Large-scale prediction of trip production is essential for origin–destination (OD) demand estimation and prediction. One of the main challenges in predicting trip production patterns lies in addressing spatial-temporal correlations and variations. Whereas many studies focus on temporal correlations, very few consider spatial adjacency between traffic analysis zones (TAZ) as explanatory variables. This research proposes a method that integrates a graph convolutional neural network (GCN) into a long short-term memory network (LSTM) to do exactly that. By introducing a nationwide graph that encodes the adjacency of TAZs, spatial heterogeneity is considered in the prediction process, and a single prediction model is trained for the entire network, thereby avoiding the need to train multiple separate models and potentially reducing overall training overhead, while increasing the prediction accuracy. Moreover, with this model, we investigate the effect of spatial scale on spatial uncertainty and prediction accuracy and analyze prediction errors, residual patterns, and their associations with socio-spatial features at different spatial scales. The findings of this research have important implications for improving OD demand prediction models and provide valuable insights into the role of spatial scale and socio-spatial features in travel demand prediction.

Road incidents, including accidents, greatly impact public safety, traffic flow, and overall transportation system functioning. Detecting and predicting incidents is crucial for effective incident management. Accurate algorithms rely on high-quality incident data sets. However, uncertainties exist due to the collection and recording process. To address this, cross-validating incident data with other datasets helps resolve inaccuracies. Additionally, enriching incident data with additional sources enables a more precise analysis of societal costs for planning purposes. In this study, we utilize traffic congestion data to examine and quantify the consequences of incidents on the Dutch highway network. First, we map match recorded incidents with related traffic patterns. Then, we label incidents as 'congestion' if significant congestion patterns were identified during or after the incidents or as 'no-congestion' if no significant congestion pattern occurred. For incidents labeled as congestion, we calculate and associate records with the congestion's duration, location, and Vehicle Loss Hours (VLH). The developed methodology has been implemented on five months of recorded data for the six most significant motorways in the Netherlands. This enriched dataset can be utilized for incident detection algorithms, analysis and management, and policy and decision-making.