Effective Cross-Continental Origin-Destination Generation

Develop a satellite-conditioned generative method that transfers origin-destination flow generation effectively across continents, including to geographically and morphologically diverse global cities.

Background

The paper evaluates RemoteCLIP, DINOv3, SatCLIP, and AlphaEarth within an otherwise fixed WeDAN graph-diffusion pipeline across United States counties, United Kingdom districts, and 14 global cities. Although some geographically grounded encoders improve transfer to the United Kingdom, no evaluated encoder achieves useful performance on the global-city benchmark; the best global CPC is only 0.122 for RemoteCLIP, while DINOv3 falls to 0.022. The authors therefore identify useful cross-continental origin-destination generation as an unresolved problem rather than a solved consequence of choosing a particular satellite vision encoder.

References

Third, no encoder transfers usefully to global cities (best CPC 0.122 for RemoteCLIP, 0.022 for DINOv3), confirming cross-continental OD generation remains an open problem.

Do Satellites See Commuters? A Critical Benchmark of Vision Foundation Models  (2609.00661 - Iqbal et al., 1 Sep 2026) in Abstract