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 245

separating impervious surface materials from other, spectrally
similar land-cover types. Indeed, dark materials like asphalt or
roofing may be difficult to distinguish from other low-albedo sur-
face types like water and shadowed surfaces (Hodgson et al ., 2003;
Dare, 2005; Lu and Weng, 2007). Other impervious materials
may be spectrally similar to bare soil (Thomas, Hendrix and
Congalton, 2003; Van de Voorde, De Genst and Canters, 2007).
Various approaches have been suggested to improve the separa-
tion of impervious from non-impervious surface types. Lu and
Weng (2009) propose a hybrid approach that combines the use
of a decision tree classifier with an unsupervised classifier and
requires intervention from an image analyst to separate impervi-
ous surfaces from water and shadow in vegetated areas. De Roeck,
Van de Voorde and Canters (2009) apply an object-based, hierar-
chical classification approach, using texture measures calculated
at the level of image objects to complement the spectral data, and
improve the distinction between impervious surfaces, shadowed
areas and bare soil. Thomas, Hendrix and Congalton (2003) and
Van de Voorde, De Genst and Canters (2007) suggest object-
based post-classification approaches to improve the accuracy of
an initial classification, using expert-based rules that rely on con-
text information. Others have proposed the use of external data,
such as elevation, to reduce the confusion between impervious
and non-impervious surface types (Hodgson et al ., 2003).
Despite the advantages of deriving impervious surface maps
directly from high-resolution satellite imagery, the limited foot-
print and the relatively high cost of such images may pose
difficulties if impervious surface maps are required for larger
areas. A less expensive and more efficient alternative to map the
spatial distribution of imperviousness in such cases is to develop
models that allow estimation of the degree of imperviousness
inside pixels of medium-resolution imagery (Landsat ETM

and water balance simulation by Liu et al . (2003). Batelaan and
De Smedt (2007) derived from the original a quasi-steady state
version, which is mainly used for recharge prediction.
Figure 18.1 (left) shows the hydrological processes simulated
in the model. These processes are computed at grid cell level, and
water balance calculations are performed in four control volumes:
interception, depression, root zone and saturated subsurface. The
processes are simulated in a cascading manner starting from a
precipitation event, followed by interception, depression stor-
age, surface runoff, infiltration, evapotranspiration, percolation,
interflow and groundwater flow. In addition, the water balance
of each grid cell is divided into impervious, vegetated, bare
soil and open water parts to account for the non-uniformity in
land-use/land-cover in individual cells. This aspect is especially
for urban areas very important since on the typical resolution
of remotely sensed data, land-cover shows strong heterogeneity
(Fig. 18.1, right).
The main input of WetSpa, to describe various hydrological
processes, is hydrometeorological data (precipitation, evapo-
transpiration and temperature) and raster maps (topography,
land-use and soil). Figure 18.2 presents the relationship between
the three main input maps, digital elevation model, land-use
and soil map, and the WetSpa model parameters, which are
derived by using physical and empirical equations. The main
outputs of the model include river flow hydrographs, which can
be defined for any location in the channel network, and spatially
distributed hydrologic characteristics of the catchment such as
evapotranspiration, recharge and runoff.

18.4.1 Surface runoff

,
ASTER). Over the past years many sub-pixel regression and
sub-pixel classification methods have been proposed to derive
the fraction of impervious surfaces within a medium-resolution
pixel directly from the spectral data, including methods based
on regression analysis (Gillies et al ., 2003; Yang and Liu, 2005;
Bauer, Loffelholz and Wilson, 2008), linear spectral unmixing
(Ji and Jensen, 1999; Phinn et al ., 2002; Rashed et al ., 2003;
Wu and Murray, 2003), artificial neural networks (Flanagan
and Civco, 2001; Wang and Zhang, 2004, Pu et al ., 2008) and
regression trees (Huang and Townshend, 2003; Yang et al ., 2003;
Xian, 2006). For a comparison of different methods the reader is
referred to Weng and Hu (2008), Yuan, Wu and Bauer (2008),
and Van de Voorde, De Roeck and Canters (2009). Although
the sub-pixel approach does not allow one to achieve the same
degree of accuracy as when high-resolution data would be used,
most studies report an average per pixel proportional error in
the estimation of impervious surfaces that varies between 10%
and 20%, depending on the characteristics of the study area and
the way the validation is done. Recently, sub-pixel approaches
have also been applied to high-resolution satellite imagery to
improve the accuracy of impervious surface maps in areas with
high proportions of mixed pixels (Mohapatra and Wu, 2007; Lu
and Weng, 2009; Wu, 2009).

+

One of the most important processes with respect to urban areas
is surface runoff, which is therefore described in more detail.
For the formulation of other processes we refer to Liu and De
Smedt (2004). Surface runoff is calculated in the model by a
modified rational method as:

C p P n ( θ/θ s ) α R s = (18.1)

where R s [LT − 1 ] is the rate of surface runoff, C p [-] is a potential
runoff coefficient, P n [LT − 1 ] is the rainfall intensity after canopy
interception, θ and θ s [L 3 L − 3 ] are actual and saturated soil mois-
ture content, and α [ - ] is an empirical exponent related to the
rainfall intensity. The potential runoff coefficient C p is a measure
of rainfall partitioning capacity, depending upon slope, soil type
and land-use combination. Default potential runoff coefficients
for different slope, soil type and land-cover under the condition of
near saturated soil moisture are interpolated from Kirkby (1978),
Chow, Maidment and Mays (1988), Browne (1990), Fetter (1980)
and Smedema and Rycroft (1988). A lookup table (Table 18.1)
has been built to relate the potential runoff coefficient to dif-
ferent combinations of slope, soil type and land-use (Liu and
De Smedt, 2004). To simplify this table, land-use classes are
reclassified into five classes as forest, grass, crop, bare soil and
impervious area. In addition, surface slope is discretized into four
classes. But, in order to estimate the potential runoff coefficient
on the basis of a continuous slope, a simple linear relationship
between potential runoff coefficient for the discretized slope
classes and surface slope is used. The potential runoff coefficients
for impervious (including open water) surface are set to 1.

18.4 TheWetSpamodel

WetSpa was originally developed by Wang, Batelaan and De
Smedt (1996) and rewritten with a focus towards flood prediction