Review of the paper “Accelerating compound flood risk assessments through active learning: A case study of Charleston County (USA)” submitted to NHESS by Terlinden-Ruhl

The study explores the application of active learning to improve the accuracy of compound flood risk assessments while minimizing computational demands. The framework leverages the input-output uncertainty related to economic damages to minimize the number of required hydrodynamic and impact simulations. Compared to traditional equidistant sampling, the proposed active learning modeling framework shows lower RMSE with lower computational time.

I would like to compliment the authors for a well-written manuscript. To the best of my knowledge, this is the first study to apply active learning techniques in the context of compound flooding, demonstrating an approach to advancing this area of research. I recommend this manuscript for publication after addressing the minor revisions outlined below. However, I am not qualified to evaluate the detailed technical aspects of active learning.

1. The authors used the “skew surge” parameter to represent the storm surge component of water levels. However, one of the key parameters for simulating flooding mentioned in the manuscript is the “duration” of the event. The skew surge is a measure of the storm surge integrated over a tidal cycle, and thus it has no duration associated with it. How are the authors defining the duration (or time series) of the skew surge? In addition, SFINCS needs a water level time series as boundary conditions to simulate coastal flooding, how do the authors generate a storm-tide hydrograph of these events using the skew surge (i.e. a single value over a tidal cycle)? Consider explaining more about these steps.
2. The methodology for obtaining the skew surge time series is somewhat unclear (see L 112). Specifically, the paper does not specify how tidal levels were determined. Were NOAA-predicted tides used, or was a harmonic analysis performed on water level data? Additionally, when removing sea-level rise, the phrase “subtracting 1-year moving average from the skew surge time series” is unclear.
3. The uncertainty bands presented in Figure 6 indicate that, in some cases, the return value of a 1000-year event has narrower uncertainty compared to a 10-year event. Does this result only reflect the uncertainty associated with the surrogate damage model for each return period? It is given that a direct comparison of confidence intervals may not be valid among different numbers of stochastic variables, as the stopping criteria are based on uncertainty. Could the authors provide an additional explanation regarding the interpretation of these uncertainty bands? What information do they provide? 
4. L 207: Authors explain that they used a Gaussian distribution to reconstruct the water levels time series. It would be better to cite a previous study or provide a proper justification for this assumption. Same for the rainfall, how well does their simplification of the hydrograph and hyetographs represent real events? Additionally, how well we can represent real events using a constant rainfall (point estimate) over the entire domain or constant hydrographs along the coast? I think this is an important point that needs to be addressed at least as a limitation of the analysis presented.
5. The authors use the SFINCS model, which was set up and validated in a previous study. While this provides a solid foundation, I believe it would be beneficial to include some additional key details about the model setup in this paper. For example, providing information on the model resolution, whether constant or spatially varying roughness was used, and how infiltration was handled (if included).
6. L 134:  The description of applying a threshold to skew surge magnitude when it “co-occurred with a higher high tide” is confusing. Since the definition of skew surge inherently depends on the high tide (or higher high tide) in a tidal cycle, further clarification would benefit the reader.
7. L 3-5: The phrase “large flood hazard when compared to the sum of the individual drivers” could be explained better.
8. L 138: Is there a specific rationale for selecting a wider (14-day) de-clustering window? Justifying this choice would strengthen the methodology section.
9. L 144: The manuscript suggests a minimum duration of 6 days for rainfall and skew surge events. Does it mean that there were no hours with rainfall less than 0.3 mm within 6 days during an event? So all the extreme events were simulated for at least 6 days? Could the authors confirm and clarify?
10. L 352: Please specify whether the given time durations refer to the computational time required “per simulation” or the total time for simulating all events.
11. In Figure 2, ensure that the colors representing coastal and inland counties are correct.

The paper is well structured, clear and complete in all its parts. Although I am not an expert in active learning, I believe that the different assumptions and the application of the models are scientifically sound.

I believe the methodology is original and I think the paper can be published with some minimal technical corrections.

In the course of the text there is more reference to the output of the procedure. I think I understood that it refers to different geographical areas. If so, I ask the authors to specify explicitly in the text

Please authors check the colors of the legend in figure 2