ECMWF和T639数值预报产品在兰州市空气质量预报应用中的对比

A comparative study of air quality forecast in Lanzhou City derived from LS-SVM models built with ECMWF and T639

利用2013-2015年兰州市空气污染逐日监测资料,分析了兰州市6种主要空气污染物PM_(10)、PM_(2.5)、NO_2、SO_2、CO和O_3的污染特征;以2014年欧洲中期天气预报中心(ECMWF)资料与T639气象要素预报产品,结合兰州市同期污染物质量浓度监测数据,分别建立了基于最小二乘法支持向量机(LS-SVM)的6种主要空气污染物未来2d的日均质量浓度预报模型;将ECMWF和T639中2015年2月1日-10月31日的气象要素与同期污染物质量浓度监测数据分别输入各模型进行试预报检验.结果表明,以ECMWF建立的预报模型对未来2 d的PM_(10)、PM_(2.5)、NO_2、SO_2和CO的日均质量浓度的预报效果优于T639,而T639对预报O_3有一定优势.用ECMWF建立的预报模型对未来24 h的空气质量指数等级和首要污染物的预报成功率为86.14%,48 h的为82.33%;T639对应的未来24 h预报成功率为83.52%,48 h的为74.43%.两种数值预报产品均可应用于基于LS-SVM预报模型的空气质量预报,其中使用ECMWF的预报产品的释用预报效果整体上更好.

Daily monitor records of air pollutant concentrations in Lanzhou City during 2013-2015 were used to analyze the pollution features of 6 major air pollutants： PM,0, PM25, NO2, SO2, CO and 03. Then 2 least square support vector machine （LS-SVM） models were built for predicting daily average concentra- tions of the 6 major pollutants for the next 24 h, 2 LS-SVM models for next 24 h to 48 h, each of them being built with either T639 and pollution dataset of 2014 or European centre for medium-range weather forecasts （ECMWF） and pollution dataset of 2014. All models underwent a forecasting trial using test samples as inputs which were derived from pollution records and ECMWF/T639 dataset consisting of processed meteorological variables from 1 Feb 2015 to 31 Oct 20 l 5. The results are as follows： The 24 h and 48 h models built with ECMWF had a better performance of predicting the daily average concentra- tions of PMI0, PM25, NO2, SO2 and CO, while models built with T639 were better at predicting 03. The corrected prediction rate for both the primary pollutant and the grade of air quality index with ECMWF model for next 24 h was 86.14%, and 82.33% for next 24 h to 48h; the corrected prediction rate with T639 model for next 24 h was 83.52%, and 74.43% for next 24 h to 48 h. T639 and ECMWF both could be used in building air forecasting models based on LS-SVM, and ECMWF showed a better overall performance.