AN ABSTRACT OF THE THESIS OF Franz Hugo Helfenstein in presented on Mathematics Doctor of Philosophy for the degree of May 1. 1986. A Free Boundary Value Problem Modeling Streambed Erosion Title: Redacted for Privacy Abstract approved: Ronald B. Guenther to derive Our purpose is twofold, equations of our model. bounded model of and to develop a solution procedure to solve the erosion streambed 2dimensional a The flow domain, which varies in time, is above by a free surface and is bounded below by an erodable streambed. An initial flow and streambed configuration given are and we would like to determine the streambed evolution. Our interaction of viscous a flow with which constraints boundary an erodable streambed. represent over Flow evolution. flow approach to modeling this process is to combine the basic a bed of sand causes the bed to undergo a distinct At slow flows, velocity, the bed the bed remains passes flat. increasing With through a set of stages: smooth, ripples, dunes, flat with chaotic flow, waves and finally antidunes. One of our main purposes is to obtain a model which has the potential of addressing this evolutionary process. We obtain a model which exhibits the general nature of streambed erosion. Specifically, we obtain classic dune migration. One indirect but significant result is that flow near vorticity of the the streambed is shown to be a fundamental factor in the erosion process. questions the which Another indirect result have arisen as is the wealth a result of this work. of open A Free Boundary Value Problem Modeling Streambed Erosion by Franz Hugo Helfenstein A THESIS submitted to Oregon State University in partial fulfillment of the requirements for the degree of Doctor of Philosophy Completed May 1, 1986 Commencement June 1986 APPROVED: Redacted for Privacy Professor of Mathematics i charge of major Redacted for Privacy Head of department of Mathematics Redacted for Privacy Dean of Gra School Date thesis is presented May 1, 1986 Typed by Franz Hugo Helfenstein for Franz Hugo Helfenstein Al.:14017LEDGELIENTS I would like to thank Ron Guenther for and help he thank my wife, gave Lea, his me in reaching this goal. for her patience, support she gave me along the way. patience, insight I would also like to understanding, and special TABLE OF CONTENTS INTRODUCTION The Problem The General Approach 1 BACKGROUND OF FLUID DYNAMICS AND SEDIMENTATION Fluid Dynamics The Navier-Stokes Equation Boundary Conditions Initial Conditions Dimensional Analysis The Reynolds Number and Other Parameters Vorticity and Potential Flow The Law of the Wall and Boundary Layers Energy Dissipation Concluding Remarks- Fluid Dynamics Sedimentation Empirical Results- Sediment Transport Empirical Results- Boundary Evolution A Brief History of Sedimentation Theory The Forces on Sediment Particles Concluding Remarks- Sedimentation 6 1 2 6 6 9 10 11 12 13 15 18 19 21 21 23 25 27 28 THE EROSION MODEL The Flow Field Streambed Erosion as a Boundary Condition Concluding Remarks- The Model 30 31 35 SPECIAL CASE FLOWS Almost Steady, Uniform Flow Potential Flow in 2 With Law of the Wall in r Concluding Remarks- Special Case Flows 46 46 49 52 A GENERAL FLOW The Flow Domain Solution Procedure- Reducing to a System of Initial Value Problems Solution Procedure- Solving the Initial Value Problems Numerically Results- Graphical Concluding Remarks- The General Flow 55 55 43 56 68 70 72 SUMMARY, OPEN QUESTIONS AND CONCLUDING REMARKS Summary Some Remarks The Model and its Solutions Improvements The Model Improvements The Numerical Scheme Open Questions Concluding Remarks 73 73 75 BIBLIOGRAPHY 80 APPENDIX The Inner Products 83 76 77 77 78 (i) LIST OF FIGURES Figure_ Page 1.0 The Erosion Problem Schematic 2.0 Potential Flow 14 2.1 The Law of the Wall 17 2.2 The Streambed Close up 17 2.3 Classification of Sediment Transport Mechanisms 22 2.4 'Major' and 'Minor' Streambed Evolutionary Stages 24 2.5 Classical Force Diagram for a Sediment Particle 28 3.0 The Flow Regions 0 and r 33 3.1 Sediment Particle Transport Mechanism 37 3.2 The Erosion Model- An Overview 44 4.0 Steady Uniform Flow 46 4.1 Almost Steady Uniform Flow- A Rippled Bed 47 4.2 Dune Migration- Almost Steady Uniform Flow 49 4.3 Potential Flow in 0 with Law of the Wall in 1 50 4.4a Dune Migration- Potential Flow/Law of the Wall 53 4.4b Dune Migration- Potential Flow/Law of the Wall 54 4.4c Dune Migration- Potential Flow/Law of the Wall 54 5.0 The Grid on R 64 5.1a Dune Migration- A General Flow 71 5.1b Dune Migration- A General Flow 71 1 A List of the Symbols Used In general, while variables subscripted with s refer to the sediment those subscripted with f refer to the fluid. is used for a critical empirical constants. value. Numbers constant; inflow at x=a A matrix block in A are critical efficiency factor large matrix ei 1d basis element constant; outflow at x=b eij 2d basis element matrix block in A kinetic energy wave speed; c=c(y*i) energy dissipation rate of fluid due to friction ci constant, empirical cz constant, empirical bed packing ratio sediment concentration Ea Ec Ef net rate energy available for erosion work rate critical energy rate level rate energy convertable to erosion work rate matrix block in A general function function; d={i'+yh')/h vector valued function da element of area 2 Fr dir local flow direction dr dr=[1+(e) ] 2 -1 fs Froude number; Fr=U /Lg force on sediment particle magnitude of gravity ds sediment diameter gravity vector ds for efficiency factor ec cs subscripts No subscript indicates a generic variable. a A as The subscript c element of arc/ength flow depth; h=ui flow domain; D=D(t)=B+1 rate of deformation tensor V. Karmen constant; k. constants =.417 ks conversion factori12 k =c c El+(e) S characteristic length qf qs Re arbitrary volume Vc critical velocity for sediment movement Vs sediment velocity neglecting resistence PS arbitrary function; K=K(x,i(x)) P1 V unit normal vector test function pressure, variable set of test functions nonfundamental parameters volumetric fluid discharge Ws x weight of particle coordinate in a sediment discharge coordinate in R resistance term coordinate in 0 Reynolds number; Re=UL/v y sediment particle curve parameter normal distance from 4 zo time, variable stress tensor coordinate in R stationary flow depth z=(a41,1,,e) a coeficient for u; u=a..eij ii horizontal component of u a s u* U0 sediment velocity along shear velocity; u*5riTirf stream gradient, degrees mean fluid velocity particle rotation rate mean flow 9c Uc Us Ua ..... a ) II ij coeficient for fluid velocity, vector U a=(a angle of repose critical flow aV surface of V mean sediment velocity 8 coeficient for A; A=Siei A del operator; A=V fluid velocity along a 2 vertical component of u ( iv ) Ax Ay grid width Xs grid height mainstream flow domain; 2=0(t) coeficiera for (15; gradient operator direction; down streambed flow domain; A r=r(t) map; A:0>R upper surface; ri=11(x,t) streambed; i=i(x,t) do initial bed form ir 3./4/59... density; p=pf+ps pf ps a fluid density sediment density common boundary of 0 and shear stress TO c boundary shear stress critical shear dynamic viscosity kinematic viscosity; V=4/Pf vorticity, scalar coeficient for h; h=4ie1 X dune wave length coefficient for w; w=X..e A FREE BOUNDARY VALUE PROBLEM MODELING STREAMBED EROSION INTRODUCTION 11( x, t) flow OW MEM soms, NOM wk. IN= .444 NNW unlit I a(x,t) ss. . , .. . . . ' The Erosion Problem Schematic Fig 1.0 The Problem Our purpose is twofold, and develop to model. to derive a model of streambed erosion a solution procedure to solve the equations of our The 2dimensional flow domain, consists region, 11=11(x.t), of two distinct regions, 0=0(0, while D = 0 is bounded above by the D(t), an + r. i=i(x,t). The mainstream flow air/fluid eroding boundary region, below by an erodable streambed, which varies in time, free r=r(t), surface, is bounded 0 and r share the common 2 a, which separates the mainstream flow from the transport boundary, At time to, (see Fig 1.0). region. for t > to. We refer region the given and we would like to determine the flow and Do in the fluid velocity is D(t) to the above problem as 'the problem' or as 'the erosion problem'. The General Apuroach Our basic approach in modeling streambed erosion is to combine the interaction of a viscous flow with is The boundary condition by the NavierStokes equation. which The viscous flow is characterized an erodable streambed. represent constraints boundary obtained by balancing sediment mass, momentum, and energy. In order to examine the erosion process mathematically, we must first have discussion. under concise terminology which will be common throughout our Streambed erosion, as a result of the general category of fluid dynamics. fluid falls we begin with Thus, the background and terminology of fluid dynamics as flow, it relates to the problem. Though dimensions the for problem all is set derivations. in two dimensions, All ambiguity is we use three removed by considering the problem to be constant in the third dimension. The theory of viscous fluids is presented in chapter II. We wish to describe flows in an open channel. We explain why we choose the NavierStokes equation to describe our flows. In general, we require some initial conditions as well as some boundary conditions. 3 How and why we choose such conditions is included in our discussion. theory. The second part of chapter II highlights sedimentation results Empirical and we describe various phenomena outlined are The major associated with streambed erosion. evolutionary and streambed minor A brief history of sedimentation changes are defined. theory is included. over Flow bed of sand causes the bed to undergo a distinct a increasing With flat. the bed remains At slow flows, evolution. flow velocity, the bed passes through a set of stages: ripples, then dunes, Past antidunes. which have not considered a general flow, models, One of our main purposes cannot address these evolutionary changes. is to have a potentially model finally and then waves then flat again with chaotic flow, capable of this addressing evolutionary process. shape evolution. streambed we discuss in chapter II, dispersion. Due the to The various stages of transport, are rolling, saltation and transport We make a distinction between sediment particle which suspension/ and mathematical difficulties associated with turbulence we shall concentrate our study on erosion resulting from rolling and saltation. The formation irregularities or of ripples fluctuating is usually turbulence. attributed It is to bed our hope that, bifurcation eventually, our model will show this evolution to be a phenomenon. We note model requirements which will achieve this aim. In chapter III we derive our model. version of the NavierStokes system to First, we give the precise be used to describe Our 4 second The flows. half of our modeling process involves depicting We derive the flow/streambed interaction. The condition. result is a boundary streambed the boundary value free 2dimensional, problem with two free boundaries. Generally, models past have been formed from a large variety of nondimensional parameters. relations These models usually attempt to match a Our with empirical heavily model strictly derived is Our empirical from balance laws. We obtain a relations are left undetermined and constants are used. streambed model which exhibits the general nature of emphasis towards is streambed. the interaction Our goal is to have a model between Our erosion. and the flow the can which process. erosion specific studied be to better understand this interaction. that the streambed is homogeneous and We allow the transport to vary but the concentration of sediment is assumed to remain constant. broad small, of consists noncohesive, relatively spherical particles. rate we assume the process of obtaining the boundary condition, In We show that under these assumptions, enough that we obtain some nice results. remains the model Specifically, we obtain classic dune movement. The complicated nature of the general model makes it prudent to consider first some special consider almost uniform, flow cases steady flow. of flows. In chapter IV we We also consider a potential in 0 combined with a logarithmic velocity profile in r. These flows yield the classic dune migration behavior. In chapter V we consider a general flow. This case is solved 5 We transform our free domain time dependent finite elements. using This fixed boundary value problem to a fixed domain. into system a solved numerically. boundary differential equations which are then ordinary of evolution solution The well as converted is as transform is required to obtain the the system this to evolution. flow boundary the contains No inverse directly. evolution This 'solution' is then presented graphically. We fundamental vorticity factor the flow near have the potential the bottom as a Another main point is the erosion process. in will that this model of model. One main point is also contrast our model with previous theory. that we have the the upgrading The last chapter presents some ideas on to address some open questions which most other models cannot. We conclude our work with a discussion of some of these open questions which have arisen as a result of this work. How can the numerical scheme be improved? For example: How can the boundary evolution be described as a bifurcation phenomenon? 6 BACKGROUND OF FLUID DYNAMICS AND SEDIMENTATION The process of erosion has been around since before man and was studied by primitive societies. erosion control on Mesa Verde. of experimented Anasazi is the father of modern the postulated law of fluid with Fluid Dynamics, the scientific study has been around for quite some time too. flows, dynamics, was it Although Euler Archemedies who buoyancy which could be considered a first step in the study of sedimentation. fluids The We use the theory to examine the forces which can induce erosion. Stokes equation is the standard description of a of viscous The Navier- general viscous fluid. A fairly complete discussion of fluid dynamics may be found in [18], [22], [24] and [33]. We highlight those aspects pertinent to the erosion problem. Fluid Dynamics The NavierStokes Equation The NavierStokes equation is coupled with a conservation of mass equation which is often refered to as the continuity equation. The continuity equation states that the net flux of mass into a 7 volume must be the same as the rate change of of mass that in volume. We derive all our relations in To setting. 2dimensions, which is set in our problem, avoid any ambiguities, 3dimensional a may be thought of as uniform in the third dimension. D. Let denoted av denote by x=(xl,xxs). u=(u(x,t),v(x,t)) Generally, we and p(x,t) be the let mass Equating the rate of change of density of the fluid. Let x=(x,y). only use field velocity the be Let points in the domain be surface of V. the in the fluid domain volume, Let V be an arbitrary, but fixed, the to mass flux across the surface of V we obtain: didt fvpdv = fv[apiatldv = fav[pui,u1da = fv[divpuldv in The first term is the total change of mass term is the The pulling of differentiation the The second inside the The third term represents the flux across the surface integral. V. result V. (2.0) last term is obtained by of applying the Gauss divergence theorem. Since V is arbitrary, we obtain the mass balance equation in D: pt flows For the type of relatively constant. Such (2.1) + div pu = 0 we will be considering, density is incompressible or divergence free flows are described by a velocity field which satisfies (2.2) div u = 0. The NavierStokes equation, derived from the Balance describe viscous flows. of Momentum Law: They are the rate of change of momentum of an arbitrary volume must be balanced by the stress over 8 the surface of the volume plus the body forces acting on the volume. [33]. V Let an arbitrary, be but fixed, V is also refered to as in the fluid domain D. fluid, be the pressure field and let g(x,t) be the density The stresses surface B Let let p=p(x,t) be of the fluid. are given by TU. (Cauchy's stress hypothesis) where n is the outward normal to V [33]. material a u=(u(x,t),v(x,t)) be the velocity field, Let volume. moving with the volume, and T stress the is tensor. Equating the the body forces per unit mass on V. time rate of change of momentum to the forces on V we obtain: fv[DgmiDt]dv = favTnda + fvplidv = fv(div T + gDldv (2.3) where DraDt = ut + (u0V)n (2.4) is called the material derivative. The first term is the momentum change of V. the right of terms The two on the first equal sign are the surface forces and body forces respectively. The last term is obtained by applying the divergence theorem to the surface integral. We assume that, for an incompressible fluid, the stress tensor the T depends linearly on the pressure and on the rate of strain of fluid D: T = pI + 2gD, D = (VU + VUr)/2 The first term is the normal stress and stress. We assume that viscosity, neglecting dependence on temperature, boundary turbulence. [33]. (2.5) the second P, is sediment is shear the constant, thereby concentration, The only body force for our problem and is 9 We gravity. let g denote gravity the vector. standard rectilinear coordinate system with the use a will We upwards direction positive. Since V is arbitrary, we obtain in D: Dpu/Dt = div T Taking to p be produces constant (2.6) pg. classical the NavierStokes equation: p(ut + (voN)u) = Vp + p.Au pg (2.7) With suitable boundary conditions, the system (2.1), (2.6) will describe moderate erosion mixture density inducing If flows. the fluid/sediment does not vary greatly from the pure fluid density, much then we can describe the flow by the NavierStokes (2.7). equations (2.2), simpler Hereafter, system will be refered to as the NS system or the incompressible this complete NavierStokes equations with the density clarified by the context. Boundary Conditions We wish to describe flows in an open channel, that is, one with an unconstrained upper surface. We call this free surface n, postulate that the stress is continuous across n, of D. If the atmospheric pressure is and the upper surface normalized to zero, the viscous effects of the air are neglected, and the surface tension of the fluid is considered negligible, we obtain on n (2.8) Tn=0. No constraint is placed directly on u on n. We obtain n(x,t) from: 10 (2.9) an/at = v(x,n,t). the standard no slip boundary condition is On a fixed surface, We call the 'solid' streambed boundary i. imposed. For fixed i, we obtain on 4: u=0 or u=0=v (2.10) For a solid but moving object (such as a ship) the no slip condition would still apply resulting in u equal to the object velocity. Initial Conditions for open In general, velocity field channel on an initial domain. an The initial domain, The pressure required to satisfy (2.9). require we flow, field can initial 00, is calculated be from the velocity field. A that flow is independent of a spatial direction is called uniform flow and flow that is independent of time is The NS equations flow. uniform or required, steady. For are steady called steady greatly simplified when the flow is no flows, initial condition is and the absence of an explicit expression for ap/at is no longer troublesome. The kinematics of steady flow are often similar to average flow especially for slow open channel flows. however, even though the flow In appears the case of erosion, steady and uniform, the streambed is in motion. We can consider the special case where the bulk of the flow is almost steady and uniform but we are forced to treat a general flow 11 as unsteady unsteady, and There is always at least a region of nonuniform. This time dependence nonuniform flow near the streambed. requires that we have some initial conditions which we choose to be u(x,y,t)=u0(x,y,0) on Do=[(x,y)I (x,0) < y ( 4(x,0)) (2.11) Dimensional Analysis analysis Dimensional is useful a Processes which are not well understood, too complicated, by reformulated flow, the fluid dynamics. in for which the physics using Buckingham the Pi Theorem. The [17]. which are assumed to affect the are mixed and matched to yield a characteristic invariant correct incorporated in relation. The are or for which a scaled model is required are often dimensional variables, independent tool dimensional nondimensional combination and the function of the The units. remaining invariants in a variables new of are functional of the first invariant (with dimension) invariants remaining (without dimension) will yield the fundamental form for the unknown if all the necessary dependencies were initially included. By way of example, suppose that the magnitude of the force, F, trying to roll a particle is a result of the fluid moving around the particle with velocity, u, and viscosity, g. about a disk shaped particle of diameter ds Taking u as constant and applying Stokes' Theorem we have 2 F=k wt(d) (w=curl u) where k is the appropriate constant. (2.12) 12 suppose Now, instead that assume a form for F, we such as 2 F=F(w,g,ds,x1,x2,...). (.1 where [F]=ML/T , denotes So, . M denotes dimensions of dimension, and T denotes dimensions L denotes dimensions of length, mass, mass length/time units has Force of time. We choose three variables as the set of fundamental parameters. variable Each succeeding fundamental be represented w, g, and Choosing set. yields: ML/T2=[w] will terms in of the as the fundamental set ds awbid ]c. There is always a unique quantity formed if the parameters dimensionally independent. In this case, ML/T =[wligi[ds] 2 F=wg(ds) f(I/ ,...). X Where dimensional quantity relation f, which is nondimensional, Reducing F=k the wg(ds)2 using w, g, ds, is and x.. functional The empirically. determined functional relation to a simple constant, as before. Thus . or non invariant an is IIx are 2 2 k, yields k would be To get a final working relation, determined empirically. The Reynolds Number and Other Parameters For NavierStokes flows involving sedimentation, the Reynolds 2 number, Re=ULp/g, and the Froude number, Fr=U /Lg, are taken to be the nondimensional parameters which best characterize the flow. U is a representative velocity, L is a representative length, and g is the magnitude of g. The Reynolds number essentially measures the 13 ratio of the inertial to the viscous forces while the Fronde number measures the ratio of the inertial to the gravity forces. Although Re and appearing in Fr are the NavierStokes the only nondimensional system, another parameters nondimensional quantity is often encountered in sedimentation theory. Let V be the kinematic viscosity stress on the boundary. Then, g/p. Let to be shear the the 'local' Reynolds number is given by Re*=u*L/v, where u* is called the shear velocity and L is a local length. Re* is assumed u*=,/to/pf. to be good a characteristic parameter for several unknown functions in sedimentation theory. usually One sets simplified to u=u(y). functions match L=ds and The bottom empirical data? to=du/dy where the flow has been line In how is: the well previous do unknown section, the functional relation f qUite often becomes f=f(Re,Re*,Fr,...). In nondimensional form the incompressible NavierStokes system becomes: Du/Bt = Vp + 2D/Re y/Fr (2.13a) div u = 0 (2.13b) where y is the direction of the gravity vector. Vorticitv and Potential Flow The vorticity of a flow is given by w, the curl of the velocity field (recall that in 2d flow w is a scalar). measure The vorticity is a of the rotational strength or mixing strength (circulation) of the flow. Irrotational flows, flows for which w=0, have some 14 special properties. An irrotational flow is circulation free. integral of u around any closed curve is zero Thus, Stokes' Thoerem. u must be the gradient of a function t or equivalently u is a potential. lines, by The t=c are the curves w=0=divu. which along (see Fig 2.0) equipotential curves the fluid moves, t may be found from At=0, The [22]. for pressure a u. stream The are orthogonal to t. direct consequence of be recovered using Bernoulli's can equation. [22]. i t=c1 t=c, \ I , t f t=c, I\ 166. flow - 1 1 I I I .. 1 .7 1 I I I - \ I I //z,/ Viz 1 Potential Flow Fig 2.0 Even for rotational flows, the NavierStokes system in a it may be advantageous to express single equation. applying the curl operator to (2.13). [33]. This is done by We obtain the vorticity 15 transport equation Dw/Dt = Aw/Re (2.14a) w=0 on n (2.14b) u=0=v on 4 (2.14c) u is easily recovered from w. [33]. The lower boundary however, is in terms of u rather than w. condition, [33]. Although mainstream flow may be approximated by potential flow, the flow in the boundary region cannot. Near the boundary, we must either use the full NavierStokes equation, some simplified version, or an empirical relation. The Law of the Wall and Boundary Lavers Our sediment transport process (that is nonconstant transport) will be restricted to the region 'near' the streambed. is often considered the boundary layer. although [31]. Our theory. We refer terminology, similar to that of classical boundary layer theory, slightly different interpretation. classical layer region There is a branch of fluid dynamics devoted strictly to boundary layer theory. to this as the classical boundary This boundary layer We reserve the has a application of theory to the erosion problem for future works. One standard empirical approach to boundary layer flow is given by the logarithmic velocity profile or the 'law of was developed by v. The normal velocity Karmen and Prandtl. [32]. the wall'. It gradient, du/dz, is taken as a function of p, T, and z, the density, 16 stress, and normal distance respectivly. By dimensional analysis, the only combination of these parameters which can occur is du/dz=i(s/p)/kz where k is an determined by v. empirical constant. Karmen. [321. (2.15) This constant was empirically Upon integration we obtain the Law of the Wall U = u* where u* is the shear velocity mentioned Karmen constant 0.417, z is thickness of stationary fluid. For of (2.16) In(z/zo) the earlier, normal fluid. [28]. the von distance and zo is the there is always a region (see Fig 2.2) 20=g/9pu* is generally used for smooth boundaries and z0=c1s/30 for sandy ds is (see Fig 2.1) a boundary composed of sediment, stationary K boundaries where is the diameter for which 65% of the sand is finer. [28]. The obvious advantage of (2.16) boundary shape. is its independence of the Note, however, that only the dominant term for u is given while assuming laminar flow conditions. 17 = u* ln(z/zo) .0* Fig 2.2 The Law of the Wall Fig 2.1 The Streambed Close up Fig 2.2 18 Energy Dissivation A discussion of energy dissipation is given by Serrin, [33]. more complete discussion is given by Lamb, [21]. We give A the highlights here. The kinetic energy, in an arbitrary volume, V, of the fluid, is given by E = fV [pu 2 /2)dv (2.17) 2 where u =u.u. Such energy is energy. The fluid volume has potential energy as well. not That is, but rather converted kinetic to all the energy dissipated directly into the fluid and surroundings, kinetic energy. dissipated due to frictional effects, must come from the We have in V E = dE/dt fvuadv) = fv8u2/dtdv = pfvusutdv. (2.18) Substituting for ut from NS we obtain E = -IvEdiv[pli(u /2 + p/p) The transport 2pusD1 20:Vu1dv. term in parentheses and is the same as for an ideal fluid. is the energy flux (2.19) due to fluid [33]. The term 2gusl) is the velocity times the stress tensor and is the energy flux due to the frictional forces. The last term represents the actual dissipation which is lost to heat. The energy is transfered from large eddies to smaller ones it and is there that the energy is finally dissipated to heat energy or available to the sediment. Applying the divergence theorem to (2.19) and infinity or uon=0 at finite surfaces we obtain in V using u=0 at 19 , 2 E = 2t1ivD:Vudv = 2g1vD dv (2.20) 2 where D =D:D. (2.20), for energy dissipation, which is a standard form for the expression may be rewritten as the BobyleffForsythe formula Du z E = w dv (2.21) 2gfav (TiTen)da. In the case of irrotational flow, we have in V E = 2gfav (u.Dn) da = 21ilv (Yu) (2.22) dv. Some consequences of (2.20), (2.21) and (2.22): In the regions of irrotational flow there is energy dissipation if and only if the flow is nonuniform. The greatest dissipation occurs where the fluid has the largest gradients. If the entire flow is irrotational, then there is no energy dissipation and hence no erosion could take place. A flow which minimizes energy loss must tend towards potential flow. Frictional dissipation of the fluid does not allow for energy to be transfered across a solid no slip boundary. If the flow across ay is steady and uniform then the dissipation depends solely on the vorticity in V. Concluding Remarks Fluid Dynamics Equations (2.4) and (2.7) with boundary conditions (2.9) and 23 Rolling and saltation occur heavier this particles manner. associated (relatively Fig (see with 2.3) turbulent all in eroding speaking) Due difficulties associated with turbulence the primarily transported in Suspension/dispersion flows. with flows we to the shall primarily is mathematical concentrate our study on erosion resulting from rolling and saltation. Empirical Results Boundary Evolution The flow over a bed of sand will cause the distinct evolution. We distinguish between changes and 'minor' evolutionary changes. remains flat. ripples, then waves and changes. Ripples This undergo a 'major' evolutionary At slow flows, the bed then dunes, then flat again with chaotic finally antidunes. These are the 'major' are distinguished from dunes in that dunes have a distinctive form and upstream. to With increasing flow velocity, the bed passes through a set of stages: flow, bed they migrate migration and downstream. the classic Antidunes migrate forms which arise are 'minor' changes. (see Fig 2.4) The formation ripples of is usually irregularities or fluctuating turbulence. dune stages, rolling transport mechanisms. increased and and In moderate the [28]. saltation transition suspension/dispersion attributed becomes to bed In the ripple and are stage, the saltation nontrivial. antidune stage, saltation and dispersion are extensive. primary In is the 24 . :*-- ////77 /,/// 7%/7777771,/7 7 ' smooth /.7/7-/ /7/). (x t) streambed- steady, uniform flow ______ n(x,t) random ripples 4(x,t) 'minor' evolution dunes- migrating downstream as 'smooth' scoured streambed- turbulent flow Fr - 1.0 - 1.5 n(x,t) 4(x,t) evolution antidunes- migrating upstream as 'minor' 'Major' and 'Minor' Streambed Evolution Stages Fig 2.4 25 Few models have made attempt an account to however, their changes. Once explain. Antidunes eventually form when the the side lee particle lift. the dunes form, is easy to migration on gradient pressure with correlated These stages appear to be strongly Froude number. The boundary shape evolves as the Froude number We shall concentrate our study on model these of the dune becomes strong enough to cause extensive 1.0-1.5. increases with antidunes forming when Fr Our all for this evolution phenomenon. will attempt to account for these evolutionary stages as well as giving an explaination for the initial dune formation. A Brief History of Sedimentation Theory A fairly complete history of recent sedimentation models is given by Allen, [1], and Vanoni, [39]. transport Bed load transport models have generally been developed from one of five main themes: average fluid velocity, streambed shear stress, probabilistic inferences, bedform celerity relationships, energetics. due Probably the first formula involving an erosion process is to Brahms, 1753, and involves the average stream velocity. found the empirical relation, 1/6 V =kW c from some basic , between a critcal velocity s and the weight of an object it can move. derived Brahms assumptions. This relation may also [4]. It is used be today 26 primarily in engineering design. Gilbert, Muller, 1914; 1948; Donat, et. al. 1929; Straub, developed MeyerPeter and 1939; formulae of the form qs=qs(Un). 2 2 qs=kU (U Uc), Other similar etc. models assume that sediment discharge properties may be characterized by the mainstream Reynolds number, local Reynolds q =q s(Re,Re*,Fr) ). s on characteristic number, from Froude number (i.e. These are empirical formulae which depend values and discharge rate correlates well arising and these not local values. with these parameters The only sediment but models expressions do not deal with distinct boundary movement but only the average effect of the stream on the sediment It was the idea of a critical force which led DuBoys, in 1879, load. to derive a discharge, qs, general equation qs=kTo(To Tc) for the in terms of a critical tractive force, Tc is the bed refinements to shear his and k model improving the expression for is an empirical constant. exist, Tc. most of which approach is attributed to H. A. Einstein, flat where To A number of concentrate on [38]. Most of the theory of sediment transport from 1947. sediment 1942, a probabilistic 1952 and Kalinske, Einstein wanted to account for the initiation of ripples on a streambed without resorting to the irregularity arguement. assumed normally distributed turbulence fluctuations to account Be for the initial uneven erosion. The bedform celerity models such as Crickmore, those of Kennedy, 1960; 1970; Korchokha, 1972; Willis and Kennedy, 1977; at. el. 27 are strictly (height, empirical length, formulae etc) to the relating dune perhaps the most extensive and one of dune celerity. his characteristics Kennedy's work is models is given here. 2 [20]. Xs = (27r/g)U gives the dune wavelength, in terms of a Xs, mean velocity, U. g is the magnitude of gravity. Models based on the energy of Bagnold, 1966. considerations These and models are flow are generally Gilbert, 1914; qs=kt°U. the largely are derived from due to theoretical of the form q5=k(t0U0 scy or MeyerPeter, 1948; Straub, 1939; et. al. have developed models based on the energy slope (stream gradient). In summary, we note that, dimensional analysis employed to arrive at the fundamental dependence Of the present considered as difficulties models, vorticity fundamental a associated does not of often buried unknowns. parameter. The mathematical with turbulence makes it difficult to deal The flow field in a constant such as the average stream velocity allowing for no local variation. only the often appear to have been with the true flow field of typical erosion flows. is is The probabilistic models are the ones which seem to deal with the boundary evolution stage from theoretical considerations. Forces on the Sediment Particle The global forces terms or on the very sediment specific are usually expressed in either terms. In global frictional force on a smooth flat surface is given by terms, the 28 wro=gdu/dz where is z normal to the (2.23) boundary. individual particles are also considered. lift the particle out of angle. In entire surface. actuallity, specific They try forces to on pivot and its position as shown in Fig 2.5. The moment is written in terms of limo, pivot The the particle weight, and the the fluid forces should apply to the We will be using a slightly different force diagram in our model derivation. INNY/M Classical Force Diagram for a Sediment Particle Fig 2.5 Concluding Remarks Sedimentation theory of sediment transport, from the theory though related, of boundary evolution. is distinct Most of the work has been 29 concentrated developed in sediment initially from empirical in nature. transport physical theory. The models, considerations, become It appears that only the probabilistic though very models have addressed the question of initial dune formation arising from a smooth streambed. There does not appear to be any accepted theory of the actual process of dune formation. of coagulation around some stand up to close scrutiny. yield ripples and evolutionary dunes. changes irregularity Increased A model must The generally held combine reasonable streambed erosion constraint. in the surface does not sediment which a belief can homogeniety address general flow the with still major some 30 THE EROSION MODEL Our basic approach in modeling streambed erosion is to the interaction of a flow with represent an erodable boundary. general viscous flow (i.e. some boundary constraints which The the combine flow characterized is NavierStokes by equation). a The boundary condition is obtained by balancing sediment mass, momentum, and energy. its The erosion model consists of open channel flow together boundary give condition overview of the problem assumptions along with erosion. some We terminology. a brief The basic pertaining to the model are noted along with the degree of their validity. is representing with derived. The erosion constraint as a Some boundary condition special cases of flows are considered in chapter IV and a general flow is considered in chapter V. We use the term erosion to encompass both the scouring, or lifting up, of sediment. D, is divided into two regions: region sediment transport in 0. deposition The sediment in 0 r. and the 0, We assume negligible behaves as if it fully dissolved and does not itself affect the sediment load in contains a mixture of this mixture as a fluid. interface, is the streambed, free, both fluid and sediment. The upper surface and is denoted by is also free, 71. and The entire fluid domain, the mainstream region, which sediment is transported, in the of is r. r We will refer to 0, the air/fluid The lower surface of and is denoted by i. 'Surface' r, will 31 refer to i while the 'bed' will refer to the streambed, interface is denoted by a. The 0/r i. Refer to Fig 1.0 for a schematic. The Flow Field We assume that the erosion process is a local result of the interaction the streambed. take between phenomenon, the the fluid and sediment along Erosion depends on the local field flow which we to be governed by the NavierStokes Equations for open channel flow. Difficulties arise concerning the sediment conservation of mass. The particles are discrete rigid masses but they cannot easily be modeled as kinematics dense points the of continuum. Fortunately, the of the flow do not depend strongly upon small amounts of sediment in solution. [28). concentrated near in 0 by incompressible the The the boundary. sediment transport is assumed This allows us to govern our flow NavierStokes equation for constant Let x=(11,12,x3) with the upwards direction positive. Through- density. out our discussion we assume the flow direction. Thus, we equate x with to be (x,y) constant where in any the x, ambiguity will be clarified. We let u=(u(x,t),v(x,t)) be the velocity field and p=p(x,t) the pressure field in D. atmospheric pressure is zero. We normalize the pressure so be that We let p=p(x,t) denote the density of the fluid (mixture) and the density of the bed sediment is denoted 32 by We ps. streambed. assume constant is p in 0 and ps is constant in the The flow in 0 is then governed by: pDm/Dt = div T + pg div u = 0 p=0 (3.0a) in 0(t) (3.0b) on 71(x,t) TU=0 lt uo in 0(t) , vo, (3.0c) on A(x,t) =vI (3.0e) y=11 given at o (3.0d) (3.00 t-0 where D/Dt:=8/8t + (u.N) is the material or spatial derivative, T = pI + 2gD, D = (Vu + VuT)/2 and (3.0h) defines the stressstrain relationship of the fluid. is the NavierStokes (a) tensor and g being the with equation gravity vector. T being the stress is the Continuity (b) equation for constant density. (c) represents normalized atmospheric pressure. (d) represents continuous stress accross auxiliary condition resulting from (d). (f) n. (e) gives is an the initial conditions. The system (3.0) is not complete boundary boundary, condition. This but requires an additional condition cannot be given on a, since the flow moves accross that interface the 04 arbitrarily. It is the sediment, not the fluid, which is trapped in U. We this, would like to extend (3.0) to include we assume that the depth of thickness of r and let h denote the 'small' relative to h. f is 'small'. flow (see Fig 3.0) depth. f. Let 6 We To accomplish denote assume the 6 is It is important to note that 8 is not being used in the sense of classical boundary layer theory. 33 h >> The Flow Regions 0 and I Fig 3.0 We interpret the region r as follows: The region r region D. be is infintesimal with respect to the This is a mathematical simplification which may physically interpreted as sediment load moving the along the bed as an impulse load. Thus, the sediment load is no longer a function of the vertical displacement but only a function of position along i. The mainstream kinematics inherent taken to process. be The simplifications the flow in kinematics the primary classical and boundary layer NavierStokes influence boundary on layer equation the transport theory to arrive at the same effects. are uses The trade 34 off this at point matching between is two sets of equations; one set for the main stream and one set for the boundary versus layer one equations of set and an infintesimal boundary layer. There is a natural boundary condition on Since S is 'small', condition which we adopt. considered as a along only flow 'distinguish between D and 0. dominated by the flow in 0. the Physically, namely, the no slip r the flow in streambed. We no longer totally the flow in D is The system (3.0) becomes: pDu/Dt = div T + pg div u = 0 p=0 (3.1a) in D(t) (3.1b) in D(t) on u(x,t) TU=0 (3.1c) on n(x,t) nt u=0=v =v1 (3.1d) (3.1e) y=11 (3.1f) on i(x,t) 110, vo Do given at t The flow domain 0 has been now completely determined by (3.1). (3.1g) extended boundary constraint, (f) has been added. to D and the no slip For a given 4, the flow is This no slip condition does not preclude erosion but rather states that there is a the bed is fixed and, Realistically, agitated to the bottom. at that level, level which at the fluid adheres to the bed. the bed has fluid moving through mixture can be it and r is an of the fluid and sediment moving along very close We have simplified the picture by assuming that the D > 0. The sediment moves along so close to the streambed that 35 'boundary' is made up of a movable solid. Streambed Erosion as a Boundary Condition We now have the flow given by condition, model. which require and (3.1) reflects the process of erosion, a boundary to complete the We obtain this condition from the conservation of mass law for the sediment. r. Let V be an arbitrary volume in of sediment (mass/volume) in Let cs be the concentration The total sediment mass in V is V. given by Iv cs dv (3.3) and the rate of mass gain is given by r dt JV cs For fine sediment, is a continuous function. cs c dtI Vs Let us (3.4) dv. dv = So, IV 8c/at dv. (3.5) s denote the velocity of sediment particles. Then, the mass flux into V is given by f ay csus.n (3.6) da where 8V denotes the boundary of V and n is the outward unit normal. By the Divergence Theorem we have IaV cu s da = I I div c u V s s dv. (3.7) Since mass is conserved, we have in r IV 8c/at dv = S V div c u s dv. (3.8) s We let 6 > 0, the width of r shrinks, and let av coincide with i, the bed. We assume that the sediment is concentrated in a thin 36 layer along i and is uniform in that layer. (3.8) becomes sfai aus/at dv =ai k =c c [1 S p s fatc u )/at + ki) dv s (P) S 2 _1/2 (3.9a) st (3.9b) 1 t is the where ad denotes the portion of av which coincides with 4. unit tangent vector to 4 and i. refers a/at us is the sediment velocity parallel to to the change along 4. (100% concentration) to ks relates the bed density the bed for a given change in 4, by c, cs packing ratio. Since V was arbitrary, ad is arbitrary and we have ac/at = 8(u u )/at SS S (3.10) ksdt. term on the left represents the total increase in sediment The mass due to locally unsteady flow. The term first the on right represents the tangential flux of sediment due to locally nonuniform (tangential) flow. The last represents term the normal flux of sediment which is a source/sink term due to erosion. We would like explicit forms for cs, and us. We restrict to quasiequilibrium erosion where changes in the sediment ourselves concentration are 'small'. A significant sediment flux accross i is always accounted for by a relative increase in the transport rate. In this way, changes in the sediment concentration remain small. this case, cs In is constant and (3.10) becomes cpdt=dus/dx (3.11) where d/dx=(8/8x + a/ay dy/dx)I y=d(x)* We cannot assume to be constant for if we were to make such us an assumption, no erosion would occur since the right side of (3.11) vanishes. There are a variety of possibilities for determining us 37 including: that the sediment rolls with the fluid. Assume depends on the energy dissipated by sediment the Assume the rate of work in transporting the fluid. (3)Usedimensionalanalysisonherep.are parameters influencing sediment transport. We consider Method both (1) and (2) and make some remarks concerning (3). (1): Obtaining expression an for us by treating the fluid as a collection of rolling, spinning fluid volumes Sediment Particle Transport Mechanism Fig 3.1 The pps. a sediment They are spun, similar volume made is up of small homogeneous particles with rolled or agitated in much the same manner as of immersed in a flow would fluid. If quickly we neglect assume the resistance, same local a ball angular 38 We derive velocity as the fluid. rolls us for the case where the sediment conglomeration of assumed to roll and spin in a similar manner. Let Us Us denotes = (l/nd ) S be Then, Ias uds (3.12) Applying Stokes' Theorem we boundary of S. the 8 Let Let u be the fluid velocity, the fluid being adhered to the particle. as (see Fig 3.1) be the mean surface velocity of S. the mean rotational velocity of S. where is a particle S of mean diameter ds rolling and spinning Consider in the fluid. sediment The volumes. spinning rolling, we treat the fluid as a That is, spins with the fluid. and have Us = (//nds) But u =nd 0 and s s (3.13) An application of the Mean Value Theorem 0=U /nd . s IS curl uda $ yields Us (3.14) wds/4n where w is the magnitude of the vorticity at (x,i(x)). We obtained (3.14) by neglecting resistance. include We the presence of resistance in as simple a way as possible by assuming Uc to be a critical threshold which must Us exceed. We obtain the expression for us us where dir = = (3.15) U dir[Us dir denotes the direction of us. [.]+ denotes the positive part of (.]. In general, U =U (d, C c movement for the sediment. s e, y,...) $ gives the resistance to That is, the resistance is a function of 39 parameters such as particle the particle weight, the bed slope, etc. Uc = k(4' + e where angle the is constant. empirical the For simplicity, we take (3.16) ec) the sediment and k is an for repose of the particle shape, size, form includes only bed slope effects. This more genera/ relation could be applying by obtained A dimensional analysis. Recall Paradox which says that there is no force D'Alembert's on a particle in a perfect constant which The takes becomes more us particle size and weight. resistance. sediment the must depend on the viscosity as well as distinct from the fluid, all As fluid. Recall that Us was derived by term these incorporate should Us properties empirical an account. into neglecting first The generalization of (3.15) is us = dir[cito Method (2): energy (3.17) c5(4' + Oc)]+. Obtaining an expression for us by equating the dissipated by the fluid with the work rate of sediment being transported The energy dissipated by the fluid must be either converted to heat or transfered to the sediment. Energy dissipation is carried out in the kinetic energy, manner: as they decompose, Energy is transfered from the larger eddies, smaller ones with little loss. following to The small eddies then transfer their usually to heat but also to sediment transport. 40 Within large the vorticity eddies, flow the is almost irrotational while dominates in the smaller eddies. is consistent with our interpretation of This explanation [21]. (2.21). From (2.21), we deduced that energy dissipation was minimized by potential flows. Let Ef by be the rate energy is dissipated the Then, fluid. the rate energy becomes available to the sediment, for transport, is given by Ea = [efficiency} x Clearly, some properties or to Energy is unavailable and lost energy]. fEf energy sediment/boundary 'unavailable' due independent of the erosion process. friction on a cohesive boundary. characteristics. the to heating Energy is Obviously, remaining energy is available to contribute by performed the sediment due to fluid lost or unavailable is of (3.18) For example: fluid which is due 'lost' to the only a portion of this the to rate work of to the 'efficiency' of the energy due transfer process. Let lost Ec Ec minimum amount be that critical, of energy initially or unavailable and let e be the efficiency of energy transfer. accounts for the which energy must whether or not erosion takes place. Ea = elEf be dissipated regardless (3.18) becomes: Ecl (3.19) The total energy dissipated by the fluid is given by (2.21) or (2.22) and may be alternatively written as E = giv 2 (to 4detlifu] dv where g is the viscosity of the fluid. (3.20) 41 The term detVu measures the dissipation due to deforming V and we assume this energy to be 'unavailable', leaving: 2 Ea = e[gw With (3.21) Ec] we assume that potential energy regards to the sediment, changes. changes are negligible relative to the kinetic energy sediment lifted is only Losses due to an infintesimal distance. heating are absorbed in Ec and e. all the work is Thus, The contained in transporting the sediment and is given by (3.22) u f s s where fs is the force on the sediment particles which causes them to roll and spin. There are a variety of ways to arrive at an expression for We give two. obvious are analysis, fs. First, we assume fs=fs(w,ds,1i,p1,P2....). wids, and g By dimensional choices as fundamental parameters. and g is given by the only quantity of force using w,ds, Thus, gw. (3.23) fs=gwf wherefrepresentsf(111,...)H.being quantities formed from p, w, ds, and pi. the nondimensional The simplest form for fs is then (3.24) fs=kgw We obtain the same expression for fs by assuming the force to result of rolling a particle where the surface force is given be a by gf(u). Applying Stokes' Theorem to f(u) kill gives the force per particle as kigw. (3.25) 42 The number of particles is constant so we again obtain (3.24). we relate the work rate With fs now given, the to available energy to obtain (3.26) E ] +/f = ergo) us s C which becomes us = dir[c w where we interpret those (3.27) r]+ grouped terms into as a result of r resistance factors. We have obtained a similar expression for Method Obtaining (3): an expression us by method (1). for from strictly us empirical considerations standard Another dimensional analysis. method for determining would The only question us would be to use be the choice of Consider that the vorticity plays a role in fundamental parameters. is a measure of the local turbulence, the fluid energy dissipation, is a measure of the local agitation at the bottom, and is related to Clearly, the rolling and spinning of the fluid and the sediment. would be a good choice for erosion. a parameter of streambed However, w does not appear to have been previously used as fundamental implied uniform expressions for parameter. fundamental a w One parameter. flow us field could Additionally, be obvious explanation for this is the most inherent in obtained using improving model would be best achieved by the replacing w models. as physical empirical a Countless fundamental reality of the constants by 43 We leave such additional empirical relations using such parameters. versions for later studies. We choose our as (3.17) working version for us for the Our final version of our model becomes remainder of this work. pft/Dt = div T + pg div u = 0 p=0 in D(t) (3.28b) in D(t) (3.28c) on n(x,t) TU=0 (3.28a) (3.28d) on 11(x,t) (3.28e) t=vIy=11 u=0=v (3.280 on i(x,t) (3.28h) cpt = dus/dx us where c2(05' + 9c (3.28g) u0, vo, D, given at t0 (3.28h) = dir[clw d/dx=(a/8x + a/ay dy/dx)ly=i00- Concluding Remarks The Model The complete model, boundary condition, consisting of a flow field and a streambed is given by (3.28). Given below are the high- lights of the model and its derivation. 1. The erosion process consists with an erodable boundary. of a flow interacting p=0 Du Dt = div T + B div u = 0 flow 0. 11017 fi".L 001' 4 lift mi. aft...111F .. The Erosion Model An Overview Fig 3.2 45 sediment The erosion process is obtained by balancing mass, energy and momentum. The erosion producing flow is governed by the Navier- Stokes equations for open channel flow. The sediment transport occurs in a 'thin' layer along the stream bed. Our model is for a relatively fine, Large discreet particles The the invalidate mass derivation as well as the assumptions used to balance obtain would round sediment. us. sediment quasiequilibrium. boundary, must transport results be in a state of Increased sediment flux, across the in an increased transport rate rather than an increased sediment concentration. The vorticity of the fundamental factor for flow the near the streambed is a erosion process in our model. The model will accept a variable flow field. Such a flow field is necessary for variable erosion. An overview of the complete model is shown in Fig 3.2. 46 SPECIAL FLOWS We consider three flow cases: almost uniform, steady flow, potential flow in 0 with 'law of the wall' in r, a general flow. Due to the complexity of a general flow, (3) is dealt with in chapter V. Almost Steady, Uniform Flow 4=0 //1 Steady, Uniform Flow Fig 4.0 If we replace the erosion condition, (3.27), with a fixed streambed condition, channel flow. If we assume a steady, (3.27g), we have in our model, standard open uniform flow over a smooth bottom, 1 would be flat also. (see Fig 4.0) In this case, u=u(y) and p=p(y) and our model, (3.27), becomes: 47 (4.0a) pgsine 0 = pu YY 0 = p (4.0b) pgcos0 u=0 on y=0 (4.0c) Tn=0 on y=h (4.0d) We solve this exactly and give the solution below. (4.1a) u = 21(sinO)y(2hy) 2g (4.1b) p = pg(cose)(hy) n(x,t) y* d(x,t 'small' /// 4. Almost Steady, Uniform Flow A Rippled Bed Fig 4.1 In actuality, such flows experience boundary turbulence. are unstable and Suppose, however, will at least that we use this flow as an approximation to the flow in the case where the streambed is relatively smooth and flat; that is, where i' is 'small'. We are, in essence, assuming this flow to be approximated by the stable flow given by (4.1). 48 boundary. small. will almost consider We (see Fig uniform steady, If the flow is too slow, then place take all. at us rippled a will be zero and no erosion flow the If over the bed gradients must be Of necessity, 4.1) flow too is strong, our In between these extreems, us is approximation will be unrealistic. nonzero and approximated by k, + kidu/dy. us Assume now that (4.2) the flow is deep enough that there exists y* constant. such that for y < y* the fluid discharge is approximately Then, i) du/dy (112)(y* qf where qf is the volumetric fluid discharge (4.3) in the region [i,y*]. This gives i) constantgy* du/dy 2 . (4.4) Substituting (4.4) into (4.2) we obtain )a/ax 3 k/(y* dus/dx (4.5) where k is the appropriate constant. Our erosion constraint (3.17) becomes t (4.6a) + cix = 0 3 C = ki(y* (4.6b) i) which is a quasilinear hyperbolic wave equation. This [41]. equation may be solved by the method of characteristics. We obtain a solution of the form i=i0(x where 40(x) is the initial downstream with speed c. (4.7) et) boundary. The initial dune moves (4.6a) indicates that the top of the dune 49 moves faster than the bottom. at the angle of In actuality, the lee side sluffs off repose of the sediment. i=460(x - ct) (see Fig 4.2). actual result Dune Migration- Almost Steady, Uniform Flow Fig 4.2 Thus, downstream with almost uniform steady, flow, assume the classical dune shape. and ripples In this scenario, if the bottom is initally flat, then no ripples would Potential Flow in 0 with Law We potential flow the Wall in the case where the consider velocity profile of in a outside region near that region. for the logarithmic velocity profile. is denoted by f form. f flow is the migrate given by a logarithmic boundary and given of the Wall' The region near the boundary We use the 'Law and the potential flow region is denoted by 0. a constant width S. (see Fig 4.3). by f is 1 --J t) flow d(x,t) Potential Flow in 2 with Law of the Vali in r Fig 4.3 < y < a) r = ((x,y)1 0 (4.8) = f(x,y)1 a < y < a = [(x',y')i x' = x be/dr, (4.9) Y' = i(x) + 8/dr} (4.10) 2 dr = [1 + (e) ] flow The in 0 is determined (4.11) in one of three ways: Method (1): Approximate uniform flow u=u(x), Method (2): Conformal mapping, Method (3): Numerically solving At=0, where u=grad t. If -11 is assumed apriori and i' is 'small' (hence &is 'small') then we can approximate the flow on a by 51 Ua = qf/h, a(x), = n(x) h (4.12) where qf is the volumetric fluid discharge and h is the depth of In general the However, numerically. for known, map, case, bed, the or case, which one maps must be obtained one for which the conformal map is is well approximated by a standard conformal then we can 'conviently' use conformal mapping. approximate bed the by In the first In the second case, we the map changes. as the bed evolves, continuously form if we assume some particular geometric particular, in conformal the 0. same the standard simple geometric form. With direct the numerical the numerical solution approach, depends on whether n is assumed apriori or taken as the continuation In the first case, we solve directly of a streamline. (4.13a) in 0 At=0 subject to the boundary'conditions: 3t/8n=unormal=0 at/an=unormal=0 at A (4.13b) at a (4.13c) In the second case, we solve the same problem but iterate on n until Bernoulli's condition, 2 u /2 + p/p + gy = constant on n, (4.14) is satisfied. Method (3) simplest. For is the deep most flows versatile while method relative to ripple height, flat and both method (1) and (3) are easy to (1) the n is almost implement. cases, some inflow and outflow constraints must be assumed. use uniform flow, method (1). is In all Here we 52 The flow in r and has a local logarithmic form. velocity, The parallel to i, is given by U = K log[(z where U zo bed, is the flow magnitude, is that distance K = IC(x,i(z)) is where necessary to + z0)/z0] (4.15) z is the normal distance above the fluid the match the velocity in where 65% of the sediment is finer than ds. r However, and with the d/30 zo is usually assumed to be of the order velocity in D. sand movement, begins for we take z=z0 to corresponds to i. from Let U = U(z;x,i(x)) denote the velocity a distance z point Let (x,i(x)). the Ua denote the velocity on a determined by the Continuity of the flow requires potential flow in 0. (4.16) Ua = U(6;x,4(x)) and K(x,i(x)) is determined from K(x,d(x)) = Ua/log[(6 We obtain u and v in + z0)/z0) (4.17) I- by u = U/dr, where ds is given by (4.11). v = Ue/dr (4.18) us is then easily obtained and hence t. Some solutions (graphically represented) are shown in Fig 4.4. Concluding Remarks Special Case Flows We have shown from these two cases that our model does indeed embody some of the major facets of dune migration. to these facets as We have 'minor' evolutionary processes. refered In contrast, 53 'major' evolutionary changes characterization to another. are (e.g. evolution from one 'major' smooth bed > ripples > dunes) (refer to Fig 2.4, page 23) In the simpler flow. above cases we have replaced the general flow with a Under these simplifications, our model is similar to most others in that no local flow variation is allowed. that when the local flow cannot vary, rate. unless Thus, It is clear neither can the local erosion although we get classical dune shape and migration, we attach a general flow to our erosion constraint our model will not have the potential to address 'major' evolutionary in the streambed. Dune Migration Potential Flow with Law of the Wall Fig 4.4a changes 54 430 ssam.ass,...naweszsmansnmsamstm:nYM .... . Dune Migration Potential Flow with Law of the Wall Fig 4.4b °Co Dune Migration Potential Flow with Law of the Wall Fig 4.4c .......... 55 A GENERAL FLOW The Flow Domain The general model for erosion was given in chapter III, (3.28). In considering the special case flows assumed uniform flow at infinity. chapter of IV, tacitly we A similar adjustment is necessary here also since we cannot deal numerically with a domain of infinite For length. we take the domain, the remainder of this discussion, D=D(t), to be the finite region D(t)=((x,y)I a < x < b, < y < n(x,t)). i(x,t) (5.0) that is, For simplicity, we assume uniform flow outside [a,b], outside the irespective inflow and outflow requiring uniform flow at the entrance respect limiting. to the inflow, this With respect to the constraint is quite limiting. not tend towards and artificial outflow, boundaries. are With boundaries. exit constraint is not very this however, additional We are excluding those flows which do a uniform flow near the exit boundary. flows (in particular, We those not tending towards a For some uniform flow) numerical 'reflections' off this artificial exit boundary may exist. We choose only flows where this reflection appears to be negligible. The erosion model on a finite interval, denoted by (*), is given by: pDu/Dt = div T + pg div u = 0 in D(t) in D(t) (5.1a) (5.1b) 56 p=0 Tn=0 ( *) (5.1c) on m(x,t) (5.1d) on n(x,t) nt u=0=v =v1 (5.1e) y=11 (5.1f) on i(x,t) (5.1g) opt = dus/dx us = dirEciw 0u/ax=0=av/ax c2(i' +)] (5.1h) at x=a and x=b (5.10 (5.1j) no, vo, Do given at to D /Ox + i'au where du /dx = (On $ s/4)1y=i(x,t). Solution Procedure Reducing to a System of Initial Value Problems unsteady flow. The system (*) depicts nonuniform, would variables: The only equation eliminate have to like u, V. p, exceptions for the a time p, the incompressibility. pressure have exclusively: we for each of the and only time dependent equations. and 1, are equation dependent Ideally pressure variable and (b), We term from (a). can use equation (b) the to The 'improved' system will time dependent equations, one for each remaining variable. Reducing (*) to a system of initial value problems is a of choosing finite elements as our solution method. result There are two main reasons for choosing finite elements over another method: (1) We have seen from almost steady, uniform flow that our erosion boundary condition becomes a nonlinear hyperbolic wave equation. 'Shocks' develop in such equations and the 57 natural approach for this phenomenon is to reformulate the problem in its weak setting. [23], [41] (2) The free surface condition is easily incorporated into our method and it simplifies the final equations. Having decided finite upon elements, apply a natural which simplifies certain aspects transformation, to a fixed domain, of our problem. we The resulting ordinary differential equation system requires some initial conditions. Numerical considerations force us to choose starting values consistent with the eventual flow. We give a brief outline of our method for solving Step (1): 'Remove' the pressure (*): term and the continuity the equation, (b), by applying the divergence operator to NavierStokes equation, (a). Step (2) Generate the weak form of this system. Step (3) Map the system from the free domain, to D, a fixed domain R. Step (4) Apply time dependent finite elements problem in the R domain. Step (5) Determine 'suitable' initial values. to the 58 result The of (1)(5) steps is a system of initial value problems set in the fixed domain, R. Step (1): Modifying (*) to 'Remove' p and equation (5.1b) Taking the divergence of (5.1 a), we obtain in D(t): div[p[ut + (u.V)u]) = div(Vp + pAu Holding p constant, pg). applying (5.1 b) and simplifying, (5.2) we have in D(t): p div[(uV)u) = Ap. (5.3) Thus, given u at time t, we have p implicitly given. Equations (a) and (b) in (*) are replaced by + gAu pint + (u101)10 = pg in D(t). where p is now the implicitly known function given by (5.4) the Poisson equation: Ap = 2p detVu p = 0 method (5.5b) on u(s,t) p=pg(ui)cose The (5.5a) in D(t) (5.5c) on i(x,t) ap/8x=0 at x=a (5.5d) 8p/ax=0 at x=b (5.5e) of solution for (5.5) is independent of the method for the other portions of our problem. Problems such as (5.5) are easily solved. [6], [29] Consequently, we proceed as if p is a known function, albeit implicitly. The system (*) has been altered to problem which we denot by (#). this new version of our 59 TU=0 (5.6a) in D(t) (5.6b) on u(x,t) nt =v1 u=0=v c pt us pg + pAu = PIUt + (5.6c) y=71 on 4(x,t) (5.6d) = dus/dx (5.6e) c2(4' = dir[c1w au/ax=0=8v/ax +c (5.6f) at x=a and x=b (5.6g) uo, vo, Do given at to where du /dx = (au lax +eauslay) Step (2): (5.6h) iy.i(x,t) and p is given by (5.5). The Weak Form of (#) We let <f,g>r1 be the inner product given by LTD fg da. We also use the equivalent notation <f,g> when no ambiguity will result. let W be our class of test on D. The exact The weak form of the problem restrictions on W will be given later. (*) defined functions We is given as: For t > to, find u and v such that for all w s W <ut,w> = <(nrusw> + <p <lew> = <v1 <dt,w> = <cp div T,w> + <g,w> ,w> (5.7a) (5.7b) y=11 dus/dx,w> (5.7c) subject to: TU=0 on n(x,t) (5.8a) u=0=v on i(x,t) (5.8b) us = dir[clw 8u18x=°=av/8x c2(4' +c)]+ (5.8c) at x=a and x=b (5.8d) 60 (5.8e) uo, vo, Do given at to and where p is given by (5.5). now We incorporate Recall the constraints (5.8) into (5.7). Green's theorem: (5.9) fD(div T)w da = ID Vw T da + faD (Tn)w ds We apply Green's theorem, (5.9), to (5.7a) and obtain <ut,w> = <(u*V)u,w> + <jolt> T,Vw> + <pTk, <p (5.10) w>OD hand where the last two products are defined by the right side of (5.9). The functions, test boundary. are w, usually chosen to vanish on the substitution is made to get zero boundary data. does Such a proceedure the standard In our case, eliminates products of the form procedure a simple For problems with nonzero boundary conditions, not quite work. The boundaries are free making the awkward. usual (constant for all time) substitution We achieve a similar end from a slightly different approach. We consider our boundary term <pTn,w>OD = <pTh,w>awo + <pTh,w)OD(4) + <pTn,w>8D(a) (5.11) <pTn,w>OD(b) where the parentheses indicate the portion of boundary to which we are restricting ourselves. The constraint (5.8a) assures Tn=0 on but u does. (5.5) in the same manner as (5.8), which admits p. Our test so that (5.12a) O. <pTn,w>OD(m) Th does not vanish on 71 Since we need not solve we need not choose a basis for w functions must only admit u and v. 61 Therefore, we can require 4. functions in W vanish on all that Thus, w=0 on 4(x,t) yields <pTn,w>(4) The uniformity of determine flow the and x=a at the actual flow there. (5.12b) o. (In fact, x=b allows to us we will choose uniform planar flow which we determine analytically.) We need not generate a finite element equation for these edges since they only solve (*) in the horizontal interior of D. need to vanish at x=a discrepancy and we x=b in essence By choosing w throwing There is no reflection when the flow becomes uniform as the outflow boundary x=b. We out the the boundary reflection mentioned above. by generated are known. are it reaches We choose w=0 at x=a and x=b and obtain (PTh'w>3D(a) = (5.12c) 0 = CpTit,w>8D(b). The result from (5.12a), (5.12b) and (5.12c) is (5.13) <pTn,w>arl = O. Our problem is now: Let W be given by W=fw(x,y)1 w e C1 (D), Given uo, vo, Do, w(x,4)=0, w(a,y)=0, w(b,y)=01. find u and v such that for t > to, (5.14) and for all w e W, <ut,w> = <(usV)u,w> + <g,w> <p <nt,w> = <viy=71,w> _1 <4t,w> = <c p du /dx,w> 1%ilw> (5.15a) (5.15b) (5.150 s where p is given implicitly by (5.5). It is this problem, (5.15), domain R. which is to be mapped to the fixed 62 Mapping D(t) to R Step (3): The problem (5.15) has two free boundaries making it awkward to Mapping D to a fixed domain will allow us to fix fix a basis for W. a basis for W. The drawbacks are, We can eliminate the need to invert additional nonlinearities. map but must we dealing with the inverse map and suffer consequences the our additional of nonlinearities. Let R be the rectangle given by R=f(I,;)i a <7 4< ; < n) < b, (5.16) and let A :D > R be the map given by (5.17a) = x The Jacobian of Ais 4). 4)/(T1 ; = (y given by JEAl = J = y, A (4(x,t) = i(a,t). We need calculation to obtain Ca,t) from R. the sediment Since a=x and i is independent that is i(x,t). _1 of (5.18) = h We are primarily interested in which never vanishes. boundary evolution, (5.17b) not perform a complicated If u is desired, however, then such a calculation must be performed. Our inner products must also be mapped to R. <f,g> > <f,g>R where <7';R = ffR f8 Y[A] da. We now suppress the R present version of our problem. and (5.19) the tilda notation and give the 63 _1 <ut3.,w> = f2<v y ,w> x 2<v ,w d> Y Y <vu ,w> + + <v d,w >)/Re + <uv ,w> Y Y Y <v ,w > x y u ,w > Y Y <h Y <ph,wI> (5.20a) <pdh,w > + <(sinO)h,w>/Fr Y <vtLvi> = [<u Y,w x> + <u + <v dh,w > 3' yy <v dh,w > + <v dh,w > y x x y ,w d> _i 2<h <v d2h,w ) Y Y I v ,w )/Re Y Y <vv ,w> + <p,w > Y Y <uhvx,w> + <uvY.dh,w> (5.20b) <(cos9)h,w>/Fr <T1t,w> = <v1 (5.20c) ,w> = <v(x01),w> Y=11 _1 <4t,w> = cp f<u d,w> + <i'h (5.20d) u '0)1y=i where h = h(x,t) = 1(x,t) d = d(x,y,t) = 4(x,t) (5.20e) + yh')/h (5.20f) and p is known implicitly. Step (4): Applying Time Dependent Finite Elements The general idea behind finite elements is the RitzGalerkin method. [7], [29] Our variables should be C (D). The finite version 'arbitrarily close' to the true of our variables should become variables as the number of basis is of elements increase. course desirable as is compact support. Finally, chosen must suffice for both the test functions and We choose a special finite basis for W. Orthogonality the the basis variables. This is the basis of 64 piecewise linear functions. The 'problem' is written in terms of the finite basis elements. The test functions now contain only a finite number of do the variables. that, as the coefficients will also be time In our case, The hope is dependent. unknowns as number the of basis elements the finite version of the variables converges to the true increase, Under certain conditions, weak solution. such convergence can be proven. [7] We proceed as follows: choose a 'suitable' finite basis for W write the entire problem in terms of the finite basis and evaluate the inner products form a system of ordinary differentia/ equations Step (4a): The Grid on R y=1 --0 y=0 flow > 1-7171A,,r; x=a i=1 i=2 -171Ai=I i=3 The Grid on R Fig 5.0 j=1so. x=b 65 We let ei denote the following basis for i=1,...I 1i + x/Ax, =1 ei x/Ax, 1+i x e [(i-1)Ax,iAxl x e EiAx,(i+1)Axl (5.21) otherwise 0 where Ax is (ba)/(I+1). We let eij denote the following basis for i=1,...I; j=1,...J 1+j-i 1+j 1+i 1+i-j 1-j + 1-i + eij(x,y) + x/Ax - y/Ay, y/Ay, x/Ax, - x/Ax + y/Ay, y/Ay, x/Ax, x x x x x x e region e region e region e region e region e region otherwise 1 2 3 4 5 6 (5.22) where Ay is 1/J. There are numerous properties of this basis. A typical such example being mu ij e e Such step. properties The = 0 if m > 1+1 or respectively n > j+1 will following be used without explicitly explaining each representation for our variables is used throughout where we employ the standard summation notation. u(x,y,0=a..(t) eij (5.23a) v(x,y,t)=P..(t) e 11(x,t)=Si(t) ei (5.23c) i(x,t)=6.(t) ei (5.23d) (71-4)=h=ti(t) ei (5.23e) w(x,y)=Xij e3 (5.23f) i=1,...I; j=1,...J All our functions are piecewise linear. tion may be easily carried out. Piecewise differentia- The following notation will be common during the remaining discussion. 66 au/ax = a.. aeij/ax = a,. e 13 13 (5.24) ij X (5.25) au/at = aa../at eij = aij eij 13 Greek letters as coefficients may be thought of We shall use the notation a = po, Po, vector. ali,...a/s). so, find a, p, 6, a such that for all X X<IP1>i where a The Discrete Version of Our Problem Step (4b): Given ao, (a11,a12 as z=(a,6,6,e). The full IP2> = (5.26) form of the inner products <IP1> and <IP2>, in their discreet version, are given in appendix 1. Step (4c): Evaluation of the inner products to form the system of ordinary differential equations Our evaluation is analogous to a much simpler case which we explain first. Given the following sort of 1dimensional inner product, must hold for all a1 We could simplify in this way: bi<ei,ei> <aiei,biej> = = which bi+1<ei.ei+1>1 aij.) The terms in the brackets, (5.27) may all be evaluated in terms of b and this example reduces to the form aEb = at.) no summation for a. where E represents the matrix formed from evaluating terms in (5.26). (5.28) 1 the bracketed 67 Since the a. (test functions) appear on both sides of the final equation, we end up with a linear system to solve. (5.29) El = [.) b If was dependent, time coefficients in (5.23) with time result would the be replace would we dependent the coefficients. constant The end system of ordinary differential equations in place of (5.29). El; (5.30) = C.) Our problem is analogous to this example except four that we have If we instead of one and we are set in 2d domain. variables momentarily denote z = (a,A,b,e) then we will ultimately obtain the form: (5.31) Al = C.) where A is the counterpart to E above. We skip all the intermediate steps and give only the final form of the problem: 0\ I-` a A, 6 sA A 0 0 A 0 0 A f2(a,0,6,e) 0 0 B 0 8 ( (0) f3 0 0 C,/ 0 if (a, 1 , (5.32) 0 \el \f4(a'ft'b,e)I or Al = f where A, B, C, and U. are given in appendix 1. x Our problem has now been reduced to solving the ordinary 68 differential equation system (5.32). need We only suitable add initial conditions to complete the problem. The Initial Condition for the Initial Value Problems Step (5): Presumably, system could we Although (5.32). attach most any fluid (numerically) eventually smooth out. to initial flow is conditions to our unstable it may In practice, it is impractical use initial conditions which are too incompatible with the flow. we take for our initial conditions the With this in mind, solution the steady flow generated when we momentarily to a related problem: fix the sediment boundary. We fix the resulting bottom flow steady with some initial shape and solve the This solution becomes our initial problem. condition at the time we turn the sediment loose. We refer to these initial conditions simply as no, po and Do. Solution Procedure Solving the Initial Value Problems Numerically Initial vlaue various methods. as (5.32) may be attacked with [7], [29], [35]. The method of choice should problems [5], such deal well with nonlinear equations. it is an We choose AdamsBashforth since implicit/explicit multistep method. more stable than the convenient, These methods are explicit, single step methods such as RungaKutta. We give the outline for solving (5.32) numerically. 69 Generate the initial conditions, Set up initial steps of multistep start, Use AdamsBashforth predictor/corrector, Find the new pressure distribution, Go to (3). We give the highlights of the procedure. Step (1): Generating the Initial Conditions Generating the initial conditions is a two step process. This allows first read in all the initial data which predetermined. problem which us to set the parameters for solving the steady flow is the second initialization step. We use and solve the fixed boundary problem by finite we reduce to domain predetermined a We That elements. is, Our output is the initial a linear equation system. flow, the initial domain being already known. Step (2): Set Up Initial Steps of Multistep Start The multistep method requires periods. These are generally explicit method. In our the obtained case, by two time solution for using single a step, we want to avoid the single step, The steady explicit methods. Our method benefits us flow is supposed to gradually go into an eroding flow. (fixed bed) here twice. Thus, by using the steady flow at both t=0 and t=ti, we satisfy the multistep requirements as well as making a smooth transition. 70 Step (3): Using AdamsBashforth Applying AdamBashforth is easy at this stage since we have the initial values at function evaluation. The tn and tn-1. The slopes are given by simple We simply run these values through our loop. only nonstandard calculation is the implicit calculation of the pressure after each pass through the loop. Step (4): Find the New Pressure Distribution The pressure distribution is necessary in evaluating the slopes. This is obtained by solving the Poisson equation for the pressure in terms of un,n, and in. We used finite elements here. Thus, we must solve the linear equation system. Results Graphical The results are presented graphically in Fig 5.1a and Fig 5.1b. 71 '30 Dune Migration A General Flow Fig 5.1a usarnesmumsuatalaLsw"...."1"...."661 i3o ewe" ISHMONI11111.11.1MINNIMMOSMINSW"*"' Dune Migration A General Flow Fig 5.1b 72 Concluding Remarks The General Solution We have developed a model for general flow. a procedure 'reasonable' solving for Given an initial dune, our similar to the special case flows' initial dune, we get a similar migration. An equilibrium However, flow also was used for the The difficulty there lies with the large variation in time scales between the evolution to flow. in that case, the flow either becomes unstable or does not undergo any appreciable erosion. flow initial rate. promote If rate and the we increase the flow and the erodability of the bed evolution' and even instability at an increased rate. up' the Certainly, rates, then we get the first improvement is a finer grid for faster flows. Our solution method for the general flow has some limitations which we note here. The flow rate must not be too excessive. The initial configuration must be fairly smooth. The outflow must tend towards uniform flow. 73 SUMMARY, OPEN QUESTIONS AND CONCLUDING REMARKS Summary Our erosion. streambed In chapter III we derived a model for basic approach was to combine the interaction of a viscous flow with boundary The which represented an erodable streambed. constraints viscous flow was characterized by the Navier-Stokes equation and the boundary condition was momentum, and energy. The erosion by obtained balancing sediment mass, The highlights of the model are: process consists of a flow subject to a boundary condition which represent an erodable streambed. The general erosion producing flow is governed by the Navier-Stokes equations for open channel flow. The boundary condition is obtained by balancing sediment mass, energy and momentum. The sediment transport occurs in a 'thin' layer the streambed. This is not the classical along boundary layer. Our model is for relatively fine, noncohesive, geneous, spherical sediment particles. homo- Large discrete particles would invalidate the mass balance derivation as well as the assumptions sediment transport velocity. used to obtain us, the 74 transport sediment The quasiequilibrium. must Increased be in sediment of state a across flux, results in an increased transport the boundary, rate rather than an increased sediment concentration. The vorticity of near flow the bottom the is a fundamental factor for the erosion process. The erosion model we obtained was a 2dimensional free boundary value problem with two free boundaries. flows well as a general flow. as The second case was: A examined special some The special flows were discussed The first case was: in chapter IV. We Almost flow. flow profile velocity logarithmic uniform steady, in conjunction with potential flow. Results from the first special shock nature of the model equations. case flow show Also, from this case we, found the flow to be too restrictive to allow for initial dune However, inherent the formation. classical dune migration pattern for an existing dune the is obtained. Results from the second special case flow are first case. The 'shock' account for to the forms as the angle of repose is attained and the solution begins to breakdown. cannot similar initial dune As in the previous formation case, we but the classic dune migration patterns are obtained. The main difference between case one and case two is that we are using an empirical flow in case two. The importance of this case is that it would allow us to upgrade our empirical constants to empirical relations. 75 In chapter V we considered a general migration our but unstable flow changes. method while some limitations which produces an has attempting attributed We this between flow rate and bed dune achieved We flow. achieve to to: evolution time scale variation the one, and rate evolutionary 'major' two, relative the of the grid for high flows or flows with an artificially coarseness high erodability. Some Remarks The Model and its Solutions One of this model aside from the distinctions major the of general flow field is the appearance near vorticity, the of the streambed, as a fundamental parameter affecting erosion. The largest particle a stream can move is called the midstream 1/6 Recall Brahms' formula, V =kW of the stream. c velocity to competence , which relates the s the weight of the particle. We show that a fundamental similar expression is obtained using to as a parameter. gw is Let fs be the force trying to roll a particle of diameter ds. proportional of the same order as To and To is known to be square of the midstream velocity, V. (V)o C Flow over evolution. a gw s the So, 1/3 W d f to s (4.17) s bed of sand causes the bed to undergo a distinct At slow flows, the bed remains flat. With increasing flow velocity, the bed passes through a set of stages: ripples, then dunes, then antidunes. flat again with chaotic flow, then waves and finally One of our main purposes was to have a model capable of 76 addressing these 'major' evolutionary changes. Erosion producing flows are generally turbulent, or at the very unsteady least and existence system, nonuniform. For uniqueness and turbulent existence flows, upon depend configuration and the Reynolds number. is, unsteady NavierStokes the the boundary Large Reynolds number, that and complex boundaries restrict the period of and uniqueness. It is just such flows which occur during the 'major' evolutionary changes. We have imposed restrictions on the class of flows with the goal of attaining a better mathematical understanding of the erosion process over precisely. the goal of attaining a model which matches data more This restricted class of flows still yields classic erosion patterns. The results of our model and its solutions are summarized here. Our special case flows have a flow field similar to many models. Our model can accept a general flow. We obtain flows dune migration and the general in both the special case flow. A fundamental parameter in our model is the vorticity near the streambed. Imurovements The Model The main improvements which we could make to the model are: 1. Replace empirical constants with empirical relations 7'7 in the derivation of the boundary condition Consider the case cs nonconstant (this could be done by assuming a simple exchange law) Consider more involved expressions for the resistence terms Take the pressure lift force into account Allow for nonconstant density in Improvements The Solution The main improvements to the solution procedure as it now stands are: Use a finer, variable mesh Use a more sophisticated method in the elimination of the pressure variable and the continuity equation Open Questions The open questions which arise from this problem are of two types: Those which deal with the model and its implications and those which deal with the solution procedures. Some of the more interesting open questions are given below. Can the model display the other 'minor' evolutionary processes, namely antidune migration? Can the model display 'major' evolutionary stages? Can the model be used to show that the evolutionary 78 stages are, in fact, bifurcation phenomenon? Can apply the classical boundary layer theory and we retain enough flow realism and still get effects? bifurcation the Note that in ELT one often assumes a boundary of constant curvature. What Is there another is the best solution approach? reasonable solution method? This problem unique is so that there is no set solution method. How does one deal with the outflow boundary condition when the flow does not become uniform? How can the difference in natural time scales for streambed versus the region flow be dealt the with more profitably? Concluding Remarks We have derived a new mode/ for streambed erosion. behavior. exhibits classical dune migration general flow field. accepts a The pressence of the general flow is necessary for certain aspects of nonuniform potentially of capable model The The model erosion. addressing Thus, erosion as this a model is bifurcation phenonenon. The fundamental parameter in this model, which distinguishes it from most others, is basic to is the vorticity near the streambed. vortex sheeting initiation of the ripples. which is The model can be possibly easily Such a term related to the 'improved' by 79 replacing empirical constants with relations. 80 BIBLIOGRAPHY Allen, J. R. L., Devolopments in Sedimentologv Sedimentary Structures, Their Character and Physical Basis, Elsevier NY (1984). Anderson, A. G., The Characteristics of Sediment Waves Formed by Flow in Open Channels, Proceedings 3rd Midwestern Conf. on Fluid Mechanics (1953). Anderson, A. G., Sedimentation in Handbook of Fluid Mechanics, 1st Ed., V. L. Streeter Editor, McGraw Hill (1961). ASCE, Sedimentation Engineering, ASCE Manuels and Reports on Engineering Practice, V. A. Vanoni Ed., (1977). Atkinson, K., Elementary Numerical Analysis, Wiley (1985). Badnold, R. A., An Approach to the Sediment Transport Problem From General Physics, USGS Professional Paper 4223 (1966). Baker, A. J., Finite Element Computational Fluid Mechanics, Hemisphere Pub. Corp. (1983). Bogardi, J., Sediment Transport in Alluvial Streams, Akademiai Kiado (1974). Carson, M. A., The Mechanics of Erosion, Pion Ltd. (1971). Chow, V. T., Openchannel Flow in Handbook of Fluid Mechanics, 1st Ed., V. L. Streeter Editor, McGraw Hill (1961). Chung, T. Y., Finite Element Analysis in Fluid Dynamics, McGraw Hill (1978). Daugherty, R. L., Basic Equations of Fluid Flow in Handbook of Fluid Mechanics, 1st Ed., V. L. Streeter Editor, McGraw Hill (1961). [13) DuBoys, P., Le Phone et Les Rivieres a Lit Affouillable, Annales des Ponts et Chaussees, Series 5, vol. 18 (1879). Einstein, H. A., The Bed Load Function for Sediment Transportation in Open Channels, Technical Bulletin 1026, USDA (1950). Goldstein, S. L., Modern Developments in Fluid Mechanics, (2 volumes) Dover (1965). Goncharov, V. N., Dynamics of Channel Flow, GIMIZ (1962). 81 Holt, M, Dimentional Analysis in Handbook of Fluid Mechanics, 1st Ed., V. L. Streeter Editor, McGraw Hill (1961). Hughes, T. J. R. and Marsden, J. E., A Short Course in Fluid Mechanics, Publish or Perish, Inc. (1976). loose, G. and Joseph, D. D., Elementary Stability and Bifurcation Theory, SpringerVerlag (1980). Kennedy, J. F., Mechanics of Dunes and Antidunes in Erodible bed Channels, J. of Fluid Mechanics, vol. 16, no. 4 (1963). Lamb, H., Hydrodynamics, 6th Ed, Cambridge (1932). Landau, L. D. and Lifshitz, E. M., Fluid Mechanics, Adison Wesley (1959). Lax, P., Hyperbolic Systems of Conservation Laws and the Mathematical Theory of Shock Waves, SIAM (1973). Meyer, R. E., Introduction to Fluid Mechanics, Wiley (1971). I251 MeyerPeter, E. and Muller, R., Formulas for Bedload Transport, (paper no. 2) Proceedings of 2nd Meeting Int'l Association for Hydraulic Research (1984). Pal, S. I., Laminar Flow in Handbook of Fluid Mechanics, 1 St Ed., V. L. Streeter Editor, McGraw Hill (1961). Parker, G., Discussion of the Theory of Minimum Rate of Energy Dissipation, J. of the Hydraulics Division, ASCE, vol. 106, no. HY7 (1980), (pg 1260-1261). Raudkivi, A. J., Loose Boundary Hydraulics, 2nd Ed., Pergamon Press (1976). Roache, P. J., Computational Fluid Dynamics, Hermosa Publishers (1972). (30] Sattinger, D., Lectures on Stability and Bifurcation Theory, Springer Lecture Notes #309 (1973). Schlichting, H. Boundary Layer Theory, McGraw Hill (1960). Schlichting, .H, Boundarylayer Theory in Handbook of Fluid Mechanics, 1st Ed., V. L. Streeter Editor, McGraw Hill (1961). Serrin, J., Mathematical Priciules of Classical Fluid Mechanics in Handbuch der Physik, vol. VIII/1, SpringerVerlag (1959). 82 Shen, H. W., Sedimentation Symposium to Honor H. A. Einstein, Ed. and Pub. by H. W. Shen, Colorado State University (1964). Stoer, J. and Bulirisch, R., Introduction to Numerical Analysis, SpringerVerlag (1980). Sumer, B. M., Lift Forces on Moving Particles Near Boundaries, J. of Hydraulic Engineering, vol 110, no. 9 (1984). Taylor, B. D., Temperature Effects in Low/High Transport Flat Bed Flows, J. of the Hydraulics Division, ASCE, vol 98, no. HY8 (1972) (pg 1427) and no. HY12 (1972) (pg 2191). Vanoni, A. V., A Summary of Sediment Transportation Mechanics, J. of Hydraulic Engineering, vol 88, no. 2 (1967). Vanoni, V. A., Fifty Years of Sedimentation, J. of Hydraulic Engineering, vol. 110, no. 8 (Aug 1984). von Karmen, T., Sand Ripples in the Desert, Technion Yearbook, vol. 6 (1947). Whitham, G. B., Linear and Nonlinear Waves, Wiley (1974). Yalin, M. S., Mechanics of Sediment Transport, 2nd Ed., Pergamon Press (1977). APPENDIX 33 APPENDIX The Inner Products A6 = q11 412 q13 q14 q21 q22 q23 q24 q211 "4- q15 416 q17 qpl q25 q26 q27 q28 429 qp2 C=441 B8=q31 q42 h./2 h./12 airAxAy 0/12 0/2 h./12 1 1 1 h./12 h./2 h./12 1 i= ,...I, j=1,...3 0/12 0/2 1 1 a11 A= a12 aij 1 a..41 aIJ 1 aIJ q210 84 2/3 1/6 1/6 2/3 1/6 1/6 2/3 1/6 B. ii 116 2/3 1/6 = Ax 2/3 1/6 1/6 2/3 1/6 1/6 2/3 1/6 2/3 1/6 2/3 1/6 1/6 C.. = Ax 1j 2/3 1/6 1/6 2/3 1/6 Axa.. [-p.1-1j-1 + 5 i-13 - 25 . . + 25.. - 5.141.3 + 0.1+1.j+1 ibs . q = 11 AxaiJ (-0i-1J-1 + 0i-1J i=1,...I; j=1,...J 1/6 2/3 25iJ-1 + 20iJ)/6 . 85 Ax3 + a fai-lj-1 '12 i-lj - 2a ij-1 i=1,...I; [-ai_11 + ai-17 - 2aiJ-1 + a. 1+1j+11/6 + 2a. }/6 j=1,...J Pii-i n13 aij+1 - ai+lj = Ax. ' + 2 5i+ii+1)/2 °i.j+1 = PiJ-1 1-13i-1J-1 i=1,...I; Ax AY °ill/2 j=1,...J + 2p g.. t-p 13 1j-1 - p ij 1 q14 = 1 Ax Ay giJ i=1,...I; 1 1 ) ij+1 j=1,...J gij = [ 5iJi [sin. - 2/ix ei-1 + jAy(ti.o. ti_i)] 86 -64 1-13ij-1 °ij+1)/ 213ij = '15 Ax (-3*7-1 Ay i=1,.../: °iJilti j=1,...J f-Pi-lj-1 °i-lj Pi+141.112 pij-1 (116 i1-1 i=1,...I; i.+11/12 j=1,...J isin(0) ti AxAy)/Fr [sin(8) ti AxAy)/2Fr i=1,...I; i=1,...I; j=1,...3 87 fli = {e2 ci + (j-1)Ay(t2 ei_i + (j-1)Ay{ti+1 fij = lei+1 fIj = CEI ti))/Ax (j-1)Ay{t1 EI-1 tI-1))/Ax 7ijlti + Yij+1ti aij = 1.YtT1-13-1ti-1 + 271-1j41-1 bij = dx{y.1-1j-1 f.1-1j-1 aiI = .f. /-13 1-13 4171-13-11-1 + Ti-13.1-1 biJ = Ax{y i-1J-1f. 1-1J c. ij y. = {y. 1-1j-1 ciJ = {y1-1J 1 y. 1-1j + 2y 17-1 + 2y. + 2y. f. ij-1 ij-1 71J-1ti f tS-1 iI-1 + 2y + 2y..1 + y 3. + 2y.-11.7f iI + 1+13 2y.. .f.. 13+1 13+1 Ti+1Ii+1)/6 + y. 2y.. 1+1j 13+1 ij-1 41+1341+1 1+1I 1+1J 1+13+1 )/6 a.. + b.. + c.. 13 q 13 13 = pl a +b.iJ + c . 1J i=1,...I; j=1,...7 Ayaiii + 2pi+li + Oi+ii+1)/6 + 1314-1 q21 Axaisti fi-1J-1 i=1,...I; j=1,...1 1-17 Pi1-1 131+111/6 38 g a.ij [-p. 1-1j-1 - + p. 25 ij-1 + 2p. ij+1 i+lj + p i+lj+1 }/6 c122 g a.1J f-P.1-1J-1 g"11 = 8'1+1 E.1 1- +i-17 - 20 i=1,...I; j=1,...J (4i+11-1)(j-1)Ay/2 exp.ij f-p.1-1j-1 + p.1-ij - + 201.1.1/6 - p + 20. + p I/6 i+lj+1 i+lj ij+1 2p q23 = Axpij [-pi_1j-1 + pi_lj - i=1,...I; j=1,...J {-a 1-1j-1 n = + 213u1/6 +a+a-a-"1-4.(2--a. 1-1j ij-1 1j ij+1 1+1j [-a1-1J-1 + a.11+ ai1-1 - aiJ1/2 i=1,...I; j=1,...J . 1+13+1 1/2 89 Ax Ay '25 g.. ) i j [-a.. ij-1 + 2a.. - a.. 1.1+1 1.1 = Ax Ay g.. = [ +a . g.13 (-a 0-1 ) "i+1 - 8i-1 1 2Ax jAY(i+1 1=1,...I; j=1,...J VI, p 2ij [-A i-1j + 20.. ij [ ReAx q26 2ReAx -Pi-1j 5i+1j) i+1 j j=1,...J gij -a27 Pij-1+2*5ij+Pijil. Pi+lj+Pi+lj+1) = g.j rP. 1j-1 41 g.. = [ ei+1 °J.j-1 'ij + A_o i+1.1 ei-1 + (j-1)AyQi+1 -i-1) ]/(4AxRe) i=1,...I; j=1,...J i-1)] 90 °i-li gij 1-5i-lj-1 Pi+li giJ 1ilj-1 gij = e.1+1 °i+ii+11 °ij-1 4 ei-1 + (j-1)dy(i+1 1-1 ) )/4AxRe j=1,...J q29= h.lj Ib.ij-1 h iJ tb ij-1ii + b..) ci_i gij = b.1j+1 I + 2*b.1j h..=(g..) /4AxAyRe4.1 (j-1)AYQi4.1 i=1,...I; j=1,...J Ay (-0.. 13 ij-1 + 20.. p..+1 1J 4210 = Ax Ay [-P iY-1 i=1,...I; j=1,...3 + p. )i. i/. 91 [cos(0) 4i AxAy)/Fr -'211 i=1,...I; j=1,...J 1 = [cos(9) all alj alJ q2 = all aij aiJ aIl aIj aIJ i=1,...I; j=1,...J AxAy)/2Fr 1=1,.. .1; j=1,...J 92 all = y11f11/2 + yijkl + y13f13/2 - Y21k2f'21 y31f31/2 als = -y1s..1f11/2 + y1J-1kl + y1Jf13/2 - y _12f' 2.) 2J - y3Jf3J/2 =£= -71j-1f1j-1/2 + yljkl + ylj+1f lj+1 =-Yi-11fi-11/2 Yilk2f'il /2 Yi+11f1+11/2 Y1+12k1 - yi+12fi+12/2 aij = 71+1j-1fi+lj-1/2 aiJ +y. 'kl = yi-1Jfi-1J/2 --1k2f* -y i+lj+ 1+ + yi+13-1f1+17-1/2 + i+1J-1kl - yi+1Jfi+1J/2 /2 all = -y aij f j-1 Ij-1 YI112fiI1 /2 +Yl -.k1 i aIJ = YI-1JfI-1J/2 j YIj-lfIj-1/2 Y, ij+1fIj+1/2 YIJfIJ/2 93 f'lj = 1.0 + fe2 (j-1)dy(0+1 = [1.0 + tei+1 (ei+1 el + (j-1)dy(0+1 0))(0+1 0) (i+1 2 0))/4(Ax) ) 0-1))(0+1 0-1))/4(Ax) 11/(4i) f'1j = 1.0 + [ei + ei]. + (j-1)dy(-0+0-1) (j-1)dy(0 20 + 0-1) 2ei + ei-1 + (j-1)dy(0+1 ei-1 + (j-1)dy(0+1 0-1))(0 ei + 0-1))/4(Ax) (ei I ei-1 + 2