[Paper Review] Natural Scales in Geographical Patterns

Telmo Menezes, Camille Roth|arXiv (Cornell University)|Apr 4, 2017

Human Mobility and Location-Based Analysis1,130 citations

TL;DR

The paper endogenously identifies a small number of natural scales in human movement by applying scale-dependent community detection to geotagged Instagram movement networks and uncovering phase transitions in partition space.

ABSTRACT

Human mobility is known to be distributed across several orders of magnitude of physical distances , which makes it generally difficult to endogenously find or define typical and meaningful scales. Relevant analyses, from movements to geographical partitions, seem to be relative to some ad-hoc scale, or no scale at all. Relying on geotagged data collected from photo-sharing social media, we apply community detection to movement networks constrained by increasing percentiles of the distance distribution. Using a simple parameter-free discontinuity detection algorithm, we discover clear phase transitions in the community partition space. The detection of these phases constitutes the first objective method of characterising endogenous, natural scales of human movement. Our study covers nine regions, ranging from cities to countries of various sizes and a transnational area. For all regions, the number of natural scales is remarkably low (2 or 3). Further, our results hint at scale-related behaviours rather than scale-related users. The partitions of the natural scales allow us to draw discrete multi-scale geographical boundaries, potentially capable of providing key insights in fields such as epidemiology or cultural contagion where the introduction of spatial boundaries is pivotal.

Motivation & Objective

• Motivate the search for endogenous, meaningful description scales in geographical patterns.
• Propose a parameter-free method to detect phase transitions in scale space.
• Show that human mobility data contains a small set of natural scales across diverse regions.

Proposed method

• Construct scale-dependent movement networks from geotagged Instagram data by taking percentiles of the distance distribution.
• Apply Louvain community detection to each scale graph and smooth the resulting partitions into geographical boundaries.
• Measure partition similarity across scales with a Rand-index-based metric to detect breakpoints.
• Define intervals as natural scales via a breakpoint detection algorithm.
• Identify prototypical scales within each natural scale interval as the most representative partitions.

Experimental results

Research questions

• RQ1 Can endogenous natural scales be detected in scale space of movement-derived networks?
• RQ2 How many natural scales arise in diverse geographical regions?
• RQ3 What is the relation between natural scales and user mobility behavior?

Key findings

• Across nine regions, the scale space splits into 2 or 3 natural scales with similar internal partitions.
• Natural scales yield multi-scale geographical boundaries that reflect movement patterns.
• Prototypical scales within each natural scale interval effectively represent the interval’s partitions.
• Boundary maps at different natural scales reveal region-specific geographic and socio-spatial structure (e.g., Belgium’s language borders, urban cores).
• Most active users span multiple scales, but natural scales are driven by scale-related behaviors rather than scale-related users.