40
ind land
a needs,
and use
veloping
1ed. The
odology
/sis and
remote
i|. vector
tisfy the
otentials
allow to
ual and
es as an
rtophoto
provides
n source
purpose,
ion. The
nt study
nize and
lena.
DSAT 7
th 15 m
gure 1).
evaluate
tion and
> images
ed level
g output
sampling
s related
nal) and
ions are
or relief
pose for
tion. On
i|. object
goal o f
eometric
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
Figure 1. LANDSAT 7 satellite multispectral (with 7 bands)
image over the central part of Lithuania
Other data sources used for research are: orthophoto maps and
vector database of the Lithuania Space Imagery Map at a scale
of 1:50 000. About 80% of Lithuania territory is covered by
compiled orthophoto maps at a scale of 1:10 000. The object of
investigation consists of 12 orthophoto maps which are
constructed at National Coordinate System LKS-94 with
resolution of 0,5 m and supplied in 77FF format.
LTDBK 50 000 vector database was created for the whole
Lithuania territory (65 000,3 sq. km) with the help of existing
maps information, digital databases together with panchromatic
orthophoto material from SPOT satellite, with geometrically
rectified multispectral space image and consist of 135 map
sheets. Resolution of digital information of the panchromatic
image is 10 meters, for multispectral images — 20 meters. The
database is created on the background of reference ellipsoid
GRS80, Transverse Mercator Projection with central meridian
Ly = 24°. The database LTDBK 50 000 is accepted as reference
investigating vector information obtained after classification of
satellite image and conversion to vector format.
3. LANDSAT 7 IMAGERY POTENTIALS IN LAND
COVER REGISTRATION
3.1 Land Cover Classification
During visual interpretation (human factor) of LANDSAT 7
satellite imagery was decided to segment existing feature into
main topographic feature classes, because human visual acuity
does not allow to identify all spectral differences in imagery
(Koarai, M., Kadowaki, T., Watanabe, N., Matsuo, K. 2001).
After visual image investigation, land cover was classified into
five types: 1) agricultural areas; 2) urban areas; 3) forestry
(vegetation); 4) hydrographic objects; 5) roads (as linear
objects). Indication of main deciphering characteristics of each
topographic object has been made in order to simplify object
interpretation. Individual areas are recognizable by its spectral
response in color bands.
The digital interpretation of satellite imagery have been made
using ERDAS Imagine software package. The major problem in
land cover supervised classification — mixed pixels there was
defined. During revision of classified image was detected
following problematic areas:
- many agricultural areas occurred as garden land cover class;
- spectral signature of sand objects appears in agriculture land
cover class.
Mixed pixel values were detected in urban and agricultural
areas. In order to avoid occurrence of such confusion
approaches are suggested: usage of multitemporal images to
457
individualize classes (especially for agricultural land cover
class); usage textural information to improve classification
accuracy; usage GIS procedures based on auxiliary data
(Parseliunas, E., 2001).
In unsupervised classification spectral classes were grouped
first, based on the numerical information in the data — 24 classes
were defined. Unsupervised classification is not completely
without human intervention. However, it does not start with a
pre-determined set of classes as in a supervised classification.
However, the objective of research also requires linear objects
such as roads, railways, highways, etc. Extraction of road axial
line is less or even not affected by any environmental or
seasonal factors. Considering the time of capture of satellite
images the road cover or railway track is quite easily
distinguishable. The extraction of road objects was performed
by manual (interactive) digitizing on the view. Digitization
process performed on merged image, which supplies ground
resolution of 15 m and keeps predefine multispectral bands. The
inaccurate in operator digitizing, also image georectification
accuracy were taken into account.
3.2 Determination of Suitability Criteria for Database
Updating
Using satellite imagery for land cover registration, it is
important to clear how large objects can be interpreted and
drawn by satellite imagery. Here was researched the possibility
of interpretation for updating topographic map at a scale of 1:50
000 using LANDSAT 7 imagery. All recognizable and
identifiable objects in satellite imagery was categorized or
segmented referring to the structure of geodata grouping in
digital vector data base LTDBK 50 000. It is defined the
thematic information presented in different data sources has
large differences in object interpretation. For this reason it is
necessary to determine and calculate the mathematical areas for
the whole territory and for the separate object classes in order to
evaluate objects variability. The distribution of interpretable
object classes' area in satellite imagery is similar to LTDBK. It
means that extracted information from LANDSAT 7 represents
the correct general disposition of geographic data. There are no
large clear deflections as well. But in some object classes there
are substantial changes of area, particularly open pits and
gardens. In research area, open pits and gardens are mostly
affected by nature and human factors. The comparison of
calculated mathematical areas have been made which gives
general overview about variability of object classes collected
from different data sources (Figure 2).
35000
20000
25000
20000
[ha]
15000
10000
Agriculture
S000
Forest
classes Garden
Quary
N LANDSAT
LTDBK 50000
LTDBK 50000
Figure 2. Areas of identified object groups
