Vous êtes sur la page 1sur 10

WATERRESOURCES

RESEARCH,VOL. 30,NO. 4, PAGES1019-1028,


APRIL 1994

Digitalelevationmodelgrid size,landscape
representation,
andhydrologicsimulations
WeihuaZhang and David R. Montgomery
Department
of Geological
Sciences,
University
of Washington,
Seattle

Abstract. High-resolution digital elevation data from two small catchmentsin the

western
United Statesare usedto examinethe effectof digitalelevationmodel(DEM)
gridsizeon the portrayalof the landsurfaceandhydrologic
simulations.
Elevation
dataweregriddedat 2-, 4-, 10-, 30-, and 90-m scalesto generatea seriesof simulated
landscapes.
Frequencydistributions
of slope(tanB), drainageareaper unit contour
length(a), and the topographicindex(a/tan B) were calculatedfor eachgrid size
model.Frequencydistributions
of a/tanB werethenusedin O'Loughlin's(1986)
criterionfor predictingzonesof surfacesaturationand in TOPMODEL (Bevenand
Kirkby,1979)for simulatinghydrographs.For bothcatchments,DEM grid size
significantly
affectscomputedtopographic
parametersandhydrographs.
While channel
routingdominateshydrographcharacteristics
for largecatchments,grid size effects
influence
physicallybasedmodelsof runoffgenerationand surfaceprocesses.
A 10-m
gridsizeprovidesa substantialimprovementover 30- and 90-m data, but 2- or 4-m data
provideonly marginaladditionalimprovementfor the moderatelyto steepgradient
topography
of our studyareas. Our analysessuggestthat for manylandscapes,a 10-m
gridsize presentsa rational compromisebetween increasingresolutionand data
volumefor simulatinggeomorphicand hydrologicalprocesses.
Introduction

Predictingspatial patterns and rates of runoff generation


andmany geomorphic processesrequires both a hydrologic
modeland characterizationof the land surface. Most physicallybasedmodelsof hydrologicand geomorphicprocesses

relyon either spatially distributedor lumped characterizationsof local slope and the drainagearea per unit contour
length[e.g., Beven and Kirkby, 1979; O'Loughlin, 1986;
Vertessyet al., 1990; Dietrich et al., 1993], and digital

hydrologicresponseusingtopographically driven models. In


this paper, we assess how grid size affects topographic
representation,derived topographic attributes, and hydrological simulations for two small catchments using highresolution digital elevation data. In contrast to previous
studies,we grid the same elevation data at several different
scalesto isolate the effect of grid size on landscape representation. Issues associated with vertical sampling resolution are not addressed

in this work.

elevation
models
(DEMs)
commonly
areused
forsuch.
Study
Areas
characterization
in a wide variety of scientific,engineering,

The study catchmentsare located at Mettman Ridge near


andplanningapplications.Althoughthe increasingavailabilCoos
Bay, Oregon, and Tennessee Valley in Marin County,
ity of DEMs allowsrapid analysisof topographicattributes
California
(Figure 1). Previous field investigations deterovereven large drainagebasins, the degreeto which DEM

gridsizeaffectsthe representation
of the land surfaceand mined the nature and distribution of geomorphicand hydrohydrological
modelinghas not been examinedsystemati- logic processesin each catchment. High-resolution digital

elevation data were generated for testing DEM-based process


models in these catchments. Field mapping in each
Digitalelevation data are stored in one of the following
formats:
as pointelevationdata on eithera regulargridor catchment reveals that the high-resolution data provide a
triangular
integratednetwork, or as vectorizedcontours reasonablyaccurate portrayal of the land surface.
cally.

storedin a digital line graph. Each of theseformatsoffers Mettman Ridge

advantages
for certainapplications,
but the gridformatis
The MettmanRidgecatchment
occupies
0.3 km2 of an
usedmostwidely.Severalrecentstudiesexploredtheeffect
area
in
which
previous
field
mapping
documented
the extent
ofDEMgridsizeon landscape
representation
[Hutchinson
of
the
channel
network
[Montgomery,
1991].
Channel
head
andDowling, 1991;Jenson, 1991;Panuskaet al., 1991;
Quinnet al., 1991].Thesestudiesshowedthat distributions locations in this area are controlled primarily by shallow
of topographic
attributesderivedfrom a DEM dependto debris flows from small unchanneled valleys [Montgomery
somedegreeon grid size. None of thesestudies,however, and Dietrich, 1988]. The catchment is highly dissectedwith
systematically
analyzedthe effectof grid size on eitherthe hillslopelengthson the order of 30-50 m [Montgomery and
statistical
characterization
of the land surface,or simulated Foufoula-Georgiou, 1993]. Slopes of 30o-40 are common,
and there are a substantial number of slopes that locally

Copyright
1994
bytheAmerican
Geophysical
Union.

exceed 45 .

Paper
number
93WR03553.
0043.1397/94/93WR-03553505.00

A 1:4800 scale topographic basemap derived from lowaltitudeaerial photographstaken prior to timber clearingwas
1019

1020

ZHANG

AND MONTGOMERY:

DIGITAL

ELEVATION

MODEL

GRID SIZE

slope
lengths
ontheorderof 30-50m [Montgomery
and
Foufoula-Georgiou,
1993].
Thetopography
oftheTennessee
Valleycatchment
is lesssteepthanthatof theMettman

ME'!-I'MA
RIDGE

Ridge
catchment;
slopes
of200-30
arecommon
andslopes
in excess of 40 are rare.
Digitalelevation
datawereobtained
fromlow-altitude
aerialphotographs
using
a stereo
digitizer
ata density
about
every10m [Dietrich
et al., 1993].Thespotelevations
were
gridded
to generate
a 5-m contourintervalmapof the
catchment
(Figure
2b).Fieldinspection
reveals
thatthedata

TENNESSEE
VALLEY

providean excellentportrayal of the land surface.


Methods

Spotelevation
datafor thetwocatchments
weregridded

Figure 1. Location map of the study catchments.

at scalesof 2, 4, 10,30, and90 m usingthegridmodule


of
Arc/Info,with griddedelevations
recordedto the nearest
centimeter.Cumulativefrequencydistributions
of threeto-

pographic
attributes,
localslope(tanB), drainage
areaper
unitcontour
length(a), andtopographic
index(a/tan

werecalculated
for eachDEM of the two studycatchments
usedas the sourceof digitalelevationdata. The basemap usingthe modelof Jensonand Domingue[1988].Their
was scannedand vectorized usingan automatedroutine to modeldefinesthe downslopeflow directionfor eachcell
reproduce
contours
identicalto thoseon the originaltopo- corresponding
to the orientation
of the neighboring
cellof
graphicmap. Althoughseveraldiscrepancies
were noted lowest elevation. The tan B for the cell is then calculated
betweenthis map and the groundsurface,herewe assume basedon the elevationdifferencebetweencells.Definition
of
that this data providesan accurateportrayalof the land the spatial distributionof flow directions allows determina.
surface(Figure 2a).

tion of the total numberof the cells that direct flow to each

cell, andthusdrainageareas.Here, a is the drainage


area

Tennessee
Valley

dividedby the grid cell dimensioncalculatedfor the centerof

The TennesseeValley catchmentoccupies1.2 km2 in


whichpreviousfield mappingalsodocumentedthe extentof

the cell. Althoughthe assignment


of all flow to a single
downslope
grid cell distortsflow pathsin both divergent

the channelnetwork [Montgomeryand Dietrich, 1989]. topographyand for slopesorientedat anglesotherthanthe


Shallow
landsliding
dominates
sediment
transport
in steep eightcardinaldirections[Quinnet al., 1991],the algorithm
hollowsand sideslopes,diffusivetransportdominates
on has been widely used in many topographicmodels[e.g.,
divergent noses, and saturation overland flow and channel Marks et al., 1984; Band, 1986; Jenson, 1991]. Several
processesdominatesedimenttransportin lower-gradient workers[e.g.,Quinnet al., 1991;CabralandBurges,1992;
valleys.The catchmentis rhythmicallydissected,
with hill- Lea, 1992]recentlyproposedalgorithmsincorporating
mul250

300

200

150

a.

180

Contour
=6

b.

Figure2. Contour
mapofthe(a)Mettman
Ridgeand(b)Tennessee
Valleycatchments.

ZHANG AND MONTGOMERY:DIGITAL ELEVATION MODEL GRID SIZE

tipledownslope-flow
directions
thataremoresuitable
for
representing
flowondivergent
hillslopes.
Whileweareeager

1021

1.0

to explorethese newer algorithms,this study doesnot


examine
their influenceon topographicrepresentation.
Hydrologic
simulations
employedthe steadystatemodel

", l.
,

TOPOG[O'Loughlin, 1986]to examinepatternsof surface


saturation
and TOPMODEL [Beven and Kirkby, 1979]to

---- 30m

t...

' .... 90m

predict
runoffproduction
to short-duration
storms.Soil
hydraulic
parameters
were estimatedfrom field measurements[Montgomery,1991].Model simulationsexploredthe
effectof DEM grid size on simulatedhydrologicresponse.

Landscape
Representation
Cumulative frequency distributions of tan B, a, and

a/tanB determinedfor eachgrid size modelreflectchanges


in both mean and local values. Comparisonof the distributionsof these topographic attributes allows direct assessmentof the influenceof grid size on landscaperepresenta-

0.0

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

tanB

1.0

tion.

Slope
Cumulativeslope distributionsare more sensitiveto DEM

gridsizefor the steeperMettman Ridgecatchmentthanfor

-.,x_

.... 30m
.....

90m

themoderategradient TennesseeValley catchment(Figure


3). For both study areas, the percent of the catchment
steeperthan a given slope systematicallydecreasesas the
0.4
DEM grid size increases, and the largest effect is for the
steepestportions of the catchments. In the case of the
Mettman Ridge catchmerit, the mean slope declines from
o.2
0.65for the 2-m grid size model to 0.41 for the 90-m grid size
model(Figure 3a). This result is consistent with, but more
pronouncedthan those of previous studies [Jenson, 1991;
0.0
0.0
0.2
0.4
0.6
0.8
1.0
Panuskaet al., 1991]. Grid size influence on slope distributanB
tionsis lesspronouncedfor the TennesseeVaLleycatchment
(Figure3b), where the mean slope is 0.34 for the 2-m grid
Figure 3. Cumulativefrequency distributionsof slopedemodeland 0.29 for the 90-m grid size model. The distribu- rived for different DEM grid sizes. (a) Mettman Ridge. (b)
tionsfor both catchmentssuggestthat grid sizessmallerthan TennesseeValley.
l0 m yield only marginalimprovementin sloperepresenta-

tion. Sincethe slope of a grid cell representsan average


slopefor the area coveredby the cell, increasingDEM grid
sizeshouldresult in decreasingability to resolvethe slope
characteristics
of steeperand more dissectedtopography.
The cumulativeslopedistributionfor the Mettman Ridge
catchment,especiallyfor the 10-m and 30-m grid size models,arestepped,while distributionsfor bothlargeandsmall
gridsizesare smoother.We suspectthat this reflectsthe
smallnumberof grid cells in large grid size modelsof this
catchment.
Fewer grid cellsfor the smallerMettmanRidge

algorithm used to compute a determines this minimum


value. While it is intuitive that larger grid size limits the
resolutionof fine-scaletopographic features, the effect on
both the mean and local a is significantfor topographically
driven hydrologic and surface process models.
TopographicIndex

The topographicindex (a/tan B) is an important component of many physically based geomorphicand hydrologic
models,as it reflectsthe spatial distribution of soil moisture,
distribution
thanfor the TennesseeValley catchment.
surface saturation, and runoff generation processes [e.g.,
Drainage
Areaper Unit ContourLength
Beven and Kirkby, 1979; O'Loughlin, 1986; Moore et al.,
Cumulativedistributionsof a alsoare sensitiveto grid size 1986].Derivation of frequencydistributionsof a/tan B is the
(Figure
4). Largergridsizesbiasin favorof largercontrib- first step for hydrologicalsimulationsin most topographiuting
areas,withcomparable
effects
inbothcatchments.
For cally driven hydrologic models.
theMettman
Ridgecatchment,
themeanvalueofa increases Grid size significantlyaffects the cumulative frequency
from20m for a 2-mgridsizeto 102m for a 90-mgridsize. distributionsof a/tan B (Figure 5). Decreasinggrid size shifts
catchment should result in a more discontinuous cumulative

In theTennessee
Valleycatchment,
the meanvalueof a the cumulative distribution toward lower values of a/tan B,
increases
from19m for a 2-mgridsizeto 120mfora 90-m with the greatesteffecton smallervalues. Again, computed
grid size.

frequencydistributionssystematicallyconvergetoward that

For eachgrid size, a singlepixel definesthe smallest of the finestgrid size. For the Mettman Ridge catchment,the
possible
valueof a andthuswherethe cumulativefrequency mean In (a/tan B) increasesfrom 3.4 for a 2-m grid size to 5.6
distributions
of a reach 100% of the catchmentarea. The for a 90-m grid size. In the TennesseeValley catchment,the

1022

ZHANG AND MONTGOMERY:

DIGITAL

ELEVATION

MODEL GRID SIZE

usingTOPMODELfor a rangeof rainfallintensities


and

1.0

baseflowsfor the study catchments.


2111

Surface Saturation

0.8

-----

lorn

....

30rn

Many hydrological,
geomorphological,
and ecological
phenomena
arecloselyrelatedto thebehaviorofthevariable

0.6

saturation
areawithina catchmerit.
By assuming
a steady

statedrainagecondition,O'Loughlin[1986]expressed
the
conditionfor surfacesaturationat any locationin a catch-

0.4

merit

as

a/tanB > At/Qo

0.2

N'N?
x

0.0

21o '

0.0

410

6.0

(l)

whereT isthemeansoiltransmissivity
ofthecatchment,
At
8.0

10.0

12.0

In(a)

is thetotal catchmeritarea,andQ0 is the runoffratefromthe


catchmerit.
The termonthe right-handsideof theequation
is

definedas the averagewetnessstateof a catchment


(W).
The total saturatedarea for a given catchmentwetnessis
simply the sum of all the local areas which have valuesof
a/tan B >

1.0

W.

The effect of DEM grid size on the computedsaturation


area can be directly examinedusing (1) and the cumulative
0.8

I'
I
[

'',
',
\,
\
\', : ',

0.6 L

0.4

;
\
"

'
lorn
.... 30

\, , ',,
,x ",
, \

distributionof a/tan B. For a given wetnesscondition,


predicted saturation areas for both catchments increasewith

.....

',,",

',

1.0

\'\ , ",,

, '

0.2

'

.... 30m

.kk '

'%'...

..... m

',,

0.0
0.0

2.0

4.0

6.0

8.0

10.0

12.0

ln(a)

Figure 4. Cumulativefrequency distributionsof the drainagearea per unit contourlengthfor differentDEM grid sizes.
(a) Mettman Ridge. (b) TennesseeValley.

0.4

0.2

x "-...
_

--...

0.0

0.0

meanIn (a/tan B) increasesfrom 4.0 for a 2-m grid sizeto 6.2


for a 90-m grid size. The influenceof grid sizeon both mean

and local valuesof a/tan B demonstrates


the potentialfor
affectingtopographicallybasedhydrologicmodelsbasedon
this parameter.
The effectof grid size on spatialpatternsof a/tanB is even

2.0

4.0

6.0

8.0

10.0

12.0

1.0

0.8

more striking(Figure6). Detailedfeaturesthat appearon


finer-gridDEMs are obscuredon coarser-gridDEMs, with a
progressive
lossof resolutionfor boththe drainagenetwork
definedby the highervaluesof a/tan and hillslopes
associated
with the lower valuesof a/tan . Degradationof
these geomorphicfeatures affects the simulationof runoff

productionand geomorphicprocesses
in topographically
driven models.
0.2

Hydrologic Simulations
We used the models TOPOG [O'Loughlin,!986] and

0.0

TOPMoDEL [Bevenand Kirkby,1979]to investigate


the
effect of grid size on hydrologicsimulations.We examined Figure 5. Cumulativefrequencydistributionsof the topoboth representation of saturated areas within a catchment graphic
index,In (a/tanB), for differentDEM gridsizes.
(a)
usingTOPOGandthe influence
on hydrographs
calculated Mettman Ridge. (b) TennesseeValley.

ZHANG

AND MONTGOMERY:

DIGITAL

ELEVATION

MODEL

GRID SIZE

1023

ii

IIi

...

..

..

..

Figure
6a. Maps
ofMettman
Ridge
showing
thespatial
distribution
ofthetopographic
index
In(a/tan
B)

forfourdifferent
DEMgridsizes:
4, 10,30.and90m.Darkershades
represent
largerIn (a/tanB).
increasing
grid size. In the MettmanRidgecatchment,
for
example,W = 180 m predictsa saturatedarea equalto

rived a functionrelatinglocal soil moisturestorageto the


topographicindex of a catchment:

about13%of the total catchmeritarea for a 2-m grid spacing,

S = + m{X - In (a/tan B) - m(8 - In (T)}


(2)
32%for a 30-m grid spacing,and 50% for a 90-m grid
spacing.
The TennesseeValley catchmentexhibitsa similar, where S is the local soil moisturedeficit, is the mean soil
although
lesspronounced,
relationbetweengrid sizeand moisture deficit of the basin, m is a parameter that characsaturated
area for a given wetnesscondition.

terizesthe decreasein soil conductivity with soil depth, and


A and 8 are the mean values of In (a/tan B) and In (T) for the
Catchment
Response
Duringa StormEvent
catchment. For locations where S > 0, the soil moisture
WeusedTOPMoDEL to explorethe effectof DEM grid
store is not filled and there is no surface saturation. For
sizeonthesimulatedhydrologicresponse
of eachcatchment
locationswith S -< 0, the soil moisture store is full, and

to a simpleshort-duration
rainfallevent.The modelpredicts
thedistributionof soil moisture and the runoff on the basisof

urfacetopographyand soil properties.A criticalassump-

surface saturation occurs.

The modelcomputesboth the relativeamountof subsur-

tion of the model is that locations with similar topography face and saturation overland runoff, as well as the spatial

andsoilproperties
respondidenticallyto the samerainfall. distributionof theserunoff processes.During a modelsimByassuming
a spatiallyuniformrecharge
rateanda quasi- ulation,the meansoilmoisturedeficitof a catchmentat time
steadysubsurface
response,
Bevenand Kirkby[1979]de- t, t, is calculatedby

1024

ZHANG AND MONTGOMERY:DIGITAL ELEVATION MODEL GRID SIZE

.'.- "':f"??'":."'"."
i '.-..-....it

-.,"' ,'.=...,..

,.,...
....

, ....

..-..... ......
':,: ..

..........': .,..

;, :..

.f,..,

.--. ......
-.

,:

....-..,..".'
"."

'.).,
........
?, .
..

--.
..,. ,
. -..., ' .

' , ,
'

..

..

..

.. .-,... ......
.. ,.......
.. '-. '
- .. , .

Figure 6b.

', = .t-] + (q,-]-

r)At

'., .

..

Same as Figure 6a except for Tennessee Valley.

(3)

where q is the total catchment runoff at time t - 1 divided


by the catchment area, r is the net recharge rate into the soil
column, and At is the computation time step. The updated S
at all points in the catchment are then computed using (2).
Any area with a soil moisture deficit larger than the incremental precipitation in a unit time step will only produce

The saturation
excessrunoffqo is the sumof excess
soil
moisture
anddirectprecipitation
thatfallson the saturated
areas. This is expressed as

qo
=Z

+r dA

whereA.,.istheareaofthecatchment
withsurface
saturation
(i.e.,S -< 0). Totalrunoffq at anytimestepisthesumof
andsurfacerunoff.While the equationfor submoisture deficit smaller than the incremental precipitationin subsurface
flowis onlyrelatedto the meanvalueof a/tanB, A,
a unit time step, or that were saturatedduring the previous surface
for saturation
excessflowis alsorelatedtothe
time step will produce both subsurfaceand saturation excess theequation
subsurface runoff,

while

those areas with either a soil

runoff. The subsurfaceflow rate q/, of the catchment is

distributionform. Thus the influenceof DEM grid sizeon

calculated by

predicted responsediffers for (4) and (5).

We first examinethe simulatedsubsurface


hydrologic

qt,= e-('-)e-g/m

(4) response
using(4) withoutconsidering
eithersurfaceflo, or

ZHANG AND MONTGOMERY:DIGITAL ELEVATION MODEL GRID SIZE


120,0 .......

0.80

?'-

.......

lOO.O

" '"
0.60

1025

--,2rn

".,.,-.

-----4m
----- lorn

"-..

.... 3o,,-,

""'

----

80.0-

'90m

0.40

lorn

....

30m

.....

OOm

60.0
..

.....................

........

--....

'--

40.0

0.20
20.0

0.00

0.0

10.0

......

20.0

30.0

40.0

'

0.0

50,0

0.0

I 0.0

time (hr)

20.0

30.0

40.0

50.0

time (hr)

0.80

100.0
'

.,

-- 2m

----4m

80.0

0.40

0.20
20.0

0.00

0.0

10.0

20.0

30.0

40.0

50.0

0.0

time ()

0.0

10.0

20.0

30.0

40.0

50.0

time ()

Figure 7. Runoffhydrographsfor differentDEM grid sizesunderdifferentrainfall and initial baseflow


conditions.(a) Mettman Ridge catchmentunder 5 mm/h rainfall and 3.5 mm/day initial base flow. (b)
Mettman Ridgecatchmentunder 100mm/hrainfalland 3.5 mm/dayinitial baseflow. (c) TennesseeValley
catchmentunder 5 mm/h rainfall and 3.5 mm/day initial base flow. (d) TennesseeValley catchment under
100 mm/h rainfall and 3.5 mm/day initial base flow.

interactions
betweensurfaceand subsurfaceflow paths.This lated hydrologicresponseconsideringboth subsurfaceand
exampleapproximatesconditionsof small initial base flow saturation excess flow. We assumed that soil parameters
andrainfallin steep catchments.For this case, it is shown were spatiallyuniform in our simulationto isolate the effect
thatbyadjusting
theinitialvaluesof themeansoildeficitg, of topographicrepresentation.We usedvaluesof 70 mm and
themodelwill producethe samerunoffhydrographs,
inde- 360 mm/h for the parameter m and surface hydraulic conpendentof the form of a/tan B distribution.
ductivity, respectively.For simplicity, we only calculated
Givenanytwo meanvaluesof In (a/tanB), A, andX2,we runoff production and ignored flow routing in channels.

havethefollowingbaseflow equations:

Simulations were conducted with four rainfall intensities (5,

qbl= e-X'-S)e-'/m

(6)

qb2= e-( : - )e-S:/m

(7)

Thenecessary
condition
for qbl = qb2is
(S2- S) = m(,X2- x )

(8)

Onceinitial values of the mean soil moisturedeficit are set


according
to this functionalrelation, the relationwill holdfor

10, 50, and 100 mm/h) sustained for 4 hours and five initial
base flow conditions(1, 2.5, 3.5, 4.5, and 9.5 mm/day). The
lowest simulated rainfall intensity (5 mm/h) occurs frequently in these areas, and the highest simulated rainfall
intensity (100 mm/h) representsan extreme event. At most
times, the antecedentbase flow rates for the study areas are
less than 4 mm/day.
The effect of grid size on computed hydrographsdepends
on both rainfall intensity and initial base flow (Figure 7). To

alltimestepsduringa storm,leadingto an identical


hydro- quantitativelyexamine these results, we normalizedpeak
computed
hydrographsusingonly the subsurfaceflow equa-

dischargescomputedwith different grid size modelsby the


correspondingpeak dischargescomputedfrom the 90-m

tion.

DEM.

graph.In other words, DEM grid size doesnot affectthe

We alsoexaminedthe effectof DEM grid sizeon simu-

A plot of normalizedpeak dischargeversusgrid sizefor a

1026

ZHANG AND MONTGOMERY: DIGITAL ELEVATION MODEL GRID SIZE

Tennessee
Valley catchment,the differencebetweenthe
peakdischarge
from grid sizessmallerthan90 m andthat
from a 90-mgrid increaseswith the rainfallintensity.
As
rainfallintensityfurther increases,however, thesedifferences
decrease.
Thisresultisexpected
considering
thatpeak

1.0

dischargewouldbe the samefor all grid sizemodelsforthe


extremecasesof (1) no surfacesaturationundera verysmall
rainfallintensityand (2) completesurfacesaturationundera
very large rainfall intensity.

0.4

Withina rangeof reasonable


rainfallintensities(e.g.,less
than 50 mm/h), peak dischargedifferencesare less than

for gridsizessmallerthan30 m for the Tennessee


Valley

O.2

0.0

0.0

catchment,and lessthan 8% for grid sizes smallerthan 10m


for the Mettman Ridgecatchment.This suggeststhat thereis
2 .o

4o.o

60.0

a DEM grid size beyond which computedhydrologic


re-

so.o

1 oo.o

grid size (m)

sponseis less sensitiveto grid size, and that this grid sizeis
approximately10m for TennesseeValley catchmentand4 m
for the Mettman Ridge catchment.

1.o

Discussion

0.8
..,..

..

..//

..,'/

The analysespresentedabove are based on the single-

direction flow-partitioning algorithm. This algorithm does


not resolve hillslope divergence, introducing artifactsthat
....'
...'
..,"'/."./*
influence cumulative frequency distributions of a and tanB.
Quinn et al. [1991] showed that for the same grid sizethe
,"
.' /
,
initialbaseflow
..-'
.,,,,"// ./'/
single-directionalgorithm yields higher tan B values, and
0.4
....
..,.,."
/",f
.o
,
therefore
lower a/tan B, than a multiple-directionalgorithm.
,
o,-'
,,-' /
/' /
-- -- 2.5ram/day
They also illustrated that DEM grid size influencesthe
.....
,.-' ..../ /"/
----- 3.5
mm/dy
0 0.2
. ' ' ...'""" /'
.... 4.5
ram/day
distributionof a/tan B for multiple-directionalgorithms,with
the percentof area havinglarger a/tan B increasingwithgrid
size. Thus the flow-partitioningalgorithm also affectstopo40.0
60,0
6.0
%0.0 graphic representation, in addition to the grid size depen0'00.0
20.0
grid size(m)
dence documentedin this paper.

. ,,,

...' ../

. .,,

The effect of DEM

Figure 8. Computed peak dischargeversus DEM grid size


for a rainfall intensity of I0 mm/h and different initial base
flows. Peak dischargesare normalized to those of the 90-m
grid size. (a) Mettman Ridge. (b) TennesseeValley.

size on the distribution

of derived

slopeshas importantimplicationsfor geomorphicand hydrologic modeling and land management decisions basedon
such models. While a coarse-grid DEM may be most practical for modeling large-scale geomorphic processes,the
coefficientsincorporated in process models and transport

laws at suchscalesare not analogousto thosemeasured


in
10 mm/h

rainfall

at various

antecedent

base flow rates is

shownin Figure 8. In general, the computedpeak discharge


increases with increasing grid size. However, as the initial
baseflow increases,the effect of grid size on the computed
discharge decreases. There are also some deviations from
the positive correlation between the computeddischargeand
the grid size, especially for the Mettman Ridge catchment.
Further examinationreveals that the deviationsin Figure 8a
reflect variations

in the back-calculated

mean initial

soil

moisture deficit. For a given initial base flow, the mean soil
moisture deficit generally decreasesas the grid size increases,but thereis a deviationfrom this trend at a 30-m grid
size, where the initial mean soil moisturedeficit is larger than
that of 10-m grid size. With a larger moisture deficit and
thereforea smallersaturatedarea, the runoff productionrate
will be smaller. For the TennesseeValley catchment,both

the computedinitial mean soil moisture deficit and peak


dischargevary more systematicallywith grid size.
The effect of rainfall intensity on the relation between

field studies.

The effect of DEM

size on the derived a/tan B will affect

prediction
of thespatialdistribution
of runoffprocesses,
and
the associated
materialtransportprocessesover a land
surface.Runoffprocesses
are governedby neitherthefinest,
nor the coarsest scale topography within a landscape.
Rather, they are governedby processesactingover intermediate scales. If the DEM grid size is too large, then many

topographic
featuressuchas hollows,low-orderchannels,
andhillslopeswill not be resolved.We suggestthat themost
appropriate
DEM gridsizefor topographically
drivenhydrologic modelsis somewhatfiner than the hillslopescale
identifiable in the field.

Anotherimplicationis for floodforecastingin a drainage


basin.Our resultsindicatethat with the samevaluesfor soil

parameters
butdifferent
DEM gridsizes,themagnitude
of
the peak runoff rate, and therefore the runoff volume,

predicted
in response
to a givenstormmaydiffersignificantly.Theseresults,however,are only for runoffproduc-

computed dischargeand the grid size is more complex


(Figure 9). Within a rainfall intensityrange of 5- to 10 mm/h

tion;hydrographs
at thebasinmouthalsoreflect
routing
of
flowthrough
thechannel
network.
Gridsizeeffects
should

for the Mettman Ridge catchment and 5- to 50-mm/h for the

be smallerfor a largedrainage
basinwhererunoffhydro-

ZHANG AND MONTGOMERY:DIGITAL ELEVATION MODEL GRID SIZE


1.00

1.00

0.80

0.80

1027

2rn

-----4m
-----

lorn

....

30m

.,..

" 0.60

0.60

0.40

0.40

0.20

0.20

0.00

!00.0

0.0

rain (rnm)

ain (mm/hr)

Figure 9. Computedpeak dischargeversusDEM grid sizefor an initial baseflow of 2.5 mrn/dayand


differentrainfall intensities.Peak dischargesare normalizedto thoseof the 90-m grid size. (a) Mettman
Ridge. (b) TennesseeValley.

graphsare dominatedby channelrouting.Even so, the usedto derive a DEM providesa guide to the grid size that
influence
of DEM grid size on predicted runoff productionis would take full advantageof the original spotelevationdata,
animportantconsiderationfor interpretinghydrologicalsim- and thusprovidethe mostfaithful landscaperepresentation.
ulationsusinga topographicallydriven model.
We suggestthat the length scale of the primary landscape
A particularlyintriguingimplicationis for calibrationand featuresof interestprovidesa naturalguideto an appropriate
validationof dynamic physically based hydrologic models. grid size. The mostbasicattribute of many landscapesis the
All hydrologicalmodels make simplifyingassumt;tionsand division imo topographicallydivergem hillslopesand cononlyapproximatereal hydrologicsystems.In practice,cal- vergentvalleys. A grid size smaller than the hillslopelength
ibrationis required to obtain acceptable correspondence is necessaryto adequatelysimulateprocessescontrolledby
between field results and model simulations. Calibrated
land form. For other processes,the most appropriate grid
parametersfor a particular catchment are then used for size for simulation models is best scaled in reference to the

hydrologicalforecasting either for the same catchment, or


processbeingmodeled.For example, a coarse(e.g., 90 m)
for a different catchment with similar physical properties.
grid size may be most appropriate for modeling orogenic
However, our results show that DEM grid size, rainfall, and
processesover large areas and long time scales.
initialbase flow all affect simulated hydrographscomputed
Our results imply that it is unreasonableto use a 30- or
withthe sameset of parametervalues.Consequently,model
90-m grid size to model hillslope or runoff generationprocalibrations
are grid size specific.
cessesin moderatelyto steep gradient topographywithout
somecalibrationof the processmodel. While a 10-m grid is

AppropriateGrid Size

a significantimprovementover 30 m or coarsergrid sizes,

Theresultsof our studyinvite the questionof what defines finergrid sizesproviderelativelylittle additionalresolution.
an appropriategrid size for simulationsof geomorphicand Thus a 10-m -gridsize presents a reasonablecompromise
hydrologic
processesusingtopographicallydriven models. between increasing spatial resolution and data handling

for modelingsurfaceprocessesin manylandThis questionis best examinedin two parts; the relation requirements
betweenland surfaceand the spot elevationdata usedto scapes.
createa DEM and that between grid size and the original
spot elevation data.

The data used to create a DEM are a filtered representa-

Conclusions

tion of landscapesampledat someregularor irregular The grid size of a DEM significantly affects both the
intervalto builda collectionof elevationdata.The spacingof representationof the land surface and hydrologicsimula-

tions based on this representation.As grid size decreases,


featuresare moreaccuratelyresolved,but faithful
theresolution
of the DEM. Decreasing
thegrid sizebeyond landscape

theoriginaldata usedto constructa DEM effectivelylimits

of a landsurfaceby a DEM dependson both


theresolutionof the originalsurveydata doesnot increase representation
grid
size
and
the
accuracyand distributionof the original
theaccuracy
of the landsurfacerepresentation
of theDEM
andpotentiallyintroducesinterpolationerrors.The relation surveydata from which the DEM was constructed.These
oftheoriginalelevationdatato thelandsurfaceis a crucial, resultshaveimportantimplicationsfor simulationsof hydrobutoftenneglected,characteristic
of a DEM. This is a logicand geomorphicprocessesin naturallandscapes.Our
problemwith many commerciallyavailableDEMs. While ability to model surface processessoon will be limited
thereare data collectionstrategies
that couldoptimize primarilyby dataqualityandotherissuesdirectlyrelatedto
underconsideration.
We suggest
thata gridsizeof
landscape
representation
(e.g.,densetopographic
samplingprocesses

inareas
ofcomplex
topography
andsparse
sampling
inareas 10 m would sufficefor many DEM-based applicationsof
withsimpletopography),
the average
spacing
of the data geomorphicand hydrologicmodeling.

1028

ZHANG AND MONTGOMERY: DIGITAL ELEVATION MODEL GRID SIZE

Acknowledgments.
Thisresearchwassupported
by NationalAeronauticsand SpaceAdministrationgrant NAGW-2652A,National
ScienceFoundationgrantIRI91-17094,andgrantsTFW FY92-010
and TFW Fy94-004 from the Sediment, Hydrology, and Mass
Wastingcommitteeof the WashingtonState Timber-Fish-Wildlife
agreement. We thank Harvey Greenberg for technical support,

SusanJensonfor makingher topographic


analysismodelavailable

Lea,N.J., An aspect
drivenkinematic
routing
algorithm,
in
Overland
FlowandErosion
Mechanics,
edited
byA. J.Parsons

and
A.D. Abrahams,
pp.393-407,
Chapman
andHall,London,
1992.

Marks,
D. M., J.Dozier,
andJ.Frew,Automated
basin
delineation
fromdigitalelevation
data,Geo-Processing,
2, 299-311,
1984.

Montgomery,
D. R., Channel
initiation
andlandscape
evolution,

to us, and Bill Dietrich, Tom Dunne, and Romy Bauerfor discussionson subjectsrelatedto the study.

Montgomery,
D.R.,andW.E.Dietrich,
Where
dochannels
begin?,

References

Montgomery,
D. R., andW. E. Dietrich,
Source
areas,
drainage

Ph.D.dissertation,
421pp., Univ.of Calif.,Berkeley,1991.

Nature, 336,232-234,

1988.

density,
andchannel
initiation,
WaterResour.
Res.,25(8),1907-

1918, 1989.
Band,L. E., Topographic
partitionof watershedwith digitalelevaMontgomery,
D. R., andE. Foufoula-Georgiou,
Channel
network
tion models,WaterResour.Res., 22(1), 15-24, 1986.
sourcerepresentation
usingdigitalelevationmodels,WaterRe.
Beven,K., Runoffproductionandfloodfrequencyin catchment
of
sour. Res., 29(12), 3925-3934, 1993.
order n: An alternativeapproach,in ScaleProblemsin HydrolP.J.Wallbrink,
G. J. Burch,andE.M.
ogy, edited by V. K. Gupta et al., pp. 107-131,D. Reidel, Moore,I. D., S.M. Machay,
Norwell, Mass., 1986.

Beven,K., andM. J. Kirkby, A physicallybased,variablecontributingareamodelof basinhydrology,Hydrol.Sci.Bull.,24, 43-69,


1979.

Cabral, M., and S. J. Burges,DEMON: A new methodfor the


automatedextractionof contributingareasand drainagenetworks

fromrectangular
DEMs (abstract),EosTrans.AGU, 73(43),Fall
Meeting suppl., 202-203, 1992.

O'Loughlin,
Hydrologic
characteristics
andmodeling
of a small

forested
catchment
insoutheastern
NewSouth
Wales:
Prelogging
condition,J. Hydrol., 83, 307-335, 1986.

O'Loughlin,
E. M., Prediction
of surface
saturation
zonesinnatural

catchments
bytopographic
analysis,
WaterResour.
Res.,22(5),
794-804, 1986.

Panuska,
J. C., I. D. Moore,andL. A. Kramer,Terrainanalysis:
Integration
intotheagriculture
nonpoint
source
(AGNPS)
pollu-

tion model,J. Soil Water Conserr.,46(1), 59-64, 1991.


Dietrich,W. E., C. J. Wilson,D. R. Montgomery,
andJ. McKean,
P., K. Beven,
andO. Planchon,
Theprediction
ofhillslope
Analysisof erosionthresholds,
channelnetworks,andlandscape Quinn,
morphologyusinga digitalterrain model,J. Geol., I01(2), 259278, 1993.

flowpathsfor distributed
hydrological
modeling
usingdigital
terrain models,Hydrol. Process.,5(1), 59-79, 1991.

R. A., C. J. Wilson,D. M. Silburn,R. D. Connolly,


and
Hutchinson,
M. F., andT. I. Dowling,A continental
hydrological Vertessy,
assessment
of a new grid-baseddigitalelevationmodelof Australia, Hydrol. Process.,5(1), 45-58, 1991.

Jenson,S. K., Applications


ofhydrologic
information
automatically
extractedfrom digitalelevationmodels,Hydrol. Process.,5(1),
31-44, 1991.

C. A. Ciesiolka,
Predicting
erosionhazardareasusingdigital

terrainanalysis,IAHS AISH Publ., 192, 298-308, 1990.

D. R. Montgomery
and W. Zhang,Departmentof Geological
Sciences,Universityof Washington,Seattle,WA 98195.

Jenson,
S. K., andJ. O. Dominque,
Extracting
topographic
structurefromdigitalelevationdatafor geographic
information
system
analysis,Photogramm.Eng. RemoteSens.,54(11), 1593-1600, (ReceivedOctober25, 1993;revisedDecember9, 1993;
1988.

acceptedDecember 20, 1993.)

Vous aimerez peut-être aussi