A significant amount of radioactive material was accidentally discharged into the atmosphere from the Fukushima Dai-ichi Nuclear Power Plant from 12 March 2011, which produced high contaminated areas over a wide region in Japan. In conducting regional-scale atmospheric dispersion simulations, the computer-based nuclear emergency response system WSPEEDI-II developed by Japan Atomic Energy Agency was used. Because this system is driven by a meso-scale meteorological (MM) model, it is difficult to reproduce small-scale wind fluctuations due to the effects of local terrain variability and buildings within a nuclear facility that are not explicitly represented in MM models. In this study, we propose a computational approach to couple an LES-based CFD model with a MM model for detailed simulations of turbulent winds with buoyancy effects under real meteorological conditions using turbulent inflow technique. Compared to the simple measurement data, especially, the 10 min averaged wind directions of the LES differ by more than 30 degrees during some period of time. However, distribution patterns of wind speeds, directions, and potential temperature are similar to the MM data. This implies that our coupling technique has potential performance to provide detailed data on contaminated area in the nuclear accidents.