Identification of sequential changes and investigation of change patterns in dynamic spatiotemporal data
MetadataShow full item record
Spatiotemporal data generally undergo fluctuations. Identifying a sequence of changes in spatiotemporal data and investigating patterns of the detected changes are commonly required across many domains for analysis, prediction, and decision-making. However, most current analytical tools do not support both functions, so there has been a limitation in comprehensively analyzing dynamics in ongoing spatiotemporal data. This research suggests an integrated change analysis framework to support the overall analysis process, from detecting a sequence of changes to investigating patterns of the detected changes, as a prototype of a decision support tool. To identify a sequence of multiple changes in ongoing spatiotemporal data, where the spatiotemporal process is unknown, this research develops a recursive change detection procedure that applies a nonparametric cumulative sum method designed to signal a shift repeatedly. The recursive change detection procedure diagnoses and estimates the actual change-point following a signal of a shift and uses the mean and variance of all observations accumulated from the estimated actual shift as the subsequent in-control mean and variance for detecting the next change. After detecting a sequence of changes, the identified changes are interpreted as explicit change types by domain-dependent definition of the change types. Spatial and/or temporal patterns of explicit changes are visualized on a set of event bands by flexibly assigning controlled dimensions to event bands, stacks, and panels, which support investigating similarities and differences of explicit changes according to the controlled space and/or time. The proposed integrated analysis of sequential changes in spatiotemporal data is applied to assist emergency facility location planning coping with time-varying demand (i.e., emergency call) patterns. A total of 12, 11, 11, and 9 sequential changes in spatial patterns of the daily emergency calls in Buffalo are identified by the recursive change detection procedure for 1998, 1999, 2000, and 2001, respectively. Spatial and temporal similarities and differences of local clusters (hot spots and cold spots) and local outliers before and after each change are investigated to obtain valuable information for location planning. A set of scenarios describing different patterns of demands and probability of each scenario are specified by exploring spatial and temporal patterns of hot spots and cold spots on event band and stack structures. Timing of relocation is identified by investigating temporal similarities of spatial demand patterns contrary to dominant spatial patterns on set of event bands. Thus, useful inputs and strategies are provided for stochastic and dynamic location planning through the integrated analysis of sequential changes in emergency calls. In this context, the integrated change analysis framework proposed in this research has the potential to support policy decisions in various service fields where analysis of dynamics in spatiotemporal data is useful.