The rapid decline of Arctic sea ice under global warming has led to increasingly heterogeneous ice conditions, with more frequent leads and open water. Although sea ice models (e.g., Icepack, the column physics package of CICE) represent sub-grid heterogeneity in sea ice, most ocean-ice coupled models still neglect sub-grid sea surface temperature (SST) heterogeneity between open-water and ice-covered fractions within a grid cell. In this study, we developed an emergent sub-grid SST parameterization for an ocean-ice coupled model to capture SST variations within individual grid cells. This approach distinguishes open-water SST from under-ice SST, explicitly represents spatial heterogeneity, and enables separate calculations of ice–ocean energy exchange processes, including basal and lateral melting. We further incorporated internal grid-scale exchange by introducing a parameterization of lateral mixing as a function of floe size and ice–ocean velocity differences. Idealized melt experiments show that the new scheme prolongs ice persistence relative to the baseline grid-averaged formulation. Under the same forcing conditions, larger floes also melt more slowly than smaller floes, consistent with results from some three-dimensional experiments. We further applied this scheme in sensitivity experiments for the Arctic Ocean during the melt season. The new parameterization improves agreement with satellite-derived ice concentration and reduces the bias by 10% relative to the baseline model.