Meredith Franklin, PhD in Statistics and Environmental Health from Harvard University
Abstract: Spatiotemporal processes exhibit complex dependencies that are often not separable. In a functional context, a variety of basis functions have been developed and applied under a penalized regression framework to effectively model spatiotemporal data over a domain. However, in many environmental applications there are additional modeling challenges including synthesizing massive high-dimensional data from multiple sources, dealing with spatial misalignment, and integrating functional forms of spatiotemporal covariates. Using several examples of estimating ground-level air pollution from remote sensing observations, the computational tools and machine learning methods to deal with dimensionality are discussed, and a distance-weighted spatiotemporal functional regression approach is presented.