The award-winning paper
What Matters for Sparse HF-Radar Current Reconstruction? Physics Constraints versus Temporal Context in Tokyo Bay

Abstract：
This study investigates how artificial intelligence can reconstruct coastal ocean surface currents when HF-radar observations are severely limited.

Using five years of Tokyo Bay HF-radar observations, the study compared conventional interpolation, machine-learning approaches, simplified physics-based constraints, and an AI model that uses recent temporal information. The results showed that incorporating recent observations substantially improved reconstruction accuracy under severe data sparsity. At only 1% retained observations, the temporal AI model achieved an RMSE of 0.0759 m/s, compared with 0.0943 m/s for an AI model using only the current observations and 0.1216 m/s for conventional interpolation.