Statistical Prediction of Winter Haze Days in the North China Plain Using the Generalized Additive Model

Abstract

AbstractWinter (December?February) haze days in the North China Plain (WHDNCP) have recently dramatically increased. In addition to human activities, climate change and variability also contributed to the severe situation and supported the possibility of seasonal predictions. In this study, using the generalized additive model (GAM), the sea surface temperature around the Alaska Gulf and sea ice area of the Beaufort Sea were selected as the predictors to establish a statistical prediction model (SPM). The difference between the current and previous year of WHDNCP (WDY) was predicted first and was then added to the observation of the previous year to obtain the final predicted WHDNCP. For WDY prediction, the root-mean-square error of the SPM using GAM was 3.01 days. In addition to the annual variation, the tropospheric biennial oscillation features and the dramatically increasing trend after 2010 were both captured successfully. Furthermore, for the final predicted WHDNCP anomalies, the long-term trend and turning points were simulated well, and the percentage of the same mathematical sign was 91.7%. Independent prediction tests were performed for 2014 and 2015, and the forecast bias was 0.86 and 0.19 days, respectively. To assess the predictive ability, recycling independent tests (including real-time hindcasts for the period 2005?15) were also applied, and the percentage of the same sign was 100%.