Abstract: Aiming at the less models available and poor prediction performance of current haze forecasting, combining statistical forecast and numerical prediction, a new haze prediction model was proposed based on hybrid stepwise multivariable and probability regression. Multivariable stepwise regression was used to control physical factors which influenced depending variable and the equation for atmosphere visibility wasgenerated. A haze prediction model based on binary variables was established using probability regression combined with factors like visibility and relative humidity. Experimental results have proved that compared with the existing CUACE, which is one of the main digital haze forecasting system, the accuracy using the hybrid regression forecasting model suggested is obviously higher.

Key words: multivariable stepwise, probability regression, forecasting model, visibility, haze forecast