Geography Reference
In-Depth Information
For example, an image with 10 cm resolution can be used
to map sediments of 10 cm or larger. In turn, this means
the image cannot be used to accurately characterise the
full distribution of sediment sizes (unless the finest sed-
iments are 10 cm in size), because the smaller sediments
are not being 'seen' by the sensor. Managers applying
the technique with panchromatic or colour imagery will
almost certainly need to charter special flights to collect
imagery at the resolution of the sediment size of interest.
For example, Carbonneau et al. (2004, 2005) chartered a
helicopter to fly at 155 m above the river to acquire 3 cm
resolution imagery.
If multispectral or hyperspectral imagery are available it
may be possible to map features smaller than the nominal
image pixel size. Rainey et al. (2003), for example, used
spectral mixture analysis to map fine sediment sizes in
an estuary. This type of 'pixel unmixing' is accomplished
based on knowledge of the spectral characteristics of the
different pure end members present within a pixel. For
example, it is easy to estimate how much black and
white paint are mixed together to create a grey colour
in a single pixel; the darker the grey, the larger the
proportion of black paint. Rainey et al. (2003) used the
optical theory that small spaces between fine grains act
as blackbody cavities. The black body behaviour fills in
the spectra for each pixel to a degree proportional to
the grain size, enabling estimates of the proportion of
sand and mud, even in pixels that were 1.75 m in size
(Table 2.2). Unfortunately, the optical theory only works
well in relatively dry sediments. As of yet, there are no
techniques for measuring sediment smaller than the pixel
size in submerged sediments. This means that issues such
as sand embededness cannot at present be characterized
with airborne imagery.
algorithms is also relatively straightforward, provided the
features are not in deep shadow.
Biotypes are receiving increasing attention from stream
managers. In Europe, biotype mapping is a mechanism for
defining biodiversity in streams under the Water Frame-
work Directive (Dodkins et al., 2005). In the United
States, many agencies use in-stream habitats to char-
acterise stream health. A number of schemes exist for
defining the different kinds of biotypes, leading to confu-
sion around definitions; one person's glide can be another
person's run. A good summary of the different biotype
classification schemes and their overlap is provided in
Table 1 of Milan et al. (2010).
Despite their growing importance, there is relatively
little research on mapping biotypes with remotely sensed
optical imagery. Researchers have achieved good results
mapping in-stream habitats using supervised classifica-
tion, a classical technique that is included on all remote
sensing software. To apply this technique, the user first
identifies a number of pixels, called training sites, on the
image that are characteristic of each class of interest (e.g.,
riffles, pools, glides). The algorithm then maps other pix-
els on the image that have spectral signals similar to the
training sites. Remotely sensed biotype maps can be more
precise than ground-based maps. Ground surveyors will
often lump many small features into a larger adjacent
features (e.g., lumping low velocity stream margins into
the riffle that dominates that stream reach). If the pixel
size is small enough, the imagery will differentiate these
many smaller extent variations.
Legleiter et al. (2002) and Marcus (2002) found that
mapping accuracy is sensitive to the number of spectral
bands used for classification. Simply put, more bands
are better. Hyperspectral imagery (discussed in another
chapter in this volume) thus provides better results than
multispectral imagery (Marcus, 2002), and imagery that
includes short wave infrared is better than imagery that
only spans the visible wavelengths (Legleiter, 2003). The
spatial resolution of imagery relative to the size of the
channel being mapped is also a crucial consideration.
Marcus et al. (2003) documented higher mapping accu-
racies with 1-m imagery in a 5th order stream than in
a 4th order stream, which in turn had higher accuracies
than in a 3rd order stream (Table 2.3). The drop in
accuracy results from the 'mixed pixel' problem, which
occurs when one pixel encompasses multiple features. In
the stream context, a 1-m pixel in a third order stream
is more likely to include portions of two units (e.g., a
glide and pool) than in the 5th order stream, where the
biotypes are much larger.
2.8 Biotypes (in-stream habitat units)
Biotypes, also called 'in-stream habitats', 'micro-habitats'
or 'morphologic units', refer to features such as riffles,
pools, glides, and exposed bars. From an optical perspec-
tive, these features vary in a number of important ways
that can be captured by remote sensing. Optical variations
associated with depth, surface turbulence, substrate size,
and vegetation associated with the units enable differenti-
ation of features that are essentially composed of the same
material - water. Likewise, the differences in composition
between exposed bars, water and vegetation make it easy
to manually map bars if the image resolution is appro-
priate. Automated mapping of bars with classification
Search WWH ::
Custom Search