Eolian dunes: Computer simulations and attractor interpretation

B. T. Werner Institute of Geophysics and Planetary Physics, University of California, San Diego, La Jolla, California 92093-0225

ABSTRACT ALGORITHM
A simple computer-simulation algorithm for the transport of In the computer simulation, dunes are built from slabs of sand,
sand by wind produces forms resembling barchan, crescentic ridge, the positions of which are constrained to lie on a square lattice
linear, and star natural dune classes. Sand is moved as slabs com- (Fig. 1). The surface elevation is proportional to the number of sand
posed of many grains that are picked up at random, transported in slabs at a lattice site. The edges of the lattice are connected by
a specified direction, and deposited (1) with a probability that de- periodic boundary conditions, whereby a slab of sand transported
pends on the local presence or absence of sand or (2) in shadow over one boundary of the lattice is brought in at the same position
zones in the lee of dunes. The simulated dune fields are interpreted on the opposite boundary. A maximum difference in surface eleva-
as complex systems, with sand-dune classes being dynamical at- tion between adjacent lattice sites, an angle of repose set to 308, is
tractors of these systems. The evolution of dunes once formed be- enforced. If the deposition of a slab of sand at a particular site would
comes decoupled from the details of eolian sand transport. violate the angle-of-repose criterion, the slab is moved down the
steepest gradient until compliance is achieved. Similarly, if erosion
INTRODUCTION of a sand slab oversteepens the slope, starting at that lattice site,
Many ideas and models have been advanced to explain the neighboring sand slabs are moved downslope one lattice site, suc-
morphology of eolian dunes in terms of the processes governing cessively going up the steepest gradient until no angle exceeds the
their initiation and evolution (reviewed by Cooke et al., 1993). Be- angle of repose.
cause of the long time required for full development, the evolu- For some of the simulations, the initial morphology is gener-
tionary sequence has been documented in only a few cases (Kocurek ated by placing sand slabs, one by one, on the lattice at random
et al., 1992). A time-dependent model relating wind and sediment locations. Alternatively, dunes with a range of configurations and
parameters to dune morphology would be helpful for addressing orientations are constructed as initial conditions.
questions regarding the efficacy of using the morphology of existing Sand slabs are transported individually in a manner that is
dune fields as paleo–climatic indicators and for inverting the eolian meant to approximate the conveyance of sand grains in saltation by
rock record to determine the details of past environmental condi- the wind over sand beds and nonerodible, rough surfaces. A sand
tions. It also would complement an existing model that relates dune slab is chosen for transport (erosion) randomly from all sand slabs
morphology and orientation to stratigraphy (Rubin, 1987). on the surface. The slab is moved a specified number of lattice sites,
Because of the complicated nature of air flow and of sand ero- l, in the transport direction and is deposited at this site with a prob-
sion, transport, and deposition over a dune, a reductionist path from ability that depends upon the number of sand slabs there. The prob-
the physics of eolian sand transport to the evolution of a dune field ability of deposition at a site with no sand slabs, pns, is less than the
is not currently feasible. Most of the models that have been pro- probability of deposition at a site with at least one sand slab, ps. This
posed for eolian sand transport and dune formation and evolution differentiation stems from the presumed greater likelihood of re-
suffer from an empirical nature that inhibits their wide application, bound for saltating grains from a stony surface vs. a sand patch
an inability to lead to testable hypotheses on the dune-field scale, or (Bagnold, 1973). If the slab is not deposited, then it repeatedly is
a lack of generality that limits consideration to highly specialized moved l sites in the transport direction until deposition, following
situations rarely encountered in nature. No fundamental, unifying which another slab is chosen randomly for transport. This procedure
theories have emerged, and no path to an all-encompassing frame- is repeated to construct the time evolution of the surface. Slab
work has been identified. movement not parallel to the transport direction originates only
A robust method for simulating the development and evolution from enforcement of the angle of repose.
of eolian dune fields is presented here. The goal is to introduce the Time t in the simulations is the number of lattice sites that have
technique, demonstrate that it produces forms that qualitatively re- been polled for slab erosion divided by the number of surface lattice
semble natural dunes, and apply a tentative interpretation to its sites in the simulation. Simulated time intervals can be related to
results. The justification for this technique is grounded in two prop- actual time intervals by specifying the physical dimensions of the
erties of systems subject to strongly nonlinear, dissipative processes slab and then scaling time so as to achieve a given mean sand flux
(e.g., Nicolis and Prigogine, 1989). (1) These systems commonly or a given mean dune-migration speed.
evolve to a finite number of steady states, termed attractors, from a
broad range of initial conditions. The existence of attractors often RESULTS
can simplify the dynamical description of the system to treatment of Application of the simple algorithm outlined above leads to the
the approach to and transition between attracting states. (2) Emer- formation of simulated dunes and dune fields with morphology and
gent behavior is self-organizing behavior on macroscopic space or behavior closely resembling those of several natural dune classes.
time scales that is consistent with the microscopic physics, but is both The results of a simulation in which forms similar to barchan dunes
simpler and essentially decoupled from the small scale. Systems develop are illustrated in Figure 2A. Under unidirectional transport
exhibiting these two properties fall into a class of nonlinear systems
termed complex systems, for which complicated behavior (such as
deterministic chaos) results from simple underlying physics. Eolian
sand transport is both nonlinear (e.g., grain flux depends nonlinearly
on wind velocity) and dissipative (e.g., energy loss in grain colli-
sions). If dune-forming systems are complex systems, these two
properties permit considerable simplification in constructing an al-
gorithm for transporting sand and building dunes. Figure 1. Side view illustrating dune-simulation transport algorithm.

Geology; December 1995; v. 23; no. 12; p. 1107–1110; 5 figures. 1107

Figure 2. A: Simulated barchan dune field
on 1000 3 1000 lattice at t 5 500 from initial
random morphology. Transport direction
toward top of map, average of three slabs
per lattice site, slab aspect ratio 5 1/3, l 5
5, ps 5 0.6, pns 5 0.4, shadow zone 15° to
horizontal (same aspect ratio, deposition
probabilities, and shadow zone for all sim-
ulations). Contour interval 5 10 slabs. B:
Simulated barchan-dune calving in same
simulation as A. Contour interval 5 5 slabs.

and starting from random morphology, dunes with a horizontal spa- (Bagnold, 1973; Rubin and Hunter, 1987). Wind regimes that have
tial scale related to l, ps and pns are spawned. The simulated dunes been proposed for linear and star dunes were implemented in the
develop the characteristic shape of a barchan, with horns facing in simulations both to subject this hypothesis to a test (within this
the transport direction, because of the inverse relation between simple framework) and to test the ability of a simple dune-building
dune-migration speed and cross-sectional size. Dune-crest termina- algorithm to generate a wider range of dune shapes. According to
tions (e.g., horns for barchans) move faster than the main body of one hypothesis, linear dunes develop in oblique reversing winds
the dune, pointing in the direction of transport so that the ratio of (Bagnold, 1973). This hypothesis for linear dunes was explored by
surface perimeter exposed to active transport to cross-sectional area applying a transport direction that varies from one side to the other,
is approximately constant and independent of lateral position along with a small mean component. Simulated linear dunes form that are
the dune. aligned along the mean transport direction (Fig. 3B). Star dunes are
The general development of barchans does not depend on the believed to form under complicated wind regimes (Fryberger and
initial placement of the sand, although the precise number, shape, Dean, 1979). Simulated dunes with some of the characteristics of
and positions of incipient dunes are sensitive to this placement. One star dunes, including arms radiating from a central core, develop
exception is a simulation beginning with sand arranged into long when the transport direction is rotated around the points of the
transverse ridges. These ridges do not break up into barchan dunes. compass (Fig. 3C).
The simulated dunes increase in size by merger between two
dunes; a small dune catches up to a large dune, and the two coalesce. INTERPRETATION
Dunes grow also by lateral linking; the horns of laterally adjacent The simulation results reported here are suggestive of a general
dunes of different sizes coalesce as the smaller dune passes the framework for eolian dunes wherein the bulk of the general features
larger. The simulated barchans continually lose sand from their and behavior are describable in terms of very simple, general prop-
horns. Dunes are found along the transport paths leading from the erties of the transport. Specifically, the results are consistent with
horns of other barchans so as to make up for this loss. Occasionally the overall hypothesis of eolian dune fields as complex systems and
a small dune detaches from the horn of a barchan, a behavior that with two particular hypotheses: (1) eolian dune classes (barchan,
has been termed calving (Fig. 2B). These processes for dune de- crescentic, linear, etc.) are attractors of the eolian sand transport
velopment and spawning and the alignment and ordering of a dune and morphology system and (2) the evolution of dunes can be de-
field observed in the simulations have counterparts in natural dune scribed as emergent behavior.
fields (Bagnold, 1973; Kocurek et al., 1992). With sufficiently high An attractor is a subset of the phase space (space of variables
sand supply, simulated dunes resembling crescentic ridges form describing the state of a system) available to a system to which it
transverse to the transport direction, evolving from initial barchans contracts, owing to dissipation, from a broader region of phase
via lateral linking (Fig. 3A). space, its basin of attraction. Therefore, if simulated eolian dune
The development of transverse dunes (barchans or crescentic classes are attractors, it should be possible to observe the evolution
ridges) is not sensitive to the parameters employed in the simula- of the system to a particular dune class from a range of initial con-
tions. Varying the transport length and the deposition probabilities ditions. To characterize the simulated dune-forming system, two
primarily alters their initial size and migration speed. Simulated phase-space variables were chosen: dune orientation angle relative
barchan dunes can form without lee shadowing, and transverse to the mean transport direction and the number of dune-crest ter-
dunes can be generated with ps 5 pns. Changes to the algorithm— minations. Orientation is one variable that clearly distinguishes dif-
including normally distributed transport perpendicular to the mean ferent dune classes. Number of terminations also distinguishes dune
transport direction and erosion or deposition probabilities that de- classes (e.g., crescentic vs. barchan) and can be a measure of the
crease or increase with increasing elevation or surface tilt— do not maturity of the dune pattern (e.g., the number of terminations de-
alter the general features of simulated dune formation. These creases dramatically as simulated linear dunes form).
changes appear to affect primarily the cross-sectional shape of the For a sequence of transport directions that resulted in simu-
dunes. lated linear dunes (Fig. 3B), simulations were performed starting
It has been hypothesized that the observed range of dune from well-formed dunes with orientations ranging from transverse
classes can be explained by recourse to a sequence of sand-trans- to longitudinal and with 0 to 70 total dune-crest terminations. The
porting winds from differing directions or with differing strengths simulated dune systems evolve to or near to the point in phase space

1108 GEOLOGY, December 1995

Figure 3. Simulated dune fields under differing sequences of transport direction evolving from initial random morphology. A: Simulated crescentic
dunes at t 5 500. Transport direction toward top of map, average of 15 slabs per lattice site, l 5 5. Contour interval 5 10 slabs. B: Simulated
linear dunes at t 5 2000, mean transport direction toward top of map, average of eight slabs per lattice site, l1 5 (2, 6) and l2 5 (2, 26) (horizontal,
vertical), duration Dt 5 40. Contour interval 5 10 slabs. C: Simulated star dunes at t 5 1600; average of 15 slabs per lattice site; l1 5 (5, 0),
l2 5 (0, 5), l3 5 (25, 0), and l4 5 (0, 25); duration Dt 5 80. Contour interval 5 10 slabs.

representing well-formed linear dunes, with orientation 08 and 0

terminations (Fig. 4). This state has the same characteristics as the
state arising from a random initial condition under the same se-
quence of transport directions. This result is consistent with the
existence of a single attractor for the system with a basin of attrac-
tion that is the entire phase space.
The nature of the hypothesized attractor varies abruptly with
changes to the parameters of the model. For example, the asymp-
totic orientation changes from longitudinal to transverse as the
obliquity of the reversing transport directions decreases below about
458 (Fig. 5). A similar but less sharp transition between barchans and
crescentic ridges is observed as the sand supply is increased. Some
tentative evidence for the simultaneous existence of two attractors
for a single sequence of transport directions has been observed. In
this case, the simulated dunes approach a longitudinal or transverse
orientation depending on initial dune orientation.
Tests for the occurrence of emergent behavior in the simula-
tions involve finding macroscopic dune behavior that is insensitive
to the details of the microscopic transport rules employed; such tests
are more qualitative than tests for attractors. The evolution of a
simulated dune field is dominated by interactions between dunes
that take several simple forms. First, because of the inverse relation
between migration speed and size, small dunes catch up with large
dunes and interact by sand transfer or merger. Second, simulated
Figure 4. Hypothesized simulated linear-dune attractor. Sample phase-
dunes lose sand from their own terminations and gain sand from space trajectories starting from well-formed dunes with range of ori-
properly aligned terminations of other dunes. Third, the focal points entations and numbers of terminations and evolving to the region of 0°
for interactions among long-crested dunes are the dune-crest ter- orientation and 0 terminations. Orientation was chosen to be the peak
minations, which (for transverse dunes) migrate faster than the of the distribution of projections of the vectors tangent to a contour
(number of slabs 5 10) onto a line rotated in 4° increments. Number of
dunes because of their smaller cross-sectional size. The dynamics of
terminations was counted by eye. Same transport parameters as Fig-
terminations in simulated dune fields resemble the dynamics doc- ure 3B.
umented for natural (Anderson and McDonald, 1990) and simu-
lated (Landry and Werner, 1994) wind ripples. A vivid illustration
of emergent behavior is calving from the horns of barchan dunes directions that are consistent with proposed hypotheses and field
(Fig. 2B). All of these behaviors depend only on the most general observations. The justification for this gross simplification of the
features of the transport (e.g., dune migration speed varying in- dynamics of eolian sand transport originates in the nonlinear, dis-
versely with size) and have been reproduced by using many diverse sipative nature of the dynamics, which gives rise to the expectation
transport rules (e.g., erosion or deposition that depends in various that dunes exhibit the property of emergent behavior of a complex
ways on elevation, slope, or lee shadowing). system, a simplification of the dynamics for large space scales and
long time scales.
DISCUSSION Dunes have been classified both by their morphology (princi-
The simple transport algorithm described above results in pally the geometry and topology of dune crests) and by their ori-
forms that resemble natural dunes under sequences of transport entation based on observational evidence (Cooke et al., 1993). A

GEOLOGY, December 1995 1109

 single point (Fig. 4). At least one more variable, possibly charac-
teristic dune size, is required to specify the macrostate of dune
systems.
The simulation algorithm described here is potentially useful in
that it can make testable predictions for complicated three-dimen-
sional morphologies and for complicated wind regimes. This char-
acteristic permits falsification of the assumptions of the model
through direct comparison to field data. In addition, the algorithm
may be useful for exploring both the consequences and dynamical
origin of the numerous hypotheses that have been advanced regard-
ing eolian dunes. For example, it should be possible to investigate
a fundamental dynamical explanation for Rubin and Hunter’s
(1987) hypothesis that dunes are oriented so as to maximize sand
transport normal to the dune crest and to test its consistency with
the simulation algorithm. Finally, mapping out the range of trans-
port regime parameters and initial conditions for particular dune-
class attractors could allow careful quantification of the uncertainty
in inverting dune shapes and orientations for wind regimes and
initial conditions.
Figure 5. For oblique reversing transport, simulated dune orientation ACKNOWLEDGMENTS
as function of ratio of transport normal to and parallel to mean trans- Supported by the Coastal Dynamics and Young Investigator Programs of
port direction. As transport varies from unidirectional to bidirectional, the Office of Naval Research (grant N00014-92-J-1446). I thank G. Kocurek,
dune orientation changes from transverse to longitudinal at a transport P. K. Haff, and R. L. Shreve for discussions that were critical to the develop-
ratio of ;1. Initial morphology in all simulations was random, average ment of the ideas described herein.
number of slabs 5 5, and orientation measured at t 5 1200.
REFERENCES CITED
Anderson, R. S., and McDonald, R. R., 1990, Bifurcations and terminations in
plausible hypothesis is that these classifications correspond to the eolian ripples [abs.]: Eos (Transactions, American Geophysical Union),
v. 71, p. 1344.
attractors of the eolian sand transport system. In this view, the var-
Bagnold, R. A., 1973, The physics of blown sand and desert dunes: London,
iables that capture the interesting dynamical behavior of the system Chapman and Hall, 265 p.
(including dune orientation and number of terminations) are those Cooke, R., Warren, A., and Goudie, A., 1993, Desert geomorphology: London,
variables for which evolution to an attractor occurs on the time scale UCL Press, 526 p.
Fryberger, S. G., and Dean, G., 1979, Dune forms and wind regime, in McKee,
associated with this interesting behavior. The vast majority of var- E. D., ed., A study of global sand seas: U.S. Geological Survey Professional
iables, which evolve faster, do not influence the dominant behavior Paper 1052, p. 305–397.
and are said to be slaved to these interesting variables. Kocurek, G., Townsley, M., Yeh, E., Havholm, K., and Sweet, M. L., 1992,
The results and provisional interpretations presented here Dune and dune-field development on Padre Island, Texas, with implica-
tions for interdune deposition and water-table– controlled accumulation:
should be viewed as tentative because of (at least) four criticisms Geological Society of America Bulletin, v. 62, p. 622– 635.
that may require further investigation. First, because the treatment Landry, W., and Werner, B. T., 1994, Computer simulations of self-organized
of aerodynamical effects is highly simplified, the initiation of a dune wind ripple patterns: Physica D, v. 77, p. 238–260.
until the development of a slipface probably is not modeled well Nicolis, G., and Prigogine, I., 1989, Exploring complexity: New York, Freeman,
313 p.
(except perhaps for calving, where sand supply, rather than aero- Rubin, D. M., 1987, Cross-bedding, bedforms and paleocurrents: Tulsa, Okla-
dynamics, may be dominant). Second, the use of periodic boundary homa, Society of Economic Paleontologists and Mineralogists, 187 p.
conditions can artificially influence the evolution of simulated Rubin, D. M., and Hunter, R. E., 1987, Bedform alignment in directionally
dunes. Increasing the size of the periodic domain does not alter varying flows: Science, v. 237, p. 276–278.
qualitative dune characteristics (general orientation and morphol- Manuscript received May 22, 1995
ogy) for the simulations described above. Third, the simple rules for Revised manuscript received August 18, 1995
erosion and deposition employed result in somewhat steep stoss Manuscript accepted August 28, 1995
slopes and a weak but unphysical dependence of cross-sectional
shape on dune size. A possible remedy is modifying the transport
rules to account for the relationships among local morphology, wind
profile, and transport flux in greater detail. Fourth, the variables
orientation and number of terminations do not describe simulated
dunes adequately as a dynamical system, as is seen from the
nonuniqueness of the phase-space trajectories emanating from a

1110 Printed in U.S.A. GEOLOGY, December 1995