

CISL’s Institute for Mathematics Applied to Geosciences (IMAGe) continued its “Beyond P-values” series of training classes with the three-day course titled “Introduction to Bayesian Statistics” on July 11-13 at NCAR’s Mesa Lab campus. This course introduced the theory and methods of Bayesian Statistics to NCAR staff and visitors. Bayesian Statistics are applied to many problems in climate science, and are especially relevant in uncertainty quantification.
The course covered some simple one- and two-parameter problems to introduce the mechanics of the Bayesian approach, and it included Markov Chain Monte Carlo (MCMC) methods for making inferences using more complicated (higher dimensional) models. For practice using these methods, participants fitted and evaluated models for a dataset of atmospheric CO2 concentrations measured in ice cores. The course incorporated the open-source statistical software package R and the open-source MCMC package Nimble that converts code written in R to C++ for improved speed.
The co-creator and lead instructor for the course, Alix Gitelman is a Professor of Statistics at Oregon State University. She was assisted by NCAR scientist and course co-creator Dorit Hammerling (CISL IMAGe), who is in charge of the overall development of the Beyond P-values series. Brian Reich, Professor of Statistics at North Carolina State University and Doug Nychka (CISL IMAGe) gave guest lectures on the selection of Bayesian priors and the value of hierarchical modeling. For additional support during in-class exercises, statistics graduate students Nathan Lenssen (Columbia University), Pulong Ma (University of Cincinnati), Matthew Edwards (University of Newcastle), and Felipe Tagle (University of Newcastle) served as coaches and provided one-on-one help for participants, while at the same time benefiting from the interactions and discussions with NCAR scientists and visitors.
IMAGe provides the Beyond P-values series of courses for NCAR staff and visitors who are interested in advancing their statistical and data analysis skills. Each course focuses on a specific statistical topic and is taught by experts in that area. This series is designed to expose participants to a variety of experts and potential collaborators, so typically multiple experienced practitioners serve as co-instructors for the participants. The instruction is data-driven and focuses on hands-on training in statistical software tools. The courses are limited to a small number of participants to provide high-quality individual support by the instructors and coaches.
