Denoising and despiking ADV velocity and salinity concentration data in turbulent stratified flows

- An optimum filter has been designed, based on minimising the mean square error.
- Filter coefficients rely on empirical correlation for the data, averaged over time samples.
- Noisy parts of the frequency spectrum are attenuated and the desired parts are amplified.
- The method interpolates between data points in time, to increase the sampling frequency.
- The concept may be applied to estimate the missing data points in space.

Due to the presence of noise and spikes in velocity measurements of turbulent flow fields, understanding the flow pattern may be seriously affected by the spurious values for zones where certain flow behaviour is not necessarily expected. In a series of laboratory experiments the velocity data recorded, using Acoustic Doppler Velocity meters (ADVs), and the salinity measurements were noticeably noisy. In despiking and denoising the velocity data a linear correlation algorithm was established, which successfully lowered the noise levels and removed the spikes. For the assessment of the method, an autoregressive model was used to generate a clean velocity signal. The spikes were generated with a uniformly random time index and a Gaussian distributed value, where White Gaussian noise was added to this simulated signal. Assessment was also undertaken on the signals generated using a three-dimensional numerical model. To enhance the comprehension of the flow field, an interpolation method for producing the missing data has also been developed, which may be deployed to increase the sampling frequency, or to produce data for the spatial domain at locations which are not included in the measurements. For salinity a different strategy was applied where a moving average procedure was carried out, as the data did not suffer from spikes and exhibited almost a constant band of noisy fluctuations.