Harnessing Deep Learning to Improve Near‐Inertial Internal Wave Forecasts

Abstract Wind‐driven near‐inertial internal waves (NIWs) play a key role in ocean mixing, yet their intermittent nature, complex generation, and propagation processes pose significant challenges in achieving accurate predictions. In this study, we utilize a deep learning (DL) model to forecast NIW velocities. Trained on 193 days of hourly NIW velocity and wind data, NIWs over the subsequent 60 days are predicted. Compared to the classic slab model, the DL model achieves a 25–59% error reduction within a 48 hr forecast window. Notably, the DL model maintains a correlation exceeding 0.80 within the first 24 hr and can effectively extend NIW predictions to multiple depths below the mixed layer. Furthermore, the pretrained model exhibits strong generalization capability, achieving a correlation coefficient of 0.82 at an independent station and during a different observation period. These findings underscore the potential of DL‐based approaches to advance NIW forecasting beyond conventional methods.