1 min in the life of a river: selecting the optimal record length for the measurement of turbulence in fluvial boundary layers

In the analysis of instantaneous velocity measurements, it is important to consider carefully the properties of the instruments as well as the sampling frequency and the record length. Although several studies have examined the influence of instrument settings on velocity measurements in rivers, little information is available on the effects of the record length on these measurements and on the turbulence parameters that can be extracted from the time series. Decisions on the record length are often made ad hoc. Here, we quantify the effects of the sampling period on the precision of turbulence statistics (e.g., mean, standard deviation, skewness, Reynolds stress, quadrant analysis) obtained from velocity measurements in fluvial turbulent boundary layers. A bootstrapping method is used to determine the standard errors (S.E.) of turbulence statistics from 19 time series. The time series were sampled in three gravel-bed rivers using both Electromagnetic Current Meters (ECM) and Acoustic Doppler Velocimeters (ADV). The record lengths ranged between 20 and 60 min. The sampling frequencies were 20 and 25 Hz for the ECMs and ADVs, respectively. We show that, for most turbulence statistics, the optimal record length (minimum sampling effort to achieve low standard errors) ranges between 60 and 90 s. We discuss the implication of these record lengths for standard errors estimates and variance losses.