River Monitoring with Ground-based Videography - Fluvial Remote Sensing for Science and Management - page 365

videos for the development of an automatic wood detec-
tion algorithm were taken from different positions in
the hydrograph of a flood on November 22 - 24, 2007
that reached a maximum flow rate just over the bankfull
(Q

16.5 Case 3 - At-a-point survey of wood
transport

16.5.1 Introduction

550m 3 /s). =

Wood is an integral component of river systems that has
a strong influence on stream ecology, sediment transport,
and geomorphology (Bisson et al., 1987, Montgomery
et al., 1996, Gurnell and Petts, 2006). It also represents
a risk for flooding and human infrastructure (Bradley
et al., 2005, Comiti et al., 2006). There is a need for
data to calibrate wood budgets and examine processes
related to the transport of wood during floods (Benda
and Sias, 2003, Hassan et al., 2005). Existing methods
that use tagging or repeat surveys of stored wood volumes
are ineffective and labour intensive in large rivers and
do not monitor the transport or accumulation of wood
during floods when access to the river is most hazardous
(Lyn et al., 2003, Moulin and Pi egay, 2004, MacVicar
et al., 2009). This section describes field tests of the use
of videography to detect and measure floating wood
transported on the surface of a river.

16.5.3 Manualdetectionandmeasurement
To test and develop an automatic wood detection algo-
rithm, it was necessary to manually scan videos and
visually identify floating wood. A semi-manual algo-
rithm written in Matlab ™ was developed to assist with
this procedure. Using the semi-manual algorithm, video
playback was stopped by the user when wood was visually
detected. The end and side points of wood pieces were
recorded using the screen cursor and the pixel locations
transformed into real coordinates using Equation (16.1).
The length, diameter, and mean position were computed
from the real coordinates. Wood volume was calculated
by assuming a cylindrical shape for the wood pieces. The
presence/absence of roots and branches were also noted.
The video was then advanced several frames and the
endpoints of the wood were relocated using the screen
cursor to allow the calculation of velocity and rotation
(Figure 16.6). The number of manually detected wood
objects was assumed to represent the actual number of
wood objects in the river. This ignores any submerged
wood, but the objective at this stage was to test an
automatic detection algorithm and the manual detection
procedure was felt to be the best data to assess missed or
false detections.
Preliminary results from the semi-manual algorithm
demonstrate the non-linear response of wood transport
to flow rate (Figure 16.7). A transport threshold exists
at approximately Q = 200m 3 /s, below which negligible
wood transport rates occur on the rising limb of the flood.
The peak inboth the frequency and volume of wood trans-
port occurs before the peak of the flood ( Q

16.5.2 Fieldsiteandapparatus
The camera was installed with a view of the Ain River, a
piedmont river with a drainage area of 3500 km 2 near the
town of Chazey sur Ain (France). Bank erosion and wood
mobilisation has been extensively studied in this river and
a high frequency of wood transport events were expected
(Pi egay, 1993, Pi egay et al., 1997, Lassettre et al., 2008).
The particular location was chosen because a gauging
station is already present and equipped with internet and
telephone access. In addition, the site is located on a high
embankment on the outside of a bend where the majority
of wood was expected to pass with an unobstructed view
of the water surface. The river is approximately 70mwide
at this location.
An Axis 221 Day/Night ™ camera was installed at
the gauging station on the Ain River in early 2007. A
Profiline ™ infrared light projector was also installed to
increase luminosity at night. Videos are transferred via the
web to remote servers and saved in 15 minute segments
in mjpeg format. The image resolution is 640x480 pixels
and the image frequency is 5 Hz. Video was recorded
during 12 floods between May 2007 and February 2009.
Of the recorded floods, five events were at or greater
than the bankfull discharge (Q bf

425m 3 /s).
Transport rates on the falling limb are much lower than
those on the rising limb and transport appears to be
negligible below a flow rate of approximately 420m 3 /s.
This result matches with an expectation that wood debris
is picked up from the bars, banks and overbank areas as
flood stage rises. These results indicate that this effect of
hysteresis is very strong and the volume of transported
material on the falling limb of the hydrograph is almost
negligible in most videos. A larger sample size is required
to accurately characterise the variability of this complex
process. However, the time required to analyse videos
using the semi-manual program is prohibitive and a fully

∼

530m 3 /s), the largest
of which occurred in April 2008 when the discharge
was twice the bankfull rate (Q

=

1080m 3 /s). The sample

=