Inference and Analysis of Climate Sensitivity via Data-Reduction Techniques: A Detailed Approach

• Activity Number: 43
• Type: Contributed
• Date/Time: Sunday, August 9, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Gabriel Huerta* and Mohammad Hattab and Charles Jackson
• Companies: University of New Mexico and University of New Mexico and The University of Texas at Austin
• Keywords: climate sensitivity ; high dimensional output ; dimension reduction ; Principal component regression ; Bayesian/Frequentist

Abstract

Climate sensitivity is a quantity of great interest in climatological research and defined as the average change in global mean temperature when CO2 is doubled. To study this quantity, we consider output/data generated by the NCAR Community Atmospheric Model (CAM 3.1). In addition to climate sensitivity, CAM 3.1 produces model runs of climate simulation generating output related to 11 different fields or variables, such as humidity, and measured over a spatial grid consisting of 8192 locations. There are only 165 of such model runs which induces a problem where the number of predictors is overwhelmingly greater than the number of observations. The main goal of this study is to introduce sensible and flexible methods to predict climate sensitivity based on the current CAM 3.1 models runs and for those available for other models. The problem is tackled via Principal Component Regression (PCR) analysis based on Bayesian and Frequentist perspectives. Specifically, the Bayesian approach is needed to solve identifiability issues that are usually ignored in the literature of PCR. Our framework provides maps to visualize the relationship between climate sensitivity and the 11 fields.