TITLE:  Resource allocation problems under uncertainty in humanitarian supply chains

STUDENT: Melih Celik

ADVISORS: Dr. Ozlem Ergun and Dr. Pinar Keskinocak

SUMMARY:

In this thesis, we consider three applications of humanitarian, emergency response and public health supply chains, namely (i) debris management, (ii) allocation of perishable commodities, and (iii) specialized nutritious foods supply chain design, where the main decisions relate to the allocation of scarce resources and there is inherent uncertainty in the environment. Our particular focus is on incorporating stochasticity into the models that address these decisions and making use of information updates as they become available.

Debris management is one of the most time consuming and complicated activities among post-disaster operations. Debris clearance is aimed at pushing the debris to the sides of the roads so that relief distribution and search-and-rescue operations can be maintained in a timely manner. In the first part of the thesis, we define the stochastic debris clearance problem (SDCP), which captures post-disaster situations where there is limited information on the debris amounts on the roads and the information is updated as clearance activities proceed. We propose a partially observable Markov decision process (POMDP) model to solve SDCP to optimality. Due to the extensive computational requirements of the POMDP model, we propose a heuristic based on a continuous time approximation, and further reduce the computational burden by applying a limited look-ahead on the search tree and using heuristic pruning. The performance of these approaches is tested on randomly generated instances that reflect various geographical and information settings and instances based on a real world earthquake scenario.

Motivated by a public health supply chain problem, we consider the allocation of a perishable item in a decentralized supply chain in the second part of the thesis. The supply chain consists of two levels: a capacitated allocation center upstream and multiple distribution locations downstream. Each distribution location serves multiple demand types with varying unit benefits accrued from satisfying the demand, and applies a certain usage policy to satisfy the uncertain demand. In particular, we focus on prioritization of higher-class demand and first-come, first-served (FCFS) usage policies. Under the assumption of no reallocation, we develop myopic algorithms that find the optimal allocation that maximizes the expected net benefit for the system. When demand arrivals follow a Poisson process, we establish a worst-case bound on the relative performance of optimal allocation under FCFS compared to that under prioritization. We also show that centralization, i.e., serving a certain demand class or all demand from a single location, can guarantee the improvement of expected net benefit over a decentralized system only if demand is prioritized and completely served from a single location. We compare the relative performances of the two usage policies, assess the effects of centralization, and compare our approaches to simpler allocation schemes using computational experiments from a dataset based on the 2009 H1N1 vaccination campaign in the state of Georgia.

Specialized nutritious foods (SNF) are aimed at the prevention and treatment of undernutrition, which leads to irreversible and long-term consequences such as hampering of body growth and reduction in productivity. Certain aspects of SNF, such as high product value, low shelf life, geographical disparity of supply and demand, limited visibility into the supply chain, and the need for refrigeration pose important challenges for efficient supply chain management. In the last part of this thesis, we present a case study on the SNF supply chain of a large scale humanitarian organization, where the aim is to determine the network structure and optimal allocation of these commodities to the distribution locations. We develop mixed integer programming models for determining the facility locations, incorporation of information updates, and equity of allocations. Using computational experiments based on data from the organization, we determine aspects of the system that have the most significant impact on the selection of network structure, and show the impact of incorporating uncertainty and information updates to our models.