ADV Data Analysis for Turbulent Flows: Low Correlation Problem

V. Martin, T.S.R. Fisher, R.G. Millar, M.C. Quick

Department of Civil Engineering, University of British Columbia,
Vancouver, B.C, Canada

Abstract
Downloaded from ascelibrary.org by Siti Fatin Mohd. Razali on 03/02/15. Copyright ASCE. For personal use only; all rights reserved.

This study was motivated by difficulties encountered when analysing ADV measurements
taken in turbulent flows in which correlations were often lower than 70% in the bottom 5 cm
of flow. The low correlations were attributed to an increase in turbulence for flows over
rough boundaries and this assumption was confirmed experimentally. The influence of the
ADV velocity range setting on correlation values was also examined. It was confirmed that a
higher velocity range setting resulted in higher correlations for measurements away from the
boundary, but in the near-bed region this was not always the case. Given the low correlation
for much of our data, it was necessary to determine the variability in average velocities and
Reynolds stresses for different degrees of filtering based on correlation. It was found that
average velocities were much less sensitive to the value of the correlation filter than
Reynolds stresses. In conclusion, for turbulent flows, ADV measurements could be used to
calculate average velocities and generate velocity profiles if measured data were edited based
on the correlation filter set to as low as 40%, provided that after filtering 70% or more data
were retained for analysis.

1. Introduction

Experimental work on gravel-bed armouring under clear water flow conditions has been
carried out in the Hydraulics Laboratory of the Department of Civil Engineering, University
of British Columbia, Canada. Velocity measurements have been made using the SonTek 16 MHz
MicroADV, to determine velocity profiles, shear velocities and roughness coefficients.

For the experiments with higher imposed shear stresses, the bed roughness increased and the
turbulence became more intense. Consequently, the velocity fluctuations became stronger
and the correlation decreased often below 70%, particularly in the lowest 3 - 5 cm of the
flow. The signal correlation is considered to be a parameter that indicates the quality of the
velocity estimate, and it is standard practice to edit and remove bad data points by setting the
correlation filter to 70% (SonTek, 1997; Lane et al., 1998). Based on this criterion and after
the removal of instantaneous velocity values with low signal correlations, there were usually
less than 70% data retained and these measurements were not used in other calculations. For
some experiments, all velocity measurements in the lowest 5 cm of flow would have to be
declined. Biron et al. (1998) demonstrated the importance of the inclusion of near-bed
measurements when determining the bed shear stress. Since the ADV data was to be used
for shear velocity determination either from velocity profiles or from Reynolds stress
measurements, “losing” data from the near-bed region represented a real concern. Thus, it
was essential to determine how much the signal correlation changed for different flow
regimes, and what would be the variability in average velocities or Reynolds stress values for
different degrees of filtering based on correlation.

Copyright ASCE 2004 HMEM 2002

Hydraulic Measurements and Experimental Methods 2002
 2. ADV Signal Correlation

Extensive explanations on ADV principles of operation can be found in SonTek Operational

Manual (1997), Lohrman et al. (1994), Lane et al. (1998), Voulgaris and Trowbridge (1998).
However, how the signal correlation is calculated and what it really means is not commonly
addressed. A detailed analysis of the ADV correlation coefficient can be found in SonTek
Technical Notes (1997), and it is not included in the SonTek Operational Manual. A
synopsis from this manuscript is given below.
Downloaded from ascelibrary.org by Siti Fatin Mohd. Razali on 03/02/15. Copyright ASCE. For personal use only; all rights reserved.

The velocities are estimated by using a technique called a pulse coherent Doppler processing,
where the ADV measures the change in phase of the return signal from two successive
acoustic pulses. The acoustic return is not a reflection from a single target, but a
superposition of the reflections from many individual particles contained in the sampling
volume. A phase coherency is achieved if all particles in the sampling volume maintain their
relative positions with respect to each other, in which case the strength and relative phases of
individual reflections would not change from one pulse to the next. However, Doppler noise
is an inherent part of Doppler-based volume backscatter systems, and there are other sources
of noise as well (detailed noise analysis is given in SonTek Technical Notes, 1997; and
Voulgaris and Trowbridge, 1998). Thus, the return signal from the second pulse is not a
phase-shifted reproduction of the first pulse, but contains a certain amount of noise, which
can be added to the coherent part of the return signal:

S1’ = S1 + N1 and S2’ = S2 + N2

In the above equations, S1’ and S2’ are the return signals from two adjacent pulses, S1 and S2
are the coherent parts of the signal, and N1 and N2 represent the random noise. The noise
adds a random error to the measured phases and to the measured velocities. If the coherent
signal power is compared to the total power, i.e. Si2 / (Si2 + Ni2), a measure of the noise is
obtained. It is this ratio that is commonly known as the ADV correlation coefficient and is a
direct measure of the Doppler noise in the velocity data.

In practice, the ADV estimates the phase shift between return signals from two successive
pulses using a complex auto-correlation function, which has a phase and a magnitude. The
phase is proportional to the distance the scatterers have travelled in the time between pulses,
and thus to the velocity. The magnitude is a measure of how similar the echoes from two
return signals are, and the normalized magnitude gives the correlation coefficient. If the
echoes from two neighbouring pulses were identical and only with a phase change, the
correlation would be perfect and equal to 100%. As the echoes become more dissimilar, the
correlation decreases.

Low correlations are associated with highly turbulent flows, air bubbles in the flow, low SNR
values, large velocity gradients within the sampling volume, the presence of large individual
particles, or interference from the boundary (SonTek, 1997; Lane et al., 1998, Finelli et al.,
1999; Goring and Nikora, 2002). In our experiments the most possible reasons for the signal
decorrelation are:

Copyright ASCE 2004 HMEM 2002

Hydraulic Measurements and Experimental Methods 2002
 - Turbulent flows - in such flows it is reasonable to assume that the particles in the sample
volume will not maintain their relative positions with respect to each other, thus adding to
the Doppler noise of the return signal. This will decorrelate the signal, and in this case
low signal correlations will not necessarily mean that the data is bad (also see Wahl,
2000b). In addition, if the turbulence eddies are of the same order as the sample volume
or smaller, the correlation will change even more, increasing the noise in velocity
measurements. The correlation coefficient for turbulent flows will be discussed in more
detail in the next section.
Downloaded from ascelibrary.org by Siti Fatin Mohd. Razali on 03/02/15. Copyright ASCE. For personal use only; all rights reserved.

- The presence of large individual particles - sound being reflected from a number of
different points close to the measuring volume confuses the instrument and decorrelates
the signal. The occasional saltating grains can be included in this category.
- Interference from the boundary - occurs when the return signal from the boundary
interferes directly with the return signal from the measuring volume, creating a “velocity
hole” at particular elevations above the bed, where noise masks the velocity signal (Lane
et al., 1998). The interference from the boundary can be recognized in the ADV output
file if the instrument attempts to change the velocity range setting. This deficiency of the
instrument can be somewhat corrected by changing the velocity range setting of the
instrument, because different settings experience the “velocity hole” at different
elevations above the bed. However, in some cases there were regions above the bed
when different velocity range settings would not reduce the noise, and those data points
had to be rejected. For this instrument and the given boundary roughness, the velocity
hole was experienced at about 1.5 to 3.5 cm above the bed.
- Large velocity gradients within the sample volume - the noise due to velocity shear
becomes important for elevations close to the boundary, i.e. y < 0.1 Y, which is
equivalent to the bottom 2 cm in our experiments (Lhermitte and Lemmin, 1994).
However, the noise values associated with this source are not significant and can be
ignored (Voulgaris and Trowbridge, 1998).

We consider the turbulence and the rough bed conditions to be the primary reasons for the
unexplained low correlations in our ADV data. The SNR values in all our measurements
were usually above 15 dB in the near-bed region, and sometimes decreased towards 10 dB
higher up above the bed.

3. Experimental Work

The experimental set-up consisted of a 15 m long, 0.5 m wide tilting flume. A tailgate at the
flume outlet was used to establish uniform flow at the beginning of each run. The test section
was 5 m long placed in the middle section of the flume. The test sediment sizes ranged from 1
mm to 32 mm, while for the developed armoured beds, D50 ranged between 4.7 and 11.1 mm,
and D84 ranged from 12.9 to 18.6 mm. There was no sediment feed in these experiments. The
velocity measurements were conducted after stable armoured beds developed and the sediment
transport ceased. For each experiment, velocity profiles were measured at a sampling rate of
50 Hz at several locations across and along the flume. These measurements were used to
calculate shear velocities (u*) and roughness heights (ks). The ADV data was also used to
obtain Reynolds stresses ( − ρ u ' w' ), which were used to plot the shear stress profiles and to

Copyright ASCE 2004 HMEM 2002

Hydraulic Measurements and Experimental Methods 2002
 estimate the total bed shear stresses (τ0), and the shear velocities on the bed (u* = (τ0/ρ)0.5).
These shear velocities were compared to shear velocities estimated from depth and slope
measurements, which were obtained with a point gauge.

To test the assumption that the correlation parameter for ADV measurements decreases with
increased turbulence, additional tests were carried out for flows over smooth and rough beds in
a smaller recirculating flume (width = 0.15 m, length = 6.0 m). Four velocity profiles for four
different boundary conditions and the same flow rates, were measured on the centerline of the
Downloaded from ascelibrary.org by Siti Fatin Mohd. Razali on 03/02/15. Copyright ASCE. For personal use only; all rights reserved.

flume (Fig. 1). Profile 1 was measured in a flow over a smooth bed 4.0 meters downstream
from the flume inlet (downstream from Profile 2 in Fig. 1a, and for the same bed conditions).
Low turbulence intensities and consequently high correlations were expected in this case.
Profile 2 was taken in a somewhat more turbulent environment for a flow over a smooth bed
where the velocities were measured just downstream from the inlet screen at the flume
entrance. For Profiles 3 and 4, small cobbles were placed on the flume bed in the middle
section of the flume (b-axis for cobbles equivalent to approximately 80 mm). Profile 3 was
taken above a cobble placed behind a couple of cobbles of the same size, while Profile 4 was
taken in the wake of a cobble placed with its long axis upright and somewhat upstream of
Profile 3 (Fig. 1b).

(a) (b)

Profile 1
Profile 2 Profile 4 Profile 3

Figure 1: Experimental set-up for testing the effects of turbulence on ADV signal
correlations (the arrows show the positions for the four profiles): (a) smooth boundary
conditions; and (b) rough boundary conditions.

The results of these measurements were analysed in terms of average correlations vs. distance
from the boundary for each point where the measurements were taken (Fig. 2). The data were
analysed using the free software, WinADV, Version 1.845 (Wahl, 2000a). Figure 2a clearly
demonstrates how average correlations decrease with measurements taken in increasingly
turbulent conditions. Profile 1 had average correlations of about 90% or higher for all distances
away from the boundary. The increase in turbulence caused by the screen for Profile 2 resulted
in lower correlations of about 20%. Profile 3 was measured above a rough boundary and this
change was expected to be followed by a further increase in turbulence. The average correlation
decreased to about 47% to 66% depending on the distance from the boundary. Profile 4 was
measured immediately downstream of an obstacle and the average correlations in this case were
only about 35 %. Overall, for all runs for flows over a rough boundary, the average correlation

Copyright ASCE 2004 HMEM 2002

Hydraulic Measurements and Experimental Methods 2002
 was below 70%, while for flows over a smooth boundary, the average correlation was 70% and
above.

Figure 2b shows the variation in average correlations for the velocity range setting of 250
cm/s. For flows over a smooth boundary (Profiles 1 and 2), the average correlation increased
for all points if compared to corresponding values of Fig. 2a. This result is consistent with
guidelines from SonTek (1997) that the correlation increases with a higher velocity range
setting, despite higher levels in instrument noise. However, for flows over rough boundaries
Downloaded from ascelibrary.org by Siti Fatin Mohd. Razali on 03/02/15. Copyright ASCE. For personal use only; all rights reserved.

(Profiles 3 and 4), the signal became more chaotic and for distances ≤ 3 cm from the
boundary the average correlation sometimes decreased. Overall, for the ADV velocity range
setting of 250 cm/s, the average correlations were higher than 70% for measurements taken
at 0.03 m or higher above the boundary, but below that depth it seemed that the velocity
setting of 100 cm/s gave more consistent results even though the correlations were lower.

(a) Velocity Range = 100 cm/s (b) Velocity Range = 250 cm/s
0.06 0.06
Distance from Boundary [m]

Profile 1 Profile 1

Distance from Boundary [m]

0.05 Profile 2 0.05 Profile 2
Profile 3 Profile 3
0.04 0.04
Profile 4 Profile 4
0.03 0.03

0.02 0.02

0.01 0.01

0.00 0.00
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Average Correlation [%] Average Correlation [%]

Figure 2: Correlation parameter for ADV measurements in flows over smooth and rough
boundaries for: (a) ADV velocity range of 100 cm/s; and (b) ADV velocity range of 250 cm/s.

These results confirm results from some other studies. For example, Voulgaris and Trowbridge
(1998) and Finelli et al. (1999) performed experiments for flows over smooth beds. Both studies
reported high correlation values, except for one point in the first study, for which the most probable
cause for low correlations was boundary interference. Another study by Nikora and Goring (2000)
was conducted in turbulent flows over rough beds, and they reported problems with low
correlations or low SNR values in the near-bed region, thus rejecting about 17% of their data.

4. Average Velocities and Reynolds Stresses

Considering that the signal correlations are often below 70% in the near-bed region in turbulent
flows over rough boundaries, numerical tests were performed to determine the sensitivity of
average velocities (Fig. 3a, c and e) and Reynolds stresses (Fig. 3b, d and f) for different levels of
filtering based on the correlation coefficient. In addition, the spike detection filter was tested using
the acceleration threshold of 1.5 g’s (Nikora and Goring, 1999; Wahl, 2000) and compared to the
values derived from the correlation filter.

Copyright ASCE 2004 HMEM 2002

Hydraulic Measurements and Experimental Methods 2002
 y=1.1cm y=1.1 cm
(a) (b)
100 100 3.5 100
90 90
3
80 80 80

Data Retained [%]

Data Retained [%]
70 2.5 70

-ρu'w' [N/m2]
Velocity [cm/s]

60 60 2 60
50 50
1.5 40
40 40
Downloaded from ascelibrary.org by Siti Fatin Mohd. Razali on 03/02/15. Copyright ASCE. For personal use only; all rights reserved.

30 1 30
20 20 20
0.5
10 10
0 0 0 0
0 20 40 60 80 100 0 20 40 60 80 100

y=3.0 cm y=3.0 cm
(c) (d)
100 100 3.5 100
90 90
3
80 80 80

Data Retained [%]

Data Retained [%]

70 2.5 70
-ρu'w' [N/m2]
Velocity [cm/s]

60 60 2 60
50 50
40 40 1.5 40
30 1 30
20 20 20
0.5
10 10
0 0 0 0
0 20 40 60 80 100 0 20 40 60 80 100

y=13.5 cm (f) y=13.5 cm

(e)
120 120 3.5 100
110 90
3
100 100 80

Data Retained [%]

90
Data Retained [%]

2.5 70
-ρu'w' [N/m2]

80 80
Velocity [cm/s]

2 60
70
60 60 50
50 1.5 40
40 40 1 30
30
20
20 20 0.5
10 10
0 0 0 0
0 20 40 60 80 100 0 20 40 60 80 100
Correlation Filter [%] Correlation Filter [%]

Velocity Reynolds stress

* % Data retained
Figure 3: Filtering based on the correlation coefficient (shaded symbols), or on the spike
detection filter with the acceleration threshold of 1.5 g’s (open symbols): (a), (c) and (e) are
the average velocities at a point; (b), (d) and (f) are Reynolds stresses.

Copyright ASCE 2004 HMEM 2002

Hydraulic Measurements and Experimental Methods 2002
 Data illustrated in Figure 3 were measured at different elevations (y) in the experiment which
had the highest shear stress and the highest roughness of the armoured bed, with D50 = 11.1 mm
and D84 = 18.6 mm. The velocity data appears to be less sensitive than the Reynolds stress to
the correlation filter. Furthermore, the velocity values calculated using the spike detection
filter (open symbols) were consistent with velocities derived using the correlation filter
(shaded symbols). However, the percentage of data retained (stars) decreased fairly rapidly
with the increase in the correlation filter value in the near-bed region, indicating more
complex turbulence structures. It appears that Reynolds stresses were much more sensitive
Downloaded from ascelibrary.org by Siti Fatin Mohd. Razali on 03/02/15. Copyright ASCE. For personal use only; all rights reserved.

to the percentage of data retained and that Reynolds stresses calculated after applying the
spiking filter were often quite different.

For determining the Reynolds stresses, data edited with the 40% correlation filter could be
included in the profile to determine the total bed shear stress if at least 70% of data are
retained for the analysis after filtering. However, this must be done with some caution,
because of the high variability and uncertainty identified in the Reynolds stress values
determined for different correlation percentages (Fig. 3b and d). It is possible that some of
the variability in Reynolds stress values might diminish with longer data acquisition, and
consequently longer averaging periods. Based on the above discussion, only data edited with
the 70% correlation filter for which at least 70% of data were retained, were used in
determining the Reynolds stress values in our experiments. For the points in the near-bed
region where average correlations lower than 70% were commonly experienced, the
Reynolds stresses decreased due to the larger roughness elements present on the bed. These
were also the data points edited with the 40% correlation filter and were not considered when
determining the total bed shear stress. The spike detection filter generally did not seem to
produce reasonable values for Reynolds stresses.

Including more data, even if the correlations are less than 70%, is preferable to using
averages based on very small number of data points. Comparison of the results using the
70% and 40% correlation filters indicated that the 40% data gave reasonable results for
velocities (Fig. 4a, triangles). Average velocities calculated using the 70% correlation filter
were often based on a fairly small number of data points (Fig. 4b, triangles), and were
probably not true representatives of the mean. The 40% correlation filter was found
preferable to the spike detection filter, because even though both methods estimated similar
velocity values (Figure 4a, circles), in the case of the spike detection filter with acceleration
threshold of 1.5 g’s, there were often less than 50% data left in the bottom 5 cm of flow (Fig.
4b, circles). To retain more data when using the spiking filter, the acceleration threshold was
varied between 1 and 3 g’s. The results were inconsistent, with some apparently realistic
points being rejected as spikes, while some obvious spikes were allowed through.

The velocity measurements in our experiments were analysed with the correlation filter set to
70% and the average velocities were kept for points for which there were at least 70% data
retained. For points with less than 70% data retained, the analysis was repeated with the
filter set to 40% correlation, and again the average velocities were kept providing that there
were more than 70% data retained. This made it possible to have velocity values closer to
the bed, which is important when determining shear velocities from velocity profiles.

Copyright ASCE 2004 HMEM 2002

Hydraulic Measurements and Experimental Methods 2002
 (a) 100 (b)
0.14
70% correlation filter
40% correlation filter
Spike detection filter 0.12 70% correlation filter

Distance from Boundary [m]

80
1.5 g's spike detection filter
0.1
60
Vx [cm/s]

0.08

40 0.06
Downloaded from ascelibrary.org by Siti Fatin Mohd. Razali on 03/02/15. Copyright ASCE. For personal use only; all rights reserved.

1:1
0.04
20
0.02

0 0
0 20 40 60 80 100 0 20 40 60 80 100
Vx [cm/s] (40 % correlation filter) Data Retained [%]

Figure 4: (a) Velocity values obtained using the 40% correlation filter vs. the 70%
correlation filter, or the spike detection filter with acceleration threshold set to 1.5 g’s; (b)
Percentage of data retained for different filtering methods vs. distance from the boundary.

5. Discussion

Low correlation values do not bias the mean velocity measurements, but only the short term
variability of the velocity data. For mean velocity measurements, correlation values as low
as 30% can be used, since averaging reduces the noise (SonTek Technical Notes, 1997).
However, this would assume that the measurement was long enough to obtain the true mean.
The measurements in our experiments were 60 - 65 seconds long at each point on the profile,
the length of measurements being constrained by the number of points taken in each
experiment (100-150). It appears that this time was either not long enough, or that the
sampling rate should have been even faster than 50 Hz to resolve the turbulence and obtain
better correlations.

Several studies examined Reynolds stress estimates from ADV measurements. Lohrman et
al. (1994 and 1995) measured Reynolds stresses between 5 and 15 cm above the bed for flow
velocities ranging from 5 to 80 cm/s, and got good agreements with LDV measurements.
They recognized that the Reynolds stress is not limited by the magnitude of the Doppler
noise, but by how well the Doppler noise is balanced in the three receive channels. They
concluded that at higher flows the importance of the noise term becomes negligible and that
the noise could be reduced by averaging, where the importance of Doppler noise diminishes
as the averaging period increases. It is not clear what were the bed conditions in these flume
studies (smooth or rough), or how long was the data acquisition and what were the
correlation coefficients (we assume better than 70%). In our experiments, for measurements
at similar depths, the average correlation coefficients were above 70%, thus the percentage of
data retained increased substantially, and the Reynolds stress values were fairly consistent
over a whole range of correlation filter values (Fig. 3f).

Copyright ASCE 2004 HMEM 2002

Hydraulic Measurements and Experimental Methods 2002
 Nikora and Goring (2000) conducted field measurements over rough beds (D50 = 13.4 mm,
and D84 = 22.1 mm). Reynolds stresses were maximum at about 5 cm above the bed and
linearly decreasing towards the water surface. They obtained reasonably good agreements
between shear velocities calculated from velocity profiles, or from Reynolds stresses, or from
depth/slope measurements. In our experiments, the Reynolds stresses were maximum at
approximately 5 cm above the bed for the highest shear stress runs, or closer to the bed for
runs with lower shear stresses. For all measured profiles, there was a region of linearly
decreasing Reynolds stresses towards the surface, which were extrapolated to the bed to get
Downloaded from ascelibrary.org by Siti Fatin Mohd. Razali on 03/02/15. Copyright ASCE. For personal use only; all rights reserved.

estimates of total bed shear stresses (τ0). The agreement between the shear velocities
calculated using the three methods was reasonably good (Ph.D. thesis in progress, Martin).

The spike detection filter in WinADV (Wahl, 2000) is still in an experimental stage, and
should be used with caution. It is interesting however, that the spike detection filter
produced similar results as the 40% correlation filter. Some recent proposed methods for
despiking the ADV data (Goring and Nikora, 2002) were not used in this study.

5. Conclusions

In the presented experiments, the mean flow velocity was of the order of 60 to 90 cm/s.
Based on the above discussion, it is recommended to set the ADV velocity range to 100 cm/s
for distances up to 3 cm above the boundary if taking measurements in flows over rough
boundaries. For distances higher than 3 cm, the ADV velocity setting of 250 cm/s
consistently gave better results. When analysing the measured data, it is common to apply a
filter for editing the instantaneous velocities based on the ADV correlation parameter.
Additional experiments demonstrated that in turbulent flows over rough boundaries, the
average signal correlation is often lower than 70%, especially close to the boundary. The
correlation parameter does not necessarily demonstrate the goodness of the measured data,
and the 70% correlation criterion should probably be lowered for removing dubious data for
these flows.

The average velocity values are reasonable estimates, even for data with lower average
correlations. The filtering criterion could be as low as 40% correlation, and yet still yield
consistent average velocities (within 3%), however there were cases where the differences
were as high as 10%. These differences were somewhat higher closer to the bed, where the
velocities are lower, and where a 1 cm/s difference in velocities represents a higher
percentage error than higher up in the flow.

The Reynolds stress estimates are much more sensitive to the value of the correlation filter,
and these data should be examined with more attention. It is important to have sufficient
time for data acquisition, so that averaging cancels the effects of the noise.

Copyright ASCE 2004 HMEM 2002

Hydraulic Measurements and Experimental Methods 2002
 6. References

Biron, P. M., Lane, S. N., Roy, A. G., Bradbrook, K. F., and Richards. K. S. (1998).
Sensitivity of bed shear stress estimated from vertical velocity profiles: The problem of
sampling resolution, Earth Surface Processes and Lanforms, 23, 133-139.
Finelli, C. M., Hart, D. D. and Fonesca, D. M. (1999). Evaluating the spatial resolution of an
acoustic Doppler velocimeter and the consequences for measuring near-bed flows,
Limnol. Oceanogr., 44, 1793-1801.
Downloaded from ascelibrary.org by Siti Fatin Mohd. Razali on 03/02/15. Copyright ASCE. For personal use only; all rights reserved.

Goring, D. G. and Nikora, V. I. (2002). Despiking acoustic Doppler velocimeter data, J.

Hydraul. Eng., 128 (1), 117-126.
Lane, S. N., Biron, P. M., Bradbrook, K. F., Butler, J. B., Chandler, J. H., Crowel, M. D.,
McLelland, S. J., Richards, K. S., and Roy, A. G. (1998). Three-Dimensional
Measurement of River Channel Flow Processes Using Acoustic Doppler Velocimetry,
Earth Surface Processes and Landforms, 23, 1247-1267.
Lhermitte, R. and U. Lemmin (1994). Open-channel flow and turbulence measurement by
high-resolution Doppler sonar, J.Atmos. Oceanic Technol., 11,1295-1308.
Lohrman, A., Cabrera, R. and Kraus, N.C. (1994). Acoustic Doppler velocimeter (ADV) for
laboratory use, Proc., on Fundamentals and Advancements in Hydr. Measurements and
Experimentation, C. A. Pugh, ed., ASCE, New York, 351-356.
Lohrman, A., Cabrera, R., Gelfenbaum, G. and Haines, J. (1995). Direct measurements of
Reynolds stresses with an acoustic Doppler velocimeter, Proc. IEEE Fifth Working
Conference on Current Measurement, 205-210.
Nikora, V. I. and Goring, D. G. (2000). Flow turbulence over fixed and weakly mobile
gravel beds, J. Hydraul. Eng., 126 (9), 679-690.
SonTek (1997). ADV Operation manual, software version 4.0, 109 pp. [Available from
SonTek, 6837 Nancy Ridge Drive, Suite A, San Diego, CA 92121.]
SonTek Technical Notes (1997). Pulse coherent Doppler processing and the ADV
correlation coefficient, 5 pp. [Available from SonTek, 6837 Nancy Ridge Drive, Suite A,
San Diego, CA 92121.]
Voulgaris, G. and Trowbridge, J. H. (1998). Evaluation of the Acoustic Doppler
Velocimeter (ADV) for turbulence measurements, J. of Atmospheric and Oceanic
Technology, 15, 272-289.
Wahl, T. L. (2000a). WinADV Version 1.845: A Windows-based viewing and post-
processing utility for ADV files; user's manuals and on-line help provided with the
software (downloadable from http://www.usbr.gov/wrrl/twahl/winadv/).
Wahl, T. L. (2000b). Analyzing ADV data using WinADV, Proc. Joint Conference on
Water Res. Eng. and Water Res. Planning and Management, Minneapolis, Minnesota.

10

Copyright ASCE 2004 HMEM 2002

Hydraulic Measurements and Experimental Methods 2002