Management Applications of Optical Remote Sensing in the Active River Channel - Fluvial Remote Sensing for Science and Management - page 27

over another. Moreover, simple methods often seem to
generate results that are nearly as accurate as more com-
plex approaches. Bustamante et al. (2009), for example,
found that the simple Water Turbidty Index of Yagamata
et al. (1988) yielded an r 2 value of 0.75 for measured ver-
sus modeled turbidity values, while a far more complex
General Additive Model yielded an r 2 of 0.79. Similarly,
Wang et al. (2009) found that one variable regression
using band 4 (near infrared) of Landsat provided excel-
lent estimates of suspended sediment concentrations in
the Yangtze River (Table 2.2).
The wide range of techniques for estimating turbidity
with optical imagery indicates the need for a more gen-
eral theoretical basis for turbidity measurement. Dekker
et al. (1997) reviewed the optical theory that underlies
turbidity mapping by remote sensing and identified sev-
eral issues of particular importance. Light interacts with
turbid waters in a non-linear manner, making it hard to
develop empirical relations that can be transferred from
one system to another. Sun angle relative to the surface
and sensor can also strongly alter turbidity estimates. The
many factors controlling turbidity and its reflected signal
have hindered development of physical models that can
be applied without local calibration data. Mertes et al.
(1993) developed an approach based on spectral mixture
analysis that holds promise, but even their technique
required calibration to laboratory data. For the fore-
seeable future, managers therefore will have to rely on
establishing empirical relations between local turbidity
and the sensor signal. Empirical approaches of this nature
can produce accurate results, but are difficult to imple-
ment in remote rivers or during flood conditions, which
are often the periods of greatest interest to managers.

system) has been used to map exposed sediment sizes at
reach scales (Hodge et al., 2009). These approaches to
ground-based mapping of sediment size are useful at the
scale of an individual plot, bar, or reach but are not
feasible over longer lengths of stream where thousands
to millions of photos or ground-based surveys might be
required to cover the entire area.
It is only recently that airborne optical imagery has been
available at sufficiently fine spatial resolutions to measure
sediment size over long lengths of stream. The coarser
resolution of these photos compared to ground images,
however, limits what they can detect.
Rather than trying to measure individual grains as is
done with ground photos, approaches for measuring sed-
iment size with aerial imagery use the image semivariance,
a statistical technique that characterises the 'graininess'
or texture of an image. The concept is simple. Areas with
larger sediment sizes have more shadows cast by the large
clasts and therefore have a more heterogeneous texture.
In contrast, surfaces with much finer sediments have less
of a size difference between clasts, have less shadowing
effect, and are more homogenous in appearance. These
variations in image texture within a given window (e.g.
35

35 pixels) can be measured by a two-dimensional
variogram (among other techniques), with higher values
indicating more brightness variation from one pixel to
the next. Carbonneau et al. (2004, 2005) discovered that
the values from a two dimensional variogram are linearly
related to grain size for both dry and submerged sedi-
ments. A linear regression between field measurements
of the median particle size, D 50 , and the two dimensional
semi-variance of the image for the same locations was
used to develop equations that could then be applied to
the remainder of the image to accurately estimate D 50
(Table 2.2). This technique can also be used to map other
percentiles of the sediment grain size distribution (D 16 ,
D 84 , etc.) so long as: (a) the window size is large enough to
get a stable semivariance signal; (b) the sediment patches
are relatively uniform at the scale of the window; and (c)
the grain size fraction (e.g. D 50 ) is larger than the image
resolution (Carbonneau et al., 2005). Using this tech-
nique with 3 cm resolution imagery, Carbonneau et al.
were able to continuously map sediment size along 80 km
of the St. Marguerite River in Quebec.
Assuming that the sediment can be seen through the
water, image resolution is the biggest limitation on mea-
suring sediment size. The general rule of thumb is that
the smallest size of sediment that can be mapped with
black and white or three band colour imagery is equal to
the pixel resolution of the image (Carbonneau, 2005).

×

2.7 Bed sediment

Mapping bed sediment size is important for documenting
in-stream habitat for fish, macroinvertebrates and other
organisms, for characterising flow resistance for hydraulic
and flood inundation models, and for modeling sediment
transport and channel stability. There is a substantial
body of literature on ground-based optical measurement
of sediment size where the pixel resolution is far smaller
than the sediment size. In this case the individual particles
can be seen with the naked eye on the imagery, so the
focus shifts to automating the procedure in order to
delineate particle boundaries and measure particle axes
(e.g. Raschke and Hryciw, 1997; Graham et al., 2005).
More recently, terrestrial laser scanning (an active sensing