rtation
ency =
Ise, i.e.
|.to. be
i| level
cts that
such as
Jon are
owcase
| share
ational
As a
rogram
A) was
nent in
mercial
antages
urban
ds. But
onomic
ngly to
ions for
change
ses and
and the
[es nOW
cycling
ventory
ant and
ntatives
nts and
elop this
d with
esources
ot been
munities
nfld.ht
I...ernational Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
1.2 Aims
The emphasis of this project is to launch the detection of
potential Brownfields sites and to provide this specific
spatial data for communities. This goal is aiming high
and also ambiguous as such resulting data are needed,
however Brownfields do not only cover land but
obviously are used land. The latter makes it extremely
difficult to work on by means of remote sensing data as
land use can only partially be assigned. Brownfields first
need to be characterized within their specific urban
regions in different geographical latitudes and climates
with different ecological and economic situations. In their
urban environment brownfields sites need to be
characterized by their form, their position and their
spatial context and can be described as consisting of
different objects such as buildings, roads, and vegetation,
and they can clearly be marked as a highly disturbed land
use.
In order to work on urban brownfields two prerequisites
are essential: a methodological approach that allows to
classify complex objects combined with high quality data
(see Barnsley, 1997).
1.3 Remote Sensing Data and Methodological Approach
For this study a panchromatic and multispectral Ikonos
imagery acquired on 02-Oct-2001 is used (Space
Imaging, LLC). Three overlapping flight paths were
taken on the very same day to cover the entire City of
Baltimore. In order to orthorectify the images the cubic
convolution algorithm was applied.
The investigation presented in this paper expands upon
an object oriented classification method as any
multispectral classification scheme will fail to detect such
a highly heterogenic object class. Image segmentation,
fuzzy classification, and structure type assignment is
performed by means of eCogniton software. This
software follows a new, object oriented approach towards
image analysis. The concept behind this software
program is that important semantic information necessary
to interpret an image is not represented in single pixels,
but in meaningful image objects and their mutual
relationships. So first the image is being structured into
user-defined homogeneous segments in any desired
resolution, then the classification procedure can follow.
The segmentation algorithm entails the simultaneous
representation of image information on different scales.
This procedure detects local contrasts and is especially
designed to work with highly textured data, such as
Ikonos, Quickbird, or digital orthophotos. The
classification process is based on fuzzy logic, to allow the
integration of a broad spectrum of different object
features such as spectral values, shape, or texture.
Utilizing not only image object attributes, but also the
relationship between networked image objects, results in
a classification scheme incorporating local context (Baatz
et al., 2000). So land use classes can also be defined as
461
“adjacent to” or “in a certain distance to” another class.
This fuzzy logic approach leads to the characterisition
and description of distinct urban land use categories
(Bauer & Steinnocher, 2001). The resulting information
is integrated in a rule system on a higher level of image
analysis on which classified land use objects are
combined to semantic structure groups, in this case
potential brownfield sites. X
The assumption underlying this approach is that potential
Brownfields sites are a land use type that follows a
certain pattern (i.e. consisting of buildings, roads or road
access, impervious surface and neglected green spaces)
so that each object can first be classified and then be
composed to a variation of structure groups).
1.4 Test Area
Brownfield sites presented in this paper will be taken
from the City of Baltimore, Maryland. Baltimore has
experienced a population decline by 11.5% between
1990 and 2000 from about 736,000 to 650,000
inhabitants adverse to the development of the Baltimore
Region which had an increase of 7.0 % during the same
period of time from more than 2,3 million people to more
than 2,5 million. Thus Baltimore City underlies an
enormously rapid change resulting in urban sprawl and
areas of conversion being located in rather central areas
(http://www .baltimorecity.gov/government/planning/cens
us/index.html).
EPA selected the City of Baltimore as one of 16 federal
Brownfields Showcase Communities in 1998. This
designation gives Baltimore preferred access to federal
resources which help building up Brownfields incentives,
and which initiate the Maryland Voluntary Cleanup Act
in February 1997. Brownfields Showcase Communities
have the following main goals: to promote environmental
protection, economic redevelopment and community
revitalization through the assessment, cleanup and
sustainable reuse of brownfields, and, to link Federal,
State, local and non-governmental action supporting
community efforts to restore and reuse brownfields
(http://www.epa.gov/swerosps/bf/showcase.htm ).
An Ikonos data set exists for the whole county of
*Baltimore City" (Maryland Department of Natural
Resources - DNR) and there is abundant ancillary data
available for the city online
(http://www.dnr.state.md.us/gis/).
The Voluntary Cleanup Program (VCP) provides data on
Brownfields properties for which the Maryland
Department of the Environment (MDE) has received an
application to participate in the VCP. It is this point data
that is used here to derive surficial test sites for the
classification scheme. Parcel boundaries are assigned to
these properties by means of matching street addresses to
each parcel. These data provide an excellent data base to
develop the classification scheme (see Figure 1).
