Deep Bayesian Techniques to Nearshore Bathymetry with Sparse Measurements

Abstract:

Nearshore bathymetry, the topography of the ocean floor in coastal zones, has played a vital role in predicting the surf zone hydrodynamics and for route planning to avoid subsurface features. Hence, it becomes increasingly important in a wide variety of applications including shipping operations, coastal management, and risk assessment. However, direct high resolution surveys of nearshore bathymetry are rarely performed due to budget constraints and logistical restrictions. One possible alternative when only sparse observations are available is to use Gaussian processes regression, or Kriging. Nonetheless, it is often difficult for traditional methods to recognize patterns with sharp gradients like those found around sand bars or submerged objects, especially when observations are sparse. In this work, we present several deep learning based techniques to estimate nearshore bathymetry with sparse, multi-scale measurements. We develop a deep neural network to directly compute posterior estimates of nearshore bathymetry, as well as a conditional Generative Adversarial Network (cGAN) that samples from the posterior distribution. We train our neural networks based on synthetic data generated from nearshore surveys provided by the U.S. Army Corps of Engineer Field Research Facility (FRF) in Duck, North Carolina. We compare our methods with Kriging on real surveys as well as ones with artificially added patterns of sharp gradient. Results show that both direct estimations and sampling by deep neural networks give better predictions than Kriging. Finally, we propose a novel method that combines deep learning based models with Kriging, and shows further improvement of the posterior estimates.