Environmental Engineering Reference
In-Depth Information
with slopes varying between 0 and 22%, with an average of 6%.
The long-term yearly precipitation and open-water evaporation
is respectively 780 and 657 mm/year and is measured at the
Royal Meteorological Institute at Ukkel, located close to the
catchment border.
The land-use map of Flanders, produced by the Agency
for Geographic Information Flanders (AGIV), was used as a
reference data set to characterize land-use within the study area.
The original map has a resolution of 20 m and was resampled
to a 30 m cell size for modeling purposes (Fig. 18.4). A Digital
Elevation Model (DEM) at 30 m resolution was interpolated from
elevation contours for every 2.5 m and the river network. Both
were digitized from topographic maps at a scale of 1:10 000. The
DEM was generated in ArcInfo using TOPOGRID. A digital soil
map of the study area was obtained from AGIV, rasterized and
reclassified into 12 USDA soil texture classes based on textural
properties.
Two image data sets were used: an ortho-rectified high-
resolution image (IKONOS), acquired on 8 June 2000, to derive
a detailed high-resolution land-cover map, and a medium-
resolution image (Landsat ETM
+
), acquired on 18 October
1999, for estimating land-cover class proportions at sub-pixel
level. The IKONOS image has four multispectral bands that
capture the reflectance in the blue, green, red and near infrared
part of the spectrum at 4 m spatial resolution, and an addi-
tional panchromatic band at 1 m resolution. The Landsat ETM
+
multispectral image includes six spectral bands in the visual-
to-mid-infrared range that were used in this study, and has a
spatial resolution of 30 m. The Landsat image was geometrically
co-registered to the IKONOS image by a first-order polynomial
transformation. The RMS error on an independent set of con-
trol points was 5.78 m, which implies that on the average the
geometric shift in
x
and
y
between IKONOS and ETM
+
data
corresponds to less than 4% of the area of an ETM
+
pixel.
18.5.2
Impervious surfacemapping
To produce a high-resolution impervious surface map of the
study area, the IKONOS image was classified using a multiple
layer perceptron classification approach (MLP). Eleven land-
cover classes were distinguished: light and dark red built-up area
(mainly red tiled roofs), light, medium and dark gray built-up area
(consisting of building materials such as concrete, asphalt, slate),
bare soil, water, crops, shrub and trees, grass and shadow. Input
variables used for defining the classification model were the four
multispectral bands, the PAN band, the NDVI and two texture
measures (local variance, binary comparison matrix) calculated
on the PAN band. Training data for the classification were
obtained by digitizing about 200 training samples for each class.
Transformations of the input bands and a selection according to
their relative contribution to the overall information content were
accomplished with NeuralWorks Predict®.Thetransformed
input variables that were retained to perform the classification
were chosen from a set of five mathematical transformations
per original input band using a genetic-based variable selection
algorithm embedded in the software.
Legend
Rivers
Open water
Industry
High density built-up/Infrastructure
City centre
Roads/Highways
Low density built-up
Non-urban
N
W
E
1
0.5
0
1
2
S
Kilometers
FIGURE 18.4
Land-use map for the Upper Woluwe River catchment.
Search WWH ::
Custom Search