Computational Methods for Hierarchical Spatial Models and Ice Sheet Model Calibration
Open Access
- Author:
- Lee, Seiyon
- Graduate Program:
- Statistics
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- May 20, 2020
- Committee Members:
- Murali Haran, Dissertation Advisor/Co-Advisor
Murali Haran, Committee Chair/Co-Chair
Ben Shaby, Committee Member
Ephraim Mont Hanks, Committee Member
Klaus Keller, Outside Member
Ephraim Mont Hanks, Program Head/Chair - Keywords:
- Spatial Statistics
Zero-inflated spatial data
Computer model calibration
sampling importance resampling
Basis Representation
computational statistics
uncertainty quantification - Abstract:
- Computer model calibration is a major component in projecting sea level rise and developing coastal flood-risk management strategies. Hierarchical spatial models have been used extensively to model spatially dependent observations across many fields such as climate science, ecology, public health, and epidemiology. The computational methods presented here have wide ranging applications in environmental sciences such as quantifying uncertainties in future sea level rise which are then used to formulate coastal risk management policies and providing researchers from various fields with a fast and readily extendable approach to fit complex hierarchical spatial models of their choice. My dissertation research focuses on developing statistical and computational methods to address pressing issues in the environmental sciences. My contributions are as follows: (1) a fast particle-based approach for calibrating a three-dimensional Antarctic ice sheet model. I developed a sequential Monte Carlo method that leverages the massive parallelization inherent to modern high-performance computing systems; (2) an efficient and extendable approach for fitting high-dimensional hierarchical spatial models. I propose a discretized and dimension-reduced representation of the underlying spatial random field using empirical basis functions on a triangular mesh; and (3) a computationally efficient method for modeling high-dimensional zero-inflated spatial observations.