Post-processing corrections can be applied to operational numerical model forecasts of weather variables to reduce forecast model errors. The post-processing technique is based on an analog post-processing correction scheme. In the analog scheme, the system searches for previous model forecasts that are similar to the current forecast, and modifies the current forecast based on errors in such previous predictions. An “abnormal (or abnormal) index” can be generated for each analog as a function of several meteorology variables (i.e., humidity, pressure and temperature) and two experience coefficients. The abnormal index can be used to exclude some errors. Furthermore, dynamic weights based on the “abnormal index” can be applied to the errors. Using the “abnormal index” weights in a post-processing technique may generate a more accurate air quality forecast.