Predicting Wine Quality Under Changing Climate: An Integrated Approach Combining Machine Learning, Statistical Analysis, and Systems Thinking

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Authors,

What are the legal and ethical implications of publishing data from the over 72,000 wine evaluations used in this study, given that they belong to a third-party institution (KGZS Maribor), and to what extent can the authors prove the right to use and the authenticity of these data as their own for scientific purposes? 

Figure 11 and so on – The data from those tables needs to be translated into English.

Thank you!

Author Response

In the file. 

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Dear authors, I have read the MS entitled "Predicting Wine Quality under Changing Climate: An Integrated Approach Combining Machine Learning, Statistical Analysis and System Thinking".

The article presents an integrated approach combining machine learning, statistical analysis, and systems thinking to predict wine quality in Slovenia under changing climate conditions. It finds that regional classification and specific monthly temperatures are the strongest predictors of wine quality, while precipitation has little statistical impact.

I have some observations in order to improve the quality of the article:

-Please explain in short the statistical methods and also the programs that were used for this. Maybe add a different subchapter for this.

-Causal Loop Diagrams (CLDs) are used, but the article does not quantify the strength or delay durations of feedback loops. Can this be improved?

-Precipitation is dismissed as statistically insignificant, yet, in reality, it is not so. Please discuss this further.

-The analysis uses average regional data, potentially masking microclimatic and vineyard-level variability, which can be seen as a limitation. Any possibilities of adding data from vineyard sensors or other microclimate data?

- Is it possible to align climatic variables with specific phenological stages of the grapevine using Growing Degree Days (GDD) or other means?

-The dependent variable is average regional wine scores, which may smooth over important variability. Explore finer resolution data, if available, or at least address the limitation this aggregation imposes on precision.

-Only raw monthly temperatures and precipitation are used. Include, if possible, GDD, diurnal temperature range, or heat stress days (>35°C) to improve model sensitivity.

-fig 4 and 5 are missing?

-consider incorporating additional bibliographical references, particularly in sections where emerging methodologies or recent developments in climate-viticulture modeling like DOI 10.20870/oeno-one.2024.58.4.7767, DOI 10.1139/CJPS-2023-0161, DOI 10.15835/nbha52113381.

Comments on the Quality of English Language

-The terms "system thinking" and "systems thinking" are both used. Which one is correct?  Please use all over text accordingly.

-“perceptual lag” and “temporal lag” are used interchangeably but could be more clearly differentiated.

-Simplify and clarify terminology. Consider revising passive constructions for better readability.

-Verify all over text for grammatical correctness, passive voice (do not use "we"), and smoother transitions.

Author Response

In the file

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This study integrates statistical analysis, machine learning (ML) algorithms, and systems thinking to assess the extent to which wine quality can be predicted based on monthly weather variables and regional classification. The quantitative dataset includes average wine scores, monthly temperatures, precipitation levels, and categorical regional data for Slovenia from 2011 to 2021. The models employed include Random Forest, Support Vector Machine (SVM), Decision Tree, and linear regression. Additionally, Causal Loop Diagrams (CLDs) were constructed to explore system dynamics.

The Random Forest model demonstrated the highest predictive accuracy (R² = 0.779), with regional classification emerging as the most significant predictor, followed by September and April temperatures. Precipitation showed no statistically significant influence on wine quality. CLD analysis highlighted the delayed effects of adaptation measures and the role of perceptual lags in vineyard management decisions.

The integration of ML with statistical analysis and CLD modeling enhances understanding of complex agroecological systems and provides robust tools for anticipating the impacts of climate variability on wine quality. Future research should incorporate phenological phases and dynamic modeling frameworks to further improve predictive capabilities.

I believe that this paper fits the aims and scope of this journal and provides useful information to readers. Therefore, this paper is suitable for publication, but the manuscript requires revisions.

Please answer the following questions.

------------------------------------------------------------

Figures 1, 2, 6, 7, 8, 9, 10, 11, 12, and 13 are shown as figures, but their content is mostly in tabular form. It would be better to convert them into tables to make things clearer and more consistent. We suggest combining and reorganizing them as Table 1 and Table 2, based on how similar they are in terms of topic or method. Also, the manuscript has a lot of figures, which might be too much for readers and make the important visual data less impactful. It is recommended that you carefully consider if each figure is necessary. If it is not necessary, you can move it to the Supplementary Data section.

Figure 14 doesn't meet the standards for publication because its resolution and visual clarity are not good enough. We suggest improving the quality of the image to make it easier to read and look professional.

Figure 4 and 5: There is no Figure.

For the reference list, please modify all references to comply with the MDPI Reference List and Citations Style Guide.

L70: “Literature rewiev” Please check this word.

L182, 235, 237, 252, 256, 282, 285, 310, 330, 353, 376, 391, 434, 471, 502: "Source: Authors' own work." If you are actually using data, figures, or findings from other sources, please be sure to clearly state the source information in the Reference List.

L607: Although the model developed in this study is based on data from Slovenia, its applicability to other regions should also be considered.

Comments on the Quality of English Language

The authors should improve their use of English, particularly in figure and table captions.

Author Response

In the file. 

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

The article entitled "Predicting Wine Quality under Changing Climate: An Integrated Approach Combining Machine Learning, Statistical Analysis and System Thinking" presents a highly relevant and innovative approach to one of the pressing challenges in modern viticulture—understanding and anticipating the effects of climate change on wine quality. By integrating machine learning (ML), classical statistical techniques, and systems thinking tools such as Causal Loop Diagrams (CLDs), the authors propose a multifaceted methodology applied to the case of Slovenian wine regions.

The abstract effectively summarizes the study’s purpose, methodological framework, and key findings. However, it would benefit from a more explicit reference to the practical implications of the work—particularly for winegrowers, policy-makers, or regional planners. A short addition describing potential applications would enhance the relevance for a broader scientific audience.

In the introduction, the authors offer a thorough and well-structured overview of the multiple factors influencing wine quality, including climatic variables and the broader concept of terroir. The text accurately highlights the complexity of wine production systems and the growing role of data-driven approaches. Yet, while the term “terroir” is invoked throughout as a key explanatory variable, its operationalization as a simple categorical region code is somewhat reductionist. This could be addressed by further acknowledging the trade-offs between explanatory richness and model interpretability.

The literature review is detailed and well-grounded, drawing on recent and classical references that span viticulture, climate science, and machine learning. Nevertheless, there is some redundancy between the introduction and this section. Additionally, while the review effectively identifies methodological gaps in the existing literature, it could engage more critically with limitations in prior ML-based studies—particularly concerning data quality, regional heterogeneity, and overfitting in small datasets.

The methodology section is a strong point of the manuscript. It is clearly written and well-documented, describing in detail the sources of data, preprocessing steps, and structure of the models. The use of over 72,000 wine scores aggregated over a decade offers a substantial empirical base. However, the study would benefit from a brief discussion on the choice of time period (2011–2021) and whether any anomalous climatic events (e.g., heatwaves or frost episodes) during that time were considered or filtered. More importantly, the section on CLDs lacks critical methodological transparency—specifically regarding how the diagrams were developed, which sources or expert insights were used, and whether any validation of the feedback structures was performed.

In the results section, the statistical analysis is comprehensive and clearly presented. Trend analysis confirms the impact of climate change on temperature dynamics, particularly in spring and summer months. The correlation and regression analyses reveal September and April temperatures as significant predictors of wine quality, which aligns with known phenological sensitivities. Conversely, precipitation emerges as a weak and inconsistent predictor—a result that the authors explain plausibly by highlighting its indirect influence via disease pressure or soil saturation. The factor analysis and subsequent regression confirm that high summer temperatures coupled with low precipitation may have a moderate but statistically significant effect on wine quality.

The machine learning models add valuable predictive insight. Random Forest outperforms other models in terms of R² and error metrics, with region emerging as the most influential predictor. Yet, the manuscript would benefit from a brief explanation of why this model performs better and whether hyperparameters were optimized. It is important to note that the small number of regional categories (n = 9) and modest sample size for ML applications introduce a risk of overfitting—this should be acknowledged.

The inclusion of Causal Loop Diagrams is one of the most distinctive features of the study. These diagrams provide a systems-based view of how climate variables, grapevine physiology, and management responses interact. The first diagram, focusing on biomass dynamics, effectively illustrates how environmental carrying capacity and resource limitations create reinforcing and balancing feedbacks. The second CLD, which captures the vineyard management system, introduces the notion of perceptual delay—a particularly useful concept when discussing adaptation lags. However, the models remain conceptual and would be strengthened by anchoring them more directly in empirical or semi-quantitative data.

In the discussion section, the authors offer thoughtful reflections on their findings. The dominance of the region variable is well contextualized as a proxy for the multidimensional nature of terroir. They also reflect appropriately on the importance of specific temperature windows (April and September) and consider how vineyard practices may mitigate or amplify the influence of climatic stressors. The discussion on the non-significance of precipitation is carefully reasoned and grounded in agronomic understanding. Still, certain interpretations—such as the role of December temperatures—could be more cautiously formulated unless further evidence is provided.

The authors rightly highlight limitations, including the use of aggregated annual data, the lack of phenology-specific climatic variables, and the delayed nature of management interventions. These reflections show awareness of the methodological boundaries of their work. However, the manuscript could go further in suggesting how the integration of time-series analysis or the inclusion of phenological phase data could improve future models.

Finally, the conclusions successfully summarize the main contributions: (1) the centrality of regional characteristics in predicting wine quality, (2) the value of integrating multiple methodological perspectives, and (3) the utility of CLDs in revealing structural and delayed system dynamics. These findings are credible and meaningful. Nonetheless, the paper could be further strengthened by articulating more clearly how this approach can be applied to other regions or scaled up for policy-making.

Conclusion of Review

In sum, the manuscript is a well-executed and innovative contribution to the fields of viticulture and climate adaptation. Its strength lies in methodological integration and thoughtful analysis. However, revisions are necessary to clarify the construction and validation of the CLDs, to provide greater methodological transparency, and to refine certain speculative claims. 

Recommendation: Minor to Moderate Revision

Author Response

In the file

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript is improved! Congratulations!

Author Response

Thank you very much for your kind words and valuable feedback. We truly appreciate your time and support throughout the review process.

Reviewer 2 Report

Comments and Suggestions for Authors

Thank you for successfully addressing many of the original concerns, including: statistical methods and tools, improving methodological transparency, causal loop diagrams (CLDs), data granularity, phenological alignment, terminology and language consistency.

I believe that the research would become even more valuable of the authors could expand on vine physiology and heat thresholds, a more detailed explanation of threshold effects (e.g., photosynthesis inhibition, enzyme activity loss) would clarify how temperature extremes impair grape quality. Also, a basic simulation modeling to semi-quantitatively test CLD assumptions would much improve the results.

Also, instead of merely noting phenological integration as future work, the authors could model proxy phenological phases (e.g., use April–May for budburst, August–September for ripening) as intermediate variables to approximate GDD impact with current data.

Comments on the Quality of English Language

The abstract remains wordy and dense. 

Tables and figures are now complete, but long blocks of statistical text could benefit from brief summaries to highlight key insights.

Some figures (e.g., CLDs) still lack descriptive captions explaining why they're important and what they show. 

Author Response

The response is in the attachment. Thank you for your valuable feedback and support through the revision process

Author Response File: Author Response.pdf