This manuscript presents a modeling framework for evaluating the joint influence of non-tidal residuals (NTR), rainfall (RF), and mean sea level variability on coastal flooding in the Gloucester City area, using the SFINCS hydrodynamic model and a copula-based statistical approach. The topic is timely and relevant, and the study is generally well-structured with a strong emphasis on scenario-based risk quantification. However, several methodological choices—particularly regarding data selection, parameter thresholds, and model assumptions—require further clarification or justification. Issues such as the generalization of AORC performance, the treatment of tropical versus non-tropical events, and simplifications in the SFINCS physics raise concerns about the robustness and generalizability of the findings. Despite these limitations, the study offers valuable insights into compound flood risk assessment. Detailed comments are provided below, which I hope will be useful in clarifying and strengthening the manuscript:

Page 5, Line 146: The sentence claiming that AORC has “higher accuracy” than other gridded rainfall datasets seems too general. For example, radar-based products like MRMS have been shown to perform as well as or better than AORC in some events, including Hurricane Harvey (e.g., Gao et al., 2021; Gomez et al., 2024). I suggest the authors either include MRMS in their comparison or rephrase the sentence to clarify that AORC’s performance advantage may depend on the region or event type.

Refs.:

Gao, S., Zhang, J., Li, D., Jiang, H., & Fang, Z. N. (2021). Evaluation of multiradar multisensor and stage IV quantitative precipitation estimates during Hurricane Harvey. Natural Hazards Review, 22(1), 04020057.

Gomez, F. J., Jafarzadegan, K., Moftakhari, H., & Moradkhani, H. (2024). Probabilistic flood inundation mapping through copula Bayesian multi-modeling of precipitation products. Natural Hazards and Earth System Sciences, 24(8), 2647-2665.

Page 6, Line 160-166: In Section 4.1, several choices such as the 3-day pairing window for NTR and RF, the 5-day declustering period, and the 350 km radius for identifying TC events are not clearly explained. It would be helpful to clarify whether these are based on physical reasoning, prior studies, or simply assumptions made for this analysis. Providing brief justifications or references would improve transparency and reproducibility.

Page 8, Line 217-228: Could the authors clarify the physical justification for uniformly scaling entire NTR and RF time series based solely on peak values? For example, does this approach preserve key timing or intensity ratios in cases with asymmetrical hydrographs or localized RF bursts?

Page 10, Line 273-275: The use of SFINCS is well-suited for handling large scenario sets; however, two model limitations warrant further discussion. First, SFINCS does not explicitly model nonlinear tide–surge interactions, which can influence the timing and amplitude of water levels in estuarine environments (e.g., Arns et al., 2020; Dullaart et al., 2023). Second, the omission of advection in the local inertia formulation may affect surge dynamics in narrow tidal channels like those surrounding Gloucester City. I recommend the authors provide a brief sensitivity analysis or comparison illustrating the impact of including vs. excluding the advection term, as SFINCS offers both options (Leijnse et al., 2021).

Refs.:

Arns, A., Wahl, T., Wolff, C., Vafeidis, A. T., Haigh, I. D., Woodworth, P., Niehüser, S., & Jensen, J. (2020). Non-linear interaction modulates global extreme sea levels, coastal flood exposure, and impacts. Nature Communications, 11, 1918.

Dullaart, J. C. M., Muis, S., de Moel, H., Ward, P. J., Eilander, D., & Aerts, J. C. J. H. (2023). Enabling dynamic modelling of coastal flooding by defining storm tide hydrographs. Natural Hazards and Earth System Sciences, 23, 1847–1862.

Leijnse, T., Dazzi, S., Yu, D., & Bates, P. D. (2021). Efficient coastal flood hazard mapping with a 2D non-inertia model. Coastal Engineering, 170, 103994.

Page 11, Line 300-301: The authors use a fixed NTR threshold of 0.63 m to yield ~5 exceedances per year, which is reasonable and aligns with past compound flood studies. However, the threshold selection could be strengthened by applying one of several recent automated, data-driven approaches developed for POT analysis, such as the Sequential Goodness-of-Fit method (Bader et al., 2018), the Extrapolated-Height Stability method (Liang et al., 2019), the L-moment Ratio Stability method (Silva Lomba & Fraga Alves, 2020), or the comparative multi-method approach applied in a coastal flood design context by Radfar et al. (2022).

Refs.:

Bader, B., Yan, J., & Zhang, X. (2018). Automated threshold selection in extreme value analysis via goodness-of-fit tests with adjustment for false discovery rate. Annals of Applied Statistics, 12(1), 310–329.

Liang, B., Shao, Z., Li, H., Shao, M., Lee, D., 2019. An automated threshold selection method based on the characteristic of extrapolated significant wave heights. Coast. Eng. 144, 22–32.

Radfar, S., Shafieefar, M., & Akbari, H. (2022). Impact of copula model selection on reliability-based design optimization of a rubble mound breakwater. Ocean Engineering, 260, 112023.

Silva Lomba, J., Fraga Alves, M.I., 2020. L-moments for automatic threshold selection in extreme value analysis. Stoch. Environ. Res. Risk Assess. 34 (3), 465–491.

Page 12, Line 328-336: While the authors maintain stratification for joint probability estimation, they combine TC and non-TC time series for event generation based on overlapping confidence intervals and similar time series shapes. Given the well-established physical differences between tropical and extratropical systems (precipitation structures, spatial scales, storm tracks), could the authors clarify how confident they are that this approach adequately preserves the distinct characteristics of these storm types?

Additionally, given the limited number of TC events, how do the authors assess whether their analysis has sufficient statistical power to detect meaningful differences? Would alternative approaches like physics-based conditioning (e.g., storm track or seasonal constraints) potentially better preserve known meteorological distinctions while addressing sample size limitations?

Page 20, Figure 10: The authors demonstrate substantial flood depth due to MSL and tidal variability using a single most-likely 0.01 AEP event. While this effectively illustrates the potential importance of these factors, could the authors comment on whether this sensitivity pattern is representative across different event types and return periods?

Additionally, given that the most pronounced effects occur along the Delaware River and Newton Creek boundaries, could the authors discuss whether the model's spatial resolution, boundary condition placement, or coastal setup might be influencing the magnitude of these sensitivities?

The manuscript “Generating Boundary Conditions for Compound Flood Modeling in a Probabilistic Framework” by Maduwantha et al. introduces a statistical framework designed to generate many synthetic but physically plausible compound events, including storm-tide hydrographs and rainfall fields, which can serve as boundary conditions for dynamic compound flood models. The framework is later applied to the case of Gloucester City in New Jersey.

The topic of the manuscript is highly relevant for quantifying flood hazard, particularly water depth resulting from the joint occurrence of storm surge and rainfall. However, the proposed framework requires further clarification and additional information to assess its validity better and ensure reproducibility by others.

The novelty of the proposed framework should be better highlighted. If I understand correctly, the essence of the proposed framework is to select joint events of NTR and rainfall from historical observations, and then fit a bivariate copula to generate “unseen” pairs. Pairs are used to amplify historical time series of NTR and rainfall over a short period of time with a temporal resolution consistent with the one required by the hydrodynamic model. An almost identical workflow was proposed by Xu, H et al (2024) "Combining statistical and hydrodynamic models to assess compound flood hazards from rainfall and storm surge: a case study of Shanghai", Hydrol. Earth Syst. Sci., 28, 3919–3930, https://doi.org/10.5194/hess-28-3919-2024. How does this framework differ from Xu et al 2024? How does this framework differ from previous studies? The novelty is hidden in the introduction.

In lines 502-503, the Authors say that  “Although measures are taken to prevent the generation of physically unrealistic events (see Section 4.3), it cannot be fully ruled out.” If this statement is true, then the Authors cannot claim that the framework generated physically plausible compound events (Abstract - Lines 16-17). This is quite an important point, and the Authors need to be transparent about the potential of the framework to generate physically plausible events or not.

The data selection procedure and its effects on the results need further clarification.

First, synchronous NTR are selected, and later on astronomical tide and mean sea level are added. Did the Authors consider using the concept of skew surge? If not, why? In addition, tide-surge interaction is mentioned but not really discussed. How relevant is it? Would the concept of skew surge solve it?

Second, multiple rainfall measurements are considered. However, it is unclear how such measurements are aggregated and how this aggregation affects the dependence between NTR and rainfall, and so the fitted copula. Moreover, which rainfall measurement is used as a reference for the lag time between peak surge and peak rainfall?

Finally, the distinction between TC and non-TC leads to copulas with different asymmetries. How do the Authors justify such differences? What happens when the TC and non-TC are combined together? I would say this is mostly relevant in paragraph 5.3 “Event Generation Process”. How do the Authors know that the 106-year event (which is also quite an interesting number!) corresponds to a TC? Given the length of the data, the 106-year event is inferred from the copula and not observed. Regardless of how the Authors track whether it is a TC or a non-TC, how sensitive are the results to the type of event? For example, what is the difference between a TC and non-TC event with the same return period? What about in terms of the drivers’ magnitude and water depth? I suggest adding some sensitivity analysis to the assumptions made, including the lag time between peak rainfall and peak surge.

Minor comments.

The “target event” is never explicitly defined, and from my personal perspective, this creates some confusion. How is it how is it selected?

ETC and non-TCI seem to be used interchangeably. I suggest checking the notation for consistency.

Line 409: the Authors say that flood depth varies in some regions. However, the case study seems to concern only one region. I would suggest checking this sentence.

The comparison between rainfall and surge is done considering duration. How did the Authors handle discrete variables when assessing correlation?