基于CMAQ模式和自适应偏最小二乘回归法的中国地区PM2.5浓度动力-统计预报方法研究

采用CMAQ模式和自适应偏最小二乘回归法相结合的动力-统计预报方法,对2014年1—12月全国252个环境监测站的PM_(2.5)浓度逐时预报值进行了滚动订正,分析了订正前后PM_(2.5)浓度的时空变化特征,重点研究该方法在中国不同地区不同季节的适用性.结果表明:CMAQ模式预报的PM_(2.5)浓度年平均和秋冬季季节平均偏差表现为非均匀空间分布特征,即辽宁、山东部分地区、川渝地区及华中、华东、华南大部分地区预报偏高,京津冀和西部大部分地区预报偏低;订正后PM_(2.5)浓度与实测值的空间分布较一致,上述偏高和偏低地区的PM_(2.5)浓度预报误差显著减小;秋冬季PM_(2.5)浓度预报和订正偏差均大于年平均值.全国区域平均PM_(2.5)浓度实测值存在明显的季节变化特征,1—3月和11—12月较大,其他月份较小;PM_(2.5)浓度预报误差较大,多数时刻预报偏低,尤其是1—3月和11—12月偏低较明显;订正后PM_(2.5)浓度与实测值较接近,而且时间变化趋势较一致,秋冬季PM_(2.5)浓度预报和订正偏差亦明显大于春夏季.秋冬季4个重点污染区域中,京津冀地区PM_(2.5)实测浓度的区域平均值较大,川渝地区次之,长三角和珠三角地区较小;珠三角地区PM_(2.5)浓度预报和订正效果较好,川渝和长三角地区次之,京津冀地区相对较差;经滚动订正后,全年和秋冬季时段PM_(2.5)浓度订正值与实测值的相关系数均显著增加,误差显著减小,尤其是秋冬季订正效果较好.川渝地区的订正改进幅度最大,长三角和京津冀地区次之,珠三角地区较小.本文方法均适用于非污染日和污染日全国范围的PM_(2.5)预报浓度订正,两种天气过程PM_(2.5)浓度的订正效果均较好;该方法对于改进京津冀地区污染日的PM_(2.5)浓度预报更有效,其他3个地区非污染日的订正改进效果优于污染日.本文研究结果可为改进空气质量预报、重霾污染天气预警和防治提供新技术途径和科学依据.

Dynamical-statistical forecasting of PM2.5 concentration based on CMAQ model and adapting partial least square regression method in China

In this study, hourly PM_2.5 concentrations at 252 environmental monitoring stations in China during January-December 2014 forecasted by the real-time running Fifth-Generation Penn State/NCAR Mesoscale Model (MM5)-Community Multiscale Air Quality (CMAQ) model system are corrected using the dynamical-statistical method based on CMAQ model and adapting partial least square regression technique. Temporal and spatial variations of PM_2.5 concentrations before and after correction are analyzed with a focus on the applicability of the dynamical-statistical method in different areas and seasons in China. It is shown that the spatial distributions of both annual and seasonal (for autumn and winter) averages of PM_2.5 concentrations forecasted by the MM5-CMAQ model system are inhomogeneous. Forecast PM_2.5 concentrations are larger than observations in parts of Liaoning and Shandong provinces, Sichuan and Chongqing provinces and most areas of Central China, East China and South China. Forecast values are smaller in Beijing-Tianjin-Hebei region and in most areas of West China. After correction, the spatial distributions of forecast PM_2.5 concentrations are in good consistence with observations, and forecast errors in the above areas decrease significantly. Forecast and corrected deviations of seasonally averaged PM_2.5 concentrations in autumn and winter in most areas of China are larger than the annual averaged values. There is an obviously seasonal variation of observed PM_2.5 concentrations, with higher values in Jan., Feb., Mar., Nov. and Dec. Forecast errors are larger, with prediction values less than observations in most of the time. Corrected PM_2.5 concentrations and its temporal variation are close to observations. Forecast and corrected deviations in autumn and winter are larger than those in spring and summer. During autumn and winter, among the four seriously polluted regions in China, Beijing-Tianjin-Hebei region shows the highest observed PM_2.5 concentrations, followed by Sichuan and Chongqing provinces, Yangtze River Delta and Pearl River Delta Regions. Pearl River Delta shows the best forecast and correction effects, followed by Sichuan and Chongqing provinces, Yangtze River Delta Region, and Beijing-Tianjin-Hebei region. Correlation coefficients between corrected PM_2.5 concentrations and observations increase remarkably, and forecast errors decrease obviously, especially for autumn and winter. Error decrease ratios after correction are the largest in Sichuan and Chongqing provinces, followed by Yangtze River Delta Region, Beijing-Tianjin-Hebei region and Pearl River Delta Region. The method presented in this paper can be applicable to the correction of PM_2.5 forecast in both polluted and clean days in China. The correction is more effective during polluted processes in the Beijing-Tianjin-Hebei region and correction effects are better during clean processes than on polluted days in other three regions. Results of this study will provide new technique and scientific basis for improving air quality forecasting, early warning and prevention of heavy haze weather.