Geography Reference
In-Depth Information
Another limitation lies in the difficulty of getting the
exact dates for images (day and hour are the best), and,
when these are available, for associated discharge. This
can have clear consequences for the detection of corri-
dor features, especially if bankfull discharge is exceeded.
This is particularly important on photo mosaics, which
often have multiple dates within the mosaic. One needs
to take care of the discharge/water level to compare fea-
tures between dates (e.g., gravel bar area, oxbows) or
when comparing features across a stream network (rain-
fall versus snow-melt or glacial hydrological regime, with
high flows occurring in different seasons whereas aerial
photographs are usually taken in summer). It is usually
misleading to simply compare aquatic areas on two sep-
arate dates because discharge is almost never the same
at the time the images were made. To address this prob-
lem, there is value in determining the discharge on the
day and hour when the photo was taken. Furthermore,
if discharge does vary from photograph to photograph,
there is also a need to determine the error in width associ-
ated with the difference. This can be done by considering
the stage-discharge relationships at the nearest gauging
station or by calibration from field surveys. However,
cross-section geometry is usually different in the area of
interest compared to the nearest gauging station site. On
the Rhine River, the error in channel width associated
with a water level difference of up to 40 cm due to dif-
ference in discharge during the photographic acquisition
was estimated from 12 selected cross-sections along the
reach. The water level variation around a reference water
level corresponds to an error between
difficulty of getting enough data for linking discharge and
planimetric characteristics.
Even if some images have the advantage of covering
large areas to provide regional biophysical representa-
tions of fluvial corridors, there are potential limitations
associated with the inherent nature of fluvial features.
Some of them are detectable by imagery only on a portion
of a hydrographic network (Figure 11.13). Channels with
a width smaller than the half diameter of riparian tree
crowns are often difficult to delineate no matter what
resolution imagery is being used. Due to the canopy
coverage in the upstream sections or to the deep water
column in the downstream sections, an exhaustive survey
of in-channel features is usually not possible. On the
Rh one catchments, 38 481 km of rivers are so narrow that
it is impossible to classify in-stream features, 5 185 km
combined exposed gravel patches and low flow channel,
and 8 048 km are only water, showing that the analysis
of in-channel features is possible on less than 10% of the
channel network. There is also a size effect that compli-
cates the detection of features that vary in either form or
characteristic size from upstream to downstream. Flood-
plain patches along large rivers are different than very
narrow floodplains observed along upper branches. Image
resolution, however, does not usually increase in upland
areas. As a consequence, image resolution provides some
clear limitations when covering the entire channel. In
many cases, a 20 to 50 cm pixel size, which is the most
common size range of archived aerial photographs, limits
the approach to rivers and large streams.
The treatment time for digitising, photo-interpreting,
and computing is also a critical issue to consider when
enlarging the spatial and temporal scale. Most histor-
ical images are presently archived as hard copies that
require scanning and georectification. There is thus a
need to develop automatic procedures for georectifying
images and extracting features, and the associated com-
putational requirements are not be trivial. In this domain
new autorectification methods are in development which
should minimise this effort, but error associated with the
procedure needs to be evaluated across a range of contexts
(Carbonneau et al., 2010). The classification procedure
applied to multiple images also requires significant com-
puter capacity to minimise the calculation time. In the
case of the Rh one network (45,000 km long), the area
of interest is covered by 6347 images (e.g, 1.9 Terabyte
of information) and most of the imagery analyses are
performed using several days of computational time on a
computer cluster. In order to minimise the required com-
putation time on the Rh one network, several steps were
5and
+
5m in
flow channel width (e.g.,
5%) (Arnaud et al., 2010).
There is also a problem of contingency of the date
at which aerial photographs are taken relative to flood
history. It is sometimes difficult to distinguish if the
observed patterns are associated with a long period of
anomalously high or low flow or with a long term trend
associated with other parameters, especially for lateral
changes and interpretation of vegetation/pioneer unit
coverage which is intrinsically linked to the temporal pat-
tern of colonisation. It is thus difficult to study variation
in channel shifting at the wider scale as photo sets used to
highlight shifting may be associated with different time
periods (effects of floods on the active channel width for
example). Similarly, variation of water surface width at a
range of flow conditions may be of interest for predicting
discharge from water area. In this case, the observer will
search for photos with contrasting discharges. However,
this is not always possible because most photos are taken
at low flow during the summer, thereby increasing the
+
/
Search WWH ::




Custom Search