International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
indication and also the possibility of controlling all natural and
technical survey condition. Among automated methods of
detection of changes the digital processing and analysis are
used. Sometimes the systems work by recognizing the
differences between compared images before classification of
detected differences; and other times, in contrast, images are
first automatically classified to produce ecological classes and
then subtracted. The second method (post classification
comparison method) is the most commonly used.
The main task in comparing the images of different years and in
discovering changes in the areas of ecosystems is to determine
the common basis of these changes, that is, the ecological trend.
This trend is defined as a translational change of a system
whereby its state on any time scale is not repeated and the
direction of its drift is constant within the investigated time
limits. The main purpose is to measure and describe the
ecological trend in terms of area change, using aerospace
images. It should be excluded the influences of different short-
terms fluctuations (for example hydrometeorological variability
in different years). Distinguishing ecological trends at this level
involves great difficulty and only a rigorous mathematical
expression is sufficient to allow scientific ecological prognoses.
2.2 Research sites
2.2.1 Caraorman-Dunavát site: As study sites has been
selected Delta’s representative areas, both for flora and fauna
diversity and for dynamic changes from the past decades. The
site has a square form of 30.72 km side and is placed in central-
western of Danube Delta and contains the following
ecosystems: ecosystem of dry land - in the south-western of
site, including surroundings of Murighiol, Dunavátu de Sus and
Dunavátu de Jos villages; ecosystem of running water -
including Sulina and Sf. Gheorghe branches, Dunavát, Tátaru
and Crisan channels; ecosystem of stagnant water - including
lakes ponds, and plugged channels; ecosystem of river banks
and marine levees - Ceamurlia, Lat and Caraorman, which had
been the domain of all kind of trees, broke up in their greatest
part and replaced with Canadian poplars. On Caraorman bank
still exists a large abandoned sand-mining; ecosystem of
marshy and flooding surfaces - is dominated by emerged
vegetation prevalent constituted of common reed and a mixture
of common reed- mace reed. Floating reed islet formation,
which is a mixture of reed roots herbs and soil, usually floating
or fixed on the bottom of depressions, has a peculiar place
within the ecosystem; ecosystem of polders - include fish
ponds, agricultural and forest polders; the zone in the north of
Sulina branch is declared area for ecological restoration.
2.22 Sulina site: The site has a square form of 24.1 km side
and is placed in central-eastern Danube Delta. This includes in
addition to Caraorman- Dunavát site, marine ecosystem -
include Black Sea coastal area. Sea proximity creates a series of
special problems. It is very important to underline that water
level variance in ecosystem of marshy and flooding surfaces is
dependent on water flow variance on Danube branches. This
ecosystem acts as a large buffer lake, which attenuates water
flow variance. Water levels difference in inner delta from april
to june has a 60-70 cm maximum.
2.3 Data sources
The study was based on available satellite images (Landsat
MSS, July 24, 1975; Landsat-TM, June 28,1984 and SPOT-
HRV, April 27, 1993), aerial photographs at 1:45,000 scale
acquired in 1984, ground radiometric measurements realized in
some representative areas of Caraorman Dunavät site, near the
Remote Sensing Laboratory of Technical University of civil
Engineering-Bucharest, at 54 km on the Sf. Gheorghe branch,
in 1984 and 1993, with a four-channel Exotech 100 AX
portable radiometer using Landsat-TM channel filters (to
facilitate the object classes separation and point out the
atmospheric effects) and complementary data (topo-
hydrographic maps, main ecosystems map, Danube Delta
vegetation and soils maps, etc). Using Landsat-MSS images
(characterized by a lower resolution) the study period extends at
18 years.
3. METHOD
3.1 Preliminary processing
One of the most important temporal reasons for change-
detection applications is the acquiring month of the imagery.
Images acquired during the period with power sun light, present
a very good contrast between various details (Jansen, 1983). For
example, in this case, the contrast from covered with vegetation
soil and uncovered is very strongly. Using the scene acquired in
the same year's period is suggested to change-detection with
purpose to reduce the problems which appear because sun-angle
difference, vegetation-phenology changes and differences in
soil-moisture. (Hall et al., 1991; Jansen et al. 1993). SPOT
HRV image was used as a reference image to ensure
compatibility of multitemporal data.
Due to lack of information regarding atmosphere characteristics
radiometric calibration has been realized for every site based on
15 invariant or quasi-invariant in time objects (Chavez, 1989):
darker zones (great lakes water and Black Sea without turbid
water), or brighter zones (Caraorman bank and beaches sand).
Landsat- MSS and TM images normalization has been done, for
every channel, using a nonlinear regression equation:
yi)=ax(i)+h (1)
where: y(i)-is radiometric value of pixel from reference image;
x(i)-is correspondent radiometric value of pixel from corrected
image.
During the registration process, have been used two reference
images, based on aerial photos at 1:45,000 scale, representing
the geometric fundament for images registration (Rignof et dl.
1991). Geometric corrections applied to photos and satellite
images have been based on 34 ground control points for
Caraorman-Dunavát site and on 28 ground control points for
Sulina site. These points have been determined, using software
PAT-MF for Aerotriangulation with independent models, at
Wild Aviolyt BC2 in the Photogrammetry, Remote Sensing and
GIS Laboratory of Moncton University, Edmundston center,
Canada.
It has been realized a resample of 4x4 for MSS and 1,5x1,5 for
TM because different sensor resolutions (80m for MSS, 30m for
TM and 20m for HRV). For low-resolution images resampling
(Landsat MSS) it has been used the nearest neighbour-method,
which does not alter the digital number of the original image
pixels. For Landsat-TM images transformation have been used
three-degree polynomials, followed by a cubic convolution. The
resample function included 16 pixels surrounding the processed
pixel, allowing local modeling of image by a polynomial
surface. Transformation accuracy varies according to resample
images. The root-mean-square (RMS) error is +0,436 pixel for
330
Internat
MSS in
HRV im
3.2 Co
Theoreti
differen:
last rang
realistic
in range
and unc:
Multisp:
which a
very ex
cambina
4: TM 4
be reali:
confirm
a signifi
In this c
used tw
Differen
a prefere
versus tl
Examini
that only
contrast
enhancei
enhance
unsupen
enhance
enhancir
values fi
1989).
direction
linear de
The w
distingui
morpho-
continen
past mez
structure
(compac
vegetatic
has been
33 Mu
For mul
method.
composi!
algorithn
Marsh,
some ar
common
remind v
been use
and a ve;
classes r
for 8. «
common
vegetatio
soil for €
running
mace ree
urban ai
