BME-based spatiotemporal local scale mapping and filtering of mortality data: The California study.

摘要

The meaningful characterization of a health-related field involves the appreciation of its spatioternporal variation at multiple scales. A rigorous description depends on the scale at which the phenomenon is considered rather than being limited by the scale at which measurement is taken. Modern geostatistical studies use various forms of hard and soft data to specify the correlation structure of the process and generate its accurate and informative spatioternporal map. The data are available at a variety of scales which are often different from the mapping scale considered. For example in many epidemiological studies, data are available at a higher level of scale (say, county scale), whereas the epidemiologist is interested in the data at a lower scale (say, locally zip code). This local scale performance may help the human exposure analysis be more meaningful. This situation requires some approach to down scale the epidemiological space/time modeling from the higher to its local scale. In this study, a new space/time mapping approach is developed, called the spatiotemporal local scale (S/T LS) approach that studies the effects of the different scales associated with the problem of interest and generates the scale-dependent maps. The theoretical support of the S/T LS approach is the well-known BME theory of spatioternporal random field analysis and mapping. Some vectorial and multi-point formulations of the BME model are considered and various analytical and numerical examples are discussed. In addition, a real-world case-study is analyzed which involves mortality data from the state of California. When applied to the mortality data set, the proposed S/T LS approach accounts for the effects of different scales (e.g., county scale vs. local zip-code scale) and generates space/time maps that are more accurate and informative than those obtained by other mapping models not accounting for the scale effect.