This paper investigates how various climate indices impacts drought assessment measured by SPEI, based on an explainable-AI framework. Below are my major concerns followed by minor comments.

Major comments:

1. While the authors claim that they use an explainable-AI framework, the methods section has limited details about SHAP and how the AI is explainable. Random Forest, while tree-based, does not embed physical mechanisms as a priori. Relevant explanations in the manuscript are also very brief. For example, there are no details about how the feature value and SHAP value work, and what information the beeswarm plot conveys in Fig. 9. The caption of Fig. 9 is also very short. Line 345 “increases in correlations with climatic indices tend to negatively affect how data points are grouped into clusters” also requires a clear physical interpretation.
2. The paper claims a methodological advancement, but the literature review gives limited coverage of studies that use conventional approach. The Discussion should (a) compare the present results with key earlier studies that relied on traditional methods, and (b) explain why the proposed framework leads to superior or complementary results.
3. Abstract line 10, it is inappropriate to state that XX% has stat sig trend because there could be spatial autocorrelation that inflate counts of significance. Same thing for Line 201-202, Line 372, Line 538. A relevant paper is Wilks, D. S. "On “field significance” and the false discovery rate." Journal of applied meteorology and climatology9 (2006): 1181-1189.
4. It is unclear what “climate indices” means. Broadly speaking, SPEI itself can also be a climate index. The authors should highlight large-scale climate variability or provide a formal definition of climate indices.
5. Table 1 lists many indices, but the manuscript does not explain why each is relevant to Sahel/African hydroclimate. Please justify the inclusion of each index or focus on a subset with documented influence on the region, similar to the description of “Sahel Precipitation”.
6. Using two particular cells in Fig. 6 and Fig. 7 is not representative. The two cell is just two out of 1335 SPEI gridded data points in the study region, and there is not a clear rationale for focusing on these cells. It is hard to follow the motivation of the analysis. While the cell in Fig. 6 has the strongest positive correlation between AMO and SPEI, the overall correlation mean is only “modest” at 0.06 (Line 239). How could it support the statement in Line 270, “AMO are closely tied to sub-regional drought dynamics”?
7. There are mismatches and typos in the manuscript. I suggest the authors carefully read their manuscript throughout. To name only a few: Line 231 refers to Fig. 5 as a “combined box and violin plot,” but Fig. 5 is a map. “ahel” in Table 1 should be “Sahel.” Line 230 describes Fig. 4 as “maps of correlations … and the most correlated climatic indices,” but Fig. 4 shows bar plots for all indices.
8. The manuscript does not specify the data sources for each climate index in Table 1.

Minor comments:

1. Abstract line 15, Why should a positive correlation necessarily imply a stronger influence on regional hydrology? Drought is part of hydrology as well; as long as a statistically significant relationship exists—positive or negative—it can affect the system.
2. Lines 19-20, “further highlights … the NTA” is confusing because the NTA is not introduced earlier.
3. Line 21, the abstract does not explain why or how the AI component is explainable.
4. Line 38-39, Gleeson et al. (2012) do not discuss temperature effects. Please check the citation or replace with a more appropriate reference.
5. Fig. 1 Consider overlaying Köppen climate‐type boundaries (or another climate‐zone map). This would help readers see whether algorithm‐identified clusters align with known climatic regions.
6. Lines 134-135, I don’t understand how “The 12-month period is long enough to capture the cumulative effect of these global drivers”. AMO operates on multi‐decadal scales, much longer than 12 months.
7. Lines 140-143, The logic is hard to follow. Clarify why having 31 indices conflicts with a 1951–2018 record, and why a “large number of indices” would undermine a robust analysis. Re-phrase to make the trade-offs explicit.
8. Fig. 3a, I suggest adding hatches or stipples to distinguish areas with and without statistically significant trends. Same thing for Fig. 5.
9. Line 210-219, When discussing the impact of climate variability on drought, indicate the direction of influence. For example, does increased aerosol loading tend to increase or decrease regional precipitation?
10. Line 266, Define the threshold for “weaker correlations” and state the correlation values, not just the IQR.
11. Line 341, what statistic of SHAP values do we use to measure the influence on clustering? I thought I should look at the mean values but here the authors cite the range.
12. Line 344, “High” and “low” should be replaced with actual correlation values (or value ranges). Note that Fig. 9 labels “feature value,” not “correlation.”
13. Lines 475-481, Spatial heterogeneity has already been discussed in lines 415-419. Avoid repetition.
14. Lines 512-519, Link the limitation of ignoring human activities to specific findings—e.g., could regions with low climate–SPEI correlation coincide with areas of extensive land‐use change or other human activities?

The manuscript titled "Decoding the Architecture of Drought: SHAP-Enhanced Insights into the Climate Forces Reshaping the Sahel" presents a robust, interdisciplinary analysis of drought patterns in the Sahel region. The authors employ a multi-method approach that combines the Standardized Precipitation Evapotranspiration Index (SPEI), Seasonal Kendall (SK) trend analysis, cross-correlation with 31 climatic indices, and a SHAP-enhanced clustering methodology using Random Forest (RF) to explore the spatial-temporal variability of drought and its climatic drivers.

Key findings include:

• A significant downward trend in SPEI-12 across 57.5% of the Sahel, particularly in the west and southeast, indicating intensified drought conditions.
• Strong negative correlations between drought severity and Global Mean Temperature (GMT) and Indo-Pacific Warm Pool (IPWP); Atlantic Multidecadal Oscillation (AMO) showed spatially heterogeneous impacts.
• The clustering analysis delineates three distinct regions with unique drought dynamics and climate-drought interactions.
• The SHAP framework reveals the differential contribution of climatic indices to drought clustering, offering high interpretability and novel insight into region-specific vulnerabilities.

1. Title and Abstract

Suggestions:

• Include quantitative results (e.g., number of clusters, correlation values) in the abstract to enhance clarity and impact.
• Slightly reduce jargon in the abstract for broader accessibility (e.g., explain “SHAP” in simpler terms before the acronym).

1. Introduction

Suggestions:

• Include a short paragraph summarizing existing clustering approaches and why SHAP-RF is a significant improvement.
• Reduce the length of some paragraphs to improve readability and flow.

1. Materials and Methods

Suggestions:

• Consider summarizing the 31 climate indices in a supplementary table only, instead of the main text, or condensing Table 1.
• Include more explanation or citation on how SHAP values are computed and interpreted in the clustering context for readers unfamiliar with explainable AI techniques.

1. Results

Suggestions:

• Provide statistical significance or validation metrics for SHAP impacts (e.g., confidence intervals or feature importance rankings).
• The explanations of Beeswarm plots can be expanded for clarity.
• Include more information on model performance (e.g., accuracy, F1-score of RF classification for clusters).

1. Discussion

Suggestions:

• Integrate more discussion on potential policy or adaptation strategies based on cluster-specific vulnerabilities.
• Acknowledge limitations such as the temporal range of the data (1951–2018), and possible bias due to data resolution or missing climatic drivers.

1. Figures and Tables

Suggestions:

• Improve color consistency and legends for clarity (e.g., avoid ambiguous shades).
• Add numerical cluster centroids or representative climate patterns for each cluster.

1. Language and Style

Suggestions:

• Consider simplifying overly dense or jargon-heavy sentences (especially in the Introduction and Discussion).
• Check for consistency in the use of abbreviations (e.g., GMT vs. Global Mean Temperature) and ensure all acronyms are introduced properly.

1. Novelty and Impact

Suggestions:

• Emphasize more clearly in the Conclusion how the framework can be generalized to other regions beyond the Sahel.