Geography Reference
In-Depth Information
gradually varied and no significant over-arching vege-
tation was present. Given the lack of uniformity in the
illumination conditions, this has led to three different
calibration and validation relationships which are clearly
visible in the figure. However, Carbonneau et al. (2006)
noted that these three relationships were parallel to each
other suggesting that the optical attenuation of the light as
it passes through the water column occurred at the same
rate in all three images. Therefore, what was required was
simply a re-adjustment of the base illumination in each
image. The complete solution involved an adaptive pro-
cedure whereby portions of dry exposed sediments were
used to establish a wet/dry interface (Carbonneau et al.,
2006). Immediately adjacent to the dry portion of this
interface, the water depth could be assumed to be zero.
This feature-based property allowed for an adaptive recal-
ibration of the image brightness which minimised drift
in the depth measurements artificially caused by changes
in the illumination of the imagery. This recalibration was
applicable to the whole image dataset.
Figure 9.10 shows the result of this correction proce-
dure when applied to the data in Figure 9.9. The three
parallel relationships have now collapsed into a single
relationship which makes the process of depth mapping
reliable. Figure 9.11 shows a typical result of the finalised
depth mapping process when applied to an image. It can
be seen that depth was set to zero at the interface of the
gravel bar and then increases smoothly up to a seemingly
constant, saturated, value of 1.5m. This saturation value
is another key limiting factor in such approaches. Unsur-
prisingly, the success and feasibility of depth mapping
from standard colour imagery depends on the visibility of
the river bed in the imagery which is dependent on water
clarity. It is therefore not uncommon to observe a satu-
ration depth below which the riverbed is not visible and
depth mapping cannot operate. In the case of Figure 9.10,
this depth was found to be roughly 1.5m. Given that
water clarity conditions can be highly variable, there is
no fixed value for this saturation depth. Even for a given
river, changes in the suspended sediment load will lead
to large changes of water clarity. Therefore, the empirical
depth mapping described in Carbonneau et al. (2006) is
only valid for depth data collected on the day of image
acquisition.
The problemofmodelling river bathymetry fromimage
data remains a complex one. Legleiter and Roberts (2009)
have used computer simulations in order to thoroughly
examine the limitations and weaknesses of this approach.
They conclude that one key parameter which could
improve image-based bathymetry mapping is improved
radiometric resolution (see Chapter 1 for a definition).
Unfortunately this seems to go against the current trends
in the instrumentation used for Fluvial Remote Sens-
ing (FRS). Rather than using bespoke imaging equip-
ment, FRS is increasingly using standard photography
equipment. The work of Legleiter and Roberts (2009)
is a clear indicator that the FRS community may soon
have to consider a move towards more advanced sen-
sors more suited to the specific requirements of riverine
environments.
9.5 Further developments in the wake
of the Geosalar project
Because of the momentum it generated, the Geosalar
project continued to stimulate research and development
after its completion in 2008. Issues related to the man-
agement and visualisation of the extraordinary volume of
data generated by the new remote sensing methods were
more specifically addressed. Work leading to the devel-
opment of in-house airborne image acquisition systems
was also conducted.
9.5.1 Integratingfluvial remotesensingmethods
The Geosalar project delivered one of the largest hyper-
spatial image databases currently available. Additionally,
it prompted the development of some important FRS
methods which were needed to analyse the images. Both
inside and outside of the Geosalar project, the significant
recent progress in fluvial remote sensing methods (see
Marcus and Fonstad, 2010) has generally focused on sin-
gle papers presenting single methods. There is currently a
severe paucity of papers which analyse the riverine envi-
ronment with a range of integrated hyperspatial remote
sensing approaches. However, the Geosalar project pro-
vided a needs-driven impetus to the development of an
integrated interface which could allow users to manip-
ulate and, crucially, analyse the large volume of data in
the image database. In 2003-2004, a prototype Fluvial
Geographic Information System (FGIS) was developed
by Geosalar researchers. The goal of this system was pre-
cisely the integration of the depth and grain size mapping
methods along with automated channel width measure-
ment done directly from the images. This early prototype
successfully produced the data seen in Figure 9.7. How-
ever, one of the key limitations of the FGIS prototype
was georeferencing of the image data. Georeferencing
can be defined as the process whereby an image raster is
mapped to real world coordinates. This process is crucial
in order to preserve the spatial relationships (e.g. the
Search WWH ::
Custom Search