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 244

an application of the Stanford Watershed Model (James, 1965).
The Stanford Watershed Model (Crawford and Linsley, 1966)
and its follow up HSPF (Bicknell et al ., 1997) utilize time series
meteorology data to simulate hydrological processes in both
pervious and impervious land segments. Although the integration
of land-use related information in lumped watershed models like
HSPF allows one to take account of differences in physical
characteristics within the catchment, linked to different types of
land-use, the lumped approach does not define physical processes
in a spatially explicit way (NOAA, 2006). Spatial interaction
between neigbboring locations is not considered in the modeling.
This can be seen as a strong limitation, especially in urbanized
areas, which are often characterized by highly complex land-use
patterns with a strong impact on runoff processes.
The need to better deal with the spatial characteristics of
hydrological processes has led to the development of distributed
hydrological models, which explicitly take into account the phys-
ical characteristics of every location, within the limits of a certain
discretization defined by the model's grid cell size. Hence, there
is a strong tendency in research towards improving hydrological
models to integrate as much as possible different hydrological
processes taking place in the atmosphere, vegetation, land sur-
face, soil and subsurface at the natural scale of a catchment.
Most models opt for a physical description of the hydrological
processes instead of e.g. statistical relationships. Physically based
models have the obvious advantage that it can be assumed that
predicted consequences of changed conditions hold within the
calibrated range of the model and more importantly that pre-
dicted consequences can be explained in function of changed
physical conditions. Early examples and trendsetters were TOP-
MODEL (Beven and Kirkby, 1979) and the SHE model (Abbott
et al ., 1986), while presently SWAT (Gassman et al ., 2007) is
widely used. For a recent review we refer to Todini (2007). Devel-
opments in GIS technology combined with an increasing use of
remote sensing as a data source for producing spatially distributed
data are the backbone of this trend in hydrological modeling.
Remote sensing provides hydrologists spatial as well as tem-
poral data on physical properties of the earth's surface that may be
used for estimating hydrological model parameters (Schmugge
et al ., 2002). Land-use datasets, which are usually obtained
through visual interpretation of aerial photography or satellite
data, are probably the most common remote sensing derived
product used in hydrological modeling. Just like in lumped
modeling approaches, also in distributed hydrological modeling
relevant model parameters, such as vegetation type and vegeta-
tion fraction, leaf area index, interception capacity, root depth
and Manning coefficient, are often directly inferred from land-
use, based on general look-up tables found in the literature (Liu
et al ., 2004). This, of course, is not an optimal approach. Because
land-use types are in most cases quite broadly defined, and have
a functional rather than a physical meaning (e.g., built-up area,
agriculture, forest), surface related parameters that are relevant in
hydrological modeling may show substantial variation within a
particular land-use patch and among patches of the same land-use
type. This is particularly so in urbanized areas, which are charac-
terized by a strong landscape heterogeneity. When using land-use
related look-up tables for hydrological parameter estimation in
such areas, spatially-explicit detail on land surface properties that
may be important in modeling and understanding of dynamic
hydrological processes is discarded (Dams et al ., 2009).
In recent years, there has been a growing interest in the devel-
opment of methods to derive various land-related parameters

that are relevant for hydrological modeling directly from remote
sensing data (Boegh et al ., 2004). A lot of work has been done
on the mapping of impervious surface distribution, both at pixel
level, using high-resolution multispectral data, as well as at sub-
pixel level, using data from medium-resolution multispectral and,
more recently, hyperspectral sensors. An overview of different
approaches for mapping impervious surface cover from remotely
sensed data is given below. Besides mapping of impervious sur-
faces to improve runoff prediction, current hydrological remote
sensing research also aims at improving spatial and temporal
parameterization of other hydrological processes. An overview
of this research is beyond the scope of this chapter, but excellent
reviews have been published on topics such as estimation of soil
moisture (Anderson and Croft, 2009; Petropoulos et al ., 2009),
mapping of evapotranspiration (Schmugge et al ., 2002; Kalma,
McVicar and McCabe, 2008; Li et al ., 2009), and flood extent
mapping (Schumann et al ., 2009).
The remainder of this chapter focuses on the use of informa-
tion on the distribution of impervious surface cover derived from
remote sensing in spatially distributed hydrological modeling.
More in particular, a case study will be presented, demonstrat-
ing the impact of different remote sensing based methods for
characterizing the distribution of impervious surfaces on runoff
estimation in an urbanized watershed, as well as the effect the
different methods have on the assessment of peak discharges
at the outlet of the catchment. Study area is the Woluwe, a
strongly urbanized watershed in the urban fringe of the Brussels
Capital Region.
In Section 18.3 a short review of different approaches for
mapping impervious surface cover from remote sensing data is
provided. Section 18.4 gives an overview of the WetSpa model, a
grid-based spatially distributed hydrological model that has been
used in the case study on the Woluwe. Section 18.5 describes the
set-up and the results of the case study, and demonstrates the need
for detailed information on surface imperviousness for spatially
explicit modeling of hydrological processes in urbanized areas.

18.3 Impervious surface
mapping

Field inventorying and visual interpretation of large-scale,
orthorectified aerial photographs are the most reliable methods
to map impervious surfaces. However, because these methods
are very time-consuming, they can in practice only be applied
to relatively small areas. Satellite imagery, obtained from high-
resolution multispectral sensors like IKONOS or Quickbird,
offers an interesting alternative for producing maps of surface
imperviousness. Although high-resolution imagery does not
provide the same level of detail as large-scale aerial photographs,
the use of automated or semiautomated image interpretation
methods, exploiting the multispectral information content of
the imagery, substantially reduces the effort that is required to
produce reliable information on the distribution of impervious
surfaces from this data.
Several studies have focused on the mapping of impervious
surface cover from high-resolution satellite data. Because of the
limited spectral resolution - most high-resolution sensors that
are commonly used have only four spectral bands: blue, green,
red, near infrared - a major challenge in these studies lies in