Uncertainty-Guided Prediction Limits of Phase-Resolved OceanWave Forecasting under Data Sparsity: Experimental andNumerical Evaluation

Accurate short-term wave forecasting is critical to floating marine structures' efficient and safe operations as well as their active control using models (e.g., digital twins) that rely on real-time, phase-resolved ocean wave loading. For moored, i.e., quasi-stationary systems informed by environmental sensors (e.g., waverider buoys, wave-sensing LIDAR), challenges arise when upstream sensor data is missing, sparse, or phase-shifted due to drift. This study investigates the resilience and performance of machine learning models—specifically TiDE and LSTM—for forecasting phase-resolved ocean surface elevations under various scenarios of data loss and uncertainty. We introduce $\tau$-trimmed models, which dynamically adapt the prediction horizon based on uncertainty thresholds derived from historical forecasts. Numerical wave tank (NWT) and wave basin experiments are used to benchmark model performance under short- and long-term data masking, spatially coarse sensor grids, and upstream phase shifts. Results show that $\tau$-trimmed models consistently reduce forecast errors and uncertainty, particularly under degraded conditions. While LSTM models achieve lower average errors, TiDE models—especially the conservative variant—exhibit greater robustness and efficiency. Phase shifts in upstream data are shown to be more detrimental than masking alone. These findings inform the design of resilient, uncertainty-aware wave forecasting systems suited for realistic offshore sensing environments.