Dynamic estimation of airport taxiway hotspots using spatio-temporal traffic and kinematic risk modeling

Abstract

At busy airports, airfield operations are among the most safety-critical phases of flight because aircraft move through complex taxiway and runway-crossing networks. Despite ground-surveillance advances, taxiway conflicts remain a concern. We develop a traffic-based method for dynamic taxiway hotspot estimation that identifies conflict-prone locations without incident or accident histories. Unlike static FAA-defined hotspots, the method represents airfield infrastructure as a directed graph and uses agglomerative clustering to identify critical intersections. Hotspot scores combine spatial utilization probabilities from Runway–Gate routes with a KDE-based temporal overlap estimated from intersection arrival times. The trajectory-level score incorporates aircraft kinematics, including relative speed and convergence geometry. We validated the model using one full day of surface-surveillance data from Hartsfield–Jackson Atlanta International Airport (1 July 2019) and cross-validated it on an independent post-pandemic day (10 September 2024). The model identified clusters corresponding to all three 2019 FAA-designated hotspots within the northern study area, ranking two of them among its top five (Recall@5 = 0.67 vs. 0.33 for an aircraft-count baseline). Cross-year validation also identified both 2024 FAA hotspots as stable or rising high-risk clusters, and the method flagged conflict-prone locations not shown in FAA airport diagrams. Arrival and departure operations yielded distinct hotspot patterns: arrival risks clustered at runway exits, while departure risks clustered at runway-entry merge points. Sliding windows capture temporal changes, enabling dynamic estimation. By combining traffic-based hotspot detection, kinematic risk integration, operation-specific analysis, and a trajectory-level collision-risk score, the method bridges offline analysis and real-time conflict-risk monitoring.