Spatial Prediction of Soil Organic Carbon Based on a Multivariate Feature Set and Stacking Ensemble Algorithm: A Case Study of Wei-Ku Oasis in China

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript ‘Spatial inversion of soil organic carbon based on multivariate 2 feature set and stacking ensemble algorithm: a case study of Wei-Ku oasis in China’ falls within the topics covered by Sustainability. The work, although complex in the analytical procedure adopted, is not particularly innovative because in recent years, after the launch of the Copernicus Sentinel missions, the literature on the use of remote sensing for the prediction of parameters such as organic C content in the soil has increased greatly. However, the multivariate technique tested increases the predictive capability of machine learning algorithms and deserves to be made known.

Before publication, it is necessary to improve the fluency of the text and make some useful changes to make the manuscript clearer from a scientific point of view. The main points to consider are listed below.

Title - Spatial inversion
The terminology is obscure to me, but it seems to be used in the field of remote sensing. If I understand correctly, spatial inversion is equivalent to spatial prediction which uses auxiliary variables that are supposed to be more or less numerically correlated with the target variable. I would suggest either including a brief explanation of the meaning of “spatial inversion” in the introduction or replacing it with “spatial prediction”.

Line 11 - Limited observation capability

I would remove this part of the sentence because it enters a strictly epistemological topic: limited observation capability is a fact in the natural sciences and is one of the main reasons why statistics was born!

L. 43 - Efficiency of remote sensing in SOC prediction I would not emphasise this aspect too much because, after all, predictions concern the C content in the first 10-15 cm of soil, whereas ground-based measurements may concern at least the 30-40 cm thickness of ploughed horizons. Remaining on costs, I personally could not find quantitative data on the lower spatial prediction costs of remote sensing and the citation [3] does not give specific information. It would be very interesting if in the future the authors made a comparison of the costs of sampling/data analysis (which, excluding survey planning, I would estimate to be around 12 working days) with those needed to acquire and process Sentinel data prior to multivariate analysis.

A final aspect, which does not only concern the manuscript, is that predictions with remote sensing almost always divide the data set into a training set and a validation set but do not use independent validation sets based on random sampling. Such an approach would increase costs but ensure a more objective assessment of the accuracy of the predictions.

L. 53 - estimation
I suggest using the term “prediction” instead of “estimation”. From a statistical point of view estimation applies to sample statistics obtained from randomised sampling designs, while prediction concerns non-random sampling such as purposive sampling used in this study.

L. 140
It is essential to include more information on the soil types present in the Wei-Ku oasis in order to understand the processes affecting soil organic matter in that area. Incidentally, does ‘tidal soil’ mean ‘fluvial soil’?

L. 153-157
The sampling design adopted is clearly purposive, but the land-use criterion for the choice of observation points is not sufficient to explain their homogeneous distribution in the area, and in my opinion more details on the selection criteria are needed.

L. 156

Please give an explanation of “garden land”.

L. 192
I would suggest not going into too much detail about the correlations between spectral bands and soil organic matter metabolism: there are many biological and biochemical processes involved and their interaction is extremely complex to be simply summarised by the share of reflected solar radiation.

Fig. 3 e Fig. 6

In the soil classifications, the three samples with SOCC > 18% would be classified as organic horizon samples. In my opinion they are to be considered outliers and should be removed from the data set.

Author Response

评论者 #1：

在发表之前，有必要提高文本的流畅性并进行一些有用的更改，以使手稿从科学的角度更加清晰。下面列出了需要考虑的要点。

1. 标题 - 空间反转。这个术语对我来说很晦涩，但它似乎被用于遥感领域。如果我理解正确，空间反转相当于空间预测，它使用应该或多或少与目标变量数值相关的辅助变量。我建议要么在引言中包括对“空间反转”含义的简要解释，要么用“空间预测”代替它。

回应：我们衷心感谢审稿人的建设性建议。作为回应，我们在标题部分将术语“空间反转”替换为“空间预测”。所有修改都已仔细实施，并以青色突出显示，以便于参考。

 

2. 第 11 行 - 观察能力有限。我会删除这部分句子，因为它进入了一个严格的认识论话题：有限的观察能力是自然科学的一个事实，也是统计学诞生的主要原因之一！

回应：我们衷心感谢审稿人的宝贵意见和深刻见解。作为对这些建议的回应，我们删除了与“有限观察能力”相关的表达，以避免不准确的认识论讨论。此外，我们还对这些陈述进行了补充和完善，以强调研究目的（有关详细信息，请参阅第 10-15 行）。

 

3. L. 43 - 土壤有机碳预测中的遥感效率我不会过多强调这方面，因为毕竟，预测涉及土壤前 10-15 厘米的碳含量，而地面测量可能至少涉及 30-40 厘米的犁地层厚度。继续在成本上，我个人无法找到关于遥感较低的空间预测成本的定量数据，而且引文 [3] 没有给出具体信息。

回应：我们对审稿人的中肯建议表示衷心的感谢。我们仔细修改了新手稿第 41-43 行的讨论，其中相关更改以青色突出显示，以便于识别。

 

4. 如果将来作者将采样/数据分析的成本（不包括调查计划，我估计约为 12 个工作日）与在多变量分析之前获取和处理 Sentinel 数据所需的成本进行比较，那将非常有趣。

回应：感谢您的深思熟虑的建议。在我们目前的研究中，采样和数据分析过程不仅包括实地调查，还包括土壤样品制备、实验室分析和数据预处理。这些任务需要大量的时间和精力。此外，Sentinel 数据的采集和处理（包括云清洁、植被指数计算和变量提取）已纳入工作流。我们同意，对野外方法和遥感方法之间的劳动力和时间成本进行详细比较是未来研究的宝贵方向，我们非常感谢您在这方面的见解。与遥感估计方法和样本数据相关的缺点已在讨论部分进行了补充。详情请参阅「4.3.局限性和未来的工作“（手稿第 698-720 行）。

 

5. 最后一个方面，不仅与手稿有关，而且遥感预测几乎总是将数据集分为训练集和验证集，但不使用基于随机抽样的独立验证集。这种方法会增加成本，但可以确保对预测的准确性进行更客观的评估。

回应：我们衷心感谢审稿人关于独立验证的建设性建议。作为回应，我们在第 721-736 行中添加了关于该方法考虑的专门讨论，这些讨论以青色突出显示，以便于参考。

 

6. L. 53 – 估计。我建议使用术语 “预测” 而不是 “估计”。从统计角度来看，估计适用于从随机抽样设计中获得的样本统计量，而预测涉及非随机抽样，例如本研究中使用的目的抽样。

回应：非常感谢您的宝贵反馈。我们已将第 53 行（现在的第 55 行）中的“估计”替换为“预测”。对于此修改，我们在整个手稿中进行了全面的检查和替换。

 

7. L. 140：必须包括更多关于 Wei-Ku 绿洲中存在的土壤类型的信息，以便了解影响该地区土壤有机质的过程。顺便说一句，“潮汐土”是指“河流土壤”吗？

回应：我们衷心感谢您的宝贵意见。在新手稿中，我们在第 147-153 行提供了对 Wei-Ku 绿洲土壤类型的详细描述。所有修改都以青色清晰突出显示。根据审稿人的意见，我们小心翼翼地将术语“潮汐土”替换为更准确的“河流土壤”，特别是在土壤分类部分，以确保土壤使用正确的学术术语。

 

8. L. 153–157：所采用的抽样设计显然是有目的的，但选择观测点的土地利用标准不足以解释它们在该地区的均匀分布，我认为需要有关选择标准的更多细节。

回应：我们衷心感谢审稿人对我们的抽样方法提出的宝贵建议。在修订后的手稿中，我们显着增强了修订手稿中第 171-174 行采样点分布的描述，并以青色突出显示以方便参考。

 

9. L. 156请解释一下“花园土地”。

回应：非常感谢您的宝贵反馈。我们用花园土地系统中植被类型的具体示例补充了手稿此信息已在手稿中进行了适当补充，并在第 174-181 行以青色突出显示。

 

10. L. 192：我建议不要过多地详细讨论光谱带和土壤有机物代谢之间的相关性：涉及许多生物和生化过程，它们的相互作用极其复杂，不能简单地用反射太阳辐射的份额来概括。

回应：我们感谢您的反馈。我们同意审稿人的观点，并根据您的建议仔细审查和修改了手稿。修改后的文本在手稿的第 214-220 行以青色突出显示，供您参考。

 

11. 图 3 e 图 6在土壤分类中，SOCC > 18% 的三个样品将被归类为有机水平样品。在我看来，它们应该被视为异常值，应该从数据集中删除。

回应：感谢您对 SOCC 数据的关注。对于 SOCC 为 >18% 的 3 个样本，我们认为保留这些数据更符合实际调查情况。这些高值来源于特定的有机质富集环境（如肥料管理、特殊水文条件、植被输入等），其值与当地实际情况相符。通过化学分析证实了数据的准确性，并参考前几年的监测数据验证了该区域存在如此高的值。为保证数据的真实性和代表性，未进行排除，客观反映土壤有机碳的自然分布特征。同时，有机碳的空间异质性在第 4 部分的第 698-720 行中讨论。

我们衷心感谢审稿人的辛勤努力。我们相信所做的修订会得到批准。我们衷心感谢您的宝贵意见和建议，这些意见和建议对提高我们的稿件质量具有重要意义。

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript takes the Wei-Ku Oasis, a typical arid oasis in Northwest China, as a case study and fills the gap in high-precision soil organic carbon (SOC) estimation in this region. The study is valuable, but the manuscript could be further improved and deepened in the following aspects:

It is suggested that the current introduction chapter be divided into two parts: introduction and overview.

在研究区域部分，手稿可以进一步强调干旱绿洲地区 SOC 研究的独特性，特别是与其他地区相比。这将有助于突出该研究的创新方面和区域意义。

在方法部分中，建议使用 OLS 回归来检查变量的显著性并增加变量的共线性检验。此外，根据反复测试，该研究分别为农田、果园和未利用土地分配了 3、5 和 0 的值。然而，这缺乏理论依据。我建议引用现有文献来支持这些值的选择。此外，SHAP 值是由 RF 和 GBM 模型的加权平均值生成的，但该公式没有说明权重（α 和 β）是如何确定的。我建议在附录或结果部分添加一个系数表来澄清这一点。

在讨论和结论部分，手稿目前缺乏对局限性的讨论。我建议在讨论的最后添加一个标题为“局限性和未来工作”的专门小节，以系统、透明地总结该研究的主要缺点。例如，数据集仅包含 95 个样本点，这可能会影响模型的鲁棒性。此外，土壤特性通常表现出空间自相关，但样本分割 （SPXY） 和模型构建（RF、GBM、Stacking）过程似乎没有解释或解决这种空间依赖性。此外，可以通过添加有关拟议方法的潜在应用途径和相关政策建议的内容来丰富讨论。

Author Response

Reviewer #2:

1. It is suggested that the current introduction chapter be divided into two parts: introduction and overview.

Response: We sincerely appreciate the reviewers for their valuable suggestions regarding the structure of the introduction. In response to this recommendation, we have meticulously reorganized the content, consolidating the original three-part introduction into two more coherent sections. The detailed revisions are highlighted in cyan from lines 37 to 134 of the manuscript.

 

2. In the study area section, the manuscript can further highlight the uniqueness of SOC studies in arid oasis regions, especially when compared to other regions. This will help to highlight the innovative aspects and regionalsignificance of the study.

Response: We sincerely appreciate the reviewer’s constructive suggestion. As recommended, we have further emphasized the uniqueness of soil organic carbon (SOC) research in arid oasis regions (Lines 157–162), The revised text has been marked in cyan for easy identification.

 

3. In the Methods section, it is recommended to use OLS regression to check the significance of the variables and to increase the collinearity test of the variables.

Response: Thank you very much for your valuable feedback. The initial comprehensive variable set of this study was meticulously selected and constructed based on the existing literature. Subsequently, we conducted an empirical analysis by applying three distinct variable screening methods. Through a systematic comparative analysis, the results showed that the variable set screened by the combination of the Boruta-Lasso algorithm could effectively resolve the collinearity problem among variables and exhibited the optimal modeling performance. In response to your suggestion, we have strengthened the explanation in this regard. For detailed information, please refer to lines 272 to 277, which are highlighted.

 

4. In addition, based on repeated testing, the study assigned values of 3, 5, and 0 to farmland, orchards, and unused land, respectively. However, there is no theoretical basis for this. I suggest citing existing literature to support the choice of these values.

Response: Thank you very much for your valuable feedback. In response, we have supplemented the methodological justification in lines 247–250 (highlighted in cyan) by citing existing literature to support the choice of these values.

 

5. In addition, the SHAP values are generated from the weighted average of the RF and GBM models, but the formula does not explain how the weights (α and β) are determined. I suggest adding a table of coefficients in the appendix or in the results section to clarify this.

 

Response: We sincerely appreciate the reviewer's constructive suggestion regarding the SHAP value calculation. In response, we have clarified the weighting mechanism in Lines 283–313 by presenting the complete formula with coefficients α and β, and added the corresponding coefficient table in Lines 561–562 (both highlighted in cyan) to demonstrate their determination through model performance-based optimization.

 

6. In the discussion and conclusion sections, the manuscript currently lacks a discussion of limitations. I propose to add a dedicated subsection entitled "Limitations and future work" at the end of the discussion to systematically and transparently summarize the main shortcomings of the study. For example, the dataset contains only 95 sample points, which can affect the robustness of the model.

Response: We sincerely appreciate the reviewer's valuable suggestion. In response, we have added a dedicated subsection titled "Limitations and Future Work" at the end of the Discussion section (Lines 698–721, highlighted in cyan). This subsection systematically addresses the study's key limitations.

 

7. In addition, soil properties often exhibit spatial autocorrelation, but the sample segmentation (SPXY) and model building (RF, GBM, Stacking) processes do not appear to account for or resolve this spatial dependence. In addition, the discussion could be enriched by adding content on potential avenues for application of the proposed approach and related policy recommendations.

Response: We extend our gratitude for your feedback. We have added a detailed discussion on the potential influence of spatial autocorrelation in soil properties. The details of this modification can be found on lines 722–737 of the new manuscript, where the relevant changes have been highlighted in cyan for easy identification.

 

We would like to extend our sincere appreciation to the reviewers for their dedicated efforts. We trust that the revisions made will meet with approval. We are truly grateful for your valuable comments and suggestions, which have been of great significance in enhancing the quality of our manuscript.

Reviewer 3 Report

Comments and Suggestions for Authors

1. Summarry and Generral Evaluetion

The paper which the author has submitted aims at performing a spatial inversion of soil organic carbon (SOC) in a specific area called the Wei-Ku oasis, which is located in arid northwestern China, and in order to do that, the authors have attempted to construct what they refer to as a multi-variate feauture space, which was achieved by integrating various types of data, including Sentinel-2A imagery, DEM, soil and climate-related variables, as well as land-use information, all of which were processed using a combination of Boruta and Lasso algorithms to select the most importent variables, and then subsequently used in ensemble learning moddeling with Random Forest, Gradient Boosting Machinne and Stacking model being applied.

Although the studdy, which present itself with clear structure and strong technical content, generally achieve its declared objective, but at the same time, it should be noted that certain sections of the manuscript does require revision or elaboration for ensuring greater consistency and interpretability in results and conclusions, which currently is not fully convincing.

2. Comments According to Manuscript Sections 

【Introduction】 (Lines 35–125)

• It could be observed that the literature reveiw part, especially somewhere from lines 56 to 91, seems more like a listing of prior work rather than a critical review, which creates difficulty in identifiying the logical transitions or identifying what precisely are the knowlege gaps that the current research aimed to fill.

• The end of the introduction, though it does mentioned the study area and method used (lines 116–125), it missed to clearly state what new innovation in technique or model structure is proposed here, which should have be clarified to highlight the contribution.

【Methodology】 (Lines 126–330)

• Although the Section 2.2.3 did explained in detail about the variable sources and types used, but one find it lacking of discussions about the co-linearity issue or standardization steps which are usually necessary when dealing with multi-source data that having different spatial and physical scales.

• There was no specefic justification regarding the selection of moddel parameters, especially mtry in RF or shrinkage in GBM; such parameters should not be choosen arbitrarily.

【Results and Analysis】 (Lines 331–525)

• While the authors have presented extensive results such as model metrics and Gaussian curves (lines 350–457), but they didn’t gave any residual analysis which would help to understand wether the errors are randomly distributed or the model tend to systematically overpredict or underpredict.

• There is an interpretation issue in SHAP analysis part (lines 504–525), because the same variable (DVI) showed opposite contributions in RF and GBM, and authors did not offer reasonable explanation why this happening, which causes confusing.

【Discussion】 (Lines 526–623)

• Most of the discussion, instead of critically explaining or interpreting results in broader context, just repeated what already shown in results section, and failed to explore, for instance, why the Stacking model have better generalisation ability—perhaps because of the meta-learner layer or cross-validation stratagy, which was not even mentioned.

• The section discussing land use data influence (lines 584–623), though it shows better moddel accuracy when TLU was included, but doesn’t really elaborate the ecological mechanism, like how land cover changes are affecting SOC through mechanisms like root litter, microbial activity or irrigation.

【Conclusion】 (Lines 626–644)

• The structure of the conclusion is quite loose and mixed, making it hard to distinguish between key findings, methods and future perspectives; the message needs to be rephrased for greater clarity.

• The limitations of the studdy is not discussed at all; for example, issues like the limited sample size, or possible transferability of the moddel to other regions could be briefly mentioned.

3. Language and Expression Issues

• Redundant or imprecise wording:

  • “achieve the spatial inversion of soil organic carbon” (line 123) → suggest replacing with “accurate spatial mapping of SOC.”

  • “prediction error was reduced by 9%” (line 22) → should clarify “compared with which model or variable set.”

  • “can furnish technical assistance for the precise fertilization of oasis agriculture” (line 30) → sounds informal; consider replacing with “provides technical support for precision fertilization in oasis regions.”

4. Refernce-Related Issues

• It is also necessery to point out that, as can be observed from the current list of citations, a substantial number of the refernces used in the manuscriptt are rather outdated, with many of them being published more than five or even ten years ago, and considering that remote sensing and machine learning techniques, especially in the context of SOC mapping, have experienced significant methodological advancements in just past three years, it would be highly advisible that the authors update their literature base by incorporating more recent publlications, particularly in the sections where state-of-art is being reviewed, otherwise the novelty of the research might be perceived as limited by the reader or reviewers.
• In addition, it must also be noted that the formatting of the references, as currently presented, are not compliying with the specific editorial style guidlines required by MDPI journals, such as missing journal volume/issue in some cases or inconsistent use of italics and author name orderings, which is something that the authors should revise strictly following the template provided by the journal editorials or MDPI style manual.

Author Response

Reviewing: 3

Comments According to Manuscript Sections:

【Introduction】 (Lines 35–125)

1. It could be observed that the literature review part, especially somewhere from lines 56 to 91, seems more like a listing of prior work rather than a critical review, which creates difficulty in identifying the logical transitions or identifying what precisely are the knowledge gaps that the current research aimed to fill.

Response: Thank you very much for your valuable suggestions. Based on your feedback, we have revised the introduction section to ensure a more natural transition, clearly identify the research gaps, and highlight the novelty of our work. The specific modifications are as follows:​ Firstly, in lines 67–70, we emphasize that in complex agricultural ecosystems, there is a lack of systematic research on evaluating the contribution of different variables to modeling accuracy. Secondly, in lines 77–79, we point out that previous studies have predominantly focused on accuracy while neglecting in-depth exploration of the model's interpretability and generalization ability. Then, in lines 98–100 and 103–109, we emphasize the necessity of constructing a spatial prediction model for soil organic carbon under heterogeneous environmental conditions, such as arid oasis ecosystems. Finally, in lines 124–134, we reiterate the significance of this research in bridging the above-mentioned knowledge gap, highlighting its unique value.

 

2. The end of the introduction, though it does mentioned the study area and method used (lines 116–125), it missed to clearly state what new innovation in technique or model structure is proposed here, which should have be clarified to highlight the contribution.

Response: We extend our gratitude for your feedback. In response, in the revised manuscript, we have enhanced the concluding paragraph of the introduction on lines 124–134, which is highlighted in cyan for easy reference, to explicitly state the novel aspects of our approach.

 

【Methodology】 (Lines 126–330)

1. Although the Section 2.2.3 did explained in detail about the variable sources and types used, but one find it lacking of discussions about the co-linearity issue or standardization steps which are usually necessary when dealing with multi-source data that having different spatial and physical scales.

Response: We sincerely appreciate the reviewer's constructive suggestions regarding multi-source data processing.  In response, we have revised the manuscript according to these suggestions. The amended text, highlighted in cyan, can be found on lines 236–238, 242–243, and 272–277 of the manuscript for your reference.​

 

2. There was no specific justification regarding the selection of model parameters, especially mtry in RF or shrinkage in GBM; such parameters should not be chosen arbitrarily.

Response: We appreciate the reviewer’s comment regarding the justification of model parameter selection. In the revised manuscript (Methods section, lines 421–424), we have clarified that all parameter values were chosen based on multiple rounds of internal testing and performance evaluation. The goal was to enhance predictive accuracy and model robustness, while minimizing overfitting. These configurations were not set arbitrarily, but were refined through empirical adjustment and validation

 

【Results and Analysis】 (Lines 331–525)

1. While the authors have presented extensive results such as model metrics and Gaussian curves (lines 350–457), but they didn’t gave any residual analysis which would help to understand whether the errors are randomly distributed or the model tend to systematically overpredict or underpredict.

Response: We sincerely appreciate your valuable suggestions regarding residual analysis. These suggestions play a vital role in enhancing the analytical quality of the thesis. In the manuscript, we have incorporated the relevant calculation formulas for standardized residuals and presented the scatter plot of the standardized residuals. The results indicate that the standardized residuals of these models are predominantly randomly distributed, thus validating the stability of the model predictions. The relevant sections have been supplemented and refined in lines 351-362 and 501-512.

 

2. There is an interpretation issue in SHAP analysis part (lines 504–525), because the same variable (DVI) showed opposite contributions in RF and GBM, and authors did not offer reasonable explanation why this happening, which causes confusing.

Response: Thank you for your comment. In the revised text (Lines 568–573), we explained that the opposite SHAP contributions of DVI in RF and GBM are due to their different algorithmic mechanisms and the spatial heterogeneity of the study area. RF reflects general patterns, while GBM captures localized effects more strongly, especially in areas where high DVI may coexist with low SOC due to soil salinization.

 

【Discussion】 (Lines 526–623)

1. Most of the discussion, instead of critically explaining or interpreting results in broader context, just repeated what already shown in results section, and failed to explore, for instance, why the Stacking model have better generalization ability—perhaps because of the meta-learner layer or cross-validation strategy, which was not even mentioned.

Response: Thank you very much for your valuable feedback. we have strengthened the discussion section in the revised version to provide critical interpretations in lines 722–737 and highlighted it in cyan.

 

2. The section discussing land use data influence (lines 584–623), though it shows better model accuracy when TLU was included, but doesn’t really elaborate the ecological mechanism, like how land cover changes are affecting SOC through mechanisms like root litter, microbial activity or irrigation.

Response: We sincerely appreciate the reviewer's insightful comments regarding the ecological mechanisms underlying land-use effects. In response, we have significantly enhanced the discussion in lines 645-695 (highlighted in cyan) .

 

【Conclusion】 (Lines 626–644)

1. The structure of the conclusion is quite loose and mixed, making it hard to distinguish between key findings, methods and future perspectives; the message needs to be rephrased for greater clarity.

We sincerely thank the reviewer for their constructive comments on the conclusion structure. In response, we have carefully revised the concluding section (Lines 739–757, highlighted in cyan) to improve its organization and clarity while maintaining all key content.

 

2. The limitations of the study is not discussed at all; for example, issues like the limited sample size, or possible transferability of the model to other regions could be briefly mentioned.

Response: We express our sincere gratitude to the reviewer for the valuable suggestion. In response, we have added a dedicated discussion of limitations in lines 698–721 and highlighted it in cyan, including considerations of sample size constraints and other limitations.

 

3. Language and Expression Issues

Redundant or imprecise wording:

(1)“achieve the spatial inversion of soil organic carbon” (line 123) → suggest replacing with “accurate spatial mapping of SOC.”。

Response: We sincerely appreciate the reviewer’s constructive suggestion. In response, we have replaced the term “achieve the spatial inversion of soil organic carbon” with “suggest replacing with ‘accurate spatial mapping of SOC.’” in line 132. All modifications have been carefully implemented and are highlighted in cyan for easy reference.

 

(2) “prediction error was reduced by 9%” (line 22) → should clarify “compared with which model or variable set.”

Response: We sincerely appreciate the reviewer's valuable suggestion. In response, we have added comparative references at line 21-24, with all modifications clearly highlighted in cyan.

 

(3) “can furnish technical assistance for the precise fertilization of oasis agriculture” (line 30) → sounds informal; consider replacing with “provides technical support for precision fertilization in oasis regions.”

Response: We sincerely appreciate the reviewer’s constructive suggestion. In response, we have replaced the term “can furnish technical assistance for the precise fertilization of oasis agriculture” with “provides technical support for precision fertilization in oasis regions.” in line 31–32 All modifications have been carefully implemented and are highlighted in cyan for easy reference.

 

4. Reference-Related Issues

(1) It is also necessary to point out that, as can be observed from the current list of citations, a substantial number of the references used in the manuscript are rather outdated, with many of them being published more than five or even ten years ago, and considering that remote sensing and machine learning techniques, especially in the context of SOC mapping, have experienced significant methodological advancements in just past three years, it would be highly advisable that the authors update their literature base by incorporating more recent publications, particularly in the sections where state-of-art is being reviewed, otherwise the novelty of the research might be perceived as limited by the reader or reviewers.

Response: We sincerely thank the reviewers for their valuable suggestions on updating the references. In response to this suggestion, we systematically modified the format of the references and replaced some of them as suggested. Meanwhile, in order to ensure the accuracy of the elaboration of relevant principles, we have retained a small number of citations from the original methodological literature in the manuscript. Furthermore, in the discussion section, especially lines 698–736 of Section 4.3, we present five of the latest literature to support this argument. All newly added literature items are marked cyan for reference.

 

(2) In addition, it must also be noted that the formatting of the references, as currently presented, are not complying with the specific editorial style guidelines required by MDPI journals, such as missing journal volume/issue in some cases or inconsistent use of italics and author name orderings, which is something that the authors should revise strictly following the template provided by the journal editorials or MDPI style manual

Response: We appreciate your feedback and fully concur with the reviewers' opinions. In accordance with the format requirements of the MDPI journal, we have meticulously reviewed and revised the reference format, and added the citation indexes. These revisions are clearly indicated in lines 770–928 of the manuscript.​

 

We would like to extend our sincere appreciation to the reviewers for their dedicated efforts. We trust that the revisions made will meet with approval. We are truly grateful for your valuable comments and suggestions, which have been of great significance in enhancing the quality of our manuscript.

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

Thank you to the author for respecting and making revisions to the manuscript based on my review comments. I recommend that the paper be accepted and published immediately.