the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 10 Sep 2025
Impacts of Air–sea Coupling on Systematic Errors in Medium-Range Winter Forecasts over the North Pacific and North Atlantic
Abstract. The impact of air-sea coupling in North Pacific and North Atlantic medium-range forecasts during winter is assessed using 20 years (1998–2017) of hindcasts produced by the Global Ensemble Prediction System (GEPS) of Environment and Climate Change Canada (ECCC). We compare an uncoupled atmospheric model (versions 5, GEPS5) with an atmosphere-ocean coupled model (version 6, GEPS6) alongside European Centre for Medium-Range Weather Forecasts Reanalysis v5 (ERA5) as the verification dataset. We find that by the third pentad, or days 11–15, coupling weakens the Aleutian Low, the Icelandic Low, and the Atlantic Subtropical High. This produces less integrated vapor transport (IVT) over the Pacific and Atlantic Oceans, whose spatial patterns are modulated by phases of Madden-Julian oscillation (MJO). Coupling also results in colder sea surface temperature (SST) over the Kuroshio Current Extension region and produces a weaker Aleutian Low due to less upward latent heat fluxes. The weaker Aleutian Low further reinforces its weakening through a positive feedback loop. Lastly, the air-sea coupling reduces the latent heat flux bias variance by 10–20 %, thus improving the IVT.
- Preprint
(6822 KB) - Metadata XML
-
Supplement
(82 KB) - BibTeX
- EndNote
Status: open (until 21 Nov 2025)
-
RC1: 'Comment on egusphere-2025-4142', Kristian Strommen, 27 Oct 2025
reply
-
RC2: 'Reply on RC1', Kristian Strommen, 27 Oct 2025
reply
It seems the formatting on my response was botched. In particular, the separation of the various minor comments didn't come out neatly. Hopefully it is clear how they are grouped, but in case of confusion I've attached a PDF of the review as well. Sorry about this.
-
RC2: 'Reply on RC1', Kristian Strommen, 27 Oct 2025
reply
-
RC3: 'Comment on egusphere-2025-4142', Anonymous Referee #2, 12 Nov 2025
reply
The paper describes the differences between seasonal hindcasts produced using the GEPS5 (uncoupled) and GEPS6 (coupled) systems. The authors identify some interesting coupled feedbacks and the paper is generally well written and presented. However, there are some significant issues that need to be addressed in order to make it acceptable for publication.
Major comments:
1) The biggest issue is that the initialisation of the SST between the two systems is so different and uses different start dates. This makes it very difficult to be certain that the changes highlighted are due to coupling and not due to the initialisation method. You need to distinguish between the impact of SST differences due to initialisation and SST differences due to running a coupled model.
2) The authors show the differences between the biases of the two systems but they don’t state whether the changes are an improvement or not. i.e. where GEPS6 is colder than GEPS5, is GEPS5 warm compared to observations and therefore GEPS6 is better or is GEPS6 too cold. I note in section 3.4 you suggest that GEPS6 is better and cite Lin et al 2019. This appears to be an internal document at ECCC that I have been unable to find online. Ideally, the bibliography should state how to get this document or you should include the SST bias plots in the current paper. It would be better to discuss this as part of section 3.1 rather than where you cite the Lin report in section 3.4.
Minor Comments:
Section 3.3: You identify the positive feedback between the cold SST and Z850. Given my comment above about initialisation, can you be sure that the colder SST in the coupled model is due to being coupled or due to the initialisation strategy. Also, how far into the forecasts does this positive feedback persist? Presumably, it must reach an equilibrium at some point.
Figure 3: It’s very hard to see the location of the continents on this figure.
Figure 6: You include this in the paper but there is no reference to it at any point in the text. I think some discussion of this figure would be interesting to understand whether the impact of the dynamical coupled model is different according to the MJO phase. Perhaps this is what you are referring to in your conclusions at line 226 (“The IVT improvement is also more significant when the MJO is active”). If so, this should be introduced and discussed before this point in the paper.
Lines 229-231: This sentence confused me. You haven’t discussed the parameterisations used for calculating the air-sea fluxes, just compared the results with different SST patterns. Are you suggesting that your results motivate further work to understand why the latent heat flux bias variance is reduced even when the SST bias variance is increased?
Citation: https://doi.org/10.5194/egusphere-2025-4142-RC3
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 561 | 47 | 11 | 619 | 31 | 15 | 23 |
- HTML: 561
- PDF: 47
- XML: 11
- Total: 619
- Supplement: 31
- BibTeX: 15
- EndNote: 23
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
GENERAL COMMENTS
This paper compares ensemble forecasts of the GEPS6, which is dynamically coupled to NEMO, with hindcasts of GEPS5, which uses persisted SST anomalies. The authors find changes in the mean and variance between the two forecasts, both in the ocean and the atmosphere. The changes are discussed in terms of the effect of air-sea coupling.
While there is some interesting analysis here, I think the paper as it currently stands has several major problems. Some of this relates to the framing of the paper (the title makes it out to be about air-sea coupling but is actually about the broader effect of using a dynamical ocean model), and some of it relates to a potentially serious confounding effect (the initialisation of GEPS6 hindcasts is apparently very different from GEPS5 hindcasts). There are also many missing references to past literature.
I flesh out these and other issues in my comments below. Major revisions will be needed to address them. I look forward to reading a revised version.
Best wishes,
Kristian Strommen
MAJOR COMMENTS
Studies that aim to really isolate coupling often deal with this by looking at things like lead-lag correlations between SSTs and wind-stress or fluxes, since correlations ignore magnitude and thus are insensitive (at least a priori) to model biases. For an example, see e.g. https://doi.org/10.1002/2016GL070559.
I think unless you want to almost completely redo the paper to follow similar methods, you need to reframe the paper to be much more specifically about the impact of using a coupled dynamical ocean model in your forecast. However, at this point it’s clear the results depend sensitively on the exact model, since this determines the model biases. Thus, I think the authors should rephrase everything to be very specifically about the comparison between GEPS6 and GEPS5. This includes mentioning GEPS somewhere in the title. Air-sea coupling should not be mentioned in the title unless considerable additional analysis along the lines of the Roberts et al. paper (or similar) is added.
MINOR COMMENTS
Section 2.3: I don’t follow the reasoning here. You say that you can’t directly compare the coupled and uncoupled forecasts because the start dates differ, and so you rather compare the two biases instead. However, unless I misunderstood something about the exact computation, this ends up being the same thing: (GEPS6-ERA5)-(GEPS5-ERA5)=GEPS6-GEPS5. So your bias difference plots are just showing GEPS6-GEPS5 anyway. I don’t think you can sidestep the problem that the initialization days are different. You just need to mention this as a confounder and discuss how much you think the results depend on it.
“To test the significance, the degrees of freedom are counted by making the following two assumptions: (a) output from different start times or different ensembles is independent”
I guess you mean different ensemble members, not different ensembles. As for the first point, this should be fine as long as the start times are relatively spaced out. Can you comment on the typical distance between start dates? The information is in the supplementary tables but it is convenient if you just state this here for the reader.
Figures 2/3: Can you make the continents more visible? Coastlines blend in with contour lines, making it hard to distinguish the two.
L140: You should include a few lines on how the shift in the Gulf Stream is very likely related to the inability of NEMO at ¼ degree to resolve the Gulf Stream properly, and cite some references for the role of model resolution. I don’t know as much about the Kuroshio current, but I’m sure model biases in this current, and likely origins of such biases, have been looked at in past studies, so it would be good to discuss these briefly as well. Alternatively, you could add this discussion to your section 4, but if so, please mention here that you will discuss these biases further in section 4.
L150: The link between the Aleutian and Icelandic lows is known and documented, see e.g. Honda et al. (2001): https://doi.org/10.1175/1520-0442(2001)014<1029:ISBTAA>2.0.CO;2 Please add some references here.
L171/172: “indicating that the coupling results in a colder SST” Can you add a comment on why this might be? This comment might be related to the above comment about past literature on Kuroshio current biases in models.
L208: Figure 3d-f should presumably refer to Figure 1d-f.
L211: “Future numerical studies are needed to gain a deeper understanding.” Figure 1f and 1i show an NAO pattern in the Euro-Atlantic. The relationship between changes in the Gulf Stream and changes in the NAO have been investigated in many past studies, see e.g. this paper and references therein: https://doi.org/10.1029/2025GL117228
More pragmatically, the NAO is the dominant mode of variability there so if you change the SSTs in this region then the atmospheric change is very likely going to project onto the NAO. Please add some comments on this, especially on the past literature.
L216: “We also notice a possible teleconnection from the Aleutian Low through the Arctic into Icelandic Low via a Rossby wave train.” Since this teleconnection is known (see above), you should rephrase to rather say that the changes to the Aleutian Low affect the Icelandic Low via a Rossby wave train, and then cite Honda again.
L233/234: “Second, there is a need for more physical understanding of how two-way coupling produces better air–sea fluxes.” There are some classic relevant studies on this. Most notably, Barusgli and Battisti (1998) needs to be mentioned here: https://doi.org/10.1175/1520-0469(1998)055<0477:TBEOAO>2.0.CO;2
In this paper they clearly explain the effect of coupling versus no coupling on heat flux and surface temperature variability. In particular, the low frequency variability in surface temperature (and, I believe, heat fluxes) will be wrong in uncoupled models due to the excess thermal damping effect they explain there. This is fundamentally related to the fact that the ocean acts as an infinite sink/source of energy in an uncoupled simulation. It seems plausible that changes in the latent heat flux bias variance you see could be related to this. You don’t necessarily need to demonstrate this decisively, but some comments at least are necessary.