科研之友
微信
新浪微博
Facebook
分享链接
摘要
Empirical regression models for next-day forecasting of PM2.5 and O-3 air pollution concentrations have been developed and evaluated for three large Chinese cities, Beijing, Nanjing and Guangzhou. The forecast models are empirical nonlinear regression models designed for use in an automated data retrieval and forecasting platform. The PM2.5 model includes an upwind air quality variable, PM24, to account for regional transport of PM2.5 and a persistence variable (previous day PM2.5 concentration). The models were evaluated in the hindcast mode with a two-year air quality and meteorological data set using a leave-one-month-out cross validation method, and in the forecast mode with a one-year air quality and forecasted weather dataset that included forecasted air trajectories. The PM2.5 models performed well in the hindcast mode, with coefficient of determination (R-2) values of 0.54, 0.65 and 0.64, and normalized mean error (NME) values of 0.40, 0.26 and 0.23 respectively, for the three cities. The O-3 models also performed well in the hindcast mode, with R-2 values of 0.75, 0.55 and 0.73, and NME values of 0.29, 0.26 and 0.24 in the three cities. The O-3 models performed better in summertime than in winter in Beijing and Guangzhou, and captured the O-3 variations well all the year round in Nanjing. The overall forecast performance of the PM2.5 and O-3 models during the test year varied from fair to good, depending on location. The forecasts were somewhat degraded compared with hindcasts from the same year, depending on the accuracy of the forecasted meteorological input data. For the O-3 models, the model forecast accuracy was strongly dependent on the maximum temperature forecasts. For the critical forecasts, involving air quality standard exceedences, the PM2.5 model forecasts were fair to good, and the O-3 model forecasts were poor to fair.
- 出版日期2016-12
- 单位清华大学
