Modeling spatio-temporal patterns of behaviors in campus using portable GPS

Abstract

Recent technological advances in portable GPS allow us to understand the complex interactions between the built environment and human behaviors with massive spatial and temporal data. This study attempted to (1) develop spatio-temporal behavior models based on associations between behavior data from GPS and the built environment, and (2) represent predicted activity in an activity density map given a scenario of building construction. For collecting behavior data, 64 undergraduate students were recruited from September to November 2013, and randomly divided into either calibration or validation samples. The 24 layers of GPS data in each hour were represented in 18,630 of a 3m x 3m grid cell within study site. The built environment of study sites was characterized by occupancy types and distance to attractions. The Poisson models showed that temporal variation of pulling and push factors of behaviors. Pearson and Spearman correlations between activity density of calibration and validation samples shows moderate level of agreement between calibration and validation samples Moderate level of agreement between calibration and validation was found. Using the estimated coefficients from the Poisson models, spatio-temporal patterns of outdoor activities were predicted and represented in an activity density map when the construction of new engineering building are completed in year 2016.The proposed methodologies estimating activity density have shown a potential usefulness of this approach in urban planning. Larger sample sizes and more sophisticated models may enhance accuracy of behavior models.