General Comments

This paper addresses the important challenge of accounting for storm splits and mergers in storm tracking algorithms. The authors demonstrate that spatiotemporal graphs are a natural fit for this problem. The storm detection, tracking, and graph creation methods are clearly explained and well justified. An excellent case is made for the use of graph-based storm tracking in storm nowcasting algorithms.

Specific Comments

To avoid confusion between the notions of predicting storm initiation and predicting development of existing storms, I recommend replacing “development” with “evolution” in at least some instances, including within the title.

In the Intro, consider additionally citing Skinner et al. (2018, Wea. and Forecasting) and Heinselman et al. (2025, Wea. and Forecasting) as examples of storm object identification and storm nowcasting studies, respectively.

The x’s in Fig. 3 are somewhat difficult to see.

Section 4: For clarity, I recommend emphasizing/reminding that the statistics presented here are valid for cell events, not for entire cell lifetimes.

L374–375: Why is the smaller sample size likely to bias the rates of increase/decrease low? Is this a post hoc assumption?

Technical Corrections

L265: Figure 1 → Figure 3

The manuscript presents an interesting and very complex topic, such as the tracking of convective cells in environments with multiple splittings and mergings. It is one of the most challenging issues in short-term forecasting due to the nature of multicell systems, with many interactions across different scales.

• Consider a more recent reference (at least add one) for VIL introduction (L131)
• Figure 2 is a bit confusing. t0 is not the current time, and it should be clearly explained. Furthermore, regarding this scheme, I have not been able to find the validity period. I mean, until when can a certain scheme be used for nowcasting?
• A doubt regarding the cell identification: have you tested in highly efficient precipitation clouds without ice particles or scarce ones? Is this technique effective in these cases? Please comment in the text.
• Another point referring to the cell identification: "Contiguous area identified from a radar image; represents an observation of a convective storm" is very vague and needs more precision. Are you considering single-threshold or multiple ones? Are you considering the same values for all the months? Note that VIL is highly sensitive to the seasonal variation. How have you solved this point?
• The presented examples are very illustrative, but I miss a more difficult situation, for instance, from May or September, with lower VIL values and, therefore, a more complex cell identification.
• Figure 6: Why do the ZDr index values not coincide with those of the other variables? (e.g. left panel, most representative cell index is 3, but for Zdr is 1)
• For the same figure, I understand that the current time is 18.05 and 10.05, respectively. Is this it? Please, clarify
• "Thus, the impact of splits and merges to the analysis depends on the definition of the cells that the analysis focuses on." I think the Authors should provide more information about this point, maybe with a discussion referring to other Authors.
• Regarding the number of splittings and mergings, have you found any correlation with the size of the cells?
• I don't understand the source of this conclusion: "It was documented that splits and merges occurred in 7.2% of all identified cells, and they were more frequent in cells with maximum VIL ≥ 20.0 kg m−2 (17.9%) or containing ZdrC features (11.7%)" (I were not able to find in the Results any comment regarding this)

Best regards