Description of the terrain surface through digital elevation models (D
EMs) strongly depends on data collection methods and DEM data structur
es. For efficiency and availability reasons regular point distribution
s are most common, which yield artefacts such as depressions and prefe
rential flow directions. These facts need to be considered when natura
l phenomena are modelled, as is shown for handling depressions and for
estimation of flow paths and upslope contributing areas. Analysis of
the main reasons for the occurrence of depressions shows that they usu
ally better reflect the terrain than their surroundings. Thus, the mos
t common remedial method of raising depressions is rejected. Algorithm
s that 'cut' a flow path from the depression through its bounding barr
ier are favoured instead. Several flow routing algorithms are evaluate
d for their behaviour in regular grids. It is shown that the multiple
flow direction (mfd) algorithm that distributes water from a grid cell
to the lower of its eight neighbours proportionally to their elevatio
n differences (slope) exhibits correct flow directions and the best ro
tation invariance. It is suggested that the estimation of upslope cont
ributing areas (TCAs) is undertaken in two steps: first, a high qualit
y flow direction data set is derived by a well-behaved mfd algorithm o
r by subgrid modelling of flow paths; secondly, the upslope contributi
ng areas are obtained by counting the upslope elements. (C) 1998 John
Wiley & Sons, Ltd.