大气科学  2018, Vol. 42 Issue (6): 1273-1285 PDF

1 中国科学院大气物理研究所大气边界层物理与大气化学国家重点试验室, 北京 100029
2 解放军 96631 部队, 北京 102208
3 中国科学院城市环境研究所城市环境科学卓越中心, 厦门 361021
4 中国科学院大学, 北京 100049
5 南京信息工程大学, 南京 210044
6 江苏省环境监测中心, 南京 210036
7 上海市气象局, 上海 200030

Application of Improved Super Ensemble Forecast Method for O3 and Its Performance Evaluation over the Yangtze River Delta Region
YAO Xuefeng1,2, GE Baozhu1, WANG Zifa1,3,4, FAN Fan1,5, TANG Lili6, HAO Jianqi1,4, ZHANG Xiangzhi6, YAN Pingzhong1, ZHANG Wending1, WU Jianbin7
1 State Key Laboratory of Atmospheric Boundary Layer Physics and Atmospheric Chemistry, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing 100029
2 96631 Army, People's Liberation Army of China, Beijing 102208
3 Center for Excellence in Urban Atomsprheric Environment, Institute of Urban Environment, Chinese Academy of Sciences, Xiamen 361021
4 University of Chinese Academy of Sciences, Beijing 100049
5 Nanjing University of Information Science and Technology, Nanjing 210044
6 Jiangsu Environmental Monitoring Center, Nanjing 210036
7 Shanghai Meteorological Service, Shanghai 200030
Abstract: Aiming at existing problems in current O3 single model forecast, an efficient superensemble forecast based on running active range (AR-SUP) is proposed and applied to the EMS-YRD (multi-model ensemble air quality forecast system for the Yangtze River Delta) O3 forecast during the study period in 2015. The performance of the newly proposed method is compared with those of R-SUP (Running Training Period Superensemble), EMN (Ensemble Mean), and BREM (Bias-Removed Ensemble Mean). The results show that compared with the other three ensemble methods, the AR-SUP exhibits significant improvement in daily O3 forecast with the RMSE reduced by 20% and 23% from that of the best single model in cool and warm seasons respectively. Further application of the AR-SUP in O3 ensemble forecast also shows high forecasting skills when the predicting time is extended to 48 h and 72 h. A number of statistical measures (i.e., reduced errors, increased correlation coefficients, and index of agreement) show that the forecasting skill has been improved at all the locations within the study region during all seasons, which indicates this method can be used to help improve the accuracy and reliability of short-term forecasts.
Keywords: O3    Ensemble air quality multi-model forecast system    Superensemble forecast    Running active range
1 引言

2 资料和方法 2.1 长三角地区多模式空气质量预报系统

 图 1 长三角地区（a）多模式系统区域设置，（b）代表空气质量监测站的分布 Figure 1 (a) Domain setting of the multi-model ensemble air quality forecast system, (b) representative air quality monitoring stations for the Yangtze River Delta

2.2 站点选取及观测资料
 ${{S}_\text{picked}}=\text{max}(D({{S}_\text{refer}}, {{S}_\text{candidate}})),$ (1)

O3因极易受气象及城市排放条件影响，短时及日变化显著（殷永泉等，2004）；同时由于O3关注短期急性健康效应，我国于2012年将日最大8 h浓度作为O3日评估标准纳入到新的《环境空气质量标准》中（中华人民共和国环境保护部，2016）。故文中O3的评估考察日最大8 h浓度，具体算法参考了美国环保署公布的8 h O3数据处理指南，即先对小时分辨率资料进行8 h滑动平均，然后逐日对滑动平均结果求最大值得到（Park, 1998）。

2.3 集成方法

（1）多模式集成平均（EMN）

 ${{V}_\text{EMN}}(j, \text{ }t)=\frac{1}{N}\sum\nolimits_{i=1}^{N}{{{F}_{i}}(j, \text{ }t)},$ (2)

（2）消除偏差的集成平均（BREM）

 ${{V}_\text{BREM}}\left(j, \text{ }t \right)=\overline{O}\left(j, \text{ }t-1 \right)+\frac{1}{N}\sum\nolimits_{i=1}^{N}{\left[ {{F}_{i}}\left(j, \text{ }t \right)-\overline{{{F}_{i}}}\left(j, \text{ }t \right) \right]},$ (3)

（3）超级集成法（SUP）及滑动训练期的超级集成（R-SUP）

Krishnamurti et al.（1999）指出，系统偏差信息还可通过模式在训练期的表现提取得到，并在预报期予以消除。通过选定一段训练期，在训练期获取各个模式的权重系数，并在训练期内得到观测平均态和模式距平，综合考虑模式的不等权重和消除偏差即得到超级集成预报（SUP）：

 \begin{align} &{{V}_\text{SUP}}\left(j, \text{ }t \right)=\overline{O}\left(j, \text{ }t-1 \right)+ \\ &\sum\nolimits_{i=1}^{N}{{{a}_{i}}\left(j, \text{ }t \right)\left[ {{F}_{i}}\left(j, \text{ }t \right)-\overline{{{F}_{i}}}\left(j, \text{ }t \right) \right]}, \\ \end{align} (4)

 $G\text{=}{{\sum\nolimits_{t=1}^{{{N}_\text{tr}}}{({{{{V}'}}_{\text{SU}{{\text{P}}_{\text{t}}}}}-{{{{O}'}}_\text{t}})}}^{2}},$ (5)