Learning to combine 12 radar receiver signals reduced wave-height estimation error in a 14-day record from Argentia, Newfoundland and Labrador. Senal Chandrasekara and Reza Shahidi describe the experiment in a 26 October preprint. It estimates sea conditions from observations, rather than forecasting future waves.[1]
High-frequency coastal radar sends radio waves along the sea surface and listens for energy scattered back by ocean waves. Arrays of receiving antennas provide different views of those echoes. Existing networks use the signals to map surface currents.[2]
Here the target is significant wave height: the average trough-to-crest height of the highest third of waves. It describes the sea state; individual waves can be considerably higher.[3]
The model learns three ways of weighting the receivers’ signals, ranks the resulting magnitudes and maps them to buoy readings. The first 66% of observations trained it; the remaining 34% supplied evaluation and setting selection.[1]
After smoothing, root mean square error was 0.192 metres, against 0.224 metres for fixed Bartlett–LOSE processing: a calculated 14.3% reduction. Correlation with buoy readings remained 0.949. The selected pipelines used different feature counts and averaging windows—12 samples for the adaptive model, 25 for the baseline—so the comparison concerns the tuned pipelines as a whole.[1]
The same evaluation observations helped select those settings, leaving no untouched final test. Both approaches also underestimated some wave-height peaks. A separate record would be needed to establish how well the reported gain carries over to new conditions.[1]
Sources & further reading
- Adaptive wave-height estimation from coastal HF radar: research preprint arXiv · 26 Oct 2026 · Accessed 30 Oct 2026
- High-frequency radar observations CeNCOOS · Accessed 30 Oct 2026
- Significant wave height National Weather Service · Accessed 30 Oct 2026