Applications of Close-range Imagery in River Research - Fluvial Remote Sensing for Science and Management - page 336

channels (bankfull width 20 m), riparian tree canopy

can hinder the survey of both the channel bed and banks.
They acquired sequences of overlapping images from a
vertically mounted non-metric camera suspended 10 m
above the river by a unipod. This methodology allowed
the generation of very accurate DEMs with a nominal
ground resolution of 3 cm.
Other studies have focused on monitoring river bank
erosion (Lawler, 1993; Barker et al., 1997; Pyle et al.,
1997). These authors used terrestrial digital photogram-
metry to reconstruct the bank profile through a high
resolution DEM. For example, Pyle et al. (1997) obtained
a spatial resolution of 2 cm, with an accuracy approxi-
mately equal to 1 cm, using oblique stereo pair images
acquired about 15 m from the eroding bank. A great
advantage of this technique is that it provides bank sur-
veys at a very fine scale, avoiding the errors due to the
interpolation process, as well as avoiding physical contact
with the unstable slopes of the bank.
When acquired from a helicopter, oblique photos
can provide qualitative information (e.g., bank char-
acters/state) all along a channel course which is not
possible by any other airborne/ground device (Piegay
and Landon, 1997). Some researchers have used archived
images and compared them with recent images to detect
decadal landscape changes (Debusshe et al. 1999; Start
and Handasyde, 2002). These authors compared a series
of photographs, mostly taken between 1952 and 1990
to highlight changes in riparian vegetation characteris-
tics due to the construction of dams. Cossin and Piegay
(2001) used oblique photos to provide a typology of
reaches according to riverscape character along a river
corridor (see details in Chapter 18).
Close range imagery can also be applied to flow velocity
measurements using particle image velocimetry (Hauet
et al. 2006). This topic is addressed in Chapter 16.

< vertical ground-based images and the rectification of

oblique images.

15.3.1 Analysisof vertical images for
particlesize

The most common methods of measuring the grain-
size distribution of surface sediments are variants of
the Wolman (1954) procedure, where grains are sam-
pled on a grid. Recent years have seen several attempts
to develop and apply image-based analysis methods of
grain-size analysis that substantially reduce field time
(and thus cost) and do not disturb the sediment being
studied. Two general approaches have been taken. The
first utilises geostatistical techniques and empirical cali-
bration to determine ensemble grain-size characteristics
from a photograph of the sediment surface (empirical
approach). The second is based on automated recogni-
tion and measurement of individual particles within an
image (object-based approach).
Empirical approaches are based on statistical measure-
ments of spatial variation in pixel intensity values, with
high-frequency variability in intensity being indicative
of small grains. The statistical measurements may be
used to derive ensemble grain-size parameters (e.g. mean
grain size) by empirical calibration against field-measured
grain-size parameters. The most successful approaches
employ measurements of the local semivariance (e.g.
Carbonneau et al., 2004) and autocorrelation (e.g. Rubin,
2004; Buscombe and Masselink, 2009), both of which
work in the spatial domain. Most recently, Buscombe
et al. (2010) have utilised the spectral properties of images
to estimate the median grain size, extending the spatial
autocorrelation approaches into the frequency domain
and requiring little calibration.
Empirical approaches have not been used widely for
ground-based assessment of fluvial sediment surfaces,
although all are potentially applicable. Local semivariance
has primarily been applied to images collected from aerial
platforms. Autocorrelation has most commonly been
applied to beach sediments, although it is increasingly
being developed for fluvial surfaces (e.g. Warrick et al.,
2009; Buscombe et al., 2010). The principal limitation of
empirical techniques is that they do not provide complete
information about the grain-size distribution, being lim-
ited to estimates of particular grain-size percentiles, for
which they have been shown to provide robust estimates
(e.g. Warrick et al., 2009; Buscombe et al., 2010).
Object-based approaches use image processing meth-
ods to identify the boundaries of individual grains in

15.3 Post-processing

The post-processing workflow of the acquired images
may involve a wide range of tools and procedures,
with different degrees of complexity depending on the
aim of the investigation. Simple, qualitative observations
are sometimes sufficient; for example when the process
recognition is the main aim of the survey. In other
applications, particularly when the objective is to extract
quantitative information from the images, more sophisti-
cated analyses are required. Here we consider approaches
to obtaining grain size and particle morphometry from