Use of impervious surface data obtained from remote sensing in distributed hydrological modeling of urban areas - Urban Remote Sensing: Monitoring, Synthesis and Modeling in the Urban Environment - page 251

0.00
0.06
0.13
0.19
0.25
0.31
0.38
0.44
0.50
0.56
0.63
0.69
0.75
0.81
0.88
0.94
1.00

FIGURE 18.5 Maps of impervious surface fractions at the cell size of the WetSpa model (30 m), obtained by sub-pixel
classification of Landsat ETM

+ data (left) and by spatial aggregation of the IKONOS-derived land-cover map to the 30 m

resolution (right).

TABLE 18.4 Mean error and mean absolute error of sub-pixel class proportions for the four major land-cover classes, for cell sizes of
30 m, 60 m and 90 m.

Mean error Mean absolute error

Cell size Impervious Vegetation Bare soil Water Impervious Vegetation Bare soil Water

30 m -0.018 0.015 -0.049 0.011 0.1030 0.1017 0.0593 0.0166

60 m -0.017 0.014 -0.048 0.011 0.0752 0.0720 0.0514 0.0152

90 m -0.019 0.015 -0.045 0.011 0.0611 0.0590 0.0467 0.0153

many urban pixels are assigned to water, although no water is
present within these pixels. This phenomenon is most clearly
observed in dense urban areas and is most likely caused by the
presence of shadow, which is spectrally similar to water. As such,
the proportion of water identified by the sub-pixel estimator
should better be interpreted as water/shade, as it may point to the
presence of both components. This may also explain the slight
overestimation of water ( + 1 . 1%). The mean absolute error for
impervious surfaces and vegetation, which are the two dominant
classes in the image, is around 10%. Aggregation to cell sizes
of 60 m and 90 m reduces this error to 7.5% and 6.1% for
impervious surfaces, and to 7.2% and 5.9% for vegetation.
Figure 18.6 summarizes the different stages in the process-
ing of the high-resolution (IKONOS) and medium-resolution
(Landsat ETM

coefficients and for estimating peak discharges at the outlet of
the catchment, using the WetSpa model. The spatial distribution
of the WetSpa model parameters for the Woluwe study area were
calculated based on the DEM, the soil map, the land-use map, and
field measurements of hydraulic properties of the river channel.
The period of simulation covers 4 days from 3 May 2005, 1.00 am
till 6 May 2005, 9.00 am. This period was selected because of
the typical occurrence of a number of spring storms. Since no
continuous discharge measurements where available, no formal
calibration of the model could be performed. However, some
global parameters were adjusted during an initial trial and error
calibration procedure, while spatial model parameters were kept
constant. In the remainder of this section the effect of the three
different land-cover input scenarios on the simulated runoff for
the different storms in the period is examined. The three discussed
scenarios correspond to a non-distributed, semidistributed and
fully distributed approach for incorporating information on
imperviousness in the runoff modeling.

+

)imagery.

18.5.3 Impact of land-cover
distribution on estimation of peak
discharges

18.5.3.1 Scenario 1: Non-distributed
approach
The first scenario that was examined corresponds to a situation
that is standard practice in hydrological modeling: the case where
no information is available about different types of urban land-
use, nor about the distribution of impervious surfaces within the
urban area. To simulate runoff for this scenario all urban classes
in the Flemish land-use map were aggregated into a single urban

Starting with the land-use map of Flanders, three scenarios were
defined to ''enhance'' the map, using the land-cover information
derived from the IKONOS and Landsat ETM + images. The three
scenarios correspond to a gradual increase of information on the
spatial distribution of land-cover within the urban area. These
scenarios were used as a starting point for calculating runoff