1998 tucker

Upload: yessy9

Post on 06-Jul-2018

226 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/16/2019 1998 Tucker

    1/14

    WATER RESOURCES RESEARCH, VOL. 34, NO. 10, PAGES 2751-2764, OCTOBER 1998

    Hillslope processes, rainage density, and landscapemorphology

    Gregory E. Tucker and Rafael L. Bras

    Department of Civil and EnvironmentalEngineering,Massachusettsnstitute of Technology,Cambridge

    Abstract. Catchmentmorphology nd drainagedensityare strongly nfluencedby

    hillslopeprocesses. he consequencesf severaldifferent hillslopeprocessaws are

    explored n a seriesof experimentswith a numericalmodel of drainagebasinevolution.

    Five different modelsare considered,ncludinga simplediffusive-advectiverocess

    transition,a runoff generation hreshold,an erosion hreshold,and two typesof threshold-

    activated andsliding. hese different hillslopeprocesses lter both the visualappearance

    of the landscape nd the predicted elationshipbetweenslopeand contributingarea. On

    the basisof the different threshold heories,we derive expressionsor the relationships

    betweendrainagedensityand environmental actorssuchas rainfall, relief, and mean

    erosion ate. These relationships ary dependingon the dominanthillslope hreshold. n

    particular, he signof the predicted elationshipbetweendrainagedensityand relief is

    positive n semiarid, ow-relief andscapes nd negative n humid landscapes ominatedby

    a saturation hresholdand/or in high-relief andscapes ominatedby simple hreshold

    landsliding.

    1. Introduction

    The structure of catchment opographydepends o a large

    extent on the interactionbetween hillslope and channelpro-

    cesses.n recognitionof this, a number of quantitativemodels

    have been developed in recent years to explore three-

    dimensional drainage basin structure and evolution. These

    modelshave made it possible o simulate he long-term mor-

    phologicconsequencesf interactinghillslopeand fluvial pro-

    cesses e.g.,Kirkby, 1986;Ahnert, 1987; Willgoose t al., 1991;

    Howard, 1994;Tucker nd Slingerland, 997], o investigatehe

    scalingpropertiesof three-dimensionalerrain [e.g., Rigon et

    al., 1992],and to explore andscape volution n the contextof

    large-scaleectonics e.g.,Lifton and Chase,1992;Beaumontet

    al., 1992;Tuckerand Slingerland, 994].To date, however,most

    models have relied on a highly simplified representationof

    hillslopeprocesses. ittle attentionhasbeen paid, for example,

    to the role of factorssuchas landsliding, ariable runoff gen-

    eration, and vegetationcover n the contextof a catchmentas

    opposed o a hillslopeprofile. In this paper, we address ome

    of these ssues heoreticallyby modeling he influenceof dif-

    ferent typesof hillslope hresholdon catchmentmorphology

    and examining he implications or drainage density.

    One of the most basicpropertiesof a landscape s its degree

    of dissection, ften expressedn termsof drainagedensity.The

    transition rom straightor convexhillslopes o concave alley

    forms s widely understood o representa transition n process

    dominance, but the nature of that transition has been debated.

    Gilbert 1909] argued hat convex-concaveormsreflect a grad-

    ual transition n processdominance rom creep to wash with

    increasing istance rom a drainagedivide.Gilbert's model was

    quantified n terms of a linear stabilityanalysis y Smith and

    Bretherton 1972] and is based on the view that valleys orm

    where flow convergence auses ill or gully excavation y run-

    off erosion o outpace nfilling by diffusiveprocesses uch as

    rain splash. he instabilitymodel hasbeen extended o include

    Copyright1998 by the American GeophysicalUnion.

    Paper number 98WR01474.

    0043-1397/98/98 WR-01474509.00

    finite-scale ffects Loewenherz, 991] and more generalpro-

    cess aws Kirkby,1987, 1993;D. Tarboton,unpublishedmanu-

    script, 1994].

    An alternativeview is that valley and channel ormation are

    controlledby geomorphic hresholds. numberof researchers

    have argued, or example, hat hillslope-valley ransitionsoc-

    cur where a threshold or runoff erosion s regularlyexceeded

    during large storms Horton, 1945; Montgomery nd Dietrich,

    1989;Willgooset al., 1991]. Data on channelhead ocations n

    watersheds n the western United States show a correspon-

    dencebetweenvalleygradientand channelhead sourcearea, a

    finding hat appears o support he thresholdmodel [Montgom-

    eryand Dietrich,1989;Dietrichet al., 1992], houghperhapsnot

    uniquely. In fact, the threshold and instability theories need

    not be mutually exclusive; ather, the two models constitute

    end-member ases, nd any given andscapemaybe threshold-

    dominatedor instability-dominated, ependingon the climate,

    relief, geology,and stageof evolution Kirkby,1993, 1994].

    A threshold or runoff erosion s only one of severaldiffer-

    ent process hresholdshat may influence andscapemorphol-

    ogy and drainage density. Drainage density and landscape

    structure may alternatively be controlled, for example, by

    thresholds or runoff generation [e.g., Kirkby, 1980; Ijjdsz-

    Vdsquez t al., 1992;Dietrichet al., 1993] or by thresholds f

    slope stability [e.g., Montgomery nd Dietrich, 1989; Howard,

    1994]. Drainage density n particular may be controlled to

    varyingdegreesby any of these hresholds, nd each different

    thresholdmay producea different functional elationshipbe-

    tween drainagedensityand factors elated to climate,geology,

    and relief. For example, the slope models of Kirkby [1980,

    1993] predict that under a humid climate, drainage density

    shoulddecreasewith increasing elief, while under a semiarid

    climate, drainage density might be independent of relief.

    Howard [1997] presentsa detachment-limitedmodel in which

    the relationshipbetween drainage densityand mean erosion

    rate dependson (1) the dominanthillslope ransportprocess

    (creep or landsliding) nd (2) the presenceor absence f a

    threshold for runoff erosion.

    2751

  • 8/16/2019 1998 Tucker

    2/14

    2752 TUCKER AND BRAS: HILLSLOPE PROCESSES, DRAINAGE DENSITY, AND TOPOGRAPHY

    In this paper we developa theoreticaland numerical rame-

    work to explore two related questions:

    1. How do different typesof hillslope hreshold nfluence

    the three-dimensionalmorphologyof the landscape,both vi-

    suallyand in terms of the slope-areastatistic?

    2. How might these different thresholds hange he func-

    tional relationshipbetween drainagedensityand factorssuch

    as rainfall, erosion rate, and relief?

    The theory is implemented numerically n the form of a

    simulation model of drainage basin evolution. We show that

    each type of thresholdprocessmplies a different relationship

    between drainagedensity expressedn terms of valley head

    sourcearea) and external actors elated o tectonics, limate,

    and material properties.Our emphasiss on valley densityas a

    measureof landscape issection,ather than on the more vari-

    able property of stream network density;however, the two

    conceptsare closely elated, as was discussed y Montgomery

    and Dietrich [1994a].As a startingpoint we consider he simple

    case n whichhillslope-valleyopography esults rom compe-

    tition between low-driven rosionand creep-typediffusive)

    hillslope transport, as envisioned by Smith and Bretherton

    [1972]. We then consideralternativecases nvolving our dif-

    ferent typesof threshold: 1) a runoffgenerationhreshold, 2)

    a simple opographicallyasedslopestability hreshold,3) a

    pore pressure-dependent lope stability hreshold hat incor-

    porates he effectsof flow convergence,nd (4) a threshold or

    erosion by surfacerunoff.

    2. Landscape Evolution Model

    Computermodelingprovides ne of the few ways o study

    the long-term morphologicconsequences f short-term pro-

    cess aws and to conductcontrolled experiments n landscape

    evolution.The model used n this study s a variant of GOLEM

    [Tuckerand Slingerland, 996, 1997] and is basedon the prin-

    cipleof continuityof sedimentmasswithin the landscape.Mass

    continuity mplies that the rate of changeof elevation of a

    point on the landscapes proportional o the local divergence

    of sediment flux,

    O (Oqsxqs•l

    D-= - 55-x oy '

    wherez is elevation, is time, U is uplift or base evel owering

    rate,andqsxandqsyare sedimentluxesn thex andy direc-

    tions, respectively.The erosion term can be divided into a

    "wash" erm, representing article erosionby surface unoff,

    and a generichillslopeerosion erm H(x, y, t), in whichcase

    (1) may be rewritten

    Oz OQs

    ot = u •xx+ H(x,y, t), (2)

    where Qs is the time-averaged lux of water-borne sediment

    and x is a vector oriented along the directionof surface low.

    In the numericalmodel, a continuousopographic urface s

    approximated s a discrete attice of cells.Water and sediment

    are routed across his surfaceby assuminghat water entering

    each cell (either from upslopeor from direct precipitation)

    flowsdownhill n the steepest irection owardone of the eight

    surroundingneighborcells. n a landscapemantled by easily

    detachable egolith, he water-bornesediment lux Q s is equal

    to the total sediment ransport capacity referred to as the

    transport-limited ase),whereas f the surfacematerialsare

    resistant o detachment e.g., bedrock,coarseor cohesive ed-

    iment,or a thickly egetated urface),Qs will generally e less

    than he sedimentransport apacitythe detachment-limited

    case). For purposesof this study, we assumea transport-

    limited condition n the sense hat sediment s always rans-

    portedat its capacity or a givenprocess; owever, he capacity

    itself may be zero below a given detachment hreshold.

    The sediment ransport ate per unit width by flowingwater,

    qs, is modeled as a power function of slope S and specific

    surfacewater dischargeq:

    qs= ks(ktqS• - Oc)•, (3)

    where ks, kt, or, j•, and •/are constantsand Oc s an erosion

    threshold.A number of commonlyused sediment ransport

    formulas anbe cast n thisgeneral orm. This equationmaybe

    expressedn terms of total sediment ransport ate in a chan-

    nel, Qs, and total surfacewater dischargeQ by assuminghat

    the width of channelized low is related to discharge ccording

    to [Leopold nd Maddock, 1953;Yalin, 1992]

    W = kwQ% (4)

    where 60 s typically --0.5 for alluvial streams that value is

    assumed enceforth).We assumehat the width-dischargee-

    lationship lsoapplies o overlandand rill flow, or which here

    is some experimentalevidence Parsons t al., 1994]. Making

    this substitution,

    Qs= ks2Q'ø(kt2QO-'ø)St3Oc), (5)

    where ks2 = kskw and kt2 = ktk,7, . If Oc = 0, this reduces

    to

    Qs= kfQmsn, (6)

    wherekf = ks k •2 m = to + ( a - a ooy andn = /33,.

    Sediment ransport s modeledusing 6) in all but one of the

    experimentsdescribedbelow, with m = n = 2.

    The simulationdomain consists f a 48-by-48cell grid that

    representsan idealized square drainage basin. This domain

    size s large enough o producehillslope-valleyopography ut

    small enough o allow for reasonable omputation imes. The

    initial condition s a nearly flat surfaceseededwith a small

    randomperturbation n the elevationof each cell. The bound-

    ary condition or the model s a single ixedoutlet n one corner

    (Figure 1). The model s drivenby a uniform effective unoff

    rate P and base evel lowering at the outlet at a rate U. The

    runoff rate represents n effectivegeomorphicaverage,and

    would not necessarily e equivalent to mean annual runoff.

    The total surface-plus-subsurfaceunoff Qt(x, y, t) is assumed

    to be proportional o contributing reaA,

    Qt(x, y, t) = PA (x, y, t). (7)

    In all but one of the casesconsideredbelow, all flow is assumed

    to travel assurface unoff Q. In one experiment, he total flow

    Q t is apportionedbetweensurfaceand subsurfacelow accord-

    ing to a topographically ased runoff generationcriterion.

    Note that in contrast o the linear relationshipn (7), it could

    be argued hat long-termeffectivedischargemight nstead ary

    less han linearlywith drainagearea (as is often the case or

    floodsof a given ecurrencenterval).Were we to account or

    such an effect, the only changewould be a reduction n the

    effective alue of "m" in the expressionsor valleygradientand

    sourcearea developedbelow.

  • 8/16/2019 1998 Tucker

    3/14

    TUCKER AND BRAS: HILLSLOPE PROCESSES, DRAINAGE DENSITY, AND TOPOGRAPHY 2753

    DRAINAGE AREA

    (Logarithmic scale)

    ZERO ':'"'"'----•"'""'""••••:.-• MAX.

    Figure 1. Simulationn which hillslope-valleyopography volves hroughsimplecompetition etween

    creep nd unoff rosion0c = 0). Eachmodel ixel s 40 m by40 m; thecatchmentize s •-3.6km .

    Howard [1994] describes detachment-limitedmodel of

    drainage asinevolution hat is similar o the presentmodel n

    that it also represents unoff erosionas a power function of

    slope and discharge. n fact, althoughwe assume ransport-

    limited conditions as long as process hresholds re exceed-

    ed), most of the findings eported n this paper would also

    apply with minor modifications o Howard"sdetachment-

    limited theory.

    3. Simulations With Varying Hillslope Processes

    In each of the simulations described below, the model is run

    with a constant ate of baselevel owering at the outlet until a

    Table 1. Model Parameter Values

    Run Description Parameter* Value

    1 no thresholds P 1 m yr 1

    2 simplehresholdandsliding

    3 saturation threshold

    U 0.1 mmyr •

    k•e 10 8 yr m 3

    kd 10 2 m2 yr •

    0c 0

    •x 40 m

    Sc 0.466 (25ø)

    T l0 s m3 yr •

    (cellwidth)1

    0.8 (38.7

    l0 s

    5OO

    1/2

    2/3

    3

    3

    4 pore pressure-.ependent tan qb

    landsliding

    5 erosion threshold

    [p•Tb pwP

    oc

    kt2

    *Unlessnoted,parametersor runs2-5 are the sameas hose or run

    1

    balancebetweenuplift and erosion s reached.This dynamic

    equilibrium condition certainly does not apply to all land-

    scapes, ut it has he advantage f providinga consistent asis

    for comparison etweendifferent models.The simulations re

    summarized in Table 1.

    3.1. Process Competition

    The long-term average rate of sediment transport by soil

    creepand relatedprocessessuchas rain splash) s commonly

    assumedo be proportional o hillslopegradient, eading o the

    well-knownhillslopediffusionequation

    H(x, y, t) = [Oz/Ot]cr kaV2z, (8)

    where the subscript"cr" refers to creep erosion. Figure 1

    shows simulated'landscapeormed by the simultaneous c-

    tion of creep equation 8)) and surface low erosion equation

    (5)), assumingc = 0 (Table 1). This s arguablyhe simplest

    model one could magine hat produceshillslope-valleyopog-

    raphy,and it is a case hat hasbeen consideredn somedetail

    in previouswork [e.g., Kirkby, 1986; Willgoose t al., .1990;

    Tarboton t al., 19.92; oward, 1994].The closest atural analog

    would be a semiarid, ow- to moderate-relief andscapewith

    predominantlyHortonian overland flow (spatially uniform

    runoff generation),sparse egetation,and loosesurfacesoils.

    Hillslope-valley ransitionsn this casenaturally end to occur

    at those points where the rate of runoff-driven gullying out-

    pacesgully nfillingby diffusiveprocessesFigure 1).

    For a one-dimensionalillslope rofile, he point of transi-

    tion from diffusion-dominated to runoff-dominated erosion

    can be analyticallydefined as the point at which the two pro-

    cesses re equally effective n transportingsediment [e.g.,

    Howard, 1997]. At the point along the slopewhere both pro-

    cesses re equallyeffective,eachprocessmust ransporthalf of

  • 8/16/2019 1998 Tucker

    4/14

    2754 TUCKER AND BRAS: HILLSLOPE PROCESSES, DRAINAGE DENSITY, AND TOPOGRAPHY

    10

    1

    lO

    2

    1003 ................................. 710 10 10 10

    Contributing rea (rn2)

    Figure 2. Plot of surface lope in the downstreamirection)

    versus ontributing rea for the simulation ictured n Figure 1.

    the total sediment lux,which or a steady tateprofile s equal

    to the erosion rate U times the distance rom the divide,x. For

    creep ransport equation 8)), this mplies hat

    U

    • Uxt kdSt or St= • xt, (9)

    wherex t is the distance o the transitionpoint and S is the

    slopeat that point.For runofferosion equation 6)), assuming

    that runoff occurs as channelized flow in rills with the total

    width of flow described y (4), that Q = Pxb (b is unit width),

    and that 0c - 0,

    • Ux=kfpmxmS•r St--2kfpm t1-m)/n.10)

    Equating he two slopes, he distance o the processransition

    point is givenby

    ( ) 1-n)/(m+n-1)

    -- kf-1/(m+n-1)p-m/(m+n-1)k•/(m+n-1).11)

    Note that this includes he casem = 2, n = 1 considered y

    Kirkby1986],orwhichxt ckd/kf)/2.Equation11)may e

    usedas an approximationor valley-head ourcearea in irreg-

    ular topography Moglenand Bras, 1994; Howard, 1997] by

    replacing he slope distancex t with upstreamcontributing

    source area As:

    ( ) 1-n)/(m+n-1)

    soc •-l/(m+n-1)pm/(m+n-1) k•/(m+n-1).

    (12)

    This expression houldbe considered nly an approximation

    becauset neglectshe effectsof slopedivergence nd/orcon-

    vergence whichwould end to alter the relative mportance f

    creep ransport),nonlocal diffusive" ransportprocesseshat

    might dampensurface nstabilities t shortwavelengthsLoe-

    wenherz, 991],and the influenceof microtopographyn flow

    aggregationDunneet al., 1995].What the analysis oespro-

    vide, however, s an indicationof the relativedegreeof depen-

    dence f source reaon the controllingarametersf, P, U,

    andk a. The predicteddependenciesor the process-transition

    model may then be comparedwith modelsbasedon different

    typesof process hreshold.

    Interestingly, he sign of the relationshipbetween source

    area and the erosion ate U (equal to the uplift rate for a

    steady tatebasin)depends n the slopeparametern. As long

    as runoff-driven ediment ransportdependsmore strongly n

    gradient than does diffusive ransport i.e., n > 1), source

    area will tend to decreasewith increasing plift rate, and vice

    versa.Many sediment ransport ormulas mplyn ranging rom

    1 and 2 [e.g.,Howard, 1980;Moore and Burch, 1986;Willgoose

    et al., 1991; Tuckerand Slingerland, 996], in which case he

    theorypredicts weak nverse elationship etweenAs and U.

    By comparison, he detachment-limitedmodel of Howard

    [1997] predictsa positive elationshipbetweenerosion ate

    and sourcearea. This reflects he assumptionhat the runoff-

    driven erosionrate E is linearly proportional o bed shear

    stress% which mplies oughly --- Sø'7.However,he de-

    tachment-limitedmodelwouldalsopredictan inverse elation-

    shipbetweensourcearea and erosion ate if erosion ate were

    instead ssumedo be proportionalo •' raised o a powergreater

    thanabout /2 (for example,3would ive oughly ---S2).

    Figure 2 shows plot of slopeversus ontributing rea for

    the simulation ictured n Figure 1. The slope-area lot shows

    a characteristic turnover that reflects the transition from con-

    vex hillslopes o concavevalleys.Similar trends have been

    observed n some slope-areadata derived rom digital eleva-

    tion data [e.g.,Turboton t al., 1991;Montgomerynd Foufoula-

    Georgiou, 993; 14qllgoose,994], thoughmany data setsshow

    a more complexpattern [e.g., jjdsz-Vdsqueznd Bras, 1995;

    Tucker,1996].

    3.2. Threshold Landsliding

    In higher-reliefsettings,hillslopedenudation s typically

    dominatedby landsliding, ith landslides ommonly riggered

    onlywhen a threshold f soilor rock strength s exceeded. his

    thresholddependences not well described y the linear dif-

    fusionequation. n the next simulation, hallow egolith and-

    sliding s modeledusinga numericalapproachbasedon the

    thresholdconcept.To describe he combinedeffectsof soil

    creepand shallowandsliding,he hillslope edimentlux term

    may be written

    qhs= kdS + oo S > Sc,

    qhs= kdS + 0 otherwise

    (whereSc is a critical alueor threshold radient hat depends

    on materialstrength),which s related o the hillslope rosion

    term in (2) by continuity f mass,

    (Oqhsqhsh 13b)

    (x, , ) =- • + Oy '

    Equations 13) are implementedn the modelby applying n

    algorithm hat triggers "landslide" t anypoint on the model

    grid where S > Sc. Each such andslide emovesust enough

    material to reduce he slope o the thresholdvalue Sc. The

    material released rom each grid cell cascades ownslope

    across he grid until it reachesa locationwhere S < Sc, at

    which point it is deposited.This process epeats teratively

    during each model time step until no oversteepened oints

    remain. The net behaviorof this algorithm s similar o that of

    models that represent andslidingusing nonlinear diffusion

    [Kirkby, 987;Anderson nd Humphrey, 990;Howard,1994].

    Figure3 shows simulationn whichhillslope rosion ccurs

    by a combination f soil creep and threshold andsliding. e-

    cause he thresholdslope s assumedo be constant cross he

    landscape,he hillsidesend to be planarand do not develop

    concavehollows.Diffusive creep dominateson the convex

    ridge tops,where curvature s high. A natural analog o this

    model might be an arid or semiarid,moderate- o high-relief

    terrain with a thin regolith cover,suchas the badlands ormed

  • 8/16/2019 1998 Tucker

    5/14

    TUCKER AND BRAS: HILLSLOPE PROCESSES, DRAINAGE DENSITY, AND TOPOGRAPHY 2755

    Figure 3. Simulation n which hillslopesare dominated by simple threshold andsliding.

    on MancosShale on the Colorado Plateau [Howard, 1997]. n

    sucha setting, he thin regolithwould tend to saturatequickly

    during arge storms,so that the effectivestrengthof material

    on side slopeswould vary relatively ittle with topography.

    At steadystate the process ransition rom threshold ands-

    liding o fluvial erosion n the model occurs t pointswhere the

    equilibrium slope for runoff erosion ust equals the stability

    thresholdSc. Upstream of this location, luvial erosionalone

    cannot maintain a gradient that is gentler than the threshold

    gradient,and as a result, the slopesare insteaddominatedby

    landsliding. he equilibrium luvial gradientcan be obtained

    by noting hat the sediment ransport ate (equation 6)) must

    equal the erosion ate U times the contributing rea. Substi-

    tuting equation 7) for discharge,he equilibriumslope-area

    relationship s

    Seq A O-m)/n, (14)

    where the exponent erm (1 - m)/n describesongitudinal

    profile concavity nd is equal to the slopeon a log-logplot of

    streamgradientversusdrainagearea [Willgooset al., 1991]. f

    Sc is spatiallyuniform, the transition rom fluvial erosion to

    landslidingccurswhereSeq = Sc. The source rea at this

    transitionpoint is given by

    kS)/(m-)

    Z•s S• l-m). (15)

    Sediment ransport heoryand data suggestm • [ 1, 2]. Thus,

    unlike the creep-dominated ase, he theory predictsa strong

    positive elationshipbetweenerosion ate and sourcearea. A

    similar result was also found by Howard [1997] for a detach-

    ment-limitedmodel. The positiveU - As relationship eflects

    the fact that a higher erosionrate implies higher relief; with

    higher relief a greater areal proportionof a catchmentwill be

    susceptibleo slope ailure.

    Figure 4a showsa slope-areaplot from the simulationpic-

    tured in Figure 3. The landslide-dominated illslopes re rep-

    resentedby a seriesof pointswith a uniform slopeangle,equal

    to the stability hreshold.The transitionbetween he landslid-

    ing and fluvial domains s abrupt, reflecting he assumption f

    a spatiallyuniform Sc. In reality, one would expectsignificant

    (a) 10

    o10

    Threshold slope

    ..

    -2

    1003 .......................... 6 ....... 7

    10 10 10 10

    ContributingArea (m2)

    (b) 10 ...........................

    16;03 104 105 106 107

    Contributing rea (rn )

    Figure 4. Slopeversuscontributingarea for simple andslid-

    ing simulations.a) SpatiallyuniformSc. (b) Spatiallyuncor-

    relatedrandomvariation n Sc (dashed inesshow he rangeof

    variation n Sc).

  • 8/16/2019 1998 Tucker

    6/14

    2756 TUCKER AND BRAS: HILLSLOPE PROCESSES, DRAINAGE DENSITY, AND TOPOGRAPHY

    SURFACE FLOW RATE

    (Logarithmic scale)

    ZERO ................•' •••••'••• ......... MAX.

    white = unsaturated

    Figure 5. Effects of saturation-excessunoff on simulatedequilibrium andscape.

    scatter about this line, due to spatial variations n material

    strengthand variations n the time elapsedsince he last slide.

    Figure 4b shows he slope-areaplot from a simulation n which

    Sc varies n spaceas an uncorrelated andom ield. The mean

    trend through he scatterof pointsslopes ownward,ndicating

    that on average,hillslopepoints with a larger upstreamcon-

    tributingarea tend to havea slightly ower gradient.The down-

    ward trend reflects the mixed influence of points that are

    dominatedby landsliding low Sc) and those hat are domi-

    natedby runoff erosion high Sc).

    3.3. Runoff Production by Saturation Overland Flow

    One of the important limiting assumptionsn the previous

    cases and in many models) is the assumptionhat runoff

    production s uniform across basin.Suchan approximations

    reasonable or arid catchments nd/or catchmentswith imper-

    meablesoils,where nfiltration-excessverland low is the pri-

    mary runoff generationmechanism. n most humid and semi-

    humidenvironments, owever, unoff s generated rimarilyby

    saturation excess and return flow and thus varies as a function

    of topography nd soil thicknessDunne,1978].There is both

    theoreticaland field evidence o suggesthat saturation hresh-

    olds may influencedrainagebasinmorphology. irkby [1980]

    andMontgomery nd Dietrich 1989]hypothesizedhat a topo-

    graphicallydependentsaturation hresholdcan mposea limit

    to drainagedensity.This problemwasexplorednumerically y

    Ijjdsz-Vdsquezt al. [1992], who found that introducing sat-

    uration threshold in a model of basin evolution altered the

    predictedcatchment ypsometric urve. n the next simulation

    we explore he long-termgeomorphic onsequencesf a topo-

    graphicallydependentsaturation hreshold n a steadystate

    catchment nd examine ts implications or drainagedensity.

    For subsurfacelow parallel to the groundsurface, he sat-

    urated subsurfacelow rate can be represented s the product

    of soil transmissivity and surfaceslopeS,

    Qss TS, (16)

    where transmissivitys the depth ntegral of hydraulicconduc-

    tivity [Bevenand Kirkby, 1979; O'Loughlin,1986]. For steady

    state runoff the total flow is the product of contributingarea

    and he effective ainfallrate (equation 7)). The overland low

    component s the total flow minus the amount that travels n

    the shallow subsurface,

    Q=PA - TS, if PA > TS, (17)

    Q = 0, otherwise.

    Saturation occurs where

    A/S -> T/P, (18)

    with P now representingainfall rate minus osseso evapora-

    tion and deeper drainage.When this runoff threshold s incor-

    porated nto the landscape volutionmodel, the model pre-

    dicts an abrupt hillslope-valleyransition Figure 5). In this

    case, he catchmentmorphologys shapedby the interactionof

    three rather than two topographicallydependentprocesses.

    Given a high enough value of T/P, the model landscape

    evolves oward a state n which saturationoccursonly within

    the valleynetwork,with an abruptbreak n topography ccur-

  • 8/16/2019 1998 Tucker

    7/14

    TUCKER AND BRAS: HILLSLOPE PROCESSES, DRAINAGE DENSITY, AND TOPOGRAPHY 2757

    ring where return flow emergesat valley heads. The abrupt

    hillslope-valleyransition s partly a result of the dependence

    of subsurfacelow on topography.On hillslopes he increase n

    gradient away rom drainagedividesalso ncreases roundwa-

    ter flow capacity, hereby nhibitingsaturation.Ultimately, the

    convergence f flow lines at some point downstream orces

    saturation.The emergence f surface low at these ocations s

    associated ith a large increase n erosiveenergy,which pro-

    ducesan abruptreduction n slope.The abruptnature of valley

    headscan be seen as a distinctoffset on a plot of slopeversus

    contributing rea (Figure6a). Points o the left of the offsetare

    unsaturated nd thereforecontrolledsolelyby creep erosion.

    Points to the right are dominatedby fluvial erosion.

    The abrupt hillslope-valleyransition n Figures5 and 6a is

    alsoof coursepartly the result of usinga single ainfall rate to

    approximatea natural sequenceof stormsof varying size and

    intensity. This assumptioncan be relaxed by introducing a

    stochastic omponent o rainfall in the model. Figure 6b shows

    a slope-areaplot from a simulation in which the model is

    driven by a seriesof random storm events.Rainfall intensity,

    duration, and interstorm arrival time are chosen at random

    from an exponentialdistribution or each storm [Eagleson,

    1978; Tuckerand Bras, 1997]. The main effect of rainfall vari-

    ability s to blur the otherwise brupt hillslope-valleyransition

    and remove the sharp offset between hillslope and channel

    points (Figure 6b). Notably, the transitionbetweenhillslope

    points thosedominatedby diffusive ransport)and the fluvial

    points thosedominatedby runoff erosion) n Figure 6b still

    occursclose o the threshold or saturationby the mean rain-

    fall event. Abrupt basal concavitysimilar to that in Figure 5

    was also a feature of the hillslopeprofile modelsexploredby

    Kirkby 1993],which ncorporated ariability n rainstormsize.

    In the absenceof other significant hresholds, alley-head

    source rea n this model s controlledby the saturation hresh-

    old when T/P > ka/U (if ka/U is large, the minimum source

    area will be imposedby diffusive ransport,as in run 1). For

    this case, he sourcearea can be approximatedas that area for

    which he equilibriumchannelgradient s equal to the gradient

    requiredor saturationo occur Seq --Ssat).Theslope t the

    pointof saturations givenby (18). Settingt equal o Seq

    (equation 14)),

    As = A 1-m)/n

    ,

    (19)

    or

    rn/(m n- 1)U1/(m n- 1)

    As - kl/(m+n_l)pm+n/(m+n_l).20)

    Equation 20) describeshe point of intersection etween he

    saturation curve for the mean storm and the fluvial scaling

    curveon a slope-area lot (Figure 6) and predicts hat source

    area for a saturation-limitedcatchmentshould be positively

    correlated with soil transmissivity nd erosion rate, and in-

    versely correlated with mean rainfall intensity and material

    erodibility.The inversedependence n erosion ate U, which

    was alsoobserved y Kirkby 1993] n a one-dimensional lope

    model, implies that drainagedensityshoulddecreasewith in-

    creasing elief in humid climates.The physicalexplanation s

    that all elsebeing equal, a higher erosion ate impliessteeper

    slopes,which in turn implies an increasedsubsurface low

    capacityand decreasedoverland flow production. However,

    higher erosion ateswould also ikely be associated ith thin-

    1

    (a) 10

    10

    10

    2

    lO

    lO

    (b)

    10

    10

    ,'

    • o ogarithmicinnedverage

    3 4 5 6 7

    10 10 10 10

    Contributing rea (rn2)

    ß ,0

    10 10

    Contributingrea (rn )

    Figure 6. Plots of slope versuscontributingarea for simula-

    tionsdrivenby saturation-excessunoff production. a) Deter-

    ministicmodel.Note the offsetat the point of saturation. b)

    Model driven by discretestorm eventswith random, exponen-

    tially distributed ntensity,duration, and interstormperiod.

    ner soils, and thus reduced transmissivity. trictly speaking,

    therefore, the erosionrate-sourcearea relationshipshould

    only apply to cases n which most subsurfacelow is carried n

    the uppermostpart of the soil column, n which caseT would

    be relatively nsensitive o soil depth.

    3.4. Pore Pressure-ActivatedLandsliding

    The topographicallyasedsubsurfacelow model equations

    (16) and (17)) has alsobeen used to model the influenceof

    variations n water table height on landslidesusceptibility.

    Montgomery nd Dietrich [1994b]developed model of pore

    pressure-inducedhallow andsliding y combininghe infinite

    slopestabilitymodel or shallow oilswith equation 16), under

    the assumption f steady tate low (equation 7)). The result-

    ing criterion predictsslope nstabilityat locationswhere

    --->---- 1 sin0, (21)

    b pwP tan

    where 0 is the slopeangle, b s the angleof internal riction,b

    is contour width, and Ps and Pw are the densitiesof soil and

    water, respectively.Equation (21) applies only to slopes

    steeper han the maximumstableangle or fully saturated oils,

    which s givenby [Montgomerynd Dietrich,1994b]

    tan 0-> [(Os- Pw)/Ps]an (b • (1/2) tan

  • 8/16/2019 1998 Tucker

    8/14

    2758 TUCKER AND BRAS: HILLSLOPE PROCESSES, DRAINAGE DENSITY, AND TOPOGRAPHY

    DRAINAGE AREA

    (Logarithmic scale)

    ZERO -:-:•• .... MAX.

    Figure 7. Simulation n which landsliding s modeled using the pore pressurestability criterion of Mont-

    gomery nd Dietrich 1994b].Elevatedpore pressure roundpointsof flow convergenceeads o the formation

    of hollowsby landslideerosion.

    To model the effectsof pore pressure-inducedandsliding

    on catchmentmorphology, 21) is used as the slopestability

    criterion n the landscape volutionmodel (Figure 7). At each

    iteration, the critical drainageareaA c is computedaccording

    to (21). If A > A c, a landslide s initiated. Landslidedebris

    then cascades ownslopeuntil reachinga locationwhere S <

    (1/2)tan 45 the maximum tablegradient or saturated oil).

    The simulated andscape Figure 7) is characterized y the

    formation of landslide-dominated hollows and an overall in-

    crease n valley density elative to the previousexperiments.

    Unlike the uniform hreshold ase Figure 3), landsliding ere

    is restricted o the hollowsand lower slopes,where the pore

    pressure ends o be high. This model might be analogouso

    moderate-relief, soil-mantled andscapes uch as the Marin

    County region studiedby Montgomery nd Dietrich [1989]. A

    slope-areaplot from the simulation Figure 8a) reflects he

    shape of the slope stability envelope,and the distribution s

    similar to that predictedby the form instabilitymodel of D.

    Tarboton unpublishedmanuscript, 994). The slope-areadi-

    agramcanbe divided nto four regions from right to left): (1)

    the fluvialscaling rend, (2) a flat region hat represents atu-

    rated andsliding,3) a second ownward caling rend, repre-

    senting partially saturated landsliding,and (4) a level or

    slightly ncreasingrend that reflectsa combination f diffusion

    and unsaturated andsliding.Notably, these scalingzonesre-

    semble he zones dentified n digital elevationmodel (DEM)

    data by Montgomery nd Foufoula-Georgiou1993] and Ijjtisz-

    Vtisquez nd Bras [1995]. Spatialvariations n soil transmissiv-

    ity (Figure8b) and n tan 45 Figure8c) introduce catter n the

    slope-area elationship ut do not significantly lter the mean

    trend.

    The landslide-dominated hollows in the simulation shown in

    Figure 7 developbecause he convergence f subsurfacelow

    around initial perturbations educes he stability threshold,

    therebyacceleratinghe local erosion ate and attractingmore

    flow. In that sense, he modified andslidingmodel exhibitsa

    type of form instabilitysimilar o that considered y Smithand

    Bretherton 1972]. However, the landslidingmodel does not

    predict form instability under all circumstances. ather, the

    tendency oward form instabilitydependson the parameters

    governing runoff generation and erosion. Figure 9 illustrates

    three possibleoutcomes or a hypotheticalcatchment. The

    figure depicts he equilibriumslope-area elationship or run-

    off erosion RS) and diffusion DS), alongwith thresholdsor

    saturation SAT) and slopestability SLP). Points n the sim-

    ulation model will tend to follow one of the two process qui-

    librium lines, exceptwhere they are limited by the slope sta-

    bility threshold.Form instabilitywill tend to occur when the

    equilibriumgradient or the rate-limitingprocess runoff ero-

    sion, andsliding, r diffusion)decreases ith increasing on-

    tributing area [Tarbotonet al., 1992]. By this criterion, runoff

    erosion line RS) is an unstableprocess, hile diffusion line

    DS) is a stabilizing rocess. ore pressure-drivenandsliding

    can act as either a neutral processor as an unstable one,

    dependingon the catchmentproperties and position n the

    landscape. he processs neutral where the thresholdgradient

    is constant reflectedby a horizontal ine in Figure 9) and is

    unstablewhere the thresholdgradient decreaseswith increas-

    ing contributing rea (the curvingportion of the stabilityen-

    velope n Figure 9).

    If the saturation hresholdratio T/P is small (becauseof

    impermeable oils, or example) Figure9a), the steeper lopes

    will tend to be uniformlysaturatedduring storms, n which case

    the solution educes o the uniform Sc case as in the simula-

  • 8/16/2019 1998 Tucker

    9/14

    TUCKER AND BRAS: HILLSLOPE PROCESSES, DRAINAGE DENSITY, AND TOPOGRAPHY 2759

    tion in Figures and 4). If T/P is arge Figure9b), sideslopes

    will tend o have ow pore pressuresnd hereforewill tend not

    to fail until they approach he stability hreshold or dry soil

    (tan 4•). Hollowsmay form where sufficient low accumulates

    to reduce he slope stability hresholdand/or generateocca-

    sionalwasherosionduring arge storms thesehollowswould

    correspondo the point labeled"valleyheads" n Figure 9b).

    The tendency or landsliding o excavatevalleys would be

    greatest, owever, n the intermediate ase Figure 9c). Here

    diffusiongivesway to landsliding long he curvingportion of

    the slope stability curve, which representspositions n the

    landscapewhere the landslide hresholddependsstronglyon

    drainagearea. The simulation n Figures7 and 8 corresponds

    to this hird case.By comparison,he parameters erived rom

    DEM databyDietrichet al. [1992]seem o fall between he two

    casesdepicted n Figures9b and 9c.

    3.5. Erosion Threshold

    Horton [1945] hypothesized hat an unchanneledhillside

    representsa "belt of no erosion"within which overland flow

    strength s below a thresholdnecessaryor erosion.Montgom-

    eryandDietrich 1989, 1992]andDietrichet al. [1993]presented

    data on channelhead locationsand terrain morphology hat

    are consistentwith this hypothesis. he conceptof a runoff

    erosion hresholdhas been explored n severalmodels Will-

    gooseet al., 1991; Howard, 1994; Kirkby, 1994; Rinaldo et al.,

    1995;Tucker nd Slingerland,997]. n run 5 (Figure 10), the

    threshold erm in equation 5) is retained.The existence f a

    (a) 10 ................................

    "- ' "_'"'",'•.'t;,.........

    • .

    03 "•'"•'x.

    101

    103

    (b)

    1oø

    LU

    0310.1

    (c)

    ..

    10 10 10 10

    10ø

    LU

    • 10 1

    03

    øJ,.g:ø%,,

    '11;-I'

    logarithmicinned verage

    10 1O 10 10

    103

    %.

    o = logarithmic innedaverage o

    10 10 1;6 10

    DRAINAGE AREA (sq. m)

    Figure 8. Slope versus contributing area for simulations

    drivenby pore pressure-sensitiveandsliding.a) Spatially ni-

    form T and Sc. (b) Spatiallyuncorrelated andomvariation n

    Sc. (c) Spatiallyuncorrelated andomvariation n T.

    (a)

    o

    o

    Valley heads

    log( drainage area )

    (b)

    SLP / / •,5' heads

    log( drainagearea )

    (c)

    Valley • •.o /

    heads-• ,•'.

    log( drainagearea )

    SLP = slopestabilityhreshold

    RS = runoff rosion quilibriumlope

    DS = hillslope iffusionquilibriumlope

    SAT = saturation threshold

    Figure 9. Effect of slope stability envelope on slope-area

    distribution.Valley forms are predicted to correspond o

    pointsof form instability,which occurwhere the equilibrium

    slopedecreases ith increasing rainage rea. (Note that by

    definition he SAT line alwaysntersectshe break n the slope

    stabilitycurve).

    runofferosion hresholdn the model Figure10) lendsgreater

    curvature to the hillslope forms, which are now controlled

    solely by creep transport. Introduction of a threshold also

    changeshe predicted elationship etween alley-head ource

    area and parameters such as erosion rate [Kirkby, 1994;

    Howard,1997].Using he samederivation hat led to equation

    (12), the source rea for the threshold ase s givenby

    kf m]A[(1-m)/Y]-/3kt--• -½+/3)=aa

    (22)

    where/5= a(1 - co). hiscomplicated xpressionescribeshe

    generalcase n whichvalley ormationbeginswhere (1) the

    thresholds exceeded nd (2) there s sufficient rosive ower

    that diffusive rocesseso not inhibit ncision.A simplerend-

    membercase s that in whichvalley ormationbeginsclose o

    the point at which the erosion threshold s exceeded.This

    might be the case, or example, n landscapes ith a resistant

    vegetation mat [Dietrichet al., 1993; Kirkby, 1994; Howard,

    1996].The source rea at the pointwhere he equilibrium ill-

    slope radient ecomesuststeep noughhatkt2P•AaS3 - 0c

    (equation 5)) is givenby

    k •./(/5/3) 1/(15+/3)

    As= •/(•+/3)/(•+U/3/(•+/3) (23)

    P /3) ,

  • 8/16/2019 1998 Tucker

    10/14

    2760 TUCKER AND BRAS: HILLSLOPE PROCESSES, DRAINAGE DENSITY, AND TOPOGRAPHY

    Figure 10. Simulated andscapewith a finite runoff erosion hreshold.

    where •5and/3 are typically --1/3 and ---2/3,respectively e.g.,

    Tuckerand Slingerland, 997]. This would be appropriate or

    large thresholds,when the thresholddominatesover stability

    considerationsKirkby,1993].Note that 0c representshe ero-

    sional resistance of the unchanneled surface rather than that of

    sedimentgrainswithin an established hannel,whichmight be

    lower. This sourcearea expressionmplies hat sourcearea in

    a threshold-limited atchmeritwill tend to vary positivelywith

    hill. lope iffusivitya function f climate ndmaterial)and 0c

    (likely a function of vegetationdensity)and to vary inversely

    with erosion rate and effective rainfall rate. The threshold

    control on valley-head ocations s illustrated n Figure 11.

    It is interesting to note that when 0c > 0, the model no

    longer predictsa simplepower law relationshipbetween slope

    and drainagearea for an equilibrium iver network though n

    practice, he slope-area elation might still closely esemblea

    power aw). With the inclusionof a threshold erm, the equi-

    librium slope-area elationship or the drainagenetwork is

    S -- A l-m)/7t_,7A 15 (24)

    where the prime on 0'c indicates that it now represents a

    threshold for sediment entrainment within an established

    channel. If the threshold term is small, the solution reduces to

    S crAt(1-m)/'•l equation14)),whereasf the hresholderm

    dominates, crA-("(•-ø'))/•. These wo end-member ases

    might correspond o fine-bed alluvial and threshold gravel

    channels, espectively Howard, 1987]. Both casesdescribea

    power-lawrelationshipbetween slope and contributingarea,

    but generally with different exponents.On the other hand,

    recent modelingwork with multiple grain sizes Gaspariniet

    al., 1997] suggestshat the dependenceon mean grain size

    implied by (24) may not apply to streamswith bimodal bed

    sediment mixtures.

    4. Discussion

    An important mplicationof the modelswe have analyzed s

    that the relationship etween drainage ensity nd factorssuch

    as climate, geology,and relief will vary in a predictableway

    dependingon the nature of geomorphicprocesses perating

    on hillslopes.We have quantified his relationship n terms of

    valley-headsourcearea rather than directly n terms of drain-

    age densityD•, but the two concepts re closely elated.Mo-

    glenet al. [1998] analyzed he relationshipbetweensourcearea

    and D• in DEM data for sevenbasins n the United States.For

    each data set, they varied the sourcearea used to extract the

    channelnetwork and compared t with the resultingdrainage

    density.The results showeda power law relationship,D• •

    A j-•, with • • 0.5, aswould be expectedgiven he dimensions

    : I IIq'

    ::::.".,.,,,

    0 , "%

    1(52 ...................................

    10 lC• 10 14

    Contributingrea rn )

    Figure 11, Slope-area plot from model run with erosion

    threshold shown n Figure 10).

  • 8/16/2019 1998 Tucker

    11/14

    TUCKER AND BRAS: HILLSLOPE PROCESSES, DRAINAGE DENSITY, AND TOPOGRAPHY 2761

    Table 2. Predicted aluesof K n the RelationDd ocR K,Assuming d oc1/A /2

    Case

    K (• = 1/2, K (•: 1/3, • (•: 2/3,

    /3-- 2/3, /3 = 2/3, /3 = .2/3,

    3' = 3)* 3' = 3)? 3' = 3/2)$

    Creep, uniform runoff generation,Oc = 0

    Creep, uniform runoff generation,{}Oc> 0

    Creep, saturation-excessunoff productionll

    Landsliding, patiallyuniform Sc

    1/3 2/5 0

    4/7 2/3 1/2

    (-1/3) (-2/5) (-2/3)

    -1 -2 -1

    *Correspondso rn = 2, n = 2 when 0c = 0 (this paper,Kirkby 1987] and Willgooset al. [1991]).

    ?Correspondso rn = 3/2, n = 2 when Oc = 0 [e.g.,Howard, 1994].

    $Correspondso the rn - 3/2, n = 1, caseconsidered y Kirkby [1993] with an additionalconstraint

    on flow width (equation 4)).

    {}Applicable nlywhen the threshold tself s not the biggest ontributor o relief (when he threshold

    term in equation 24) is smallrelative o the other term).

    IIAssumes is independentof relief.

    of length and area. For the sake of argument,we will assume

    thatDd oc •-ø's.Assuminglarger alueof '1,such sDd

    A•- • [MontgomeryndDietrich, 989],would mplya stronger

    relationshipbetween drainagedensityand the various actors

    that control sourcearea, but would not alter the sign or the

    relative magnitudeof the dependenceon different factors.

    4.1. Drainage Density and Relief

    Of particular interest s the relationshipbetween drainage

    densityD,• and topographic elief R. Topographic elief is not

    one of the governingariablesn our theoretical odel or

    source area, but it can be deduced from three of those vari-

    ables,U, P, and kf. There existmanydifferent elief-based

    measures, uchas otal relief (maximumelevationabovecatch-

    ment outlet), relief ratio (total relief divided by catchment

    length), and local relief (maximumelevationdifference n an

    analysiswindowof specified ize). n a large catchment,most

    of the reliefwill arise rom the elevation hange long he main

    streambetween he outlet and headwaters.Recognizing his,

    we can define a "drainage network relief" R,• as the total

    elevationifferenceetweenheoutlet nda poin{ longhe

    main streamat a distance L upstreamof the outlet, where L

    is the total length of the main stream and p is a suitable

    proportion e.g.,90%). The main streammaybe definedas he

    longeststream n the basin [Hack, 1957] or as the streamwith

    the largestcontributing rea at any givenconfluence. he

    statistichas severaladvantages:t is relativelyeasy o measure

    in a DEM; it can be readilyobtained rom theoreticalmodels;

    and because treamprofilesare typically ogarithmic, t does

    not depend stronglyon catchmentsize. On the other hand,

    because f the assumptionhat mostof the relief occurswithin

    the luvial etwork,heR,•statisticrovidesnapproximation

    of total relief only at spatialscales hat are significantlyarger

    than typicalhillslope-valleywavelengths.

    Drainage network relief R d can be derived rom the model

    as follows.Assuminga typicaldrainagebasingeometry, t can

    be shown hat (14) impliesa power aw relationship etween

    relief and steadystate erosion ate, U,

    e oc kp m)n•l, (25)

    with a dependence n basin size only if the slope-areaexpo-

    nent (1 - m)/n is significantly ifferent from 0.5. Among

    other things, his relationshipprovidesa meansof comparing

    the theoreticalmodelwith empiricalstudies f denudation ate

    as a function of relief [Schummand Hadley, 1961; Schumm,

    1963;Ahnert, 1970; Summerfield nd Hulton, 1994]. Equation

    (25) may be substituted nto the source-areaexpressions

    (equations12), (t5), (20), and (23)) to solve or source rea as

    a function of relief. Table 2 listsvaluesof the parameter K in

    the power aw relationD,• ocR,• for typicalvaluesof m, n, 3,

    /3, and3,.Note hatn isderived nder heassumptionhatk,•,

    kf, P, andotherparametersreprimarily ontrolled yclimate

    and/or material propertiesand therefore do not vary system-

    aticallywith R,•. For most of the parameters, his s probablya

    reasonableassumption.Soil transmissivity , however,might

    be expected o varywith relief owing o associated ariations n

    regolith hickness. or this reason, he saturation-excessunoff

    modelshouldbe considered pplicable nly o landscapes ith

    thick soils (which generallyexcludesmountainous errain)

    and/or soils in which most subsurface runoff travels in the

    upper ew centimeters. or such andscapeshe model predicts

    an inversecorrelationbetween elief and drainagedensity,as

    compared with landscapesdominated by Horton overland

    flow, in which case he model prescribeshe opposite elation-

    ship. This finding is consistentwith Kirkby's 1987] stability

    analysis, which predicted a positive relationship between

    sourcearea and valley gradient or humid (saturationdomi-

    nated) settings nd an inverse elationshipor semiarid infil-

    tration excess-dominated)ettings for the casen = 2).

    In the first case n Table 2 (creep-dominated,Hortonian

    runoff with a negligible hreshold),drainage density s inde-

    pendent of relief only when n •- 1. Thus an observation hat

    drainagedensitydoes vary with relief could not by itself be

    taken as evidence or thresholdbehaviorwithout ndependent

    knowledgeof the climate and of the physical elationshipbe-

    tween slopeand sediment ransportefficiency.

    Notably, although he theory predictsa positivecorrelation

    between D,• and R,• for Hortonian, creep-dominated and-

    scapes,t also predictsa strong nversecorrelation or land-

    slide-dominated nes. Such a difference appears o be sup-

    ported by observations.n low- to moderate-relief settings,

    moststudies howa positivecorrelationbetweenD,• and relief.

    Schumm 1956] found a positivecorrelationbetweenD,• and

    the relief ratio in a humid badlands.Montgomery nd Dietrich

    [1989, 1994a] eporteda positivecorrelationbetweendrainage

    densityand valley gradient n stream-headhollows n the west-

    ern United States. A positive correlation between drainage

    density nd slopehasalsobeenobservedn experimentalwork

    [Schumm t al., 1987]. n contrast o the data from thesemostly

    (thoughnot exclusively)ow- to moderate-reliefcases,Oguchi

    [1997] found an inverserelationshipbetween relief and the

    densityof deepvalleys n mountainouserrain n Japan. Ogu-

  • 8/16/2019 1998 Tucker

    12/14

    2762 TUCKER AND BRAS: HILLSLOPE PROCESSES, DRAINAGE DENSITY, AND TOPOGRAPHY

    Slope/area trend for WE38 catchment

    1.000 - ' '

    o.oo -+

    0.010

    0.001

    101 102 103 104 105 106 107

    Drainage Area (square meters)

    Figure 12. Slope versuscontributingarea obtained from a

    5-m-resolution igital elevationmodel (DEM) of the WE-38

    watershed,centralPennsylvania, .S.A. Each point represents

    an averageof 100 consecutive EM pixels,arranged n order

    by contributing rea.

    chi also analyzed different methods for measuringD a and

    showed hat the results depend on the criteria used. If only

    well-definedvalley orms are identified as drainages, here is a

    clear inverseD a-R relationship,whereas f all contour nflec-

    tions are considered, the relationship disappears. In the

    present contextwe are concernedwith D a as a measureof

    landscape issection, o the first method s more meaningful.)

    The model resultssuggesthat the varyingbehaviorof Da with

    respect o R may reflect differencesn process ominanceon

    hillslopes.Given that landsliding s active n the basinsstudied

    by Oguchi 1997] and colleagues, e suggesthat their findings

    reflect an increasingareal proportion of planar, landslide-

    dominatedhillslopes n steeper errain. This interpretation s

    also consistentwith the study of Howard [1997], in which he

    documented a change from a positive to a negative R-Da

    relationshipwith increasing elief in badland andscapes n the

    Colorado Plateau, U.S.A.

    4.2. Drainage Density and Climate

    Each of the modelswe have considered redictsa positive

    correlation between D d and the rainfall parameter P. The

    predictedsensitivitys greatest or landslide-dominated atch-

    ments and for catchments in which channel network extent is

    limited by a saturation hreshold.However, he predictedsen-

    sitivity o P doesnot account or changesn other factors hat

    are also ikely to be correlatedwith climate.Most studies f the

    relationshipbetween climate and D d show either an inverse

    relation [Melton, 1958] or a humped curve in which D a in-

    creaseswith mean annualprecipitationunder semiarid o arid

    climates,and decreases nder more humid climates Gregory

    and Gardiner,1975;Moglenet al., 1998].Moglenet al. [1998]

    emphasizehe importanceof vegetationas a limiting actoron

    D a. In the context of this model, vegetationwould serve to

    increaseboth surface esistanceOcand transmissivity , both

    of which end to decrease a (equations 23) and (23)). There

    is alsoevidence hat creepdiffusivity a maybe correlatedwith

    climate. ield estimatesange rom -2 x 10 4 m2/yr n the

    aridAravaValley Enzel tal., 1996] o -10 -2 m2/yrn humid

    to subhumid ettings n the United States Nash, 1980;Hanks

    et al., 1984;Rosenbloom ndAnderson,1994].Higher diffusiv-

    ity under a humid climate would tend to decreaseD d, and

    converselyor an arid climate. ncreasedweatheringrates un-

    der a humid climatewould also end to increase egolith thick-

    ness, hereby increasingT and decreasingD a. All of these

    factorsmay contribute o counterbalancinghe positiveeffects

    of increased ainfall on Dd. In addition,changesn precipita-

    tion variabilitywould be likely to influencedrainagedensity

    [Tucker ndSlingerland,997]. heparametersf andP lump

    information about both mean rainfall and rainfall variability,

    so that P is not necessarily quivalent o mean annual rainfall

    or runoff. Given the importanceof large, relatively nfrequent

    events n shaping andscapes,t is possible hat the geomor-

    phicallyeffective unoff rate (if indeed here is sucha thing)

    actually ncreasesunder an arid climate with highly variable

    precipitation e.g.,Knox, 1983].

    4.3. Process "Fingerprints" in Slope-Area Data

    Each of the different processes xploredhere has an impact

    on the slope-area elationship,suggestinghe possibility hat

    slope-areadata may be used o discriminate etweendifferent

    geomorphicprocess egimes Tarbotonet al., 1992;Dietrich et

    al., 1993; Montgomery nd Dietrich, 1994] as well as to test

    theories of slope and landscapedevelopment [e.g., Kirkby,

    1993].n eachof themodels onsidered,illslope.process(es)

    introduceone or more inflectionpoints n the mean slope-area

    trend. Such inflectionshave been observed n many data sets

    [Tarbotonet al., 1991; Montgomery nd Foufoula-Geourgiou,

    1993;Wiltgoose,994; jjdsz-Vdsqueznd Bras, 1995],and their

    significance as been debated. jjdsz-Vdsqueznd Bras [1995]

    recognized hat a commonpattern in many (thoughnot all)

    data setswas the presenceof two apparentlyparallel scaling

    trends,with the larger one representing he main valley net-

    work (Figure 12). Of the variousmodels, he slope-arearend

    producedby the landslide-dominated imulationsmost closely

    reproduces his pattern. Montgomery nd Foufoula-Georgiou

    [1993]alsonoted the presence f multiple nflectionsn slope-

    area data, thoughwith somewhatdifferentpatterns han those

    in Figure 12 and in the data of Ijjdsz-Vdsqueznd Bras [1995],

    and they interpreted the larger-area nflection as a transition

    from debris flow-dominated channels to alluvial channels. The

    data of Figure 12 are from a moderate-relief catchment n

    Pennsylvania n which landsliding s not today a significant

    process.However, t is possible hat the landslide-like signa-

    ture" reflectsmassmovementby solifluctionunder a perigla-

    cial climateduring he last glaciation.Alternatively, he pattern

    may be controlledby low-gradient, ow-area pixelswithin the

    largervalleys.Further simulationmodelingmay help elucidate

    the origin of these ypesof pattern in slope-areadata, though

    at present his approach s hamperedby a lack of DEM data of

    sufficient esolution o properlymap andscape tructure t the

    hillslopescale.

    5. Conclusions

    The modelswe have explored n this paper providea frame-

    work for understanding he impact of different hillslopepro-

    cesses nd process hresholdson the structureof terrain. A

    numberof important ssues emain unanswered,ncluding he

    role of vegetation,whichhasa clear mpacton drainagedensity

    and sedimentyield, and the role of short-termvariability in

    rainfall and runoff. Our analysis as considered nly equilib-

    rium landscapes, nd it shouldbe emphasized hat thesemod-

    els apply o landscapeshat have had sufficient ime to adjust

  • 8/16/2019 1998 Tucker

    13/14

    TUCKER AND BRAS: HILLSLOPE PROCESSES, DRAINAGE DENSITY, AND TOPOGRAPHY 2763

    to prevailing onditionsbase evel owering ate, rainfall,etc),

    but do not necessarily escribewhat might happen for exam-

    ple to drainage density) in the short term in response o

    changes n environmentalconditions.Such issuesmight be

    addressedhrougha combination f analysis f high-resolution

    DEM data, dating of geomorphic eatures in specific and-

    scapes, nd modelingof transient andscape tatesand process

    interactions.

    Acknowledgments. This researchwas supportedby the U.S. Army

    ConstructionEngineering Research Laboratory (DACA 88-95-R-

    0020) and the Army ResearchOffice (DAAH 04-95-1-0181).This

    work hasbenefited rom discussionsith Ying Fan, Nicole Gasparini,

    Stephen Lancaster, Jeff Niemann, Daniele Veneziano, and Kelin

    Whipple, and from thoughtful eviewsby Alan Howard, Glenn Mo-

    glen, David Montgomery,and David Tarboton.

    References

    Ahnert, F., Functionalrelationships etweendenudation, elief and

    uplift n largemid-latitudedrainagebasins,Am. . Sci.,268, 243-263,

    1970.

    Ahnert, F., Process-responseodelsof denudation t differentspatial

    scales,CatenaSuppl.,10, 31-50, 1987.

    Anderson,R. S., and N. F. Humphrey, nteractionof weatheringand

    transportprocessesn the evolutionof arid landscapes,n Quanti-

    tativeDynamic Stratigraphy, dited by T. Cross,pp. 349-361, Pren-

    tice-Hall, EnglewoodCliffs, N.J., 1990.

    Beaumont, C., P. Fullsack, and J. Hamilton, Erosional control of active

    compressionalrogens,n ThrustTectonics,ditedby K. R. McClay,

    pp. 1-18, Chapmanand Hall, New York, 1992.

    Beven,K. J., and M. J. Kirkby,A physically ased ariablecontributing

    area modelof basinhydrology, ydrol.Sci.Bull., 24(1), 43-69, 1979.

    Dietrich, W. E., C. J. Wilson, D. R. Montgomery, J. McKean, and

    R. Bauer,Erosion hresholds nd andsurfacemorphology, eology,

    20, 675-679, 1992.

    Dietrich, W. E., C. J. Wilson, D. R. Montgomery, and J. McKean,

    Analysis of erosion thresholds,channel networks, and landscape

    morphologyusing a digital terrain model, J. Geol., 101, 259-278,

    1993.

    Dunne, T., Field studiesof hillslope low processes,n HillslopeHy-

    drology, ditedby M. J. Kirkby,pp. 227-293, JohnWiley, New York,

    1978.

    Dunne, T., K. X. Whipple, and B. F. Aubry, Microtopographyof

    hillslopesand initiation of channelsby Horton overland flow, in

    Natural and Anthropogenic nfluences n Fluvial Geomorphology,

    Geophys. onogr.Ser.,vol. 89, editedby J. E. Costaet al., pp. 27-44,

    AGU, Washington,D.C., 1995.

    Eagleson,P.S., Climate, soil, and vegetation,2, The distributionof

    annualprecipitation erived rom observed tormsequences, ater

    Resour. Res., 14, 713-721, 1978.

    Enzel, Y., R. Amit, N. Porat, E. Zilberman, and B. J. Harrison, Esti-

    mating he agesof fault scarpsn the Arava, Israel, Tectonophysics,

    253, 305-317, 1996.

    Gasparini,N.M., G. E. Tucker, and R. L. Bras,Downstream ining:A

    drainage asinperspectiveabstract), os Trans.,AGU, 78(46), Fall

    Meet. Suppl.,F282-F283, 1997.

    Gilbert, G. K., The convexity f hilltops, . Geol., 17, 344-350, 1909.

    Gregory, K. J., and V. Gardiner, Drainage densityand climate, Z.

    Geomorphol., eue Folge,19, 287-298, 1975.

    Hack, J. T., Studiesof longitudinalstream profiles n Virginia and

    Maryland, U.S. Geol. Surv.Prof Pap., 294-B, 97 pp., 1957.

    Hanks, T. C., R. C. Bucknam, K. R. Lajoie, and R. E. Wallace,

    Modificationof wave-cutand faulting-controlledandforms, . Geo-

    phys.Res., 89, 5771-5790, 1984.

    Horton, R. E., Erosionaldevelopment f streams nd their drainage

    basins;Hydrophysical pproach o quantitativemorphology,Geol.

    Soc. Am. Bull., 56, 275-370, 1945.

    Howard, A.D., Thresholds n river regimes, n Thresholdsn Geomor-

    phology, ditedby D. R. Coates,and J. D. Vitek, pp. 227-258, Allen

    and Unwin, Winchester, Mass., 1980.

    Howard, A.D., Modelling luvialsystems: ock-, gravel-and sand-bed

    channels, n River Channels, dited by Richards,pp. 69-94, Basil

    Blackwell, New York, 1987.

    Howard, A.D., A detachment-limitedmodel of drainagebasin evolu-

    tion, Water Resour. Res., 30, 2261-2285, 1994.

    Howard, A.D., Thresholds and bistable states in landform evolution

    models abstract),Eos Trans.AGU, 77(17), SpringMeet. Suppl.,

    S136, 1996.

    Howard, A.D., Badland morphologyand evolution: nterpretation

    using a simulationmodel, Earth SurfaceProcessesandforms,22,

    211-227, 1997.

    Ijj•isz-V•isquez, ., andR. L. Bras,Scaling egimes f localslope ersus

    contributing rea n digitalelevationmodels,Geomorphology,2(4),

    299-311, 1995.

    IjjJsz-V•isquez,E. J., R. L. Bras, and G. E. Moglen, Sensitivityof a

    basin evolutionmodel to the nature of runoff productionand to

    initial conditions, Water Resour. Res., 28, 2733-2741, 1992.

    Kirkby, The stream head as a significantgeomorphic hreshold, n

    Thresholdsn Geomorphology,dited by D. R. Coates, and J. D.

    Vitek, pp. 53-73, Allen and Unwin, Winchester,Mass., 1980.

    Kirkby, M. J., A two-dimensional imulationmodel for slope and

    streamevolution, n HillslopeProcesses,dited by A.D. Abrahams,

    pp. 203-222, Allen and Unwin, Winchester,Mass., 1986.

    Kirkby, M. J., Modelling some nfluences f soil erosion, andslides nd

    valleygradienton drainagedensity nd hollowdevelopment, atena

    Suppl.,10, 1-14, 1987.

    Kirkby,M. J., Long erm nteractions etweennetworks nd hillslopes,

    in ChannelNetworkHydrology, ditedby K. Beven,and M. J. Kirkby,

    pp. 255-293, John Wiley, New York, 1993.

    Kirkby, M. J., Thresholdsand instability n stream head hollows:A

    model of magnitudeand frequency or washprocesses,n Process

    Modelsand TheoreticalGeomorphology,dited by M. J. Kirkby, pp.

    295-314, John Wiley, New York, 1994.

    Knox, J. C., Responses f river systemso Holocene climates, n Late

    Quaternary nvironments f the UnitedStates, ol. 2, The Holocene,

    editedby H. E. Wright and S.C. Porter, pp. 26-41, Univ. of Minn.

    Press,Minneapolis,1983.

    Leopold,L., and T. Maddock,The hydraulicgeometryof streamchan-

    nels and some physiographicmplications,U.S. Geol. Surv. Prof

    Pap., 252, 1953.

    Lifton, N. A., and C. G. Chase,Tectonic,climaticand lithologic nflu-

    enceson landscaperactal dimension nd hypsometry:mplications

    for landscape volution n the San Gabriel Mountains, California,

    Geomorphology, , 77-114, 1992.

    Loewenherz, D. S., Stability and the initiation of channelizedsurface

    drainage:A reassessmentf the shortwavelength imit, J. Geophys.

    Res.,96(B5), 8453-8464, 1991.

    Melton, M. A., Correlation structureof morphometricpressureof

    drainage basins and their controlling agents,J. Geol., 66, 35-56,

    1958.

    Moglen, G. E., and R. L. Bras, Simulationof observed opography

    usinga physically-basedasin evolutionmodel, Hydrol. and Water

    Resour.Rep. 340, Ralph M. ParsonsLab., Mass. Inst. of Technol.,

    Cambridge, 1994.

    Moglen, G. E., E. A. B. Eltahir, and R. L. Bras, On the sensitivity f

    drainagedensity o climate change,WaterResour.Res.,34, 855-862,

    1998.

    Montgomery, D. R., and W. E. Dietrich, Sourceareas, drainage den-

    sity,and channel nitiation, WaterResour.Res.,25, 1907-1918, 1989.

    Montgomery, D. R., and W. E. Dietrich, Channel initiation and the

    problem of landscapescale,Science, 55, 826-830, 1992.

    Montgomery, D. R., and W. E. Dietrich, Landscapedissectionand

    drainage area-slope hresholds, n ProcessModels and Theoretical

    Geomorphology,dited by M. J. Kirkby, pp. 221-246, John Wiley,

    New York, 1994a.

    Montgomery,D. R., and W. E. Dietrich, A physically asedmodel for

    the topographic ontrol on shallow andsliding,WaterResour.Res.,

    30, 1153-1171, 1994b.

    Montgomery, D. R., and E. Foufoula-Georgiou, Channel network

    source epresentation singdigital elevationmodels,WaterResour.

    Res., 29, 3925-3934, 1993.

    Moore, I. D., and G. J. Burch, Sediment ransport capacityof sheet

    and rill flow:Applicationof unit streampower heory,WaterResour.

    Res., 22, 1350-1360, 1986.

    Nash, D. B., Morphologicdatingof degradednormal fault scarps, .

    Geol., 88, 353-360, 1980.

    Oguchi,T., Drainage densityand relativerelief in humid steepmoun-

  • 8/16/2019 1998 Tucker

    14/14

    2764 TUCKER AND BRAS: HILLSLOPE PROCESSES, DRAINAGE DENSITY, AND TOPOGRAPHY

    tainswith frequent slope ailure, Earth Surface rocessesandforms,

    22, 107-120, 1997.

    O'Loughlin, E. M., Prediction of surfacesaturationzones n natural

    catchments, Water Resour. Res., 22, 794-804, 1986.

    Parsons,A. J., A.D. Abrahams, and J. W. Wainwright, On determin-

    ing resistanceo interrill overland low, WaterResour. es.,30, 3513-

    3521, 1994.

    Rigon, R., A. Rinaldo, and I. Rodriguez-Iturbe,On landscape elf-

    organization, . Geophys. es.,99, 11,971-11,993,1994.

    Rinaldo, A., W. E. Dietrich, R. Rigon, G. Vogel, and I. Rodriguez-

    Iturbe, Geomorphologicalignatures f varying limate,Nature,374,

    632-634, 1995.

    Rosenbloom,N. A., and R. S. Anderson,Hillslope and channelevo-

    lution in a marine terraced landscape,Santa Cruz, California, J.

    Geophys.Res., 99, 14,013-14,030, 1994.

    Schumm,S. A., Evolutionof drainagesystems nd slopes n badlands

    at Perth Amboy, New Jersey,Geol. Soc.Am. Bull.., 67, 597-646,

    1956.

    Schumm,S. A., The disparitybetweenpresent atesof denudationand

    orogeny,U.S. Geol. Surv.Prof. Pap., 454-H, 13 pp., 1963.

    Schumm,S. A., and R. F. Hadley, Progress n the applicationof

    landform analysis n studiesof semiarid erosion, U.S. Geol. Surv.

    Circ., 437, 1-14, 1961.

    Schumm,S. A., M.P. Mosley,and W. E. Weaver,Experimental luvial

    Geomorphology,13 pp., John Wiley, New York, 1987.

    Smith, T. R., and F. P. Bretherton, Stability and the conservation f

    mass n drainagebasinevolution,WaterResour.Res.,8, 1506-1529,

    1972.

    Summerfield, M. A., and N.J. Hulton, Natural controls of fluvial

    denudation ates in major world drainagebasins, . Geophys. es.,

    99, 13,871-13,883, 1994.

    Tarboton, D. G., R. L. Bras, and I. Rodr/guez-Iturbe,On the extrac-

    tion of channel networks rom digital elevation data, Hydrol. Pro-

    cesses, , 81-100, 1991.

    Tarboton,D. G., R. L. Bras, and I. Rodriguez-Iturbe,A physicalbasis

    for drainagedensity,Geomorphol., , 59-76, 1992.

    Tucker, G. E., Modeling the large-scalenteractionof climate, ecton-

    ics, and topography,Tech.Rep. 96-003, Earth Syst.Sci. Cent., Pa.

    State Univ., UniversityPark, 1996.

    Tucker, G. E., and R. L. Bras, The role of rainfall variability n drain-

    age basin evolution: mplicationsof a stochasticmodel (abstract)

    Eos Trans.A GU, 78(46), Fall Meet. Suppl.,F283.

    Tucker, G. E., and R. L. Slingerland,Erosional dynamics, lexural

    isostasy,nd ong-lived scarpments: numericalmodelingstudy, .

    Geophys. es., 99, 12,229-12,243, 1994.

    Tucker, G. E., and R. L. Slingerland,Predictingsediment lux from

    fold and thrust belts, Basin Res., 8, 329-349, 1996.

    Tucker, G. E., and R. L. Slingerland,Drainage basin response o

    climate change,WaterResour.Res.,33, 2031-2047, 1997.

    Willgoose,G. R., A statistic or testing he elevationcharacteristics f

    landscapesimulationmodels,J. Geophys.Res., 99, 13,987-13,996,

    1994.

    Willgoose,G. R., R. L. Bras,and . Rodriguez-Iturbe,A model of river

    basin evolution, Eos Trans. AGU, 71, 1806-1807, 1990.

    Willgoose, G. R., R. L. Bras, and I. Rodriguez-Iturbe,A physically

    based coupled network growth and hillslope evolution model, 1,

    Theory, WaterResour.Res.,27, 1671-1684, 1991.

    Yalin, M. S., River Mechanics, 19 pp., Pergamon,Tarrytown,N.Y.,

    1992.

    R. L. Brasand G. E. Tucker correspondinguthor),Departmentof

    Civil and EnvironmentalEngineering,Room 48-108, Massachusetts

    Institute of Technology,Cambridge,MA 02139. (e-mail: gtucker@

    mit.edu)

    (ReceivedOctober27, 1997;revisedApril 22, 1998;

    acceptedMay 1, 1998.)