Abstract
Remote coastal inlets present significant opportunities for tidal energy development but are often underserved due to the prohibitive cost and logistical complexity of traditional site characterization methods. This paper introduces a hybrid framework that integrates sparse field measurements with one-dimensional (DYNLET) and two-dimensional (Delft3D) physics-based hydrodynamic models to characterize flow behavior in Kootznahoo Inlet, a remote site in Southeast Alaska. High-resolution spatial velocity and bathymetry data were collected over a condensed campaign using vessel-mounted ADCPs and RTK-GPS, enabling calibration of both models. To extend temporal reach, we implemented advanced deep learning models—Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Transformer networks—trained solely on tidal elevation data obtained by pressure transducers. The BiLSTM model showed superior short-term accuracy, while the Transformer model demonstrated strong long-range forecasting stability. These AI-predicted tidal elevations will serve as dynamic boundary conditions for future Delft3D simulations, eliminating the need for extended in-situ measurements. This integrated AI-physics strategy offers a scalable, cost-effective solution for high-resolution site assessment in data-scarce environments, bridging the gap between physics-based fidelity and data-driven foresight to accelerate the deployment of marine energy systems.