This paper presents an interpretable graph neural network (GNN)-based real-time air traffic controller (ATCO) workload assessment framework to address the limitations of existing complexity metrics that fail to capture subtle operational factors beyond the sheer number of aircraft in increasingly congested airspace. It uses an attention-based model that predicts the number of upcoming clearances, which are instructions issued by controllers to aircraft, from interactions in static traffic scenarios. The key is to derive aircraft-specific workload scores by systematically removing aircraft and measuring their impact on model predictions. The proposed framework outperforms ATCO-inspired heuristics and provides more reliable scenario complexity estimates than existing baselines. This tool provides a new way to analyze and understand complexity sources by attributing workload to specific aircraft, which can be applied to controller training and airspace redesign.
