This manuscript assesses the skill of the seasonal streamflow forecast derived from the

operational forecast system SYS8, based on CNRM-CM6-1. This model incorporates an advanced river routing model that interacts with the land and atmosphere via surface/sub-surface runoff, aquifer exchange and open water evaporation to predict river streamflow. Results show that the coupled forecast system is more skillful than the ESP in predicting streamflow for the summer and winter seasons. The manuscript particularly emphasizes the enhanced land water storage initial conditions, especially in summer and in large basins where the low-flow response is influenced by soil water storage. This manuscript demonstrates the potential to utilise direct global streamflow forecasts issued by a global climate model fully coupled with a river-floodplain model, and shows the capacity of an operational seasonal forecast system based on a General Circulation Model to deliver relevant streamflow predictions. I have a few comments below.

1. The author evaluated summer and winter streamflow forecast one month in advance, I wonder whether two or even three lead months are considered (not 3-month mean)? Are the results consistent with the conclusions in the manuscript, or are the results of the online coupled forecast system used superior to the ESP, as the authors state in the introduction, “while modified versions of ESP can improve streamflow predictions for shorter lead times, their skill decreases faster over time compared to NWPB systems”, maybe for longer lead times, the effectiveness of online coupled forecast system in predicting seasonal streamflow improves more.
2. Figure 1 demonstrates comparison of model configurations, but it is not sufficiently intuitive to understand and what the numbers in the figure represent is not explained. Also, “to generate the benchmark hindcast Offline_ICL, the land-river model ISBA-CTRIP is forced by ERA5 historical climate (Figure 1) so that each year produces one of the 25 forecast members” (Line 143), why does each year produce one of the 25 forecast members? The author mentioned 25 members several times, what specifically does members refer to?
3. In chapter 3.2, the author shows the performance of the atmospheric seasonal forecast is presented in Figures 5 and 6, in particular, precipitation and near-surface temperature. Please highlight in the figures where the author mentioned in the paragraph. In Figure 5, the ACC of global precipitation is overall lower in Online_ICLnud than in Online_ICL in summer, especially in South America and Australia, also Online_ICLnud has more blue areas than Online_ICL in winter that means more negative ACC of precipitation and near-surface temperature. Can the author give some explanations?
4. Line 50: “Conversely, in regions dominated by rainfall, FCAs tend to significantly influence…”, what does FCAs mean?
5. Lines 147-149: “For example, to apply the L3OCV selection method to the hindcast of 1993, only forcing from 1996 to 2020 ensures 25 members. For the hindcast of 2000, only forcing from 1992 to 1998 and 2003 to 2020 is used.” The previous article refers to the period from 1993 to 2017, please confirm the range.
6. Line 172: The values written in the article is not the same as in Figure 2.
7. I suggest the color bar is divided by 0, which makes it possible to visualize the changes in the indicator more clearly in Figs.3-4, and whether the horizontal coordinates of the last column of Figs.3-4 are displayed incorrectly.

Review comments for Skilful Seasonal Streamflow Forecasting Using a Fully Coupled Global Climate Model by Narváez-Campo, G. and Ardilouze, C.:

This study assesses the SYS8 (Météo-France) fully coupled forecast system for seasonal streamflow prediction. Results show higher skill than the Ensemble Streamflow Prediction (ESP) approach, particularly in summer (due to improved land water storage) and winter (better climate anomaly predictions). The results of this study are valid and highlight the operational value of global GCM-based SSF. Overall, the study is methodologically sound, but major revisions are required, particularly in the interpretation of results, which needs to be more structural. Specific comments are as follows:

Section 2.4: More explanation is needed here. Is streamflow bias correction applied only to the online models? If so, how is the comparison fair when offline models are not post-processed?

Section 3.1, Figure 3: Why do panels (c, f, i) all use percent mean bias (%) as the x-axis? This makes the figure difficult to interpret. The same issue applies to Figure 4.

Line 212: Clarify what is meant by "dryest regions" here. The same applies to Figure 4.

Line 213: The argument for this figure is not very clear to me. The current phrasing implies that the positive/negative bias directions remain unchanged from ICL to ICLnud, which is not necessarily the case. The reduction of negative bias refers to that some original blue-marked points in (a) got red points in (b), but as I understand, the figure (b) shows the difference in the absolute value of bias, meaning the ICLnud bias can be either negative or positive. While the claim that bias is reduced is still valid. Maybe try to rephrase the argument, and it would be useful to also show the bias of ICLnud, perhaps in an appendix.

Lines 232-235: The description for locations is inconsistent, sometimes referring to latitude, sometimes to country names, which is difficult to follow. Also, the interpretations themselves sometime do not match with each other. For example, in Line 234, it is stated that Europe shows improved precipitation predictions, but the previous sentence states degradation in the north of 40°N. Similarly, Australia is within the range that is described as showing improved correlation as it is south of 20°S, but it shows degradation in the figure. In general, this figure is difficult to interpret. Consider either adding highlighted boxes on the plot to clearly mark the areas being discussed, or use a better way to specify the area in the text.

Line 253: "From Online_ICLnud to Online_ICL, the ACC is slightly reduced, especially for basin outlets north of 40°N." Is this correct? It seems like the opposite may be true, please verify.

Line 255, Figures 7: Are the online system results in this figure bias-corrected or not? Some explanation would help to understand.

Figure 7: The red color is used to represent better performance, which is somehow difficult to remember. Consider either adjusting the color scheme or adding a note in the legend to show the optimal side.

Figure 12: There are red curve lines overlapping with the legend text.

Line 300: "Arid regions" here, which specific areas are being referred to? This is not clear to me.

Line 313: The phrase "remains limited for most conrespecttinents" likely contains a typo. Please check or clarify.