Back to search

Clemson University

Multi-Stage Stochastic Programming for Disaster Relief Logistics Under Forecast Uncertainty

Abstract

dc:description.abstract

Hurricanes are among the most destructive annual disasters in the United States, presenting interdependent challenges for evacuation planning and relief-supply logistics. Coordinating evacuation and relief operations is crucial to ensure the timely and effective movement of at-risk populations and the delivery of essential supplies. This dissertation develops and evaluates three progressively advanced multi-stage stochastic programming (MSSP) frameworks that integrate evacuation and relief-item pre-positioning while explicitly accounting for uncertainty in hurricane forecasts. Chapter 1 introduces a fully adaptive MSSP model for the integrated hurricane relief logistics and evacuation planning (IHRLEP) problem. The model simultaneously optimizes evacuation flows and inventory pre-positioning over time as new forecasts become available. The hurricane process is approximated as a Markov chain (MC), with transition dynamics derived from autoregressive models of historical forecast errors issued by the National Hurricane Center (NHC). Case studies based on Hurricanes Florence and Ian demonstrate that adaptive MSSP policies significantly reduce out-of-sample (OOS) costs compared to static two-stage models and offer insight into the benefits of adaptivity under varying lead times and resource conditions. Chapter 2 extends the IHRLEP framework to a distributionally robust (DRO) MSSP model to enhance worst-case OOS performance without relying on parametric assumptions about the hurricane process. Historical forecast errors are used in conjunction with non-parametric kernel regression to generate a family of plausible conditional distributions for storm track and intensity. A tractable MC discretization supports decision-making that hedges against worst-case scenarios while maintaining asymptotic optimality. Computational experiments show that the DRO-based MSSP approach achieves stronger robustness than models based on a single nominal distribution. Chapter 3 incorporates the sequential nature of rolling forecasts by comparing three modeling strategies: (i) an MSSP based on the Martingale Model of Forecast Evolution (MMFE), (ii) a time-series (TS) MSSP that combines an autoregressive intensity model with a MC for track, and (iii) a rolling horizon (RH) approximation that updates decisions based on the latest forecast without explicitly modeling uncertainty. Using real forecast archives, the MMFE-based MSSP demonstrates superior accuracy in capturing forecast evolution and achieves the lowest expected OOS cost. Meanwhile, the RH approach offers a simpler online decision-making tool at the expense of decision quality. Together, these three studies contribute a comprehensive suite of modeling and algorithmic tools for improving the reliability, robustness, and responsiveness of coordinated evacuation and relief-logistics planning under hurricane forecast uncertainty. This work advances the field of stochastic disaster operations research and provides actionable insights for emergency managers facing high-stakes, time-sensitive decisions.

Degree

thesis:*
Name thesis:degree_name
Doctor of Philosophy (PhD)
Level thesis:degree_level
Dissertation
Discipline thesis:degree_discipline
Industrial Engineering

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Bhattarai, Sudhan
Contributors dc:contributor
  • Dr. Yongjia Song
  • Dr. Thomas Sharkey
  • Dr. Hamed Rahimian
  • Dr. Qi Luo

Subjects

dc:subject × 8

Identifiers

dc:identifier.*
Repository record dc:identifier
https://open.clemson.edu/all_dissertations/4070

Chain of custody

source
Harvested from
Clemson University
Base URL
open.clemson.edu/do/oai/
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Bhattarai, Sudhan. Multi-Stage Stochastic Programming for Disaster Relief Logistics Under Forecast Uncertainty. Dissertation thesis, https://open.clemson.edu/all_dissertations/4070