LSTM Prediction of Air–Sea Interaction
Visiting Researcher (Summer Intern), UT Dallas Flow Turbulence and Dynamics Lab
Advisor: Dr. Kianoosh Yousefi · June – August 2026
Overview
Wind over ocean waves produces turbulent airflow that is difficult to measure and expensive to simulate. Over the summer I trained sequence models to predict air–sea interaction dynamics from turbulent flow time-series data.
Method
- Developed LSTM-based deep learning models on turbulent flow time-series data.
- Researched and integrated dynamic mode decomposition (DMD) filtering of the data before training.
- Processed large time-series flow datasets for training and evaluation.
Results
| Metric | Baseline | With DMD | Change |
|---|---|---|---|
| Median R² | 0.66 | 0.81 | +22.7% |
| Mean R² | 0.44 | 0.64 | +45.5% |

After the summer the lab retained me part-time to continue this work; see CNN reconstruction of airflow over waves.