Multimarginal optimal transport

Bohan - [Research Posts]

Multi-marginal Optimal Transport for Sea Ice Dynamics Prediction

Robertson Channel

B. Zhou and M. Parno [1] utilize a general multi-marginal optimal transport (MMOT) framework to obtain the continuous representation of discrete-in-time data. One possible application is the prediction on the sea ice dynamics. Given observations (called as marginals in the general framework) at different time-stamps, our method provides a prediction on the sea ice dynamics in a continuous time. This provides a solution to some stage in the Lagrangian Observation Mapping. The python package with its description can be found [2].

Using the SAR data (every 6 days) obtained from Alaska Satellite Facility on the Robertson Channel (thanks to our group member J. Park at Dartmouth), the video shows the prediction dynamics every 2 day. Even finer predictions are possible within a few minutes on a personal computer.

  1. Zhou and Parno, Efficient and Exact Multimarginal Optimal Transport with Pairwise Costs, 2022.
  2. Parno and Zhou, MMOT2D python package, 2022.
Last updated: 2022-Nov-05