This study systematically investigates meteorological-to-groundwater drought propagation at the global scale using hyper-resolution (~1 km) groundwater simulations and classifies groundwater responses into three types to explore the global patterns and drivers of multi-year groundwater droughts. The authors also combine model simulations with observational data, which is an interesting and valuable attempt. Overall, this study provides useful insights into our understanding of meteorological-to-groundwater drought propagation. Before recommending the manuscript for publication, I have several comments and suggestions that I hope the authors will carefully consider and appropriately address.    

Main comments:  

1.The term “drought propagation” in the title is rather broad, whereas the manuscript primarily examines propagation from meteorological drought to groundwater drought. The authors may consider specifying “meteorological-to-groundwater drought propagation” in the title for greater precision.    

2.Please clarify how all reported global percentages were calculated. Are these based on the number of groundwater grid cells or on area-weighted fractions? If the former, please demonstrate that differences in grid-cell area do not bias the reported global percentages. I would also recommend replacing expressions such as “35% of the world” with a more precise description of the denominator.    

3.The validation indicates that GLOBGM tends to simulate longer groundwater droughts than observed, particularly against GRACE (mean duration of 9.9 months in the simulations versus 5.5 months in GRACE). This is highly relevant because the central result of the manuscript is the prevalence of multi-year groundwater drought and the classification of systems based on drought duration. Could you quantify how this apparent persistence bias affects the estimated fraction of the globe experiencing multi-year groundwater drought (35%) and the proportion classified as meteo<GW? At minimum, I think this uncertainty needs to be discussed much more explicitly when interpreting the global percentages.  

 4.The authors mentioned that long-term groundwater decline violates the stationarity assumption underlying the standardized anomaly approach and may artificially produce very long drought events. However, given that drought duration is the central diagnostic of this study, I do not think that simply recommending interpretation alongside a trend mask is sufficient. Could the authors quantify the sensitivity of the main results after excluding grid cells with strong groundwater trends, or at least stratify the reported percentages by trend magnitude? In particular, it would be important to know whether the reported 35% global multi-year fraction and the meteo<GW category are disproportionately associated with strongly declining groundwater systems.        

Other comments:  

1.Line 123: Please briefly discuss the sensitivity of the results to the 12-month threshold used to distinguish ND and MYD. For example, would the major spatial patterns remain similar using 18 or 24 months?    

2.Line 134: Has the sensitivity of the results to the drought threshold been tested (e.g., −0.5, −1.0, −1.5)? Since event duration can be highly threshold-dependent, a brief sensitivity analysis would strengthen the robustness of the results.    

3.Line 179: “DRR” should presumably be “DDR”.  

 4.Line 191: I would consider replacing “confidence bounds” with “sensitivity bounds” or another term unless a formal confidence level can be associated with these intervals.    

5.Line 217: The term “groundwater drought predictability” appears not suitable for the ROC-AUC analysis. As currently implemented, the ROC-AUC quantifies the ability of contemporaneous SPEI-12 values to discriminate groundwater drought from non-drought months, rather than predictive skill in a forecasting sense. Unless an out-of-sample or lead-time prediction framework is introduced, I suggest using terms such as “discriminatory ability” or “meteorological indicator skill” and moderating the forecasting interpretation.      

6.The groundwater simulations are at ~1 km, whereas recharge is ~10 km and SPEI is 0.5°. Please discuss more explicitly how this substantial scale mismatch affects interpretation of the apparent fine-scale spatial variability in drought propagation and response types.    

7.Please explain how the three “representative” time series in Fig. 5 were selected. Were they selected using objective criteria or chosen visually? Providing their geographic locations would also be helpful.    

8.Line 403: The use of “tipping points” seems speculative based on the current analysis. Unless a threshold/nonlinear transition is explicitly demonstrated, I suggest describing this as a hypothesis or potential mechanism more cautiously.