The Unite States is a country with thousands of lakes and lake has very different surface characteristics and atmospheric boundary layer structures from land. Thus, accurate simulating surface temperature and the atmospheric boundary layer depth is extremely important for air quality forecasting over lakes and surrounding regions. Currently the NOAA Environmental Modeling System (NEMS) Non-hydrostatic Multi-scale Model on the Arakawa staggered B-grid (NMM-B) is coupled with the Community Multi-scale Air Quality (CMAQ) modeling system for operational air quality forecasting. However, the surface ozone is often over-predicted by the coupling system over lakes such as Lake Michigan and surrounding regions. In this study, we examine several important meteorological fields that may cause surface ozone over-predictions, including surface temperature, winds and the planetary boundary layer height (PBLH). The simulated meteorological inputs are evaluated with surface observational data and satellite measurements. The simulated chemical species (e.g., surface ozone and PM2.5) are compared with AIRNow observational data. Several sensitivities are conducted to examine the impact of different PBL schemes and shallow-convection scheme on surface ozone predictions. Finally, how to improve air quality forecasting over lake regions is discussed.