Title: Statistical Spatio-Temporal Models with Applications to Natural Processes

Date: April 8th, 2025

Time: 3:30 PM – 5:00PM

Location: Georgia Freight Bureau Conference Room (ISyE Groseclose 226)

Meeting Link: https://gatech.zoom.us/j/7228364844

Guanzhou Wei

Industrial Engineering PhD Student

H. Milton Stewart School of Industrial and Systems Engineering

Georgia Institute of Technology

Committee:

Dr. Xiao Liu (Advisor), H. Milton Stewart School of Industrial and Systems Engineering

Dr. Jianjun Shi, H. Milton Stewart School of Industrial and Systems Engineering

Dr. Yu Ding, H. Milton Stewart School of Industrial and Systems Engineering

Dr. Kamran Paynabar, H. Milton Stewart School of Industrial and Systems Engineering

Dr. Shuai Huang, Department of Industrial and Systems Engineering, University of Washington

Abstract:

In response to increasingly frequent extreme natural events (e.g., floods, hurricanes, and wildfires), numerous Earth observation programs have been launched in recent decades. As the volume, resolution, and complexity of Earth-monitoring data increase, both opportunities and challenges arise in modeling and understanding the underlying natural processes. This thesis aims to develop statistical spatial-temporal models for analyzing and understanding critical natural processes using Earth observation data. 

In Chapter 2, we explore power-line fire risk quantification for power delivery infrastructures. We propose a new spatio-temporal point process that captures both the instantaneous and historical effects of key environmental covariates on power-line fire risk, as well as the spatio-temporal dependency among different segments of the power delivery network. In Chapter 3, we develop a physics-informed statistical spatio-temporal model for wildfire aerosol propagation, leveraging multisource remote-sensing data streams and the advection-diffusion equation that governs the process. In Chapter 4, we extend a recently proposed PDE-based statistical spatio-temporal model by incorporating a data-flipping method. This approach ensures that the physical spatial process becomes fully periodic and has a complete waveform without boundary discontinuities. Thus, the Gibbs phenomenon is eliminated even when the Fourier series is truncated in the PDE-based statistical spatio-temporal model.