几种空气质量预报方法对冬季预报效果的评估与对比

The Evaluation and Comparison of Several Methods for Air Quality Forecasting in Winter

基于2016年冬季泰州市环境空气质量自动监测数据,定量评估NAQPMS模式、CMAQ模式和人工订正对污染物质量浓度和空气质量等级的预报效果。结果表明,模式预报和人工订正对各污染物预报的相关系数由高到低排列为PM_(2.5)、PM_(10)、NO_2、SO_2、O_3-8h,颗粒物预报效果最好。除O_3-8h外,NAQPMS对各项污染物预报的相关系数R为0.47～0.82,CMAQ为0.75～0.81,人工订正为0.43～0.78,3种预报方式均能准确反映污染物浓度的变化趋势;模式预报、人工订正对O_3-8h预报相关系数均<0.4。在发生颗粒物污染过程时,人工订正结果相对更为准确。NAQPMS、CMAQ和人工订正对空气质量等级24 h预报准确率分别为38.9%、41.1%和35.6%,NAQPMS对优类别的预判准确率较高,CMAQ、人工订正对良类别的预判准确率较高。对比不同时效的预报效果,24 h预报时效的准确率高于48和72 h。提出,城市空气质量预报可采用集合预报方式,综合1～2种运行较稳定的主流预报模式预报结果,预报员对模式模拟结果进行人工修订,提高预报准确率。

The forecasting effects of the pollutants concentration and the air quality grade by NAQPMS model, CMAQ model and artificial correction were quantitative evaluated, respectively, based on the automatic monitoring data of environmental air quality in the winter of 2016 in Taizhou. The results showed that the correlation coefficient of forecasting for each pollutant order from high to low were ： PM2.5 and PM10, SO2 , NO2 , 03 - 8h, whether if by the model prediction or by artificial correction. The forecast effect of particulate matter was the best. Except for 03 -8h, the correlation coefficient between the predicted concentrations and the measured concentrations of pollutants was in 0.47 ~ 0.82 by NAQPMS, while the correlation coefficient was in 0.75 ~ 0.81 by CMAQ and the correlation coefficient was in 0.43 ~ 0.78 by artificial correction. The variation trend of pollutants concentration could be well reflected through all of the three prediction methods. The correlation coefficient for 03 -8h were less than 0.4 by model prediction and by artificial correction. When the particulate pollutions happened, the artificial correction results were more accurate. The 24 h prediction accuracy of NAQPMS, CMAQ and artificial correction for air quality grade was 38.9% , 41.1% and 35.6%, respectively. NAQPMS performs better on excellent level, CMAQ and artificial correction perfl）rm better on good level. Compared with the forecast effect of different aging time, the accuracy of the forecast of 24 h was higher than 48 and 72 h. The accuracy of city air quality forecast can be improved by ensemble forecast method, as comprehensive 1 ~ 2 kinds of mainstream model prediction results with steady running, and forecasters make artificial revision to the model sinmlation results.