Predictiing Historical Climate with a Reduced Rank Model with an Orthogonal Predictive Process

• Activity Number: 404
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): John Tipton*+ and Mevin Hooten
• Companies: Colorado State University and Colorado State University
• Keywords: Predictive Process ; Spatial Statistics ; Paleoclimate ; Sparse Data

Abstract

Most statistical reconstructions of paleoclimate use current and historical instrumental records to train a model of paleoclimate variables inferred from records of climate proxies such as tree rings, lake levels, coral bands, boreholes, etc. Because the instrumental record is short relative to the periodicity of m\ any climate signals, any extension of the climate signal can be of vital use. A source of pre-processed instrumental records in the United States from 1895 to the present is from PRISM (Parameter-elevation Relationships on Independent Slopes Model). Between 1820 and 1895 in the upper midwest of the United States is a record of historical US fort temperature measurements that are relatively sparse. We propose a Bayesian principal components reduced rank model with an orthogonalized spatial predictive process to reconstruct the historical temperature surfaces from an extremely sparse historical temperature record. This is done with a cross-validation to prevent unwarranted departures from the mean surface due to the sparsity of the data.