Deep learning techniques for nearshore and riverine bathymetry estimation using water-surface observations

Riverine bathymetry and near-shore bathymetry are of crucial importance for shipping operations, recreational safety, coastal management, military operations, and flood risk management. However, direct high resolution surveys of bathymetry are difficult to perform due to budget constraints and logistical restrictions. Several recent efforts have been made to use sparse bathymetry measurements and indirect observations to estimate high-resolution bathymetry profiles. However, these techniques are computationally expensive for large-scale problems. In this work, two deep learning techniques are used for riverine bathymetry and near-shore bathymetry estimation problems. In the first method, we use a deep autoencoder that takes the flow velocity measurements and boundary conditions for a riverine problem as input and estimate the river bathymetry profile. In the second method, we combine Kriging with a deep neural network, which has been known for its ability to recognize non-linear and complex patterns, to estimate near-shore bathymetry with sparse measurements. In order to demonstrate the accuracy of these techniques, we applied them to two synthetic test cases and compared the results with traditional methods in terms of accuracy and computational cost. Results indicate that these algorithms can provide accurate estimation for bathymetry profiles with a smaller computational cost compared to traditional methods.