空气质量数值预报优化方法研究

基于源清单“Nudging”修正方法和XGBoost算法对徐州市2016年12月13个监测站点的PM_2．5、PM₁₀、O₃、SO₂、NO₂、CO等6种污染物浓度预报值进行修正,并分析了修正前后模式预报改善效果.在源清单“Nudging”修正部分,本文结合IDW空间插值算法对SO₂、NO₂、CO等3种污染物浓度预报值进行修正,与修正前后模拟结果相比,采用同化源模拟的预报浓度值与观测值的相关系数提高了0.06~0.27不等,平均绝对误差和均方根误差减少的幅度较为明显,平均相对偏差（MFB）和平均相对误差（MFE）均在理想水平范围内,NO₂修正效果最好,其次是SO₂和CO.基于XGBoost算法的统计修正部分,本文结合WRF气象预报要素建立统计回归模型,对6种污染物进行统计修正,经滚动修正之后,预报偏低或偏高现象得到很大的改善,除了SO₂之外,相关系数均提高到0.6~0.7左右,各项误差统计指标改进幅度非常明显.总体而言,本文采用的两种修正方法对中小尺度空气质量数值预报改进效果非常明显,反映了此优化方案的可行性和科学性.

A revised approach to air quality forecast based on Models-3/CMAQ

In this study, the forecast values of hourly PM_2.5、PM₁₀、O₃、SO₂、NO₂、CO concentrations at 13 environmental monitoring stations in Xuzhou city during December 2016 were corrected using nudging scheme and XGBoost algorithm, and improvement model prediction before and after correction were analyzed. A method combining nudging scheme and IDW interpolation algorithm was adopted by modifying the forecast values of SO₂、NO₂、CO concentrations, results showed that the correlation coefficient between the predicted concentration and the observation simulated by the assimilation source increased by 0.06~0.27, and the mean absolute error and the root mean square error decreased obviously, the average relative deviation (MFB) and average relative error (MFE) were within the ideal range, had best effect on NO₂ followed by SO₂ and CO. The part of statistical revision which based on XGBoost algorithm, by introducing WRF meteorological forecast elements established a statistical regression model, which could be used for modifying the forecast values of PM_2.5、PM₁₀、O₃、SO₂、NO₂、CO concentrations. Results showed that lower or higher than normal conditions were greatly improved,with the exception of SO₂, the correlation coefficient increased to about 0.6~0.7, the reduction of the error of statistical indicators was very obvious.