General comments

The manuscript "Exploring the use of seasonal forecasts to adapt flood insurance premiums" by Viet Dung Nguyen et al. represents a significant and innovative contribution to the field of flood insurance by integrating seasonal flood forecasts into the adjustment of flood insurance premiums. This approach is both timely and highly relevant in the context of a potential increase in flood risks associated with climate change.

The manuscript stands out for its novel integration of hydrological and economic models. The use of non-stationary flood frequency analysis coupled with the Dynamic Integrated Flood Insurance (DIFI) model showcases an interdisciplinary approach that is important for addressing challenges in this kind of climate risk management. Additionally, the study's focus on Germany provides a valuable case study that offers insights into the application of these models in a specific geographic and climatic context.

The paper is well-structured, progressing logically from a detailed introduction that sets the stage for the research, through to the methodology, results, and a comprehensive discussion. The authors also discuss the limitations and potential implications of their findings, which is crucial for a balanced scientific discourse.

That said, the paper has several minor issues that need to be resolved before publication.

Specific comments

• Seasonal forecast choice: the authors explain that there are two approaches for seasonal flood forecasting, one based on dynamical models and another one based on statistical approaches. The authors note in line 70, "seasonal climate forecast from dynamical or statistical models have seen great advancements in recent years," and again in line 89, "Given the progress in recent years in seasonal flood forecasting methods…" However, the manuscript stops short of detailing these advancements or providing specific examples and references to substantiate these claims.
  • I’d suggest that the authors further develop these claims (with specific examples and references).
  • Linked to the previous point, I’d also suggest including some lines or a paragraph justifying the authors’ choice of going for a statistical approach with past data instead of directly using data from dynamical seasonal models. Factors might involve the comparative skill of statistical models against dynamical systems in the study area, the availability and resolution of historical data, limitations inherent to dynamical models that make them less suited to the specific objectives of this research, the availability of hydrological models or the scope of the paper as a proof of concept.
• Model selection: the choice of the non-stationary flood frequency model and the DIFI model is central to your study. It would be beneficial to provide a more detailed justification for selecting these specific models over other available models.
  • How do these models compare in terms of their predictive accuracy, computational efficiency, and sensitivity to different climatic and hydrological conditions?
  • Regarding the GEV distribution used from Steirou et al. 2022, have you checked how good does it fit to the 136 stations data? In Steirou et al. 2022 it seems that what is checked is the degree of improvement by using climate information in the location parameter in the GEV adjustment. But there is an inherent uncertainty on which is the best distribution fit when adjusting for return periods (a regular problem in hydrology, climate, and meteorology fields).
    • I’d suggest including some mention to this uncertainty. No need to perform additional computations, but Q-Q plots representing the observed quantiles vs. the theoretical quantiles from several distributions are a good tool to visualize this complexity.
• Methods and data section: this section, although extensive, currently falls short in providing a comprehensive and detailed description of the datasets and models employed. Data such the one coming from the Climatic Research Unit (CRU) and the Global Runoff Data Centre (GRDC), receive late mentions (line 339 for CRU and line 380 for GRDC) without prior introduction or elaboration on their characteristics. Similarly, models like GLOFRIS are introduced abruptly (line 278), leaving a gap in the reader's understanding of the methodological underpinnings of the study. I’d suggest reviewing this section and try to complete the missing information by, for example, creating a specific sub-section for ‘Data and Models’, leaving the ‘Methodology’ in another sub-section.
• Model calibration and validation: I’d also suggest elaborating more on the calibration and validation processes for your models. This is particularly important for the flood forecasting model, where predictive accuracy is key. How do the models perform in terms of key metrics like, i.e. root mean square error (RMSE) or skill scores against observed flood events? I’d suggest including more detail / discussion on these aspects (i.e. referencing other works).
• Inter-model interactions: your methodology involves integrating multiple complex models. A more detailed discussion on how these models interact, particularly how uncertainties in one model propagate through to others, would significantly enhance the robustness of your approach. How do uncertainties in the seasonal forecast model affect the accuracy of the flood frequency model and subsequently the DIFI model?
• Limitations: in the ‘Recommendations’ sub-section there is already a discussion on some of the limitations but, generally, they lack a bit of detail. For instance, in line 334-336: ‘although both models are similar in conceptional terms, they apply different sub-models and datasets for calculating risk. Future studies should apply a more consistent model chain’; although we can go to the original papers and check what these differences are and why they are important, it would be worth including some more specificity on why their use could make the model chain somehow less ‘consistent’.
• Spatial and temporal coverage: your study focuses on Germany, which has specific hydrological and climatic characteristics. How transferable are your findings to other geographical regions with different hydrological and climatic conditions?

Technical corrections

• ‘Policyholders’ in line 65 needs a hyphen.
• Figure 2 lacks the rivers (or they are barely visible, compared to figure 4).

Dear Authors, first thank you for your study on how seasonal flood forecasts can be integrated into the calculation of flood insurance premiums. You have done an excellent job of presenting your research and findings clearly and concisely. Overall, I believe that this manuscript makes an important contribution to our understanding of an underexplored area of flood risk management. With some minor revisions and additions, this paper has the potential to be an impactful publication in its field. 

Comments:

1. The section detailing the methods, data used, and the explanation of the seasonal forecasting model could be expanded to enhance clarity. Specifically, information on the lead time of the forecasts, the sources of predictability, and how these elements influence the model's accuracy and reliability should be more thoroughly explained.

2. The discussion section would benefit from an expanded exploration of human drivers within the model chain. 

Figure 2 Clarity: It is mentioned that rivers should be displayed with blue lines in Figure 2, but they are not visible.

L260 – I suggest avoiding commenting on results that are not shown.

L265 - Statistical Significance?