• A data-driven speed prediction approach is presented for non-recurrent events. • A novel data-driven congestion propagation prediction approach is presented. • The developed approach identifies the spatiotemporal impacts of incidents. • The proposed approach captures impacts of non-recurrent events on arterials. This paper proposes a speed prediction algorithm and a data-driven network congestion propagation approach that accounts for the spatiotemporal dynamic effects of non-recurrent events. Multi-year observed travel time, collision, and weather data provide essential insights into historical spatial and temporal patterns of road closures and their ramifications and are thus harnessed to discern the underlying traffic propagation dynamics. Unlike the common practice in the literature that focuses solely on the travel time variation of the link subject to non-recurrent events, the developed congestion propagation approach determines the time of occurrence, maximum propagation effect time, and flow recovery time by tracking how congestion moves over time and space across multiple road segments near the one experiencing an incident (i.e., impact area). The performance of the impact area speed prediction and congestion propagation framework is examined using historical INRIX travel time, weather, and collision data collected at the Calgary Road network. The results of the speed prediction model indicate that the developed speed prediction model can provide highly accurate link speed predictions during non-recurrent events. This finding is important as the accuracy of the proposed model is calculated during non-recurrent events, which display more complex traffic conditions than the recurrent events. The primary improvement of the developed speed and congestion propagation prediction models is its ability to capture multi-dimensional impacts of non-recurrent events in a road network, which is vital for road resilience, safety, and vulnerability analysis.
Esfe et al. (Fri,) studied this question.