Date of Award

May 2018

Degree Type

Thesis

Degree Name

Master of Science

Department

Mathematics

First Advisor

Sergey Kravtsov

Committee Members

Clark Evans, Vytaras Brazauskas

Abstract

In this thesis, Wiener filtering of gridded surface-temperature time series from observations and climate model simulations is performed by using multi-channel singular spectrum analysis (M-SSA) in order to isolate non-stationary climate signals. The contributions to the singular spectrum from shorter-term internal climate variability, treated in this context as noise, are estimated by fitting to the data spatially extended stochastic models, which are subsequently used to produce synthetic ensembles of surface temperature time series and the corresponding synthetic M-SSA spectra. The full spectra are weighted by the signal-to-noise ratios and transformed back to physical space to obtain reconstructions of the non-stationary signal. This methodology was first tested using the twentieth century simulations from the Community Earth System Model Large Ensemble Project, for which the forced climate signal can be reliably estimated by taking the ensemble average over the 40 available climate realizations, then applied to individual model ensembles as well as the overall ensemble from the Coupled Model Intercomparison Project Phase 5 and, finally, to the observational surface-temperature time series from Twentieth Century Reanalysis. The method is shown to successfully recover the low-frequency (decadal or larger time scales) component of the forced signal in model simulations, but fails to isolate shorter-term variability associated with volcanic eruptions. The secular signals estimated from model simulations and observations exhibit large differences, which indicates the presence, in observations, of a pronounced multi-decadal variability with a distinctive spatiotemporal structure absent in any of the model simulations.

Download

Share

COinS