SATELLITE-DERIVED BATHYMETRY USING CONVOLUTIONAL NEURAL NETWORKS AND MULTISPECTRAL SENTINEL-2 IMAGES

Abstract. Satellite-Derived Bathymetry (SDB) has been used in many applications related to coastal management. SDB can efficiently fill data gaps obtained from traditional measurements with echo sounding. However, it still requires numerous training data, which is not available areas. Furthermore, the accuracy problem arises considering linear model could address non-relationship between reflectance and depth due bottom variations noise. Convolutional Neural Networks (CNN) offers ability capture connection neighbouring pixels non-linear relationship. These CNN characteristics make compelling be for shallow water extraction. We investigate of different architectures using window sizes band combinations. use Sentinel-2 Level 2A images provide values, Lidar Multi Beam Echo Sounder (MBES) datasets are as references train test model. A set in-situ subimage pairs extracted perform training. The compared transform applied two other study Resulting ranges 1.3 m 1.94 m, coefficient determination reaches 0.94. generated a size 9x9 indicates compatibility reference depths, especially at areas deeper than 15 m. addition both short wave infrared bands four visible improves overall SDB. implementation pre-trained provides similar results depending on conditions.