You are on page 1of 9

Estimating Surface Water Flow Distribution for Urban Runoff Simulation

Using Gridded Digital Terrain Models*


Petter Pilesj
Senior Lecturer, Remote Sensing & GIS Laboratory. Department of Physical Geography,
University of Lund, P.O. Box 118, S 221 00, Lund, Sweden. Phone: +46 46 222 9654,
E-mail:petter.pilesjo@natgeo.lu.se

Qiming Zhou
Associate Professor, Department of Geography, Hong Kong Babtist University,
Kowloon Tong, Kowloon, Hong Kong. Phone: +852 23395048
E-mail: qiming@hkbu.edu.hk

Introduction
Research over the last few decades has greatly improved our understanding of processes of
runoff generation and erosion. In humid landscapes, runoff is responsible for accelerated erosion
of agricultural land, the expansion of channel and rill networks, and decreased water quality in
streams, at substantial economic and environmental cost to society. In an urban environment,
however, the man-made landscape tends to be paved by artificial materials and human-induced
vegetation, thus the surface runoff is fundamentally controlled by the swage system and surface
morphology.

The land surfaces, however, have a complex three-dimensional topography and spatially variable
surface pavement and vegetation, which make prediction of the patterns of runoff difficult.
Empirical models of regional runoff estimation are incapable of predicting patterns of flow in
complex terrain. Furthermore, for a model to have a universal application it must be based upon
simple but realistic physical principles, and not upon empirical equations. The advent of detailed
digital terrain models (DTM) and GIS technology now make it possible to apply physical
principles to predict spatial patterns of runoff and flow accumulation under a range of
environmental conditions (Moore, et al., 1994).

In an urban environment the pattern of runoff is strongly controlled by topography and man-
made drainage systems, while the sub-surface infiltration plays a relatively less important role.
Specialist digital terrain models have been used to predict soil moisture and runoff generation
(e.g. Moore, et al., 1988) controlled by the surface flow. These models reduce the landscape to a
set of flow tubes perpendicular to contours, thus simplifying water movement to one dimension.
However, the terrain models are difficult to construct and cannot be applied to areas larger than
10 km2. A better approach for problems of land management is to predict runoff using grid-based
DTM which represents the main stream of today's digital elevation data supply.

Gridded DTM should be immediately applicable to three important aspects of hydrology in an


urban system. Firstly, patterns of overland flow can be predicted, secondly the concentration of

*
In Proceedings of the 3rd International Workshop on Urban 3D/Multi-media Mapping, Shibasaki, R. and Shi, Z.
(eds.), 12-14 September, Tokyo, CDROM.
overland flow into channels can be simulated, and thirdly, patterns of surface wash, rill, and
channel erosion can be predicted from the earlier results.
One foundation of the GIS-based terrain modelling is the algorithms that interpret the digital
elevation data into some critical parameters in the hydrological process of surface water, such as
slope and aspect (Bolstad and Stowe, 1994; Hodgson; 1995; Skidmore, 1989), terrain
topographic characteristic (Lee, 1994; Skidmore 1990; Tribe, 1991), soil wetness (OLoughlin,
1986), surface/volume approximation (Kalmar, et al., 1995), radiance and hillshading (Giles, et
al., 1995; Lapen and Martz, 1993; Zhou 1992), runoff modelling (Maidment et al. 1996), surface
water flow direction (Holmgren, 1994), flow accumulation and catchment area (Freeman, 1991;
Martz and Garbrecht, 1992; Pilesj, et al., 1998), topographic index (also known as ln(a/tan)
index, Quinn, et al., 1991, 1995) and drainage networks (Mark, 1984; Band, 1986; Meisels, et
al., 1995). It has been recognised, however, that some surface feature modelling algorithms may
produce poor results which lead to some significant artificial facts (known as artifacts) in the
output of interpretation. One of the main reasons is that the assumptions on the runoff pattern of
surface water implied by these algorithms are often over-simplified (Wolock and McCabe Jr.,
1995; Pilesj and Zhou, 1996). For example, commonly used gridded GIS flow direction models
assume water flows towards one of eight possible directions, resulting in highly distorted
drainage patterns.

Various improvements to the hydrological modelling algorithms have been reported. Among
these Pilesj, et al. (1998) showed that algorithms for estimation of flow distribution over a
gridded DTM should preferably include a judgement of the local topographical form (e.g.
convexities and concavities). However, the form-based algorithm proposed by them includes
advanced mathematical operations, and is therefore both computational demanding and
sometimes difficult to implement for the end users. A less-complicated and more efficient
method, therefore, is proposed in this paper.

The Aim of the Study


The aim of the study is to develop an algorithm that estimates surface flow distribution from all
cells (except for the border cells) in a gridded DTM. The distribution will be divided between the
eight neighbour cells surrounding all examined cells of the DTM. Each of the neighbour cells
can receive between 0 and 100 percent of flow from the centre cell (the cell being assessed), as
long as the sum of the distributed flow equals 100 (or 0 if the centre cell is a depression/sink).

Methodology
Like most of the proposed flow distribution algorithms that are
widely used today, the proposed algorithm is based on a filtering 3
procedure. Each cell in the DTM (except for the border cells) is
examined individually. The examination is based on the elevation 2
value of the cell as well as the elevation values of its eight
neighbours. The methodology described below will be based on this 1
3 x 3 kernel.

The nine numbers in the kernel all represent elevation values in a 0 1 2 3


point on the surface. In Figure 1 the positions of the nine elevation
points refer to locations in a local coordinate system, in which the Figure 1. The nine elevation
points in the kernel refer to
upper-left point has the x and y location of [0.5, 2.5], the upper mid locations in a local
point has the location of [1.5, 2.5], etc. coordinate system.

In order to estimate flow distribution from the centre cell (x = 1-2; y = 1-2, see Figure 1), the
elevation values between the nine known points have to be known. This can be done by
interpolation. Pilesj, et al. (1998) proposed an establishment of a second order trend surface in
order to create a continuous surface covering the kernel. However, an obvious disadvantage of
the trend surface is that it often changes the elevation values of the nine known points, which is
not at all desirable. Instead, we now propose that the continuous surface inside the x and y
coordinates 0.5-2.5 is estimated by the application of triangulated partition, which can be
regarded as the special application case of Triangulated Irregular Network (TIN) technique.

Triangulated partitioning

If we connect the centre point of the kernel to all its eight neighbour
points with straight lines, and frame the area by applying straight 3
lines at the border (Figure 2) we can look upon the kernel as a
continuous surface. The surface consists of eight facets, each with a 2
constant slope and aspect.
1
The idea behind the proposed algorithm for estimation of flow
distribution is to simulate flow from different locations on the centre
cell, and then examine in which neighbouring cells where these flow 0 1 2 3
paths end up. In Figure 3 we see three examples of this:
Figure 2. By connecting
known elevation point with
1. The flow from point 1 is easterly if the aspect of the facet is 90
straight lines we create a
degrees. The flow then crosses the edge to the next facet, where it continuous surface.
continues in a northeast direction It ends up (on the border line)
continues in a northeast direction. It ends up (on the border line)
in the upper right cell which means that point 1 in the centre cell 3
drains to the upper right cell.
2 1
2. If the aspect of the facet where point 2 is located is 270 degrees it
drains directly to the west, and the flow from point 2 ends up in 2
3
the cell left of the centre cell. 1

3. If the facet where point 3 is located is 225 degrees, and the facet
to the left has an aspect value of 135 degrees, the flow path is
0 1 2 3
initially southwest and then south. This means that point 3 in the
centre cell drains to the cell below the centre cell. Figure 3. Three examples of
flow paths from the centre
cell, based upon a
triangulated facet
approximation of the kernel.

By calculation of flow paths from a large number of points (e.g. 80) 2

in the centre cell, and examine where on the border line they end up,
it is possible to estimate the flow distribution from the centre cell
with a close approximation. Figure 4 gives an example of a
1.5
distribution of 80 points in the centre cell, resulting in 10 cells in
each facet. A total number of 80 points imply that each flow path
corresponds to 1.25 percent of the total flow from the centre cell.
1
1 1.5 2

Figure 4. An example of 80
starting points located in the
centre cell for estimation of
flow paths to neighbouring
cells.

The estimation of flow distribution from the centre cell can therefore be implemented by the
following steps of data processing:

1. Calculations of aspect values for each of the eight facets.

2. For all 80 points: calculation of flow path over the facet. In some cases (e.g. Point 2 in Figure
3) we will reach the borderline.

3. For all flow paths that do not reach the borderline at the above step (e.g. Point 1 in Figure 3):
calculation of flow path over adjacent facet. This step will be repeated until we reach the
borderline.
4. For all flow paths that are trapped between two facets (e.g. Point 3 in Figure 3): calculation
of flow path along the edge between the facets. This will be repeated until we reach the
borderline.

5. For all 80 end points of flow path: calculation of number of paths received by each
neighbouring cell. The number multiplied by 1.25 will correspond to the estimated flow into
that direction in percent.

Calculation of aspect value for each of the eight facets

Each individual triangular facet has the constant slope and aspect. The facet can thus be
described as a plane in a 3-dimensional space as:

z(x, y) = b0 + b1x + b2y (1)

where the coefficients b0, b1 and b2 can be resolved by solving an equation system based on the
three corner points. The aspect value for the facet can then be calculated according to:

b2 b
= 180 arctan + 90 1 (2)
b1 b1

where is the aspect of the facet in degrees. Here, the aspect is defined as the deviation from
north, and increases clockwise.

Calculation of flow path over a facet

When the aspect value of a facet is known it is relatively straightforward to calculate the flow
path. The flow path itself is not of any greater interest, but the end point of the flow is. If it ends
on the borderline, the flow analysis is ready. While the end point is on the border between facets,
further calculations are demanded.

Coordinates of an end point can be calculated for given the starting point and the aspect value,
since:

y 2 y1
tan = (3)
x 2 x1

where x1; y1 and x2; y2 are the coordinates of the starting point and the end point respectively.

The examples of this, applied on the upper left facet of the kernel, is presented in Figure 5 and
described below. Please note the examples assume three different aspect values of the facet.
a) If = 45, the flow path is represented by Vector 1 in Figure e
5, and the x and y coordinates of the end point (Point b) are 2.5
g d b
[1.3, 2.5]. Using Equation 3, given known , the coordinates 1
of the starting point (Point a [1.1, 2.3]) and the y coordinate of a
the end Point b (2.5), the Equation 3 can be easily solved, and 2
h
the x coordinate at the end point is calculated to 1.3.
c
b) If = 135, the flow path is represented by Vector 2 in Figure
5, and the x and y coordinates of the end point (Point c) are f
1.5
[1.5, 1.9]. In this example the x coordinate of the end point 0.5 1.5
instead of the y coordinate is known. Note that the use of a
Figure 5. Three examples of
known y coordinate (2.5 from example a)) is not valid since it how end points can be
corresponds to a case where = 315 (Point d). calculated on facets with
different aspect values.

c) If = 225, the case becomes more complicated to calculate the end point (Point h in Figure
5). However, since it is known that the end point will be located on the diagonal line between
Point f and g, and along this line we have:

x xf = y yf

where xf and yf represent x and y coordinates at Point f, respectively, the relationship


between the x and y coordinates of the end point can be solved as:

1 .5 x = y 1 .5 y = 3 x (4)

We can now replace y2 in Equation 3 by (3 - x2) and thus to solve x2 and y2 ([0.9, 2.1]). Note
that the use of a known x or y coordinate (1.5 or 2.5) is not valid here.

The example presented above is performed on the upper left facet of the kernel. Although the
other seven facets (Figure 2) have different corner coordinates the procedure in calculating the
end points are the same, because the x and y coordinates on the border as well as the relationship
between x and y coordinates on the diagonal line are always known.

Calculation of flow path over the adjacent facet

All flow paths over adjacent facets can be calculated according to the methodology presented
above. The only difference is that the starting point is not given by the initial position as the
previous cases, but by the end point from the previous calculation. Normally a new end point
will be calculated, either on the borderline or on the edge with the adjacent facet. In the latter
case the calculation will be repeated until we reach a borderline.
Calculation of flow path between two facets

If a flow path cannot cross the edge between facets, the b


case of "flow is trapped at the edge" is found. An
example of this is shown in Figure 6, where the left
facet has an aspect value of 45 degrees and the right
facet has an aspect value of 315 degrees. In this case the
flow path along the edge (representing a valley line) is
assumed until the borderline is reached. The end point
of flow path is easily calculated by comparison of the
a
two elevation values at Point a and b where the lower
point (in this case Point b), is the end point. Figure 6. If the flow is trapped at the edge
of two facets it will follow the valley line.

If the flow direction on the edge is not towards the borderline (e.g. towards Point a, i.e. the
centre point, in Figure 6) it will be directed through the steepest downhill path from the centre
point.

Calculation of number of paths received by each neighbouring cell

In order to estimate the flow proportion from the centre cell to the eight neighbours, the number
of end points on the borderlines are counted. Depending on position on the borderline, each point
contributes 1.25 percent flow in one of the eight directions. Referring to Figure 3, for example,
the following range of x and y coordinates are used in order to decide which neighbouring cell to
which the individual end point should be linked:

upper left cell: x = 0.5 and y = 2-2.5 or x = 0.5-1 and y = 2.5


upper cell: x = 1-2 and y = 2.5
upper right cell: x = 2-2.5 and y = 2.5 or x = 2.5 and y = 2-2.5
left cell: x = 0.5 and y = 1-2
right cell: x = 2.5 and y = 1-2
lower left cell: x = 0.5 and y = 0.5-1 or x = 0.5-1 and y = 0.5
lower cell: x = 1-2 and y = 0.5
lower right cell: x = 2-2.5 and y = 0.5 or x = 2.5 and y = 0.5-1

Discussion and Concluding Remarks


Flow paths over flat surfaces, on whole kernel or individual facets, are treated according to the
methods presented in Pilesj, et al. (1998). This means that the flow is assumed to take the
shortest path over the flat surface in the direction towards the closest downhill facet.

The algorithm presented here could deliver more efficiency in hydrological modelling
computation while the sub-surface infiltration is less important. It will also present a more
realistic result comparing with some commonly used hydrological modelling algorithms used in
today's GIS. The main reason for this is that it combines the advantage of the form based
algorithm presented by the previous study (Pilesj, et al., 1998) with a triangulated partitioning
algorithm that does not alter the initial elevation values. Another improvement is that it estimates
flow from the whole area of the target grid cell (by the use of the 80 sample points) instead of
from the centre point only. Although one may argue that the use of sample points would still
cause inaccuracy and uncertainty in the outcome, the close approximation can ensure that these
uncertainty will be minimal comparing with other possible sources of errors.

References
Band, L.E., 1986. Topographic partition of watersheds with digital elevation models, Water
Resources Research, 22: 15-24.

Botlstad, P. V., Stowe, T., 1994. An evaluation of DEM accuracy: elevation, slope, and aspect,
Photogrammetric Engineering and Remote Sensing, 60: 1327-1332.

Freeman, T.G., 1991. Calculating catchment area with divergent flow based on a regular grid,
Computers and Geosciences, 17: 413-422.

Giles, P. T., Chapman, M. A., Franklin, S. E., 1995. Incorporation of a digital elevation model
derived from stereoscopic satellite imagery in automated terrain analysis, Computers and
Geosciences, 20: 441-460.

Hodgson, M., 1995. What cell size the computed slope/aspect angle represent? Photogrammetry
Engineering and Remote Sensing, 61: 513-517.

Holmgren, P., 1994. Multiple flow direction algorithms for runoff modelling in grid based
elevation models: an empirical evaluation, Hydrological Processes, 8: 327-334.

Kalmar, J., Papp, G., Szabo, T., 1995. DTM-based surface and volume approximation,
geophysical applications, Computers and Geosciences, 21: 245-257.

Lapen, D. R., Martz, L., 1993. The measurement of two simple topographic indices of wind
sheltering-exposure from raster digital elevation models, Computers and Geosciences,
19: 769-779.

Lee, J., 1994. Digital analysis of viewshed inclusion and topographic features on digital elevation
models, Photogrammetric Engineering and Remote Sensing, 60: 451-456.

Maidment, D.R., Olivera, J.F., Calver, A., Eatherral, A. and Fraczek, W., 1996. A unit
hydrograph derived from a spatially distributed velocity field, Hydrological Processes.

Mark, D.M., 1984. Automated detection of drainage networks from digital elevation models,
Cartographica, 21: 168-178.
Martz, L.W. and Garbrecht, J., 1992. Numerical definition of drainage network and
subcatchment areas from digital elevation models, Computers and Geosciences, 18: 747-
761.

Meisels, A., Raizman, S., and Karnieli, A., 1995. Skeletonizing a DEM into a drainage network,
Computers and Geosciences, 21: 187-196.

Moore, I.D., O'Loughlin, E.M. and Burch, G.J., 1988. A contour-based topographic model for
hydrological and ecological applications, Earth Surface Processes and Landforms, 13:
305-320.

Moore, I.D., Grayson, R.B. and Ladson, A.R., 1994. Digital terrain modelling: a review of
hydrological, geomorphological, and biological applications, in Beven, K.J. and Moore,
I.D. eds., Terrain Analysis and Distributed Modelling in Hydrology, John Wiley & Sons,
Chichester, UK: 7-34.

OLoughlin , E. M., 1986. Prediction of surface saturation zones in natural catchments by


topographic analysis, Water Resources Research, 22: 794-804.

Pilesj, P., Zhou, Q. and Harrie, L., 1998. Estimating flow distribution over digital elevation
models using a form-based algorithm. Geographical Information Sciences, 4: 44-51.

Quinn, P.F., Beven, K.J. and Lamb, R., 1995. The ln(a/tan) index: how to calculate it and how
to use it within the TOPMODEL framework, Hydrological Processes, 9: 161-182.

Quinn, P.F., Beven, K.J., Chevallier, P. and Planchon, O., 1991. The prediction of hillslope flow
paths for distributed hydrological modelling using digital terrain models, Hydrological
Processes, 5: 59-79.

Skidmore, A.K., 1989. A comparison of techniques for calculating gradient and aspect from a
grided digital elevation model, International Journal of Geographical Information
Systems, 3: 323-334.

Skidmore, A.K., 1990. Terrain positions mapped from a gridded digital elevation model,
International Journal of Geographical Information Systems, 4: 33-49.

Tribe, A., 1991. Automatic recognition of valley heads from digital elevation models, Earth
Surface Processes and Landforms, 16: 33-49.

Wolock, D.M. and McCabe Jr., G.J., 1995. Comparison of single and multiple flow direction
algorithms for computing topographic parameters in TOPMODEL, Water Resources
Research, 31: 1315-1324.

Zhou, Q., 1992: Relief shading using digital elevation models, Computers and Geosciences, 18:
1035-1045.

You might also like