International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
e Industrial / commercial building 
e  Paved parking lots — few cars 
e Neglected green spaces 
Potential Brownfields sites 111 
e Industrial / commercial building 
e Neglected green spaces 
So several cycles are run to detect potential Brownfields sites as 
structured image objects. At this stage, structure groups without 
buildings have not been successfully assigned. It shows that 
Brownfields sites that are just covered with impervious surface 
are very difficult to be traced and the scheme has to be im- 
proved for this kind of surficial structure. The result is shown in 
Figure 4 with the same subset as the image and classified result 
above (see Figure. | and 
3). 
  
Figure 4. The black lines show correctly presented structure 
groups. The black outlined area including hatch marks just 
above the grouped brownfields marks where another 
brownfields site is located but cannot be detected with the 
presented method yet. 
3. DISCUSSION 
The land use classification of potential urban brownfields in the 
study region produces promising results. Using Ikonos imagery 
and working with an object-based classification approach, such 
as the software program eCognition provides, are essential tools 
to approach the investigation. There are a manifold factors that 
impact the classification and its individual steps including the 
timing of data acquisition (not only the phenological stage for 
green spaces, but also the day of the week for the presence of 
cars), the availability and quality of spatial test sites (point data 
such as street addresses are insufficient, parcel boundaries in 
terms of polygons are a prerequisite), and background 
information on the brownfields test sites (timing of cleanup and 
possible redevelopment). Some of the more distinct 
classification errors are between shadow and dark roofs, 
residential areas and urban parks. Shadows are also an 
especially important consideration with the use of this high 
spatial resolution imagery, particularly shadowing of urban 
canopy into streets, alleys, and onto adjacent green spaces. As 
no multitemporal imagery is available for this particular study, 
the visual comparison with orthophotos or the knowledge of the 
urban structure helped to improve the first classification step 
(see Goetz, et al., 2003). To validate the assigned brownfields 
sites for the available test sites a visual check and image 
overlay can confirm. In order to improve the overall 
classification scheme and to validate the data for the whole city, 
ancillary GIS information such as data to urban green 
infrastructure, protected lands, private conservation properties, 
just to mention a few, need to be taken and compared with the 
resulting structure type. 
Investigations on potential urban brownfields sites by means of 
remote sensing data is a brand new field with no comparing 
projects. As the issue is compelling in many cities worldwide 
and high resolution imagery gain weight for urban problem 
solving it can be assumed that this issue is going to be tackled 
more often and that technical improvements for their detection 
will accelerate. 
4. CONCLUSION 
Brownfields sites have been recognized as important urban 
features that need to be detected in early stages, cleaned up, and 
redeveloped. Traditional approaches to mapping and monitoring 
these critical sites rely on ownership information, active 
neighbourhoods, realtors, to mention some, and are usually 
recorded as point data. This is not very practical for planning 
purposes, especially when the public issue goes beyond realtor 
interests and stand for healthy, vibrant neighborhoods. Landsat 
and ASTER data are simply not of sufficient spatial resolution 
(30m, 15m) to adequately map brownfields sites. So very high 
resolution imagery is needed to work in this field. The up-to- 
date pros and cons of the presented remote sensing approach are 
listed below. 
Major advantages: 
e Potential brownfields sites can be assigned for a 
whole city. 
e Once a step-wise approach is worked out it can easily 
be modified for specific characteristics (e.g. 
Brownfields in a virtual city are most likely to be 
found next to water, or just below the local foothills, 
or they always consist of an older pavement than 
other lots do). 
e Potentially assigned sites can be compared with 
available GIS data for a city. 
e Verified sites can be marked in maps and checked by 
local authorities. 
Major disadvantages: 
e Remote sensing only offers a bird's eye perspective 
on a critical and challenging issue. 
e Very high resolution data sets need to be available. 
e There is no single assignment for such sites and their 
detection is highly knowledge-based. 
e The detection is not operational yet and still part of 
basic research. 
464 
Intern 
Refer 
Barns 
recogi 
land c 
21019 
Bauer 
classi! 
GeoB. 
Goetz 
EU 
cover, 
mid-A 
pp. 1S 
Howl: 
Canto 
Autun 
Johns: 
satelli 
Vol. € 
Refer 
Donn: 
Sensi 
Lond 
Refer 
Baatz 
M., Si 
Guide 
5.1. 
Ikono 
[magi 
S23 / 
This 
Max-] 
the C« 
USA.