Reconstructing global air cargo flows using demand disaggregation: A multi-objective optimization approach

Abstract

This paper presents a constrained calibration framework that reconstructs itinerary-level air cargo flows by disaggregating Origin-Destination (OD) demand across feasible road-air routing options under behavioral and operational assumptions. The model integrates observed leg-level payloads with detailed supply attributes, such as airline network structure, flight schedules, aircraft capacities, and first- and last-mile road-feeder services, to capture hub roles, carrier strategies, transshipment constraints, and catchment area effects. For each OD pair, we generate choice sets of feasible itineraries subject to transfer rules, hub sequencing, airport geography, and journey-time bounds. Itinerary attractiveness is determined by a constant-elasticity term that combines generalized time and schedule depth, and flows are assigned using an attractiveness-based allocation, while ensuring routing feasibility and capacity limits are enforced. Calibration is posed as a scalarized dual-objective non-linear optimization that balances accuracy in observed leg loads (via absolute deviation penalties) against over-allocation of capacity (via hinge penalties), yielding capacity-consistent reconstructions at the network scale. Applied to a large real-world schedule and capacity snapshot, the framework reproduces realistic leg loads and itinerary patterns, delivering interpretable insights, including load factors, hub throughput, transit-time distributions, indirect routing, catchment areas, and network imbalances. In practice, the model functions as a demand-to-itinerary disaggregation layer that (i) can feed downstream optimization, emissions inventories, and policy analysis, or (ii) can be embedded within a joint network-design loop in which capacity, timing, and disaggregation co-evolve. Validation against publicly available leg-level data and robustness analyses support the approach. In the absence of itinerary-level ground truth, results are interpreted as model-implied, feasibility-consistent reconstructions for decision support and scenario testing (e.g., capacity shocks or schedule changes).