Analyzing travel destinations distribution using large-scaled GPS trajectories: A spatio-temporal Log-Gaussian Cox process

Under the increasingly serious city diseases, it is very important to deeply understand the spatio-temporal correlation between travel destination and urban built environment. The spatio-temporal point process has been widely used in many specific fields to explore the evolution of events. We employ a spatio-temporal Log-Gaussian Cox process model (LGCP) to analyze travel destination, which consists of three different components: a spatial component λ(s), a temporal component μ(t), and a separable log-Gaussian stochastic intensity field exp{Υ(s,t)}, to model and predict the destination. The spatial component and temporal component are modeled by the Poisson log linear regression model combined with covariates of interest separately. Spatial covariates including land use, demographics, travel origins and road network, and temporal covariates including the periodic function, weekdays and weekend variation, are selected as influencing factors to analyze the impact of building environment on travel destinations under different combinations of spatial and temporal covariates. Then, an extensible Markov chain Monte Carlo (MCMC) algorithm is applied to simulate Gaussian random field, and the fitted LGCP model can be used to obtain the prediction distribution of Gaussian random field beyond the last time point of data observation. In the experiment, taxi destination data collected in Shenzhen city May 6 to May 23, 2019 are used to validate modeling performance. The results in experiment shows the space–time positions of destination simulated by LGCP model are very similar to those observed. This study can provide strategy for urban planners and transportation administration to conduct reasonable policy to balance travel demand and public resources.