基于时空关联的细颗粒物成分空间推测方法及应用研究

Ambient fine particles (PM2.5) has caused obvious environment, health and climate effect. However, one of the basic scientific problems that the temporal and spatial distribution characteristics of PM2.5 chemical compositions are not clear. This study tries to obtain its temporal and spatial variability under designed sampling strategy from two aspects: (1) meteorological factors such as wind speed, wind direction, relative humidity, solar radiation, etc. affect the dispersion rate and generation speed of secondary particles, are the dominant factors causing dynamic changes. A generalized additive model (GAM) will be established to quantitatively express the effects of meteorological factors on PM2.5 components based on continuously sampling at a fixed site; 2) land use characteristics and spatial distribution of pollution sources are main factors causing spatial variation of PM2.5 components. This study will capture such spatial variation pattern by purposive sampling, explore the relationship between PM2.5 components and their co-variables, and finally build a land use regression model (LUR) to describe of spatial variability of PM2.5 components. Adjusted by time series' trend, we can achieve the modeling of the spatio-temporal variability, and the results will be verified by an independent set of samples. This study aims to establish an efficient spatio-temporal predictive mapping method based on the integration of GIS technology, remote sensing and atmospheric analysis, and finally solve the above basic scientific problems and obtain first-hand PM2.5 components distribution data. The results from this proposed study will greatly improve quantitative evaluation of air pollution on health risk and uits ncertainty, and will also provide important data and scientific basis for pollution control measures.

大气细颗粒物（PM2.5）引发了显著的环境、健康及气候效应，但其“化学成分时空分布特征不明晰”这一基本科学问题尚未解决。本项目拟通过设计采样方案分别表征PM2.5成分的时间和空间变异性，在空间模型上叠加时间序列的调节作用，实现兼顾时-空变异的推测建模，并对结果进行验证。具体方法为：1）气象要素影响着扩散速率及二次颗粒物的生成，是引起时间动态的主导因素，拟建立气象与定点连续采样PM2.5成分之间的广义可加模型（GAM），定量表达气象-成分时间序列；2）污染源的空间分布及周边地表状况是成分空间变异的决定因素，通过布设空间代表性样点，捕捉成分的空间变异，探寻其与空间协同变量的关系，建立刻画成分空间变异的土地利用回归模型（LUR）。最终将时空模型进行叠加及验证。本研究旨在融合GIS、遥感和大气分析技术，建立高效的细颗粒物成分时空推测方法，获取一手分布数据。为定量评价大气污染健康风险及其不确定性、有针对性地采取污染控制措施提供重要数据和科学依据。

大气污染物的精细时空预测建模对大气环境治理和人群暴露评价具有重要意义。本课题围绕该方向开展了一系列研究，首先，在对空间插值理论、时间序列预测、时空预测理论方法研究和总结的基础上，凝练和提出了融合多源观测数据的大气污染物及颗粒物组分空间推测方法框架；其次，综合利用地面大气污染常规观测数据、布点采样数据、地面基础地理信息、遥感、气象等多源数据，构建了各协同变量与大气污染物的定量关系模型、基于深度学习的大气污染物时间序列预测模型、基于土地利用回归及机器学习的空间预测模型等，实现了对大气污染物及颗粒物组分的空间推测建模，同时分析了输入变量及样点位置和数量对建模精度的影响；最后，运用所构建的模型，以北京和石家庄为例，开展了模型应用研究，验证了所构建方法的有效性，将优化后的土地利用回归方法开发形成了自动化预测制图软件eLUR，大大提高了模型的精度和效率。项目共发表标注基金课题的论文共19篇，其中17篇为SCI收录，2篇为中文核心；参编专著《空气污染人群健康风险评估方法及应用》、取得软件著作权1项，培养硕士研究生3名，参加会议学术报告或特邀报告5次。所开发的eLUR软件已免费开放下载，应用于环境健康领域一线工作。