Spatial Autoregressive Model for Estimation of Visitors' Dynamic Agglomeration Patterns Near Event Location.

Area of Science:

• Urban dynamics and mobility studies
• Geospatial data analysis
• Computational social science

Background:

• Ubiquitous mobile computing generates massive traffic data for understanding collective movement.
• Crowd formation and dispersal in populated areas are critical aspects of urban dynamics.
• Previous models often lack the granularity to capture dynamic agglomeration patterns.

Purpose of the Study:

• To develop a model for analyzing dynamic visitor agglomeration patterns at events.
• To utilize aggregate Global Positioning System (GPS) location data for crowd movement analysis.
• To enhance the understanding of collective human mobility in social spaces.

Main Methods:

• Collected aggregate GPS location data from mobile phones at a 250 m spatial resolution.
• Developed spatial autoregressive models incorporating two-step adjacency matrices.
• Applied models to represent visitor movement between geographic grids around an event site.

Main Results:

• The proposed spatial autoregressive models demonstrated a higher goodness-of-fit compared to models without spatial or temporal autocorrelations.
• The models effectively represented visitor movement dynamics within the studied event area.
• A significant decrease in predictive accuracy was observed when using estimated prior-period endogenous variables.

Conclusions:

• Spatial autoregressive models with adjacency matrices are effective for analyzing dynamic crowd movement patterns.
• Incorporating spatial and temporal autocorrelations significantly improves model fit for visitor agglomeration.
• Model performance for prediction is sensitive to the accuracy of historical data inputs.