1: \documentclass[pre,twocolumn,showpacs,preprintnumbers,floatfix]{revtex4}

2: \usepackage{graphicx,citesort,psfrag,amsmath,amssymb}

3: 

4: \newcommand{\rem}[1]{\textbf{#1}}

5: \newcommand{\ve}[1]{\mathbf{#1}}

6: \newcommand{\dd}[2]{\frac{\partial #1}{\partial #2}}

7: \newcommand{\tdd}[2]{\partial #1/\partial #2}

8: \newcommand{\avg}[1]{\left\langle #1 \right\rangle}

9: 

10: \sloppy

11: 

12: 

13: \begin{document}

14: 

15: \bibliographystyle{apsrev}

16: 

17: %\preprint{}

18:  

19: \title{Minimal model for aeolian sand dunes}

20: 

21: \author{Klaus Kroy}

22: \affiliation{Dept.\ of Physics and Astronomy, University of

23: Edinburgh, EH9 3JZ, United Kingdom}

24: \author{Gerd Sauermann, and Hans J. Herrmann}

25: \affiliation{ICA-1, Universit\"at Stuttgart, Pfaffenwaldring 27, 70569

26: Stuttgart, Germany} 

27: 

28: \date{\today} 

29:              

30:  

31: \begin{abstract}

32:  We present a minimal model for the formation and migration of aeolian

33:   sand dunes. It combines a perturbative description of the turbulent

34:   wind velocity field above the dune with a continuum saltation model

35:   that allows for saturation transients in the sand flux. The latter

36:   are shown to provide the characteristic length scale. The model can

37:   explain the origin of important features of dunes, such as the

38:   formation of a slip face, the broken scale invariance, and the

39:   existence of a minimum dune size. It also predicts the longitudinal

40:   shape and aspect ratio of dunes and heaps, their migration velocity

41:   and shape relaxation dynamics. Although the minimal model employs

42:   non--local expressions for the wind shear stress as well as for the

43:   sand flux, it is simple enough to serve as a very efficient tool for

44:   analytical and numerical investigations and to open up the way to

45:   simulations of large scale desert topographies.

46: \end{abstract}

47:  

48: \pacs{45.70.Mg, 45.70.Qj, 47.27.-i, 51.10.+y}

49: 

50: %\keywords{Suggested keywords}

51: 

52: \maketitle

53: 

54: 

55: \section{Introduction}

56:   

57: Sand dunes develop wherever sand is exposed to an agitating medium

58: such as air or water that lifts grains from the ground and entrains

59: them into a surface flow. The diverse conditions of wind and of sand

60: supply in different regions on Earth give rise to a large variety of

61: shapes of aeolian dunes \cite{Bagnold41,Pye90,Lancaster95}. Moreover,

62: dunes have been found on the sea--bottom and even on Mars

63: \cite{malin-etal:98,Thomas99}.  Despite the long history of the

64: subject, the underlying physical mechanisms of dune formation are

65: still not very well understood.  How are aerodynamics (hydrodynamics)

66: and the particular properties of granular matter acting together to

67: create dunes?  How is the shape of a dune maintained when it moves?

68: Since the macroscopic phenomena of interest are separated by many

69: orders of magnitude from the grain scale and involve various coupled

70: nonlinear processes such as turbulent air flow and grain hopping

71: (``saltation''), one is bound to devise some simplified models in

72: order to address such questions. We will argue that approximate

73: numerical models can only be successful if based on a sound

74: qualitative understanding of the problem. Therefore, our main aim is

75: to identify the key mechanisms underlying dune formation and migration

76: and incorporate them into a working minimal model of aeolian sand

77: dunes, and we will emphasize generic aspects over the more specific

78: details. For definiteness, the reader may find it helpful to think of

79: isolated transverse dunes or crescent--shaped barchan dunes as major

80: applications of the model. A schematic sketch of the height profile of

81: a barchan is shown in Fig.~\ref{fig:barchan}. The broad phenomenology

82: of aeolian and submarine land forms provides a large number of

83: different characteristic structures that can certainly not all be

84: described by the same simple model developed with the specific

85: examples of barchan or transverse dunes in mind. However, we expect

86: that our approach is amenable to future adaptations that make it

87: applicable to a broader class of sand topographies on the one hand,

88: and for quantitative investigations of more specific questions on the

89: other hand. Although the minimal model refers only to rather generic

90: properties of the wind velocity field and the laws of aeolian sand

91: transport, it can make interesting predictions about the surface

92: profile, the development and position of the slip face, dune migration

93: etc.\ that are insensitive to the simplifying assumptions. The main

94: features of the model were already briefly presented in a recent

95: Letter \cite{kroy-sauermann-herrmann:2002}.  The present contribution

96: gives a more comprehensive discussion of the model and tries to

97: communicate its precise definition as well as its major predictions to

98: an interdisciplinary readership. The model, as presented here, is

99: restricted to a two--dimensional ($2d$) slice of a dune parallel to

100: the unidirectional wind.  (A generalization to $3d$ problems is in

101: preparation.) A further restriction is the neglect of ripples and

102: direct slope effects onto the sand transport outside slip

103: faces. Although they have successfully been incorporated into

104: continuum sand transport models

105: \cite{hoyle-woods:97,hoyle-mehta:99,prigozhin:99} similar to our own

106: \cite{sauermann-kroy-herrmann:2001}, we chose to disregard them for

107: the present purpose and leave their integration to future work.

108: 

109: The paper is organized as follows. In the next introductory section we

110: summarize some background knowledge and basic definitions. We will

111: also introduce a naive ``zeroth order'' description of the wind shear

112: stress and the induced aeolian sediment transport. Its instructive

113: failure to produce dune--like steady--state solutions will be a guide

114: for identifying two relatively small effects (the upwind shift of the

115: maximum of the shear stress with respect to the topography and the

116: saturation transients in the sand flux) as key ingredients of a proper

117: description of structure formation by aeolian sand transport.  We will

118: moreover derive the scaling behavior of the migration velocity for

119: translation invariant heaps and dunes of different size but similar

120: shape based on very general grounds. Sections~\ref{sec:wind} and

121: \ref{sec:saltation} are devoted to the definition of the minimal

122: model, i.e., to the modeling of the air shear stress exerted onto a

123: heap of sand and the induced sand transport, respectively. The first

124: step builds on turbulent boundary layer calculations developed in a

125: series of publications mainly by Hunt and coworkers

126: \cite{jackson-hunt:75,%

127: sykes:80,zeman-jensen:88,hunt-leibovich-richards:88,Carruthers90,weng-etal:91},

128: the second one on a previous contribution

129: \cite{sauermann-kroy-herrmann:2001} by the present authors. Only the

130: most pertinent results of these earlier developments will be

131: summarized here.  In the remainder, we will derive some important

132: predictions of the model for the central slice of a barchan dune or

133: transverse dune. In particular, we will demonstrate that there is a

134: minimal dune size. Although we will thereby gain interesting results,

135: these are rather meant to be illustrative examples of possible

136: applications of the model. By no means do we attempt to provide a

137: complete analysis of its predictions, and it should become obvious

138: that much more remains still to be done.  Finally we will summarize

139: our main results and speculate about probable consequences of the

140: present $2d$ theory for $3d$ topographies.

141: 

142: \begin{figure}[t]

143:   \includegraphics[width=\columnwidth]{sketch.eps} 

144: \caption{Sketch of a barchan dune. Sand is eroded by the wind on the

145: upwind or ``stoss'' side and transported to the brink. Strong

146: deposition occurs due to flow separation behind the brink. On the

147: downwind or ``lee'' side, sand slides down at the angle of repose

148: (about $32^{\circ}-35^{\circ}$) over a concave slip face.}

149: \label{fig:barchan}

150: \end{figure}

151: 

152: 

153: \section{General}\label{sec:general}

154: \subsection{Aeolian sand transport}\label{sec:aeolian}

155: Before going into the description of the model, we want to recall some

156: general background and to introduce some quantities of major interest.

157: First of all, for convenience, we will usually refer to dunes without

158: slip face as heaps. Further, we will sometimes find it helpful to

159: focus on isolated heaps or dunes on bedrock, although most of our

160: discussion is not restricted to this situation.  

161: 

162: The key quantity for the description of the formation and migration of

163: sand dunes and heaps is the local horizontal surface velocity $v(x,t)$

164: of a sand height profile $h(x,t)$ at all positions $x$ and times

165: $t$. Via mass conservation it can be related to the erosion rate

166: $\nabla q(x,t)$ (negative erosion is deposition), where the sand flux

167: $q(x,t)$ is defined as the mass of sand transported per unit of time

168: across a hyper--plane transverse to the wind direction. More

169: precisely, since we want to specialize our discussion to a $2d$ slice

170: parallel to the unidirectional wind velocity, the hyperplane is a

171: vertical line and $q$ is a flux per unit width. Mass conservation then

172: takes the form of a continuity equation for the height profile

173: \begin{equation}\label{eq:mass}

174: \varrho_{s}\frac{d h(x,t)}{dt} = -\frac{\partial q(x,t)}{\partial x} 

175: \end{equation}

176: with $\varrho_{s}$ the density of the sand bed. 

177: 

178: With Eq.(\ref{eq:mass}) one can write the position dependent migration

179: velocity at a given time $t$ as

180: \begin{equation}\label{eq:v_def}

181: v(x) = \varrho_{s}^{-1} \; \frac{q'}{h'}

182: % = \varrho_{s}^{-1} \; \frac{dq}{dh} 

183: \;,

184: \end{equation}

185: where we have introduced the shorthand notation $f'(x)\equiv

186: df(x)/dx$.  At this stage we can already get some physical insight by

187: observing that this equation needs special attention at the top of a

188: heap or dune, where we expect the denominator to vanish. For $v$ to

189: remain finite at the crest as required in the steady state, there are

190: in general only two possibilities. Either the sand flux $q$ is fine

191: tuned so that the erosion $q'$ vanishes in exactly the same way as the

192: slope $h'$, or the profile $h(x)$ is not differentiable at the

193: crest. As the reader may already anticipate and will be verified

194: below, both cases have their physical realizations, the former

195: in heaps or small dunes with smooth crests and the latter in large

196: dunes with a slip face that terminates in a sharp brink.

197: 

198: The problem we face, if we want to calculate the dynamic evolution of

199: desert topographies, is the closure of Eq.(\ref{eq:mass}) or

200: (\ref{eq:v_def}) by expressing the flux $q(x,t)$ in terms of the

201: height profile $h(x,t)$ and the external wind and boundary

202: conditions. Since for the applications we have in mind, the migration

203: velocity is very small compared to the speed of elementary sand

204: transport processes (grain hopping etc.) and the wind speed, the

205: topography can be assumed to be stationary for considerations

206: concerning the wind and sand transport dynamics. This allows one to

207: subdivide the problem of calculating $q(x)$ into two independent

208: steps. First, one needs to know the stationary wind velocity above a

209: given topography. More precisely, what is required is the shear stress

210: $\tau$ exerted by the wind onto the ground. And secondly, one needs a

211: model that predicts the stationary sand flux $q(x)$ for a given

212: stationary $\tau(x)$, schematically

213: \begin{align}

214: h(x) &\to \tau(x) \label{eq:task1} \\

215: \tau(x) &\to q(x)\;. 

216: \label{eq:task2}

217: \end{align} 

218: Computing the derivative $q'$ and integrating the mass conservation

219: Eq.(\ref{eq:mass}) then closes the model and allows one to predict the

220: development of the surface profile in time.  Since aeolian dunes

221: typically have relatively gentle slopes outside their slip face, we

222: will at the present stage restrict the scope of the minimal model to

223: this case and disregard in Eq.(\ref{eq:task2}) the direct slope

224: effects $h'(x) \to q(x)$ onto the flux outside the slip face.

225: 

226: In special cases, the relations (\ref{eq:task1}), (\ref{eq:task2}) are

227: phenomenologically and theoretically well established.  For a flat

228: surface, $h(x)\equiv$ const., it is well known

229: \cite{landau-lifshitz:fm} that the mean turbulent wind velocity

230: increases logarithmically with height above the surface. It can be

231: characterized by a single characteristic velocity, the ``shear

232: velocity'' $u_*$ defined by $u_*^2\equiv\tau_0/\varrho_{a}$ with

233: $\tau_0$ the (suitably time averaged) shear stress and $\varrho_{a}$

234: the density of air. Since the shear stress of the air is transmitted

235: to the surface, the latter can mobilize grains on a surface covered

236: with sand, if it exceeds a threshold value $\tau_t$. As a result, the

237: wind entrains some grains into a surface layer flow.  The grains

238: advance mainly by an irregular hopping process (``saltation''),

239: thereby reducing the wind velocity in the surface layer. Via this

240: feed--back mechanism a unique relation between the shear stress $\tau$

241: and the sand flux $q$ is established in the equilibrium state.  If

242: $\tau$ is not too close to the threshold, this relation can

243: approximately be represented as \cite{Bagnold41}

244: \begin{equation}\label{eq:bagnold} 

245: q_s \propto \tau^{3/2} \;.     

246: \end{equation} 

247: Although a host of more accurate descriptions have been discussed in

248: the literature

249: \cite{Lettau78,Sorensen91,Pye90,ActaMechanica91,sauermann-kroy-herrmann:2001}

250: and one of them will be part of our definition of the minimal model

251: below, the simpler Eq.(\ref{eq:bagnold}) will be sufficient for our

252: qualitative discussion in the first part of the paper. The index $s$

253: in Eq.(\ref{eq:bagnold}) emphasizes that such local relations are

254: restricted to situations where the flux is saturated, that is, equal

255: to its equilibrium transport capacity. This is certainly not the case

256: near a boundary between uncovered and covered ground or on sloped

257: beds. Neglecting this restriction for the moment,

258: Eq.(\ref{eq:bagnold}) predicts that the shear stress perturbation

259: \begin{equation}\label{eq:tauhat} 

260:   \hat\tau(x) \equiv \tau(x)/\tau_0-1

261: \end{equation} 

262: above a modulated topography $h(x)$ is responsible for flux gradients

263: $dq_s/dx$ that cause erosion and deposition and thus --- according to

264: Eqs.(\ref{eq:mass}), (\ref{eq:v_def}) --- migration of the sand surface.

265: Explicitly closing the model by assuming that the shear stress is an

266: affine function of the modulation of the topography ($\hat \tau

267: \propto h$) leads to what we call the ``zeroth order'' model, which

268: will briefly be analyzed in the next paragraph.

269: 

270: \subsection{The ``zeroth order'' model}\label{sec:zero}

271: The zeroth order model is given by

272: \begin{align}

273: \hat\tau \{h(x)\} & \to \hat\tau(h)\propto h(x)/L \label{eq:tau0} \\ 

274: q\{\tau(x)\}  & \to q(\tau)= q_s(\tau) \label{eq:q0}

275: \end{align}

276: where we have used the curly brackets to indicate a general functional

277: dependence and introduced a characteristic length scale $L$ of the

278: topography to normalize the height profile. (The motivation for the

279: latter step will become clear in the next section.) The zeroth order

280: model assumes \emph{local} relations in

281: Eqs.(\ref{eq:task1}),(\ref{eq:task2}). It approximates the wind shear

282: stress perturbation by its ``affine'' contribution (proportional to

283: the profile $h$ that causes the perturbation) and replaces the true

284: sand flux $q$ by its saturated value $q_s$, thereby neglecting

285: saturation transients.  This model is so simple that its qualitative

286: predictions for an arbitrary smooth heap of sand can easily be

287: anticipated without doing any actual calculations. 

288: 

289: Combining Eq.(\ref{eq:v_def}) with

290: Eqs.(\ref{eq:bagnold})--(\ref{eq:q0}) one obtains a surface velocity

291: that increases with height ($dv/dh \geq 0$) due to the nonlinearity

292: of Eq.(\ref{eq:bagnold}). This implies that the upwind (or ``stoss'')

293: slope tends to decrease and the downwind (or ``lee'') slope tends to

294: increase. Since $dq/dx \propto dh/dx$ by the chain rule, there is no

295: erosion or deposition at the top of a smooth heap, which therefore

296: keeps its initial height. Obviously, integrating forward in time will

297: eventually increase the lee slope up to the angle of repose, where

298: surface avalanches have to be introduced and a slip face of constant

299: slope develops.  If the latter reaches the crest the above argument

300: for the persistence of the height can no longer be applied, because

301: the slope at the crest is then ill defined. Since there is so far

302: nothing to stop a further decrease of the windward slope, the model

303: dune will then start to decrease in height and finally flatten

304: out. The steady--state solution is a flat surface.

305: 

306: The simple argument shows that the zeroth order model --- although it

307: gives some clue as to the origin of the slip face --- is insufficient

308: for a proper qualitative understanding of dunes. However, some

309: important lessons can be learned from it that will be helpful in our

310: further investigation of the problem. First, even with a very

311: simplistic model any reasonably heap--like initial condition will

312: quickly develop into a dune--like shape with a slip face. Secondly,

313: although the latter may seem to converge to a steady--state solution

314: for intermediate times, it finally turns out to be unstable and

315: flattens out. The discussion of the migration velocity in

316: Section~\ref{sec:aeolian} suggests that small deviations from

317: Eqs.(\ref{eq:tau0}) and (\ref{eq:q0}) at the brink can make an

318: important difference. Obviously some caution is needed in judging the

319: success of numerical models of dune formation. Unless stability has

320: explicitly been demonstrated, they may be suspected to fail in a

321: similar way as the zeroth order model when integrated over

322: sufficiently long times (which has actually not been checked for some

323: models that can be found in the literature) or to be sensitive to

324: numerical errors at the brink. Detailed numerical modeling should

325: therefore be preceded by a sound qualitative understanding of the

326: mechanisms underlying dune formation.  We will argue in

327: Sections~\ref{sec:wind}, \ref{sec:saltation} that to this end a subtle

328: balance between two small deviations from

329: Eqs.(\ref{eq:tau0}),(\ref{eq:q0}) and especially non--local

330: contributions in Eq.~(\ref{eq:task2}) have to be taken into account.

331: 

332: \subsection{Migration velocity}\label{sec:v}

333: Before entering a detailed discussion of the minimal model, it is

334: worth pausing for some general thoughts as to what can be said about

335: the shear stress and the speed--up of the wind above an obstacle,

336: without actually doing the (somewhat involved) calculation. 

337: 

338: A basic property of strongly developed turbulence is its dilation

339: invariance or scale free structure. Whereas general Navier--Stokes

340: flow is invariant under a scale transformation that keeps the Reynolds

341: number constant, strongly turbulent flow (for ``infinite'' Reynolds

342: number) allows for infinitely many such similarity

343: transformations. Landau and Lifshitz \cite{landau-lifshitz:fm} took

344: advantage of this fact for deriving the logarithmic velocity profile

345: mentioned above by an elegant scaling argument. The logarithmic

346: velocity profile suggests that the speed--up of the wind and therefore

347: also the shear stress perturbation above a heap of given shape should

348: itself be logarithmically dependent on its size. But how do they

349: depend on the shape of the obstacle?  Since the flow itself does not

350: provide any characteristic length scale, the dimensionless quantity

351: $\hat \tau$ defined in Eq.(\ref{eq:tauhat}) can only depend on a

352: dimensionless characterization of the profile $h(x)$. In other words,

353: to lowest order in the perturbation, it must be a linear functional of

354: the derivative $h'$ and can be written as

355: \begin{equation}\label{eq:scaling}

356: \hat \tau(\xi) = 

357: 	\varepsilon \,{\cal T}\left\{f'(\xi)\right\}\;, \qquad

358: 	\varepsilon \equiv H/\! L \;,

359: \end{equation}

360: with a dimensionless profile function

361: \begin{equation}\label{eq:profile}

362: f(\xi)\equiv h(x)/H \qquad \xi \equiv x/\! L \;. 

363: \end{equation}

364: and a scale--free (and necessarily non--local) linear functional

365: ${\cal T}$.  This reasoning can be repeated for the dimensionless

366: velocity and pressure perturbations.  Intuitively, the scaling $\hat

367: \tau \propto \varepsilon$ for a flat smooth obstacle ($\varepsilon \ll

368: 1$) can be understood from Fig.~\ref{fig:scaling}. When the air flows

369: over the obstacle, the velocity close to the obstacle is deflected by

370: an angle $\varepsilon$ whereas it remains constant far above the

371: obstacle. For incompressible flow, continuity translates this into a

372: speed--up of order $\varepsilon$ and (via Bernoulli's law) into a

373: corresponding pressure drop near the top of the heap. This in turn

374: causes a shear stress perturbation $\hat \tau$ of the same order.

375: 

376: \begin{figure}[t]

377: \psfrag{H}{$H$}

378: \psfrag{L}{$L$}

379: \psfrag{e}{$\varepsilon$}

380: \psfrag{u}{$u$}

381: \psfrag{ee}{$u+\varepsilon u $}

382: \begin{center}

383: \includegraphics[width=\columnwidth]{airflow.eps}

384: \end{center}

385: \caption{Schematic sketch of the deflection of the wind velocity $u$

386: above a flat heap of aspect ratio $\varepsilon \equiv H\!/\!

387: L\ll1$. The characteristic length scale $L$ is in this context

388: conventionally often identified with  the half length at half

389: height of the heap. The vertical deflection causes a speed--up above

390: the top of the heap. This is accompanied by a pressure perturbation

391: that is negative above the top of the heap and positive at its tails.

392: Due to turbulence, the flow pattern is asymmetric even above a

393: symmetric heap.}

394: \label{fig:scaling}

395: \end{figure}

396: 

397: These general considerations already allow us to predict the scaling

398: of the migration velocity $v$ with dune size if we assume that dunes

399: of different size have roughly similar shapes $f(\xi)$ and aspect ratios

400: $\varepsilon$, which is indeed suggested by the scale invariance of

401: the turbulent wind field and by observations.  Inserting

402: Eq.(\ref{eq:scaling}) into Eq.(\ref{eq:v_def}), and again

403: approximating Eq.(\ref{eq:task2}) by a local sand transport law $q

404: \equiv q(\tau)$, we find

405: \begin{equation}\label{eq:v}

406:  v  \frac{dq}{d\hat\tau}

407: 	\frac{d\hat\tau/dx}{\varrho_{s}\varepsilon f'}

408: 	=\frac{dq}{d\hat\tau}\frac{{\cal

409: 	T'}\{f'\}}{\varrho_{s} L f'} \propto \frac1L \;.

410: \end{equation}

411: The final proportionality strictly holds only if the steady--state

412: shape $f(\xi)$ and aspect ratio $\varepsilon$ are scale

413: invariant. However, it can be expected to be robust and rather

414: insensitive against violations of exact scale invariance.  First, the

415: normalized steady--state shapes $f(\xi)$ are strongly constrained by

416: the requirement that they render $v(x)\equiv v$, independent of $x$,

417: along the heap. Therefore, they should to a first approximation be

418: independent of size, which is indeed borne out by the minimal model

419: (Fig.~\ref{fig:shapes}) and empirical observations

420: \cite{sauermann-etal:2000}.  Moreover, the dependence $v\{ f'(\xi) \}$

421: is rather indirect and can therefore be expected to be weak. Secondly,

422: Eq.(\ref{eq:bagnold}) suggests that for gently sloped obstacles

423: ($\varepsilon \ll 1$) the dependence of $dq/d\tau$ on the aspect ratio

424: $\varepsilon$ also is not very pronounced. And finally, --- due to the

425: above mentioned scale invariance of turbulence --- a scale invariant

426: aspect ratio can reasonably be expected for large dunes. In fact, we

427: will show below that the minimal model predicts that the aspect ratio

428: of small heaps is not constant but rather decreases proportional to

429: their height. But this also implies that the latter becomes too small

430: to have a very significant effect on the above argument.  Note,

431: however, that only for strictly scale invariant dunes, Eq.(\ref{eq:v})

432: becomes identical to the often quoted observation that dunes migrate

433: with a speed inversely proportional to their height

434: \cite{speed}. Since the deviations of large dunes from scale

435: invariance are not very pronounced, the difference between these

436: predictions is not very strong except for small dunes and heaps.

437: Presently available field data are maybe not accurate enough to

438: clearly distinguish between $v\propto 1/L$ and $v\propto 1/H$, though

439: some data support $v\propto1/L$, most notably the comprehensive study

440: of barchan dunes in southern Peru by Finkel \cite{Finkel59}. As we

441: will show below, our numerical results for the minimal model clearly

442: favor $v\propto1/L$.

443: 

444: We also note that together with Eq.(\ref{eq:bagnold}), Eq.(\ref{eq:v})

445: moreover predicts that the migration velocity grows non--linearly with

446: (as the third power of) the wind velocity. A more accurate relation

447: can be obtained from the minimal model as described below, but the

448: qualitative conclusion is the same. Dunes can migrate farther in a

449: short period of exceptionally strong wind than during much longer

450: periods of gentle winds. Finally, we should mention that some caution

451: is needed when identifying the characteristic length scale $L$ in

452: Eq.(\ref{eq:v}). In our discussion, we have so far assumed that

453: $f(\xi)$ is a smooth function, which is not the case for dunes with

454: slip face. Below we will argue that in this case $f$ should be

455: identified with the envelope of the dune and its separation bubble and

456: $L$ with the characteristic length scale of this envelope. For a

457: barchan dune the latter practically coincides with the total length of

458: the dune from its windward end to the tips of its horns (cf.\

459: Fig.~\ref{fig:barchan}).

460: 

461: In contrast to the overall migration velocity of a translation

462: invariant dune, the position dependent migration velocity $v(x)$ that

463: determines that shape is much harder to obtain since it requires a

464: precise knowledge of the non--local functional ${\cal T}$ in

465: Eq.(\ref{eq:scaling}). This will be provided in the following section.

466: 

467: \section{Wind shear stress}\label{sec:wind}

468: \subsection{Surface shear stress on a smooth heap}\label{sec:theory}

469: The discussion in the preceding paragraph showed that --- in contrast

470: to the assumption in Eq.(\ref{eq:tau0}) --- the dependence of the

471: shear stress on the height profile is non--local. Although it will

472: turn out that this shortcoming of Eq.(\ref{eq:tau0}) is not

473: responsible for the failure of the zeroth order model, it should by

474: now have become apparent that further progress can hardly be achieved

475: without a rather detailed understanding of the turbulent wind field

476: above heaps and dunes. For dunes with a slip face that typically has a

477: slope of about $32^{\circ}-35^{\circ}$ and terminates in a sharp

478: brink, the situation is similar to the textbook example of a backward

479: facing step, which has the reputation of a test case for numerical

480: turbulent models. Even if a commercial turbulent solver is used, the

481: accurate calculation of the shear stress e.g.\ on a barchan dune is a

482: non--trivial task and quite demanding in computer time and memory, and

483: the most interesting long--time dynamics of dunes is therefore

484: difficult to access. For this reason, we want to focus on flat smooth

485: heaps, first. In this case, one can apply an analytical perturbation

486: theory for turbulent boundary layer flow over smooth hills that has

487: been developed over the last decades \cite{jackson-hunt:75,sykes:80,%

488: zeman-jensen:88,hunt-leibovich-richards:88,Carruthers90,weng-etal:91}.

489: Though the calculation is essentially a formalization of the intuitive

490: description accompanying Fig.~\ref{fig:scaling}, it requires a highly

491: non--trivial boundary layer construction that we will not recapitulate

492: here. The interested reader is referred to the original literature. We

493: merely quote the final result for the $x-$component (along the main

494: wind direction) $\hat \tau_x$ of the surface shear stress perturbation

495: above a profile $h(x,y)$

496: \cite{hunt-leibovich-richards:88,weng-etal:91},

497: \begin{equation}

498: \label{eq:hunt} 

499: {\cal F}_{xy}\{\hat\tau_x\} = \frac{A k_x( k_x +

500: i B |k_x|)}{(k_x^2 + k_y^2)^{1/2}} \, {\cal F}_{xy} \{ h(x,y) \}

501: \;.  

502: \end{equation} 

503: We have abbreviated the Fourier transformation from the space

504: variables $x$, $y$ to the respective wave numbers $k_x$, $k_y$ by

505: ${\cal F}_{xy}$. For simplicity the logarithmic $k-$dependence of the

506: parameters $A$ and $B$ was neglected.  The latter are then given

507: by

508: \begin{equation}\label{eq:a_b}

509: \begin{split}

510:  A &= 

511: \frac{\ln \left(\Phi^2/\ln \Phi\right)^2}{2(\ln \phi)^3}

512:  [1+\ln \phi+2\ln(\pi/2)+4\gamma] \\

513:  B &= \pi/[1+\ln\phi+2\ln(\pi/2)+4\gamma_E] \\

514:   \phi & \equiv 2\kappa^2\Phi/\ln\phi 

515: \end{split} 

516: \end{equation}

517: and depend logarithmically on the ratio $\Phi\equiv L/z_0$, where $L$

518: is the characteristic length of $h(x,0)$ (for this purpose

519: conventionally often identified with half the length at half height or

520: about one fourth of the characteristic wavelength) and $z_0$ is a

521: measure of the surface roughness (typically an effective length

522: somewhat below the linear dimension of the latter). We also have

523: introduced the von K\'arm\'an constant $\kappa\approx 0.4$ and Euler's

524: constant $\gamma_E\approx 0.577$.  A practical approximation for

525: $\phi$ is obtained by iterating (twice) the implicit equation for

526: $\phi$ and closing it by dropping the remaining $\ln\phi$.  The

527: dependence of $A$ and $B$ on $L$ is depicted in

528: Fig.~\ref{fig:ab}. Obviously, as long as $L/z_0$ does not change by

529: orders of magnitude (e.g.\ due to vegetation), this extremely weak

530: scale dependence is negligible for our purposes, and $A$ and $B$ can

531: be regarded as constant theoretical or phenomenological parameters.

532: For definiteness we often work with the values $A=4$ and $B=0.25$

533: (approximately obtained for $L/z_0 = 10^5 \dots 10^6$). Although these

534: values may differ somewhat depending on the particular application in

535: mind, or on the presence or absence of ripples, and may

536: phenomenologically be somewhat different from the theoretical

537: prediction, this does not affect our general conclusions.

538: 

539: \begin{figure}[tb]

540:    \psfrag{L}{$10^{-5} L/z_0$}

541:    \psfrag{A}{$A$, $B$} 

542: \begin{center}  

543:    \includegraphics[width=\columnwidth]{a_b_log.eps} 

544: \caption{The

545:    theoretical prediction for the dependence of the parameters $A$

546:    (solid line) and $B$ (dashed line) of Eq.(\ref{eq:a_b}) on the

547:    ratio of the dune size $L$ to the roughness length $z_0$.}  

548: \label{fig:ab}

549: \end{center}

550: \end{figure} 

551: 

552: For the following discussion we want to specialize Eq.(\ref{eq:hunt})

553: to the central slice of a transverse or barchan dune along the wind

554: direction. To this end we evaluate Eq.(\ref{eq:hunt}) for the central

555: slice $h(x)$ of a heap that has a Gaussian shape with standard

556: deviation $\sigma$ in the transverse direction parameterized by $y$,

557: \begin{equation}\label{eq:gauss}

558: h(x,y)=h(x)e^{-y^2/2\sigma^2} \;.

559: \end{equation}

560: This approximation is technically useful, and although it may seem

561: relatively crude for a particular real dune, it typically does not

562: introduce any noteworthy derogation of the results compared to a more

563: accurate description. The Fourier coefficients of the shear stress

564: $\hat\tau(x)\equiv \hat\tau_x(x,0)$ on the central slice along the

565: wind direction become

566: %\begin{equation}

567: \begin{multline}

568: {\cal F}\{\hat\tau(x)\}= \frac{\sigma A}{\sqrt{2\pi}} k(k +

569: iB |k|) \times

570: \\

571:  e^{-\frac14k^2\sigma^2}

572: K_0\!\left(\frac{k^2\sigma^2}4\right){\cal F}\{h(x)\}

573: \end{multline}

574: %\end{equation}

575: Here, $K_0$ denotes a modified Bessel function and the Fourier

576: transforms are one--dimensional, so that we can drop the redundant

577: $x-$subscripts.  

578: 

579: For transverse dunes ($\sigma/L\to \infty$), we obtain the two

580: equivalent expressions

581: \begin{subequations}\label{eq:wind}

582: \begin{align}

583: {\cal F}\{\hat\tau^\infty (x)\} &= A(|k|+iB\, k){\cal F}\{h(x)\} \;. 

584: \label{eq:fourier_wind} \\

585:  \hat\tau^\infty(x) &= A \left[ h'(x)\otimes (\pi x)^{-1} + B h'(x)\right]

586:  \;. \label{eq:real_wind}

587: \end{align}

588: \end{subequations}

589: For the real space version we have abbreviated a convolution integral

590: according to

591: \begin{equation}

592: \label{eq:convolution}

593: f\otimes g \equiv \int_{-\infty}^{\infty}\!\!\!\!  d\xi \; f(\xi)\,g(x-\xi) \;.

594: \end{equation} 

595: Evaluation for arbitrary $\sigma$ gives two correction terms

596: \begin{equation}

597: \label{eq:correction}

598: \hat\tau^\sigma=\hat\tau^\infty - A (h\otimes \Delta_1 + B

599: h'\otimes \Delta_2) \;,

600: \end{equation}

601: with 

602: \begin{align}

603: \label{eq:deltas}

604: \Delta_1 &=

605: \frac{U\left(\frac32,1,\frac{x^2}{2\sigma^2}\right)}{4\sqrt \pi

606: \sigma^2}+\frac{\frac{\sqrt{\pi}}2

607: U\left(\frac12,0,\frac{x^2}{2\sigma^2}\right)-1}{\pi x^2} \\ 

608: \Delta_2 &=\frac1{\sigma \pi}\int_{0}^{\infty}

609: \!\!\!\!d\xi\,\cos\left(\frac{\xi x}\sigma\right) \left[1-\frac{\xi

610: e^{\xi^2/4}}{\sqrt{2\pi}} K_0\left(\frac{\xi^2}4\right)\right]

611: \end{align}

612: two even functions depicted in Fig.~\ref{fig:corrections} that are

613: flat for $\sigma/L\to\infty$ and become peaked for $\sigma\simeq

614: L$. (The confluent hypergeometric $U$ functions

615: \cite{abramowitz-stegun} have been introduced to rephrase the

616: sine--part of the Fourier integrals.)

617: 

618: 

619: \begin{figure}[t]

620: \psfrag{x}{$x/L$}

621: \psfrag{De}{$\Delta_1$} 

622: \psfrag{Ds}{$\Delta_2$} 

623: \begin{center}

624:  

625: \includegraphics[width=\columnwidth]{d2.eps} 

626: \includegraphics[width=\columnwidth]{d1.eps}

627: \caption{The peaked functions $\Delta_1$ and $\Delta_2$ of

628: Eq.(\ref{eq:deltas}) for $\sigma=1$, 2, 5. The area under the peaks

629: remains constant (0 and $1/2\pi$), while the peak heights decrease

630: proportional to $\sigma^{-2}$ and $\sigma^{-1}$, respectively.}

631: \label{fig:corrections} 

632: \end{center} 

633: \end{figure} 

634: 

635: Since the correction terms in Eq.(\ref{eq:correction}) are numerically

636: small, we may --- given a reasonable localized heap shape in the wind

637: direction --- approximately replace both functions $\Delta_1$ and

638: $\Delta_2$ by delta functions, thus arriving at

639: \begin{equation}\label{eq:approxcorrect}

640: \hat\tau_\sigma\approx \hat\tau_\infty - A [c_1(\sigma) h + B

641: c_2(\sigma) h'] \;.

642: \end{equation}

643: In this approximation they are seen to give merely a

644: $\sigma-$dependent renormalization of the asymmetry parameter $B\to

645: B(\sigma) \lesssim B(\infty)= B$ and to add a (trivial) term

646: $c_1(\sigma) h(x)$ within the brackets of

647: Eq.(\ref{eq:real_wind}). Numerically, one can estimate $L\,c_1(

648: L/\sqrt2)$ and $c_2( L/\sqrt2)$ to be about 0.2 (cf.\

649: Fig.~\ref{fig:hunt_jackson}). The exact $\sigma-$dependence of the

650: coefficients is determined by the shape and extension of the heap in

651: the $x-$direction, because the area under the peaks $\Delta_1$ and

652: $\Delta_2$ is constant and independent of $\sigma$, while the peak

653: height decreases proportional to $\sigma^{-2}$ and $\sigma^{-1}$,

654: respectively. Since both corrections vanish for $\sigma/L\to\infty$

655: and do not contribute any substantial new effects to

656: Eq.(\ref{eq:wind}), they may for simplicity be omitted altogether in

657: the following discussion that mainly aims at a qualitative

658: understanding.  Fig.~\ref{fig:hunt_jackson} moreover shows that they

659: can approximately be mimicked by a renormalized parameter $A$ in

660: Eq.(\ref{eq:wind}) for the central slice of a symmetric heap. This

661: leads to the important conclusion that the wind shear stress on the

662: central slice of a $3d$ symmetric heap and on a heap with a profile

663: that is constant in the transverse direction, are qualitatively the

664: same and quantitatively similar, which was not \emph{a priori}

665: obvious.  Together with the fact that on a gently sloped obstacle the

666: transverse components of the shear stress are small compared to its

667: longitudinal components, this suggests that the predictions of

668: Eq.(\ref{eq:wind}) apply in a first approximation to any slice of a

669: dune parallel to the wind direction. In this sense, the study of

670: Eq.(\ref{eq:wind}) is representative.  Summarizing the foregoing

671: discussion we can say that to gain a qualitative understanding of dune

672: formation by aeolian sand transport one may focus on

673: Eq.(\ref{eq:wind}) as a model for the wind shear stress. We therefore

674: analyze this equation in some detail in the next paragraph.

675: 

676: \begin{figure}[t]

677: \psfrag{x}{$x/L$}

678: \psfrag{ta}{$\hat \tau(x)$} 

679: \begin{center} 

680: \includegraphics[width=\columnwidth]{hunt_jackson.eps} 

681: \caption{Shear stress perturbation above the central slice of a $3d$

682: symmetric ($\sigma=L/\sqrt 2$) Gaussian heap. The plot compares two

683: approximations to Eqs.(\ref{eq:hunt}), (\ref{eq:correction}) (points):

684: (i) Eq.(\ref{eq:approxcorrect}) with $Lc_1\approx c_2\approx0.2$

685: (solid line), and (ii) Eq.(\ref{eq:wind}) with $A$ renormalized by a

686: factor 0.8 (dashed line). While (i) is practically indistinguishable

687: of Eq.(\ref{eq:hunt}) on the present level of accuracy, the simpler

688: approximation (ii) already captures most of the $3d$ effects.}

689: \label{fig:hunt_jackson} 

690: \end{center} 

691: \end{figure} 

692: 

693: \subsection{Properties and consequences of Eq.(\ref{eq:wind})}\label{sec:properties}

694: 

695: A scaling analysis of Eq.(\ref{eq:wind}) immediately reveals

696: that $\hat \tau$ is indeed of the general form anticipated on general

697: grounds in Eq.(\ref{eq:scaling}). The amplification of the shear

698: stress at the top of a smooth profile is thus determined by its aspect

699: ratio $\varepsilon=H\!/\!L$ and is essentially independent of the

700: absolute height $H$. It only has a very weak logarithmic dependence on

701: the absolute size of the dune through the prefactors $A$ and $B$ given

702: in Eq.(\ref{eq:a_b}). Moreover, for a symmetric profile

703: $f(-\xi)=f(\xi)$, $\hat \tau$ is the sum of a symmetric part and an

704: anti--symmetric part, i.e., the flow over the heap has a symmetry

705: breaking component that is a consequence of turbulence.  The origin of

706: the symmetric and antisymmetric parts of $\hat \tau$ can intuitively

707: be understood as follows. As we have pointed out in

708: Section~\ref{sec:v} (see Fig.~\ref{fig:scaling}), the streamlines have

709: to be compressed above the heap if the perturbation is not to be

710: transmitted to infinite height, and as a consequence, there is a

711: corresponding increase in the shear stress. For the laminar average

712: flow, this speed--up and the associated decrease in atmospheric

713: pressure above the heap are symmetric for a symmetric heap as is the

714: corresponding shear stress perturbation, which accounts for the

715: dominant symmetric part of $\hat \tau$. On the other hand, the inertia

716: of the turbulent velocity fluctuations around this laminar main flow

717: contributes an asymmetric resistance to deflections of the flow. It

718: counteracts the upturn of the streamlines on the windward side and the

719: downturn on the lee side. Formally, this effect enters the

720: perturbative calculation of $\hat \tau$ through the Reynolds stress.

721: 

722: Further insight can be gained from special analytical

723: solutions to Eq.(\ref{eq:wind}).  For the normalized heap

724: profiles

725: \begin{equation}

726:    \label{eq:hills}

727: \begin{split}

728:    f_L(\xi) & = \frac1{1+\xi^2}\\

729:    f_G(\xi) & = \exp\left(-\xi^2\right)\\ 

730:    f^n_C(\xi) & = S(\xi) \cos^n \xi  

731: \end{split}

732: \end{equation}

733: with

734: \begin{equation}

735:     S(\xi)\equiv  

736:    \begin{cases}

737:      1 & |\xi| \leq \pi/2 \\

738:      0 & |\xi| \geq \pi/2   \;, 

739:    \end{cases} 

740: \end{equation}

741:  we obtain

742: \begin{equation}

743: \label{eq:solutions}

744: \begin{split}

745: {\hat\tau_L}  

746: &= A (1 - 2B \xi -\xi^2)f_L^2(\xi) \\ 

747: {\hat\tau_G}

748: &=2A \left[\pi^{-1/2} -  \xi(B + \text{erfi}\, \xi) 

749: f_G(\xi)\right] \\

750: {\hat\tau_C}

751:  &=\frac{A}{\pi} \cos(2\xi) \left[\,\text{Si}(\pi+2\xi) +

752: \text{Si}(\pi-2\xi)\right] \\ & \quad

753: -\frac{A}{2\pi}

754: \sin(2\xi)\left[\,\text{Ci}(\pi+2\xi)+\text{Ci}(-\pi-2\xi) \right.  \\ 

755: & \quad \phantom{-A\sin(2x)[}\left.

756: -\text{Ci}(\pi-2x) - \text{Ci}(-\pi+2x) \right] \\

757: & \quad  -2B \, S(\xi)\,\cos \xi\,\sin \xi \qquad (n=2)

758: \end{split}

759: \end{equation}

760: with erfi the imaginary error function and Si and Ci the sine and

761: cosine integral functions, respectively \cite{abramowitz-stegun}. The

762: result given for $\hat \tau_C$ is for the special case $n=2$.  Both

763: the profiles of Eq.(\ref{eq:hills}) and the corresponding solutions of

764: Eq.(\ref{eq:wind}) given in Eq.(\ref{eq:solutions}) are shown in

765: Fig.~\ref{fig:hills}.

766: 

767: \begin{figure}[t]

768: \psfrag{x}{$x$}

769: \psfrag{h}{\mbox{}\;\; $f$} 

770: \psfrag{t}{$\!\!\!\!\tau/\tau_0$} 

771: \begin{center}

772:  

773: \includegraphics[width=\columnwidth]{hills2.eps} 

774: \caption{\emph{Lower curves:} The normalized profiles of

775: Eq.(\ref{eq:hills}), the Lorentzian $f_L$ (dashed), Gaussian $f_G$

776: (solid), and cosine $f^2_C$ (dotted). \emph{Upper curves:} The

777: corresponding surface shear stresses $\tau(x)/\tau_0$ from

778: Eq.(\ref{eq:solutions}) with $A \varepsilon = 0.8$, $B=0.25$.}

779: \label{fig:hills} 

780: \end{center} 

781: \end{figure} 

782: 

783: \begin{figure}[tb]

784:     %\psfrag{T}[l]{$\hat \tau_G$, $f_G$} 

785:     \psfrag{X}{$\xi$}

786:     \psfrag{f}{$f_G$} 

787:     \psfrag{t1}{$\hat \tau_{\rm sym}$}

788:     \psfrag{t2}{$\hat \tau_{\rm asy}$} 

789:     \psfrag{t3}{$\hat \tau_G$}

790: \begin{center}  

791:     \includegraphics[width=\columnwidth]{sym_asym.eps}

792:     \caption{The symmetric and asymmetric parts $\hat \tau_{\rm sym}$

793:     and $\hat \tau_{\rm asy}$ of the shear stress perturbation $\hat

794:     \tau_G$ of Eq.(\ref{eq:solutions}) for the Gaussian profile $f_G$

795:     of Eq.(\ref{eq:hills}) with $A\varepsilon=0.8$ and $B=0.25$. Note

796:     the small windward shift of the maximum of the shear stress with

797:     respect to the crest of the heap caused by the asymmetric

798:     contribution proportional to $B$.}

799: \label{fig:sym_asym}

800: \end{center}

801: \end{figure}

802: 

803: The plots of $\tau$ show that as a rule of thumb one can estimate the

804: relative magnitude of the shear stress perturbation at the top of the

805: heap by $A\varepsilon$. The plots share several crucial properties not

806: present in the affine approximation $\hat\tau\propto h$ of the zeroth

807: order model. At the tails of the profiles in Fig.~\ref{fig:hills}, the

808: shear stress decreases below its asymptotic value $\tau_0$ on the

809: plane.  This effect is particularly pronounced for the profile $f^n_C$

810: that has a discontinuity in its second derivative.  Further, as a

811: consequence of the second term in Eq.(\ref{eq:wind}), the surface

812: shear stress is not symmetric even for symmetric profiles like those

813: in Eq.(\ref{eq:hills}). Fig.~\ref{fig:sym_asym} displays the symmetric

814: and asymmetric parts $\hat \tau_{\rm sym}$ and $\hat \tau_{\rm asy}$

815: of the shear stress perturbation $\hat \tau_G$ for the Gaussian

816: profile $f_G$, separately. The asymmetric contribution to $\hat \tau$

817: is small compared to the symmetric one. For the profile $f_L$ the

818: corresponding shift $\delta x_\tau$ of the location of the maximum of

819: the shear stress with respect to that of the maximum of $f_L(x)$ can

820: be calculated analytically,

821: \begin{equation}\label{eq:shift}

822: \delta x_\tau/\! L=2\,(1 + B^2)^{1/2}\,\sin [\arctan (B)/3] -B

823: \;.

824: \end{equation}

825: It is indeed found to be very small, because $B$ is small and thus

826: $\delta x_\tau/\!  L\approx -B/3$ typically amounts to a length of

827: about a few percent of the total heap length. Nevertheless, it is a

828: crucial element in the modeling of aeolian sand transport, as will now

829: be demonstrated.

830: 

831: \begin{figure}[t]

832:   \psfrag{v}[lr]{$v(x)$} \psfrag{x}{$x/L$}

833: \begin{center}   

834:   \includegraphics[width=\columnwidth]{v_s2.eps}

835:   \includegraphics[width=\columnwidth]{v_s3.eps} 

836: \caption{The position

837:   dependent surface migration velocity $v(x)$ in arbitrary units

838:   according to Eqs.(\ref{eq:v_def}), (\ref{eq:wind}) with $A=4$,

839:   $B=0.25$. \emph{Upper part:} For the profile $f^2_C$ (gray) and

840:   varying aspect ratios $\varepsilon = 0.01$ (dashed), $0.1$ (solid),

841:   $0.19$ (dotted) and the local flux relation Eq.(\ref{eq:q0}).

842:   \emph{Lower part:} For the profile $f^{1.65}_C$ (gray) with

843:   $\varepsilon=0.1$ (solid), $0.05$ (dashed) and the non--local flux

844:   relation Eq.(\ref{eq:flux}). For simplicity, $q_s$ was represented

845:   by Eq.(\ref{eq:bagnold}) and $l_s\approx 0.1$ was taken constant.}

846: \label{fig:v}

847: \end{center}

848: \end{figure} \noindent

849: 

850: For a qualitative estimate of the effects of Eq.(\ref{eq:wind}) onto

851: the sand transport over a dune, it is useful to consider once again

852: the local zeroth order model for aeolian sand transport,

853: Eq.(\ref{eq:q0}), i.e.\ a completely saturated flux $q=q_s(\tau)$ with

854: $q_s$ given by Eq.(\ref{eq:bagnold}). (Below, we will show that this

855: is asymptotically valid on large dunes in strong winds.) The distinct

856: features of Eq.(\ref{eq:wind}) that are missing in the zeroth order

857: approximation for the shear stress, Eq.(\ref{eq:tau0}), are then

858: easily seen to have potentially profound effects on the shape

859: evolution.  First, due to the depression of the shear stress at the

860: tails of the profiles, deposition rather than erosion may occur at the

861: windward foot. Secondly, due to the asymmetric contribution in

862: $\tau(x)$ there can be a net deposition on a symmetric heap of sand.

863: In particular, the shift of the position of the maximum shear stress

864: with respect to the top of the heap allows deposition at the top of

865: the heap. For an initially flat heap of sand there is thus the

866: possibility of a steepening of the windward slope and mass

867: growth. This implies that a plane sand surface is unstable against

868: modulations. To illustrate this effect, we used

869: Eq.(\ref{eq:solutions}) to calculate the migration velocity $v(x)$ of

870: a cosine--shaped heap of sand $f^2_C(x)$ according to

871: Eqs.(\ref{eq:v_def}), (\ref{eq:bagnold}), (\ref{eq:q0}), and

872: (\ref{eq:wind}). The latter is shown in the upper part of

873: Fig.~\ref{fig:v}.  The decrease of $v(x)$ on the lee side reveals the

874: anticipated self--amplifying tendency of the unstable lee slope to

875: steepen, since $v'<0$ implies that $v$ increases with height. This

876: gives rise to the formation of the slip face.  More interestingly,

877: Eq.(\ref{eq:wind}) renders $v(x)$ approximately constant over almost

878: the whole windward side if $\varepsilon=H\!/\!L$ is close to a certain

879: value determined by the values of the coefficients $A$ and $B$ in

880: Eq.(\ref{eq:wind}).  Slightly better constancy can be achieved for

881: slightly lower $n$ (with slightly larger $\varepsilon$) but \emph{not}

882: for the profiles $f_G$ and $f_L$, for which $v(x)$ is always

883: non--uniform.  Let us finally consider the dashed and dotted lines in

884: the upper part of Fig.~\ref{fig:v}. They were obtained for a smaller

885: and a larger aspect ratio and represent a steepening ($v'<0$) and

886: flattening ($v'>0$) of the windward side, respectively. In other

887: words, profiles with larger/smaller windward slopes are driven towards

888: the solution with constant windward $v(x)$ and a stable optimum

889: windward slope different from zero.  Altogether, Fig.~\ref{fig:v} thus

890: suggests that the coupled Eqs.(\ref{eq:bagnold}) and (\ref{eq:wind})

891: drive a heap of sand towards a ``dune'' with a cosine--like windward

892: profile of a preferred aspect ratio, and a slip face on the lee side.

893: 

894: 

895: 

896: From the qualitative analysis presented so far, it is not yet obvious

897: that this process converges to a translation invariant steady--state

898: solution.  Several previous studies using similar descriptions either

899: did not scrutinize the long time behavior of their models

900: \cite{Wippermann86,van_dijk-arens-boxel:99}, or failed to obtain

901: stable dunes \cite{zeman-jensen:88,Stam97}.  To obtain a consistent

902: general model for dune formation under general influx and wind

903: conditions, the present wind model Eqs.(\ref{eq:hunt}),

904: (\ref{eq:wind}) has to be appropriately adapted to situations

905: with flow separation above slip faces. And, most importantly, the

906: saturated--flux approximation Eq.(\ref{eq:q0}) of the ``zeroth order''

907: model has to be abandoned. These steps will be discussed in the

908: following subsection and in Section \ref{sec:saltation}, where we will

909: also explain the lower part of Fig~\ref{fig:v}. This will complete the

910: definition of the minimal model.  Its numerical solutions will be

911: presented in Section \ref{sec:solution}.

912: 

913: 

914: \subsection{Flow separation}\label{sec:separation}

915: The wind model as discussed so far works fine for smooth heaps with

916: gentle slopes. However, as we have already mentioned, its application

917: to dune profiles with slip faces and sharp brink lines is not

918: straightforward. The perturbative turbulent boundary layer approach

919: leading to Eq.(\ref{eq:hunt}) does not account for flow separation, a

920: phenomenon that occurs at sharp edges and steep slopes to prevent an

921: extreme bending of the streamlines \cite{landau-lifshitz:fm}. (For

922: some of the technical terms involved in this section, the reader is

923: referred to Fig.~\ref{fig:separation}.)  Instead of bending the

924: streamlines around sharp edges, re--circulating eddies separate from

925: the (on average) laminar main flow, thereby creating an effective

926: envelope that diverts the main flow on a smooth detour around the

927: obstacle. See Figs.~\ref{fig:separation} and \ref{fig:fluent} for a

928: schematic sketch and a numerical calculation of a typical velocity

929: field, respectively. Fortunately, it turns out that dune formation and

930: migration do not in general depend very sensitively on the details of

931: this complicated process. Or in other words, there is a large number

932: of interesting problems of aeolian sand transport for which these

933: details are largely irrelevant, and for which their somewhat realistic

934: physical representation would create a huge overhead in complexity

935: (especially in $3d$) to an otherwise tractable problem. It was

936: therefore suggested earlier \cite{zeman-jensen:88} that for the

937: purpose of calculating the shear stress on the windward side of a

938: dune, one may to a good approximation represent flow separation on the

939: lee side by the following heuristic method. A wind model such as

940: Eq.(\ref{eq:wind}) restricted to smooth, gently sloped objects is

941: applied to the envelope

942: \begin{equation}\label{eq:envelope}

943: \tilde h(x)=\text{max} \{h(x),s(x)\}

944: \end{equation}

945: of a dune $h(x)$ and a phenomenologically defined separation bubble

946: $s(x)$. This disregards the fact that the separating streamline does

947: not represent a solid boundary of the same roughness as the original

948: object, but the corresponding errors are expected to be small.

949: Typically, one wants $s(x)$ to be a mathematically simple smooth

950: continuation of the dune profile. It is, however, crucial that the

951: latter respects some major phenomenological properties of flow

952: separation \cite{badbubble}.  Although this is by no means a rigorous

953: procedure, one can test its predictions for selected cases against

954: numerical solutions of various turbulence models. The hope is that via

955: this approach, one can eventually get a qualitative understanding of

956: the mechanisms and phenomena involved in dune formation and migration,

957: leaving certain quantitative aspects to a more elaborate (and much

958: more laborious) future analysis.

959: 

960: \begin{figure}[t]

961:   \psfrag{s}[b]{streamlines} 

962:   \psfrag{sb}[cb]{separation bubble}

963:   \psfrag{ss}[cb]{separating streamline}

964:   \psfrag{d}{dune}

965: \begin{center}

966:   \includegraphics[width=\columnwidth]{separation.eps}

967:   \caption{Sketch of the central slice of a barchan dune and the

968:   separation bubble. The shear stress on the windward side of the dune

969:   is calculated by applying Eq.(\ref{eq:wind}) to the

970:   phenomenologically defined envelope of the dune and the separation

971:   zone.}

972: \label{fig:separation}

973: \end{center}

974: \end{figure}

975: 

976: In the spirit of the minimal model we want to parameterize the

977: separating streamline $s(x)$ in the simplest form that obeys

978: physically motivated boundary conditions at its detachment and

979: reattachment points $x_d$ and $x_r$. At detachment, the slope of the

980: separating streamline must match the slope of the dune. Moreover, also

981: the curvature must be continuous there, since discontinuities in

982: curvature are detected by Eq.(\ref{eq:wind}) and cause kinks in $\tau$

983: and discontinuous steps in the erosion/deposition as is e.g.\ the case

984: for the profile $f^2_C$. For the reattachment point, there are no

985: comparable restrictions to the slope and curvature, since the

986: separation bubble is not sharply defined there, and the model aims at

987: a realistic description of the conditions in the wake region only

988: insofar as they affect the shear stress on the \emph{windward}

989: side. On the lee side, inside the separation bubble, the shear stress

990: can simply be set to zero \cite{rainout}, since it is typically below

991: the threshold for aeolian sand transport.  Therefore, the choice of

992: the reattachment matching condition is a matter of convenience rather

993: than physical significance in the present model. However, we want

994: $s(x)$ to reproduce some common phenomenological knowledge about flow

995: separation. First, from many numerical calculations it is known that,

996: at high Reynolds numbers, the turbulent boundary layer reattaches at a

997: distance of about $6H$ after a backward--facing step of height

998: $H$. Secondly, it has often been observed experimentally that in

999: strongly turbulent flows over hills and symmetric triangular

1000: obstacles, flow separation sets in if the backward slope exceeds an

1001: angle of about $14^\circ$. Although, in both cases the exact numerical

1002: values depend on various factors such as the surface roughness and the

1003: Reynolds number, they shall be treated as fixed phenomenological

1004: constants at the present stage. A model that fulfills all the above

1005: requirements is a third order polynomial with continuous slopes at the

1006: boundaries and a maximum negative slope of $\tan 14^\circ$. The

1007: boundary conditions

1008: \begin{equation}\label{eq:bc}

1009: \begin{split}

1010: s(x_d) & = h_d \equiv h(x_d) \qquad \; s(x_r) = 0 \\

1011: s'(x_d) & = h_d' \equiv h'(x_d) \qquad s'(x_r) = 0 \;. \\

1012: \bar s' & \equiv \text{max}\{-s'(x)\} = \tan 14^\circ = 0.25

1013: \end{split}

1014: \end{equation}

1015: constrain the third order bubble parameterization to be of the form

1016: \begin{equation}\label{eq:sepbubble}

1017:   s(z)= (2 h_d + h_d' L_b)z^3 -(3 h_d + 2 h_d' L_b) z^2 + h'_d L_bz +

1018:   h_d

1019: \end{equation}

1020: with $z\equiv (x-x_d)/L_b \in [0,1]$.  With the further abbreviation

1021: $\nu\equiv h_d'/\bar s'$, we can express the length $L_b\equiv

1022: x_r-x_d $ of the bubble as

1023: \begin{equation}\label{eq:sx}

1024: L_b =\frac{3h_d}{h_d'} \frac{1 - \nu - \sqrt{1 +  \nu}}{3- \nu} \approx

1025:   \frac{3h_d}{2\bar s'}\left(1+ \frac  \nu4 + \frac{ \nu^2}8\right) \;,

1026: \end{equation}

1027: where the final approximation for small $h_d'$ is sufficient for our

1028: purpose (and numerically better behaved as the exact expression). A

1029: subtlety of such a separation bubble parameterization is the fact that

1030: the slope at $x_d$ determines the length of the bubble, which in turn,

1031: via Eq.(\ref{eq:wind}) influences the curvature at $x_d$. In other

1032: words, the presence of the bubble introduces a non--local feed--back

1033: between the slope and the curvature at the brink, which we believe is

1034: physically reasonable. In Fig.~\ref{fig:bubbles} we give some examples

1035: of separation bubbles for different boundary slopes $h_d'$, while

1036: Fig.~\ref{fig:tau_bubble} illustrates the application of the above

1037: discussion for the calculation of the shear stress.  It shows an

1038: example for a dune profile $h(x)$ with slip face and the separation

1039: bubble $s(x)$, together with the shear stress $\tau(x)$ resulting from

1040: Eq.(\ref{eq:wind}) if $h$ is replaced by the envelope $\tilde h$.

1041: 

1042: \begin{figure}

1043:    % \psfrag{L in h0}[t]{$(x-x_d)/h_d$}

1044:    % \psfrag{s in h0}{$s(x)/h_d$}

1045:      \psfrag{X}[t]{$(x-x_d)/h_d$}

1046:      \psfrag{T}{$\frac{\displaystyle s(x)}{\displaystyle h_d}$}

1047:  \begin{center}

1048:     \includegraphics[width=\columnwidth]{bubbles.eps} 

1049:     \caption{Separation bubbles with a maximum negative slope of 0.25 

1050:     according to Eq.(\ref{eq:sx}) for varying initial slopes $-0.25 \leq 

1051:     h_d' \leq 0.25$. (The aspect ratio of the plot was stretched for 

1052:     presentation.)}

1053:       \label{fig:bubbles} 

1054: \end{center}

1055: \end{figure}

1056: 

1057: 

1058: \begin{figure}

1059:     \psfrag{x in L}[t]{$x / L$}

1060:     \psfrag{h in h0, tau in tau0}{$h / H$, $\tau / \tau_0$}

1061:     \psfrag{hxxxxxx}[lc]{\footnotesize $h$}

1062:     \psfrag{sxxxxxx}[lc]{\footnotesize $s$}

1063:     \psfrag{tauxxxx}[lc]{\footnotesize $\tau$}

1064:  \begin{center}   

1065:     \includegraphics[width=\columnwidth]{sepbub_tau.eps}

1066:     \caption{The windward profile $h(x)$ of a dune with slip face and

1067: the separation bubble $s(x)$ form together a smooth effective

1068: obstacle, defined by the envelope $\tilde h (x)$. To calculate the shear stress

1069: $\tau(x)$ on the windward side of the dune, $\tilde h$ is substituted

1070: for $h$ in Eq.(\ref{eq:wind}). In the region of re--circulation

1071: the surface shear stress $\tau$ is set to zero \cite{rainout}. 

1072: Without the separation bubble, $\tau(x)$ would develop a sharp

1073: singularity at the brink.}

1074: \label{fig:tau_bubble}

1075: \end{center}

1076: \end{figure}

1077: 

1078: We have performed several series of numerical fluid dynamics

1079: calculations in $2d$ and $3d$ with the commercial fluid dynamics

1080: solver Fluent~5 \cite{Fluent5} using the $k\epsilon$ and

1081: \emph{large--eddy} turbulent closure models to confirm the general

1082: picture outlined above and our particular implementation of the

1083: separation bubble. The differences between numerical and theoretical

1084: predictions for the shear stress on the windward side of various

1085: dune-- and heap--like objects in $2d$ and $3d$ were quantitatively

1086: small and not more significant than other neglected terms. Moreover, a

1087: comparison of predictions obtained from Eq.(\ref{eq:sepbubble}) with

1088: wind measurements on a barchan dune in Brazil \cite{sauermann:phd}

1089: showed good agreement. Therefore, we are confident that the proposed

1090: mathematical description of the wind shear stress captures the

1091: relevant aspects in the spirit of the minimal model. As an example for

1092: the numerical fluid dynamics calculation, we show in

1093: Fig.~\ref{fig:fluent} the flow velocity in the symmetry plane of a

1094: $3d$ barchan dune obtained with Fluent~5 \cite{Fluent5}. The wind is

1095: blowing from left to right. The boundaries were chosen to be periodic

1096: in the transverse direction. At the influx boundary, the velocity was

1097: fixed by imposing a logarithmic velocity profile. The wind profile at

1098: the outflux boundary is not known \emph{a priori}. Although, for high

1099: Reynolds numbers the latter is expected to affect the solution only

1100: close to the boundary, it is well known that different choices for the

1101: outflux boundary condition as well as different discretization schemes

1102: may lead to quantitatively different results \cite{wesseling:2000}.

1103: Here, we chose to set the derivative of the velocity normal to the

1104: outflux boundary to zero. The surface profile was represented as a

1105: solid boundary with constant roughness length. Finally, along the top

1106: boundary we imposed the velocity of the undisturbed logarithmic inflow

1107: profile. The whole calculation was performed on a grid that had an

1108: exponentially growing mesh size in the vertical direction. A

1109: considerable grid refinement was necessary in the wedge--like region

1110: of the separation bubble close to the brink.

1111:  

1112: \begin{figure}[tb]

1113:   \begin{center}

1114:    

1115:     \includegraphics[width=\columnwidth]{velocity_cut_black.eps}

1116:     \caption{Cut along the symmetry plane of a $3d$ barchan dune. The

1117:       velocity vectors calculated numerically with a commercial 

1118:       fluid dynamics solver \cite{Fluent5} clearly display the flow 

1119:       separation at the brink and a large eddy in the wake region.}

1120:       \label{fig:fluent}

1121: \end{center}

1122: \end{figure}

1123: 

1124: These remarks complete the first task of constructing a model for the

1125: calculation of the wind shear stress on a given dune profile as

1126: outlined in Eq.(\ref{eq:task1}). By deriving the linear

1127: Eq.(\ref{eq:wind}) for the shear stress and combining it with

1128: the heuristic separation bubble, we have obtained an approximate but

1129: numerically extremely efficient model for the wind shear stress on

1130: dunes. This is a crucial step in the construction of a minimal model

1131: of aeolian sand dunes, since the enormous complexity of the turbulent

1132: air flow over structured terrain otherwise severely restricts the

1133: possible applications of the model.

1134: 

1135: Going back to the upper part of Fig.~\ref{fig:v} with the above

1136: discussion in mind, we can re--interpret this figure in order to

1137: anticipate the behavior of the surface migration velocity $v(x)$ of a

1138: dune with slip face. If, for qualitative purposes, $f^2_C$ is

1139: interpreted as the envelope of a dune and its separation bubble, we

1140: can conclude that the slip face must be located near the sharp drop of

1141: $v(x)$ slightly upwind from the top of the envelope. This is indeed

1142: consistent with observations for large dunes. Together with the good

1143: representation of the windward profiles of large dunes

1144: \cite{sauermann-etal:2000} by $f^n_C$ ($n\approx 2$), it suggests that

1145: the given description becomes qualitatively correct in the limit of

1146: large dune sizes.  The next section is devoted to the discussion of

1147: important subtleties related to the fact that dunes are not typically

1148: in this limit.

1149: 

1150: 

1151: \section{Sand flux}\label{sec:saltation}

1152: As outlined in Eq.(\ref{eq:task2}), the second task in the

1153: specification of the minimal model is to find a prescription for

1154: calculating the sand flux $q(x)$ for a given topography $h(x)$ and

1155: shear stress $\tau(x)$. So far, we have been using the local

1156: saturated--flux approximation Eq.(\ref{eq:q0}) in our qualitative

1157: arguments. However, a closer look at the predictions obtained within

1158: this approximation reveals a number of inconsistencies.  First, as we

1159: have already noted in the discussion of Fig.~\ref{fig:v}, the use of

1160: Eq.(\ref{eq:q0}) together with the complete wind model of

1161: Section~\ref{sec:wind} leads to the odd prediction of deposition at

1162: the windward foot of an isolated heap or dune, where the shear stress

1163: decreases. This defect of Eq.(\ref{eq:q0}) has been noticed in the

1164: literature before (see e.g.\ Refs.\

1165: \cite{Wiggs96,sauermann-kroy-herrmann:2001}). Previous numerical

1166: studies tried to avoid this problem by focusing onto the short time

1167: behavior and by introducing \emph{ad hoc} heuristic methods such as a

1168: ``smoothing operator'' \cite{Wippermann86} or an ``adaptation length''

1169: \cite{van_dijk-arens-boxel:99}. The reason for the problem is that the

1170: saturated--flux approximation breaks down at a boundary

1171: ground/sand. As another shortcoming, we want to mention that the model

1172: as discussed so far predicts a universal scale invariant dune shape

1173: with a brink that is displaced slightly upwind from the maximum of the

1174: envelope, leading always to a positive slope at the brink.  A glance

1175: at a real dune field proves that the latter is not always the case and

1176: careful measurements \cite{sauermann-etal:2000} have revealed

1177: systematic deviations from scale invariance.  Though less obvious, it

1178: turns out that the reason for this discrepancy lies again in the

1179: saturated--flux approximation. Both mentioned problems are thus

1180: naturally resolved by introducing a slightly more general sand

1181: transport law that allows for saturation transients.

1182: 

1183: \subsection{Saturation transients}

1184: The saturated--flux approximation Eq.(\ref{eq:q0}) assumes that the

1185: flux is everywhere equal to the equilibrium transport capacity $q_s$

1186: of the wind. However, due to variable wind or sand conditions, the

1187: actual sand flux $q$ is in general different from $q_s$. These

1188: deviations are called saturation transients, because they quickly

1189: relax to zero under homogeneous conditions. We have recently

1190: demonstrated \cite{sauermann-kroy-herrmann:2001} that this relaxation

1191: occurs within a characteristic length scale, called the

1192: \emph{saturation length} $\ell_s$, which is related to (but distinct

1193: from) the mean saltation length of the grains. It was moreover shown

1194: how the introduction of saturation transients cures the problem of

1195: deposition at the windward foot of an isolated sand dune.  Here, we

1196: only summarize the most pertinent results of this earlier development

1197: in order to demonstrate how a size dependence of the dune shape

1198: naturally results as a consequence of saturation transients.

1199: 

1200: The sand transport model of Ref~\cite{sauermann-kroy-herrmann:2001} is

1201: based on a mean--field like description of saltation. It models a

1202: typical grain that is accelerated by friction with the air and slowed

1203: down by dissipative interactions with the bed. The average properties

1204: of the complicated splash process

1205: \cite{Anderson91,Nalpanis93,Rioual2000} are subsumed into two

1206: dimensionless parameters, an effective restitution coefficient

1207: $\alpha$ for collisions with the bed, and a kinetic coefficient

1208: $\gamma$ that characterizes the relaxation of the density of saltating

1209: grains to its saturated value. Together with an effective height for

1210: the wind--grain interaction that enters only logarithmically, these

1211: are the only phenomenological parameters of the model. They have been

1212: determined by a comparison with experiments and grain scale

1213: simulations.  Formally, the model consists of two coupled differential

1214: equations for mass and momentum conservation, and a modified turbulent

1215: closure relation that accounts for the feedback of the saltating

1216: grains on the wind velocity.

1217: 

1218: 

1219: For the present purpose, the model can be simplified by taking

1220: advantage of the fact that the prevailing conditions in applications

1221: to dunes are typically well described by the steady--state

1222: ($\partial/\partial t\simeq 0$) version. Further, the relaxation of

1223: the typical sand transport velocity can be assumed to be fast compared to

1224: the variations in the density of mobilized grains in the saltation

1225: layer. Approximating the latter by its saturated value for the

1226: calculation of the effective wind speed $u_{\rm eff}$ via the modified

1227: turbulent closure, one can decouple the mass and momentum conservation

1228: equations.  The whole model can then in a reasonable approximation be

1229: reduced to a single differential equation

1230: \begin{equation}

1231:   \label{eq:flux}

1232:   \ell_s\partial q/\partial x = q( 1-q/q_s)  

1233: \end{equation}

1234: for the sand flux $q(x)$.  The shear stress dependent parameters

1235: \begin{equation}\label{eq:ls}

1236:   \ell_s= l/(\tau/\tau_t-1)\;, \qquad q_s = \rho_s u_s 

1237: \end{equation}

1238: are immediately identified as the saturation length and the saturated

1239: flux, respectively.  The equation for $q_s$ generalizes

1240: Eq.(\ref{eq:bagnold}) to arbitrary wind speeds. In the following we

1241: specify the explicit expressions for both quantities as they result

1242: from the sand transport model of

1243: Ref.~\cite{sauermann-kroy-herrmann:2001}, but the structure of

1244: Eq.(\ref{eq:flux}) is thought to be more general and independent of

1245: the precise form of Eq.(\ref{eq:ls}). Again, $\tau(x)$ is the position

1246: dependent shear stress discussed in Section~\ref{sec:wind} and $\tau_t

1247: \approx 0.1\; \text{kg\,m}^{-1}\text{s}^{-1}$ is the estimated impact

1248: shear stress threshold that corresponds to a critical shear velocity

1249: $u_{*t}\approx 0.28 \;\text{m\,s}^{-1}$ \cite{Owen64}.  (For

1250: simplicity, we do not introduce the additional threshold for purely

1251: aerodynamic entrainment here, but allow instead for a small residual

1252: influx even if the latter is nominally zero.) To make the underlying

1253: structure of the model more palpable, we have expressed $\ell_s$ in

1254: terms of another characteristic length scale $l\equiv 2\alpha

1255: u_s^2/(g\gamma)$, which (up to a numerical factor) is the average

1256: saltation length of the grains. The latter --- but not $\ell_s$ ---

1257: must always be considerably smaller than the dune length for the model

1258: to be applicable.  Further, we have decomposed $q_s$ into the

1259: saturated density $\rho_s = 2\alpha(\tau-\tau_t)/g$ and the effective

1260: sand transport velocity at saturation $u_s = u_{\rm eff} - \delta u$

1261: with $u_{\rm eff}$ the effective wind velocity that accelerates the

1262: grains, given by

1263: \begin{equation}\label{eq:ueff}

1264:  u_{\rm eff}\kappa\sqrt{\varrho_{a}} = 2\sqrt{\tau_t+(\tau

1265: -\tau_t)/\zeta} + (\ln\zeta'-2)\sqrt{\tau_t}\;. 

1266: \end{equation}

1267: By $g$ we have denoted the gravitational acceleration and from

1268: Ref~\cite{sauermann-kroy-herrmann:2001} we adopt the (approximate)

1269: numerical values $\alpha=0.35$, $\gamma=0.2$, $\zeta= 8$, $\zeta'=

1270: 200$, and $\delta u= 1.8\;$m/s for the lag velocity of the grains. We

1271: note that these numerical values are not completely independent of

1272: each other and of the mentioned value for the impact threshold

1273: $\tau_t$, due to the calibration of the sand transport model with

1274: experimental data \cite{sauermann-kroy-herrmann:2001}.  For

1275: convenience we show a plot of the saturation length $\ell_s$ obtained

1276: with these values in Fig.~\ref{fig:ls}.  This completes the definition

1277: of the sand transport part of the minimal model on gently sloped

1278: ground.

1279: 

1280: \begin{figure}[tb]

1281:  \psfrag{T}{$\tau/\tau_t$}

1282:  \psfrag{L}{$\ell_s \;\text{[m]}$}

1283:   \begin{center}

1284:     \includegraphics[width=\columnwidth]{ls.eps} 

1285:     \caption{The

1286:     saturation length $\ell_s$ in meters as a function of the shear

1287:     stress exerted by the wind onto the sand bed. This function 

1288:     sets the natural length scale for dunes and heaps.}

1289: \label{fig:ls}

1290: \end{center}

1291: \end{figure}

1292: 

1293: 

1294: \subsection{Consequences}

1295: Before we complete the general model definition by a brief paragraph

1296: on slip faces, we want to point out some implications of the model as

1297: developed so far.  First, note that the full expression for $q_s$

1298: given in Eq.(\ref{eq:ls}), contains Eq.(\ref{eq:bagnold}) as a

1299: limiting case for strong winds but is better approximated by

1300: $q_s\propto \tau-\tau_t$ for moderate wind speeds. For weak flux

1301: gradients and strong winds, one may set the left hand side of

1302: Eq.(\ref{eq:flux}) to zero, leading to $q=q_s$. This is

1303: typically the case on most of the windward slope of a large dune,

1304: where the left hand side of Eq.(\ref{eq:flux}) can roughly be

1305: estimated by $\ell_s q_s/L \ll q_s$.  The local saturated--flux

1306: approximation with Eq.(\ref{eq:bagnold}) for $q_s$, which we have

1307: applied throughout our qualitative discussion so far, is thus

1308: asymptotically valid for large dunes and strong winds (except near the

1309: windward foot of an isolated dune). This is what one might have

1310: expected in the first place, and the reader may wonder at this point

1311: how the saturation transients and their characteristic length scale

1312: $\ell_s$ can have the claimed importance. How can $\ell_s$ affect the

1313: shape of a dune that is typically about two orders of magnitude

1314: larger? This apparent puzzle is now easily resolved by going back to

1315: Figs.~\ref{fig:hills}, \ref{fig:sym_asym} and Eq.(\ref{eq:shift}) and

1316: by observing that the symmetry breaking shift $\delta x_\tau$ of the

1317: location of the maximum of the shear stress with respect to that of

1318: the maximum of the height profile (or envelope) that is responsible

1319: for the finite windward slope and growth of dunes, is also of the

1320: order of some per cent of the total dune length. In summary, the

1321: longitudinal profile of dunes and heaps is determined by the

1322: competition of two quantitatively small but qualitatively crucial

1323: effects, one related to turbulent wind flow and the other to sediment

1324: transport. This may be the reason, why its explanation proved elusive

1325: for a long time.

1326: 

1327: To get a qualitative idea of the consequences of the introduction of

1328: the generalized non--local flux law in Eq.(\ref{eq:flux}) as

1329: replacement for Eq.(\ref{eq:q0}), we want again to go back to our

1330: discussion of the surface migration velocity of the cosine shaped heap

1331: $f^n_C$ in Fig.~\ref{fig:v}. Let us for the moment adopt a crude

1332: approximation and replace the expression for $q_s$ given in

1333: Eq.(\ref{eq:ls}) by its simpler limiting form $q_s \propto \tau^{3/2}$

1334: introduced in Eq.(\ref{eq:bagnold}). We also neglect the variation of

1335: $\ell_s$ on the dune and replace it by a (fine--tuned) constant

1336: $\ell_s\approx 0.1\, L$.  With an influx (about $0.7\, q_s$ for the

1337: solid and $0.8\, q_s$ for the dashed line in the lower part of

1338: Fig.~\ref{fig:v}) one can thus achieve a fairly constant surface

1339: velocity over the \emph{whole} length of a cosine shaped heap.  Again

1340: the constancy is slightly better for $n<2$ than for $n=2$. It is

1341: further improved by reducing the slope of the heap well below the

1342: optimum windward slope of the dune obtained for $\ell_s\to0$, as seen

1343: from a comparison of the solid line and the dashed line. We also note

1344: in passing that the influx needed to maintain the shape is increasing

1345: with decreasing slope. Together, the plots in Fig.~\ref{fig:v}

1346: confirms our claim that even an $\ell_s\ll L$ may visibly affect the

1347: overall shape of aeolian dunes.  Although, for the example shown in

1348: the lower part of Fig.~\ref{fig:v}, the saturation length is only

1349: about $1/30$ of the heap length, the slip face instability is

1350: evidently completely washed out. Altogether, this strongly suggests

1351: the existence of translation invariant cosine shaped heap solutions

1352: for the model.  The ultimate proof will be provided by the numerical

1353: results presented in Section~\ref{sec:solution}, where the full form

1354: of $q_s$ and $\ell_s$ according to Eq.(\ref{eq:ls}) will be used, but

1355: the present crude approximation already illustrates the main point,

1356: and also demonstrates that the behavior is a generic consequence of

1357: Eq.(\ref{eq:ls}) and insensitive to the detailed form of the

1358: parameters $\ell_s(x)$ and $q_s(x)$ that may phenomenologically be

1359: somewhat different from the model prediction without affecting our

1360: general conclusions.

1361: 

1362: An immediate consequence of the foregoing discussion is the existence

1363: of a minimum dune size. For small enough dunes, the slip face

1364: instability is washed out by the saturation transients. One may also

1365: arrive at this conclusion from an analysis of heaps. To this end we

1366: observe that the value of the saturation length $\ell_s$ is a property

1367: of the wind velocity and the saltation kinetics and depends on the

1368: topography only indirectly through the variable shear stress

1369: $\tau$. Moreover, it is apparent from Fig.~\ref{fig:ls} that this

1370: dependence becomes weak for strong winds. On the other hand, the

1371: symmetry breaking shift $\delta x_\tau$ is proportional to the

1372: absolute size of the heap (or envelope) and not directly dependent on

1373: the wind velocity. For the special profile $f_L$, this was verified

1374: analytically in Eq.(\ref{eq:shift}). As we have seen, a smooth heap

1375: can only be a translation invariant solution of the model if the lag

1376: (of order $\ell_s$) of $q(x)$ with respect to $q_s(x)$ and the shift

1377: $\delta x_\tau$ are fine--tuned to guarantee a vanishing erosion rate

1378: at the top of the heap. From this we expect heaps to obey

1379: \begin{equation}\label{eq:heap_aspect}

1380: \ell_s \simeq \delta x_\tau \propto L \approx \text{const.}

1381: \end{equation}

1382: to a first approximation. This condition can only be fulfilled if the

1383: aspect ratio $\varepsilon$ of heaps grows proportional to their size

1384: (i.e.\ roughly $\varepsilon\propto H$). Hence, in contrast to large

1385: dunes with slip face, for which we have argued that they are

1386: asymptotically scale invariant ($\varepsilon \sim$ const.), heaps must

1387: have a strongly size dependent aspect ratio. As a consequence,

1388: translation invariant heap solutions obviously cannot exist beyond a

1389: certain critical size. A slip face will develop when the shear stress

1390: on the lee side of the heap drops below the threshold value $\tau_t$,

1391: or at the latest, when the lee slope exceeds the critical slope for

1392: flow separation. This will be further analyzed in

1393: Section~\ref{sec:solution}.  Finally, we note that the steady--state

1394: flux of a heap can be estimated by the observation that the outflux is

1395: essentially determined by the strength of the reduction

1396: $\tau_0-\tau_{\rm min}$ of the shear stress at the lee end of the

1397: heap. According to Eq.(\ref{eq:scaling}), the latter is (for a given

1398: shape) proportional to the aspect ratio $\varepsilon$. For qualitative

1399: purposes, the outflux $q^{\rm out}$ may thus be estimated in the

1400: saturated flux approximation with

1401: Eqs.(\ref{eq:ls}),(\ref{eq:ueff}),(\ref{eq:scaling}) as

1402: \begin{equation}\label{eq:outflux}

1403: q_s^{\rm out} \propto \tau_{\rm min} -\tau_t \propto

1404: \varepsilon_c-\varepsilon \;,

1405: \end{equation}

1406: where we have assumed $\tau_{\rm min}/\tau_t \lesssim 2$ (fulfilled

1407: for moderate wind speeds and/or heaps near the critical heap size)

1408: to linearize the expression for $q_s(\tau)$ given in

1409: Eq.(\ref{eq:ls}). Here, $\varepsilon_c \propto \tau_0-\tau_t$ is the

1410: critical aspect ratio for which the shear stress on the lee drops

1411: below the threshold and the outflux vanishes. Note that the latter

1412: increases with increasing shear stress whereas the heap length

1413: decreases according to Eqs.(\ref{eq:ls}), (\ref{eq:heap_aspect}). The

1414: effects of the two trends onto the critical heap mass could therefore

1415: partially cancel unless the lee slope exceeds the critical slope for

1416: flow separation.

1417: 

1418: \subsection{Slip face}\label{sec:slipface}

1419: We have argued above that for large heaps ($L\gg\ell_s$), aeolian

1420: sediment transport tends to increase the downwind slope until it

1421: reaches the angle of repose of the grains. At this point, any further

1422: increase of the lee slope initiates avalanches that restore a slope

1423: slightly below the static angle of repose and eventually create a slip

1424: face of a roughly uniform slope of about $32^\circ - 35^\circ$. Since

1425: the physical modeling of this process itself is not a major objective

1426: of the present contribution, we can choose between different possible

1427: implementations for this phenomenon. In $2d$ it is possible to

1428: represent the slip face as boundary condition for the sand transport.

1429: It is uniquely determined by its fixed uniform slope and mass

1430: conservation. However, with regard to a future generalization of the

1431: present $2d$ model to $3d$ we chose a more realistic implementation

1432: based on a widely--used avalanche model \cite{Bouchaud94}. The

1433: formulation of this model bears some close similarities with the sand

1434: transport model presented in the preceding paragraph, and thus

1435: suggests itself as a natural extension of the latter to the slip

1436: face. This completes the definition of the minimal model that will be

1437: solved numerically in the next section.

1438: 

1439: \section{Solution of the model}\label{sec:solution}

1440: Apart from the model definition, the preceding sections have provided

1441: some qualitative insights into the main mechanisms responsible for

1442: dune formation and migration. Now we are prepared to study numerically

1443: the quantitative predictions of the model. Again, we emphasize that we

1444: only can explore some major features of the model in the present

1445: report, leaving many interesting questions and more systematic and

1446: quantitative parameter studies for future work.

1447: 

1448: For convenience, the solution procedure of the minimal model is

1449: summarized as a flow chart in Fig.~\ref{fig:flowchart}. One starts

1450: with an initial profile $h(x,t=0)$ (typically $f_G$ or $f_C^2$),

1451: checks whether a separation bubble has to be added for the calculation

1452: of the shear stress, then obtains the latter from

1453: Eq.(\ref{eq:fourier_wind}) and uses the result as input for the

1454: iterative solution of the sand transport equation

1455: Eq.(\ref{eq:flux}). This finally gives the erosion/deposition needed

1456: to update the surface profile. Technically, Eq.(\ref{eq:fourier_wind})

1457: is implemented as a Fast--Fourier--Transform algorithm, and for the

1458: integration of Eqs.(\ref{eq:flux}) and (\ref{eq:mass}) an upwind

1459: discretization scheme is used. Simulation times can be reduced by

1460: using an adaptive time step.

1461: 

1462: \begin{figure}[tb]

1463:   \psfrag{initial}[c][c]{\tiny $h(x,t=0)$}

1464:   \psfrag{bubble}[c][c]{\small Eq.(\ref{eq:sepbubble})}

1465:   \psfrag{tau}[c][c]{\small Eq.(\ref{eq:fourier_wind})}

1466:   \psfrag{flux}[c][c]{\small Eq.(\ref{eq:flux})}

1467:   \psfrag{mass}[c][c]{$\genfrac{}{}{0pt}{1}{\text{Eq.(\ref{eq:mass})}\; \&}

1468:     {\text{\tiny avalanches}}$}

1469:   \begin{center}   

1470:     \includegraphics[width=\columnwidth]{flowchart.eps}

1471:     \caption{Solution of the minimal model.}

1472:       \label{fig:flowchart} 

1473: \end{center}

1474: \end{figure}

1475: 

1476:  

1477: \subsection{Steady--state shapes}

1478: The scheme of Fig.~\ref{fig:flowchart} can be iterated for different

1479: influx boundary conditions. For all of the numerical calculations

1480: presented below, we chose periodic boundary conditions. They are the

1481: natural choice for studies of the steady--state shapes.  To

1482: investigate the mass balance under prescribed influx conditions, on

1483: the other hand, one has to apply open boundary conditions. 

1484: 

1485: 

1486: \begin{figure}[tb]

1487:   \psfrag{h in m}[][90]{$h(x)$ [m]} 

1488:   \psfrag{x in m}{$x$ [m]}

1489:   \begin{center} 

1490: \includegraphics[width=\columnwidth]{heaps.eps}

1491: \includegraphics[width=\columnwidth]{dunes.eps}

1492:   \caption{Steady--state heaps (upper plot) and dunes (lower

1493:   plot). The aspect ratio is stretched for better visualization.}

1494:   \label{fig:shapes}

1495: \end{center}

1496: \end{figure}

1497: 

1498: 

1499: Fig.~\ref{fig:shapes} shows steady--state solutions of the model for

1500: initial profiles $f_G$ of different mass. These solutions are obtained

1501: for fixed wind conditions with parameters $A=3.2$ and $B=0.25$

1502: appropriate for the central slice of a $3d$ (symmetric) heap or of a

1503: barchan dune. The shear velocity $u_*=0.4$ m/s lies well above the

1504: impact threshold. (The situation very close to or below the threshold

1505: would need special attention.) As anticipated above, large dunes

1506: become asymptotically scale invariant. The asymptotic master curve for

1507: the windward profile is practically indistinguishable from the profile

1508: $f_C^n$ ($n\lesssim 2$), and the slope at the brink is indeed

1509: positive. Its average windward slope is inversely proportional to the

1510: value of the parameter $A$ given in Eq.(\ref{eq:a_b}). Due to the

1511: additional terms in the expression Eq.(\ref{eq:correction}) for the

1512: shear stress on dunes with a finite width, somewhat steeper average

1513: windward slopes are predicted for barchan dunes than for transverse

1514: dunes under identical influx and wind conditions. However, a detailed

1515: quantitative comparison is probably beyond the scope of the present

1516: semi--quantitative implementation. More important are the remarkable

1517: qualitative predictions of the model. In particular, the fact that

1518: dunes with slip face are only stable above a certain (wind dependent)

1519: critical size, whereas smooth steady--state heaps only exist below a

1520: critical size, deserves attention. We also note that the steady state

1521: is not always unique. There is a hysteretic regime, where the initial

1522: conditions can select one of two possible steady--state shapes and

1523: accordingly the masses for the two sets of profiles in

1524: Fig.~\ref{fig:shapes} are not all distinct. The largest heaps in the

1525: upper plot were obtained from flat initial profiles $f_G$, whereas the

1526: smallest dunes with slip face in the lower plot were obtained from

1527: steeper initial profiles $f_G$ of the same mass. Especially, the dune

1528: with a negative slope at the brink could only be obtained from steep

1529: initial conditions. Since under natural wind and sand conditions, the

1530: initial conditions themselves will generally be heaps or dunes close

1531: to the steady state, one can say that the model predicts a critical

1532: heap size for slip face formation and a critical dune size for slip

1533: face destruction. In both cases the slip face is finite as a

1534: consequence of flow separation.  The latter also allows a dune to be

1535: somewhat higher than a heap of the same mass, since its effective

1536: volume as seen by the average air flow is increased by the separation

1537: zone. As anticipated, the aspect ratio of the dunes is asymptotically

1538: constant, whereas it is strongly size--dependent for heaps. This

1539: effect can be seen more quantitatively in the representation of

1540: Fig.~\ref{fig:H_V}, where the height $H$ of the steady--state heaps

1541: and dunes is plotted versus the product $HL$ of their height and

1542: length $L$. Clearly, heaps are better described by $H\propto H L$ as

1543: predicted in Eq.(\ref{eq:heap_aspect}), whereas large dunes approach

1544: the scaling limit $H\propto \sqrt{H L}$.

1545: 

1546: \begin{figure}[tb]

1547:   \psfrag{flat}{\small flat}

1548:   \psfrag{steep}{\small steep}

1549:   \psfrag{H}[rl]{$H$ [m]}

1550:   \psfrag{V}[t][b]{$H L$ [m$^2$]}

1551:   \psfrag{infit}{\small $\propto H L$}

1552:   \psfrag{asfit}{\small $\propto \sqrt{H L}$}

1553:   \begin{center}   

1554:     \includegraphics[width=\columnwidth]{H_HLrl.eps}

1555:    \caption{Steady--state heights $H$ versus the product of the height

1556:   and length of the heaps and dunes. In the hysteretic

1557:   regime, flat and steep initial conditions have to be distinguished.}

1558:     \label{fig:H_V}

1559: \end{center}

1560: \end{figure}

1561: 

1562: \begin{figure}[tb]

1563:   \begin{center} 

1564: 

1565:   \psfrag{ustar035xxxx}{\tiny $u_*=0.35$}

1566:   \psfrag{ustar040xxxx}{\tiny $u_*=0.40$} 

1567:   \psfrag{ustar045xxxx}{\tiny $u_*=0.45$} 

1568:   \psfrag{ustar050xxxx}{\tiny $u_*=0.50$}

1569:   \psfrag{propto1hxxxx}{\tiny $\propto 1/H$}

1570:   \psfrag{propto1Lxxxx}{\tiny $\propto 1/L$} 

1571:   \psfrag{height in m}[t][b]{$H$ [m]} 

1572:   \psfrag{length in m}[t][b]{$L$ [m]} 

1573:   \psfrag{vd in m per year}{$v$ [m/yr]} 

1574:   \psfrag{Lvd in m per year}{}   

1575:   \includegraphics[width=0.45\columnwidth]{2d_v_vs_h.eps}

1576:   \includegraphics[width=0.45\columnwidth]{2d_v_vs_L.eps}

1577:   \caption{Migration velocities predicted by the minimal model for

1578:   steady--state dunes of different size at various wind

1579:   velocities. The caption gives the shear velocity $u_*$ in m/s. The

1580:   numerical data are compared to the scaling laws $v\propto H^{-1}$

1581:   (left) and $v\propto L^{-1}$, where $L$ is the length of the

1582:   envelope of the dune and its separation bubble as described in the

1583:   main text. (Note that the migration of real dunes is

1584:   substantially slower due to the small fraction of wind days per

1585:   year.)}  \label{fig:v_H_L}

1586: \end{center}

1587: \end{figure}

1588: 

1589: \subsection{Migration velocity}

1590: For the overall migration velocity of steady--state dunes with a scale

1591: invariant profile, we derived on general grounds the simple scaling

1592: prediction $v\propto L^{-1}$ in Section~\ref{sec:v}. We have also

1593: given some arguments why this prediction should be rather robust

1594: against relaxing the condition of shape invariance, in contrast to the

1595: relation $v\propto H^{-1}$ that can only be inferred from it if the

1596: scaling assumption holds exactly. Here, we check these predictions

1597: for the steady--state solutions numerically. Fig.~\ref{fig:v_H_L}

1598: shows the numerically obtained migration velocity for dunes fitted to

1599: the scaling relations $v\propto H^{-1}$ (left) and $v\propto L^{-1}$

1600: (right).  As we have mentioned above, one has to take for $L$ the

1601: characteristic length of the envelope rather than that of the dune

1602: alone. For simplicity, we estimate $L$ by adding $6H$ to the

1603: horizontal length from the windward foot of the dune to its crest,

1604: thus neglecting the weak slope dependence of the size of the

1605: separation bubble. Obviously, the $L^{-1}-$scaling is superior for

1606: moderate winds and small dunes where $v\propto H^{-1}$ systematically

1607: fails to describe the data. This is also supported by field data

1608: \cite{Finkel59}. Both fits become identical in the scaling limit $L\gg

1609: \ell_s$. Due to the decrease of $\ell_s$ with the wind speed, the

1610: latter is reached for smaller dunes at stronger winds.

1611: 

1612: \subsection{Stability}

1613: We have already pointed out that the choice of different boundary

1614: conditions for the flux allow a separate discussion of shape and mass

1615: stability. This is of practical importance, since (in $2d$) all

1616: steady--state shapes are unstable with respect to mass changes. If the

1617: influx of a steady--state solution deviates slightly from its

1618: corresponding steady--state flux, this solution will start to either

1619: shrink until it has flattened out or grow without bound.  Though the

1620: latter effect could (but need not) be a peculiarity of the vanishing

1621: outflux for $2d$ dunes with slip face, at least the former generalizes

1622: to $3d$ heaps. Despite the fact that the steady--state shapes are

1623: (locally) stable attractors for the shape evolution under periodic

1624: flux conditions, mass stability can in general not be achieved under

1625: open boundary conditions. The situation is clarified in

1626: Fig.~\ref{fig:q_V}. It depicts the steady--state sand flux $q$ over

1627: the bedrock as a function of aspect ratio. The numerical results

1628: nicely confirm our theoretical expectation from

1629: Eq.(\ref{eq:outflux}). For all dunes with slip face the flux vanishes

1630: identically in $2d$, whereas in general it grows with decreasing size

1631: for smooth heaps.  For open boundary conditions, the line in

1632: Fig.~\ref{fig:q_V} can be interpreted as an unstable phase boundary

1633: (with hysteresis) between infinitely growing and shrinking

1634: solutions. For example, a heap with influx slightly below the

1635: steady--state, will shrink a bit. To remain close to the steady--state

1636: shape, it will therefore mainly reduce its height, whereas its length

1637: will stay almost constant. Due to the reduced aspect ratio

1638: $\varepsilon$, $\hat \tau$ decreases in magnitude and the shear stress

1639: depression at the lee boundary is less pronounced. As a consequence

1640: there is less deposition on the downwind slope and the outflux is

1641: higher, so that the heap shrinks even more etc. A completely analogous

1642: reasoning applies to the opposite case of higher influx. The

1643: corresponding shape attractors are the scale invariant asymptotic dune

1644: shape and the flat surface, respectively.

1645: 

1646: The above discussion explains, why isolated smooth aeolian sand heaps

1647: are rarely observed as distinct features of desert topographies. Under

1648: approximately stationary wind and influx conditions they only exist as

1649: transient states that either vanish or develop into dunes with slip

1650: face. Under variable wind and influx conditions, the situation is less

1651: clear and deserves a detailed study of its own. For example, the model

1652: predicts that during a period of strong wind all dunes are driven

1653: towards the asymptotic shape. After a subsequent period of weak winds,

1654: finite size effects become more pronounced, and small dunes may

1655: develop longitudinal profiles like those in the hysteretic regime or

1656: even loose their slip face. Again, the case $\tau_0 \approx \tau_t$ of

1657: a shear stress close to or below the threshold shear stress needs

1658: special attention.

1659: 

1660: The prevailing wind conditions as well as recent changes in the wind

1661: velocity are thus encoded in a complicated but comprehensible way in

1662: the shapes of the dunes in a dune field. This is a promising direction

1663: for further studies. One may hope that by systematic studies

1664: along these lines one will in the future be able to infer flow

1665: conditions in remote or uncomfortable places (e.g.\ on the sea bottom

1666: or on other planets) by analyzing dune

1667: shapes.

1668: 

1669: \begin{figure}[tb]

1670:   \psfrag{flat}{ flat} 

1671:   \psfrag{steep}{ steep} 

1672:   \psfrag{qin}[c]{$q^{\rm out}$ [kg\,m$^{-1}$s$^{-1}$]}  

1673:   \psfrag{eps}{$H/L$}

1674:   \psfrag{growing}{$\stackrel{\displaystyle

1675:   \text{growing}}{\text{solutions}}$}

1676:   \psfrag{shrinking}{$\stackrel{\displaystyle

1677:   \text{shrinking}}{\text{solutions}}$} 

1678: \begin{center}

1679:   \includegraphics[width=\columnwidth]{q_aspect.eps}

1680:   \caption{Steady--state outflux under periodic boundary

1681:   conditions. In the hysteretic regime, steep and flat initial

1682:   conditions have to be distinguished, as in Fig.~\ref{fig:H_V} The

1683:   figure may also be read as a phase diagram for the situation with

1684:   open boundary conditions. In this case the steady--state solutions

1685:   --- though attractors for the shape --- are unstable with respect to

1686:   mass fluctuations.}  \label{fig:q_V}

1687: \end{center}

1688: \end{figure}

1689: 

1690: 

1691: \subsection{Relaxation dynamics}\label{sec:relaxation}

1692: As a first step towards an understanding of the effects of variable

1693: wind speeds (for constant wind direction), this section is devoted to

1694: an exploratory study of the transient shape evolution. We restrict

1695: ourselves to periodic boundary conditions leaving the richer phase

1696: space of open boundary conditions for future

1697: studies. Fig.~\ref{fig:growth} shows two extreme scenarios. A flat

1698: initial condition with a mass greater than the critical mass for slip

1699: face formation (a), and a steep initial condition with a mass below

1700: the critical mass for slip face destruction (b). The steep initial

1701: condition gives rise to the temporary formation of a slip face that is

1702: finally washed out by the saturation transients, whereas the flat heap

1703: steepens until the shear stress on the lee falls to the threshold

1704: value.  This causes complete deposition on the lee side of the

1705: heap. Whether this happens before or with the onset of flow separation

1706: depends on wind and influx conditions. As it also depends on the

1707: precise numerical values of some of the phenomenological parameters of

1708: the model, a detailed parametric study is again beyond the scope of

1709: the present contribution.

1710: 

1711: \begin{figure}[t]

1712:   \begin{center}  

1713:     \includegraphics[width=\columnwidth]{comb_slipface.eps}

1714:     \caption{Growth histories for two Gaussian heaps of different mass

1715:       and initial aspect ratio for periodic flux boundary conditions.

1716:       Note that the distances to reach the steady--state shape are

1717:       different.  (The aspect ratio of both plots was rescaled by a

1718:       common factor for better visualization.)}

1719: \label{fig:growth} 

1720: \end{center}

1721: \end{figure}\noindent

1722: 

1723: Although the times to reach the steady state are apparently somewhat

1724: longer for the flat initial condition, it is evident from

1725: Fig.~\ref{fig:growth} that the relaxation dynamics is in general

1726: relatively fast even if the initial condition is far from the steady

1727: state shape. Large dunes under low influx conditions as they prevail

1728: e.g.\ in fields of isolated dunes should therefore be well described

1729: by an adiabatic approximation assuming that (except after drastic

1730: changes in the wind and sand conditions as they occur during sand

1731: storms) the dune is practically in a steady state.  Apart from the low

1732: influx, this also relies on the fact that virtually no sand is lost

1733: over the slip face. For a large isolated $3d$ barchan dune this

1734: implies that most of the sediment transported over the dune is

1735: actually trapped in a tread--milling flux, and only a small portion of

1736: the total flux is contributed by and contributes to the external flux.

1737: Hence, under steady wind conditions these dunes are in a quasi steady

1738: state and thus very close to their true steady--state

1739: shape. Investigation of the steady--state properties is therefore the

1740: starting point also for the study of their time evolution. Moreover,

1741: this suggests that a comparison of our steady--state shapes to shapes

1742: obtained in field measurements is justified. In fact, the calculated

1743: shapes agree nicely with recent measurements for barchan dunes

1744: \cite{sauermann-etal:2000}. The situation is less clear for small

1745: heaps, where mass losses can be of the order of the total flux and may

1746: thus lead to significant differences between the steady--state and the

1747: transient shapes under vanishing influx.

1748: 

1749: In the remainder of this section, we want to investigate more closely

1750: the mechanism that drives the shape relaxation. As we have pointed

1751: out, the positions of the maximum of the sand flux and of the maximum

1752: of the profile must coincide in the steady state to make the

1753: erosion/deposition vanish at the crest. We have shown that for small

1754: heaps, this can be achieved by a fine--tuning of $\delta x_\tau$ to

1755: about $\ell_s$. In contrast, for large dunes and strong winds, $\delta

1756: x_\tau \gg \ell_s$, and the steady--state condition can only be met

1757: with a singularity at the crest. This important difference is

1758: exemplified by Figs.~\ref{fig:deltaheap} and \ref{fig:deltadune}. Both

1759: figures show the evolution of the height and the displacements $\delta

1760: x_\tau$ and $\delta x_q$ of the locations of the maximum of the shear

1761: stress and of the maximum of the sand flux from the location of the

1762: top of the sand profile, respectively. The distance between both

1763: displacements is the lag of the flux with respect to the shear stress

1764: due to the saturation transients, and is therefore closely related to

1765: the saturation length $\ell_s$ for smooth surface profiles.  It

1766: guarantees the proper vanishing of the erosion rate $q'$ at the top of

1767: a steady--state heap where $\delta x_\tau$ is finite, but vanishes for

1768: large steady--state dunes, where the slip face ends in a sharp brink

1769: singularity at which the grains fall into the wake and are quickly

1770: deposited.

1771: 

1772: \begin{figure}[tb]

1773:   \begin{center}  

1774:   \psfrag{h in m}{$H$ [m]}

1775:   \psfrag{delta in m}{$\delta x_{\tau, q}$ [m]}

1776:   \psfrag{time}{time}

1777:   \psfrag{xmaxtau}{$\delta x_\tau$}

1778:   \psfrag{xmaxqxx}{$\delta x_q$}

1779:   \includegraphics[width=\columnwidth]{2d_displacement_vs_time_bump.eps} 

1780:  \caption{The figure shows the transient evolution of various

1781:   interesting length scales for a heap. \emph{Lower part:} Height of

1782:   the heap. \emph{Upper part:} Distance of the locations of the

1783:   maximum of the shear stress and of the maximum of the sand flux from

1784:   the position of the top of the heap. In the steady state, the

1785:   erosion/deposition vanishes at the crest.}  \label{fig:deltaheap}

1786: \end{center}

1787: \end{figure}

1788: 

1789: \begin{figure}[tb]

1790:   \begin{center}  

1791:   \psfrag{h in m}{$H$ [m]}

1792:   \psfrag{delta in m}{$\delta x_{\tau, q}$ [m]}

1793:   \psfrag{time}{time}

1794:   \psfrag{xmaxtau}{$\delta x_\tau$}

1795:   \psfrag{xmaxqxx}{$\delta x_q$}

1796: 

1797:   \includegraphics[width=\columnwidth]{2d_displacement_vs_time.eps}

1798:   \caption{The figure shows the transient evolution of various

1799:   interesting length scales for a dune that develops out of a smooth

1800:   heap as in Fig.~\ref{fig:growth}. \emph{Lower part:} Height of the

1801:   dune. \emph{Upper part:} Distance of the locations of the maximum of

1802:   the shear stress and of the maximum sand flux from the position of

1803:   the top of the crest. The lag between shear stress and sand flux

1804:   vanishes, when the slip face reaches the crest.}

1805:   \label{fig:deltadune}

1806: \end{center}

1807: \end{figure} \noindent

1808: 

1809: \section{Summary and outview}\label{sec:sum}

1810: In summary, we have shown that a simple minimal model for the

1811: wind--driven sediment transport over a sand dune is capable of

1812: explaining several important features of desert topographies. Among

1813: them are the migration velocities of heaps and dunes, their shape

1814: along the wind direction and the existence of a minimal dune size and

1815: a maximum heap size.

1816: 

1817: As we have emphasized throughout this contribution and demonstrated by

1818: the numerical solutions in the preceding section, the symmetry

1819: breaking part of the shear stress exerted by turbulent air flow on an

1820: obstacle, and local deviations of the sediment flux from its

1821: equilibrium transport capacity (``saturation transients''), are the

1822: essential ingredients in the modeling of aeolian sand dunes. It is

1823: exactly the balance of these two relatively small effects that is

1824: responsible for the relaxation of arbitrary initial conditions into a

1825: characteristic dune or heap shape. Their neglect was responsible for

1826: the failure of the naive zeroth order model discussed in

1827: Section~\ref{sec:zero}. In hindsight we can say that it is not so much

1828: the quantitative errors but the omission of this \emph{qualitatively}

1829: important mechanism, what makes the zeroth order model an insufficient

1830: description. In contrast, taking this balance properly into account,

1831: makes the minimal model structurally stable against the neglect of

1832: less significant quantitative details of the same order of magnitude.

1833: 

1834: \begin{figure}[tb]

1835:   \psfrag{L}{$L$ [m]}

1836:   \psfrag{ustar}{$u_*/u_{*t}$}

1837:   \begin{center}

1838:   \includegraphics[width=\columnwidth]{shape_diagram.eps}

1839:   \caption{Qualitative shape diagram that could be useful

1840:   in the analysis and comparison of field studies. The migration

1841:   velocity is constant along rising lines, whereas falling lines

1842:   indicate invariant dune shape.} \label{fig:shapediagram}

1843: \end{center}

1844: \end{figure}

1845: 

1846: This direction was recently pursued further in an effort to calculate

1847: analytically certain steady state shapes of dunes and heaps by

1848: ``linearizing'' the minimal model \cite{andreotti-claudin:2002}. One

1849: may as well wish to proceed also into the opposite direction. After

1850: the basic mechanism is understood, more elaborate dune models can be

1851: constructed by putting some of the neglected details back into the

1852: description. Detailed parametric studies of such a refined model for a

1853: certain dune type and comparison to field data would be very useful to

1854: test some of the less generic predictions of the underlying sand

1855: transport model \cite{sauermann-kroy-herrmann:2001}, such as the shear

1856: stress dependence of the saturation length $\ell_s$

1857: (Fig.~\ref{fig:ls}). This is important, since, as we have shown, the

1858: variable parameter $\ell_s$ sets the characteristic length scale with

1859: respect to which dunes and heaps can be said to be large or

1860: small. Phenomenological knowledge about $\ell_s$ is still very

1861: limited. More detailed studies could further be helpful to map out

1862: quantitative shape diagrams, of the type sketched qualitatively in

1863: Fig.~\ref{fig:shapediagram}. These diagrams could be useful not only

1864: for the validation of the model, but also for the comparison of field

1865: data from different places with different prevailing wind and sand

1866: conditions. The migration velocity is constant along the rising lines

1867: in Fig.~\ref{fig:shapediagram}, which were obtained from

1868: Eq.(\ref{eq:v}) using $q\approx q_s$ together with

1869: Eq.(\ref{eq:ls}). They allow for example a comparison of the migration

1870: velocities of dunes of different sizes that are exposed to (on

1871: average) identical winds. Or one may infer the average wind speed from

1872: measurements of sizes and migration velocities in a dune field. The

1873: falling lines in Fig.~\ref{fig:shapediagram} are lines of constant

1874: shape, assuming that the latter is determined by $\ell_s/L$. They may

1875: thus be used for correlating wind speeds with dune shapes. In general

1876: (in particular for the full $3d$ problem), such shape diagrams will be

1877: more complex since the influx is an additional important variable that

1878: we have neglected here, as it vanishes for $2d$ dunes in the steady

1879: state.

1880: 

1881: Moreover, as we pointed out, there are still many consequences of the

1882: present model that await a systematic investigation.  And a major

1883: future task is finally the generalization of the present discussion to

1884: the $3d$ case. A promising route could be the construction of an

1885: effectively sliced model that allows one to use the proposed model for

1886: the separation bubble and to keep the time--limiting calculation (the

1887: integration of the flux equation) effectively one--dimensional. The

1888: smaller transverse currents could be inferred from the sliced

1889: solution. A generalization of the flux equation to the $2d$ surface of

1890: a $3d$ dune is also feasible \cite{sauermann:phd}.  A more ambitious

1891: task will eventually be the simulation of dune fields.  Whereas the

1892: existence of a minimum dune size could be obtained by an analysis of

1893: the shape stability alone, the question of a possible existence of a

1894: characteristic or maximum dune size in a dune field, depends on the

1895: mass balance of a dune in the complicated environment provided by the

1896: other dunes, and is much more difficult to answer

1897: \cite{lima-etal:2002}.

1898: 

1899: \begin{acknowledgments}

1900: We gratefully acknowledge financial support by the Deutsche

1901: Forschungsgemeinschaft through contract No HE 2732/1-1, and helpful

1902: discussions with Ken Andersen and Philippe Claudin.

1903: \end{acknowledgments}

1904: 

1905: %\bibliography{journals,basics,dune,special,unpub}

1906: 

1907: \begin{thebibliography}{40}

1908: \expandafter\ifx\csname natexlab\endcsname\relax\def\natexlab#1{#1}\fi

1909: \expandafter\ifx\csname bibnamefont\endcsname\relax

1910:   \def\bibnamefont#1{#1}\fi

1911: \expandafter\ifx\csname bibfnamefont\endcsname\relax

1912:   \def\bibfnamefont#1{#1}\fi

1913: \expandafter\ifx\csname citenamefont\endcsname\relax

1914:   \def\citenamefont#1{#1}\fi

1915: \expandafter\ifx\csname url\endcsname\relax

1916:   \def\url#1{\texttt{#1}}\fi

1917: \expandafter\ifx\csname urlprefix\endcsname\relax\def\urlprefix{URL }\fi

1918: \providecommand{\bibinfo}[2]{#2}

1919: \providecommand{\eprint}[2][]{\url{#2}}

1920: 

1921: \bibitem[{\citenamefont{Bagnold}(1941)}]{Bagnold41}

1922: \bibinfo{author}{\bibfnamefont{R.~A.} \bibnamefont{Bagnold}},

1923:   \emph{\bibinfo{title}{The physics of blown sand and desert dunes.}}

1924:   (\bibinfo{publisher}{Methuen}, \bibinfo{address}{London},

1925:   \bibinfo{year}{1941}).

1926: 

1927: \bibitem[{\citenamefont{Pye and Tsoar}(1990)}]{Pye90}

1928: \bibinfo{author}{\bibfnamefont{K.}~\bibnamefont{Pye}} \bibnamefont{and}

1929:   \bibinfo{author}{\bibfnamefont{H.}~\bibnamefont{Tsoar}},

1930:   \emph{\bibinfo{title}{Aeolian sand and sand dunes}}

1931:   (\bibinfo{publisher}{Unwin Hyman}, \bibinfo{address}{London},

1932:   \bibinfo{year}{1990}).

1933: 

1934: \bibitem[{\citenamefont{Lancaster}(1995)}]{Lancaster95}

1935: \bibinfo{author}{\bibfnamefont{N.}~\bibnamefont{Lancaster}},

1936:   \emph{\bibinfo{title}{Geomorphology of desert dunes}}

1937:   (\bibinfo{publisher}{Routledge}, \bibinfo{address}{London},

1938:   \bibinfo{year}{1995}).

1939: 

1940: \bibitem[{\citenamefont{Malin et~al.}(1998)\citenamefont{Malin, Carr,

1941:   Danielson, Davies, Hartmann, Ingersoll, James, Masursky, McEwen, Soderblom

1942:   et~al.}}]{malin-etal:98}

1943: \bibinfo{author}{\bibfnamefont{M.~C.} \bibnamefont{Malin}},

1944:   \bibinfo{author}{\bibfnamefont{M.~H.} \bibnamefont{Carr}},

1945:   \bibinfo{author}{\bibfnamefont{G.~E.} \bibnamefont{Danielson}},

1946:   \bibinfo{author}{\bibfnamefont{M.~E.} \bibnamefont{Davies}},

1947:   \bibinfo{author}{\bibfnamefont{W.~K.} \bibnamefont{Hartmann}},

1948:   \bibinfo{author}{\bibfnamefont{A.~P.} \bibnamefont{Ingersoll}},

1949:   \bibinfo{author}{\bibfnamefont{P.~B.} \bibnamefont{James}},

1950:   \bibinfo{author}{\bibfnamefont{H.}~\bibnamefont{Masursky}},

1951:   \bibinfo{author}{\bibfnamefont{A.~S.} \bibnamefont{McEwen}},

1952:   \bibinfo{author}{\bibfnamefont{L.~A.} \bibnamefont{Soderblom}},

1953:   \bibnamefont{et~al.}, \bibinfo{journal}{Science}

1954:   \textbf{\bibinfo{volume}{279}}, \bibinfo{pages}{1681} (\bibinfo{year}{1998}).

1955: 

1956: \bibitem[{\citenamefont{Thomas et~al.}(1999)\citenamefont{Thomas, Malin, Carr,

1957:   Danielson, Davies, Hartmann, Ingersoll, James, McEwen, Soderblom

1958:   et~al.}}]{Thomas99}

1959: \bibinfo{author}{\bibfnamefont{P.~C.} \bibnamefont{Thomas}},

1960:   \bibinfo{author}{\bibfnamefont{M.~C.} \bibnamefont{Malin}},

1961:   \bibinfo{author}{\bibfnamefont{M.~H.} \bibnamefont{Carr}},

1962:   \bibinfo{author}{\bibfnamefont{G.~E.} \bibnamefont{Danielson}},

1963:   \bibinfo{author}{\bibfnamefont{M.~E.} \bibnamefont{Davies}},

1964:   \bibinfo{author}{\bibfnamefont{W.~K.} \bibnamefont{Hartmann}},

1965:   \bibinfo{author}{\bibfnamefont{A.~P.} \bibnamefont{Ingersoll}},

1966:   \bibinfo{author}{\bibfnamefont{P.~B.} \bibnamefont{James}},

1967:   \bibinfo{author}{\bibfnamefont{A.~S.} \bibnamefont{McEwen}},

1968:   \bibinfo{author}{\bibfnamefont{L.~A.} \bibnamefont{Soderblom}},

1969:   \bibnamefont{et~al.}, \bibinfo{journal}{Nature}

1970:   \textbf{\bibinfo{volume}{397}}, \bibinfo{pages}{592} (\bibinfo{year}{1999}).

1971: 

1972: \bibitem[{\citenamefont{Kroy et~al.}(2002)\citenamefont{Kroy, Sauermann, and

1973:   Herrmann}}]{kroy-sauermann-herrmann:2002}

1974: \bibinfo{author}{\bibfnamefont{K.}~\bibnamefont{Kroy}},

1975:   \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Sauermann}},

1976:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{H.~J.}

1977:   \bibnamefont{Herrmann}}, \bibinfo{journal}{Phys.~Rev.~Lett.}

1978:   \textbf{\bibinfo{volume}{88}}, \bibinfo{pages}{054301}

1979:   (\bibinfo{year}{2002}).

1980: 

1981: \bibitem[{\citenamefont{Hoyle and Woods}(1997)}]{hoyle-woods:97}

1982: \bibinfo{author}{\bibfnamefont{R.~B.} \bibnamefont{Hoyle}} \bibnamefont{and}

1983:   \bibinfo{author}{\bibfnamefont{A.~W.} \bibnamefont{Woods}},

1984:   \bibinfo{journal}{Phys.~Rev.~E} \textbf{\bibinfo{volume}{56}},

1985:   \bibinfo{pages}{6861} (\bibinfo{year}{1997}).

1986: 

1987: \bibitem[{\citenamefont{Hoyle and Mehta}(1999)}]{hoyle-mehta:99}

1988: \bibinfo{author}{\bibfnamefont{R.~B.} \bibnamefont{Hoyle}} \bibnamefont{and}

1989:   \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Mehta}},

1990:   \bibinfo{journal}{Phys.~Rev.~Lett.} \textbf{\bibinfo{volume}{83}},

1991:   \bibinfo{pages}{5170} (\bibinfo{year}{1999}).

1992: 

1993: \bibitem[{\citenamefont{Prigozhin}(1999)}]{prigozhin:99}

1994: \bibinfo{author}{\bibfnamefont{L.}~\bibnamefont{Prigozhin}},

1995:   \bibinfo{journal}{Phys.~Rev.~E} \textbf{\bibinfo{volume}{60}},

1996:   \bibinfo{pages}{729} (\bibinfo{year}{1999}).

1997: 

1998: \bibitem[{\citenamefont{Sauermann et~al.}(2001)\citenamefont{Sauermann, Kroy,

1999:   and Herrmann}}]{sauermann-kroy-herrmann:2001}

2000: \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Sauermann}},

2001:   \bibinfo{author}{\bibfnamefont{K.}~\bibnamefont{Kroy}}, \bibnamefont{and}

2002:   \bibinfo{author}{\bibfnamefont{H.~J.} \bibnamefont{Herrmann}},

2003:   \bibinfo{journal}{Phys.~Rev.~E} \textbf{\bibinfo{volume}{64}},

2004:   \bibinfo{pages}{031305} (\bibinfo{year}{2001}).

2005: 

2006: \bibitem[{\citenamefont{Jackson and Hunt}(1975)}]{jackson-hunt:75}

2007: \bibinfo{author}{\bibfnamefont{P.~S.} \bibnamefont{Jackson}} \bibnamefont{and}

2008:   \bibinfo{author}{\bibfnamefont{J.~C.~R.} \bibnamefont{Hunt}},

2009:   \bibinfo{journal}{Q. J. R. Meteorol. Soc.} \textbf{\bibinfo{volume}{101}},

2010:   \bibinfo{pages}{929} (\bibinfo{year}{1975}).

2011: 

2012: \bibitem[{\citenamefont{Sykes}(1980)}]{sykes:80}

2013: \bibinfo{author}{\bibfnamefont{R.~I.} \bibnamefont{Sykes}},

2014:   \bibinfo{journal}{J.~Fluid Mech.} \textbf{\bibinfo{volume}{101}},

2015:   \bibinfo{pages}{647} (\bibinfo{year}{1980}).

2016: 

2017: \bibitem[{\citenamefont{Zeman and Jensen}(1988)}]{zeman-jensen:88}

2018: \bibinfo{author}{\bibfnamefont{O.}~\bibnamefont{Zeman}} \bibnamefont{and}

2019:   \bibinfo{author}{\bibfnamefont{N.~O.} \bibnamefont{Jensen}},

2020:   \bibinfo{type}{Tech. Rep.} \bibinfo{number}{M-2738},

2021:   \bibinfo{institution}{Ris{\o} National Laboratory} (\bibinfo{year}{1988}).

2022: 

2023: \bibitem[{\citenamefont{Hunt et~al.}(1988)\citenamefont{Hunt, Leibovich, and

2024:   Richards}}]{hunt-leibovich-richards:88}

2025: \bibinfo{author}{\bibfnamefont{J.~C.~R.} \bibnamefont{Hunt}},

2026:   \bibinfo{author}{\bibfnamefont{S.}~\bibnamefont{Leibovich}},

2027:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{K.~J.}

2028:   \bibnamefont{Richards}}, \bibinfo{journal}{Q. J. R. Meteorol. Soc.}

2029:   \textbf{\bibinfo{volume}{114}}, \bibinfo{pages}{1435} (\bibinfo{year}{1988}).

2030: 

2031: \bibitem[{\citenamefont{Carruthers and Hunt}(1990)}]{Carruthers90}

2032: \bibinfo{author}{\bibfnamefont{D.~J.} \bibnamefont{Carruthers}}

2033:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{J.~C.~R.}

2034:   \bibnamefont{Hunt}}, in \emph{\bibinfo{booktitle}{Atmospheric Processes over

2035:   Complex Terrain}}, edited by

2036:   \bibinfo{editor}{\bibfnamefont{W.}~\bibnamefont{Blumen}}

2037:   (\bibinfo{publisher}{Am. Meteorological Soc.}, \bibinfo{address}{Boston},

2038:   \bibinfo{year}{1990}), vol.~\bibinfo{volume}{23}.

2039: 

2040: \bibitem[{\citenamefont{Weng et~al.}(1991)\citenamefont{Weng, Hunt, Carruthers,

2041:   A., Wiggs, Livingstone, and Castro}}]{weng-etal:91}

2042: \bibinfo{author}{\bibfnamefont{W.~S.} \bibnamefont{Weng}},

2043:   \bibinfo{author}{\bibfnamefont{J.~C.~R.} \bibnamefont{Hunt}},

2044:   \bibinfo{author}{\bibfnamefont{D.~J.} \bibnamefont{Carruthers}},

2045:   \bibinfo{author}{\bibfnamefont{W.}~\bibnamefont{A.}},

2046:   \bibinfo{author}{\bibfnamefont{G.~F.~S.} \bibnamefont{Wiggs}},

2047:   \bibinfo{author}{\bibfnamefont{I.}~\bibnamefont{Livingstone}},

2048:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{I.}~\bibnamefont{Castro}},

2049:   \bibinfo{journal}{Acta Mech. Suppl.} \textbf{\bibinfo{volume}{2}},

2050:   \bibinfo{pages}{1} (\bibinfo{year}{1991}).

2051: 

2052: \bibitem[{\citenamefont{Barndorff-Nielsen and

2053:   Willetts}(1991)}]{ActaMechanica91}

2054: \bibinfo{editor}{\bibfnamefont{O.~E.} \bibnamefont{Barndorff-Nielsen}}

2055:   \bibnamefont{and} \bibinfo{editor}{\bibfnamefont{B.~B.}

2056:   \bibnamefont{Willetts}}, eds., \emph{\bibinfo{title}{Aeolian Grain Transport

2057:   1 {\&} 2}}, Acta Mechanica Suppl. (\bibinfo{publisher}{Springer},

2058:   \bibinfo{address}{Wien, New York}, \bibinfo{year}{1991}).

2059: 

2060: \bibitem[{\citenamefont{Lettau and Lettau}(1978)}]{Lettau78}

2061: \bibinfo{author}{\bibfnamefont{K.}~\bibnamefont{Lettau}} \bibnamefont{and}

2062:   \bibinfo{author}{\bibfnamefont{H.~H.} \bibnamefont{Lettau}}, in

2063:   \emph{\bibinfo{booktitle}{Exploring the world's driest climate}}, edited by

2064:   \bibinfo{editor}{\bibfnamefont{H.~H.} \bibnamefont{Lettau}} \bibnamefont{and}

2065:   \bibinfo{editor}{\bibfnamefont{K.}~\bibnamefont{Lettau}}

2066:   (\bibinfo{publisher}{Madison}, \bibinfo{address}{Center for Climatic

2067:   Research, Univ. Wisconsin}, \bibinfo{year}{1978}).

2068: 

2069: \bibitem[{\citenamefont{S{\o}rensen}(1991)}]{Sorensen91}

2070: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{S{\o}rensen}},

2071:   \bibinfo{journal}{Acta Mechanica (Suppl.)} \textbf{\bibinfo{volume}{1}},

2072:   \bibinfo{pages}{67} (\bibinfo{year}{1991}).

2073: 

2074: \bibitem[{\citenamefont{Landau and Lifshitz}(1963)}]{landau-lifshitz:fm}

2075: \bibinfo{author}{\bibnamefont{Landau}} \bibnamefont{and}

2076:   \bibinfo{author}{\bibnamefont{Lifshitz}}, \emph{\bibinfo{title}{Fluid

2077:   Mechanics}}, vol.~\bibinfo{volume}{6} of \emph{\bibinfo{series}{Course of

2078:   Theoretical Physics}} (\bibinfo{publisher}{Pergamon Press},

2079:   \bibinfo{address}{London}, \bibinfo{year}{1963}).

2080: 

2081: \bibitem[{spe()}]{speed}

2082: \bibinfo{note}{This has to be distinguished from the statement that the

2083:   migration velocity of a dune with slip face is proportional to the flux over

2084:   the brink divided by the height, which is (trivially) true if losses over the

2085:   slip face are negligible, but useless unless the flux over the brink is

2086:   specified.}

2087: 

2088: \bibitem[{\citenamefont{Finkel}(1959)}]{Finkel59}

2089: \bibinfo{author}{\bibfnamefont{H.~J.} \bibnamefont{Finkel}},

2090:   \bibinfo{journal}{Journal of Geology} \textbf{\bibinfo{volume}{67}},

2091:   \bibinfo{pages}{614} (\bibinfo{year}{1959}).

2092: 

2093: \bibitem[{\citenamefont{Abramowitz and Stegun}(1965)}]{abramowitz-stegun}

2094: \bibinfo{author}{\bibfnamefont{M.}~\bibnamefont{Abramowitz}} \bibnamefont{and}

2095:   \bibinfo{author}{\bibfnamefont{I.~A.} \bibnamefont{Stegun}},

2096:   \emph{\bibinfo{title}{Handbook of Mathematical Functions}}

2097:   (\bibinfo{publisher}{Dover}, \bibinfo{address}{New York},

2098:   \bibinfo{year}{1965}).

2099: 

2100: \bibitem[{\citenamefont{Wippermann and Gross}(1986)}]{Wippermann86}

2101: \bibinfo{author}{\bibfnamefont{F.~K.} \bibnamefont{Wippermann}}

2102:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Gross}},

2103:   \bibinfo{journal}{Boundary Layer Meteorology} \textbf{\bibinfo{volume}{36}},

2104:   \bibinfo{pages}{319} (\bibinfo{year}{1986}).

2105: 

2106: \bibitem[{\citenamefont{van Dijk et~al.}(1999)\citenamefont{van Dijk, Arens,

2107:   and van Boxel}}]{van_dijk-arens-boxel:99}

2108: \bibinfo{author}{\bibfnamefont{P.~M.} \bibnamefont{van Dijk}},

2109:   \bibinfo{author}{\bibfnamefont{S.~M.} \bibnamefont{Arens}}, \bibnamefont{and}

2110:   \bibinfo{author}{\bibfnamefont{J.~H.} \bibnamefont{van Boxel}},

2111:   \bibinfo{journal}{Earth Surf. Process. Landforms}

2112:   \textbf{\bibinfo{volume}{24}}, \bibinfo{pages}{319} (\bibinfo{year}{1999}).

2113: 

2114: \bibitem[{\citenamefont{Stam}(1997)}]{Stam97}

2115: \bibinfo{author}{\bibfnamefont{J.~M.~T.} \bibnamefont{Stam}},

2116:   \bibinfo{journal}{Sedimentology} \textbf{\bibinfo{volume}{44}},

2117:   \bibinfo{pages}{127} (\bibinfo{year}{1997}).

2118: 

2119: \bibitem[{bad()}]{badbubble}

2120: \bibinfo{note}{An inapropriate boundary condition appears to be the main reason

2121:   for the failure of the original attempts to model flow separation in this way

2122:   \cite{zeman-jensen:88}.}

2123: 

2124: \bibitem[{rai()}]{rainout}

2125: \bibinfo{note}{For most applications it is a matter of convenience how exactly

2126:   the shear stress is set to zero inside the separation zone. For example, one

2127:   may choose a form that mimics the theoretically expected

2128:   \cite{sauermann-kroy-herrmann:2001} deposition of sand on the slip face

2129:   without further adaptation of the model. In $2d$ one may also treat the brink

2130:   as a boundary for the flux equation (see Section~\ref{sec:slipface}).}

2131: 

2132: \bibitem[{\citenamefont{{Fluent Inc.}}(1999)}]{Fluent5}

2133: \bibinfo{author}{\bibnamefont{{Fluent Inc.}}}, \emph{\bibinfo{title}{{FLUENT

2134:   5}}} (\bibinfo{year}{1999}), \bibinfo{note}{finite volume solver}.

2135: 

2136: \bibitem[{\citenamefont{Sauermann}(2001)}]{sauermann:phd}

2137: \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Sauermann}}, Ph.D. thesis,

2138:   \bibinfo{school}{{U}niversit{\"a}t {S}tuttgart} (\bibinfo{year}{2001}).

2139: 

2140: \bibitem[{\citenamefont{Wesseling}(2000)}]{wesseling:2000}

2141: \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Wesseling}},

2142:   \emph{\bibinfo{title}{Principles of Computational Fluid Dynamics}}

2143:   (\bibinfo{publisher}{Springer}, \bibinfo{address}{Berlin, Heidelberg, New

2144:   York}, \bibinfo{year}{2000}).

2145: 

2146: \bibitem[{\citenamefont{Sauermann et~al.}(2000)\citenamefont{Sauermann, Rognon,

2147:   Poliakov, and Herrmann}}]{sauermann-etal:2000}

2148: \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Sauermann}},

2149:   \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Rognon}},

2150:   \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Poliakov}}, \bibnamefont{and}

2151:   \bibinfo{author}{\bibfnamefont{H.~J.} \bibnamefont{Herrmann}},

2152:   \bibinfo{journal}{Geomorphology} \textbf{\bibinfo{volume}{36}},

2153:   \bibinfo{pages}{47} (\bibinfo{year}{2000}).

2154: 

2155: \bibitem[{\citenamefont{Wiggs et~al.}(1996)\citenamefont{Wiggs, Livingstone,

2156:   and Warren}}]{Wiggs96}

2157: \bibinfo{author}{\bibfnamefont{G.~F.~S.} \bibnamefont{Wiggs}},

2158:   \bibinfo{author}{\bibfnamefont{I.}~\bibnamefont{Livingstone}},

2159:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Warren}},

2160:   \bibinfo{journal}{Geomorphology} \textbf{\bibinfo{volume}{17}},

2161:   \bibinfo{pages}{29} (\bibinfo{year}{1996}).

2162: 

2163: \bibitem[{\citenamefont{Anderson}(1991)}]{Anderson91}

2164: \bibinfo{author}{\bibfnamefont{R.~S.} \bibnamefont{Anderson}},

2165:   \bibinfo{journal}{Acta Mechanica (Suppl.)} \textbf{\bibinfo{volume}{1}},

2166:   \bibinfo{pages}{21} (\bibinfo{year}{1991}).

2167: 

2168: \bibitem[{\citenamefont{Nalpanis et~al.}(1993)\citenamefont{Nalpanis, Hunt, and

2169:   Barrett}}]{Nalpanis93}

2170: \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Nalpanis}},

2171:   \bibinfo{author}{\bibfnamefont{J.~C.~R.} \bibnamefont{Hunt}},

2172:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{C.~F.}

2173:   \bibnamefont{Barrett}}, \bibinfo{journal}{J.~Fluid Mech.}

2174:   \textbf{\bibinfo{volume}{251}}, \bibinfo{pages}{661} (\bibinfo{year}{1993}).

2175: 

2176: \bibitem[{\citenamefont{Rioual et~al.}(2000)\citenamefont{Rioual, Valance, and

2177:   Bideau}}]{Rioual2000}

2178: \bibinfo{author}{\bibfnamefont{F.}~\bibnamefont{Rioual}},

2179:   \bibinfo{author}{\bibfnamefont{A.}~\bibnamefont{Valance}}, \bibnamefont{and}

2180:   \bibinfo{author}{\bibfnamefont{D.}~\bibnamefont{Bideau}},

2181:   \bibinfo{journal}{Phys.~Rev.~E} \textbf{\bibinfo{volume}{62}},

2182:   \bibinfo{pages}{2450} (\bibinfo{year}{2000}).

2183: 

2184: \bibitem[{\citenamefont{Owen}(1964)}]{Owen64}

2185: \bibinfo{author}{\bibfnamefont{P.~R.} \bibnamefont{Owen}}, \bibinfo{journal}{J.

2186:   Fluid. Mech.} \textbf{\bibinfo{volume}{20}}, \bibinfo{pages}{225}

2187:   (\bibinfo{year}{1964}).

2188: 

2189: \bibitem[{\citenamefont{{J. P. Bouchaud and M. E. Cates and J. Ravi Prakash and

2190:   S. F. Edwards}}(1994)}]{Bouchaud94}

2191: \bibinfo{author}{\bibnamefont{{J. P. Bouchaud and M. E. Cates and J. Ravi

2192:   Prakash and S. F. Edwards}}}, \bibinfo{journal}{J. Phys. France I}

2193:   \textbf{\bibinfo{volume}{4}}, \bibinfo{pages}{1383} (\bibinfo{year}{1994}).

2194: 

2195: \bibitem[{\citenamefont{Andreotti and Claudin}()}]{andreotti-claudin:2002}

2196: \bibinfo{author}{\bibfnamefont{B.}~\bibnamefont{Andreotti}} \bibnamefont{and}

2197:   \bibinfo{author}{\bibfnamefont{P.}~\bibnamefont{Claudin}},

2198:   \emph{\bibinfo{title}{Selection of dune shapes and velocities. part 2: A

2199:   two-dimensional modelling}}, \eprint{cond-mat/0201105}.

2200: 

2201: \bibitem[{\citenamefont{Lima et~al.}()\citenamefont{Lima, Sauermann, Herrmann,

2202:   and Kroy}}]{lima-etal:2002}

2203: \bibinfo{author}{\bibfnamefont{A.~R.} \bibnamefont{Lima}},

2204:   \bibinfo{author}{\bibfnamefont{G.}~\bibnamefont{Sauermann}},

2205:   \bibinfo{author}{\bibfnamefont{H.~J.} \bibnamefont{Herrmann}},

2206:   \bibnamefont{and} \bibinfo{author}{\bibfnamefont{K.}~\bibnamefont{Kroy}},

2207:   \emph{\bibinfo{title}{Modelling a dune field}}, \eprint{cond-mat/0202528}.

2208: 

2209: \end{thebibliography}

2210: 

2211: 

2212: \end{document}

2213: