Arctic Amplification (AA) robustly exists in observations and model simulations. AA shows significant seasonality and inter-model spread. Previous studies found that surface type would influence the spatial distribution and seasonality of AA via several processes such as albedo, surface heat fluxes and heat capacity. This study extends the discussion of influence of surface type to inter-model spread in AA from a process-based perspective. The Arctic is divided into four surface types: ice cover, ice retreat, ice free and land. The magnitudes and inter-model spread in AA are calculated from the difference between abrupt-4 × CO₂ and pre-industrial experiments of 17 CMIP6 models. The ice retreat region shows largest seasonal contrast of AA, while the ice cover region exhibits most significant inter-model spread in Arctic warming in January-February-March (JFM).

The result reveals that the seasonal energy transfer mechanism, which means the heat temporarily stored in summer and then released in winter, and the change of effective thermal inertia (ETI) are key factors responsible for the maximum winter warming of AA. The transition from sea ice to open ocean accompanied with increased ETI suppresses summer warming while the winter warming is amplified by releasing the heat stored in summer and accelerating the rate of heating due to small ETI. Therefore, the ice-retreat regions with the largest sea ice loss which implies extraordinary change of ETI produce the most significant seasonal contrast of AA, while the ice-free ocean reveals a weaker and almost uniform seasonal warming. Compared to other regions, the ice cover region shows maximum inter-model spreads in JFM because there is a stronger inter-model spread of oceanic heat transport process while the weaker upward surface turbulent sensible and latent heat fluxes tend to suppress the inter-model spread. The relatively small inter-model spread in summer is caused by the inter-model spread in ice-albedo feedback offsetting the spread in oceanic heat transport process.