A space-time flow LISA approach for panel flow data

Abstract

Spatial flow data represent meaningful spatial interaction (SI) phenomena between geographic regions that are often highly dynamic. However, most existing flow analytical methods are cross-sectional, and there is a lack of methods to measure spatiotemporal autocorrelation of flow data. To fill this gap, we proposed a new localized spatial statistical method called Space-Time Flow LISA. The method design is a combination of two existing method families, namely space-time LISA and Spatial Flow LISA. A critical component of the method is the space-time weight matrix of flow data that blends pairwise spatial and temporal connectivities. We design three versions of the matrix, namely contemporaneous, lagged, and hybrid. We evaluate the method using both synthetic data and a case study of U.S. interstate migration from 2005 to 2017. The method is found to have high efficacy in finding spatiotemporal local autocorrelation patterns. Unlike the Spatial Flow LISA that tends to detect short-distance migration corridor havens (‘HH’ flows) and long-distance migration corridor deserts (‘LL’ flows), the Space-Time Flow LISA is less impeded by the distance between flow origin and destination, as they can pick up local patterns that are less spatially explicit but temporally dependent. In addition, the new method is able to detect time-sensitive patterns such as the outmigration from Louisiana forced by Hurricane Katrina in 2005. By integrating spatial, temporal, and attributive associations into a one-step analysis, the proposed Space-Time Flow LISA can illustrate the spatiotemporal structure of flow phenomena well, and reveal dynamic distribution changes over time.