You are on page 1of 10

Journal of Hydrology (2007) 332, 30– 39

available at www.sciencedirect.com

journal homepage: www.elsevier.com/locate/jhydrol

How does modifying a DEM to reflect


known hydrology affect subsequent terrain analysis?
a,*
John Nikolaus Callow , Kimberly P. Van Niel a, Guy S. Boggs b

a
University of Western Australia, School of Earth and Geographical Sciences, 35 Stirling Highway,
Crawley, Western Australia 6009, Australia
b
Charles Darwin University, School of Science and Primary Industries, Darwin 0909, Australia

Received 2 September 2005; received in revised form 26 April 2006; accepted 23 June 2006

KEYWORDS Summary Many digital elevation models (DEMs) have difficulty replicating hydrological pat-
Digital elevation model; terns in flat landscapes. Efforts to improve DEM performance in replicating known hydrology
Geographic information have included a variety of soft (i.e. algorithm-based approaches) and hard techniques, such
system; as ‘‘Stream burning’’ or ‘‘surface reconditioning’’ (e.g. Agree or ANUDEM). Using a represen-
Hydrology; tation of the known stream network, these methods trench or mathematically warp the original
Hydrological modelling DEM to improve how accurately stream position, stream length and catchment boundaries rep-
licate known hydrological conditions. However, these techniques permanently alter the DEM
and may affect further analyses (e.g. slope). This paper explores the impact that commonly
used hydrological correction methods (Stream burning, Agree.aml and ANUDEM v4.6.3 and
ANUDEM v5.1) have on the overall nature of a DEM, finding that different methods produce
non-convergent outcomes for catchment parameters (such as catchment boundaries, stream
position and length), and differentially compromise secondary terrain analysis.
All hydrological correction methods successfully improved calculation of catchment area,
stream position and length as compared to using the DEM without any modification, but they
all increased catchment slope. No single method performing best across all categories. Differ-
ent hydrological correction methods changed elevation and slope in different spatial patterns
and magnitudes, compromising the ability to derive catchment parameters and conduct sec-
ondary terrain analysis from a single DEM. Modification of a DEM to better reflect known hydrol-
ogy can be useful, however knowledge of the magnitude and spatial pattern of the changes are
required before using a DEM for subsequent analyses.
ª 2006 Elsevier B.V. All rights reserved.

Introduction

* Corresponding author. Tel.: +61 8 6488 3435; fax: +61 8 6488 Detecting surface hydrology features in flat landscapes
1054. using Digital Elevation Models (DEMs) is a known problem,
E-mail address: jcallow@segs.uwa.edu.au (J.N. Callow). and new analysis methodologies (Garbrecht and Martz,

0022-1694/$ - see front matter ª 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.jhydrol.2006.06.020
How does modifying a DEM to reflect known hydrology affect subsequent terrain analysis? 31

1997; Gallant and Dowling, 2003) and ways to recondition racy and precision in many areas, flatter regions which have
DEMs for improved performance have been suggested a mean drop per pixel up to two orders of magnitude lower
(Hutchinson, 1989; Hellweger, 1996; Soille et al., 2003). than the catchment mean require a DEM with centimetre to
Whilst some of these methods are soft solutions (e.g. flow sub-centimetre accuracy. DEMs of this accuracy are not
algorithms), others permanently alter the DEM. Soft meth- available for this study area, or for many others at the
ods are relatively well understood (Tarboton, 1997; Jones, catchment or sub-catchment scale.
2002; Tarboton and Ames, 2002) and while impacts of DEM The minimum resolution and precision of a DEM are
error on secondary terrain analyses have been studied (Fish- important when analysing surface hydrology (Quinn et al.,
er, 1998; Holmes et al., 2000; Van Niel et al., 2004), the im- 1991; Hutchinson and Dowling, 1994; Wolock and Price,
pact of permanent alteration of a DEM on any further 1994; Zhang and Montgomery, 1994; Gyasi-Agyei et al.,
analysis has not been investigated. This paper explores the 1995; Gallant and Hutchinson, 1997; Walker and Willgoose,
impact that altering DEM cell values to better represent 1999; McMaster, 2002). However, the required level of res-
the known hydrology of a landscape (hydrological correc- olution and precision is often not available, particularly for
tion) has on the overall nature of a DEM, and how the vast, flat landscapes. In these circumstances, methods of
changes impact on any further terrain analysis. It demon- DEM hydrological correction must be used. However, further
strates different, non-convergent outcomes for secondary analyses are often conducted. There is little understanding
terrain analyses due to the type of method used. of the effect that the introduction of additional error by
DEMs are widely used for modelling surface hydrology. hydrological correcting a DEM has on subsequent analysis.
Analyses include the automatic delineation of catchment In the study catchment, land managers have diverted
areas (O’Callaghan and Mark, 1984; Martz and De Jong, flow by constructing earth banks, successfully rerouting flow
1998), development of terrain characteristics (Moore within and between sub-catchments. Thus, alteration of the
et al., 1991) and drainage networks (Fairfield and Leymarie, DEM to replicate the new hydrological reality of the study
1991), the detection of channel heads (Montgomery and Die- catchment is required. The rest of the paper discusses sur-
trich, 1988; Montgomery and Dietrich, 1992), estimating face reconditioning methods and explores their impact on
hydrology and soil moisture (Beven and Kirkby, 1979; the nature of the DEM and subsequent terrain analy-
O’Loughlin, 1986; McKenzie et al., 2003; English et al., ses (slope) conducted using the hydrologically corrected
2004), determination of flow accumulation (Peuker and DEMs.
Douglas, 1975) and flow direction and routing (Tarboton,
1997; Tarboton, 2002), and automated extraction of param-
eters for hydrological or hydraulic modelling (Doan, 2000; Study setting and methodology
Ackerman, 2002).
The accuracy with which a DEM is able to replicate the The Kent River catchment in southwest Western Australia
hydrological reality of a catchment is determined by the (Fig. 1) drains an ancient and very flat landscape, where
scale of capture (i.e. cell size), the precision (i.e. vertical stream gradients fall only tens of centimetres per kilometre
accuracy and relative accuracy between adjacent, up- (Salama et al., 1993; Callow and Smettem, in press). A DEM
stream and downstream cells) and strength of the landscape of the study region was purchased from the Department of
gradient (i.e. flatness) (Gyasi-Agyei et al., 1995; Quinn Land Information (DLI). Generated as part of the Land Mon-
et al., 1991; Hutchinson and Dowling, 1994; Wolock and itor Project (Allen and Beeston, 1999; Caccetta et al.,
Price, 1994; Zhang and Montgomery, 1994; Gallant and 2000), the DEM has a grid size of 10 m and a vertical accu-
Hutchinson, 1997; Walker and Willgoose, 1999; McMaster, racy of ±1 m (Caccetta et al., 2000). This DEM is at the high-
2002). Algorithms used to represent and extract real-world er end of available, catchment-scale DEMs in terms of the
processes from a DEM are also significant (Tarboton, 1997; grid size and vertical accuracy, compared to products by
Jones, 1998; Tarboton and Ames, 2002; Gallant and Dow- professional agencies such as the United States Geological
ling, 2003). Survey (USGS 30 m DEMs), Shuttle Radar Topography Mission
Thus, a number of methods for DEM improvement have (SRTM 30 m or 90 m products) and British Ordnance Survey
been suggested, such as the removal of spurious sinks (Jen- (Land Form Profile 10 m DEMs). While DEMs created from
son and Dominque, 1988; Soille et al., 2003), incorporation Light Detecting And Ranging (LIDAR) methods offer better
of vector stream data for stream burning (Maidment, 1996; resolution and precision, they are not typically available
Mizgallewicz and Maidment, 1996; Saunders, 1999) or sur- at the catchment scale and are usually produced for a spe-
face reconditioning (Hutchinson, 1989, 2004; Hellweger, cific project, rather than being widely available.
1996). Further problems in delineating surface hydrology While pit-filling and flow direction algorithms are well
are caused by human impacts that have altered catchment known to affect hydrological analysis (e.g. Tarboton,
hydrological processes. For example, the incorporation of 1997; Tarboton and Ames, 2002; Soille et al., 2003), this pa-
road data into modelling of surface runoff has been identi- per focuses on effects of hydrological correction algo-
fied as a significant process (Croke et al., 1999; Luce, 2002; rithms. All DEMs used in this paper were prepared using a
Duke et al., 2003). pit-filling algorithm (Jenson and Dominque, 1988), and flow
Flat landscapes, such as the study catchment, which has direction calculated using a D-8 algorithm (O’Callaghan and
a mean drop per pixel of 0.32 m, require a vertical accuracy Mark, 1984). This is the most commonly used approach when
of 32 cm for a 10 m DEM, according to the Gyasi-Agyei crite- preparing and processing DEMs for hydrological analysis
ria (Gyasi-Agyei et al., 1995). The best catchment-scale DEM (Saunders, 1999; Jones, 2002). Hydrological modelling with
available for the study catchment has a vertical accuracy of a pit-filled version of the raw land monitor DEM revealed
±1 m and, while it can delineate hydrology with good accu- disagreement with the expected water catchment bound-
32 J.N. Callow et al.

Figure 1 The upper Kent River study catchment, showing how the raw DEM incorrectly routes flow out of the Kent River catchment
and into the neighbouring Frankland River catchment to the west.

aries and stream network data in some (flatter) sub-catch- proaches that are likely to have variable effects on second-
ments. The raw DEM routes flow through a series of lakes ary terrain analysis.
in the northwest of the catchment, and into the neighbour- Stream burning was developed to improve the replica-
ing Frankland River catchment (Fig. 1). Field work and con- tion of stream positions by using a raster representation of
sultation with local farmers and land managers identified a vector stream network to trench known hydrological pat-
how flow should be routed through the lakes and channels terns into a DEM at a user specified depth (Maidment, 1996;
and into the Kent River catchment, as depicted in Fig. 2. Mizgallewicz and Maidment, 1996; Saunders, 1999). Depths
This study investigates three algorithms for modifying a of 1 m, 2 m, 5 m, 10 m and 100 m deep were used to assess
DEM to reflect known hydrology; ‘Stream burning’, ‘Agree’ the trenching depth required to correct hydrology. Stream
and ‘ANUDEM’. While a large number of algorithms have burning has been used successfully in many other projects
been developed for this purpose, these algorithms were se- (Saunders, 1999), but with some limitations, such as dis-
lected as they have been widely used (e.g. Saunders, 1999; torted watershed boundaries (Saunders, 1999) and the cre-
Renssen and Knoop, 2000; Turcotte et al., 2001; Doll and ation of parallel streams (Hellweger, 1996). Stream
Lehner, 2002; Soille et al., 2003) and represent three scales burning also creates a discrepancy between the original
of DEM modification. That is, Stream burning only modifies DEM and the trenched ‘‘stream’’ cells, leading to a dramatic
the elevation value of stream cells within a DEM, while the jump in elevation, which is likely to affect derived proper-
Agree algorithm modifies the value of streams cells and the ties such as slope particularly when a deep trench is re-
surrounding area within a user defined distance. The ANU- quired to correct hydrology. The advantages of this
DEM algorithm can potentially modify a whole landscape. method are its simplicity, computational efficiency and
These algorithms therefore provide three contrasting ap- the restriction of changes to fewer cells in the landscape.
How does modifying a DEM to reflect known hydrology affect subsequent terrain analysis? 33

Figure 2 Expected flow directions and catchment boundary from field work and people with expert knowledge.

Agree (Hellweger, 1996) uses a raster representation of stream network data for processing in ANUDEM v5.1 using
the known stream network to lower the landscape across default values for roughness penalties and standard errors
a user specified horizontal buffer distance and depth as well (Hutchinson, 2004). ANUDEM v4.6.3 (Hutchinson, 2003)
as burning a stream at a selected depth. This method ad- was also used through the TOPOGRID command in ArcGIS
dresses the parallel drainage issues through creation of an 9, with the data prepared as above and processes using sug-
even sloped surface across the buffer distance, but in doing gested default values. Both results were output as 10 m
so smooths the landscape, creating an even gradient and grids.
uniform aspect perpendicular to the stream channel. A The modified DEMs created using the four different
greater number of cells are changed and a drop between flat methods described above were then pit-filled (Jenson and
areas and stream cells still occurs where a channel is Dominque, 1988), before a stream network was extracted.
trenched. After unsuccessful trial settings of 250 m, 500 m This was created using a critical contributing area value of
and 750 m with drops of 1 m, 2 m and 5 m across the plain 1.5 km2, as determined from application of the area/slope
with 1 m and 2 m drop at the channel, a successful setting method (Tarboton et al., 1991) and checked against stream
of 1000 m buffer distance, vertical drop of 5 m across the head position on aerial photography. The catchment bound-
buffer plain and a channelised depth of 2 m was selected, ary was delineated from an outlet point at the junction of
which replicated the true site hydrology well. this sub-catchment with the main channel of the Kent River.
ANUDEM (Hutchinson, 1988; Hutchinson, 1989; Hutchin- Catchment area and stream networks derived for each DEM
son, 2003; Hutchinson, 2004) creates a smooth surface with- were compared visually against the expected result and val-
out sinks by imposing a global drainage condition via a ues for catchment area, stream length, mean and maximum
iterative drainage enforcement algorithm, which is based slope used to quantify the success of different methods.
on input data that can include irregularly spaced elevation The successful and least severe (i.e. the shallowest
data points (spot heights), contour lines, streamlines, sink trench) Stream burning, Agree and ANUDEM v4 and ANUDEM
points (lakes) and cliff lines. Elevation across the entire v5 DEMs were then further analysed to determine the spatial
DEM can be altered when creating a new surface that en- pattern and magnitude of changes caused by hydrological
forces drainage and eliminates abrupt jumps between the correction. An assumption was made that while the raw
stream and non stream cells. Although ANUDEM allows the DEM contained error (causing the incorrect routing of flow),
user a larger range of input variables to correct hydrology, it represented the best representation of real catchment
for the sake of simplicity and consistency across the three topography. A residual grid was created by subtracting the
different methods, only the stream network data was used various hydrologically corrected grids from the original
to enforce drainage. ArcGIS 9 was used to create a 2 m con- DEM (pit-filled version of the raw DEM). Three-dimensional
tour dataset (with the contour interval representing half the plots with histograms and values for mean, standard devia-
vertical accuracy of the original DEM) and the UNGENERATE tion, total root mean squared (RMS) error, minimum and
command used in ARC/INFO to prepare the contour and maximum change were used as a means of assessing the
34 J.N. Callow et al.

magnitude and spatial patterns of changes caused by hydro- raphy was flattened across the buffer zone. The resulting
logical correction. Slope (Horn, 1981) (in percent) was cal- surface is on average 0.70 m lower, with more cells altered
culated from the raw and corrected DEMs, with residuals than any other method and the highest RMS error (2.277)
calculated as above to determine the effects that hydrolog- and range. ANUDEM v4 and v5 produced a slightly lower sur-
ical correction on a subsequent terrain analysis. face ( 0.43 m and 0.42 m, respectively) (Fig. 4c and d).
ANUDEM v5 performed slightly better than ANUDEM v4, with
a lower residual, standard deviation, range and RMS error,
Results though results were very similar. For the ANUDEM results
(Fig. 4c and d), slight undulation in the error surface away
Stream burning required a 10 m deep trench in this flat from the stream network results from conversion of grid
landscape to correct hydrology. Fig. 3b and c identifies that elevation data to contour data for processing by the ANU-
trenching up to 5 m has little effect on stream position and DEM algorithm and back to a DEM. Stream burning produced
catchment area. Trenching depths of 1 m, 2 m and 5 m mar- the least change in elevation, with ANUDEM v5 and v4 per-
ginally increased stream length, while catchment area re- forming equally second best. Agree caused the most signif-
mained the same as the original DEM. Stream burning with icant change in elevation compared to the original surface.
10 m and 100 m trench replicated stream network length The slope residuals plots in Fig. 5 identify the spatial pat-
( 1%) and position (Fig. 3d) the best out of all methods, terns of changes caused by the different hydrological cor-
with catchment area close to the expected value ( 7%). rection methods on a secondary terrain variable. The ways
While trenching 10 m or 100 m produced identical catch- that different hydrological correction methods have altered
ment areas and similar stream lengths, the deeper trench elevation to improve hydrology (Fig. 4) are reflected in the
caused mean (2.4–5.2%) and maximum slope (59.7– spatial patterns and magnitudes of change to a secondary
562.8%) to increase significantly compared to the original terrain variable (slope). Stream burning increased mean
DEM. Also of note are the truncated catchment boundaries slope by 0.05% and altered the least number of cells, but
in the lower catchment (Fig. 3d and e), a characteristic it also caused the largest increase in individual cell slope
downfall of Stream burning as identified by Saunders (1999). of all methods.
Agree, ANUDEM v4 and ANUDEM v5 all improved the While Agree had the greatest effect on elevation, flat-
delineation of catchment boundary and stream locations tening the lower portion of the catchment (see Fig. 4b). This
over the original DEM. Fig. 3f shows that the catchment smoothing of the landscape across the buffer distance re-
delineated by Agree is close to the expected result (+3%) duced overall slope, and counteracted the effects of the
and has corrected the issue with the truncated catchment 2 m trench which increased slope adjacent to the stream
boundary in the lower catchment. Stream length was cells. The resulting error surface has a lower mean slope
slightly shorter than expected ( 3%), and stream position ( 0.02%) than the original (pit-filled) DEM, and low RMS
replication was poorer than other successful methods, as (0.623) and range. If Agree was used in a flatter landscape,
an area of overlapping buffer distance caused drainage to without the dissected terrain in the lower part of the sub-
meander tortuously. Smoothing topography across the buf- catchment, smoothing of the landscape across the buffer
fer distance mitigated the effects of Stream burning, result- distance would be less significant and mean slope would
ing in the smallest increase in mean slope (+3%), thought increase.
maximum slope increased significantly (+172%). ANUDEM v4 and v5 performed relatively well, with mini-
ANUDEM v4 and v5 performed similarly, producing catch- mal increases in slope (0.15% and 0.21% steeper, respec-
ment areas close to the expected result ( 3% and 7%, tively). RMS was lowest for ANUDEM v5 (1.347), with
respectively), with ANUDEM v4 and Agree the best at repli- ANUDEM v4 very close (1.349). ANUDEM v5 has a tighter
cating catchment area (ANUDEM v4 3%, Agree +3%). Repli- range than ANUDEM v4, with both close to results for Agree
cation of stream length was acceptable for both ANUDEM (ANUDEM v5 has a tighter range than Agree). The results for
methods (+5%, +6%, respectively), as was stream position the ANUDEM residuals are also affected by the minor
though some differences in the upper catchment are evi- changes away from the stream network used to warp eleva-
dent (compare Fig. 3g and f to 3a). Results for increased tion. Based on the raw statistics for the error surfaces,
mean and maximum slope did, however, show some differ- Agree appears to create the least change, however, visual
ence between the two methods. ANUDEM v4 performed bet- inspection of error surfaces reveals that both ANUDEM
ter than ANUDEM v5 for mean slope (+6% to +9%), but methods (Fig. 4c and d) create smooth error surfaces in
ANUDEM v5 performed best at minimising increases in max- comparison to both Agree and particularly Stream burning.
imum slope (+37% to +57%) slope. Both ANUDEM v4 and v5
performed well in replicating stream length, position and
catchment area and were consistent with the true hydrology Discussion and conclusions
of the catchment.
Residual grids of elevation (Fig. 4) show the spatial pat- DEMs are an important tool for spatially modelling hydrol-
tern and magnitude of changes caused by the different ogy, but surface reconditioning of DEMs is often required
hydrological correction methods, compared to the original to improve their representation of true hydrology. However,
(pit-filled) DEM. Stream burning with a 10 m trench changed correction of the surface hydrology has been shown in this
the least number of cells, with the lowest mean difference study to further impact terrain analyses, and selection of
compared to the original DEM ( 0.01 m lower), lowest the method chosen will lead to very different results.
range and RMS (0.265) (Fig. 4a). Agree altered cells across Stream burning was the simplest method, altering the least
a larger area, and caused significant lowering where topog- number of cells and performing well for replicating stream
How does modifying a DEM to reflect known hydrology affect subsequent terrain analysis? 35

Figure 3 The catchment area and stream position expected from expert knowledge (a), against the results from the original DEM
data (b), unsuccessful Stream burning to 5 m (c), successful Stream burning at 10 m and 100 m (d & e, note: similar results with
different depth of trench), Agree (f), ANUDEM v4.6.3 (g) and ANUDEM v5.1 (h).
36 J.N. Callow et al.

Figure 4 Residual surface showing difference in elevation between the original (pit-filled) DEM and hydrological corrected (pit-
filled) DEM created using Stream burning (a), Agree (b), ANUDEM v4.6.3 (c) and ANUDEM v5.1 (d). Note that vertical axis has been
reversed for the purposes of display.

length and position, but performed poorly for calculating Agree fundamentally changed the way that the landscape
catchment area and significantly affected subsequent ter- was represented in the DEM. This was evident from the
rain analysis of slope proximal to the channel network. change in elevation (See Fig. 4b). ANUDEM v4 and v5 pro-
Agree performed well for determining catchment area and duced very similar results to each other, performed consis-
reasonably for stream length. While mean slope was close tently across most of the criteria, and maintained landscape
to the expected result, hydrologically correcting using integrity. ANUDEM v4 was equal best for determining catch-
How does modifying a DEM to reflect known hydrology affect subsequent terrain analysis? 37

Figure 5 Residual surface showing difference in slope between the original (pit-filled) DEM and hydrological corrected (pit-filled)
DEM created using Stream burning (a), Agree (b), ANUDEM v4.6.3 (c) and ANUDEM v5.1 (d).

ment area (with Agree), while ANUDEM v5 caused the least hydrology with the accuracy required for modelling hydro-
increase in maximum slope, though the other methods were logical processes across the study catchment. While the
superior for determining stream length and mean catch- performance of the raw DEM for delineating stream posi-
ment slope (refer to Table 1). tion and catchment area in this portion of the study area
The dataset used in this study is at the better end of is poor, in other equally flat parts of the catchment its per-
available catchment scale DEMs, when compared with formance was acceptable. Elsewhere, the raw DEM repli-
those used in other studies, in both its grid resolution cated ‘‘natural’’ hydrology accurately, but the actions of
and vertical accuracy. However, it is unable to replicate land managers in diverting flow means that hydrological
38 J.N. Callow et al.

Table 1 Results achieved by the original DEM and different hydrological correction methods in replicating known hydrological
conditions and catchment parameters
Hydrologically correct Catchment area (km2) Stream length (km) Mean slope (%) Max slope (%)
Expected result 69.5 34.7 2.4 16.9
Original DEM No 11.9 ( 83%) 5.2 ( 85%) 3.5 (+46%) 12.5 ( 26%)
Trench 1 m No 11.9 ( 83%) 5.2 ( 85%) 3.6 (+50%) 12.6 ( 25%)
Trench 2 m No 11.9 ( 83%) 5.7 ( 84%) 3.6 (+50%) 15.2 ( 10%)
Trench 5 m No 11.9 ( 83%) 5.7 ( 84%) 3.8 (+58%) 31.8 (+88%)
Trench 10 m Yes 64.5 ( 7%) 34.2 ( 1%) 2.7 (+14%) 59.7 (+253%)
Trench 100 m Yes 64.5 ( 7%) 34.3 ( 1%) 5.2 (+117%) 562.8 (+3230%)
Agree Yes 71.6 (+3%) 33.7 ( 3%) 2.5 (+3%) 46 (+172%)
ANUDEM v4 Yes 67.6 ( 3%) 36.3 (+5%) 2.6 (+6%) 26.6 (+57%)
ANUDEM v5 Yes 64.5 ( 7%) 36.9 (+6%) 2.6 (+9%) 23.15 (+37%)
The most successful method for each parameter is identified in bold.

correction was required to replicate the new hydrological Australia Postgraduate Award for the first author. Sugges-
reality of the catchment. tions of two anonymous referees improved the final
The difficulty of delineating drainage direction in this manuscript.
landscape is further highlighted because delineation of the
correct flow paths required the consultation with people
who have decades of expert knowledge of the area and
References
the fact that there was even some discrepancy between
Ackerman, C.T., 2002. HEC-GeoRAS 3.1. US Army Corps of
these experts (that was finally reconciled). Using only the Engineers, Institute for Water Resources. Hydrologic Engineering
stream position predicted by experts with knowledge of Centre, Davis, CA.
the area, four different methods were able to hydrologically Allen, A. Beeston, B., 1999. The land monitor project: a multi-
correct the DEM so that the catchment area and stream net- agency project of the Western Australian Salinity Action Plan
work matched or approximated the predicted result. If the supported by the Natural Heritage Trust, WALIS.
purpose of a study is to create a DEM that is able to replicate Beven, K., Kirkby, M.J., 1979. A physically based, variable contrib-
the stream position, length and catchment boundaries, then uting area model of basin hydrology. Hydrological Sciences
all methods were relatively successful in achieving this goal. Bulletin 24 (1), 43–69.
Caccetta, P.A., Allen, A. Watson, I., 2000. The land monitor
Many projects involving the use of DEMs require accurate
project, Land Monitor Project.
development of secondary parameters such as aspect,
Callow, J.N., Smettem, K.R.J., in press. Channel response to a new
slope, wetness index or predicted rates of water movement hydrological regime in southwestern Australia. Geomorphology.
through the landscape. Surface reconditioning compromises Croke, J.C., Hairsine, P., Fogarty, P.J., 1999. Runoff generation
these outcomes. In this study, the precision and minimisa- and re-distribution in logged eucalyptus forests, south-eastern
tion of additional error was different between the methods Australia. Journal of Hydrology 216, 56–77.
and for different parameters. The variable performance of Doan, J.H., 2000. Geospatial Hydrologic Modelling Extension HEC-
these models indicates the importance of assessing their GeoHMS. US Army Corps of Engineers, Institute for Water
suitability for deriving specific secondary attributes on a Resources, Hydrologic Engineering Centre, Davis, CA.
case-by-case basis. Doll, P., Lehner, B., 2002. Validation of a new global 30-min drainage
direction map. Journal of Hydrology 258 (1–4), 214–231.
Hydrologically correcting DEMs is a necessary step for
Duke, G.D., Kienzle, S.W., Johnson, D.L., Byrne, J.M., 2003. Improv-
many hydrological modelling projects. All of the methods
ing overland flow routing by incorporating ancillary road data into
provided a significant improvement for the delineation of digital elevation models. Journal of Spatial Hydrology 3 (2), 1–27.
the catchment boundary and stream position on the original English, P., Richardson, P., Glover, M., Cresswell, H., Gallant, J.,
dataset, by warping the dataset. However, the DEM may 2004. Interpreting Airborne Geophysics as an adjunct to Hydro-
ultimately be altered in such a way that further hydrological geological Investigations for Salinity Management: Honeysuckle
calculation cannot be performed with the certainty and Creek Catchment, Victoria. 18/04, CSIRO Land and Water.
accuracy required to model further hydrological processes Fairfield, J., Leymarie, P., 1991. Drainage networks from grid
in the catchment. digital elevation models. Water Resources Research 27 (5), 709–
717.
Fisher, P., 1998. Improved modelling of elevation error with
geostatistics. GeoInformatica 2 (3), 215–233.
Acknowledgements Gallant, J.C., Dowling, T.I., 2003. A multiresolution index of valley
bottom flatness for mapping depositional areas. Water
Resources Research 39 (12), 1–4, ESG.
The authors wish to than Mike Hutchinson (CRES – Austra-
Gallant, J.C., Hutchinson, M.F., 1997. Scale dependence in terrain
lian National University) for making ANUDEM v5.1 available analysis. Mathematics and Computers in Simulation 43 (3–6),
for use in this project. Thanks also to Murray Gangell and 313–321.
Rob Webb for sharing their expert knowledge of flow direc- Garbrecht, J., Martz, L., 1997. The assignment of drainage
tions. This study was completed as part of a Ph.D. Thesis at direction over flat surfaces in raster digital elevation models.
the University of Western Australia, with the support of an Journal of Hydrology 193, 204–213.
How does modifying a DEM to reflect known hydrology affect subsequent terrain analysis? 39

Gyasi-Agyei, Y., Willgoose, G., de Troch, F.P., 1995. Effects of Moore, I., Grayson, R., Ladson, A., 1991. Digital terrain modelling: a
vertical resolution and map scale of digital elevation models on review of hydrological, geomorphological, and biological appli-
geomorphological parameters used in hydrology. Hydrological cations. Hydrological Processes 5, 3–30.
Processes 9, 363–382. O’Callaghan, J.F., Mark, D.M., 1984. The extraction of drainage
Hellweger, R., 1996. Agree.aml. Center for Research in Water networks from digital elevation data. Computer Vision, Graphics
Resources, The University of Texas at Austin, Austin, TX. and Image Processing 28, 323–344.
Holmes, K.W., Chadwick, O.A., Kyriakidis, P.C., 2000. Error in a O’Loughlin, E.M., 1986. Prediction of surface saturation zones in
USGS 30-meter digital elevation model and its impact on terrain natural catchments by topographic analysis. Water Resources
modelling. Journal of Hydrology 233 (1–4), 154–173. Research 22 (5), 794–804.
Horn, B.K.P., 1981. Hillshading and the reflectance map. Proceed- Peuker, T., Douglas, D., 1975. Detection of surface-specific points
ings of the IEEE 69 (1), 14–47. by local parallel processing of digital terrain elevation data.
Hutchinson, M.F., 1988. Calculation of hydrologically sound digital Computer Graphics and Image Processing 4, 375–387.
elevation models. In: Third International Symposium on Spatial Quinn, P., Beven, K., Chevallier, P., Planchon, O., 1991. The
Data Handling. International Geographical Union, Columbus, OH. prediction of hillslope flow paths for distributed hydrological
Hutchinson, M.F., 1989. A new procedure for gridding elevation and modelling using digital terrain models. Hydrological Processes 5,
streamline data with automatic removal of spurious pits. Journal 59–79.
of Hydrology 106, 211–232. Renssen, H., Knoop, J.M., 2000. A global river routing network for
Hutchinson, M.F., 2003. ANUDEM Version 4.6.3. Australian National use in hydrological modeling. Journal of Hydrology 230 (3–4),
University. Centre for Environmental Studies, Canberra. 230–243.
Hutchinson, M.F., 2004. ANUDEM Version 5.1. The Australian Salama, R.B., Farrington, P., Bartle, G.A., Watson, G.D., Hawkes,
National University (ANU), Canberra, Australia. G., 1993. Salinity investigations in selected wheatbelt catch-
Hutchinson, M.F., Dowling, T.I., 1994. A continental hydrological ments of Western Australia: base data from east Perenjori
assessment of a new grid-based digital elevation model of catchment, Technical Memorandum 93/9. Division of Water
Australia. Hydrological Processes 5 (1), 45–58. Resources, CSIRO.
Jenson, S.K., Dominque, J.O., 1988. Extracting topographic struc- Saunders, W., 1999. Preparation of DEMs for use in environmental
ture from digital elevation data for geographic information modelling analysis, 1999 ESRI User Conference. ESRI Online, San
systems. Photogrammetric Engineering and Remote Sensing 54 Diego, CA.
(11), 1593–1600. Soille, P., Vogt, J.V., Colombo, R., 2003. Carving and adaptive
Jones, K.H., 1998. A comparison of algorithms used to compute hill drainage enforcement of grid digital elevation models. Water
slope as a property of the DEM. Computers and Geosciences 24 Resources Research 39 (12), 1366–1379.
(4), 315–323. Tarboton, D.G., 1997. A new method for the determination of flow
Jones, R., 2002. Algorithms for using a DEM for mapping catchment directions and upslope areas in grid digital elevation models.
areas of stream sediment samples. Computers and Geosciences Water Resources Research 33 (2), 309–319.
28, 1051–1060. Tarboton, D.G., 2002. Terrain analysis using digital elevation
Luce, C.H., 2002. Hydrological processes and pathways affected by models (TauDEM), Utah State University, Logan, UT, USA.
forest roads: what do we still need to learn? Hydrological Tarboton, D.G., Ames, D.P., 2002. Advances in the mapping of flow
Processes 16, 2901–2904. networks from digital elevation data. World Water and Environ-
Maidment, D., 1996. GIS and hydrological modelling: an assessment mental Resources Congress, Orlando, FL.
of progress. In: Third International Conference on GIS and Tarboton, D.G., Bras, R.L., Rodriguez-Iturbe, I., 1991. On the
Environmental Modelling, Santa Fe, NM, 20–25 January, 1996. extraction of channel networks from digital elevation data.
Martz, L., De Jong, E., 1998. CATCH: A FORTRAN program for Hydrological Processes 5, 81–100.
measuring catchment area from digital elevation models. Turcotte, R., Fortin, J.-P., Rousseau, A.N., Massicotte, S., Ville-
Computers and GeoSciences 14 (5), 627–640. neuve, J.-P., 2001. Determination of the drainage structure of a
McKenzie, N., Gallant, J. Gregory, L., 2003. Estimating water watershed using a digital elevation model and a digital river and
storage capacities in soil at catchment scales. 03/03, CRC lake network. Journal of Hydrology 240 (3–4), 225–242.
Catchment Hydrology. Van Niel, K.P., Laffan, S.W., Lees, B.G., 2004. Effect of error in the
McMaster, K.J., 2002. Effects of digital elevation model resolution DEM on environmental variables or predictive vegetation mod-
on derived stream network positions. Water Resources Research elling. Journal of Vegetation Science 15 (6), 747–756.
38 (4), 13–19. Walker, J.P., Willgoose, G.R., 1999. On the effect of digital
Mizgallewicz, P.J. Maidment, D., 1996. Modelling agricultural elevation model accuracy on hydrology and geomorphology.
transport in Midwest Rivers using Geographic Information Water Resources Research 35, 2259–2268.
Systems. Online Report 96-6, Centre for Research Water Wolock, D.M., Price, C.V., 1994. Effects of digital elevation model
Resources, University of Texas, Austin, TX. map scale and data resolution on a topography-based watershed
Montgomery, D., Dietrich, W., 1988. Where do channels begin? model. Water Resources Research 30, 3041–3052.
Nature 336, 232–234. Zhang, W., Montgomery, D., 1994. Digital elevation model grid size,
Montgomery, D., Dietrich, W., 1992. Channel initiation and the landscape representation and hydrologic simulations. Water
problem of landscape scale. Science 225, 826–830. Resources Research 30, 1019–1028.

You might also like