An exhaustive elucidation of the heat transfer performances and mechanisms of supercritical LNG under water-bath heating conditions is paramount for optimizing heat transfer and designing related heat exchangers. This study employs experiments and numerical simulations to explore the heat transfer mechanisms and prediction methods of supercritical LNG flowing horizontally under buoyancy. The experimental investigation was executed within a small SCV using the N2-Ar mixture as a substitute medium for safety considerations. Concurrently, the accuracy of the numerical model employed is rigorously authenticated through a comparative analysis with the empirical findings obtained from the experimental endeavor. The investigation delves into the impacts of operating conditions, structural parameters, and medium composition on heat transfer through numerical simulations. The influence mechanism of buoyancy on heat transfer is meticulously explicated by examining the distribution of flow fields and thermophysical properties within the boundary layer. The introduction of a novel criterion, Bufull, derived from the analysis of wall shear stress variation caused by buoyancy, facilitates a quantitative assessment of the impact of buoyancy on heat transfer. Additionally, a simplified buoyancy criterion, FBu, is introduced, dependent solely on fundamental operating parameters and tailored for engineering applications, with a determined critical value of 1/ FBu = 60.3. Based on the dramatically variable thermophysical and transport properties, that occurs near the critical point, as well as buoyancy correction, a correlation is formulated for predicting mixed convection heat transfer, exhibiting an MAE of merely 3.04% and accurately predicting 81.28% of the data within a ± 5% margin of error, surpassing existing correlations. Finally, the proposed heat transfer calculation method is validated using five groups of operational data from two in-service SCVs. The maximum absolute discrepancy between the computed outlet temperature and the monitored value is 1.95K. The findings of this paper can support practical applications.