Statistical Downscaling of Sea Surface Temperature from Global Climate Models

• Activity Number: 413 - Analyses of Environmental Data
• Type: Contributed
• Date/Time: Thursday, August 12, 2021 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Ayesha Ekanayaka* and Emily Kang and Peter Kalmus and Amy Braverman
• Companies: University of Cincinnati and University of Cincinnati and NASA Jet Propulsion Laboratory and Jet Propulsion Laboratory, California Institute of Technology
• Keywords: Downscaling; local approximate Gaussian process; Global Climate Models; Spatio-temporal dependencies

Abstract

Sea Surface Temperature (SST) is an essential factor in climate science. Numerous studies reveal that rising SST is a severe threat for many marine ecological systems, particularly for coral reef systems. Thus, projected potential changes in future SST will help conservation plans. Global Climate Models (GCMs) provide future climate estimates including SSTs. However, these projections are generally made at coarse resolutions (~1 degree) and hence they become too coarse to capture fine-scale variations in a regional SST process. Therefore, downscaled versions of GCM projections are highly demanded in regional studies. We propose a statistical downscaling technique to produce high-resolution (~0.01 degree) SSTs for the Great Barrier Reef (GBR) area. Our method involves local approximate Gaussian Process (laGP) and hence we computationally benefit from high parallelizability. Moreover, the proposed method is developed paying thorough attention to spatio-temporal dependencies and it is capable of providing uncertainty estimates, which is often a lacking skill in previous downscaling approaches. The performance of our method is compared with the state-of-art downscaling method for SSTs.