Examination of the Mathematical Models Used in Atmospheric Source Apportionment Studies

9523731 霍普克 PI开发和测试了一些不同的数学技术，可以帮助从成分数据集中提取空气中颗粒物大气传输的源/受体关系的信息。这些方法包括用于将高维、多元素组成数据投影到二维或三维中的技术;以及通过创建条件概率场将气象信息以反向轨迹计算的形式整合到分析中。对于多变量空气质量数据的解释，已经探索了神经网络的使用，以及其与最小生成树技术的结合。 在本项目中，将继续发展、测试和应用新的多元统计方法来处理大气样本成分数据，主要是颗粒和降水。与来自世界各地的大气科学家建立了广泛的合作关系，这些科学家已开始进行实地研究，包括在大湖地区、纽约北部、切萨皮克湾以及澳大利亚、芬兰和瑞典进行测量。PI的分析方法将为检查数据提供新的工具，并帮助从这些数据库中提取最大的有用信息。在将继续进行的开发和测试项目中，有正矩阵因子分解和多模态因子分析，用于查看几种不同类型的3向数据表。

9523731 Hopke The PIs have developed and tested a number of different mathematical techniques that can help extract information on source/receptor relationships in the atmospheric transport of airborne particulate matter from compositional data sets. These methods include techniques for projecting high dimensionality, multielemental composition data into two or three dimensions; and the integration of meteorological information in the form of back trajectory calculations into the analysis by creating a conditional probability field. For the interpretation of multivariate air quality data, the use of Neural Networks has been explored, as well as its combination with the minimal spanning tree technique. In this project, the development, testing, and application of new multivariate statistical methods to atmospheric sample composition data, primarily particles and precipitation, will be continued. Extensive collaborations have been established with atmospheric scientists from all parts of the world who have initiated field studies, including measurements in the Great Lake area, upstate New York, the Chesapeake Bay, as well as Australia, Finland and Sweden. The PI's analysis methods will provide new tools for examining data, and help extract maximum useful information from these data bases. Among the development and testing projects that will be continued are Positive Matrix Factorization, and multimodal factor analysis for looking at several different types of 3-way data tables.