Spatio-Temporal Reconstruction of Climate from Large Pollen Data Sets

• Activity Number: 620 - Spatial and Spatiotemporal Modeling in Climate and Meteorology
• Type: Contributed
• Date/Time: Thursday, August 1, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics and the Environment
• Author(s): John Tipton* and Basil Davis and Manuel Chevalier and Philipp Sommer
• Companies: University Of Arkansas and University of Lausanne and University of Lausanne and University of Lausanne
• Keywords: pollen; paleoclimate; spatio-temporal model; multinomial data; low-rank approximation

Abstract

We present a Bayesian Hierarchical model framework for the prediction of spatio-temporal climate from large pollen proxy datasets. Our work builds upon prior work using overdispersed multinomial functional regression models and extends these efforts to generate predictions across regional and continental scales while making relatively few assumptions. The development of a computationally efficient estimation framework makes fully Bayesian inference possible while the development of statistical software makes the model accessible to practitioners. We demonstrate the performance of our model using simulation and demonstrate how site-level inference can deviate from regional, continental and global trends.