Surfzone Topography-informed Deep Learning Techniques to Nearshore Bathymetry with Sparse Data
summary
Uses a conditional GAN trained on FRF Duck NC surveys (240 of 265 surveys, ~10,000 augmented samples) to interpolate nearshore bathymetry from sparse point measurements, producing sharper sandbar jumps than Kriging and reporting lower RMSE than Kriging on 15 held‑out surveys. Adds Gaussian perturbations for training, averages ~100 GAN realizations for a posterior mean, and combines the GAN estimate with residual kriging (iterating 2–3 times) to further reduce errors versus standalone methods.
AI-written summary · about
chapters (9)
- 0:00Introduction to talk
- 1:30Deep learning for bathymetry
- 4:01Kriging prior knowledge
- 7:42Data and training setup
- 9:14Comparison with kriging
- 11:45Combining neural networks and kriging
- 13:55Conditional generative adversarial networks
- 19:26Conclusion and future work
- 21:00Q&A session
related talks
44mRobust Estimation and Generative Adversarial Nets
41mLarge Scale Field Delineation
19mMultispectral Image Restoration with GAN
01h 00mTutorial on Deep Learning IV
52mDeep Generative Models And Unsupervised Methods For Inverse Problems
32mEnsembles of GANs as a Data Augmentation Technique for Alzheimer research
42mUsing Generative AI and Foundation Models to Predict Above Ground Biomass
18mDance of the Ancestors...