Restricted (Penn State Only)
Kim, Eun Kyeong
Graduate Program:
Doctor of Philosophy
Document Type:
Date of Defense:
December 12, 2017
Committee Members:
  • Alan MacEachren, Dissertation Advisor
  • Alan MacEachren, Committee Chair
  • Andrew M. Carleton, Committee Member
  • Donna J. Peuquet, Committee Member
  • Zhenhui Jessie Li, Outside Member
  • Clio M. Andris, Committee Member
  • bursts
  • burstiness
  • burstiness measure
  • exploratory spatio-temporal data analysis
  • spatio-temporal event analysis
  • contrail outbreaks
  • geo-located tweets
  • geo-tagged tweets
  • geographically weighted regression
  • GWR
  • exploratory data analysis
  • temporal regularity
  • temporal homogeneity
  • burst analysis
Exploring space-time structures is essential to understanding geographic events that are localized in space and time. The need to understand a range of space-time events has prompted development of many exploratory spatio-temporal data analysis (ESTDA) methods that include spatiotemporal cluster detection (e.g. space-time scan statistics). The focus of this dissertation is on extending ESTDA methods and their application to include attention to temporal and spatio-temporal burstiness of dynamic, geographical scale phenomena. This focus on burstiness complements and extends existing research on spatial and spatio-temporal clusters. The concept of clusters can be defined in multiple ways; many statistical methods to detect clusters statistically define a clustered pattern as one that rejects a completely random process (CRP). CRP is based on the density of events within specific space and time. The dissertation introduces a new concept of temporal clusters, ‘bursts’ that do not just reject CRP, but indicate a characteristic temporal pattern in which short term high activity periods labeled as ‘bursts’ alternate with extremely lengthy inactive periods. More specifically, this dissertation develops a new set of ESTDA statistics that characterize temporal burstiness, or frequency-invariant temporal regularity/irregularity, and applies the new statistics to analyzing geographic events including wildfires, geo-tweeting behaviors, and jet contrail outbreaks. In Chapter 1, I lay out the context for the overall work and describe the research goals and objectives. In Chapter 2, I review relevant literature on ESTDA statistics including point pattern analysis methods as well as the concepts and methods for burst analysis, and position the indicators of temporal burstiness presented here in the context of existing ESTDA methods. In Chapter 3, I develop a novel burstiness measure for event data with finite sample sizes, addressing the issue of small sample size effects of burstiness metrics as initially proposed in statistical physics. In Chapter 4, I propose a set of local indicators of temporal burstiness that integrates spatial containers filtering input data points into the burstiness measure developed in Chapter 3; the resulting methods are applied to analysis of wildfire data as an illustration of their utility. In Chapter 5, I apply the proposed methodology in a more comprehensive application, for characterizing temporal regularity/irregularity patterns of contrail outbreaks in the conterminous United States. Geographically weighted regression (GWR) is adopted to explore relationships between the upper-troposphere (UT) meteorological factors, temporal burstiness, and frequency of contrail outbreaks. The concluding chapter, Chapter 6, highlights contributions, challenges, and future work. Overall, the components of this dissertation provide a methodology to explore the temporal burstiness of geographic events and help reveal new temporal aspects of geographic phenomena.