To accurately predict the future trajectory of Arctic sea ice in a warming climate, an explicit understanding of ice-ocean-atmosphere interaction processes is essential. While previous melting simulation studies have focused on the process where atmospheric radiative forcing dominates surface melting during the early melt season, a distinct temporal decoupling occurs in late summer and autumn; paradoxically, the sea ice melt rate peaks even as the atmospheric heat source is effectively blocked, indicating that atmospheric heat alone has significant limitations in explaining this large-scale melting. In this study, we analyzed the spatiotemporal evolution of the sea ice edge and subsurface ocean heat using time-series and high-resolution observational data from the Laptev Sea slope, presenting analytical results that demonstrate the mechanism driving autumn sea ice melt transitions from the atmosphere to the ocean. According to our newly identified 'ocean-driven autumn feedback loop' conceptual model, ocean heat accumulated during the summer is actively subducted in the form of mesoscale eddies through baroclinic instability occurring at the sea ice edge in autumn. Specifically, we detailed the dynamical process where a continuous upward ocean heat flux from the subsurface heat reservoir, protected from atmospheric cooling, induces basal melting of the sea ice. To verify this, we conducted a simultaneous spatiotemporal analysis of net surface radiation, water temperature at a depth of 40m, and sea ice trends from June to October 2020, comparing them with observational data. The analysis confirmed that strong vertical mixing caused by baroclinic instability acts as the core mechanism for the phenomenon where sea ice melt reaches its maximum in September in the absence of atmospheric radiative heating. Consequently, we showed that integrating this ocean-driven feedback into climate models is essential to accurately simulate future Arctic sea ice decline and prolonged melt seasons.

Arctic Sea Ice,Basal melting,Ice-albedo feedback,Baroclinic Instability,Mesoscale eddies