Methods for comparison and analysis of spatiotemporal fields

This dissertation develops three methods for spatiotemporal fields data, each designed to address research topics in climate science. The first two methods are similarity measures for evaluating the differences between climate models and observational datasets. The last method is a multivariate spatiotemporal model which characterizes the joint evolution of observed climate processes. Chapter 2 introduces the spherical convolutional Wasserstein distance (SCWD) to more comprehensively measure differences between climate models and observational data. This new similarity measure accounts for spatial variability using convolutional projections and quantifies local differences in the distribution of climate variables. We apply SCWD to evaluate the historical model outputs of the Coupled Model Intercomparison Project (CMIP) members by comparing them to observational and reanalysis data products. Additionally, we investigate the progression from CMIP phase 5 to phase 6 and find modest improvements in the phase 6 models regarding their ability to produce realistic climatologies. Chapter 3 introduces the sliced elastic distance, a new metric which considers potential time misalignment between climate models and observational data. Sliced elastic distance decomposes differences in the local evolution of climate processes into shape differences (amplitude), timing variability (phase), and bias (translation). We apply the sliced elastic distance to rank CMIP phase 6 precipitation models by their similarity to observational data at both global and regional scales. Using intermediate calculations from our method, we perform an in-depth phase analysis of the Indian summer monsoon to identify timing biases in the onset and retreat of the monsoon season in each CMIP6 model. Finally, Chapter 4 introduces a novel multivariate space-time dynamic model to quantify relationships in the joint evolution of atmospheric processes. This model captures spatial variation using a flexible set of basis functions for which the coefficients are allowed to vary in time through a vector autoregressive (VAR) structure. The model is cast in a Bayesian dynamic linear model (DLM) framework and estimated using a customized MCMC sampling approach. We apply this model to study the relationship between aerosols, radiation, and temperature following the 1991 Mt. Pinatubo eruption and highlight when such a model is advantageous over simpler univariate models.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Statistics

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2024

Author dc:creator

• Garrett, Robert Charles

Contributors dc:contributor

• Li, Bo
• Douglas, Jeffrey A
• Shand, Lyndsay
• Harris, Trevor A

• Spatial Statistics
• Functional Data Analysis
• Dynamic Linear Model
• Climate Model Evaluation

Statement dc:rights

• Copyright 2024 Robert Garrett

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/125782

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F125782