基于WRF-Chem模式的华东区域PM_(2.5)预报及偏差原因

WRF-Chem based PM_(2.5) forecast and bias analysis over the East China Region

介绍了基于WRF-Chem在线区域化学/传输模式和INTEX-B与MEIC人为源排放清单建立的华东区域大气环境数值预报业务系统,并利用PM_（2.5）监测数据评估了该系统在2013年11月1日~2014年1月31日和2014年11月1日~2015年1月31日2个细颗粒物高浓度阶段的业务预报效果.结果表明,华东区域大气环境数值预报业务系统具有较好的预报效果：2个阶段和3个时效（24,48,72h）的效果基本相当,不同阶段和预报时效的相关系数基本在0.7以上;各城市的相关系数基本在0.5以上,约1/2城市的平均预报偏差在25%以内,污染预报的CSI/TS评分达到0.55.预报效果呈现出一定的区域性特征,华东中北部相关系数较高,北部的偏差中值和均方根误差较大;不同城市之间的表现呈现出明显的差异.分类检验和重点城市检验的结果都显示,预报效果随污染程度加重而降低.数值预报多数较观测偏低,约3/4呈现负偏差,平均偏低20%~30%.气象-污染的双向反馈作用的不足可能是引起重污染预报效果下降的主要原因,人为源清单的不确定性也影响了其准确性的提升,因此还需要进一步提升该系统对高PM_（2.5）污染的预报能力.

An operational forecasting system for atmospheric environment over East China was introduced in this paper. The system was established based on the WRF-Chem Model, an online coupled regional chemical transport model. Anthropogenic emission inventory was composed of MEIC and INTEX-B data. The forecasting performance for fine particles （PM2.s） was evaluated during two high-concentration time periods （November 1St, 2013-January 31^st, 2014 and November 1st, 2014-January 31 st, 2015）. Evaluation results could be summarized with the following features. 1） The numerical forecasting system had generally good performance of regional PMz5 forecasts. The performance was comparable during the two periods and three forecast lengths of 24-hour, 48-hour, and 72-hour. The integrated correlation coefficients （Rs） were greater than 0.7, and CSI/TS score for PMz5 pollution days was 0.55. For the cities evaluated, Rs were mostly over 0.5, and about half of locations had mean biases below 25%. 2） Regional differences could be found in the performance of different cities. Better performance of Rs was found over northern and central part of the domain, whereas relatively larger errors occurred over the northem areas. 3） When evaluated for pollution category and major cities, the model showed degraded performance during heavy PM2.5 pollution episodes. 4） PM2.5 concentrations tended to be under-estimated in general. About 3/4 of the daily biases were negative and most mean biases were between -20% and -30%. 5） Lack of feedback from pollution to meteorology as well as errors in the emission inventory were likely the main reasons leading to lower forecast capability during heavy PM2.5 pollution episodes. Therefore, further improvements were required to forecast accurately under these severe conditions.