Modelling Catchment and Fluvial Processes and their Interactions - Environmental Modelling: Finding Simplicity in Complexity - page 177

modelling are usually between 5 and 50 m in resolu-
tion. Many hydrological parameters have been shown to
be sensitive to DEM resolution, coarser DEMs leading
to reduced slope and more sparse and smoothed river
topologies (see Chapter 5). Increased catchment wetness
and peak flow in the TOPMODEL has been observed
using coarser representations of topography (Zhang and
Montgomery, 1994) and a number of other models have
been shown to be sensitive to DEM resolution, although
the sensitivity is greatest for models with a high temporal
resolution (e.g. Yang et al ., 2001). Thus, there needs to
be appropriate balance between the spatial and temporal
resolution of a raster hydrological model.
Digital elevation models are usually constructed by the
interpolation of a raster grid from point-based altitude
data. Such data are acquired over small areas using field
survey with an EDM (electronic distance measurement)
system or, over larger areas directly from the computer-
ized interpretation of stereo orthophotographic surveys
or, most commonly, through the digitization of contour
lines on existing cartography (with the contours usually
having been derived from the prior manual interpreta-
tion of orthophotographic survey). Increasingly, DEMs
are available from radar (radio detection and ranging)
sensors such as used during the SRTM mission (Farr and
Kobrick, 2000), from satellite orthoimagery such as the
ASTER 2 GDEM 3 available for the area between 83 ◦ Nand
83 ◦ S, and from laser-based LIDAR (light detection and
ranging) at smaller scales.
Alternatives to the raster interpretation of topographic
data include triangular irregular networks (TINs -
Palacios-Velez and Cuevas-Renaud, 1986) - which vary
in size according to the topographic complexity - and
contour-based models (O'Loughlin, 1986; Moore et al .,
1988). Contour-based models discretize a topographic
surface into irregular-shaped cells made of two consecu-
tive contours connected by two lines, which represent the
steepest slope connecting the two contours. Raster cells
might not be the best representation for hydrological
purposes but they are certainly the most practical for
integration with pixel-based remotely sensed datasets
and raster GIS calculation tools and are thus the most
popular (if not always the most accurate) representation
of topography.
Digital elevation models for the area around Mount
Everest in Nepal are shown in Figures 11.1 and 11.2.

Figure 11.1 is the SRTM DEM 4 and Figure 11.2 is the
ASTER GDEM Version 2. 5 Both are displayed using
SAGAGIS. 6 In order to define the catchments one must
first define the potential flow paths for water.
Potential flow paths are defined according to the dif-
ference in elevation between the cell in which the flow
originates and the elevation of neighbouring cells. The
algorithms used can be classified into those that flow
into a single neighbouring cell and those with apportion
flow to multiple neighbours. The algorithms commonly
used include the so-called D8 algorithm (O'Callaghan
and Mark, 1984), the MF (multiple flow direction) algo-
rithm (Quinn et al ., 1991), the D- ∞ algorithm (Tarboton,
1997), the digital elevation model network - DEMON
(Costa-Cabral and Burges, 1994), the Rho-8 algorithm
(Fairfield and Leymarie, 1991) and the kinematic routing
algorithm (Lea, 1992). The D8 algorithm uses a nearest
neighbour approach to define the relationship between
a central cell and its eight contiguous neighbours (four
sides, four vertices) and defines the steepest down-slope
gradient along which all of the water is deemed likely to
flow. There are only eight possible flow directions, 45 ◦
apart, and all water flows to one of them. This algorithm
is used in many geographical information systems (GIS)
and some hydrological models but does tend to produce
rather linear flow networks when applied in areas of
smoothed or shallow topography. An example of a D8
drainage network is superimposed on a DEM for Tambito
in southern Colombia in Figure 11.3.
The Rho-8 algorithm randomly assigns the direction
of flow to downslope neighbours weighted according
to the degree of slope. The MF algorithm also uses
the eight nearest neighbours but then apportions flow
to all lower neighbouring cells based on the relative
magnitude of their slopes. The aspect-based kinematic
routing algorithm assigns flow direction according to the
calculated aspect. A plane is fitted to the corners of the
cell through interpolation of the elevations of the pixel
centres. Flow is routed as if it were a ball rolling along this
plane. The DEMON and D- algorithms use surface-

fitting algorithms to determine a surface-slope vector for
the central pixel and the direction of this vector is used
to proportion flow between the two neighbouring cells,

∞

4 Available from csi.cgiar.org and www.kcl.ac.uk/geodata.
5 The ASTER GDEM is a product of the Japanese Ministry for Econ-
omy, Trade and Industry (METI - www.ersdac.or.jp/GDEM/E/)
and NASA (http://asterweb.jpl.nasa.gov/gdem.asp).
6 System

2 Advanced Spaceborne Thermal Emission and Reflection Radio-
meter.
3 See www.gdem.aster.ersdac.or.jp/.

for Automated Geoscientific Analyses, www.saga-gis

.org/en/index.html.)