Skip to main content

PhD Seminar: Ricardo Batista, "Copula-based Optimal Split Questionnaire Design and Road Network Change Detection using a Hidden Markov Random Field"

Jul 24, 2023 - 9:00 AM
to Jul 24, 2023 - 10:00 AM

Speaker: Ricardo Batista, PhD Candidate, Department of Statistics, Iowa State University

Title: Copula-based Optimal Split Questionnaire Design and Road Network Change Detection using a Hidden Markov Random Field

Abstract: In many survey settings, split questionnaire designs (SQDs) achieve better performance than distributing a full questionnaire. The ability to ask subsets of questions allows practitioners to manage respondent burden and allocate more sample units to questions with higher uncertainty, for instance, leading to a more optimal outcome. Current methods of computing the optimal SQD can either handle mixed response or employ an objective function based on the data's Fisher information matrix (FIM) but not both. This study fills the gap by outlining a joint distribution for SQDs composed of possibly mixed-response items, thus unlocking optimality criteria based on the FIM. This manuscript also introduces an objective function that extends the notion of relative variability captured by the coefficient of variation to categorical data.

Maintaining the road network up to date is important to Geographic Information System applications such as navigation and urban planning. Image-based updating has gained prominence given the proliferation of relatively low cost remote-sensed imagery. But challenges remain as methods based solely on images must wrangle with incomplete structures and invalid topologies created during the feature extraction phase. This study introduces a change detection method that leverages the entire aerial image time series during feature extraction and change detection by modeling the road network over time using a hidden Markov random field defined on a spatiotemporal graph. Our experimental results on imagery provided by the National Resources Inventory demonstrate satisfactory change detection sensitivity and specificity.