南昌市空气质量预报产品检验分析及统计订正试验

Inspection Analysis and Statistical Correction Test of Nanchang Air Quality Forecast Product

利用南昌市环境空气质量监测数据,对比分析了WRF-Chem模式和国家级空气质量预报指导产品对6种污染物浓度的预报效果,并采用时序法、时刻法和标准化法3种训练样本构建方案,利用BP神经网络法对WRF-Chem模式和国家级空气质量预报指导产品6种污染物浓度的预报结果进行订正试验。结果表明:1)WRF-Chem模式预报的6种污染物浓度的预报误差整体比国家级空气质量预报指导产品的预报误差要小,即WRF-Chem模式的预报效果优于国家级空气质量预报指导产品。2)WRF-Chem模式6种污染物浓度预报值与观测值的均方根误差的日变化均呈波动形式,除了O3在10时开始升高到18时达到峰值以外,其余的污染物均是从10时开始下降到16时或18时达到谷值。国家级指导产品6种污染物浓度预报值与观测值的均方根误差日变化则略有不同,除了NO2和O3分别在08时和20时达到谷值以外,其他4种污染物均是在14时达到谷值。3)采用标准化法对CO、SO2、PM10、PM2.5集合订正后的误差比WRF-Chem模式的要小;时刻法、时序法对NO2、SO2、PM2.5集合订正后的误差比WRF-Chem模式的要小,预报效果对单一模式预报结果有一定改进作用。

Based on the air quality monitoring data in Nanchang, the forecasting effect data of WRF Chem model and national air quality forecasting guidance products of the concentration of six pollutants, three kinds of training samples were used to build up the scheme, and the BP neural network method was used to correct the prediction results of six pollutant concentra tions in the WRF Chem model and the national air quality forecasting guidance product. The results showed that： 1） The pre diction error of the six pollutant concentrations predicted by the WRF Chem model was smaller than that of the national air quality forecast guidance product, thus, the prediction effect of the WRF Chem model was considered to be better than that of the national air quality forecast guidance product. 2） The daily variation of the root mean square error of the predicted value of the six pollutants in the WRF Chem model was fluctuating, the pollutants decreased from 10：00 and reached the valley at 16：00 or 18：00, except for 03 , which increased at 10：00 and reached the peak at 18：00. The daily variation of the root mean square error of the predicted values of the six pollutants in the state level guidance products was slightly different. NO2 and 03reached the valley at 08：00 and 20：00, respectively, while the other four pollutants reached the valley at 14：00.3） The coffee tion error of CO, SO2 , PM10, and PM2 ＋ sets by normalization method was smaller than that of the WRF Chem model. The er rot ratio of NO2 , SO2 , and PM2＋ sets corrected by time method and time series method was smaller that that by WRF Chem model, and the forecast effect was improved on the single model forecast result.