Reproducing Weekly Crash Patterns Using a Minimalist Driver Error Model

Abstract

Microscopic traffic simulation models are increasingly used for safety assessment, yet their ability to reproduce realistic crash patterns remains poorly understood. This study compares weekly patterns of real crashes at one of Berlin’s most accident-prone intersections (“Schlesisches Tor”) with simulated crashes generated using a simple “blackout” driver error model in SUMO. Building on prior work, we analyze both temporal trends and spatial distributions of crashes between observed and simulated data.The simulation was set up using averaged weekly traffic demand derived from multi-year municipal count data. The road network, imported from OpenStreetMap, was refined regarding lane geometry, lane assignments, and cycle paths. Since signal timing data were unavailable and varied over time, the traffic light phases were manually adjusted based on field observations. The accident model induces a 3-second loss of situational awareness during which drivers continue moving at constant speed, with probabilities calibrated for different traffic modes (car, truck, bicycle).Simulated crash patterns closely match observed weekly trends with distinct morning and evening peaks on weekdays and flatter curves on weekends. However, there is a slight overestimation of crashes on Mondays/Tuesdays and weekends, while daily temporal profiles align well.Spatially, simulated rear-end collisions cluster along main roads, matching observations. However, the simulation overpredicts turning-area collisions and underpredicts conflicts at pedestrian crossings, revealing a lack of pedestrian counts and possible limitations in modeling pedestrian behavior (e.g., hesitation, red-light crossing). These discrepancies highlight that while vehicle interactions are reasonably captured, pedestrian-vehicle interaction mechanics remain inadequately represented.In conclusion, this study demonstrates that a simple blackout-based microscopic model can reproduce both temporal and spatial crash patterns across different weekdays without complex behavioral rules. However, accurate modeling of vulnerable road users requires richer empirical data on pedestrian behavior.