Assessing the Influence of Land Use Patterns and Temporal Granularity on Urban Travel Demand Forecasting Performance

Accurate forecasting of urban travel demand is critical for improving transportation planning and management. This study evaluates the influence of land use patterns and temporal granularity on forecasting performance. Using large-scale origin–destination data from Shenzhen, we develop an attribute-augmented spatio-temporal graph convolutional network (AST-GCN) that integrates point-of-interest (POI) information, holiday indicators, and temporal segmentation. Empirical results show that incorporating POIs and finer temporal granularity significantly improves prediction accuracy, particularly during peak commuting periods. Shorter time intervals better capture short-term fluctuations but require greater computational resources. These findings highlight the importance of integrating diverse data sources and balancing temporal resolution with computational efficiency. The study provides new evidence for data-driven travel demand forecasting and offers practical insights for urban traffic management and planning.