Image Utilisation for the Study and Management of Riparian Vegetation: Overview and Applications - Fluvial Remote Sensing for Science and Management - page 215

extent; 95% and 85% respectively for stands in leaf-off
and leaf-on conditions (Townsend and Foster, 2002;
Hess and Melack, 2003). Inundation conditions have also
been mapped with some Moderate Resolution Imaging
Spectroradiometer (MODIS) Images by Sakamoto et al.
(2007) in the Mekong delta. As with LiDAR data, such
information can improve riparian vegetation mapping
like soil moisture estimations given by RADARSAT-1
Synthetic Aperture Radar (SAR) images (Makkeasorn
et al., 2009) or by Spot images (Muller and Decamps,
2001). Lastly, hyperspectral data can give information
about vegetation phenology and physiology, and thus
indirectly participate in water budgeting by quantifying
evapotranspiration (Ringrose, 2003; Loheide and
Gorelick, 2005, Nagler et al. 2005a; Bawazir et al., 2009).

order to be properly detected with at least 50% of pure
pixels (i.e. with pure information over at least half of
its surface). However, a good perception (e.g. 80% of
pure pixels or more) is only possible with objects whose
diameter is of 16 pixels or more. For example, Spot
Panchromatic data (resolution 5 m) may provide good
information on circular objects with diameters of at least
80 m (area 5 ha). Similarly, a large tree with a canopy

diameter of 20 m can only be well-detected with pixels
< 1 m. For analysing a shrub of two metres in diameter,
pixels must be

∼ 0 .

10 cm. However, these recommendations
are optimistic as tree canopies are almost never circular
or homogeneous in reflectance, compared to agricultural
plots. The presence of shadow in the canopy at the time
of acquisition of the data also can confuse results. Gaps
in the canopy introduce a signal from the understory
vegetation or the soil into the spectral profile. This local
heterogeneity is highly variable from one area to the
other and cannot be estimated a priori in natural riparian
woodlands where several vegetation strata typically exist
(Figure 10.4). In a plantation plot, trees of the same age
and the same species (often the same clone) are planted
at regular intervals (typically every 4 or 5 metres) and
therefore quantification might be more feasible.

<

10.3 Season and scale constraints
in riparian vegetation studies

10.3.1 Choosinganappropriate timewindow
fordetectingvegetationtypes

The spectral contrasts existing between plant species
depends on the phenological stages of the local vege-
tation. Three phenophases, directly related to biomass
changes, can be easily described for each wood category:
leafing, full foliage and leaf falling (Muller, 1995). In the
growing season, the effective time window for detecting
such contrasts is relatively limited and explains the inap-
propriateness of most images (Figure 10.3). The most
obvious differences are detected either in spring (leaf-
ing) or in autumn (leaf fall) when changes correspond
to the modification of biomass with possible time shifts
from one species to another and with corresponding
increases or decreases in the near infrared bands (TM4 or
Spot 3) or in the Short-wave infrared (TM5 or SPOT4).
Unfortunately, in the visible bands, little information
can be easily extracted due to low signal levels and poor
differentiation between species. In addition, other impor-
tant phenophases are not detectable (e.g. flowering, seed
ripening, dispersal, and additional leafing phases).

10.3.3 Spatial/spectral equivalence for
detectingchanges

If one considers that a pixel of vegetation with a surface
S has a reflectance R and is affected by an internal change
(e.g. leafing, dieback, leaf fall, flowering etc.) characterised
by a reflectance R over a surface S , the resulting new re-
flectance R of the pixel may be roughly approximated by:
R × R × S + (S - S ) S = R × (10.1)

The corresponding modification of the reflectance of the
pixel by the disturbing factor is:
(R −

(R − S /

= ×

R) R) S (10.2)

This equation simply reminds us that, in order to be
detected by a pixel (S fixed), change relies both on the
reflectance contrast (R -R) of the disturbing factor and on
the surface S affected by this change. For example, dou-
bling the reflectance change of the pixel requires doubling
either the reflectance contrast of the disturbing factor, or
its area, or the product of both. A change in reflectance
can also be compensated by an inverse change in area
and thus be undetectable. The equation also indicates
that to detect the same change with a pixel size twice the
size of another (e.g. using images of 20 m against 10 m

10.3.2 Minimumdetectableobject size
inthe riparianzone

The quality of the perception of an object can be roughly
estimated by simply considering the proportion of pure
and mixed pixels in its description (Table 10.3). For
example, a circular object (typically a tree) must have
a diameter at least 8 times larger than the pixel size in