Geography Reference
In-Depth Information
water surface reflections. This case study was concerned
with two sources of error: the 'bed effect', related to the
visibility of the channel bed and the 'flare effect', caused
by light reflecting or glinting off the water surface. The
bed effect occurs when LSPIV similarity index is biased
by immobile bed particles. This situation arises when the
water is clear and shallow and when the concentration of
tracer particles at the water surface is low. Both of these
conditions frequently occur in the field during moderate
or low flows. The result is an error characterised by
a zero or near-zero flow velocity. The flare effect occurs
when the PIV algorithmmomentarily tracks water surface
reflections and interprets them as flow velocity vectors
having a magnitude and orientation that corresponds
with the characteristics of the flare. The result of this flare
effect is an error characterised as random noise. In this
section a filtering method is introduced that allows the
removal of erroneous instantaneous velocity vectors from
the record. The effectiveness of the filtering technique is
illustrated with a field study conducted in two contrasting
river environments in the United Kingdom and Canada.
surface within the camera field of view using a DGPS
(Leica RTK DGPS) on the River Wharfe and a total
station (Leica TC 307) on the St-Charles River.
16.4.3 Datafiltering
Georeferenced mean flow velocities were measured
approximately one centimetre below the water surface
over a period of 60 seconds using a Valeport current meter
(model 002) on the River Wharfe (n
14) and a Marsh-
McBirney current meter (model Flow Mate) on the St-
Charles River (n = 19). These measurements were used
as reference values for flow velocities at the water surface.
Instantaneous measurements of surface flow velocity
were obtained using LSPIV within interrogation areas of
40 × 40 pixels centered on each location in the image
where velocities were measured with a current meter. As
in the previous example, the cross-correlation coefficient
was used as the similarity index. The resulting instan-
taneous velocity vectors in pixels/s were orthorectified
using the control points of the water surface in order to
yield estimates of instantaneous velocities in cm/s (Fujita
et al., 1998).
Instantaneous velocities were filtered based on the
construction and analysis of the spatial probability density
function (PDF) (Figure 16.4a). For each interrogation
area, the spatial PDF was constructed by discretising
the velocity vector data space into 1
=
16.4.2 Fieldsiteandapparatus
The first field site was on the River Wharfe, a small mean-
dering gravel bed river located near Leeds, UK. The active
channel width of the studied section was approximately
10 meters and the maximum water depth was 1m. The
weather on the sampling day was cloudy and the water
was dark and turbid with abundant naturally occurring
tracing elements floating at the water surface (foam and
leaves)(Berube et al., 2004). No additional tracer material
was added to the water surface. The second field site was
on the St-Charles River, a larger gravel bed river located
near Quebec City, Canada. The active channel width
of the studied section was approximately 30m and the
maximum flow depth was 0.6m. The weather was sunny
and the water was clear. Due to an absence of natural
tracer particles, biodegradable shampoo foam was used
to increase seeding.
Oblique digital video records of the water surface were
obtained using a 3-CCD mini-DV digital video camera
(Sony VX-1000, 480 × 720 pixels) mounted on a 1.5m-
high tripod positioned on the river bank. The height of
the riverbank on the Wharfe and St-Charles rivers was
respectively 1.5 and 3m. The video records were obtained
for a period of 1min at 30 Hz on the River Wharfe and
2min at 15 Hz on the St-Charles River. Videos were saved
in mpeg4 compression format. At each site, a minimum
of five ground control points were measured at the water
1cm/sbinsand
computing the density of data points in each tile.
×
N xy
N t
PDF ( v x , v y )
=
(16.3)
where v x is the streamwise velocity, v y is the lateral
velocity, N xy is a count of the instantaneous velocity
estimates in the v x v y bin and N t is the total number of
instantaneous velocity estimates. Velocity estimates were
filtered by removing all values where the PDF was less
than a given percentile. In these tests a threshold value of
the 90th percentile was found to be sufficient to preserve
instantaneous velocity vectors due to natural turbulent
fluctuations of the flowwhile removing erroneous velocity
vectors created by bed and/or water surface flare effects
(Figure 16.4b). The result of this filtering procedure is to
isolate the primary peak of the spatial PDF from which
the mean velocity vector is then calculated.
16.4.4 Results
Significant correlations were obtained at both sites
between current meter and unfiltered PIV measurements
of
surface flow velocity (River Wharfe, R 2
=
.
0
76,
 
Search WWH ::




Custom Search