PhD · NTNU · spatial statistics
Classifying water masses in three dimensions
Adaptive sampling in 3-D: a vehicle porpoises through depth layers to classify river and ocean water, choosing each move by how much it reduces classification error.
- 3-DGaussian random field, updated in real time
- IEEE JOEJournal of Oceanic Engineering, 2023
The problem
A river plume is not a surface: fresh water lies on top of salt water, and the boundary between them tilts and folds with depth. Mapping it from one depth misses most of it.
The approach
- A 3-D Gaussian random field models salinity, updated in real time as the vehicle measures.
- The goal is not a map but a classification: is this water river or ocean? The design criterion is the expected reduction in misclassification, the excursion probability around the threshold.
- The vehicle porpoises (yo-yos) between depth layers, and each leg is chosen where it reduces that error most.
The method is published in the IEEE Journal of Oceanic Engineering (2023). A follow-up I co-authored, on efficient real-time 3-D sampling of a plume front, appeared in Frontiers in Marine Science (2024).