Geography Reference
In-Depth Information
intensive and, as a result, a multi resolution sampling
strategy was developed. The process was as follows: (i)
mature cottonwood forest stands were mapped using a
Quickbird image isolating deciduous from conifers, (ii)
forest establishment age was assessed for these stands
using a time-series of 50 cm aerial photographs taken
between 1948 and the present (Figure 8.15a and b),
(iii) from this map mature cottonwood patches (35-45
years) were selected, (iv) from these mature cotton-
wood stands dead branches were detected using available
hyperspatial images (Figure 8.13). As the hyperspatial
data provided a sufficient spatial resolution that dead
wood branches could be individually identified with the
naked eye an image processing methodology was devel-
oped that used not only the spectral properties of the
dead wood, but also textural and structural properties
discernible from the images (see Dunford et al., 2009).
Figure 8.15c shows a plot of dead branch density versus
plot elevation. A positive and significant correlation is
observed thus supporting the initial hypothesis. In this
example, hyperspatial imagery greatly reduced the field
effort needed to characterise the dieback of cottonwoods
and provided highly useful additional textural and struc-
tural parameters to assist the remote characterisation of
dead branches.
Our second example comes from restoration work on
the Malourdie reach of the Rh one River in France. This site
is a former channel which was dredged in 2003 in order
to recreate water ponds with low velocity on the margins
of the fast flowing main channel in order to provide habi-
tat for fish spawning and rearing and to improve local
biodiversity (Figure 8.16). In such a context, monitoring
is essential in order to evaluate the intervention's success
and to estimate its life span and evolution. Hyperspatial
ULAV campaigns were done in April 2006 and March
2007 supported with field surveys providing ground con-
trol and water depth measurements (119 in 2006 and
59 in 2007; Figure 8.16a). By applying regression models
linking the water depth measurements to the radiomet-
ric values from the hyperspatial imagery, predictions of
water depth could be generated (Figure 8.16b). This pro-
vided bathymetric map covering the full restored reach
covered by the imagery for each of the years surveyed
(Figure 8.16b - April 2006). The quality of two models
ranged from 0
considerably reduced through time and from this data a
trend has been extrapolated (Figure 8.16c). This model
is used then to predict bathymetry at a range of times
(Figure 8.16d).
8.4.2 Repeatedsurveys throughtime
A further advantage of flexible platforms is their capa-
bility to perform repeated hyperspatial surveys allowing
seasonal changes and inter-annual changes to be evalu-
ated. Multidate approaches bring with them a number
of constraints in term of acquisition which must be con-
sidered by operators when designing campaigns. One
of the key constraints is non-uniform illumination dur-
ing different survey dates which can lead to difficulties
when comparing images from different epochs. Causes for
these differences include atmospheric condition, season
and time of day. In addition channel spectral proper-
ties may also vary as a result of changes in either the
channel bed (e.g., biofilm development, spatial varia-
tion in bedload remobilisation), surface conditions (e.g.,
algae development, laminar versus turbulent flow condi-
tions) or water column conditions (variations in discharge
and suspended sediment concentration). All these vari-
ations become factors that complicate analyses methods
which are based on the spectral properties of the image
such as bathymetric mapping or vegetation phenology
classification.
Here we discuss a multidate study of the Mollon
reach of the Ain river, France. The site was surveyed 10
times during 2010 mainly during the months of April to
September. The range of discharge during the six month
acquisition period was 20 to 80 m
3
.s
−
1
with a mean annual
discharge of 120 m
3
.s
−
1
. From these repeated surveys dur-
ing the spring/summer period, it is possible to map the
vegetation patches and monitor monthly changes involv-
ing patterns of growth and senescence (Figure 8.17). It is
also possible to map the colonisation of the former chan-
nel by floating vegetation between June and July 2010
(Figure 8.17, pink square in the upper right corner and
Figure 8.18) and the early spring colonisation of herba-
ceous vegetation on the gravel bar and its senescence at
the beginning of the summer. The use of image data from
different dates allows for the classification to distinguish
different vegetation patches - thus separating high, low
and sparse riparian vegetation patches. However, as the
georeferencing of each photo is not perfect, the images
do not overlay perfectly. Therefore the boundaries of the
different land-use types are fuzzy, and may often require
a filter to mitigate for these spatial errors.
R
2
89. Residual errors of models
have been also estimated, ranging from 9 to 12.5 cm, and
mapped showing a homogeneous spatial distribution of
residuals error. Sedimentation is also monitored at the
site every two years. The initial year following the restora-
tion, the sedimentation rate was significant but this has
.
71
≤
≤
0
.
Search WWH ::
Custom Search