grammatical errors in the Conclusions have been corrected, for example, the phrase “During last three-decade” has been consistently revised to “During the past three decades”. In addition, inappropriate capitalization issues (e.g., the capitalization of common nouns such as “Conversion”) have been corrected to conform to standard academic writing conventions. We also made minor adjustments to wording and phrasing to improve grammatical accuracy, linguistic coherence, and overall academic rigor of the Conclusions section. These revisions do not alter any of the original research findings or quantitative results (Conclusions section, Lines 523–545).

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

 

Comments for author File: Comments.pdf

Author Response

To Reviewer #3

Comment: This article provides an excellent and timely assessment of soil erosion over the last 30 years within the Qilian Mountain National Park, which is part of the highly sensitive alpine-arid ecosystem located at the northeastern end of the Tibetan Plateau. Additionally, utilizing multiple sources of remote sensing data, Random Forest, and structural equation modelling to examine and understand spatiotemporal changes in erosion and their causes offers an advanced insight into this process. The manuscript is well structured, has a comprehensive methodology, and is well-drafted. Despite these attributes there are several areas requiring improvement; specifically, validation based on ground data, limitations of resolution provided via satellite images, and alternative interpretations of the anthropogenic factors affecting erosion. A subsequent review of each of these key areas has been provided along with suggestions to improve the quality of your manuscript.

Response: We sincerely thank the reviewer for the positive and encouraging evaluation of our manuscript. We appreciate the recognition of the scientific value of this study, particularly regarding the integrated use of multi-source remote sensing data, Random Forest, and structural equation modelling to investigate long-term soil erosion dynamics in the Qilian Mountain National Park.

We fully acknowledge the reviewer's comments concerning ground-based validation, spatial resolution limitations of satellite data, and the interpretation of anthropogenic drivers. In response, we have revised the manuscript to clarify the scale-dependent applicability of our approach, explicitly discuss the limitations associated with data resolution and the lack of field-based validation, and further refine the interpretation of human activity indicators within a regional-scale context. These issues are now more clearly addressed in the Methods, Discussion, and the “Limitations and Future Perspectives” sections.

 

Comment 1: Resolution Limitations: The study mentions that remote sensing datasets operate at a 1 km resolution. This means they struggle to accurately capture micro-scale erosion. Because of this limitation, it tends to underestimate erosion in small, steep areas.

Response: We sincerely thank the reviewer for the professional comment on the limitations of the spatial resolution of the remote sensing data. We fully agree that a spatial resolution of 1 km has inherent limitations in characterizing fine-scale erosion processes, particularly in potentially underestimating erosion intensity on steep slopes and in localized high-risk areas.

In the revised manuscript, we have clarified and expanded the discussion of resolution-related limitations. Specifically, in the Discussion and future perspectives section (Section 4.6, Lines 486–497), we explicitly state that a 1 km resolution is insufficient to capture micro-topographic features, such as narrow gullies, steep slopes, and patchy exposed soil surfaces, which are often critical for erosion initiation and development. As a result, erosion risk in localized high-risk zones may be conservatively estimated at the regional scale.

At the same time, we emphasize that the adopted spatial resolution remains appropriate for the regional-scale and long-term analysis ( Discussion and future perspectives Section 4.6, Lines 486–497) conducted in this study. The consistency of erosion patterns across time and space at this resolution supports the robustness and comparability of the results for identifying broad spatial gradients and long-term trends, rather than fine-scale process dynamics.

 

Comment 2: Uncertainty Quantification: The outputs from RUSLE lack any detailed sensitivity analyses or confidence intervals. This absence makes it hard to interpret the uncertainty around the erosion estimates.

Response: Thank you for the reviewer’s important comment regarding the lack of explicit uncertainty quantification in the RUSLE-based erosion estimates. We fully acknowledge that characterizing uncertainty is essential for interpreting model outputs at the regional scale. In response to this concern, we have added a one-at-a-time (OAT) sensitivity analysis in the revised manuscript to quantitatively assess how uncertainties in individual RUSLE factors propagate to soil erosion estimates.

Specifically, under the assumption that other factors remain constant, we systematically perturbed rainfall erosivity (R), soil erodibility (K), slope length and steepness (LS), and the cover-management factor (C) within plausible uncertainty ranges (R ±15%, K ±20%, LS ±10%, and C ±25%). The resulting variations in estimated soil erosion intensity were quantified and summarized. This sensitivity analysis and its results have been added to the Supplementary Materials (Table S3), and the corresponding methodological description has been included in the Discussion section (Section 4.6, Lines 497–505).

The results indicate that the RUSLE outputs exhibit a nonlinear response to uncertainty in individual factors. Among the tested factors, uncertainties in the cover-management factor (C) and soil erodibility (K) exert the strongest influence on the magnitude of erosion estimates, whereas the sensitivity to LS is relatively lower. It is important to note that, although uncertainty in individual factors affects the absolute values of erosion intensity, it does not alter the spatial pattern of erosion nor the relative comparisons among different ecosystem types discussed in the manuscript. This clarification ensures that the main conclusions regarding spatial variability and ecosystem-level differences remain robust despite inherent model uncertainty (Table S3).

 

Table S3. One-at-a-time sensitivity analysis of RUSLE factors using factor-specific perturbation ranges

Factor perturbed	Perturbation range	Mean erosion (t·ha⁻¹·yr⁻¹)	Relative change (%)
Baseline	—	277	—
Rainfall erosivity (R)	−15% ~ 15%	262~292	−15.00 ~15.00
Soil erodibility (K)	−20% ~ 20%	257~297	−20.00 ~20.00
Slope length–steepness (LS)	−10% ~ 10%	267~287	−10.00 ~10.00
Cover-management (C)	−25% ~ 25%	252~302	−25.00 ~25.00

Note: Sensitivity analysis was conducted using a one-at-a-time perturbation approach. Each RUSLE factor was independently varied within a conservative uncertainty range (R ±15%, K ±20%, LS ±10%, and C ±25%) while keeping the other factors constant. Values represent mean relative changes in soil erosion intensity across the study area.

 

Comment 3: Causal Pathways: The results from PLS-SEM are helpful, but they could be clearer. Adding ecosystem specific visualizations in supplementary materials, like path diagrams that separate by land type, would enhance understanding...

Response: We sincerely thank the reviewer for the valuable suggestion regarding the clarity of the causal path representation in the PLS-SEM analysis.

It should be noted that, in the original manuscript, we have already constructed and presented ecosystem-specific PLS-SEM pathway models. Specifically, Figure 6 displays separate SEM results for cropland, forest, shrubland, grassland, and bare land ecosystems. Each sub-panel explicitly illustrates the relative importance of driving factors identified by the random forest analysis, as well as the corresponding PLS-SEM pathways linking climate variables, vegetation, and human activity factors to soil erosion. This figure and its interpretation are described in detail in the Results section (Section 3.4, Lines 331–347) and further discussed in the Discussion section (Section 4.4, Lines 401–423).

Given that ecosystem-specific pathway diagrams are already fully presented in the main text, we did not duplicate the same diagrams in the Supplementary Materials in order to avoid redundancy and to maintain a concise manuscript structure. Instead, we ensured that Figure 6 is sufficiently detailed and clearly annotated to allow direct comparison among ecosystem types.

 

Comment 4: Representation of Human Activity: The Human Footprint Index, along with GDP and population density proxies, does a decent job. Still, their effectiveness is limited due to the absence of local land use variables. Factors like grazing pressure and land ownership would provide more specificity in policy implications.

Response: We agree with the reviewer's perspective that the Human Footprint Index (HF), together with GDP and population density, essentially represents macro-scale proxy indicators of human activity intensity at the regional level. These indicators are primarily used to characterize the overall level of human disturbance and its spatial distribution patterns, but they are not able to further distinguish more detailed land management processes, such as grazing intensity, land tenure, fencing practices, or specific cultivation methods. This limitation indeed constrains the degree of refinement that can be achieved in management-oriented policy recommendations.

In response to this issue, we have implemented targeted revisions and clarifications in the revised manuscript. First, in the Data Sources section (Section 2.2, Lines 120–165) and in Section 4.6 (Limitations and Future Perspectives, Lines 493–525), we explicitly clarify the applicable scale and limitations of human activity indicators. We emphasize that these variables are intended to characterize gradients of human disturbance and development intensity at the national park scale, rather than to represent specific land management measures.

Second, in Section 4.5 (Policy and Management Implications, Lines 464–492), we improve the specificity and practical relevance of policy recommendations by differentiating management priorities across ecosystem types. For example, erosion control and vegetation stabilization measures are prioritized in bare land and highly sensitive areas; regulation of grazing pressure is emphasized in grassland and shrubland ecosystems; and soil and water conservation practices are strengthened in cropland areas. This approach enhances the applicability of policy recommendations while remaining within the explanatory boundaries of the available data.

Finally, in Section 4.6 (Limitations and Future Perspectives, Lines 493–525), we further emphasize that future studies should prioritize the integration of more detailed ground-based management data, such as grazing intensity statistics, land-use management records, and field survey data, and should combine these with UAV observations or high-resolution satellite imagery. Such efforts would improve the representation accuracy of human activity factors and support more management-oriented and operational policy analyses.

 

Comment 5: Validation: There is no validation using field measurements or erosion plot data, which is critical for assessing the realism of model estimates.

Response: We thank the reviewer for the critical and valuable comment regarding field-based validation. We agree that field observations derived from erosion plots, sediment traps, or long-term ground monitoring stations play an irreplaceable role in evaluating the absolute accuracy and physical realism of model-based erosion estimates. However, Qilian Mountain National Park is characterized by complex topography, high elevation, diverse ecosystem types, and a vast spatial extent. At present, long-term and spatially representative datasets of erosion plots or sediment flux measurements covering the entire study area are not available. As a result, direct quantitative validation based on field observations cannot be conducted at this stage.

Under this constraint, the primary objective of this study is not to precisely calibrate absolute erosion rates, but rather to focus on identifying regional-scale spatial patterns of soil erosion, long-term temporal dynamics, and the relative roles of different driving mechanisms. To avoid over-interpretation of the model results, the lack of field-based validation has been explicitly identified as a key limitation in the revised manuscript (Section 4.6, Limitations and Future Perspectives, Lines 493–525).

In addition, we describe feasible alternative validation approaches adopted in this study. These include the spatial consistency between high-erosion-risk areas and steep slopes, bare land, and regions with low vegetation cover, as well as the consistency between erosion intensity levels across different ecosystem types and results reported in previous studies conducted in the Qinghai–Tibet Plateau and adjacent alpine–arid regions (Section 4.6, Lines 493–525). It is explicitly stated that these comparisons are intended to support the plausibility and robustness of the results, rather than to serve as direct field-based validation of erosion magnitude.

Furthermore, the revised manuscript emphasizes that future research will prioritize the establishment of long-term erosion monitoring plots in representative ecosystem types and high-risk erosion areas. By integrating erosion plots, sediment collection devices, rainfall simulation experiments, and UAV observations with high-resolution satellite imagery, multi-scale and multi-method field validation frameworks will be developed to progressively improve the reliability of regional-scale erosion models in estimating absolute erosion rates (Section 4.6, Lines 493–525). We again sincerely thank the reviewer for this constructive suggestion.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Corrections have been made and explanations given.

Still, the authors mention limitations in the availability of long-term data on historical land management practices, disturbance intensity, and detailed human activity processes, or that detailed records of ecological restoration engineering are not consistently available.

For their purpose, they mention that for regional-scale analyses, the resolution is sufficient to capture the spatial distribution of erosion hotspots, or that the approach is consistent with the objectives of regional-scale assessment. Also, my inquiries for a more detailed technical explanation of steps taken to reach the results are covered by the sources the data was taken from. Although there is a park administration who probably has some data, I do not know how available this information for public / academia is.

I will give an OK, but let the decision for the editorial board, preferably combined with the opinions of at least one other reviewer.

Kind regards!

Author Response

Comment 1:Corrections have been made and explanations given.

Still, the authors mention limitations in the availability of long-term data on historical land management practices, disturbance intensity, and detailed human activity processes, or that detailed records of ecological restoration engineering are not consistently available.

For their purpose, they mention that for regional-scale analyses, the resolution is sufficient to capture the spatial distribution of erosion hotspots, or that the approach is consistent with the objectives of regional-scale assessment. Also, my inquiries for a more detailed technical explanation of steps taken to reach the results are covered by the sources the data was taken from. Although there is a park administration who probably has some data, I do not know how available this information for public / academia is.

I will give an OK, but let the decision for the editorial board, preferably combined with the opinions of at least one other reviewer.

Response:We thank the reviewer for this clarification. We note that, to the best of our knowledge, detailed land management or erosion-related data that may be held by park management authorities are not publicly accessible or consistently available to the academic community. Consequently, such datasets could not be incorporated into the present study. This constraint has been explicitly stated in the revised manuscript as a limitation related to data availability (Section 4.6, Limitations and future perspectives, Lines 511–513).

 

Author Response File: Author Response.pdf