| 02 Jan 2026
The impact of stochastic sea ice perturbations on seasonal forecasts
Abstract. Sea ice ensemble forecasts can be highly underdispersive, meaning that the ensemble spread is notably lower than the average forecast error. One common strategy to address underdispersion is to add stochastic perturbations to the forecasts. We detail the implementation of a stochastically perturbed parameter (SPP) scheme for SI3, the sea ice component used by the Integrated Forecast System (IFS), the forecast model used and developed by the European Centre for Medium-Range Weather Forecasts (ECMWF). We then evaluate its impact on seasonal forecasts of northern hemisphere summer and winter. The inclusion of SPP is found to enhance ensemble spread for sea ice concentration (SIC) and sea ice thickness (SIT) forecasts by around 10 % relative to an unperturbed forecast, which results in a better calibrated probabilistic forecast. Some small but robust changes to the mean state are also found, including a general decrease in the mean SIC and a redistribution of the winter ice from the central Arctic to the ice edge. These changes reduce or increase the mean bias depending on the region. Changes to the mean and spread of the sea ice result in changes to the mean and spread of air temperature up to at least 850 hPa, altering the mean air temperature biases of the model. An apparent consequence of this is a significant increase in seasonal forecast skill of 500 hPa geopotential (Z500) over the Euro-Atlantic domain in winter, which partially projects onto the North Atlantic Oscillation. We conclude that sea ice stochastic perturbations can be a valuable contribution to increased reliability of seasonal forecasts of the sea ice itself and can impact seasonal forecasts of the atmosphere at high and mid latitudes.
Summary
This manuscript presents an implementation and evaluation of a stochastically perturbed parameter (SPP) scheme applied to the sea ice component of ECMWF's Integrated Forecast System (IFS). The authors assess the impact of this scheme on ensemble spread, mean state, and seasonal forecast skill for both sea ice and atmospheric variables across the Northern Hemisphere (NH) during winter and summer. The main finding is that applying the SPP scheme enhances ensemble spread for both sea ice concentration (SIC) and sea ice thickness (SIT) by approximately 10% relative to unperturbed forecasts. However, this change in the mean can, in some cases, increase bias in the model. Beyond spread, SPP induces systematic changes to the mean state, most notably a general reduction in SIC and a redistribution of ice from the central Arctic toward the ice edge during the winter. These sea ice changes seem to also affect the overlying Arctic atmosphere, producing temperature changes of roughly 0.5°C at 850 hPa, particularly around the Greenland, Barents, and Kara seas. The inclusion of the SPP scheme also results in a substantial increase in the anomaly correlation coefficient (ACC) for 500 hPa geopotential heights (Z500) over the sub-polar North Atlantic during winter, suggesting better seasonal forecast skill there. The authors hypothesize that this change in ACC arises since the SPP scheme increases sea ice at the ice edge during winter, which increases the meridional temperature gradient and also decreases warm biases near Greenland and the nearby seas. Thus, the correction comes from compensating biases in the model. The authors also demonstrate that sea ice perturbations are considerably more effective at generating ensemble spread in uncoupled ocean forecasts than in fully coupled simulations, a finding with important implications for data assimilation applications.
Overall Opinion
The study presents in an interesting way of testing ways of correcting the underdispersive nature of sea ice forecasts in numerical forecast models, in this case the IFS. The nature of how changing SIC and SIT impact the overlying atmosphere and aspects of the seasonal forecast is quite interesting and very timely, particularly as the community continues to debate the potential link of Arctic sea ice variability on mid-latitude NH weather variability. The manuscript is written very well, and the explanations are well constructed. The authors did an excellent job as well in presenting hypotheses and then testing them with different experiments, statistical tests, and toy models. These were well-appreciated and should be done more in scientific papers. Well done!
Overall, the paper is scientifically sound and well-explained. There are some constructive elements of the paper that I think should be addressed to tighten the message of the paper and ultimately improve the presentation and reception of the work. As such, I am recommending major revisions before the manuscript is suitable for publication.
Major Revisions
1. Minimize/Remove Analyses for the Southern Hemisphere (SH). The main focus of the paper is on the Arctic sea ice and the effects changes in the sea ice forecasts have on NH winter and summer atmospheric variables and seasonal forecasts. The authors also included some similar analyses for the SH sea ice in Supplemental Material and refer to it throughout the paper. While maybe interesting, I found it distracting and overwhelming when reading the paper. The paper has 16 main text figures and 11 Supplementary Figures - that's a lot to process as a reader (along with the text). As it is, there is so much to say about the NH to fill this paper that the SH results are not really important. Furthermore, while there appears to be some similarities in the application of the SPP application for SIC in the SH, the authors have not gone further to explore, for example, seasonal forecast impacts in the Southern Hemisphere or other implications for changing the SIC/SIT forecasts. My suggestion is that the authors remove some/all of the figures on SH SIC/SIT and reserve those comments for the discussion and conclusions.
2. Z500 Forecast "Skill" Result. The change in ACC for Z500 in the NH winter is arguably one of the top results in the paper. However, there are a couple of things that I think are important to mention. First, ACC is not "forecast skill" - it is just correlation and measures spatial phase agreement but says nothing about model performance on the field. For example, ACC scores can be high but amplitudes could be off by orders of magnitude in the forecast. Furthermore, the sample size for evaluating the correlation is of concern. So, I would be careful about using the term forecast "skill" in relation to ACC.
Next, while the number of members and simulations may make for a large total sample size, the effective degrees of freedom for statistics is an issue because of temporal and spatial correlations in Z500. I'm not familiar with the three-way correlation test used by the authors for checking statistical significance, so maybe serial correlation is already accounted for. For example, improving Z500 forecasts over the North Atlantic will also also change downstream heights over Northern Europe and into Eurasia, so those Z500 changes are not independent for testing (Figure 16). The authors should address this and redo their statistical significance to account for the serial correlation.
3. Hypothesis for Z500 Skill Changes. The authors offer a hypothesis as to why wintertime Z500 forecasts may improve with the SPP scheme – i.e., that cooling in the model over Greenland and the Kara and Barents Seas strengthens the meridional temperature gradient and thus improves Z500 forecasts. They test this mechanism with a bootstrapping test (Figure 16c), but the results are less than convincing. Setting aside the significance issue (see previous comment), the correlation is weakly negative (-0.28), which really means that the change in sea ice edge explains ~8% of the variability in Z500 in the subpolar North Atlantic. That leaves 92% of the variance unexplained. So, while the hypothesis is interesting, there is a lot of unexplained variance left. I think this needs to be acknowledged and also related to whether or not the improvements in Z500 skill really are meaningful via their proposed mechanism.
4. Summer Results are De-emphasized throughout the Paper. The authors conduct their SPP experiments on winter and summer conditions in the Arctic, but the authors focus very highly on the winter results. That is somewhat understandable, but I think there are a few more points that could be made about the summertime results. While the impacts of the SIC/SIT changes in summer have minimal effects on the Arctic summertime atmosphere because of how close the atmosphere is to freezing, the authors do not offer alternative mechanisms for why there are significant SIC changes in the summer (see Figure 2). Moreover, the conceptual model in Section 5.2 is less applicable for the summer, but the authors do not offer an alternative mechanism for explaining the summertime results. The authors should conceptualize a model for explaining their summertime results better, as we have observed that Arctic extremes in the summer also impact mid-latitude NH weather patterns.
Minor Revisions
1. Data Availability. The data statement is incomplete as presented, as it does not show a repository for the authors' experiments and/or code for analyses.