A Future Scenario Prediction for the Arid Inland River Basins in China Under Climate Change: A Case Study of the Manas River Basin

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This manuscript investigates the future climate scenarios of the Manas River Basin using CMIP6 models and multi-model integration methods. The study applies the quantile mapping technique to correct biases and evaluates temperature and precipitation changes under different SSP scenarios. However, several key issues need to be addressed. Therefore, major revisions are necessary before this manuscript can be considered for publication in Sustainability.

1. iThenticate shows a Percent Match of 31%. I recommend adjusting the text to reduce the similarity rate.  
2. The first occurrence of "IPCC" in the main text should be spelled out as "Intergovernmental Panel on Climate Change." Similar errors appear in several other places; the author should carefully check.  
3. The literature review should outline the differences between various machine learning methods and their respective advantages and disadvantages.  
4. Why did you integrate multiple machine learning methods for evaluation? Does this approach lead to better fitting results?  
5. The introduction section needs a brief description of the study area's selection.  
6. Figure 1 lacks latitude and longitude information.  
7. There is a lack of discussion on research limitations. The author can elaborate from perspectives such as data and scenario settings. For example, "we selected only the three most representative scenarios—SSP126, SSP245, and SSP585" (10.3390/buildings14072165).  
8. A comparison with other similar studies is missing.  
9. Can the research findings support real-world management? The author could elaborate on this aspect.

Author Response

Review expert comments reply:

Question 1: iThenticate shows a Percent Match of 31%. I recommend adjusting the text to reduce the similarity rate.

Response:Repetitive content has been modified and the article has been polished. We are grateful for the reviewers' comments.

Question 2: The first occurrence of "IPCC" in the main text should be spelled out as "Intergovernmental Panel on Climate Change." Similar errors appear in several other places; the author should carefully check.

Response:We have made the revisions and are grateful for the reviewers' suggestion.

Question 3: The literature review should outline the differences between various machine learning methods and their respective advantages and disadvantages.

Response:Thank you for your valuable suggestion to enhance the literature review. As recommended, we have expanded the Introduction to systematically outline the differences, advantages, and limitations of key machine learning methods (e.g., Random Forest, Support Vector Machines, Artificial Neural Networks) in climate modeling contexts.

Question 4: Why did you integrate multiple machine learning methods for evaluation? Does this approach lead to better fitting results?

Response:We appreciate the reviewer’s question regarding the integration of multiple machine learning methods. Our study employs a multi-model ensemble approach to enhance the robustness and accuracy of climate predictions by leveraging algorithmic diversity. Different machine learning models (e.g., random forest, support vector machines, neural networks) exhibit complementary strengths in capturing the nonlinear relationships inherent to complex climate systems. For instance, random forests excel at handling interaction effects in high-dimensional data, while neural networks are superior in global pattern fitting. By applying a weighted ensemble strategy, we synthesize the strengths of individual models, mitigating biases that may arise from single-algorithm approaches (e.g., overfitting to outliers or underestimating trend variations). Empirical results demonstrate that the ensemble method significantly improves prediction accuracy compared to the best single model—reducing the root mean square error (RMSE) by 12.3% and increasing the spatial correlation coefficient by 8.5% in historical temperature simulations. In 3.1.2. Quantitative evaluation of GCM simulation ability, RF,ANN,SCM and WSM do show better simulation ability.Furthermore, the dispersion of multi-model outputs provides policymakers with quantifiable uncertainty information, enhancing the practical utility of our findings. Although this study does not exhaust all possible algorithm combinations, our methodology aligns with frameworks such as the IPCC’s multi-model assessments and has been widely validated in climate science . Future work will focus on optimizing dynamic weight allocation strategies and designing specialized ensemble models for extreme events to improve climate risk early-warning capabilities. We thank the reviewer for their suggestion.

Question 5: The introduction section needs a brief description of the study area's selection.

Response:We have made the revisions and are grateful for the reviewers' suggestion.

Question 6: Figure 1 lacks latitude and longitude information.

Response:We have made the revisions and are grateful for the reviewers' suggestion.

Question 7: There is a lack of discussion on research limitations. The author can elaborate from perspectives such as data and scenario settings. For example, "we selected only the three most representative scenarios—SSP126, SSP245, and SSP585" (10.3390/buildings14072165).

Response:The manuscript has been revised as requested. In the conclusion, a brief discussion is provided regarding issues such as the possible neglect of the unique socio-economic or policy-driven climate trajectories, resolution, and extreme climate events specific to arid inland basins. We are grateful for the reviewers' suggestions.

Question 8: A comparison with other similar studies is missing.

Response:Add 3.4 Comparison with other inland river basins in Northwest China to make a comparative analysis with other inland river basins in Northwest China, thanks for the reviewer's comments.

Question 9: Can the research findings support real-world management? The author could elaborate on this aspect.

Response:It has been added in the abstract, introduction and conclusion, thanks to the reviewers for their comments.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

Dear authors,

Please see the attached file.

Comments for author File: Comments.pdf

Comments on the Quality of English Language

English expressions could be further optimised for better communication.

Author Response

Review expert comments reply:

Question 1: Terminology and Standardization Errors

（1）Check whether the scenario names "SSP1.2-6, SSP2.4-5, SSP5.8-5" comply with standardized naming conventions. The correct format should be:SSP1-2.6, SSP2-4.5, SSP5-8.5(SSP stands for Shared Socioeconomic Pathway; the numbers represent radiative forcing values in W/m²). Note that both incorrect and correct formats appear throughout the text!

Response:We have made the revisions and are grateful for the reviewers' suggestion.

（2）Misused Terminology: Check if the term is described in accordance with the

statistical definition of QM terminology.

Response:We have made the revisions and are grateful for the reviewers' suggestion.

（3）Formula Ambiguities:

• Justify the fixed parameter "0.999": Cite literature supporting this value or provide theoretical rationale.

Response:We have made the revisions and are grateful for the reviewers' suggestion.

(4)Validate formula logic: If all models are ranked first (Ranki=1), verify whether CRI equals 1 (range: 0-1).

Response:We are grateful to the reviewers for their help in improving the methodological rigor. The revised CRI formula has been mathematically verified and tested through examples, ensuring its logical consistency and the rationality of the value range. The relevant modifications have been clearly marked in the text.

Question 2:Formatting and Layout Deficiencies

（1）Missing Line Numbers: Add continuous line numbers to facilitate precise feedback.

（2）Figure/Table Issues:

• Center-align figures and ensure resolution ≥300 dpi.
• Correct subfigure labels (e.g., Fig.9 (a) Minimum temperature; (b) Precipitation).Figure 1: Center the layout, improve resolution, and fix caption typos (e.g., "Figure 1. Location map of the research area.." → "Figure 1. Location map of the research area.").

Response:(1),(2) have been modified,We have made the revisions and are grateful for the reviewers' suggestion.

Question 3: Logical and Structural Weaknesses

（1）Vague Model Screening Criteria: Section 3.1.1 claims "eliminating poorly performing models" but lacks quantitative thresholds.

Response:Thank you for your important suggestions on the transparency of model screening. We have supplemented detailed quantitative screening criteria in Section 3.1.1 of the revised manuscript (The correlation coefficient between minimum temperature and maximum temperature < 0.97, and the correlation coefficient of precipitation < 0.5). Since the core indicators of the Taylor diagram can be divided into correlation coefficient, standard deviation and root mean square error. Among them, the correlation coefficient is used to measure the closeness of the linear relationship between two variables and can be used to visually represent the gap between model data and observed data. The two indicators of standard deviation and root mean square error can be seen from Figure 3 and Table S7 that the fitting degree with observed data is poor, so four models with poor performance were excluded, so as to reflect better simulation performance of the MME model. These supplementary contents enhance the rigor of the method section.

Thank you for helping us improve the scientific standardization of the paper.

（2）Insufficient Data Presentation:

• Add multi-metric comparison tables to display model performance across RMSE, SS, KGE, and TSS.
• It is advised to include a detailed score table for each model across all metrics.

Response:We sincerely appreciate the reviewer’s suggestion to enhance the transparency of model performance evaluation. 3.1.1. Taylor diagram and linear trend analysis mainly utilize the r, RMSE, and STD of each model to analyze their performance, and analyze the trend characteristics of temperature and precipitation simulation results. Linear trend fitting analysis is conducted, and Table S7 has been added to statistically present the r, RMSE, and STD of each model; 3.1.2. Quantitative evaluation of GCM simulation ability mainly evaluates the simulation ability of each model quantitatively, selects the optimal model for future climate prediction, and Table S8 has been added to statistically present the TSS, SS, IVS, KGE, and CRI score values.

（3）Lack of Mechanistic Analysis: Most of the descriptions in the original text are of

phenomena, lacking mechanistic explanations.

• It is suggested to evaluate whether these years experienced extreme climate events (such as droughts/floods) and to analyze the reasons for model failure.

Response:Thank you for the reviewers' comments. We have cited relevant literature to prove this point and have made revisions and supplements. In years when the deviations between observational data and model data are significant, there are extreme weather events. Due to the systematic errors in CMIP6 data, the simulation effect of extreme weather is relatively weak. This is mainly caused by the following two aspects:(1) insufficient sensitivity to synoptic-scale dynamics, such as anomalies in the West Asian High and associated moisture transport deficits during droughts , and (2) inadequate representation of orographically triggered convective processes during extreme rainfall events . Furthermore, coarse-resolution parameterizations fail to resolve mesoscale interactions between topography and atmospheric circulation, leading to misrepresented precipitation extremes. To address these shortcomings, future efforts should prioritize convection-permitting regional climate models and advanced ensemble techniques that integrate machine learning for bias-aware dynamical downscaling. Such approaches could enhance the physical fidelity of extreme event projections, particularly in topographically complex arid regions.

• Does it align with the actual situation, is it measured data? Or is there supporting

data to argue this point!

Response:We appreciate the reviewer's emphasis on empirical validation and have addressed this by incorporating foundational mechanistic studies (Roe, 2005) on orographic precipitation dynamics alongside regionally analogous works (He et al., 2023) to theoretically contextualize topographic controls, while simultaneously cross-validating our results with the HRLT high-resolution precipitation dataset (Qin et al., 2022) derived from advanced satellite-gauge merging algorithms.

• Is there specific data supporting this conclusion? It is suggested to add a simple

description of the mechanisms!

Response:Thank you for the reviewers' comments. I have revised the conclusion and enhanced its coherence. Additionally, I have added a simple description of the mechanisms.

 

Question 4:The description in the analysis paragraph (such as the conclusion section) is very confusing. It is suggested to organize it in a point-by-point, hierarchical progression.

Response:We have made the revisions and are grateful for the reviewers' suggestion.

 

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

This manuscript evaluates historical temperature and precipitation in the Manas River Basin using nine CMIP6 models, five multi-model integration methods. The results are expected to provide reference value for the sustainable development of China’s inland river basin under future climate change. There are some comments to improve the quality of the manuscript.
1. In 3.1.1, the authors excluded four models with poor performance, but did not provide specific quantitative thresholds (e.g., correlation coefficients, RMSE thresholds). It is recommended that the quantitative criteria for model selection be described.
2. Where is (a) and (b) in Fig. 9? Why are 9a and 9b labeled the same as “Minimum temperature”? In Fig. 9 and 11, the color-coded ranges and geographic identifiers (e.g., mountain ranges, deserts) of the spatial distribution map are not clearly labeled. It is recommended that the figure be rechecked and the legend added.
3. The manuscript focuses on long-term average changes in rainfall. Did the authors consider the impact of future extreme heat, drought or heavy rainfall events on precipitation and temperature?
4. The terminology is inconsistent, SSP5-8.5 appears in Table 1, and SSP5.8-5 appears in the main text. It is recommended that the nomenclature of the SSP/RCP scenarios be harmonized, and the manuscript language needs to be refined and corrected.
5. What is the sensitivity of the QM bias correction to the prediction results? It is suggested to compare the results of different bias correction methods.
6. Some of the figures in the manuscript should be modified to be clearer, as in Fig. 2, Fig. 3.
7. Why are future projections categorized as “recent (2015-2060)” and “forward (2061-2100)”? How did the authors get the segmented timeline of 2060?
8. The SSP scenarios contain demographic, economic, and technological assumptions. Do the authors consider their effects on climate projections?
9. The manuscript is excessively repetitive. It is strongly recommended to revise and embellish the language.

Comments on the Quality of English Language

The language of this manuscript could be improved.

Author Response

Review expert comments reply:

Question 1: In 3.1.1, the authors excluded four models with poor performance, but did not provide specific quantitative thresholds (e.g., correlation coefficients, RMSE thresholds). It is recommended that the quantitative criteria for model selection be described.

Response:Thank you for your important suggestions on the transparency of model screening. We have supplemented detailed quantitative screening criteria in Section 3.1.1 of the revised manuscript (The correlation coefficient between minimum temperature and maximum temperature < 0.97, and the correlation coefficient of precipitation < 0.5). Since the core indicators of the Taylor diagram can be divided into correlation coefficient, standard deviation and root mean square error. Among them, the correlation coefficient is used to measure the closeness of the linear relationship between two variables and can be used to visually represent the gap between model data and observed data. The two indicators of standard deviation and root mean square error can be seen from Figure 3 and Table S7 that the fitting degree with observed data is poor, so four models with poor performance were excluded, so as to reflect better simulation performance of the MME model. These supplementary contents enhance the rigor of the method section.

Thank you for helping us improve the scientific standardization of the paper.
Question 2: Where is (a) and (b) in Fig. 9? Why are 9a and 9b labeled the same as “Minimum temperature”? In Fig. 9 and 11, the color-coded ranges and geographic identifiers (e.g., mountain ranges, deserts) of the spatial distribution map are not clearly labeled. It is recommended that the figure be rechecked and the legend added.

Response:Has been modified as required. The figure mainly shows the extent of changes in the lowest and highest temperatures and precipitation. If too many labels are added, it will cause the extent of the changes to be unclear. Therefore, the dividing lines between the mountainous area and the oasis, and between the oasis and the desert have been added. Thank you for the reviewer's suggestions.
Question 3: The manuscript focuses on long-term average changes in rainfall. Did the authors consider the impact of future extreme heat, drought or heavy rainfall events on precipitation and temperature?

Response:We sincerely appreciate the reviewer’s insightful question regarding the impacts of extreme climate events. In this study, our primary focus was to evaluate long-term trends in average temperature and precipitation under different SSP scenarios, as these metrics provide foundational insights into baseline climate shifts critical for water resource planning and agricultural adaptation in arid inland basins. While extreme events (e.g., heatwaves, droughts, and heavy rainfall) were not explicitly analyzed, we acknowledge their profound implications for regional vulnerability and resilience.

The omission of extreme event analysis stems from two key considerations:

1. Scope and Data Limitations: The CMIP6 models used in this study prioritize large-scale, long-term climate trends. While they capture mean climate variables robustly, their resolution and parameterization of extreme events—particularly at regional scales—remain uncertain. Many extreme indices (e.g., consecutive dry days, heatwave duration) require higher-resolution datasets or tailored statistical downscaling, which were beyond the scope of this work.
2. Methodological Focus: Our objective centered on optimizing multi-model ensemble techniques to reduce uncertainties in baseline projections. Extending this framework to extremes would necessitate additional validation against observational extremes and specialized bias-correction methods, which we plan to address in future research.

We fully agree with the reviewer that extreme event analysis is vital for comprehensive climate risk assessment. As a next step, we will integrate extreme climate indices (e.g., R99p for heavy rainfall, TXx for extreme heat) using high-resolution regional climate models and leverage machine learning to improve extreme event attribution. These enhancements will bridge the gap between mean trends and localized impacts, offering actionable insights for policymakers.

Thank you for highlighting this critical gap, which will significantly strengthen the practical relevance of our research.
Question 4: The terminology is inconsistent, SSP5-8.5 appears in Table 1, and SSP5.8-5 appears in the main text. It is recommended that the nomenclature of the SSP/RCP scenarios be harmonized, and the manuscript language needs to be refined and corrected.

Response:We have made the revisions and are grateful for the reviewers' suggestion.
Question 5: What is the sensitivity of the QM bias correction to the prediction results? It is suggested to compare the results of different bias correction methods.

Response:Thank you for your valuable suggestion regarding the sensitivity analysis of Quantile Mapping (QM) bias correction and the comparison of different correction methods. We appreciate the opportunity to clarify these points.

In this study, QM was selected as the primary bias correction method due to its widespread application in climate science and proven effectiveness in aligning model outputs with observed data distributions, particularly for variables like temperature and precipitation. The QM process adjusts the quantiles of model-simulated data to match those of observations, thereby reducing systematic biases while preserving the model’s projected climate trends. Our supplementary materials (e.g., Text S5,Figure S1,Table S6 in the Appendix) demonstrate that QM significantly improved the agreement between corrected model outputs and historical observations.

Regarding sensitivity, the QM correction’s performance was robust across different SSP scenarios. For example, under SSP5-8.5, the corrected temperature trends retained the projected warming magnitude while aligning more closely with observed interannual variability. However, we acknowledge that QM’s sensitivity to extreme values or non-stationary climate conditions (e.g., accelerating warming post-2050) warrants further investigation.

While our study did not explicitly compare QM with alternative bias correction methods (e.g., Delta Change, Empirical Quantile Mapping), this omission was primarily due to the scope focusing on optimizing multi-model ensemble performance rather than method intercomparison. Nonetheless, we fully agree that such a comparison would enhance methodological transparency. In ongoing work, we are evaluating multiple correction techniques (including machine learning-based approaches) to quantify their impacts on projection uncertainties. Preliminary results suggest that QM outperforms simpler methods (e.g., linear scaling) in preserving extreme value statistics, but hybrid approaches may offer additional advantages.

We will incorporate a brief discussion of these points in the revised manuscript to address the reviewer’s concern and cite recent studies that benchmark QM against other techniques. Thank you again for this constructive feedback, which strengthens the methodological rigor of our work.
Question 6: Some of the figures in the manuscript should be modified to be clearer, as in Fig. 2, Fig. 3.

Response:We have made the revisions and are grateful for the reviewers' suggestion.
Question 7: Why are future projections categorized as “recent (2015-2060)” and “forward (2061-2100)”? How did the authors get the segmented timeline of 2060?

Response:Thank you for your insightful question regarding the division of future projections into “near-term (2015-2060)” and “long-term (2061-2100)” periods. This segmentation was primarily guided by the following considerations:

1.The temporal division aligns with common practices in climate modeling studies, where projections are often partitioned into intervals to analyze evolving trends under different socioeconomic and emission scenarios. Many CMIP6 models provide outputs in standardized periods (e.g., 30- to 50-year windows), and 2015-2060 represents a mid-century horizon, while 2061–2100 captures end-of-century dynamics. This approach facilitates comparison with existing literature and IPCC assessment frameworks.

2.The near-term period (2015-2060) corresponds to actionable planning windows for climate adaptation and mitigation strategies, such as China's Xinjiang decarbonization target (e.g., carbon neutrality by 2060). In contrast, the long-term period (2061-2100) reflects the compounding effects of delayed climate action, enabling assessments of irreversible impacts (e.g., glacier loss, sea-level rise).

3.By 2060, intermediate emission scenarios (e.g., SSP2-4.5) may approach key climatic tipping points (The response of snow cover to climate change has a lag effect. The cumulative effect of snow cover melting over 50 years (such as glacier retreat and changes in meltwater runoff) is consistent with the research objective “Effects of climate change on precipitation and temperature.”). Separating the analysis before and after this timeframe allows for clearer identification of inflection points in temperature and precipitation trends.

4.The unique characteristics of the Manas River Basin: ecological protection initiatives (e.g., the restoration of Lake Manas) and hydraulic engineering projects (e.g., the Kenswat Reservoir) within the basin are planned with a design life of approximately 50 years. The forecast results for the period 2015-2060 can directly inform the assessment of project benefits and support sustainable water resources management.

4.Dividing the 85-year span (2015-2100) into two roughly equal intervals (45 and 39 years) ensures sufficient data length for robust trend detection while minimizing inter-decadal variability noise. This balance enhances the reliability of both near- and long-term projections.

While the choice of 2060 as a boundary year is somewhat arbitrary, it balances scientific utility, policy relevance, and computational feasibility. Future work could explore sensitivity analyses with alternative time divisions (e.g., 2050 or 2070) to assess the robustness of our findings. We appreciate your feedback and will clarify this rationale in the revised manuscript.
Question 8: The SSP scenarios contain demographic, economic, and technological assumptions. Do the authors consider their effects on climate projections?

Response:Our study utilized CMIP6 model data, which inherently incorporates the socioeconomic assumptions of SSP scenarios (e.g., SSP1-2.6 and SSP5-8.5). These assumptions—including population growth, economic development, and technological choices—indirectly influence climate predictions through their impacts on greenhouse gas emission pathways. For instance, the SSP5-8.5 scenario assumes high fossil fuel dependency, rapid population growth, and limited technological innovation, which collectively determine input parameters such as CO2 concentrations and aerosol emissions in the models. While the study did not explicitly deconstruct the independent effects of specific socioeconomic variables, their aggregated impacts are reflected in climate simulations via emission-driven mechanisms.

In our methodology, we focused on optimizing the statistical relationships among climate variables (temperature, precipitation) through multi-model ensemble techniques (e.g., the weighted ensemble method) and machine learning algorithms (e.g., random forest). These approaches rely on the intrinsic integration of socioeconomic assumptions within the CMIP6 outputs but do not quantify the isolated influence of individual factors (e.g., population policies or technological breakthroughs). For example, variations in model responses to energy transition rates under the same SSP scenario may affect projections, yet our primary goal was to evaluate the efficacy of ensemble methods in reducing uncertainties rather than dissecting socioeconomic drivers in detail.

We acknowledge certain limitations. First, sensitivity experiments were not conducted to distinguish climate responses under SSP scenarios with similar radiative forcing levels (e.g., SSP2-4.5 vs. SSP3-7.0), limiting our understanding of the independent roles of socioeconomic pathways. Second, the global SSP assumptions were not dynamically coupled with regional human activities (e.g., irrigation expansion or urbanization in the Manas River Basin), potentially underestimating localized climate feedbacks from socioeconomic changes (e.g., humidity increases due to irrigation). These factors may affect the regional applicability and policy robustness of our projections.

Future research will prioritize: (1) designing sensitivity experiments to disentangle the contributions of population, economic, and technological factors to climate projections; (2) developing coupled climate-socioeconomic models that integrate dynamic feedbacks from regional water management and land-use decisions. We sincerely appreciate the reviewer’s insights, which will enhance the scientific value of our work for policy applications and provide more precise decision-making support for sustainable development in arid inland river basins.
Question 9: The manuscript is excessively repetitive. It is strongly recommended to revise and embellish the language.

Response:To ensure the language quality of this article, it was re-polished using MDPI’s Author Services. We are grateful for the reviewers' suggestions.

 

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

This work used CMIP6 models with different integration methods to evaluate and predict past/future temperature and precipitation changes in Manas River Basin. The work is comprehensive, noval and adds value as reference for future development plan for basin areas. I only have a few minor comments.

1. I see a lot of contents in the supplementary information. However, these contents are never referenced in the manuscript. Please make sure all sections in the SI are properly referred in the main manuscript.
2. For the models chosen in Table 2, please provide an explanation on why these models were selected for this study. 
3. The letters x and y have been used in multiple models representing different things. Please distinguish them by adding subscripts. For example y_RF to represent the simulation output from random forest model
4. On page 7 before reference [54], it seems like a name of the author is missing. Please correct it. 
5. Figure 9 is missing the subtitle (a) and (b) on the figures. Please add it to the figure. Currently is very hard to understand because is hard to distinguish which ones belong to figure 9(a) and which ones belong to figure 9(b).
6. On page 14, above Figure 9, please paraphrase this sentence: through longitudinal comparison and analysis of the recent and long-term maximum.... This sentence is very long and difficult to read and comprehend. Please paraphrase it and chop it into a few shorter sentences. 
7. In section 3.2.2, the author said temperature increased continuously from north to south. However, figure 9 indicates that in all scenarios temperature is decreasing from north to south. Please explain the inconsistency or correct the typo. 

Author Response

Review expert comments reply:

Question 1:I see a lot of contents in the supplementary information. However, these contents are never referenced in the manuscript. Please make sure all sections in the SI are properly referred in the main manuscript.

Response:We have made the revisions and are grateful for the reviewers' suggestion.

Question 2:For the models chosen in Table 2, please provide an explanation on why these models were selected for this study.

Response:We sincerely appreciate your thoughtful feedback and provide the following clarifications regarding the model selection criteria:

1.The selected CMIP6 models originate from diverse countries (e.g., Australia, China, Canada, Europe, Russia, Japan, Germany, France), representing major global climate modeling centers. This diversity helps mitigate systemic biases arising from single-model structures or parameterization schemes, thereby enhancing the robustness and generalizability of the multi-model ensemble (MME) results.

2.The chosen models have been successfully applied in other typical arid inland river basins, such as the Hotan River Basin (He, C.;Luo, C.;Chen, F.;Long, A..Tang, H. CMlP6 multi-model prediction of future climate change in the Hotan River Basin. Earth Science Frontiers 2023, 30 (03), 515-528.) and the Shiyang River Basin (Dai, J.;  Hu, H.;  Mao, X..Zhang, J. Future climate change trends in the Shiyang River Basin based on the CMlP6 multi-model estimation data. Arid Zone Research 2023, 40 (10), 1547-1562.). Given the climatic and hydrological similarities between these basins and the Manas River Basin (our study area), the selected models are well-suited for arid inland regions in northwestern China (see Text S6 in the Supplementary Materials).

3.While downloading CMIP6 models from the official website, some models exhibit issues such as incomplete temporal coverage or missing variables. However, the models selected in this study have complete temporal sequences.All models meet the criteria for completeness and continuity across both the historical period (1979–2014) and future scenarios (2015–2100), ensuring temporal consistency in our analysis.

In summary, our model selection balances diversity, data quality, and historical performance to ensure the applicability of findings to arid inland river basins like the Manas River Basin, while providing reliable inputs for the multi-model ensemble framework.

Thank you once again for your constructive comments, which have strengthened the clarity and rigor of our work.

Question 3:The letters x and y have been used in multiple models representing different things. Please distinguish them by adding subscripts. For example y_RF to represent the simulation output from random forest model

Response:We have made the revisions and are grateful for the reviewers' suggestion.

Question 4:On page 7 before reference [54], it seems like a name of the author is missing. Please correct it.

Response:We have made the revisions and are grateful for the reviewers' suggestion.

Question 5:Figure 9 is missing the subtitle (a) and (b) on the figures. Please add it to the figure. Currently is very hard to understand because is hard to distinguish which ones belong to figure 9(a) and which ones belong to figure 9(b).

Response:We have made the revisions and are grateful for the reviewers' suggestion.

Question 6:On page 14, above Figure 9, please paraphrase this sentence: through longitudinal comparison and analysis of the recent and long-term maximum.... This sentence is very long and difficult to read and comprehend. Please paraphrase it and chop it into a few shorter sentences.

Response:We have made the revisions and are grateful for the reviewers' suggestion.

Question 7:In section 3.2.2, the author said temperature increased continuously from north to south. However, figure 9 indicates that in all scenarios temperature is decreasing from north to south. Please explain the inconsistency or correct the typo.

Response:We have made the revisions and are grateful for the reviewers' suggestion.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Authors

The revision seems properly addressed my recommendations. I have no further suggestions, and the paper can be considered for publishing.

With best regards

The reviewer

Reviewer 3 Report

Comments and Suggestions for Authors

I don't have any more comments.