Geography Reference
In-Depth Information
of substrate quality (and other parameters) for very long
reaches of a channel. However, methods such as the RHS
have some key limitations. First they require more or
less continuous access to the river banks. Whilst this
might be possible in many European rivers which now
flow through agricultural or urban landscapes, many key
salmon rivers of the world, notably in Scandinavia and
North America, flow through undeveloped forested areas
and, as a result, complete access to the entire channel
from the ground is often difficult. Another limitation
with ground based approaches is that they frequently use
visual appraisal methods in order to save time in the field
and allow for greater distances and channel lengths to
be sampled. Whilst this strategy does recognise the need
to sample rivers over increasing scales as advocated by
Fausch et al. (2002), the resulting data is hard to reconcile
with documented habitat preferences which are collected
by manual sampling of individual clasts. Furthermore,
the type of qualitative data collected by visual appraisal
methods is also incompatible with physical models of
sediment transport which could allow the prediction and
validation of habitats distribution at catchment scales.
There is therefore a clear need for substrate quantifica-
tion approaches which are capable of delivering accurate
and quantitative measurements of substrate with little or
no field effort over riverscape scales.
Such requirements can be fulfilled by remote sensing
approaches and this section examines the contribution
of the Geosalar project and others to the quantitative
assessment of sediment size with a focus on habitat.
ground survey which could then be analysed in order
to estimate sand content in a quantitative manner. The
problem that must be solved is the manner in which the
image information can be converted to a quantitative
measurement of sand area.
In remote sensing and image analysis terms, this is a
segmentation problem requiring the delineation of a fea-
ture in an image based on pixel properties. Once a feature
is segmented, its type can be attributed in the process
commonly called 'classification'. For example, a classic
classification application is the simple identification of
land-use types (e.g. forest, urban and/or agricultural) in
an air photo or a satellite image. Basic segmentation and
classification approaches rely on the brightness values in
the image and assume that an object or land-use of any
given type will most likely be of a specific colour or grey
level. For example, trees are assumed to be green, water
is assumed to be dark and sediment is generally light grey
with a slight reddish hue depending on exact composition.
However, in the case of sand identification, a close exam-
ination of ground imagery reveals that the solution may
not be straightforward. Figure 9.2a shows a small image
covering 20
20 cm of a sand patch with a few protrud-
ing gravels and cobbles. Visually, the identification of the
clasts in the image is quite natural. One could therefore
expect that the frequency distribution of pixels brightness
values should reveal a clear bimodal distribution with
one mode for the sand and a second mode for the clasts.
However, the interpretation of Figure 9.2b is much more
ambiguous. While this histogram does have two poorly
distinguished modes, the modal tails show considerable
overlap. Figure 9.2c shows the result of the application
of Otsu's segmentation algorithm (Otsu, 1979) which
was specifically designed for bimodal histograms. It can
clearly be seen that while the clast is effectively delineated,
many sand particles were also delineated. This is simply
due to the fact that colour, or in this case grey level, is not
the key distinctive parameter. Sand grains having a colour
similar to that of the clast will be falsely identified as clasts.
Given that lithology exerts a dominant control on clast
colour, the presence of sand grains with the same colour
as the clasts in any given river is highly likely. Carbon-
neau et al. (2005b) therefore hypothesised that another
image property could act as a better discriminator in the
classification process.
Close observation of Figure 9.2a clearly shows that sand
is characterised by its mixed colour. Grains of sharply
differing colour are in close spatial proximity. This leads
to the suggestion that the defining feature of a patch
of sand is not its average colour but the variability of
×
9.3.1 Superficial sanddetection
The detrimental effects of fine sediment in fluvial gravels
on salmonid habitat quality are well documented (e.g.
Chapman et al., 1986). During the incubation phase, it
has been shown that the presence of fine sediment within
the gravel matrix reduces the survival rate of the eggs
(Wu, 2000; Soulsby et al., 2001). Furthermore, Cunjak
et al. (1998) suggested that during the juvenile life stage,
sand deposition on the surface of the bed could block
access to the large interstitial voidspaces used by juvenile
salmon as shelter during overwintering. Clearly, from a
purely visual perspective, it is relatively easy to detect
the presence of sand on the surface of a gravel bar
and therefore it can be deduced that an image of sand
deposited on gravels will somehow have the information
which allows for the identification of the sand patch.
A method could therefore potentially be devised whereby
terrestrial photographs are taken in the field during a
Search WWH ::
Custom Search