9
WATER RESOURCES RESEARCH, VOL. 30, NO. 12, PAGES 3531-3539, DECEMBER 1994 Self-organized river basin landscapes: Fractal and multifractal characteristics Ignacio Rodriguez-Iturbe,1 Marco Marani,2 Riccardo Rigon,3 and Andrea Rinaldo2 Abstract. In recent years a new theory of the evolution of drainage networks and associated landscapes has emerged, mainly in connection with the development of fractal geometry and of self-organized criticality (SOC) concepts. This theory has much improved our understanding of the mechanisms which determine the structure of natural landscapes and their dynamics of evolution. In the first part of this paper the main ideas in the theory of landscape self-organization are outlined, and some remarkable features of the resulting structures are presented. In the second part we apply theoretical tools developed in the context of multifractal fields to the study of the scaling properties of the field of elevations of SOC landscapes. We observe that such landscapes appear to be more complex than simple self-affine fractals, although in some cases a simple fractal framework may be adequate for their description. We also show that multiple scaling may emerge as a result of heterogeneity in the field properties reflecting climate and geology. 1. Introduction The global frameworks of fractal analysis [Mandelbrot, 1983] of fluvial structures [La Barbera and Rosso, 1987, 1989; Tarboton et al., 1988; Rodriguez-Iturbe et al., 1992a, b, c; Rinaldo et al.f 1992; Turcotte, 1992; Chase, 1992; Die tier and Zhang, 1992; Ouchi and Matsushita, 1992; Lavallee et al, 1993; Rinaldo et al., 1993; Ijjasz-Vasquez et al., 1993; Rigon et al., 1993], and of self-organized criticality [Bak et al., 1987, 1988; Bak and Creutz, 1993] have been the foundations of a new theory about the evolution of drainage networks and their associated landscapes. It has been shown [Rodriguez-Iturbe et al., 1992a; Rinaldo et al., 1992] that optimal channel networks (OCNs) obtained by minimizing the local and global rates of energy expenditure evolve automatically from arbitrary initial conditions to network configurations exhibiting fractal and multifractal character istics indistinguishable from those found in nature. Although denditric structures obtained through optimality conditions had been introduced before [Howard, 1971, 1990], we be lieve this was the first time fractality was observed in optimal structures. Sun et al. [1994a] have studied the original model of Rodriguez-Iturbe et al. [1992a] and Rinaldo et al. [1992], corroborating the fractal aspects of the planar structure of networks developed under this framework. Chance and the rules of optimum energy expenditure are 'Department of Civil Engineering, Texas A&M University, Col lege Station. 2Istituto di Idraulica "G. Poleni," Universita di Padova, Padua, Italy. 3Dipartimento di Ingegneria Civile e Ambientale, Universita di Trento, Trent, Italy. Copyright 1994 by the American Geophysical Union. Paper number 94WR01493. 0043-1397/94/94WR-01493505.00 the elements which control the evolution and final structure of the drainage network. Although the particular outcome of an individual process is given by randomness, the interplay of such elements in the extended and dissipative system which is the drainage network results in the same fundamen tal principles of evolution. The fractal characters of the systems obtained from these principles are found not to be dependent on the initial conditions, thus suggesting that natural fractal structures like river networks may arise as a joint consequence of optimality and randomness [Rodriguez- Iturbe et al., 1992c]. An important step forward was taken by Rinaldo et al. [1993], who observed that OCNs are a particular case of self-organized critical structures. The theory of self- organized criUcality [Bak et al, 1987, 1988] provides in fact a broad framework for the dynamics of fractal growth and an appealing scheme for the evolution of drainage networks [Rinaldo et al., 1993] which, unlike the theory of OCNs, does not explicitly state a global optimality principle as the guiding evolution concept. Again, the characteristics of the state toward which the system evolves are independent of the initial conditions and are very similar to those observed in nature. In the mentioned scheme [Rinaldo et al., 1993; Rigon et al., 1994], landscapes evolve through the interplay of fluvial (erosive) and hiilslope (diffusive) processes. The erosion process acts on the system, maintaining it in a critical state (i.e., a state in which the shear stress due to water flow cannot exceed a critical value), while a special kind of diffusive process is considered as preserving the fractal nature of the elevation field. We believe that this use of the theory of self-organized criticality provides a clear link between the fractal landforms observed in nature and the dynamics responsible for their growth. The quantitative characterization of self-organized critical (SOC) landscapes and their comparison with natural terrain properties (including power laws in the probability distribu tion of cumulated areas [Rodriguez-Iturbe et al., 1992b] and of stream lengths [Tarboton et al., 1988; Rodriguez-Iturbe et 3531

Self-organized river basin landscapes: Fractal and ...pdodds/files/papers/others/1994/rodriguez-iturbe199… · Self-organized river basin landscapes: Fractal and multifractal characteristics

  • Upload
    others

  • View
    2

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Self-organized river basin landscapes: Fractal and ...pdodds/files/papers/others/1994/rodriguez-iturbe199… · Self-organized river basin landscapes: Fractal and multifractal characteristics

WATER RESOURCES RESEARCH, VOL. 30, NO. 12, PAGES 3531-3539, DECEMBER 1994

Self-organized river basin landscapes:Fractal and multifractal characteristicsIgnacio Rodriguez-Iturbe,1 Marco Marani,2 Riccardo Rigon,3and Andrea Rinaldo2

Abstract. In recent years a new theory of the evolution of drainage networks andassociated landscapes has emerged, mainly in connection with the development offractal geometry and of self-organized criticality (SOC) concepts. This theory has muchimproved our understanding of the mechanisms which determine the structure ofnatural landscapes and their dynamics of evolution. In the first part of this paper themain ideas in the theory of landscape self-organization are outlined, and someremarkable features of the resulting structures are presented. In the second part weapply theoretical tools developed in the context of multifractal fields to the study of thescaling properties of the field of elevations of SOC landscapes. We observe that suchlandscapes appear to be more complex than simple self-affine fractals, although in somecases a simple fractal framework may be adequate for their description. We also showthat multiple scaling may emerge as a result of heterogeneity in the field propertiesreflecting climate and geology.

1. Introduct ion

The global frameworks of fractal analysis [Mandelbrot,1983] of fluvial structures [La Barbera and Rosso, 1987,1989; Tarboton et al., 1988; Rodriguez-Iturbe et al., 1992a,b, c; Rinaldo et al.f 1992; Turcotte, 1992; Chase, 1992;Die tier and Zhang, 1992; Ouchi and Matsushita, 1992;Lavallee et al, 1993; Rinaldo et al., 1993; Ijjasz-Vasquez etal., 1993; Rigon et al., 1993], and of self-organized criticality[Bak et al., 1987, 1988; Bak and Creutz, 1993] have been thefoundations of a new theory about the evolution of drainagenetworks and their associated landscapes. It has been shown[Rodriguez-Iturbe et al., 1992a; Rinaldo et al., 1992] thatoptimal channel networks (OCNs) obtained by minimizingthe local and global rates of energy expenditure evolveautomatically from arbitrary initial conditions to networkconfigurations exhibiting fractal and multifractal characteristics indistinguishable from those found in nature. Althoughdenditric structures obtained through optimality conditionshad been introduced before [Howard, 1971, 1990], we believe this was the first time fractality was observed in optimalstructures. Sun et al. [1994a] have studied the original modelof Rodriguez-Iturbe et al. [1992a] and Rinaldo et al. [1992],corroborating the fractal aspects of the planar structure ofnetworks developed under this framework.

Chance and the rules of optimum energy expenditure are

'Department of Civil Engineering, Texas A&M University, College Station.

2Istituto di Idraulica "G. Poleni," Universita di Padova, Padua,Italy.

3Dipartimento di Ingegneria Civile e Ambientale, Universita diTrento, Trent, Italy.Copyright 1994 by the American Geophysical Union.Paper number 94WR01493.0043-1397/94/94WR-01493505.00

the elements which control the evolution and final structureof the drainage network. Although the particular outcome ofan individual process is given by randomness, the interplayof such elements in the extended and dissipative systemwhich is the drainage network results in the same fundamental principles of evolution. The fractal characters of thesystems obtained from these principles are found not to bedependent on the initial conditions, thus suggesting thatnatural fractal structures like river networks may arise as ajoint consequence of optimality and randomness [Rodriguez-Iturbe et al., 1992c].

An important step forward was taken by Rinaldo et al.[1993], who observed that OCNs are a particular case ofself-organized critical structures. The theory of self-organized criUcality [Bak et al, 1987, 1988] provides in facta broad framework for the dynamics of fractal growth and anappealing scheme for the evolution of drainage networks[Rinaldo et al., 1993] which, unlike the theory of OCNs,does not explicitly state a global optimality principle as theguiding evolution concept. Again, the characteristics of thestate toward which the system evolves are independent ofthe initial conditions and are very similar to those observedin nature. In the mentioned scheme [Rinaldo et al., 1993;Rigon et al., 1994], landscapes evolve through the interplayof fluvial (erosive) and hiilslope (diffusive) processes. Theerosion process acts on the system, maintaining it in acritical state (i.e., a state in which the shear stress due towater flow cannot exceed a critical value), while a specialkind of diffusive process is considered as preserving thefractal nature of the elevation field. We believe that this useof the theory of self-organized criticality provides a clear linkbetween the fractal landforms observed in nature and thedynamics responsible for their growth.

The quantitative characterization of self-organized critical(SOC) landscapes and their comparison with natural terrainproperties (including power laws in the probability distribution of cumulated areas [Rodriguez-Iturbe et al., 1992b] andof stream lengths [Tarboton et al., 1988; Rodriguez-Iturbe et

3531

Page 2: Self-organized river basin landscapes: Fractal and ...pdodds/files/papers/others/1994/rodriguez-iturbe199… · Self-organized river basin landscapes: Fractal and multifractal characteristics

3532 RODRIGUEZ-ITURBE ET AL.: SELF-ORGANIZED RIVER BASIN LANDSCAPES

al., 1992b], multifractal spectra of new growth site measures, i.e., width functions [Rinaldo et al., 1992, 1993], andscaling relationship of slopes with cumulated area [Tarbotonet al., 1989; Rigon et al., 1993]) was the subject of previouspapers [Rinaldo et al., 1993; Rigon et al., 1994]. This paperdescribes a systematic analysis of the fractal properties ofthe resulting topographies and a comparison with measuresof the roughness of natural terrains.

2 . A Conceptua l Framework forLandscape Self-Organization

We will now briefly recall a general scheme for theevolution of drainage networks in which the structure of thefield of elevations organizes itself, evolving through a sequence of critical states [Rinaldo et al., 1993] under theaction of fluvial and hillslope processes.

Fluvial erosion may be dynamically described by anequation like dz/dt = af(r - rc), with/(:c) = 0 for x < 0(where t is the local shear stress, tc is the critical thresholdfor erosion, and a provides the timescale for the erosionprocess). The shear stress t is proportional to yVz, where yis the flow depth and Vz is the local slope. From localoptimality principles and empirical evidence we have that y<* Q05 [Rodriguez-Iturbe et al., 1992a]. Thus at the /th site,Ti a Qi v*Zi and whenever t,- > rc erosion exists. The localoptimality principle also leads to the scaling of slopes withdischarge as Vz,- = Qi~0'5, or, using drainage area as asurrogate of discharge, as Vz,- ~ A/~ .

Rinaldo et al. [1993] studied the dynamic evolution resulting from local exceedance of an erosion threshold. Briefly,the evolution rules are as follows. Each site / in a two-dimensional lattice has two associated variables, the elevation z; and the total flow Q{ surrogated by the draining areaAi. An arbitrary initial planimetric network is obtained bychoosing the elevations of the landscape randomly or in asystematic way. For each site the shear stress is thencomputed as r,- « A°-5Az,-, Az being the drop along thedrainage direction fixed by the steepest descent (since latticelengths are equal to one, Az is equal to the slope). Thecomputed t,- are compared with a given threshold value tc,and the exceedances are identified throughout the lattice.The maximum exceedance is placed, say, at they'th site, andthe elevation of this site is reduced to the value which yieldst = tc. Drainage directions are then recomputed because ofthe modified elevation at the site./. The process is repeateduntil the system evolves to a critical state in which there areno exceedances of t above tc. This critical state is thenperturbed at random by adding elevation to a node. Theperturbation may lead to a readjustment of the structure, andthis is repeated until further perturbations do not inducevariations in the configuration of the system.

It is important to note that the total energy dissipation ofthe system decreases during the evolution and stabilizes atvalues very close to those obtained in OCNs, whose dynamics are controlled by the explicit minimization of the totalrate of energy expenditure. An explanation of the equivalence of OCNs and self-organized critical networks (SOCs)relies on the equivalence of minimizing the total rate ofenergy expenditure and the total potential energy of thesystem [Ijjasz-Vasquez et al., 1993]. In fact, the evolution ofself-organized networks leads the system to the state ofminimum possible potential energy compatible with the

restriction that slopes scale with area as Vz,- = A/ . Therole of randomness is crucial as the perturbations are thedriving mechanism that maintains the evolution of the system.

As described above, the evolutionary strategy from anyinitial condition used by Rinaldo et al. [1993] is one in whichthe maximum exceedance of the threshold shear stress isidentified and the elevation of the site is reduced so that thelocal shear is at the critical value. Two different evolutionarystrategies have also been implemented. The first one is arandom strategy in which a site is chosen at random amongall those t exceeding rc and the elevation is then reduced tomake t = tc. The dynamics then continues as in the originalstrategy described before. The second strategy is a parallelprocessing one in which all sites with r > tc have theirelevations simultaneously updated to make r = rc. Although conceptually different, the three evolutionary strategies produce self-organized networks with identical fractaland multifractal characteristics. Furthermore, in all casesthe total energy expenditure is very close to that obtained forOCNs in similar domains [Rinaldo et al., 1993].

Threshold dependent (TD) processes as in the case ofcritical shear stress are most important in the carving andevolution of river channels. Nevertheless, often thresholdindependent (TI) processes act on the formation of thelandscape of a river basin, especially at the hillslope scale.The lowering of elevations throughout the basin may in thiscase be described by a field equation like

dz— =a/(r-Tc) + /3F(Vz).at

0)

The first term of the right-hand side describes TD fluvialerosion which is zero when t < tc. The second term modelsthe TI processes which are essentially related to the gradientof the elevation field, Vz. The TI processes act everywhere,but since a » B [Rigon et al., 1994] the first term plays thecentral role whenever erosion forces are acting.

Rigon et al. [1994] present a scheme for landscape self-organization which accounts for the presence of TI processes and is an extension of the one described in theprevious section. After the readjustments following a number of perturbations leading to a SOC landscape a diffusioncomponent (TI) is introduced in the scheme. The site i withthe largest drop to one of its neighbors/, Azy, is identifiedand its elevation z, decreased by a quantity proportional toAz,y. The mass removed from z, is redistributed among theneighbors j with elevation zt lower than z,-. Mass redistribution is done according to

Zj ~ Zj

kSnn

0 < v < \ , (2)

where k G nn indicates that the sum is over the nearestneighbors lower than /. The points draining into i changetheir Vz while the points./ change both z and Vz, and thus awhole set of adjustments occurs following a TI activity atany point. The above procedure produces a lowering of theslopes which results in smoother hillslopes and gentlerlandscapes when the iterations continue in time. Indeed,without the TI processes the hillslopes of the resultinglandscapes look unrealistically rough or at least highly

Page 3: Self-organized river basin landscapes: Fractal and ...pdodds/files/papers/others/1994/rodriguez-iturbe199… · Self-organized river basin landscapes: Fractal and multifractal characteristics

RODRIGUEZ-ITURBE ET AL.: SELF-ORGANIZED RIVER BASIN LANDSCAPES 3533

Plate 1. Self-organized landscape resulting from both TD and TI processes [after Rigon et al., 1994].

uncommonly so. One of the relevant effects of TI processesis the lowering of the fractal dimension of the landscape[Rigon et al., 1994]. An example of the effect of repeatedcycles of TD and TI processes is shown in Plate 1. Both vand the length of the iterative process are important, andthrough their effects a large variety of landscapes can beobtained.

3. Fractal and Multifractal Descriptorsof Landscapes

Let us consider the asymptotic field of elevations z(x, t =«>) = z(x) produced by the above processes, defined at everypoint x in a subset of U2. If z(x) has the property

dz(x, + Ae) - z(x,) = Aw[z(x2 + e) - z(x2)], (3)

where e is a unit vector, A is a separation distance, X|, x2 aretwo arbitrary points, and the d above the equals signindicates equality in distribution (the exponent H is calledthe Hurst exponent [Mandelbrot, 1983; Feder, 1988]), it issaid to be simple scaling [Lavallee et al., 1993]. The unique

fractal dimension D characterizing the simple scaling field isrelated to Hurst exponent H via the relationship D = 3 - H[Feder, 1988]. It is important to note that a simple scalingfield is postulated to have constant statistical characteristicswith respect to position and orientation. Equation (3) holds,in fact, for any choice of x12 and e.

Many examples of fractal analyses of topography andbathymetry measured along a linear track can be found in theliterature [e.g., Malinverno, 1989; Turcotte, 1992] documenting the fractal nature of natural landscapes. Most of thesestudies are based on monofractal (or, simple scaling) modelsthat do not seem entirely consistent with the properties ofmeasured field data. Recent contributions [Schertzer andLovejoy, 1989; Lavallee et al., 1993] give a different interpretation of the fractal characters observed in real topographies, arguing that geographical fields are generally multi-fractal and that inconsistencies are inevitable when theseentities are forced into narrower geometric frameworksinvolving a single fractal dimension. In view of these considerations the characteristics of the SOC landscapes introduced in the previous sections will be analyzed with the

Page 4: Self-organized river basin landscapes: Fractal and ...pdodds/files/papers/others/1994/rodriguez-iturbe199… · Self-organized river basin landscapes: Fractal and multifractal characteristics

3534 RODRIGUEZ-ITURBE ET AL.: SELF-ORGANIZED RIVER BASIN LANDSCAPES

* ■ ■ ' ' ' " " ' ' ' ' ' " ' ^ ^ ^ ^

120 ■ j y & $ - v - - ; N

100 • m80 H p .60 _ . , . " , •

40 g*sn*?> . .

2 0

ScIp -0

20 4 0 60 80 100 120

Figure 1. Plot of the set of points 5 that belong to predetermined ranges of elevations. Isolines are identified atdifferent heights for an elevation field resulting from TDprocesses only. The solid area refers to the set 5s2, i.e., theset of points whose elevation is lower than 2 (z(x) ^ 2).Different levels of shading indicate sets S where 2 < z(x) <3 and so on. /

purpose of linking the dynamics of fractal growth throughSOC processes and the resulting fractal or multifractalstructures.

There are many ways to characterize rough surfaces. Oneway to study the fractal properties of a field is to define,through the introduction of a threshold value of elevation T,sets that have theoretically determined characteristics withrespect to fractal geometry. One such set is the exceedanceset S>.7, introduced as the set of all the points x satisfyingz(x) > T. Likewise, one can define the perimeter set PT asthe subset of the points of 5^T having at least one nearestneighbor belonging to the set S<T. In general, since PT is asubset of Ss.T, then D(PT) < D(S>T). In the case of asimple scaling structure the sets S>r are characterized by asingle fractal dimension D(5>r) = D while D(PT) = 2 - H,both values being independent of T [Lavallee et al., 1993].These properties provide a means of checking whethersimple scaling arises for z(x).

Figure 1 shows the sets of points which belong to differentranges in elevation, from which we have obtained the abovesets 5 defining the elevation field z(x, ta) for a SOClandscape obtained when only TD actions are operating, i.e.,whose field equation is

dz— = af(r- tc).a t

(4)

The initial condition z(x, 0) was obtained by Eden [1961]growth [Rinaldo et al., 1993]. Here rc = 1 on a 128 x 128lattice. Elsewhere [Rigon et al., 1994], we have shown thatthe statistical features of the landscape are independent ofthe threshold value tc. Figure 2 shows contour lines of the

120

100

Figure 2. As in Figure 1, for the elevation field in Plate 1,where both TD and TI processes were allowed to occur.

landscape z(x, tB = °°) shown in Plate 1, obtained by a cycleof TI processes defined by the field equation

dz— =/3F(Vz),at

(5)

acting from the initial condition z(x, ta). The resulting fieldis described in Figure 1.

Box counting has been applied to the graphs in Figures 1and 2 to determine the fractal dimension of each contourline. For each contour, box-counting [Mandelbrot, 1983;Feder, 1988] data are represented reasonably well (Figure 3)by a straight line whose slope is the fractal dimension D(PT)in a log-log plot of the number of boxes N(S) that cover thewhole contour versus the box size 5. This indicates that anycontour of a self-organized landscape may be considered aself-similar fractal. The main results of the analyses areillustrated in Table 1. We note that the fractal dimensions ofthe isolevel lines obtained for different contour elevations

o

6.5

6

5 i

5

4.5 DX

4

3.5

3

1 \

- 4 . 5 - 3 . 5 - 3

log 6-2.5

Figure 3. Typical dependence of A/(5) on 5 found in thebox-counting procedure applied to one of the isolines ofFigure 1 (TD processes only).

Page 5: Self-organized river basin landscapes: Fractal and ...pdodds/files/papers/others/1994/rodriguez-iturbe199… · Self-organized river basin landscapes: Fractal and multifractal characteristics

Table 1. Elevations, Fractal Dimensions (Box Counting)of the Border Sets, and Correlation Coefficients of Box-Counting Scaling for TD and TI SOC LandscapesContour TD Actions TI Actions

z(x) = T D(PT)TD r D(PT)Tl

1 1.38 0.99 1 . 3 0 0 . 9 91.5 1.42 0.96 1 . 3 0 0 . 9 82 1.43 0.98 1 . 3 0 0 . 9 82.5 1.43 0.94 1 . 2 9 0 . 9 73 1.39 0.96 1 . 2 6 0 . 9 73.5 1.37 0.99 1 . 2 2 0 . 9 64 1.35 0.92 1 . 2 1 0 . 9 4

are slightly different. Similar results were observed for theoptimal topographies obtained by Sun et al. [1994b] byenforcing an artificial perfect scaling of slopes to an optimalchannel network [Rodriguez-Iturbe et al., 1992b, c; Rinaldoet al., 1992]. The contours near the average height have verysimilar fractal dimensions, and, overall, the surfaces produced seem to show the signs of a relatively complexself-affine fractal. However, in the above analysis, samplesize effects are important both at the higher and at the lowerelevations, thus partially impairing the related results. In therange of elevations where the sample size is constant weobserve a consistent behavior analogous to the topographiesobtained by Sun et al. [1994b]. Moreover, we will show laterthat other analyses suggest that multiple scaling does notoccur in the topographic field obtained via SOC procedureswith space-constant field properties. Fast Fourier transformanalyses [e.g., Cochran et al., 1967; Feder, 1988; Malin-verno, 1989], roughening measurements [e.g., Mandelbrot,1983; Feder, 1988; Matsushita and Ouchi, 1989; Dubuc etal., 1989a, b] and height-height correlation measurements[e.g., Mandelbrot, 1983; Feder, 1988; Sun et al., 1994a, b]also agree with the results shown by the contours near theaverage elevation.

Figure 4 shows the log-log plot of the power spectrum(S(k)) (ensemble-)averaged over all south-north (S-N) andeast-west (E-W) topographic profiles obtained from verticalcuts (or transects) through a SOC landscape produced by TDactions (equation (4)). The data can be fitted by a straight line

V0.1

0.001

0.01

Figure 4. Power spectrum averaged over all S-N and E-Wtransects of the TD 128 x 128 SOC landscape considered.The slope in the log-log plot is -B = -1.6, yielding D - 1.7for the (mean) fractal dimension of the transects.

2 y

> Variation Q\ CO

.s i / 2-D=0.5o-1 y / 1

1.4

0 0 . 2 0 . 4 0 . 6 0 . 8 1 1 . 2

Log box size

Figure 5. Results of the variation method analysis appliedto a transect of the TD 128 x 128 SOC landscape considered.Averaging over all transects gives D = 1.65 for the meanfractal dimension [after Rigon et al., 1994].

(i.e., S(k) « k~p) with slope -B(B = 1.6) giving, as shownlater, D = (5 - B)I2 = 1.7 ± 0.1 [Feder, 1988].

Figure 5 shows the results of the variation method analysis[Dubuc et al., 1989a, b] on a single vertical cut of the surfaceof a TD 128 x 128 SOC landscape with rc = 1. The rangevariation of the height z within a box of size L scales asproportional to LH (D = 2 - H) if the object is simplescaling. Once ensemble averaging is performed, the resultingfractal dimension is D = 1.65 ± 0.05.

The analysis of mean deviation, first introduced for self-affine fractals by Matsushita and Ouchi [1989], studies thedeviation a2(L) as a function of the horizontal length scale Las

°-2(LH^2 2 (^-(zhr (6)i j = \

where tq is the elevation at the lattice sites x = (/, j) and{z)l is the average height for the surface within a grid of sizeLxL For a self-affine fractal, a is related to L by a powerlaw a oc LH. Figure 6 shows the log dependence of o- on Lfrom which we derive D - 1.66 ± 0.05. We have repeatedall the above analyses for the TD + TI processes describedin (1) leading to the landscape in Plate 1 and have obtained

1 •-0.25

S-

-0.3 .<Y> S .I-0"35COa_J -0.4

• S >.-< i2-D=0.34i

-0.45

. ^ "1.1 1.2 1.3 1.4

LogN

1.5 1.6 1.7

Figure 6. Dependence of mean deviation o\L) on the horizontal length scale L obtained applying the Matsushita andOuchi [1989] analysis to the TD 128 x 128 SOC landscapeconsidered. This procedure gives the value D = 1.66 for themean fractal dimension [after Rigon et al., 1994].

Page 6: Self-organized river basin landscapes: Fractal and ...pdodds/files/papers/others/1994/rodriguez-iturbe199… · Self-organized river basin landscapes: Fractal and multifractal characteristics

3536 RODRIGUEZ-ITURBE ET AL.: SELF-ORGANIZED RIVER BASIN LANDSCAPES

1.7 dzat=- «/(r - rc) r^^.ryT^^

1.6 , V//vwAq 1.5

1.4

AAA V % = «/(r - re) + 0F(Q, Vz)

1.3

0 1 0 2 0 3 0 4 0 5 0 6 0E-W transect number

Figure 7. Values of the fractal dimension for different E-Wtransects of TD and TD + TI topographies.

consistent results for the mean fractal dimensions. Nevertheless, we observed a nonnegligible range in the individualfractal dimensions of the vertical cuts of the topography andin the contour lines at different elevations. Figure 7 shows aplot of the fractal dimensions for different E-W one-dimensional tracks of the topographies of TD and TD + TItopographies. One notices that the self-affine characteristicsof the evolving surface rendered by the TI type of self-organization are less rough. The above discussion suggestedthat the complexity of the self-affine fractal surface neededfurther analysis.

There are other ways to determine the scaling nature of afield, namely tho£e related to the definition of generalizedvariograms [Lavallee et al., 1993], which are essentially ageneralization of height-height correlation analyses. The qthmoment of a range, or generalized variogram, Ce/(X), isdefined as

C,(A) = <|z(x)-z(x + r)|%,_A, (7)where the angle brackets are the ensemble averaging operator and A is the separation distance. The qth moment will besaid to be scaling if the following relation holds:

CJ\) = \Ki«)CJ\)o:\KW. (8)It is worth noting that the definition of C^A) given aboveimplies that the random field considered is stationary, sincethe qth moment is supposed to depend only on the separation distance, or lag, A. The implications of this hypothesiswill be considered in greater detail later on in this paper.

In the case of simple scaling the exponent K(q) is linear,i.e.,

K{q) = qH, (9)as can be shown by taking qth powers of the moduli of bothsides of (3) and taking ensemble averages. In multifractals amore general form for K(q) has to be assumed: We will referto this case as multiple scaling.

Restricting ourselves, without loss of generality, to thecase of elevation values sampled along a one-dimensionaltrack, we will consider the relation between the form of thepower spectrum obtained from the Fourier transform of thedata and the exponent K(q) governing the scaling of themoments. It is known that for a series whose graph isself-affine the power spectrum has the form of a power law,

i.e., E(k) « k~P, in a suitably chosen interval for thefrequency k. The Wiener-Khinchine theorem states that thepower spectrum is the Fourier transform of the autocorrelation function R(\) = (z(x + A)z(*)) (a function of the lagonly, for stationary random processes), which also needs bepower law as R(\) « A^_1 [Falconer, 1985] for a self-affinegraph z(x). The autocorrelation function can, in turn, berelated to the second order moment C2(A) via

<|zU + A) - z(x)\2) = 2(z(x)2) - 2/?(A), (10)

giving B = 1 + K(2). In the case of simple scaling £(2) =2H and B = 1 + 2H. Since the fractal dimension of the"profile" of z(x) is D = 2 - H, we have D = (5 - B)/2.These relations are commonly used in evaluating the fractaldimension of a two-dimensional profile and can be useful indetermining the fractal or multifractal nature of such aprofile. Lavallee et al. [1993] have contended, in fact, thatsince for multifractals K(2) ¥= 2K(X) = 2H, the aboverelation between D and B will not hold in virtually allapplications to real geophysical profiles. In fact, to justifyD = 2 - H we only need C,(A) = A^C^l), while D =(5 - B)I2 also requires that C2(A) = A2WC2(1).

We have applied the analysis of moments to the SOClandscapes introduced earlier, computing qth moments fromaverages along both S-N and E-W transects. This amounts toconsidering the field as isotropic and transects as realizationsof a (stationary and ergodic) random process. Statisticalparameters of the process itself can thus be computed onthese realizations. The scaling properties of the momentsobtained from the usual 128 x 128 TD landscape are shownin Figure 8. The related K(q) versus q graph is shown inFigure 9. It can be seen that the moments follow a scalingbehavior remarkably well and that an almost perfect simplescaling behavior is exhibited by the K(q) versus q curves.The exponent that characterizes the simple scaling behaviorwas found to be H = 0.303 ± 0.001 in the case of TDprocesses, and H = 0.368 ± 0.004 forTD + TI processes,showing that TI processes are responsible for a lowering ofthe fractal dimension of landscapes (recall that D = 2 - H)[Rigon et al., 1994]. These facts suggest that transects fromSOC elevation fields obtained through TD and TI actions canbe considered as realizations of a stationary random processand that, as such, their study is relevant to the fractalcharacterization of the fields themselves. The results ob-

A&•I T

I

+

Figure 8. Scaling of ^th moments in a SO elevation field onwhich only TD processes were allowed to occur. The lowestcorrelation coefficient was found to be R = 0.99.

Page 7: Self-organized river basin landscapes: Fractal and ...pdodds/files/papers/others/1994/rodriguez-iturbe199… · Self-organized river basin landscapes: Fractal and multifractal characteristics

90U3S3Jd sin SuuBniBA3 ui psyaxs aq sjojsjaqj pinoqs 3Jbo•paAoiduia uaaq pBq (rBpouiiq "8-a) ppg pioqsajqj Ajbuou-bjsuou b uaqM Apo paunooo X\JB9\0 Suiposripw psuijoj-jad sasApuB aqj jo sjiuiii AjiAUisuas aqj uiquM isbo] je'sajnpiujs Suireos ajduiis uns jnq jaqSnoj sajBaio (6 ajn8i j)PP9 pioqsaiqi 'Ajbuoipp Aubdusubp 'apBUBA ArreijBds y

•(8 QinSivj) sjinsaj spouiojdui jo AjubuojpjsreqojS b puB 'ainpiuis ^jo/apu aqi Aq paonpui Aujbuou-bjsuou aqj joajsp jou ssop AipmjBU jBqj (sjpmj S"N PueM-3 3uop "3-i) saxB ajBuipjooo Suop sjnpaoojd SuyduiBSaqj si Jl snqi &661 *7» 'a «*&* -€661 "/» l» opnjum -6861•wiw^Al P"» W*0 '.6861 '7» >* »oioqjDjL\ bojb oSbuibjppappumo qjiM sodojs jo Suipos oj anp 'suopoojip A\oyaqj 3uop paidures uaq/w DOJJBAaia jo sjusuisjoui aqj u; All-jbuoijbjsuou S3lldiui ft) ui 3x ppqsaiqj mBJSDOO SuiApapunub qjiM uoijbziub3jo-jps sadBDspirei «pB| ui -paMssqo uoaqSBq Suipos aidijinui puB sarquA ppy aqj jo sjuauiajoui aq)io Ajubuoijbjsuou jo ABidjajui anqns y sodBOspuB[ pjnpujo sssuqSnoj sqj joj paAiosqo oSubj oqj ui 81 sppy Suijmsaj3qJ jo qoB3 3uizuapBJBqo suoisuounp pjobjj sjSuis aqi•suoipsJip sjBUipjooo Suore pojduiBS uaqM sjuouiojoui oqj joAjubuoijbjs poijsijBjs pus Suipos agOnf! Jiqiqxo oj pojsaSSns3jb soiisuapBJBqo snoauaaouioq AipijBds qjiM sassaaojdIT pub ril q3nojqj paurejqo sppy uoijbaop DOS snu.l1 - ( Z ) ) ^ j o i u j o j o q j A q p a n ? a A a i

s. SuirBOsninui pqj Pub (960 = H Supq uoissajSaj -reaunam 10 juaragjaoo uoijBpiJOD uinuiiuiui aqj) -ioiABqaq Suireosavbi J3A\od b qopui sjuauioui pqj uaas aq ubo \\ "ApAipodssj'7i pus n sainSy ui UAvoqs 3JB uoijBpj b snsjoA (b)x sqiPub sjusuioui oqi jo uuoj 3qi -iOUUBiu iBnsn aq) ui pazAiBUBSBM '7 31BU UI UMOqS 'Pp9 UOIJBAOp dflj, Qm J3MO0CO = a* P"B Ol M J3ddn) - 3l 3JB SS3JlS JB3MS IBD!IP°am joj s9n,BA u3soqo aqi Pub 'ppno aqj q3nojqj lBUo3e.pam 3uoib papiAip si uasoqo ppu aqi -ss3J;s JB3qs ptDijuo3qj io uonnquisip lepouiiq Aiq3iq b Aq pazuapmqo aoiwiWxW^°unJ s™ (U) Ac> P3SS9jdx9 ""TO"*!* U0'S0J3qt aqi 'saniauaSojapq i3JB3i3 jo spsjjs aqj Jsai 01

•pajapisuoo aq AuauaSojapq jo aaaSap JaqSiq b jigoBid 351BJ Abiu JoiABqaq SuqBOS ajdiuis ujojj saanyBdop jBqiSuijsaSSns 'sppy sno3uo3ouioq jo osbo aqi ui uBqi psure«BApsop ssai 'pej m 'si b Pub {b)x u33Avpq UOUBpi jwuiiauj. -3lduiis psjspisuoo aq (ApAisnpuoo jou q3noqiiB) jmsAbui sajnpaus 3uireos asoq/A (6 aJnSij ujojj 300'0 + 083'0

•ppy aqj Aq psouapiAa ssiyadojdSuireos-sidninui oqi azyBnsiA oj pajPid uasq sbM ()d30J3)uiOJM qiiM) ;y Jssq jbou., v -pwapisuoo Apni iiioqjjou u. ppyU0UBA9P IBJIUBU 3qJ joj SAJnO ft snsJSA (b)% 'Ol wn3!i

•sasssoojd u + ai snoauoSojapqjo ssBO aqi ui 3000 + 0830 = H P"* 'suoipb u + QlJOJ f000 + 89£*0 = H 'SP3JJ3 Ql Joj Z000 T £0£"0 = Haq oi punoj sba\ uopBpJ SunBOS ajduiis sqj jo juauodxa aqi•juapiAS sjoui JoiABqsq Surreos oiduiis o^bui oi ponojd usaqSBq (jdaojajui OJ3Z qjiM) )y jsoq jbouij y 'sossaooad rj, + Qlsno3U33oj3pq puB 'u + Ol 'Ol ™°*} ApAipodsoj 3ui-linsaj ssdBospuBi 30S J0J saAJno b snsjsA (b)x 6 aJnSij

01 0

K * ) * — ' ) / » -■§

. , ^

9 :

( ^ - - ) / " = |

0

(*A,©Ufl' + (8A—t>/to"

soi

ST

3

t SZ

f

= //) sodBOspuBi Joq3noj AiiuBoyiuSis ui JinsaJ sanpuoSo-japq piiiu,, qons jo spojja oqi 3JnjBU ui poAJOsqo asoq)qopiu 'pqi sonsiws sonpojd oi [^661 "/" ^'7 «*«*»] UMOU>l s?ussoqo ppy aqi 6 ainSij ui UA\oqs sjb (ft66l "l*> '3 uoSty]spxid z oi renba sjbos rej33pi qjiM sjnpaijs uoijbpjjod,Buu'3uodx3 Pub 'Z'Q = $* 'I = W qi!M (II) ui (x)Jx ppy1BUJJOU301 b qjiM psuiBjqo) ajnpaooad 30S W OJ P3«lddBsisAibub aouBUBAOO P3ZiiBJ9Uo3 aqj jo siinsoj sqi [t766l«•» p «/oSiy]"ainpiujs uoijBpiioo b qjiM uoipunj uiopuBJb jo JBqi si jsjy psjapisuoo 3± jo AujiqeuBA reinads aqj puB

(II) ( ( x ) J x - x ) / « = ^3SB0 siqj UI SI

uoipnba ppy aqi -sjapuiBJBd ppy aqi jo uopnqujsip iBijBdsuuojiunuou b qSnojqj AjwuaSojapq ppoui yiM sm 'SimMoqs oi -psAOiqoB uoisojo jo aajSap am qjiM '-aM 'qjdopqjiM osjB jnq aoBjjns aqj uo uouisod qjiA\ Apo pu sauBAAjSuojis Ayadojd siqi 'ajnpu ui 'sb suoijbuijoj iBoiSopaSIBJnpu jo aiPBiu iios oq; jo Ainiqipojo aqj ui A^puoSojapqjo pnpojdAq b oq Abui SuirBOSiunui pqj paSSns 3A\

uoisuaiuip ibpbjj aiSuis b SuiAjpsds Aq paquossp aq jouAbui JeiauaS ui sadBOspuB] jBjnpu jo saiyadojd oupuioaSaqi pqi saqdiiii pbj siqi b Pub (b)x uoaMpq uoiyepjjBauyuou b '-3-i 'SuiiBOSiipui jo sajnpuSis aqi A\oqs paapuipaipms sppy JBunpu aqj jem 'AjSupuiAuoo ojmb 'Moqs(Oi ainSy) suoippduioo jno jo puB [£661 "V> & 9?ll™°l\wniBjajii aqi jo sjinsai aqx -soiydBjaodoi paj jo satyadojdSuipos aqj isai oj ( HI 01 £ x 0S3 ^S pxid) opuoizBM odiSoj-030 oiziajos Aq paonpojd Ajmi ujaqyou jo (niQ) dBUJUIBJJ31 iBjiSip sqj jo uoiyod -Ui5|-003Z B paipms 3ABq 3M

UOISSnDSIQ 'fr

•sjaq psjs-pisuoo sb ssdBospuBi OOS au^s aqi uiojj us^bj uoipAapjo' sauq jnojuoo snoiJBA jo sisAjbub jbpbjj snoiAOJd JnomiM juajsisuoo 3iB sjuauioui qjft jo Apms am qSnojqj paurei

/.£££ sHdVDSQNVi MisvahhawQ3ZINVOHO-J13S -TVia sa^nii-zsnoraao^

Page 8: Self-organized river basin landscapes: Fractal and ...pdodds/files/papers/others/1994/rodriguez-iturbe199… · Self-organized river basin landscapes: Fractal and multifractal characteristics

•ppy aqj Aq psouapiAS ssijjsdojdSuipos-aidijpui sqj aznensiA oj 'oi ajnSij ui sb 'pajjojduaaq SBq (jdaojsjui oiaz qjiM) jy jssq jb3ui| y -3 ajeu joppy snoau38oj9jaq oqj joj SAjno b snsj9A (b)% 'z\ aanSij

'96'0 = X 9q oj punojSBM JU9pyj909 UOIJBpjJOO JS9MOJ 3qj, -J ajB|J ui UA\oqs9dBDspuBi DOS 3M1 JOJ SJU9UIOIU qj£ jo Suireag 'U ajnSij

V001

• 01=b ,

8=b ,

9=b 0

M> .3=b .

. I=b 0

01

•98J9UJ9 oj Suireos gjdijpuj jojpgp99U si sgijjgdojd ppy jo AjpugSojgjgq jeqj sjssSSns siqx•9J9 'gjBUiip 'ASopgS 'ojjubui jios oqj jo ASopqjy pssodxajo ABjdiajui aqj ui AjpuaSoiajaq pjnjBU Suijooyaj 'ssij-jsdojd ppy ojqBUBA AyBijeds jo osbd jBpoijjBd b u; sjnooo

AjipjDBJjijpuj jBqj UA\oqs 3ABq 3^ -sasssoojd snoouoSouioqXneijBds jo suijsj ui paunqdxa aq jouubd ssdBospuBj pjnjBUu; punoj AyBjuouiusdxs joiABqsq Suipos-sjdijpui snq^

"Ajub-uoijbjs ajepjsod jBqj spoj Apo Suisn uaqM Suiposupui jo

•ssaijs JBaqs reopuo aqj jo uounqujsip ppouiiq b uiojj Suijpssj adBospuB] OOS Q± 'Z aPId

09

S3dV3SQNV3 NISV8 H3AIH Q3ZINV0H0-J33S "IV J.3 Saanil-ZSnOIHQOH 8£se

Page 9: Self-organized river basin landscapes: Fractal and ...pdodds/files/papers/others/1994/rodriguez-iturbe199… · Self-organized river basin landscapes: Fractal and multifractal characteristics

RODRIGUEZ-ITURBE ET AL.: SELF-ORGANIZED RIVER BASIN LANDSCAPES3539

ReferencesBak, P., and M. Creutz, Fractals and self-organized criticality, in

Fractals and Disordered Systems, edited by P. Bak, Springer-Verlag, New York, in press, 1993.

Bak, P., C. Tang, and K. Wiesenfeld, Self-organized criticality: Anexplanation of 1//noise, Phys. Rev. Lett., 59, 381-385, 1987.

Bak, P., C. Tang, and K. Wiesenfeld, Self-organized criticality,Phys. Rev. A, 38(1), 364-374, 1988.

Chase, C. G., Fluvial landsculpting and the fractal dimension oftopography, Geomorphology, 5, 39-57, 1992.

Cochran, W. T., J. W. Cooley, D. L. Favin, H. D. Helms, R. A.Kaenel, W. W. Lang, G. C. Maling, D. E. Nelson, C. M. Rader,and P. D. Welch, What is the fast Fourier transform?, IEEETrans. Audio Electroacoust., 15(2), 1967.

Dietler, D., and Y. Zhang, Fractal aspects of the Swiss landscape,Physica A, 191, 213-219, 1992.

Dubuc, B., J. F. Quiniou, C. Roques-Carmes, C. Tricot, and S. W.Zucker, Evaluating the fractal dimension of profiles, Phys. Rev.A, 59(3), 1500-1512, 1989a.

Dubuc, B., S. W. Zucker, W. Tricot, J. F. Quiniou, and D. Wehbi,Evaluating the fractal dimension of surfaces, Proc. R. Soc.London A, 425, 113-127, 1989b.

Eden, M., A two-dimensional growth process, in Fourth BerkeleySymposium on Mathematical Statistics and Probability, vol. IV,Biology and Problems of Health, edited by J. Neyman, pp.223-239, University of California Press, Berkeley, 1961.

Falconer, J., The Geometry of Fractal Sets, Cambridge UniversityPress, New York, 1985.

Feder, J., Fractals, Plenum, New York, 1988.Gupta, V. J., and E. Waymire, Statistical self-similarity in river

networks parameterized by elevation, Water Resour. Res., 25(3),463^76, 1989.

Howard, A. D., A simulation of stream networks by headwatergrowth and branching, Geogr. Ann., 3, 29-50, 1971.

Howard, A. D., Theoretical model of optimal drainage networks.Water Resour. Res., 26(9), 2107-2117. 1990.

Ijjasz-Vasquez, E. J., R. L. Bras, I. Rodriguez-Iturbe, R. Rigon,and A. Rinaldo, Are river basins optimal channel networks?, Adv.Water Resour., 16, 69-79, 1993.

La Barbera, P., and R. Rosso, The fractal geometry of rivernetworks (abstract), Eos Trans. AGU, 68, 1276, 1987.

La Barbera, P., and R. Rosso, On fractal dimension of rivernetworks. Water Resour. Res., 25(4), 735-741, 1989.

Lavallee, D., S. Lovejoy, D. Schertzer, and P. Ladoy, Nonlinearvariability of landscape topography: Multifractal analysis andsimulation, in Fractals in Geography, edited by N. Lam and L.De Cola, pp. 158-192, Prentice-Hall, Englewood Cliffs, N. J.,1993.

Malinverno, A., Testing linear modes of sea-floor topography, PureAppl. Geophys., 131, 139-155, 1989.

Mandelbrot, B. B., The Fractal Geometry of Nature, W. H.Freeman, New York, 1983.

Matsushita, M., and S. Ouchi, On the self-affinity of various curves,Physica D, 38, 246, 1989.

Ouchi, S., and M. Matsushita, Measurement of self-affinity onsurfaces as a trial application of fractal geometry to landformanalysis, Geomorphology, 5, 115-130, 1992.

Rigon, R., A. Rinaldo, I. Rodriguez-Iturbe, E. Ijjasz-Vasquez, andR. L. Bras, Optimal channel networks: A framework for the studyof river basin morphology, Water Resour. Res., 29(6), 1635-1646,1993.

Rigon, R., A. Rinaldo, and I. Rodriguez-Iturbe, On landscapeself-organization, J. Geophys. Res., 99, 11,971-11,993, 1994.

Rinaldo, A., I. Rodriguez-Iturbe, R. Rigon, R. L. Bras, E. Ijjasz-Vasquez, and A. Marani, Minimum energy and fractal structuresof drainage networks, Water Resour. Res., 28, 2183-2195, 1992.

Rinaldo, A., I. Rodriguez-Iturbe, R. Rigon, E. Ijjasz-Vasquez, andR. L. Bras, Self-organized fractal river networks, Phys. Rev.Lett., 70, 822-826, 1993.

Rodriguez-Iturbe, I., A. Rinaldo, R. Rigon, R. L. Bras, and E.Ijjasz-Vasquez, Energy dissipation, runoff production and thethree-dimensional structure of channel networks, Water Resour.Res.,28(4), 1095-1103, 1992a.

Rodriguez-Iturbe, I., E. Ijjasz-Vasquez, R. L. Bras, and D. G.Tarboton, Power law distributions of mass end energy in riverbasins, Water Resour. Res., 28(4), 988-993, 1992b.

Rodriguez-Iturbe, I., A. Rinaldo, R. Rigon, R. L. Bras, and E.Ijjasz-Vasquez, Fractal structures as least energy patterns: Thecase of river networks, Geophys. Res. Lett., 19(9), 889-892,1992c.

Schertzer, D., and S. Lovejoy (Eds.), Scaling, Fractals, andNon-linear Variability in Geophysics, D. Reidel, Norwell, Mass.,1989.

Sun, T., P. Meakin, and T. J0ssang, A minimum energy dissipationmodel for river basin geometry, Phys. Rev. E, 49(6), 4865-4872,1994a.

Sun, T., P. Meakin, and T. J0ssang, The topography of optimaldrainage basins, Phys. Rev. A, in press, 1994b.

Tarboton, D. G., R. L. Bras, and I. Rodriguez-Iturbe, The fractalnature of river networks. Water Resour. Res., 24(8), 1317-1322,1988.

Tarboton, D. G., R. L. Bras, and I. Rodriguez-Iturbe, Scaling andelevation in river networks, Water Resour. Res., 25(9), 2037-2051, 1989.

Turcotte, D. L., Fractals and Chaos in Geology and Geophysics,Cambridge University Press, New York, 1992.

M. Marani and A. Rinaldo, Istituto di Idraulica "G. Poleni,"Universita di Padova, via Loredan 20, 1-35131 Padova, Italy.

R. Rigon, Dipartimento di Ingegneria Civile e Ambientale, Universita di Trento, Mesiano di Povo, 1-38050 Trento, Italy.

I. Rodriguez-Iturbe, Department of Civil Engineering, TexasA&M University, College Station, TX 77843.

(Received November 22, 1993; revised April 25, 1994;accepted June 8, 1994.)