AI workflows
AI-Driven MBES Roll Prediction
HydroStack has developed a deep learning approach that predicts a continuous roll angle signal per MBES ping directly from acoustic data. The model is designed to improve geometric reconstruction, reduce surface artefacts, and support more reliable downstream bathymetric products.
In practical terms, this capability:
- Predicts directly interoperable roll time series for standard workflows
- Improves depth geometry rather than only matching roll targets
- Supports QA validation against reference motion data
- Helps reduce rework on challenging datasets
By automating roll estimation directly from MBES data, teams can cut hours of manual correction and validation from each project, reduce costly reprocessing cycles, and move from raw soundings to delivery-ready surfaces significantly faster.
Conceptually, the model treats roll estimation as a data-driven inversion problem: inferring the motion signal that best explains the observed acoustic structure while preserving physical consistency in corrected depth outputs.
MBES roll prediction examples
These visuals illustrate pre-correction versus corrected results and show the predicted roll signal generated directly from MBES pings.