Ensemble and enhanced PM10 concentration forecast model based on stepwise regression and wavelet analysis

Abstract

An ensemble and enhanced PM10 (particulate matter with a diameter less than 10 μm) concentration forecast model was established in eastern China based on data from 2005 to 2009. The enhanced model consists of a single stepwise regression forecast model and a combined forecast model based on wavelet decomposition and stepwise regression. Six individual forecast results were obtained with a combined model that can predict PM10 concentrations at multiple scales. By decomposing variables into detailed and approximated components in six scales and with the application of stepwise regression, the best-fitted forecast models were established in each component of the different scales. Then, the predicted results of the detail and approximation components were reconstructed in each scale as the enhanced prediction. A regional model was established for eastern China. The accuracy rate of each forecasted result by the regional model was calculated using testing data from 2010 based on the needs of operational forecasting. Precision evaluations were also performed. A comparatively higher accuracy was obtained by the combined model. The advantage of predicting the PM10 concentration with the combined model had wide spatial and temporal suitability. An enhanced forecast model was established for each city of eastern China with improvements, where all the predicted results in each city were evaluated by the accuracy rate and precision validation. In each city, the best-fitted model with the highest precision was selected and combined in an ensemble. The ensemble and enhanced forecast model had a significant improvement in accuracy rate and the highest precision of PM10 concentration forecasting in eastern China.