Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Manuscript Comments

Summary

This manuscript tackles the non‑grain changes of cropland in Yunnan Province using multi‑source data (crop yield, land use, remote sensing imagery and population statistics) from 2001–2021. The authors conducted a dynamic spatiotemporal clustering framework (DSTCM) using time-series analysis, spatial autocorrelation, multiple regression and the CLUE‑S model. It displayed the temporal trends, spatial heterogeneity and influencing factors of non‑grain conversion, and predicted the trends with different scenarios. This paper reported that the non‑grain index decreased with fluctuations, and strong spatial clustering of non‑grain conversion was shown in the north‐western and south‐western Yunnan. The significant drivers includes soil fertility, cropping structure, GDP level and technology factor. The model achieved good accuracy of the forecasting period. The topic was relevant to food security, and the research methods and results were valuable. However, it should be further revised before accepted by the journal of Remote Sensing.

Major Issues and Recommendations

• The abstract should be improved with concise expression. Currently, the expression was not very concise, and the logical was not very clear. The reginal issue of non-grain conversion should be improved in key sentence in the Abstract part and the thorough literature summary in the Introduction part, which located in the high mountain and gorge area of the Yunnan-Guizhou Plateau under the background of karst landform. What is the difference of non-grain conversion with other regions?
• Definition of the “non‑grain index” should be improved. The key indicator is poorly defined. Equation (4) is incomplete, symbols (e.g., weights , NDVI variables) are not explained. The manuscript did not provide a clear mathematical formulation or rationale for the weighting scheme.
• Introduction of multi‑source data processing was not very clear. The details of quality control and data harmonization were lacking. For example, how to handle the differences in spatial resolution and temporal coverage across Landsat, Sentinel and statistical data? How to address the missing values and smooth the time series data? If possible, please provide a flowchart of data processing steps.
• Model assumptions and parameter determination. The parameter settings of employed models were lacking. For example, please add the stationarity tests, order selection (AIC/BIC) and diagnostic checks for time series models. driving factors analysis and its uncertainty should be calibrated and validated for CLUE‑S. The SVM parameters (C = 1, γ = 0.1) should be given more details, such as model tuning processing and cross‑validation with goodness‑of‑fit statistics.
• Implications of results and : Some statements appear contradictory or unclear. The authors state that the non‑grain index declined while the non‑grain conversion process intensified, which is confusing. Clarify whether a lower index value indicates more or less non‑grain conversion. In the spatial analysis, high Moran’s I and high Geary’s C values are reported simultaneously for some regions; explain the meaning of these measures and how they can be reconciled. Provide confidence intervals or uncertainty estimates for key results and predictions, and discuss how sensitive the conclusions are to data and model choices.
• The figures and tables should be improved. Tables listing model parameters (e.g., Table 2) should list the variables definition, as well as units and sample periods. Maps should include scale bars and coordinate grids. The text annotations in the illustrations should be clearly visible when the document is viewed at 100%, such as Figure 2. Some maps should be supplemented to show spatial heterogeneity, such as non-grain parcel distribution maps, spatial clustering maps, etc.
• English language should be improved and many formatting issues should be revised. The manuscript contains many long sentences and grammatical errors etc. Formula numbering is inconsistent and some equations are not properly formatted. Partial references were cited without complete journal names and page numbers. The whole paragraphs should be improved in the concise and logical style, which is more friendly to readers.
• Study area description. The subdivision of Yunnan into five regions required justification, included a table summarizing the area, population and dominant crops of each region and provided a location map.
• Notation. Provide a symbol table or explain all variables when equations are introduced. Several variables showed without explanation.

Overall Evaluation and Recommendation

The manuscript addressed a hot topic: non‑grain conversion of reginal cropland by integrating multi‑source data and various analytical methods. With substantially major revision, this manuscript could be reconsidered to be accepted by Remote Sensing for publication.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript examines the non-grain conversion of cultivated land in Yunnan Province utilizing multi-source data (crop yield, land use, remote sensing imagery, and population statistics) spanning 2001–2021. The authors develop a dynamic spatiotemporal clustering model (DSTCM) that integrates time series analysis (ARMA/ARIMA/SARIMA), spatial agglomeration metrics (Moran's I, Geary's C, Getis-Ord Gi*), multiple linear regression, and the CLUE-S model to characterize temporal fluctuations, spatial patterns, and driving factors of non-grain conversion, while forecasting future scenarios influenced by climate, policy, and market dynamics. Key findings include a phased decline in the non-grain index with fluctuations; pronounced spatial clustering in central, southwestern, and southeastern Yunnan; significant influences from soil fertility, crop types, GDP levels, and technology; and model validation accuracies exceeding 98% for 2022–2024, projecting a continued index decline to 2035 under optimized conditions. The research is pertinent to remote sensing applications in agricultural monitoring and aligns with emerging concerns in food security amid land use shifts. The multi-method fusion and scenario-based predictions represent a notable advancement over static analyses. Nevertheless, to align with Remote Sensing's rigorous standards for methodological transparency, validation robustness, and integration with state-of-the-art techniques, the manuscript requires substantial refinements in index formulation, data fusion protocols, and comparative benchmarking against recent advancements.

(1)Definition of the “non-grain index”: The foundational metric lacks comprehensive explication, rendering interpretations ambiguous. Equation (4) aggregates sub-indices from diverse data sources but omits explicit variable derivations (e.g., NDVI computation in Eq. 7, land cover change index), weighting methodologies, and sensitivity assessments. For instance, how are weights (α, β, γ, δ) calibrated—via expert judgment, entropy methods, or empirical fitting? This opacity hampers reproducibility and comparability with contemporary indices in Yunnan-focused studies, such as those employing Google Earth Engine for cropland mapping with >93% accuracy.To enhance scientific validity, furnish a complete algorithmic workflow, including pseudocode for index computation, and validate against alternative formulations (e.g., incorporating hyperspectral indices for crop discrimination).

(2)Transparency in multi-source data processing: While preprocessing steps (e.g., radiometric correction via ENVI, interpolation for gaps) are outlined, critical details on harmonization across heterogeneous sources are absent, potentially introducing biases in spatiotemporal fusion. For remote sensing data (Landsat/Sentinel at 30m), specify cloud masking protocols (e.g., Sen2Cor or CFMask) and temporal compositing strategies to mitigate Yunnan’s frequent cloud cover, which could skew NDVI trends. Moreover, report confusion matrices or producer/user accuracies for land cover classification, and quantify error propagation from raster-to-vector conversions. Recent Yunnan cropland studies highlight the efficacy of multi-sensor fusion on Google Earth Engine platforms for sub-decadal mapping. Supplement with a detailed data pipeline diagram and uncertainty metrics (e.g., Monte Carlo simulations) to bolster reliability.

(3)Model assumptions and parameter selection: The ensemble of models (time series, spatial statistics, regression, CLUE-S with SVM optimization) is ambitious, yet justifications for selections and hyperparameter tuning are insufficient, risking overfitting or underperformance. For ARIMA/SARIMA, detail differencing orders (d), ACF/PACF diagnostics, and stationarity tests (e.g., KPSS alongside ADF). In CLUE-S, elucidate transition rule derivations, elasticity parameters, and calibration against historical data, particularly for topographic constraints in Yunnan’s karst landscapes. SVM parameters (C=1, γ=0.1) warrant grid search or Bayesian optimization evidence; compare with alternatives like random forests in analogous non-grain analyses. Incorporate cross-validation folds, residual plots, and scenario sensitivity analyses to affirm model robustness beyond reported Kappa values.

(4)Interpretation of results: Certain assertions exhibit inconsistencies or require deeper contextualization, undermining clarity. The "decline-fluctuation-decline-rise" phasing in the non-grain index (Fig. 3) contrasts with demographic pressures implying escalation, necessitating explicit linkage to external drivers like the Returning Farmland to Forest Program's impacts on non-flatlands. Spatial metrics (e.g., Moran's I=0.57) indicate clustering, but reconcile with Geary's C interpretations and provide local indicators (LISA maps) for hotspot validation. Predictions (Figs. 7–8) claim high accuracy (>98%), yet overlook confidence bands or ensemble forecasting to address uncertainties in climate scenarios. Elaborate on index directionality (lower values denoting reduced non-grain) and sensitivity to assumptions, drawing parallels with recent NPP trend analyses in Yunnan.

(5)Statistical rigour in the analysis of driving factors: t-tests (Tabs. 6–7) assess regional influences but neglect essential diagnostics such as multicollinearity (VIF), heteroscedasticity (Breusch-Pagan), or interaction terms, potentially inflating significance (e.g., p=0.047 for crop types). The geodetector method, mentioned abstractly, should be fully implemented to quantify factor interactions (q-statistics) and compared with structural equation modeling for causal pathways. Differentiate proximate (e.g., soil fertility) from underlying drivers (e.g., policy), avoiding over-reliance on p-values near thresholds. Align with recent non-agriculturalization impact studies in Yunnan, which integrate GIS for multi-factor attribution. Augment with bootstrapped confidence intervals and path analysis to delineate hierarchical influences.

(6)Presentation of figures and tables: Visual aids are informative yet deficient in annotations and standardization. Figures (e.g., 3–8) require legends for color ramps, scale bars, and north arrows on maps; temporal plots should include error shading. Tables (e.g., Tab. 3 factor classification) omit units and sources; Tab. 8 scenario weights need justification via AHP or similar. To improve accessibility, adopt high-resolution formats and embed quantitative summaries (e.g., trend slopes). Reference exemplary presentations in recent Yunnan vegetation productivity mappings.

(7)English language and formatting: The text features protracted sentences, inconsistent terminology (e.g., "non-grainization" vs. "non-grain conversion"), and grammatical lapses (e.g., missing articles, awkward phrasing like "the non-grain conversion trend also poses challenges"). Equations are misnumbered or malformed (e.g., incomplete symbols in Eqs. 1–3); references lack uniformity (e.g., incomplete DOIs, varying styles). Professional proofreading is imperative, alongside adherence to journal guidelines for SI units, figure captions, and LaTeX formatting for equations.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

This study integrates time series analysis, spatial agglomeration analysis, and the CLUE-S model to systematically investigate the spatiotemporal pattern characteristics and driving factors of non-grain conversion of cultivated land in Yunnan Province from 2001 to 2021, and predicts the trends from 2025 to 2035.

Although the manuscript is generally very detailed, it has many issues in terms of organization and writing approach. 
First, it fails to concisely and clearly elaborate on the technical innovations and research methods proposed in this study. Second, a large amount of contents exhibit significant redundancy.
Third, while the manuscript analyzes the changing trends of non-grain conversion of cultivated land in Yunnan from multiple perspectives, the connection between these conclusions and the technical roadmap is not well emphasized. 
Finally, whether there are significant differences among the results of different analyses seems to lack separate discussion.
Other suggestions are as follows.

1. Line 81, what does "significant positive spatial correlation" refer to? This statement is ambiguous.  

2. The second paragraph of Section 1 only cites 7 previous references ([4] to [10]), which is insufficient to convincingly illustrate the identified shortcomings (lines 86–90). Details should be supplemented to clarify the research gaps more specifically, such as the nature of specific regional differences, which influencing factors may interact with each other, and the specific aspects included in the spatiotemporal distribution differences among various non-grain conversion regions. These discussions should be integrated with the innovations of this study in the following paragraph to highlight its distinctive features.  

3. The first sentence of the last paragraph in Section 1 is overly lengthy. It should be rewritten, possibly split into shorter sentences, to strengthen the logical flow of the paragraph. Redundancies in this paragraph should be reduced. Ensure that specific operations are linked to specific innovations, and avoid unsubstantiated claims. For example, instead of emphasizing "…from multiple dimensions and angles," clarify which specific angles are addressed in this paper. Similarly, specify the "multi-dimensional natural and socio-economic factors" incorporated into the CLUE-S model and the "various policy intervention and market fluctuation scenarios," etc. Additionally, the writing style of the first section resembles a government report rather than an academic paper.  

4. Please provide supplementary information about the remote sensing data used, such as the number of observations, observation periods, and data intervals.  

5. An overall flowchart is needed in Section 3, which should outline the complete procedure of the proposed methodology to illustrate the integration of the three involved methods.  

6. For Equation (7), what is the basis for defining this new RSI feature? Why not use 𝑁𝐷𝑉𝐼, 𝐿𝐶𝐼, and 𝑉𝐹𝐶 as independent variables in this research? Similarly, please explain the rationale behind Equation (8): why are GDP and population combined instead of being treated as independent variables?  

7. DSTCM appears to be the key innovation of this study and should be emphasized in the abstract and introduction. An overall flowchart is recommended to illustrate its principle. The second paragraph of Section 3.2 is excessively long; while it aims to highlight the advantages of the proposed DSTCM, it fails to explain how these advantages are achieved. Some content in this paragraph should be moved to the discussion section, supported by convincing details from comparative experimental results.  

8. Tables 2 and 3 should be explained with the aid of figures. Additionally, redundant content in the appendices should be removed. Table 2 also has formatting issues.  

9. Multiple formatting issues exist, including those related to tables, references, and statements.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

This manuscript focuses on extraction and prediction of spatiotemporal pattern characteristics of farmland non-grain conversion in Yunnan Province. Spatial autocorrelation analysis and CLUE-S model were analyzed. It is is of great practical signiicance for promoting agricultural and rural modernization and regional sustainable development, but it still needs revision and check before formal publication. Here are some specific suggestions for the authors:

1. In the Figure 1, authors is recommended to give a scale bar and north arrow for both the Yunnan and China map, which are a basic and important elements for a good map. In particular, in the map of upper right corner of the Figure 1, China's administrative boundaries should include the South China Sea islands. It is better to have a approval number for the China's administrative boundaries.
2. Line 192. “….geometrically corrected to ensure spatial location accuracy.” As for the geometric correction, authors can give some more information. For example, How many control points are selected for geometric correction? What is the RMS value?
3. Line 194. As for the land use classiication, what kind of the classiication was conducted for the image? What is the classification accuracy of land use for the different time points 2001 and 2021, which will impact the prediction accuracy.
4. For the Section of 4.2 and Figure 4.

Some spatial autocorreation index such as Moran’s I, G-statistic, Geary's C have P value and Z score to testing the significance for judge the clustering or dispersion. So if authors can give these values or explain for this, it would be better.

6. For the Section of 4.3 and Figure 6.

Some inluencing factors such as GDP level is a strong correlated with Cost of agricultural production, Prices of agricultural products and Agricultural subsidy policy. In addition, another some geographical factors such as the DEM, roads, rivers and streams, etc also might influence the conversion about non-grainification of cultivated land. So authors could be recommended to give some principles of selecting the driving factors.

7. English should be polished, especially for description of the calculation formula.

Comments for author File: Comments.pdf

Comments on the Quality of English Language

Not Good. English should be polished.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

This manuscript was well modified to response the comments. However, few issues should be further improved as follows:

1) For the last paragragh of the Introduction part, the importance was emphasized. It was suggested to add the stepwise introduction of research process.

2) The Conlusion part should be put more emphsis and further well-organized again for readers.

3) Could you give an spatial distribution maps of non-grain identification result for Yunnan Province?

4) The authors have highlighted the importance of study area in the Yunnan-Guizhou Plateau karst region. Could you improve the Figure 1 with the relationship of Yunnan Province, Karst region of study area and Yunnan-Guizhou Plateau, which can help readers to make a good understanding the importance of this study.

5) Figure 4 was suggested to be improve with the vertical axis title and unit.

Author Response

Please see the attachment.

Author Response File: Author Response.docx