Data Fusion Approaches for Analysing Many Spatial Outcomes Jointly - Semiparametric Mixture Methods for Correlated Counting Processes and Joint Outcome Analyses

点击翻译按钮获取中文摘要

This work develops new statistical tools to address serious issues in data analysis and brings these to bear in important applications related to water, surveillance and forestry. Statistical methods for using related sources to build information are in critical demand for advancing science; indeed this gap is a major obstacle in environmetrics. The focus of this proposal is to address this gap in several related projects as described below. The proposed tools will serve to jointly map several diseases or mortality rates in order to identify the commonalities in the spatial structure across the maps. Visualizing rates is important because this may provide cues to suggest factors which may be linked to mortality or disease, for further detailed study. As well, maps provide an overall description of mortality or disease rates that can identify disease inequities or areas of resistance to disease. The strength of the methods developed is their capacity to take into account spatial aspects of the data, and any connections which exist because of spatial configurations. By so doing, the methods are able to build better estimates for any specific region, because they use all the information in the map, and across maps, in a combined analysis, as appropriate. More importantly, the spatial connections across the map are themselves of interest in order to understand the disease process. We will propose new ways for analyzing natural processes which result in data that have many zero values, which is common in ecological studies. This research fills the gap in analytical methods for handling this sort of disparate data. It will help to address situations where there is a spatial connection between the zeros and between the non-zero counts, as well as across these two compartments of the data, with this spatial connection changing over time. In some contexts, the data contain both an abundance of zeros and high extreme values, such as in studies of river flow in intermittent streams, where floods occur. The difficulty of dealing with extremes in both ends of the data has long frustrated hydrologists. New statistical tools will be brought to bear to aid in models which are based on hydrological principles to solve these challenges. Importantly, the research considers the development of tools to automatically isolate specific kinds of information from satellite images. This will raise the efficiency and effectiveness of earth surveillance. New methods will be developed to model mixed spatial data sources, such as from satellite data and ground observations, and improve surveillance of forest fires using satellite images. We will also study novel ways of extracting information from satellite data to estimate soil moisture for agricultural purposes. The proposed research will serve to increase expertise in Canada in such complex data analysis. Students trained will gain significant experience, increase development of new methods as well as the marketability of their knowledge and skill sets within academia, industry and at government agencies.