Learning Optimal Drifter Release Locations for Surface Current Estimation in the Levantine Mediterranean

We address the problem of designing an optimal drifter‐release strategy in the Levantine Mediterranean in order to obtain accurate estimates of surface currents. Building on the assimilation framework of Issa et al. (2016), we generate synthetic drifter trajectories from an ocean re-analysis velocity field (Mediterranean Sea Physics Reanalysis. MEDSEA_MULTIYEAR_PHY_006_004) and assimilate them to quantify the improvement, or “gain,” in the reconstructed flow. For each hypothetical launch point we compute a time and space averaged gain, thereby producing a map that links initial drifter locations to the expected percentage correction of the background field. These gain maps are then used to train a machine-learning model based on a U-NET architecture, which learns to predict, from a given background velocity field alone, a spatial map of the anticipated correction associated with any drifter launching point. The resulting tool provides a fast surrogate for expensive observing-system simulation experiments and directly suggests optimal release locations tailored to the instantaneous flow. We compare our strategy with deployments based on random placement and on seeding along the unstable manifolds of the background flow.