Assessing the advantages and limitations of Coastal Depth Estimation using Machine Learning techniques

<p>Bathymetry plays an important role in ship navigation and as an input to numerical models that carry out the accurate estimation of various coastal processes such as storm surge, associated coastal flooding, and sediment dynamics like erosion-accretion of coastline. However, in-situ measurements of depths become challenging by virtue of the acquisition costs involved in deploying conventional methods, such as ship-based echo-sounding and LiDAR based techniques. Resultantly, several studies have been conducted related to Satellite Derived Bathymetry estimation using Optical Remote Sensing and empirical approaches, and most recently Machine Learning (ML) techniques. The advantage of ML methods is that they account for the non-linear relationship that exists between depths and reflectance in complex coastal environments. The present study evaluates the relative performance of 13 different linear and non-linear ML approaches, (1) Least absolute shrinkage and selection operator (LASSO) (2) Least-angle regression (LARS) (3) LASSO-LARS regression (4) Automatic relevance determination regression (ARD) (5) Bayesian ridge regression (BRR) (6) Multilayer perceptron (MLP) (7) K-nearest neighbors (KNN) (8) Support vector regression (SVR) (9) Random forest regression (RF) (10) Extra Trees regression (ET) (11) Gradient boosting regression (GBR) (2) Bagging regression (BR), for depth estimation using Landsat 8 OLI imagery in Labuan, Malaysia. The estimated depths are compared with in-situ measurements and various descriptive statistics are reported. It is observed that for all the study areas, ET, KNN, RF and BR consistently provide better results in comparison to other algorithms. The primary aim of the study is to highlight the best available ML methods that can be used with Medium Resolution satellite imagery to derive bathymetry as well as discussing the pros and cons of using ML for coastal bathymetry estimation .&#160;</p>