Professor Wang Yaping's team has made new progress in high-precision coral reef bathymetric inversion research based on multi-source satellite data fusion and deep learning. Taking the Jiuzhang Atoll in the South China Sea as the study area, they proposed a novel bathymetric inversion model that combines Generative Adversarial Network (GAN)-based data augmentation with a stratified Convolutional Neural Network (CNN). This model effectively addresses the challenges of optical signal attenuation and insufficient training samples in deep water (>15 m). The relevant findings were published in the CAS Q1 top journal IEEE Transactions on Geoscience and Remote Sensing (TGRS) under the title: "Advancing Coral Reef Bathymetry: A GAN-Augmented and Stratified CNN Analysis of Fused ICESat-2 and Sentinel-2 Dataset."
Research Background
Although coral reefs cover less than 1% of the ocean floor, they harbor extremely high biodiversity. High-resolution bathymetric maps of coral reefs are fundamental for hydrodynamics and geomorphology studies in these areas and are crucial for ecological monitoring, conservation, and navigation safety in the context of global climate change. However, traditional ship-based or airborne measurements are costly and limited in coverage. Satellite-Derived Bathymetry (SDB) techniques based on multispectral satellites (e.g., Sentinel-2) and spaceborne LiDAR (e.g., ICESat-2) have gained significant attention in recent years. However, in waters deeper than 15 meters, the accuracy of traditional data-driven SDB models decreases significantly due to rapid optical signal attenuation and the scarcity of in-situ reference data. This bottleneck poses a serious challenge for fine-scale mapping of deep-water zones (15–30 m, often where biodiversity is most enhanced).
Research Results
Leveraging high-precision bathymetric data from ICESat-2 and multispectral imagery from Sentinel-2, the team innovatively proposed a deep learning bathymetric inversion framework integrating data augmentation and stratified modeling, successfully overcoming existing technical bottlenecks. Specifically, the study introduces GANs to strategically augment scarce samples in deep-water areas, effectively addressing the core challenge of insufficient training data for the deep-water segment. Furthermore, a stratified CNN was constructed to account for the varying optical attenuation at different depths, enabling the model to intelligently adapt to the complex optical environment from shallow to deep water.
The results demonstrate that this method produced a high-resolution bathymetric map of the Jiuzhang Atoll, which was well-validated against in-situ multi-beam echo sounder data. Validation against ICESat-2 samples yielded a correlation coefficient (R²) of 0.95, a Mean Absolute Error (MAE) of 0.75 m, and a Root Mean Square Error (RMSE) of 1.30 m within the 0–19 m depth range. Further external validation using independent multi-beam data showed that the stratified CNN model achieved an overall RMSE of 2.40 m and an MAE of 1.64 m, indicating strong application potential.
This study, by synergistically utilizing active LiDAR and passive multispectral remote sensing data, demonstrates the effectiveness of GAN-based data augmentation and stratified CNNs in improving bathymetric inversion accuracy in deep water. It provides new technical support for fine-scale coral reef topographic mapping, coastal ecosystem management, habitat monitoring, and disaster mitigation.
Article Information
Ziyao Chen from the School of Marine Science and Engineering, Nanjing Normal University, is the first author of the paper. Dr. Jin Li from East China Normal University and Professor Yaping Wang from Nanjing Normal University are the co-corresponding authors. The research was supported by the Young Faculty Research Innovation Capacity Support Project (ZYGXQNJSKYCXNLZCXM-O2).
Article Link
Chen, Z., Wang, L., Feng, W., Gu, Y., Li, J., & Wang, Y. P. (2026). Advancing Coral Reef Bathymetry: A GAN-Augmented and Stratified CNN analysis of fused ICESat-2 and Sentinel-2 dataset. IEEE Transactions on Geoscience and Remote Sensing. https://doi.org/10.1109/TGRS.2026.3659873