New method improves wave height measurements from radar data
A spatially-adaptive neural-network approach reduces errors when estimating significant wave heights.
Researchers developed a new way to estimate significant wave height (SWH) from high-frequency (HF) radar data. This method uses a spatially-adaptive neural-network to improve accuracy.
The work extends a previous physics-guided linear ordered-statistics estimator (LOSE). This previous method mapped the largest rank-ordered magnitudes of Bartlett-beamformed voltages to buoy-observed SWH.
The new approach uses three trainable complex spatial filters. These filters are initialized around the buoy look direction. The spatial weights are optimized using gradient-based updates.
Significant wave height (SWH) is the measurement being retrieved. High-frequency (HF) radar is the tool used for data collection. Bartlett-beamformed voltages are the values mapped to buoy-observed SWH.
Testing used a 13.385-MHz, 12-receiver element WERA radar at Argentia, Canada. The adaptive configuration reduced root-mean-square error by approximately . It maintained a Pearson correlation coefficient at .
The researchers note that feature counts and smoothing windows were selected using the evaluation interval. Independent testing is needed to establish the generality of the improvement.
Key facts
- The adaptive configuration reduced root-mean-square error by approximately .
- The method maintained a Pearson correlation coefficient at .
- Testing involved a 13.385-MHz, 12-receiver element WERA radar at Argentia, Canada.