资源环境科技发展态势分析平台

Recent innovations in remote sensing technologies and retrievals offer the potential for predicting ultrafine particle (UFP) concentrations from space. However, the use of satellite observations to provide predictions of near-surface UFP concentrations is limited by the high frequency of incomplete predictor values (due to missing observations), the lack of models that account for the temporal dependence of UFP concentrations, and the large uncertainty in satellite retrievals. Herein we present a novel statistical approach designed to address the first two limitations. We estimate UFP concentrations by using lagged estimates of UFP and concurrent satellite-based observations of aerosol optical properties, ultraviolet solar radiation flux, and trace gas concentrations, wherein an expectation maximization algorithm is used to impute missing values in the satellite observations. The resulting model of UFP (derived by using an autoregressive moving average model with exogenous inputs) explains 51 and 28% of the day-to-day variability in concentrations at two sites in eastern North America.