Abstract: A model is developed to prognosticate the heavy fog. For creating a objective forecasting-heavy-fog model, the circulation background information must be transformed into numeral format. Ignored the fog classification and corresponding synoptic patterns, it uses various physical variations to describe the circulation background on the formation mechanism of the heavy fog which is valuable in not only easily quantifying, but also realizing the forecast's objectivity and automatity. The algorithm of the variations conversion and their combinations presented. The concept and its formula of the directional transporting are very effective to dynamical quantifying of relevant background field. In the fog-forecasting model, based on BP neural networks technique, the forecast factors and forecast index and better describe the whole background, in which developed the heavy fog, it consists of numerous information and has effective application. The design of the neural networks and the parameterization scheme would be a good reference to the construction of the other forecast systems。