You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

The 3rd ISPRS Workshop on Dynamic and Multi-Dimensional GIS & the 10th Annual Conference of CPGIS on Geoinformatics
Chen, Jun

ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS", Bangkok, May 23-25, 2001
id GIS data, with the
lery in an integrated
i by utilising a digital
J in the procedure is
rried out for both the
-inally, the nature of
learly identified.
mapped up to now
national scales has
oatial resolution of
er maps generated
jery, with a spatial
if detail in the scene
ailability of recent
tial resolution has
leometric detail and
397) for local and
ly to generate other
is enhanced (11 bit
*re generally limited
ion sensors (seven
jssociated with an
ally, an increase in
mage) generating a
ation on a per-pixel
les classify land use
als within an image,
lost similar spectral
suits that are 'noisy'
sification (so called
are classified as
spectral and spatial
pe of the fields and
ormation about the
pies, roads fields. If
nd registered to the
vithin boundaries of
elds. For instance,
r IKONOS bands of
as features defining
Jormally, the use of
thin a geographical
ides facilities for
3mple, checking the
, ensuring that lines
cal lines that end
Fig. 1 "Color composite image of IKONOS data - Fasano (Italy)
All these considerations suggest that replacing the coarser
spatial resolution satellite sensor imagery with finer ones would
lead to an increase in per-field classification accuracy, especially
for relatively small sized fields (Harris and Ventura, 1995;
Westmoreland and Stow, 1992).
However, the method used to integrate remotely sensed imagery
with field boundaries is also meaningful. Such integration may
be carried out in three stages (White, 1995):
before classification (pre-classifierstratification),
during classification (classifier modification)]
after classification (post-classifiersorting/).
Many examples of these procedures employed only the latter
two. Westmoreland and Stow (1992) integrated, in a single
stage, cartographic data with remote sensed imagery during
classification to assess land use change on per-field basis.
Alternatively, other studies obtained land cover on a per-pixel
basis before integrating the classified image with cartographic
data for per-field classification (White, 1995).
The ways with which the two data models, raster and vector, are
integrated in a RS/GIS system can be summarised in:
a) separate database, cartographic and image processing
systems with facilities to transfer data between them;
b) two software packages (image processing and GIS) with a
shared user interface and dynamic links.
c) a single software package with shared processing.
In this study the b) type of implementation is performed, because
of the high level of the customisable statistical and logical tools
and the possibility of modifying per-pixel classification using the
spatial information derived from external data.
This research was carried out in the Italian territory (about 4 Km
per 5 Km) which includes part of the municipal district of Fasano
(Brindisi) in the Apulia region (Fig. 1). It represents a critical area
because of urbanisation over the last thirty years, still continuing
now. Such evolution is been apparent with long period built-up
area phenomena on a local scale (second home), with rise and
spread of tourist places (“Safari” zoo, fair grounds) and, finally,
with agricultural transformations (olive-groves dominating over
natural areas, such as ilex and Mediterranean bush).
The géomorphologie and settling features, with the town and the
ancient fortified farms that are located on the plain (100 m a.s.l.
quota) and the villas that are strewn around the hills (300-400 m
a.s.l. quota), justify the choice of this area because it permits the
testing of the validity of well-established methodologies of
classification on a over-regional scale, but barely investigated on
new generation satellite sensors imagery. Moreover, such
studies allow opening of new unexplored techniques applicable
on a large scale by adding meaningful inputs to those
multidisciplinary studies connected to decision-making and
planning activities.
The data sources acquired for the analysis consisted in:
the 4 bands IKONOS raster data (Acquisition Date/Time:
2000-05-12 / 09:14), processed using the remote sensing
image processing system ERMAPPER;
the digital topographic map in a scale of 1: 5000 (Date
1998), stored and processed by means of the geographical
information system ARC/INFO;
the ground reference data obtained with a land use survey
(January 2001).
4.1 Data processing
To facilitate the integration of the IKONOS and vector data it was
necessary to register both data sets to a single map coordinate
system (Mather, 1995), in this case Gauss-Boaga map
projections and Roma40 datum, by identifying 30 common