Analysis of sentinel-2 image time series for enhancing satellite-derived bathymetry

This is the authors' abstract. We don't add key points for this paper.

Fine bathymetry is essential for oceanographic and geomorphological research. Traditional methods like shipborne sonar and airborne LiDAR have limitations in spatial coverage and high costs. In contrast, satellite images provide a scalable and cost-effective way to conduct large-area observations. Advanced algorithms for satellite-derived bathymetry (SDB) have been developed, including log-transformed single band (LSB), band ratio (BR), Lyzenga polynomial (LP), third-order polynomial (TOP), and back propagation (BP). However, clouds, waves, and sun glint make it challenging to select high-quality images. This study investigated the Anda Reef in China, Oahu Island in the U.S., and Wyborn Reef in Australia. We quantified and ranked cloud cover from Sentinel-2 time series images and analyzed three image fusion methods – mean synthesis, median synthesis, and maximum outlier removal synthesis – alongside five SDB methods – LSB, BR, LP, TOP, and BP. Results revealed that the BP outperformed the other SDB methods, achieving the lowest RMSE of 0.449 m. Fusing multi-temporal images using maximum outlier removal synthesis enhanced SDB precision, reducing RMSE by up to 1.367 m compared to mean and median synthesis. Additionally, fusing images before inversion proved more effective, yielding an RMSE reduction of up to 0.085 m.