ul 2004
scanner
nt and
analysis
nalysis.
s and
phthora
entified
namomi
s. They
5, using
ebhardt
ature to
station.
ion by
levels
aster in
ect soil-
rmis on
trichum
rot of
lopholus
rds) and
n does
species.
or-borne
ead and
ides and
nted out
studying
molaena
camara,
Schinus
ecies are
remotely
variables,
tation of
rubs and
detection
possible.
jues, GIS
nderstory
ps. These
ntial sites
jptible to
ASION
probable
. A major
of pre and
invasion
behaviour
n is more
ecological
sffectively
f a given
system or
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
community. There is a need to evaluate disturbances not in
terms of the elements of a given regime, but rather in terms of
ecological effects.
Estimating animal species numbers, population size and related
features is rather difficult in comparison to plants. However,
Kolar & Lodge (2001) indicated clear relationships between the
characteristics of releases and the species involved, and the
successful establishment and spread of invaders. Allen &
Kupfer (2000) developed a modified change vector analysis
(CVA) using normalized multidate data from Landsat TM and
examined Adelges piceae infestation. Mineter et al. (2000)
showed that parallel software frameworks could speed up both
the development and the execution of new applications. Luther
et al. (1997) pointed out the importance of logistic regression
techniques to develop models for predicting forest susceptibility
and vulnerability and to assess the accuracy of the susceptibility
and vulnerability forecasts. Using an integrated multimedia
approach in the vegetation database for invasive species
provides a unique way to represent geographic features and
associated information on interrelationships between flora,
fauna, and human activities (Hu et al., 1999). Predictions of
malaria risk mapping (Kleinschmidt et al, 2000) and
microbiological risk assessment for drinking water (Gale, 2001)
are some examples of risk mapping and prediction that have
been done in the field of biological invasions. Applications of
these promising quantitative approaches in an Integrated GIS
environment may allow us to predict patterns of invading
species more successfully. For the monitoring and control of
insect pests such as screw-worm (cattle pest), desert locust
(rainfall-dependent agricultural pest) and armyworm, satellite
facilities and the simultaneous use of their data could be used
for a wide variety of purposes (Barrett, 1980). Crops can be
regularly monitored to predict their economic yields, regional
early warning for famine or pest infestation and phenological
mapping of natural vegetation (Steven et al., 1992). The cost of
monitoring with colour aerial photography is within the
affordable range of most control programmes (Benton &
Newnam, 1976).
4. ISSUES OF SPATIAL AND TEMPORAL SCALE AND
ACCURACY
Scale is one of the central issues in invasion ecology. All
observations depend upon the spatial scale, size of the study
area investigated and resolution of the remote sensor. Habitat
evaluation of a species is influenced strongly by spatial scale
(Cogan 2002; Trani, 2002). There is no "correct" scale; it
depends on survey purpose (Trani, 2002). The variations in the
landscape patterns are scale-dependent (Rescia et al, 1997).
However, in most of the cases, landscape scale is used as an
appropriate scale for modelling.
McCormick (1999) pointed out the importance of scale and
colour infrared-photographs while mapping Melaleuca
quinquenervia. Carson et al. (1995) found that the LANDSAT
TM and SPOT data with ground resolution of 30 and 20 meters
respectively, are not considered useful for mapping at species
level, unless the stand of an invasive species is large enough.
Multi-date imagery therefore appears to improve mapping and
modelling the infestation pattern of canopy dominant species
(Bren, 1992, Hessburg et al., 2000, Mast et al., 1997). Medd &
Pratlety (1998) assessed the relevance of precision systems for
weed management. Bren (1992) examined the invasion of
Eucalyptus camaldulensis into an extensive, natural grassland
673
in a high flood frequency site using 45 years time series aerial
photographs (taken in 1945, 1957, 1970, and 1985)
extrapolated model showed the almost complete extinction of
extensive grass plains. Hessburg et al. (2000) used aerial photo
(from 1932 to 93) of interior northwest forests, USA and found
emergent non-native herb lands. Mast et al. (1997) provided a
quantitative description of the Pinus ponderosa tree invasion
process at a landscape scale using historical aerial photography,
image processing and GIS approaches. Welch er al. (1988)
applied GIS to analyse aerial photo for monitoring the growth
and distribution of 13 aquatic emergent, submergent and free-
floating species. They produced vegetation maps using large
scale (1:8000-1:12000 scale) color infrared aerial photographs
of different years (1972, '76, '83, '84, 85).
Current developments in sensor technology have the potential
to enable improved accuracy in the mapping of vegetation and
its productivity. Rowlinson et al. (1999) indicated that using
manual techniques to identify infested riparian vegetation from
1:10,000 scale black and white aerial photographs yielded the
most accurate and cost-effective results. The least accurate data
sources for this purpose were aerial videography and Landsat
thematic mapper (TM) satellite imagery. High spatial (less or
equal to 1m) but low spectral resolution remote sensing data
appeared to be useful in mapping invasive Chinese tallow trees
with an accuracy of greater than 95 percent (Ramsey et al.,
2002). Medlin et al. (2000) could detect infestations of Senna
obtusifolia, Ipomoea lacunosa, and Solanum carolinense with
at least 75% accuracy using multispectral digital images.
Vrindts et al. (2002) distinguished seven weed species with a
more than 97% correct classification using a limited number of
wavelength band ratios. Everitt et al. (2001b) noted that
Juniperus pinchotii had lower visible and higher near-infrared
(NIR) reflectance than associated species and mixtures of
species allowing a mapping accuracy of 100 percent. Lass et al.
(2000) tested accuracy of detection of a homogenous population
of Centaurea solstitialis at different spatial resolution. Their
result showed a low commission and ommission errors with
0.5m spatial resolution than 4.0m. Very-high spatial resolution
(0.5 m) colour infrared (CIR) digital image data from colour-
infrared digital camera imagery showed potential for
discriminating Acacia species from native fynbos vegetation,
other alien vegetation and bare ground (Stow et al., 2000). In
cases where different spatial resolutions resulted in equal
detection accuracy, the larger spatial resolution was selected
due to lower costs of acquiring and processing the data.
All these studies noted the importance of image resolution,
spectral characteristics, superiority of lower scaled aerial
photographs and images. It also shows clearly that for accurate
mapping of invasive species it is important to take the
phenological stage into account in aiming of taking the aerial
photographs or images.
Although high spectral and spatial resolution provide the ability
to classify canopy dominant species, precise classification of a
species is still difficult. Several such studies of the spectral
properties of invasive species have been derived, mostly from
low altitude aerial photography or field spectrographs.
However, the information reaching the remote observer will be
minimum. Other factors like atmospheric noise, humidity,
shadow, contribution from soil add to the confusion and the
chance of discrimination of separate species low (Price 1994).
Furthermore, variation in orientation of leaves, age of a leaf,
variation in leaf area index, different slopes of the locations
where the individuals are found could make the spectral