The paper provides an extensive study on downscaled retrospective forecast in the Northwest Atlantic Ocean from GFDL global model using a 1/12 configuration based on MOM6, previously designed and assessed by the Authors in another dedicated paper.

The methodology used for assessing the forecast is very interesting and quite comprehensive as well as the process-oriented analysis, supported by discussed results.

I very much appreciate the opportunity to review this manuscript. This paper focuses on evaluating seasonal predictability of surface temperature and salinity and bottom temperature over the North America East Coast by using a dynamically downscaled model forecast system (MOM6-NWA12) and compared with the parent SPEAR model forecasts. Detailed discussions about sources of improved prediction skill from downscaling are included for the Northeast U.S. region, as well as discussions on the effect of long-term warming trends over this area. Besides, this paper also contributes a useful discussion on the ensemble size for reasonable prediction skill when predicting SST. This is a very important work with high quality contributing to the research field, and I only have a few minor comments on this work:

1. Some description about seasons in the Result section are confusing, not sure if authors are talking about initialization seasons or forecast seasons. For example, on lines 237-239, “the downscaled model has skill greater than persistence and SPEAR across a wide range of times, except in the winter…”. It is not clear if “winter” here refers to the initialization month of December or forecast months in winter.

2. Lines 267-268, the Southeast U.S. LME, as shown in Figure 1, is not narrow compared to most other LMEs. I also question on its dominance by the Gulf Stream, as Gulf Stream is usually referred to the western boundary current north of Cape Hatteras (so north of the Southeast U.S. LME). 

3. Description of the forecast-observation mean bias (for Figs 5-7) could be more focused on those forecasts that have significant forecast-observation correlation coefficient (Figs 2-4). 

4. Lines 278-284: authors could just write out the season name, instead of “first season”, “last season”, and “seasons 0 and 2”.

5. Lines 294-295: (1) “remaining three regions” -> “remaining four regions”? (2) “aside from the increased spread in the Southeast U.S.” not sure why it is “increased” when comparing with SS and NEUS based on Figure 9, please consider rephrasing this sentence.

6. Line 340: “mid-Atlantic Bight” -> MAB

7. Figure 11: Please indicate correlation significance in the figure for each panel. Figure 11 shows the correlation in the GOM is minimum at Lead 6 but increases at Lead 7. Could you please explain it? Please consider adding the location of the Scotian Shelf box in Figure 1.

8. Figure 12-13: please consider adding correlation significance in each correlation map.

9. Figure 20: Do predictions of bottom temperature also require approximately 4 ensemble members to provide a reasonable compromise between computational costs and prediction skill?

Attached as pdf.