Spatio-temporal modelling of traffic accident risk in road networks: a point process approach

Traffic accident data provide an important basis for understanding road safety problems, and their spatio-temporal structure on a road network can be analysed within a Geographic Information System (GIS) framework by incorporating GIS-derived covariates into a spatio-temporal point process model to explain intensity variation. This study applies parametric spatio-temporal point process modelling to traffic accident data from Nganjuk Regency, Indonesia, to examine how accident intensity varies across space and time as a function of road geometric characteristics and climatic conditions. Spatial intensity is modelled using maximum composite Poisson and logistic likelihoods, while temporal intensity is analysed using Poisson and generalized Poisson models to accommodate overdispersion from unobserved temporal heterogeneity. The results show that the maximum composite Poisson likelihood performs better for the spatial aspect, while the generalized Poisson model is better for the temporal modelling. The best model indicates that distance to traffic lights and road type are the dominant spatial determinants of accident intensity on the linear network, whereas air temperature, humidity and rainfall significantly influence weekly accident intensity, indicating structured temporal variation. Integrating spatial and temporal components within the GIS framework reveals that accident risk arises from the interaction between road-network characteristics and seasonal environmental conditions, with persistent concentration along arterial roads and a mid-year peak.