Abstract:In recent years,severe haze pollution has been damaging human health,traffic security,the ecosystem and social economy in eastern China.In addition to the haze forecast within 1 week,seasonal haze prediction provides scientific support for longer periods to the decisions of emission reduction.In this study,taking the annual increment as the predictand,monthly prediction models were trained for the Beijing-Tianjin-Hebei and Yangtze Delta regions.The performances of the built models were similar,with 2 days of root-mean-square error and a>80% simulation rate of the anomalies' mathematical sign.In the real-time seasonal prediction for Beijing-Tianjin-Hebei haze days in the winter of 2016,the results with respect to the climate mean(the previous year) were completely (mostly) accurate.During the winter of 2017,the predicted biases for the December and January haze days in the Yangtze River Delta were very small,and the bias of February was nearly 2 days.