From Airspace Degradation to Deconfliction: Predicting Safety-Critical Traffic States through the Evolution of Multiscale Complexity Metrics

Growing air traffic demand is pushing airspace management towards the limits of its operational capacity, with steadily increasing workload for air traffic control officers, with direct consequences for safety. Understanding which traffic conditions give rise to safety-critical situations is therefore an essential open problem. Here we compare a large number of deconfliction events with equivalent conflict-free scenarios, extracted from thirteen weeks of air traffic control data over the Swedish Flight Information Region. We describe the traffic structure before each event through thirty sector-complexity metrics computed at multiple spatial and temporal scales, and use them as inputs of Machine Learning models. We find that the need of deconflictions can be predicted with an accuracy reaching $70\%$, but only when the traffic is described at the sector spatial scale. Predictability increases monotonically in the minutes preceding the clearance, showing that situations requiring deconflicting actions are not sudden occurrences, but rather the endpoint of a progressive degradation process. Additionally, the information needed for such prediction mainly resides in specific characteristics of the overall traffic structure, and not only in the conflicting pair. These results numerically demonstrate that safety-critical airspace states can be anticipated from the collective behaviour of the traffic, paving the way for early-warning systems that flag degraded, safety-relevant conditions before an intervention is required.