- 1The Ohio State University, Earth Science, columbus, United States of America (hsu.771@osu.edu)
- 2The Ohio State University, Earth Science, columbus, United States of America (moortgat.1@osu.edu)
Shallow water bathymetry is vital for understanding coastal ecosystems, managing marine resources, and monitoring environmental changes. However, global mapping remains challenging due to the limited penetration of optical and near-infrared light in water, which is rapidly absorbed and scattered by suspended particles and water molecules. Other electromagnetic frequencies, such as microwaves, do not penetrate deeply enough, rendering photogrammetric methods ineffective for underwater mapping.
High-accuracy methods like airborne LiDAR, sonar, and ICESat-2 (a spaceborne altimetric LiDAR) provide detailed bathymetric measurements but are limited by sparse spatial coverage and infrequent revisits. This study combines the strengths of airborne LiDAR and ICESat-2 data to train Machine Learning models for bathymetry extraction from Sentinel-2 multispectral imagery. Sentinel-2 offers global coverage, 10-meter resolution, and a ~5-day revisit cycle, presenting a scalable solution for large-scale mapping. Atmospheric corrections were applied to Sentinel-2 data, and ICESat-2 data were adjusted for tidal and refraction effects. Using Machine Learning models, we evaluate whether smaller ICESat-2-derived training datasets can achieve comparable accuracy to those trained on airborne LiDAR data, which provide a more comprehensive depth range.
In the past, correlations between the logarithm of Sentinel-2 blue-green band ratios versus depth has been widely used in bathymetric studies. We seek to improve prediction accuracy from optical imagery by incorporating other nonlinear relationships and leveraging additional spectral bands, allowing for more robust modeling across varying environmental and water conditions.
Our research underscores the complementary strengths and limitations of ICESat-2 and airborne LiDAR for bathymetric modeling and highlights the potential of Sentinel-2 for global, repeatable bathymetry. Achieving accurate and frequent mapping could revolutionize coastal monitoring, enabling applications such as disaster impact assessments and change detection after events like oceanic landslides, volcanic eruptions or earthquakes.
Keywords: Shallow water bathymetry, ICESat-2 ATL03, Airborne LiDAR, Sentinel-2, Random Forest, Coastal mapping
How to cite: Hsu, H. J. and Moortgat, J.: Enhancing Shallow Water Bathymetry Using Machine Learning with ICESat-2, Airborne LiDAR, and Sentinel-2 Imagery, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10185, https://doi.org/10.5194/egusphere-egu25-10185, 2025.