Abstract. During boreal summer, prevailing southerlies traverse the sharp sea surface temperature (SST) front in the northwest Pacific (NWP) Ocean, creating a stable air–sea interface characterized by surface air temperature (SAT) higher than SST, which promotes the frequent occurrence of advection fog. However, long-term shipborne observations reveal that during episodes of advection fog, SAT usually decreases below SST, with a peak relative frequency (∼ 34.5 %) in all fog observations before sunrise and a minimum relative frequency (∼ 18.8 %) before sunset. From a Lagrangian perspective, this study employs a turbulence-closure large-eddy simulation (LES) model to trace a fog column across the SST front and investigates how SAT drops below SST during an advection fog event. The LES model, incorporating constant solar radiation, successfully simulates the evolution of advection fog and the negative difference between SAT and SST. Simulation results show that once the near-surface air condenses, thermal turbulence is generated by strong longwave radiation cooling (LWC) at the fog top. The influence of LWC on the fog layer surpasses the cooling effect of the near-surface mechanical turbulence ∼ 2 h after the fog formation while the fog column is still positioned over the SST front. When the fog column arrives at the cold flank of the SST front, the top-down-developing mixed layer induced by the LWC reaches the surface, causing SAT to drop below SST. The LES model with diurnal solar radiation successfully simulates the observational diurnal variation in SAT and SST (SAT-SST) during the fog event, suggesting that the model captures the essential processes responsible for negative SAT-SST. This study highlights the significance of fog-top cooling and its associated thermal turbulence in the evolution of advection fog. Given the challenges faced by numerical weather prediction models in forecasting sea fog, our findings suggest that observations of negative SAT-SST during advection fog episodes present an opportunity to enhance the performance of these models in simulating the thermal turbulence induced by the LWC at the fog top.