The manuscript addresses causal relationships between climate mode indices. Even assuming that statistical associations between indices reflecting very distinct spatial and temporal scales of variability is a valid exercise to infer climate dynamics, in my opinion the manuscript needs to be revised for improved clarity and readability. The quality of the figures needs to be substantially improved.

Title: IPO should be replaced by Interdecadal Pacific Oscillation

The structure of section 2 is very confusing, including both data description and results before the actual description of the methodology used to produce the results. It’s not clear why Figure 1a) is not a separate figure and instead is mixed with results in figures 1b)-d). Furthermore, Figure 1a) is poorly described, the legend includes text such as “mw (x5), lp, nh=7, which is not described in the caption (and it’s not clear whether really needed or it’s relevance). What is gained from such smoothed signals in a relatively short time series is not evident, and should be better described in the manuscript.

The data description and it’s presentation in table 1 needs to be improved in order to clarify how each climate index was indeed calculated and at what the temporal resolution. The minimalist caption in Table 1 should be improved. What does the * means in RMM1 and RMM2?

Table 2 should be better described both in terms of the caption itself and in the text. The meaning of the “All years” column in Table 2 is not clear. More importantly, the objective criteria by which the positive and negative IPO phases are identified should be clearly and unambiguously stated.

The posterior probability plots should be improved (Figures 1b)-d), Figures 3, 4, and 5 by adding axes (as in Figure 1b), for consistency), and particularly to improve readability of the posterior probabilities, as in the current configuration it is not possible to effectively distinguish between low and intermediate probabilities. Maybe just providing the two highest probability ranges and distinguishing between them in another way other than width of the line (for example by using solid and dashed or dotted lines) would enable to reader to actually see the results, with the current design it’s almost impossible.

This study uses Bayesian structure learning to examine causal relationships between major tropical and extratropical climate modes and identifies notable regime-dependent differences in climate dynamics. The work offers an interesting perspective on how these relationships may vary across regimes. However, in its current form, the manuscript would benefit from clearer calarify, and stronger justification of methodological choices. I therefore recommend further revision before the study can be considered for publication.

1. While the Introduction provides substantial background on IPO, ENSO, and causal discovery methods, the core scientific question is not stated sharply enough. For example, it is unclear whether the primary objective is to demonstrate that causal network structures differ between IPO phases.

2. Several inferred links, particularly involving the MJO, SAM, and NAO, are described descriptively (e.g., as edges in the graph), but their physical interpretation remains vague. The manuscript would benefit from deeper discussion on what lagged statistical causality implies in the context of atmospheric teleconnections and how these findings relate to established mechanisms.

3. The classification of IPO phases is based on a 30-year moving window and manual selection (Table 2), but the specific criteria and sensitivity of this approach are not well explained. Similarly, the choice of a 6-month maximum lag in the structure learning is not justified in terms of known climate dynamics. Both choices are critical to the causal analysis and should be more clearly motivated and discussed.

4. Several figures (e.g., Figures 1b–d, 3, 4, 5) are difficult to interpret due to visual overloading and unclear encoding. Posterior probabilities are represented solely by line width, which is hard to distinguish visually, especially without a legend or clear thresholding. Some axes and labels are inconsistent or missing. I suggest using line styles or colors to improve clarity.

5. The manuscript mixes data, methods, and results in ways that make it difficult to follow. For instance, Section 2 presents results (e.g., posterior causal networks in Figures 1b–d) before the causal inference method is properly introduced in Section 3. Figure 1a, showing IPO phase identification, also appears abruptly without a clear explanation of the smoothing/filtering methods or its connection to the subsequent analysis. I recommend reorganizing the manuscript so that methods are presented before results, and separating Figure 1 into more coherent units with clearer captions and contextual framing.