多源异构空气污染集合大数据复杂模型的不确定性分析是空气污染防治领域面临的一个重要难题。本项目利用跨学科的可视分析技术，从全新视角分析空气污染集合大数据的不确定性特征及演变过程，探索空气污染跨地域分布模式与相关性，解决时变、多源、高维的复杂集合大数据分析难度极高的问题。通过建立统一空气污染多源异构时空大数据模型，提出基于时空位图索引的不确定性量化表示方法，降低数据模型复杂度。引入邻域直方图思想及分布差异性评价标准，提出不确定性特征提取方法，研究运动信息检测与时序增量更新策略，实现在线时序特征跟踪；同时提出文本信息时空传播模型，对跟踪结果进行验证和反馈，提高准确性。构建多维度评价的全局相关体，引入视觉反馈机制，提高不确定性全局相关性分析精度，同时有效降低计算复杂度。构建多场集合数据的多变量可视化隐喻表达，建立交互式层级可视化平台。本项目研究成果能够为空气污染防治的相关研究提供新思路和理论依据。

The uncertainty analysis of air pollution complex model for multi-source heterogeneous ensemble data is an important problem in the air pollution control field. From a new perspective, this project uses interdisciplinary visual analysis technology to analyze the uncertainty features and evolution procedure of air pollution for ensemble data, and explore the distribution pattern of different regions and correlation of air pollution, as well as solve the complex analysis problems for big data with time-varying, multi-source, high dimensional characters. Through establishing the unified spatiotemporal model for air pollution multi-source heterogeneous big data, the uncertainty quantitative representation method based on spatiotemporal bitmap index is proposed, so as to reduce the complexity of data model. According to the neighborhood histogram idea and distribution difference evaluation criterion, the uncertainty feature extraction method is proposed to study motion detection and sequential incremental updating strategy, so that it is capable to implement the online time sequential feature tracking. Furthermore, the spatiotemporal diffusion model for text information is put forward to verify and feedback the result of tracking, thus the accuracy will be improved. By constructing the global correlation volume with multidimensional evaluation information, and utilizing the visual feedback mechanism, the precision of global correlation computation for uncertainty analysis is improved highly and the computational complexity is reduced effectively. Finally, the visual metaphorical expression method for multi-field multivariate ensemble data is presented, and hierarchical interactive visualization platform is built to help experts to explore the hidden features in the big data. The research results of this project can provide new ideas and theoretical basis for the study of air pollution prevention and control.