设为首页 加入收藏 登录旧版
多元线性回归方法对北京地区PM2.5预报的改进应用
Application of Ensemble Forecast and Linear Regression Method in Improving PM2.5 Forecast over Beijing Area
投稿时间:2018-01-22  修订日期:2018-05-08
DOI:10.19316/j.issn.1002-6002.2019.02.06
中文关键词:  PM2.5  集合预报  多元线性回归方法  空气污染红色预警
英文关键词:PM2.5  ensemble forecast  multiple linear regression  red alert of air heavy pollution
基金项目:国家重点研发计划(2016YFC0208803);北京市科委重大专项(D17110900150000)
作者单位
潘锦秀 北京市环境保护监测中心, 大气颗粒物监测技术北京市重点实验室, 北京 100048 
晏平仲 中国科学院大气物理研究所, 大气边界层和大气化学国家重点实验室, 北京 100029 
孙峰 北京市环境保护监测中心, 大气颗粒物监测技术北京市重点实验室, 北京 100048 
李云婷 北京市环境保护监测中心, 大气颗粒物监测技术北京市重点实验室, 北京 100048 
刘保献 北京市环境保护监测中心, 大气颗粒物监测技术北京市重点实验室, 北京 100048 
王占山 北京市环境保护监测中心, 大气颗粒物监测技术北京市重点实验室, 北京 100048 
董瑞 北京市环境保护监测中心, 大气颗粒物监测技术北京市重点实验室, 北京 100048 
摘要点击次数: 30
全文下载次数: 0
中文摘要:
      利用多元线性回归方法(REG)将多模式空气质量预报系统中3个模式(CMAQ、CAMx和NAQPMS)对北京市2016年PM2.5的预报结果和观测数据进行集合,并对集合结果进行评估。结果表明:①不同模式的预报结果不尽相同,均能够反映2016年北京地区PM2.5随时间的变化趋势,CMAQ、CAMx和NAQPMS相关系数为0.6~0.9,标准化平均偏差为-0.6~0.6。3个模式对重污染峰值预报都存在偏差,NAQPMS预报偏差低于其他模式;②基于多元线性回归集成预报模型能显著提高日均PM2.5预报的准确率,能较好地改进不同季节模式整体高估或者低估的系统性偏差现象,春季国控平均偏差由-23 μg/m3改善至-2.3 μg/m3,冬季平均偏差降低近20 μg/m3;③利用多元线性回归方法对2016年红色预警期间小时PM2.5订正结果显示,相关系数提高了0.13,均方根误差降低了20~30 μg/m3,并且对峰值浓度有较好的调整,预报峰值更为接近实况峰值,特别是对北部地区的改进效果较为明显,反映了实际观测数据对空气质量数值模式预报修正的研究意义和可行性。
英文摘要:
      In order to improve the forecast performance,air quality monitoring PM2.5 data of Beijing and the simulated data from air quality operational forecasting system(consisting of CMAQ,CAMx and NAQPMS numerical models) of 2016 were collected using multiple linear regression method(REG)and the effectiveness of the improvements were evaluated.The results indicated that different models have their own forecast skills and can reflect the variation of PM2.5 in 2016 with the correlation coefficient of 0.6-0.9 and the standardized mean deviation of -0.6-0.6. However,the three models all had deviation for heavy pollution forecast and the deviation of NAQPMS was lower than the other two. Besides,REG brougth significant improvement of the average daily PM2.5forecast.It improved the overall overestimation or underestimation in different seasons. The average deviation of PM2.5 was improved from -23 to -2.3 μg/m3 in spring and reduced 20 μg/m3 in winter. And,REG was used to calibrated hourly PM2.5value during red alert period of air heavy pollution. The results showed that the correlation coefficient was increased by 0.13 and that the root mean square error was decreased by nearly 20-30 μg/m3. Moreover,the peak concentration was more closer to the reality during red alert period. And,the improvement effect was more obvious in the northern region which reflects the significance and feasibility of the method.
查看全文  查看/发表评论  下载PDF阅读器