ANNOUNCEMENTS
Accurate bathymetric data are critical for coastal management, marine navigation, habitat conservation, and catastrophe risk assessment. Traditional approaches like shipborne echo sounding and LiDAR, while accurate, are expensive, time-consuming, and have limited spatial coverage. This work investigates the use of Satellite-Derived Bathymetry (SDB) to estimate shallow water depths in the Lakshadweep Archipelago by combining multispectral Sentinel-2 imagery and ICESat-2 ATL03 photon data with machine learning approaches.
Sentinel-2 Level-1C data were atmospherically corrected using the Case-2 Regional Coast Colour (C2RCC) processor to obtain normalized water-leaving reflectance (RHOWN) values, particularly in the blue and green bands. ICESat-2 data were preprocessed by removing high-confidence photons, converting ellipsoidal heights to orthometric using the EGM2008 geoid, and applying a refraction correction to account for light bending at the air-water interface.
The study tested five machine learning models—Random Forest (RF), Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost), Neural Network, and a Hybrid model combining RF and XGBoost—at several test sites in the Lakshadweep Islands, including Minicoy, Kalpeni, Kavaratti, and the Agatti-Bangaram cluster. The Random Forest approach surpassed all other models in terms of accuracy and robustness, because to its ensemble nature and ability to simulate complicated nonlinear interactions. The trained RF model was then used to generate bathymetric predictions for 13 locations across the archipelago, yielding spatially continuous depth maps with great fidelity.
This study validates the viability and efficiency of applying machine learning to integrate Sentinel-2 and ICESat-2 data for SDB. The method offers a scalable and economical substitute for producing bathymetric data in isolated island settings where conventional surveys are not feasible. The results lend credence to the further use of RF-based SDB approaches in environmental monitoring, climate change impact assessments, and coastal and marine spatial planning.
Keywords: Bathymetry, ICESat-2, Sentinel-2, Random Forest, Satellite Derived Bathymetry (SDB).
