The Statistics Seminar speaker for Wednesday, March 7, 2018, will be Joe Guinness, a visiting assistant professor within Cornell's Department of Biological Statistics and Computational Biology. Guinness studies modeling and computational issues that arise in the analysis of large spatial-temporal datasets, with a focus on applications in earth sciences, including soil, weather, and climate. He received his Ph.D. in 2012 from the University of Chicago, and is visiting us from NC State University, where he is an assistant professor in the Department of Statistics.
Talk: Spatial-Temporal Modeling in Environmental Statistics
Abstract: This talk will give a brief overview of several projects involving modeling and computation for big spatial-temporal datasets, with applications in the environmental sciences. This includes general purpose statistical methodology for fast and accurate interpolation of heterogeneously sampled spatial-temporal datasets, applications to remote sensing, prediction of impacts of climate change on pine growth, and a study of arsenic binding in complex soil environments. The remainder of the talk will focus on an application of spatial-temporal modeling to the compression of climate model output. Data storage costs have become a limiting factor for climate research that relies on high resolution numerical simulations. Our recently introduced statistical compression methods target the storage of a lower-dimensional summary of the full dataset and a statistical (probabilistic) model for the full dataset given the lower-dimensional summary. We argue that this framework is capable achieving large compression ratios and is appropriate for climate model output, which contains valuable information about future large-scale trends, whereas projected small-scale variation is meaningful only through its statistical characteristics. This project is a collaboration with Dorit Hammerling of the National Center for Atmospheric Research.