Identifying multiscale spatio-temporal patterns in human mobility using manifold learning.

Area of Science:

• Human mobility and migration
• Network science
• Geospatial analysis

Background:

• Human mobility is fundamental to societal organization and adaptation to risks.
• Mobile phone Call Detail Records (CDR) offer vast data for studying mobility but present analytical challenges.
• Human mobility is multiscale, exhibiting diverse patterns in timing, magnitude, and spatial extent.

Purpose of the Study:

• To identify and characterize spatio-temporal scales and patterns of human mobility.
• To leverage advanced analytical techniques for extracting meaningful insights from large CDR datasets.
• To explore the applications of mobility pattern analysis in urban planning, infrastructure design, and hazard risk management.

Main Methods:

• Utilized Call Detail Records (CDR) data from the Orange mobile network in Senegal.
• Applied spectral graph wavelets, a manifold learning technique, for unsupervised data analysis.
• Employed dimensionality reduction to reveal underlying mobility patterns.

Main Results:

• Identified seasonal variations in human mobility patterns.
• Revealed mobility patterns linked to large-scale, short-term religious events.
• Demonstrated the effectiveness of spectral graph wavelets in uncovering multiscale mobility dynamics.

Conclusions:

• Manifold learning methods, such as spectral graph wavelets, provide novel insights into human mobility.
• Understanding human mobility patterns is critical for effective urban planning and infrastructure development.
• These findings have significant implications for hazard risk management, particularly in the context of climate change-induced migration.