Taking Shenzhen city as an example, the statistical and physical relationship between density of pollutants and various atmospheric parameters are analyzed in detail, and a space-partitioned city air pollution potential prediction scheme is established based on it. The scheme considers quantitatively more than ten factors at surface and planetary boundary layer (PBL), especially the effects of anisotropy of geographical environment, and treats wind direction as an independent impact factor. While the scheme treats the prediction equation respectively for different pollutants according to their differences in dilute properties, it considers as well the possible differences in dilute properties at different district of the city under the same atmospheric condition, treating predictions respectively for different district. Finally, the temporally and spatially high resolution predictions for the atmospheric factors are made with a high resolution numerical model, and further the space-partitioned and time-variational city pollution potential predictions are made. The scheme is objective and quantitative, and with clear physical meaning, so it is suitable to be used in making high resolution air pollution predictions.