Models’ ability to reproduce the Arctic summer intraseasonal mode and its associated sea ice responses

Internal atmospheric variability profoundly influences the Arctic summer climate, yet its critical subseasonal-to-seasonal (S2S) dynamics remain underexplored. Using the observational baseline of a 21-day westward-traveling intraseasonal Rossby wave, we evaluate its representation across physical and hybrid prediction paradigms, including CESM2, seven operational S2S models, and NeuralGCM. Models exhibit distinct structural limitations: physics-based models maintain a realistic ~21-day periodicity but fail to reproduce the wave’s spatial stagnation over the Canadian Arctic Archipelago, whereas NeuralGCM captures this localized phase structure but artificially accelerates the periodicity by ~5 days. Furthermore, predictive skill is severely limited, peaking at 11 days in UKMO/SEAS5 before degrading rapidly below useful thresholds by days 7–9 in the remaining models. This constrained atmospheric predictability triggers substantial downstream biases in sea-ice simulations, highlighting the broader cryospheric consequences of errors in atmospheric circulation. As for this bias source, we find that models fundamentally distort the physical drivers of ice loss, attributing 84.0% of the response to dynamics and only 16.0% to thermodynamics, a stark deviation from the observationally constrained baseline of 63.7% versus 36.3%. We trace this dynamic–thermodynamic imbalance to an anomalous landward shift in the high-pressure center. This displacement misdirects surface winds, fails to force the anticyclonic Beaufort Gyre, and induces a localized radiative deficit, thereby artificially inflating dynamic ice accumulation while suppressing thermodynamic melt.