Length-squared L-function for identifying clustering pattern of network-constrained flows

Abstract

The network-constrained flow is defined as the movement between two locations along road networks, such as the residence-workplace flow of city dwellers. Among flow patterns, clustering (i.e. the origins and destinations are aggregated simultaneously) is one of the cities’ most common and vital patterns, which assists in uncovering fundamental mobility trends. The existing methods for detecting the clustering pattern of network-constrained flows do not consider the impact of road network topology complexity on detection results. They may underestimate the flow clustering between networks of simple topology (roads with simpler shapes and fewer links, e.g. straight roads) but with high network intensity (i.e. flow number per network flow space), and determining the actual scale of an observed pattern remains challenging. This study develops a novel method, the length-squared L-function, to identify clustering patterns of network-constrained flows. We first use the L-function and its derivative to examine the clustering scales. Further, we calculate the local L-function to ascertain the locations of the clustering patterns. In synthetic and practical experiments, our method can identify flow clustering patterns of high intensities and reveal the residents’ main travel behavior trends with taxi OD flows, providing more reasonable suggestions for urban planning.

Keywords

• network-constrained flow
• clustering pattern
• network distance
• ripley’s k-function
• length-squared l-function