大气复合污染海量多源观测同化与集合预报方法研究

High accuracy air pollution forecast is significantly valuable for the air pollution prevention and control in China. This project aims to develop advanced chemical data assimilation methods and ensemble forecast approaches with focusing on the improvement of PM2.5 and O3 forecast. Firstly, in order to overcome the limitations of the current algorithm in dealing with strong-nonlinearly chemical data assimilation issues, a revised assimilation algorithm with stepwise and smoothing correction will be developed. Secondly, inversely estimating the reactive precursor emissions of PM2.5 and O3 will be investigated through taking into account the uncertainties of chemical reaction reactions. Thirdly, an advanced assimilation scheme that enables the assimilation of multi-source and massive observation data will be designed to provide balanced constraints for the initial conditions and emissions of air pollutants. Finally, new air quality ensemble forecast method including both super ensemble forecasts and machine learning will be developed and its potentials in improving the PM2.5 and O3 forecast will be explored. This project will provide new advanced approaches for improving the air pollution forecasts in China, and it will also bring out a high-accuracy reanalysis dataset for the health impact researches. As a result, this project could serve as an important support for the scientific goal of the whole research project.

减小大气污染预报预警不确定性对我国当前大气污染防控具有非常重要的现实意义和科学价值。本项目围绕我国PM2.5和O3预报问题，研发误差分步订正的集合同化算法和耦合化学反应不确定性的源反演方法，突破大气污染资料的强化学非线性同化难题和强化学活性污染物的源反演问题，开展卫星、地基和雷达多源观测数据的误差评估和自适应质量控制，发展海量多源观测的多尺度平衡约束同化方案，减小PM2.5和O3预报关键前体物的初始浓度场和源排放数据不确定性。进而，研发多模式多扰动超级集合预报方法，探索针对海量多源观测和超级集合预报的机器学习集成预报，提升PM2.5和O3预报特别是3-5天预报的准确率。本项目成果将为我国大气复合污染应对提供先进的预报预警新方法，为健康效应研究提供高精度的多源观测再分析数据集，从而为联合重大研究计划的科学目标提供重要支撑。本项目研究内容也属于重大研究计划拟开展的重点研究内容之一。

本项目围绕大气复合污染的资料同化和集合预报问题，开展了多源观测数据的误差估计和质量控制、同化融合与源清单反演、大气污染集合预报三大内容的研究，破解了多污染协同同化、强化学污染物源反演等难题，研发了面向我国大气环境地面监测网络的全自动异常数据识别方法、六项常规污染物观测数据的协同同化方法、强活性污染物的源清单反演方法，建立了相应的软件系统。集成应用上述方法和软件系统，建立了我国首个大气污染再分析数据集并公开共享，数据文件总下载量超30万次，作为6大数据之一入选重大研究计划数据集成项目“中国大气复合污染多源信息综合数据共享平台研发（92044303）”。项目首次实现中国地面氨气浓度观测数据对中国大气氨排放进行了高空间分辨率的反演优化，获得了中国区域氨的逐月反演排放量。项目研发的常规污染物观测数据协同同化软件在中国环境监测总站、京津冀、长三角和珠三角区域中心等单位业务应用，大幅提升了空气质量预报预警的准确率，成为空气质量预报业务的核心软件，支撑2018年青岛上合峰会、上海进口博览会等重大国际活动的空气质量保障工作。项目发表论文18篇，其中SCI论文13篇，获批专利1项，申请软件著作权2项，培养博士研究生5名，硕士研究生4名，获国家科技进步二等奖2人次。