|
STATEMENT
OF THE PROBLEM
Monitoring the extent of impervious areas within the limits of the
City of Columbia is a requirement of the city government to support:
1) Estimates of urban development and
2) The use of hydrologic and water quality models applied to urban
watersheds.
PRESENT SITUATION
The City of Columbia wants to have an estimate of the impervious
area as a measure of urban development. The same information is
also needed as an input parameter to hydrologic models.
POTENTIAL BENEFITS
Monitoring of the extent of impervious surface areas in the City
of Columbia provides city management with a basic parameter to measure
urban development and helps the efforts to meet the more strict
EPA regulations of water quality in urban watersheds.
TECHNICAL APPROACH
TAKEN
The project was carried out using
as data source the four IKONOS-2 multispectral bands (0.45–0.53
mm, 0.52 – 0.61mm, 0.64 – 0.72 mm, 0.77 – 0.88
mm) with a 4 m spatial resolution. The IKONOS-2 data was obtained
on 30 April 2000 at 10:48 AM. Sun azimuth was 138.74 degrees and
sun elevation was 60.74 degrees. The nominal data collection azimuth
was 352.42 degrees and the nominal data collection elevation was
62.89 degrees.
Figure 1. Map showing the outline of the City of Columbia
and the IKONOS coverage used for impervious surface mapping.
Figure2: Multispectral IKONOS
image (RGB=4,3,2) with 4 meter spatial resolution
sharpened with IKONOS 1 meter panchromatic image of part of the
City of Columbia. Inset box shows outline of image used in Figure
3.
A Supervised Maximum Likelihood Classifier
was used to map impervious surfaces. Training sites were defined
initially for ten classes. Impervious surfaces were included into
two classes of concrete surfaces (roads and parking areas) and one
class for asphalt surfaces. The other seven classes that were defined
included two classes with different types of types of bare ground
(associated with different exposed soil horizons and poorly developed
grass), two classes for grass (well and fairly well developed grassland),
two classes for water bodies (identified by the presence or absence
of sediments in suspension) and woodlands. Single-pixel training
sets (Gong and Howarth, 1990) with approximately 100 pixels in each
were established for each class. The classification results were
combined into six major classes. The impervious areas were combined
into one class. Bare soil was kept in two classes. One of them could
be associated with urban and rural areas where the ground surface
was in part covered by vegetation consisting of a mixture of poorly
developed grass and stressed or dead crops and the other was associated
with construction sites alone. Woodlands and water were the other
two classes in the final ground cover map.
Figure 3 is a representative area outside downtown Columbia that
illustrates in more detail the six land cover/land use classes used
to present the results of the supervised classification to detect
impervious classes. When this figure is evaluated in reference to
Figure 4 that displays the classification results it is possible
to show the significance of the spatial resolution in obtaining
good classification results. All the impervious surfaces within
the apartment complex are identified with great accuracy. The same
can be said for the major roads located in the area.
Figure3: IKONOS 4-meter multispectral
image sharpened with IKONOS 1-m panchromatic band showing the Broadway
Village apartment complex in east Columbia.
Figure4: Result of supervised
maximum likelihood classification of IKONOS multispectral images
(4 meter) showing impervious surfaces (red). The overall accuracy
of this classification is approximately 91%.
Accuracy assessment was based on
the evaluation of classification results for the pure pixels selected
a priori throughout the image. The confusion or error matrix produced
for the Columbia area had an overall accuracy level of 91%.
Table1: Confusion or error
matrix for the classification of impervious surface in the City
of Columbia using IKONOS multispectral data (30 April 2000).
ANTICIPATED
IMPLEMENTATION PROBLEMS IN URBAN GOVERNMENT
The government agency interested in implementing this approach to
be used on a routine basis should have the appropriate image processing
hardware and software in place. Once the training and testing sites
for supervised classification of impervious surface have been defined,
it will be possible to produce land cover maps cover maps from IKONOS
multispectral data sets showing impervious surface with more than
90% accuracy.
ADDITIONAL WORK
The information contained in IKONOS pan band (0.45 – 0.90
mm) still has to be evaluated in combination with the multispectral
bands.
PROJECT PARTICIPANTS
Dr. Aderbal C. Corrêa and Mr. Janggam Adhityawarma of the
Center for Environmental Technology (CENTECH), Department of Civil
and Environmental Engineering, carried out the project for ICREST.
|
