Chapter 1

Vorticity and Potential Vorticity
In this chapter we use potential vorticity to unify a number of the concepts we have previously encountered. We derive the general form of potential vorticity conservation, and then appropriate approximate forms of the potential vorticity conservation law, corresponding to the familiar quasigeostrophic and planetary geostrophic equations. In order to keep the chapter essentially selfcontained, there is some repetition of material elsewhere.

1.1
1.1.1

Vorticity
Preliminaries
ω= ×v

Vorticity is defined to be the curl of velocity, and normally denoted by the symbol ω. Thus (1.1)

Circulation is defined to be the integral of velocity around a closed loop. That is C= v · dl. (1.2)

Using Stokes’ theorem circulation is also given by C= ω · dS (1.3)

The circulation thus may also be defined as the integral of vorticity over a material surface. The circulation is not a field like vorticity and velocity. Rather, we think of the circulation around a particular material line of finite length, and so its value generally depends on the path chosen. If δS ˆ is an infinitesimal surface element whose normal points in the direction of the unit vector n, then n· ×v = 1 δS v · dl (1.4)

1

2

CHAPTER 1. VORTICITY AND POTENTIAL VORTICITY

where the line integral is around the infinitesimal area. Thus the vorticity at a point is proportional to the circulation around the surrounding infinitesimal fluid element, divided by the elemental area. (Sometimes the curl is defined by an equation similar to (1.4).) A heuristic test for the presence of vorticity is then to imagine a small paddle-wheel in the flow: the paddle wheel acts as a ‘circulationmeter’ and so rotates if vorticity is non-zero.

1.1.2

Simple axisymmetric examples

Consider axi-symmetric motion, in two dimensions. That is, the flow is confined to a plane. We use cylindrical co-ordinates (r, φ, z) where z is the direction perpendicular to the plane. Then u z = ur = 0 uφ = 0 (1.5) (1.6)

(a) Rigid Body Motion
The velocity distribution is given by uφ = Explicitly, ω = ωz × v = ωz · k 1∂ 1∂ 2 (ruφ ) = (r ) = r ∂r r ∂r = 2 r (1.7)

(1.8)

That is, the vorticity of a fluid in solid body rotation is twice the angular velocity of the fluid.

(b) The ‘vr ’ vortex
This vortex is so-called because the tangential velocity distribution is such that the product v r is constant. That is: K uφ = (1.9) r where K is a constant determining the vortex strength. Evaluating the z–component of vorticity gives K 1∂ 1 ∂ (ruφ ) = r =0 (1.10) wz = r ∂r r ∂r r except when r = 0, when the expression is singular and the vorticity is infinite. Obviously the paddle wheel rotates when placed at the vortex center.

1.2. THE VORTICITY EQUATION
c d θ uϕ r
1

3

a

b

r

2

Figure 1.1: Evaluation of circulation in the axi-symmetric vr vortex. The circulation around path a − b − c − d − a is zero. This result does not depend on the size of the path; thus the circulation around any infinitesimal path not enclosing the origin is zero, and thus the vorticity is zero everywhere except at the origin.

We can obtain the result another way by calculating the circulation u· dl around an appropriate contour, the contour a-b-c-d-a in fig. 1.1. Over the segments a-b and c-d the velocity is orthogonal to the contour, and so the contribution is zero. Over b-c we have Cbc = and over d-a we have K θr2 = Kθ r2 (1.11)

K θr1 = −Kθ (1.12) r1 Thus the net circulation Cbc + Cda is zero. The result is independent of r1 and r2 and θ , so we may shrink the contour to an infinitesimal size so that it encloses an infinitesimal surface element. On this, by Stokes’ theorem, the vorticity is zero. Thus the vorticity is everywhere zero, except at the origin, and circulation around any closed contour not enclosing the origin is zero. Cda = −

1.2

The Vorticity Equation
1 ∂v −v×ω =− p+ν ∂t ρ

Recall that the momentum equation may be written
2

v

(1.13)

Taking its curl gives the vorticity equation ∂ω − ∂t × (v × ω) = 1 ( ρ × p) + ν ρ2
2

ω

(1.14)

whence we obtain ∂ω + (u · )ω = (ω · )v − ω ∂t ·v+ 1 ( ρ × p) ρ2 (1. for then 1 1 dp ( ρ × p) = 2 ρ × ρ =0 2 ρ ρ dρ The second term on the right hand side of (1. Dt ∂x ∂y ∂z (1.20) · v = 0 and. . (There is no commonly-used name for this quantity. (1. the vorticity equation is. (1.) For a homentropic fluid p = p(ρ). then Dω = (ω · )v. p × ρ vanishes and we have the simple and elegant form.21). but we shall call it the non-homentropic term. Dt If the fluid is also incompressible. VORTICITY AND POTENTIAL VORTICITY (For the rest of the chapter we will neglect viscosity.21) The terms on the right-hand-side of (1.’ In order that mass be conserved. Dt (1. or if the density is a function only of pressure (a ‘homentropic’ fluid) the right-hand-side vanishes.20) or (1. if the fluid is being ‘stretched.) If the density is constant. ∂u ∂u ∂u Dωx = ωx + ωy + ωz .15) As vorticity is the curl of velocity its divergence vanishes.19) Dt ρ ˜ where ω = ω/ρ.17) The last term on the right-hand-side is sometimes called the baroclinic term. directly from (1.22) The first term on the right-hand-side acts to intensify the x-component of vorticity if the velocity is increasing in the x-direction — that is. D ˜ ˜ ω = (ω · )v.17).21) are conventionally divided into ‘stretching’ and ‘twisting’.13) may be written × (u × ω) = (ω · )v − (v · )ω −ω ·v+v ·ω (1. Consider a single Cartesian component of (1.16) (1. The divergence term may be eliminated with the aid of the mass-conservation equation Dρ +ρ ·v =0 (1.4 CHAPTER 1.18) Dt to give ˜ Dω 1 ˜ = (ω · )v + 3 ( ρ × p) .

24) 1.26) We let η = ωz = ω · k Both the stretching and twisting terms vanish in two-dimensional turbulence. THE VORTICITY EQUATION 5 stretching in one direction corresponds to contracting in one or both of the other directions.28) ψ.1. Let this be the z-direction. each material parcel of fluid keeps its value of vorticity even as it is being advected around. the vorticity increases. w = 0. are zero. we use the incompressibility condition to define a streamfunction ψ such that u=− and ζ = 2 ∂ψ . or the vorticity measured in an inertial reference frame. Thus.27) where Dζ /Dt = ∂ζ /∂t +u· ζ .25) Only one component of vorticity non-zero. The second and third terms involve the other components of vorticity and are called twisting terms: vorticity in the x-direction is being generated from vorticity in the yand z-directions.29) . ∂y (1. Furthermore.1 Two-dimensional flow In two-dimensional flow the velocity field is confined to a plane. Dζ = 0. Since ωa = ωr + 2 and is a constant. because the conservation of angular momentum demands faster rotation as the moment of inertia falls. denoted by u. ∂x v= ∂ψ .2. (1. We return to an informative topological interpretation of vorticity evolution shortly. for incompressible homentropic flow. and the rate of change of any quantity normal to that plane. The velocity normal to the plane. specification of the vorticity completely determines the flow field. Dt (1. is v = u = ui + vj.23) (1. and the two-dimensional vorticity equation becomes. in two-dimensional flow vorticity is conserved following the fluid elements. Then the velocity. The results derived above apply to the absolute vorticity. To see this. this is given by ω=k ∂v ∂u − .2. the vorticity equation in a rotating frame of reference is Dωr 1 = [(ωr + 2 ) · ]v − (ω + 2 ) · v + 2 ( ρ × p) Dt ρ or D Dt ωr ρ = 1 (ωr + 2 ) · ρ v+ 1 ( ρ × p) ρ3 (1. (1. ∂x ∂y (1. and.

VORTICITY AND POTENTIAL VORTICITY Given the vorticity. We will give a straightforward proof.30). as we see in later sections and chapters.34) .32) The primary restrictions are that the flow be inviscid.32) is the velocity relative to the inertial frame.31) ∂t · k. or the circulation is conserved ‘following the flow. 2 ψ) = 0 (1. The vorticity is stepped forward one time-step using a finite-difference representation of (1. we note that in the presence of rotation the vorticity equation becomes ∂ζ + u · (ζ + f ) = 0 (1. That is. numerical integration of (1. the velocity in (1. Finally. both in the numerical solution and in diagnostic analysis.31) reduces to (1.26) is a process of time-stepping plus ‘inversion.’ That is D Dt v · dl = 0 .28). and background rotation plays where f = 2 no role. the last is the most important for geophysical fluids. and the vorticity ‘inverted’ using (1. the Poisson equation (1. beginning with the inviscid momentum equation: Dv 1 = p − φ. (1.32) through the integral gives D Dt v · dl = = Dv D dl · dl + v · Dt Dt 1 − p + φ · dl + v · dv ρ (1. that body forces be representable as potential forces. Similarly. Dt ρ (1. If f is a constant.33) where φ represents the body forces on the system.3 Kelvin’s Circulation Theorem Kelvin’s circulation theorem states that under certain circumstances the circulation around a material fluid parcel is conserved.29).30).29) and (1.29) is solved for the streamfunction.6 CHAPTER 1. The notion that complete or nearly complete information about the flow may be obtained by ‘inverting’ one field plays a very important role in geophysical fluid dynamics.28). the material derivative D/Dt is reference-frame indifferent. then (1. and the velocity fields obtained through (1.30) ∂t plus (1. 1. and that the flow be homentropic. Of these. The circulation is defined with respect to the inertial frame of reference. Applying the material derivative in (1.’ The vorticity equation may then be written entirely in terms of the streamfunction ∂ζ + J (ψ.

38) (ωr + 2 ) · dS = 0. entropy remains constant on that same material loop as it evolves and so circulation is preserved.35) Using Stokes’ theorem. the appropriate forms of the circulation theorem are D Dt and (vr + × r) · dl = 0.1 The ‘frozen-in’ property of vorticity At a slightly deeper level Kelvin’s circulation can be seen as a consequence of the topological character of the vorticity field. if Ds/Dt = 0. Such a line obviously connect fluid elements and therefore we can define a co-incident material lines. the flux of vorticity through a material surface is constant. We can draw a vortex line through the fluid. In non-homentropic flow. because they are integrals around a closed loop. (1. then if s = constant around the loop the density is a function only of pressure and the integral in (1. Further. The first term vanishes for a homentropic fluid.36) That is. and if Ds/Dt = 0. which is everywhere in the direction of the local velocity.1. In a rotating frame of reference. KELVIN’S CIRCULATION THEOREM 7 using the momentum equation and D(δl)/Dt = δv (see next section). .35) vanishes.37) D (1.3. it is conserved if the material path is in a surface of constant entropy. Dt where vr and ωr are the velocity and relative vorticity as measured in the rotating frame. In some ways this is the more natural form of Kelvin’s circulation theorem because it is really a consequence of the topological properties of vorticity. The second and third terms vanish separately. Since the equation of state is of the general form p = p(ρ. 1. the circulation is not generally conserved. and in particular that vortex lines are tied to material lines. (1. A material-line simply connects material fluid elements. This definition is analogous to that of a streamline. A vortexline is a line drawn through the fluid which is everywhere in the direction of the local vorticity. However. As the fluid moves the material lines deform. the circulation theorem may be written D Dt ω · dS = 0 (1. s. s). for such a fluid we may define a function G such that dG/dp ≡ 1/ρ and then 1 p · dl = ρ = = dp ρ dG dp dp dG = 0.3.

40) (1. and the vortex lines evolve in a manner determined by the equations of motion. t) (1. Dt δt which follows from the definition of the material derivative in the limit δt → 0.41) D δl = (δl · )v (1.42) Dt Comparing this with (1. at some initial time we can define an infinitesimal material line element parallel to the vorticity at that location.2 it is apparent that δl(t + δt) = l + δl + (v + δv)δt − (l + vδt) = δl + δvδt Substituting into (1. t) = Aω(x.39) gives But since δv = (δl · )v we have that D δl = δv Dt (1. See text for details. a vortex line always contains the same material elements — the vorticity is ‘frozen’ to the material fluid. δl being the infinitesimal material element connecting l with l + δl. we see that vorticity evolves in the same way as a line element. The rate of change of δl following the flow is given by D 1 δl = δl(t + δt) − δl(t) . From fig. v + δv . v δl (t + δt) l + δt v Figure 1.8 CHAPTER 1. .19)..39) (1. Consider how an infinitesimal material line element δl evolves. It follows from the diagram that D δl /Dt = δv. l + δl + δt (v + δv) δl l . The remarkable property of vorticity is that at later times the vortex lines remain co-incident with the same material lines that they were initially associated.e.43) . To see what this means. VORTICITY AND POTENTIAL VORTICITY l + δl .2: Evolution of an infinitesimal material line δl from time t to time t + δt. 1. Put another way. δl(x. i.

and is in that direction. for all subsequent times.2 Implications for circulation. i. because the volume of fluid is conserved. say ∂u/∂x > 0. and δV ∝ ωδS (1. We can now see why. t ). t ) = A−1 δl(x .46) The right hand side is just the circulation around the surface.e. The vorticity equation tells us that vorticity is enhanced if the velocity is in the direction of vorticity is increasing.3).3: Stretching of material lines leads to vorticity amplification. the plane of the surface is orthogonal to the vorticity. vorticity is amplified.44) 1.) Now consider a vortex tube (a collection of vortex lines) passing through a surface whose normal vector is parallel to the direction of vorticity (that is. Then. Let the volume of a small material box around the surface be δV . However. But the volume of the material.3. δV . must remain . the length of the material lines be δl and the surface area be δS. the material lines converge. Then δV = δl δS (1. If the corresponding material tube is stretched. the surface shrinks. KELVIN’S CIRCULATION THEOREM 9 Figure 1. That is ω ∝ δl. (1. and the integral of vorticity over a material surface (the circulation) remains constant. If.. then material lines in the x-direction are being stretched.1. ω(x .45) But the vorticity through the surface is proportional to the length of the material lines. (This mechanism is important in the dynamics of tornadoes. where A is a (dimensional) constant. 1. the magnitude of the vorticity of that fluid element (wherever that particular element may be in the fluid) remains proportional to its length. See fig. Since the vorticity is amplified in proportion to the length of the physical line element. then of course vorticity is amplified.3.

Thus. and the circulation remains constant.4 * Vortex Stretching and Viscosity: a Simple Example In this section we take a slight detour from our main theme and give a simple example of how vortex stretching and the associated amplification of vorticity can be balanced by dissipation (viscous effects) leading to a steady state. We will verify the circulation theorem for incompressible.52) (1.48) by the area of the infinitesimal material surface element δS δS Dζ = −δSζ Dt 2 ·u (1. VORTICITY AND POTENTIAL VORTICITY constant if the fluid is incompressible (recall that the constancy of density implies the constancy of a material volume). although vorticity has been amplified by stretching.51) · v. DδV /Dt = δV (1.50) gives Dζ DδS δS = −η Dt Dt or D (ζ δS) = 0 Dt verifying the circulation theorem. At some point choose the z-direction to be parallel to the vorticity. .47) Then the vorticity equation is Dζ ∂w =ζ Dt ∂z which.51) in (1. using the mass conservation equation. homentropic flow one other way. directly from the vorticity equation. may be written Dζ ∂u ∂v = −( + ) = −ζ Dt ∂x ∂y 2 (1.48) ·u (1.53) (This is analogous to the result for rate of change of a volume element. Thus. ζ ) (1.50) Now. the vortex lines are closer together and the product ωδS remains constant and circulation is conserved. 1. the rate of change of the surface area is given by D δS = δS Dt 2 ·u (1.) Using (1. 0.10 CHAPTER 1. This kind of balance occurs in tornadoes and other similar natural phenomena. ω = (0.49) where 2 · u = ∂u/∂x + ∂v/∂y is the two-dimensional divergence. including the common-or-garden bath plug vortex. Multiply (1.

1.4. The azimuthal velocity is unspecified. Vorticity The vorticity is only non-zero in the vertical direction: ωφ = ωr = ∂uz ∂ur − =0 ∂z ∂r 1 ∂uz ∂uφ − =0 r ∂φ ∂z 1 ∂ ∂ur (ruφ ) − = ζ (r) r ∂r ∂φ ωz = (1. ·v = 1 ∂vφ ∂vz 1 ∂ + + (ur r) r ∂φ ∂z r ∂r 1 ∂ = 0+α− (αr 2 ) = 0 2r ∂r (1. the second term the ‘swirl. Our goal is to find the form of uφ produced by a balance between vortex stretching and dissipation. The flow is converging toward the center. and we may expect that only through the introduction of viscosity may be achieve a steady state. the radial velocity depends only on the radial co-ordinate and the vertical velocity depends only on the vertical co-ordinate. .55) This holds no matter the form of the azimuthal field. z): 1 v = (− αr. * VORTEX STRETCHING AND VISCOSITY: A SIMPLE EXAMPLE 11 1. and to satisfy mass continuity it is accelerating upward at a constant rate.4. therefore only the radial dependence of the azimuthal velocity produces a vorticity. provided it is a function only of r. The vortex stretching term will produce vortex amplification.’ and the third term the upward motion which will produce vortex stretching.56) In words. or core.1 Posing the Problem We suppose that there is an axi-symmetric swirling flow. Mass Conservation This velocity field is consistent in that it satisfies the constant density form of the mass conservation equation. uφ . φ. this in the vertical direction. If the given rate of convergence is α. then the velocity field is in cylindrical co-ordinates (r. except that we shall suppose that it is only a function of the radial co-ordinate. around some center. αz) (1.54) 2 The first term represents the radial converge. Explicitly.

vorticity distribution in which amplification due to vortex stretching is balanced by viscous dissipation. and supposing a convergence rate of α = 1 cm s−1 per meter. the flow becomes irrotational away from a delta function at the origin.54) this simplifies to 1 ∂ζ 1 ∂ − αr = ζα + ν 2 ∂r r ∂r r ∂ζ ∂r (1. as determined by an e-folding scale.57) ∂ζ uφ ∂ζ ∂uz ∂ζ ∂ζ + ur + + uz =ζ +ν ∂t ∂r r ∂φ ∂z ∂z 2 ζ (1. the constant C must be zero. is given by ro = 2 ν α 1/2 (1.63) As viscosity tends to zero. whence ζ = ζ0 exp − αr 2 4ν (1.60) 1 ∂ζ +C − αζ r 2 = νr 2 ∂r (1. Might this model apply to a tornado? Using a molecular viscosity.61) To avoid a singularity at the origin. or as the intensity of the convergence increases.2 The solution The vertical component of the vorticity equation contains only the stretching term. which is rather small.4. VORTICITY AND POTENTIAL VORTICITY 1.62) This is the steady state. ν ∼ 10−5 m2 s−1 . and is. Aside from uncertainties in the convergence rate. we find 1 ∂ ∂ − α (ζ r 2 ) = ν 2 ∂r ∂r which integrates once to r ∂ζ ∂r (1. Dζ ∂uz =ζ +ν Dt ∂z or 2 ζ (1.59) Combining the vortex stretching and advection terms. . In fact. radially symmetric. the main error here is supposing that vortex stretching is balanced by molecular diffusion of momentum. there are likely to be small scale turbulent motions which in some ill-defined way greatly enhance the effective viscosity and thicken the rotational core.58) In the steady state and using (1. then we find ro ∼ 3 cm.12 CHAPTER 1. Vorticity falls very quickly away from a rotational core whose thickness ro .

64) turn out to be the final steady solutions of the initial value problem for almost any initial vorticity distributions. The main restrictions are that ζ → 0 faster than 1/r 2 as r → ∞. which is indeed irrotational.56) for vorticity.5. the process of convergence and the associated vortex stretching produce a very tightly bound vorticity distribution. Since the first term decays much faster than the second.1 Homentropic flows. Then using the scaler evolution equation in conjunction with the vorticity equation gives us a scalar conservation equation. Interestingly. First. Requiring uφ = 0 at the origin gives A = 2νζ0 /α. furthermore. POTENTIAL VORTICITY CONSERVATION 13 Velocity Field Integrating the expression (1. or to take care of the evolution of fluid elements. (Of course non-conservative forces and viscosity also lead to circulation non-conservation. (i) From Kelvin’s theorem for an infinitesimal volume we write Kelvin’s theorem as: D [(ω · n)δA] = 0 Dt (1. 1. for r >> ro (or for ζ0 = 0) the swirling velocity field goes as 1/r. such as is found in the atmosphere and ocean.1. the distributions (1. like tornadoes. we find the radial velocity field to be uφ (r) = − 1 2ν αr 2 A ζ0 exp − + r α 4ν r (1. such as vorticity itself.64) where A is a constant of integration. Thus. It is important not only in obvious cases of vortex intensification.5 Potential Vorticity Conservation Although Kelvin’s circulation theorem is a general statement about vorticity conservation.65) . Second. For non-homentropic flow this scalar field must be chosen in a special way (it must be a function of the entropy alone). This is potential vorticity. it is not satisfied for non-homentropic flow. it is not always very useful statement for two reasons. The basic idea is that we can use a scalar field that is being advected by the flow to keep track of.) It turns out that it is possible to derive a beautiful conservation law that overcomes both of these failings and one. but the restriction to homentropic flow can then be avoided.5. it is a not a statement about a field. but this applies to virtually all conservation laws and does not diminish them. The process of vortex stretching is very common.62) and (1. no matter how diffuse the initial vorticity. but is believed to be responsible for the cascade of energy to small scales in threedimensional turbulent motion. that is extraordinarily useful in geophysical fluid dynamics. 1. and that the initial circulation is finite and non-zero (its value determining ζ0 ).

1. where n is a unit vector normal to an infinitesimal surface δA (see fig.69) ˜ where ω = ω/ρ.67) (1.66).4).4: An infinitesimal fluid element. (ii) From the frozen-in property From (1. (1.14 CHAPTER 1. that is D Dδχ (χ1 − χ2 ) = =0 (1.66) Since χ is conserved on material elements. Dt Since n = χ /| χ | and the infinitesimal volume δV = δhδA we have that ω · nδA = ω · χ δV | χ| δh (1. where χ is a conserved tracer satisfying Dχ = 0. VORTICITY AND POTENTIAL VORTICITY n δh Χ δΑ Χ+δΧ Figure 1. bounded by two isosurfaces of the conserved tracer χ .65) becomes ρδV D ω · χ =0 (1.68) δχ Dt ρ or D ˜ (ω · χ) = 0 Dt (1. the difference between χ between two infinitesimally close fluid elements is conserved.70) Dt Dt . and the mass ρδV of the volume element is also conserved. The volume is bounded by two isosurfaces of values χ and χ + δχ .

Dt where (1. For atmospheric and oceanic applications. POTENTIAL VORTICITY CONSERVATION But δχ = χ · δl where δl is the infinitesimal vector connecting the two fluid elements. Dt Following Salmon (1998) suppose there are three independent scalars χi . In homentropic flow the very generality and ubiquity of the conservation law seems to make it less useful.74) Dθ =0 (1. the conserved potential vorticity is Q = (ω + 2 )/ρ where ω is the relative vorticity and the rate of rotation of the coordinate system. The vorticity in the potential vorticity conservation law is the absolute vorticity.74) is the most useful and form of potential vorticity conservation. Dt ρ For nonhomentropic flow the potential vorticity equation becomes.72) ˜ where Q = (ω · χ ) is the potential vorticity.3 Potential vorticity in isentropic coordinates Dχi = 0.5.5. if the equation of state can be written in the form p = p(s.76) . because the required use of a special conserved scalar imparts more information to the potential vorticity conservation law. 2.71) to obtain again DQ =0 Dt (1.1.73) However. 1. and the equation has profound consequences. It turns out that potential vorticity is a much more useful quantity for nonhomentropic flow than for homentropic flow.71) But since the line element and the vorticity (divided by density) obey the same equation. (1. ρ) where s is the entropy.75) Dt and θ = θ (s). 1. and if we demand that the conserved scalar be a function of entropy s alone then we recover D ˜ (ω · θ) = 0 .2 Nonhomentropic flow 1 DQ = 3 χ · ( ρ × p). In a rotating frame of reference.5. i = 1. (1. we can replace the line element by vorticity (divided by density) in (1. Thus D ( χ · δl) = 0 Dt 15 (1. 3 that each satisfy (1.

Then. 1.e. We consider a few such cases here. and D (1. the direction parallel to s) is still conserved.82) ( θ × A · s) = 0.77) may be equivalently written Q = (Q1 . take the entropy s as one of the coordinates. ..80) The conservation law may then be written D ( Dt θ × A) = 0 (1.78) ×A (1. atmospheric and oceanic models are cast in a ‘semi-Lagrangian’ form. the potential vorticity in direction of increasing s (i.77) (1. if we use the χi to define a coordinate system. Q2 . Dt Often. In this case the potential vorticity conservation takes a useful and simple form. the conservation law is lost. . and we shall discuss this case more later on. the conserved potential vorticity is the curl of the velocity in θ coordinates. However. and let v = A1 χ1 + A2 χ2 + A3 χ3 then (see problem 2) the potential vorticity field (1. there are three conserved potential vorticities.81) That is. VORTICITY AND POTENTIAL VORTICITY Then. . the full Navier-Stokes equations are far too complex to be of direct use for ocean-atmosphere modelling. ∂θ1 ∂θ2 ∂θ3 (1. in which the vertical coordinate is a function of entropy (typically density in the ocean or potential temperature in the atmosphere). However. For a nonhomentropic fluid. appropriate approximate forms of potential vorticity are conserved by the respective approximate models — Boussinesq. Now.79) = ∂ ∂ ∂ . ˜ Qi = (ω) · χi . Q3 ) = where θ θ (1.16 CHAPTER 1.6 Potential Vorticity Conservation in Approximate Models As we saw in previous chapters. and leave others as problems for the reader. hydrostatic and so on. for a homentropic fluid.

2 The hydrostatic Boussinesq equations In the hydrostatic approximation the vertical momentum equation is ∂φ =b ∂z (1.90) .6.85) in Cartesian coordinates we obtain. that · v = 0.. This can be transformed into a more revealing form by writing it as Qh = ρ z vx − vz ρx ρz − uy − u z ρx ρz . or equivalently Q=ω· T (1. POTENTIAL VORTICITY CONSERVATION IN APPROXIMATE MODELS 17 1. density need not play a role in the definition of a conserved potential vorticity.e. From these two considerations. conserved if the flow is inviscid and adiabatic) when advected by the three-dimensional velocity. in Cartesian coordinates.87) (1. Q = (vx − uy )bz + (wy − vz )ρx + (uz − wx )ρy (1.6.85) 1. since the equation for vorticity itself is isomorphic to that for a line element if volume is conserved. However. i. for the equation of state is typically of the form ρ = ρ0 (1 − α(T − T0 )). (1. inviscid.1 The Boussinesq equations Recall that the Boussinesq equations are an incompressible set (by our definition.84) (in the absence of heat sources).86) and that this is a Lagrangian conserved quantity for unforced.e.89) Qh = (vx − uy )ρz − vz ρx + uz ρy .83) The thermodynamic equation is often written as an equation for the buoyancy b = −gδρ/ρ0 . Dρ/Dt = 0 (1. it is plain that the appropriate form for potential vorticity is Q = ω · b.88) where φ = p/ρ0 and b = −gδρ/ρ0 is the buoyancy. the Boussinesq equations are not homentropic. adiabatic flow.. (1. Expanding (1.1. It can be shown that (problem 3) that the appropriate potential vorticity is. and that this is quasi-conserved (i.) Thus. (1.6.

ρ or b. although the extension to the compressible case is straightforward. (iii) Variations in the Coriolis parameter are small. written informally. the three assumptions we will have to make are. potential vorticity is simply the horizontal vorticity evaluated on a surface of constant density.) Thus. Other geostrophic models (e. and make the corresponding approximations. VORTICITY AND POTENTIAL VORTICITY But the terms in the inner brackets are just the horizontal velocity derivatives at constant ρ.g. (ii) Basic state stratification that does not vary in the horizontal.. (i) Small Rossby number. Derivatives with respect to z are implicitly taken at constant x and y.92) Thus.18 CHAPTER 1. since x and y are Eulerian coordinates whereas the ‘vertical’ coordinate. (These relationships follow from standard rules of partial differentiation. frontal geostrophy) may be derived with other assumptions replacing (ii) and (iii). planetary geostrophy.91) with a similar expression for ∂u/∂y|ρ . multiplied by the vertical derivative of density. and also assume hydrostasy. note that ∂v ∂x = = ∂v ∂x ∂v ∂x + ∂v ∂z ∂z ∂x ρ z ρ ∂v ∂ρ − ∂z ∂x z z / ∂ρ .1 We will do this in the Boussinesq approximation. 1. is Lagrangian. This is a ‘semi-lagrangian’ version of potential vorticity. it is revealing (and easier) to begin with potential vorticity. . we obtain Qh = = ∂ρ ∂z ∂ρ ζρ ∂z ∂v ∂x − ∂u ∂y ρ ρ (1.7 Quasi-geostrophy. Scaling theory guides the approximations. ∂z (1. The second and third assumptions are peculiar to quasi-geostrophy. The basic idea is that we begin with the potential vorticity equation and make approximations both to the potential vorticity and to the advecting velocity. Note that we could replace density by buoyancy b in the above derivation. As before. To see this. revisited Rather than deriving the quasi-geostrophic approximation from the momentum equations.

q = (βy + ζ )N 2 + f0 bz .96) Using geostrophic and hydrostatic balance (∂φ/∂z = b) gives thermal wind balance f0 × with associated scaling ∂u =− b ∂z (1. the dynamically important part of potential vorticity is given by. because the basic state is unvarying in the horizontal. Since in chapter ?? we proceeded asymptotically. and that. Without approximation. in Cartesian coordinates (purely for simplicity) in a rotating frame of reference the potential vorticity is ˆ Q = (f0 + βy + ζ )(bz + bz ) − (vz bx − uz by ) = ˆ ˆ f0 bz + (βy + ζ )bz + f0 bz + (βy + ζ )bz − (vz bx − uz by ) .97) f UL (1. here we shall be more informal.98) H Note that this only applies to the perturbation buoyancy. (1.93) Then.7. y. At low Rossby number geostrophic balance (f × u ≈ − φ) gives the for pressure scaling φ ∼ f UL (1. (1. z. we write the stratification as: ˆ b = b(z) + b (x.94) ˆ We will subsequently drop the primes on the perturbation b. (1.95) and indeed this is correct. although some find such formalities obfuscating. More formally we scale the equations and perform an asymptotic expansion in powers of the Rossby number.101) (1. ∂z (1. we anticipate that the terms on the second line are in decreasing order of size. Geostrophic and thermal wind balance imply if we define a streamfunction ψ by b∼ ψ =φ f0 then ug = k × ψ and b=f ∂ψ . REVISITED 19 and these may be formalized by the use of nondimensionalization and asymptotics. and write N 2 (z) = −bz . t).1. From the assumptions above.100) (1. since the first term is constant in space and time. QUASI-GEOSTROPHY.99) .

102).104) where the terms in curly brackets indicate the scales of the terms multiplying them relative to the first term. However. (1. Assuming that βy ∼ ζ /f ∼ Ro then. (1. in the low Rossby number limit. then the term f0 bz will dominate the evolution of potential vorticity.105) 2 We see that if F 2 /Ro is not O(1).] To derive the quasi-geostrophic approximation. as indeed it is for large-scale flows in the ocean and atmosphere.98). . it is the effects of stratification. I prefer to think of this expression in terms of the Froude number.20 CHAPTER 1. This approximation relates to the scale of the motion.102) where F = U/N H is the Froude number. it cannot −1 be larger than the deformation scale by O(Ro . condition on stratification is equivalent to demanding that the horizontal scale of the motion must be the deformation scale or smaller. The condition F 2 /Ro ∼ Ro is equivalent to Ro L2 λ2 = O(Ro ). it is a demand that the stratification be strong. Thus. (1. q = (βy + ζ )N 2 + F2 f0 bz 2 Ro (1. We return to this point shortly. This can be formalized by nondimensionalizing all of the variables using (1.106) where λ = N H /f is the radius of deformation. save for the nondimensional factors in curly brackets. the potential vorticity is: Q = f0 N 2 + {Ro } (βy + ζ )N 2 + F2 F2 f0 bz + {Ro } 2 2 Ro Ro (βy + ζ )bz − (vz bx − uz by ) . F 2 /Ro is assumed small. that is if F 2 /Ro is not small. [This expression is sometimes written |bz | 1 ∼ 2 N Ro Ri (1.104) is a constant.103) where Ri = N 2 /(∂u/∂z)2 is the Richardson number. although to demand that it be small may seem a little unmotivated for it has little to do with geostrophic balance. Rather. rather. the lowest order approximation to Q that will contribute to its dynamical evolution is.96).104) are then all O(1) nondimensional quantities. (1. because it not the shear of the flow that is central here. The terms in (1. VORTICITY AND POTENTIAL VORTICITY The ratio of the size of the perturbation buoyancy to the basic state buoyancy is then given by bz f UL F2 |bz | = 2 ∼ 2 2 = ˆ N H N Ro bz (1. In this limit (which arises in the planetary geostrophic approximation) all dynamics involving relative vorticity are absent. Since the first term in (1.

107) Dt ∂z where q = f0 N 2 and the advective derivative is (at this stage) fully three-dimensional. The potential vorticity equation is thus Dq ∂q ˆ +w =0 (1. ∂qg + ug · q g = 0 ∂t where qg = (βy + ζ ) + ∂ ∂z f0 b N2 (1. REVISITED 21 Similarly. ˆ ∂q ∂q ˆ + ug · q + w =0 ∂t ∂z f0 bz .111) (1.112) Eliminating w between (1.112) (left as an exercise for the reader: problem 4) gives the quasi-geostrophic potential vorticity equation. q = (βy + ζ ) + ˆ ∂b ∂b + ug · b + w =0 ∂t ∂z where (1. if |βy| ∼ f . then the dominant dynamically active term is βyN 2 and the leading order term in the evolution equation would become βv = 0.114) .110) (1. N2 is the approximation to (perturbation) potential vorticity in the quasi-geostrophic limit. it is not a very dynamically rich equation. Now. Let us now consider how the advecting velocity should be approximated. Although this graphically illustrates the constraining effect that the β–effect has on meridional motion. since ˆ f is nearly constant.108) and the vertical advection in (1. This equation becomes. QUASI-GEOSTROPHY.110) and (1. That is ∂u ∂v U + ∼ Ro ∂x ∂y L Thus the proper scaling for the vertical velocity is.7. We can simplify this further with the help of the thermodynamic equation.109) U L (1.1. the horizontal velocity is nearly non-divergent. L (1. after divinding by N 2 .107) is important only in advecting the basic state potential vorticity q. w ∼ Ro UH .113) (1.

in terms of streamfunction. S. (1. CHAPTER 1.22 or. in which the particle relabelling symmetry that gives rise to potential vorticity conservation is discussed. v. K. Oxford University Press. if a little more formal. Salmon.114) (or (1.116) 5. 2. and derive quasi-geostrophic theory in isentropic coordinates.114) or (1. G. b) = (−ψy . From the streamfunction. 1967. 1998. R. ψ + 4.115) is not an approximation to the Ertel potential vorticity. .115) There are a number of important points to be made: 1. Chapter 4 contains a brief discussion of potential vorticity. ψx . VORTICITY AND POTENTIAL VORTICITY qg = (βy + 2 ψ) + ∂ ∂z f02 ∂ψ N 2 ∂z (1. That approximation is given by (1. An Introduction to Fluid Dynamics. The so-called quasi-geostrophic potential vorticity given by (1. Nevertheless. (1. that here. N 2 ∂z (1. This contains an extensive discussion of vorticity and vortices. Vallis (1996) gives a derivation similar to. Further Reading Batchelor. Cambridge University Press. and chapter 7 a longer discussion of Hamiltonian fluid dynamics. The streamfunction may be diagnosed from the potential vorticity by the solution of the threedimensional elliptic equation: 2 + f02 ∂ ψ = qg − βy. the velocity and buoyancy are easily diagnosed as elements of the vector: (u. Charney and Stern (1961) were the first to clarify the relationship between quasi-geostrophic potential vorticity and Ertel potential vorticity. Geophysical Fluid Dynamics. f ψz ).117) Notes 1.115)) is more useful than (1.111) because it is conserved when advected by the horizontal geostrophic velocity and because 3.111). qg may be written as the three-dimensional divergence of a vector J where J = k (f02 /N 2 )∂ψ/∂z.

or some other way.79) is the same as that defined by (1.1.77). a square) with segments neither parallel nor orthogonal to the radius.g.112)) to obtain the so-called quasi-geostrophic potential vorticity equation (1. Show explicitly that the circulation around it is zero. Show.110) and (1.77). conserved if the flow is adiabatic and inviscid) when advected by the three-dimensional flow. REVISITED 23 Problems 1.113). Derive the quasi-geostrophic potential vorticity equation appropriate in the atmospheric case — i. 3. Show that potential vorticity defined by (1. . allow density to vary by O(1) in the vertical. 4. For the vr vortex. (Some may think this problem is perverse. that in the hydrostatic Boussinesq approximation the quantity given by (1. or the ‘quasi-Boussinesq’ approximation in height coordinates.. QUASI-GEOSTROPHY. You may do this either using pressure coordinates.e. 5. and not enclosing the origin.89) is quasi-conserved (i..e.7. beginning with the momentum equations. Eliminate the vertical velocity between the thermodynamic and potential vorticity equations ((1. choose a contour of arbitrary shape (e. Hint: begin with (1.) 2.

VORTICITY AND POTENTIAL VORTICITY .24 CHAPTER 1.

1 The Closure Problem Although it is impossible to predict the detailed motion of each eddy. What this means precisely. the mean state may not be changing. consider the weather system. These have their origins in classical papers by Kolmogorov and Oboukhov in 1941. For example.1 The individual eddies come and go. we know that next summer will 25 . we certainly have some skill at predicting their mean state—the climate.1. we ask ‘what is the problem?’ Then we will consider the classical scaling theories of turbulence in both two and three dimensions. Although we cannot predict these very well. in which the storms. For example. and how the motion of the atmosphere and ocean connects to the standard and some more modern theories of turbulence is the subject of the next two chapters. The scales of much motion of interest are certainly significantly larger than the dissipation scale (the scale at which molecular viscosity becomes important) by several orders of magnitude and. anti-cyclones. Loosely. First. to a large degree. simply motion of a forced-dissipative fluid subject to various constraints such as rotation and stratification. The circulation of the atmosphere and ocean is of course. and. at may if not all scales. hurricanes. and are intrinsically unpredictable. turbulence is high Reynolds number fluid flow. dominated by nonlinearity.Chapter 2 Basic Theory of Incompressible Turbulence A turbulent flow is one that is dominated by eddies with a spectrum of sizes between some upper and lower bounds. We won’t consider many of the variations and subtleties associated with the classical theory (such as the effects of intermittency) except in so far as they may affect geophysical flows. 2.—constitute the eddies. for the two-dimensional case. and before considering turbulence in a geophysical context. in section 2. Following this we will discuss geophysical issues. with both spatial and temporal disorder. the motion is highly nonlinear. This seems to accord with our definition of turbulence. fronts etc. Kraichnan in 1967.

26

CHAPTER 2. TURBULENCE, BASIC THEORY

be warmer than next winter, and that in California summer will be drier than winter. We know that next year it will be colder in Canada than in the Mexico, although there might be an occasional day when this is not so. We would obviously like to be able to predict the mean climate without necessarily trying to predict or even simulate all the eddies. We might like to know what the climate will be like one hundred years from now, without trying to know what the weather will be like on February 9 2056, plainly an impossible task. Even though we know what equations determine the system, this task proves to be very difficult because the equations are nonlinear, and we come up against the ‘closure’ problem. Let us try to write an equation for the mean flow. The program would be to first decompose the velocity field into mean and fluctuating components, v =v+v (2.1)

Here v is the mean velocity field, and v is the deviation from that mean. The mean may be a time average, in which case v is a function only of space and not time. It might be a time mean over a finite period (e.g a season if we are dealing with the weather). Most generally it is an ensemble mean. Note that the average of the deviation is, by definition, zero; that is v = 0. We then substitute into the momentum equation and try to obtain a closed equation for v. To visualize the problem most simply, we carry out this program for a model nonlinear system which obeys du + uu + νu = 0 (2.2) dt The average of this equation is: du + uu + νu = 0 (2.3) dt The value of the term uu is not deducible simply by knowing u, since it involves correlations between eddy quantities u u . That is, uu = uu + u u = uu. We can go to next order to try (vainly!) to obtain a value. First multiply (2.2) by u to obtain an equation for u2 , and then average it to yield: 1 du2 + uuu + νu2 = 0 (2.4) 2 dt This equation contains the undetermined cubic term uuu. An equation determining this would contain a quartic term, and so on in an unclosed hierarchy. Most current methods of ‘closing the hierarchy’ make assumptions about the relationship of (n+1)’th order terms to n’th order terms, for example by supposing that: uuuu = uu uu − αuuu (2.5)

where α is some parameter. Such assumptions require additional, and sometimes dubious, reasoning. Nobody has been able to close the system without introducing physical assumptions not directly deducible from the equations of motion.

2.2. THE KOLMOGOROV THEORY

27

In the constant density (say ρ = 1) Navier-Stokes equations, the x-momentum equation for an averaged flow is ∂u ∂p + (v · )u = − − ·vu. (2.6) ∂t ∂x Written out in full in Cartesian co-ordinates, the last term is ·vu = ∂ ∂ ∂ uu + uv + uw ∂x ∂y ∂z (2.7)

These terms, and the similar ones in the y- and z- momentum equations, represent the effects of eddies on the mean flow. They are known as Reynolds stress terms. The ‘problem’ of turbulence might be considered to be to find a representation of such Reynolds stress terms in terms of mean flow quantities. However, it is not at all clear that any reasonable general solution (or parameterization) even exists, short of computing the terms explicitly.

2.2
2.2.1

The Kolmogorov Theory
The physical picture

Consider high Reynolds number incompressible flow which is being maintained by some external force. Then the evolution of the system is governed by ∂v 1 + (u · )v = − p+F +ν ∂t ρ and ·v =0 (2.9) Here, F is some force we apply (i.e., we are stirring the fluid). A naïve scale analysis of these equations indicates that the relative sizes of the inertial terms on the left-hand-side to the viscous term is the Reynolds number U L/ν. To be explicit let us consider the ocean, and take U = 0.1 m/s, L = 1000 km and ν = 10−6 m2 s−1 . Then Re ∼ 1011 , and apparently we can neglect the viscous term on the right hand side of (2.8). But this can lead to a paradox. The fluid is being forced, and this forcing is likely to put energy into the fluid. We obtain the energy budget for (2.8) by multiplying by v and integrating over a domain. If there is no flow into or out of our domain, the inertial terms in the momentum equation conserve energy and we obtain, d dt or ρv 2 dV = ˆ dE = dt F · v + νv ·
2 2

v

(2.8)

v dV

(2.10)

F · v − νω2 dV

(2.11)

28

CHAPTER 2. TURBULENCE, BASIC THEORY

ˆ where E is the total energy. If we neglect the viscous term we are led to an inconsistency, since the forcing term puts energy in (F · v > 0), but there is nothing to take it out! Thus, energy keeps on increasing. What is amiss? It is true that for motion with a 1000 km length scale and a velocity of a few centimetres per second we can neglect viscosity when considering the balance of forces in the momentum equation. But this does not mean that there is no motion at much smaller length scales— indeed we are led to the inescapable conclusion that there must be some motion at smaller scales in order to remove energy. Where and how does this motion occur? Boundaries are one important region. If there is high Reynolds number flow above a solid boundary, for example the wind above the ground, then viscosity must become important in bringing the velocity to zero in order that it can satisfy the no-slip condition at the surface. In the ‘boundary layer’ the viscous terms must be at least of the same order as the inertial terms, that is ∂u ∂u ∂ 2u u (2.12) +v ∼ν 2 ∂x ∂y ∂y which implies the viscous terms may be important in boundary layer of approximate size L∼ ν U (2.13)

This is a very small number for geophysical fluids, of order millimetres or less. The other way that energy may be dissipated is through energy dissipation in the interior of the fluid. In order to do this the fluid must somehow generate motion on very small time and space scales. How might this happen? Suppose the forcing acts only at large scales, and its direct action is to set up some correspondingly large scale flow, composed of eddies and shear flows and such-like. Then typically (a mathematician would like to say ‘generically’) there will be an instability in the flow, and a smaller eddy will grow. At first it will grow exponentially, because during the period the eddy is small the large scale flow may be treated as an unchanging shear flow, and the disturbance while still of small amplitude will obey linear equations of motion similar to those applicable in idealized Kelvin-Helmholtz instability. The exponential instability clearly must be drawing energy from the large scale quasi-stationary flow, and it will eventually saturate at some finite (as opposed to infinitesimal) amplitude. Although it has grown in intensity, it is still typically smaller than the large scale flow which fostered it (remember how the growth rate of the shearing instability got larger as wavelength of the perturbation decreased). As it reaches finite amplitude, the perturbation itself may become unstable, and smaller eddies will feed off its energy and grow, and so on. The process has been encapsulated in the following ditty, attributed to L. F. Richardson, Greater whorls have lesser whorls, which feed on their velocity. And lesser whorls have smaller whorls, and so on to viscosity.

2. sometimes called the outer scale. and energy is drained away.e. it is possible to predict what the energy spectrum is. Energy is transferred to smaller and smaller scales in a cascade-like process (fig. the latter condition precludes the presence of solid boundaries but can be achieved in a periodic domain (see e. the intensity of the motion as a function of wavenumber.2). on the size of eddies which can be achieved.2 2.) .2. These eddies grow.2 Inertial range theory Given the above picture. If the transfer occurs between eddies of similar sizes (i. (This naturally puts an upper limit. 2.g. eddies are generated which are sufficiently small that they feel the effects of viscosity. fig. The eddies in reality are embedded in each other.1: Schema of a ‘cascade’ of energy to smaller scales: eddies at a large scale break up into smaller scale eddies. 2.1). and develop still smaller eddies. i. where it becomes dissipated. It seems that this picture of turbulence was envisioned by Richardson in the first part of this century. Let us suppose that the flow is statistically isotropic and homogeneous.2. thereby transferring energy to smaller scales.e. and was quantified by Kolmogorov and Oboukhov.. The picture which emerges if of a large scale flow which is unstable to eddies somewhat smaller in scale. THE KOLMOGOROV THEORY 29 L1 decreasing scale L2 L2 L3 L3 L3 L3 Figure 2. Finally. it is ‘spectrally local’) the transfer is said to be a cascade. There is a flux of energy from the large scales to the small scales.

If β = 0 the flow will not be isotropic. the Fourier transformed field—with similar ˜ identities for v and w. Then we assume that there is a range of scales intermediate between the large scale and the dissipation scale where neither forcing nor dissipation are explicitly important. We write (2.2: A Cartesian periodic domain. depends on the nonlinear transfer of energy being . known as the locality hypothesis.e. k is a scalar wavenumber 2 2 2 given by k 2 = kx + ky + kz . and.16) where E(k) is the modal energy. BASIC THEORY y x ƒ=ƒo+ßy Figure 2.14) where u is the coefficient of the k’th wavenumber —i. t) = u(k. E the energy per unit volume. The idea is to suppose that the forcing scale is much larger than the dissipation scale. The energy in the fluid is given by (assuming density is unity) ˆ E= E dV = = 1 2 1 2 (u2 + v 2 + w 2 ) dV ˜ ˜ (u2 + v 2 + w 2 ) dk ˜ (2.30 CHAPTER 2. Fluid that leaves on one side enters with the same properties on the other side. t)eik·x dk ˜ (2.15) ˆ using Parseval’s theorem. allowing a statistically homogeneous flow. We calculate the energy spectrum of a turbulent fluid by a simple physical assumption. If we decompose the velocity field into Fourier components. TURBULENCE. and E is the energy spectral density. but may still be homogeneous. We suppose that the turbulence is homogeneous and isotropic. Here E is the total energy.15) as ˆ E = E(k) dk (2. first articulated by Kolmogorov. in two dimensions. because of the assumed isotropy. This assumption. then we may write u(x.

undetermined by the theory. E Energy spectrum. T2 T2 (2.17) Dimensional analysis then gives us the form of this function (see table 2. and an energy dissipation rate.17). (2.19) This is the famous ‘Kolmogorov -5/3 law’. is E(k) = Kε 2/3 k −5/3 . the left hand side has dimensionality L3 /T 2 . In (2. perhaps slightly more physical. Another. which is the time taken for a parcel with energy E(k) to move a distance 1/k. no physical processes are important except inertial ones.3). E(k) = f (ε. E(k) ∼ ε 2/3 g(k) L3 L4/3 ∼ g(k). enshrined as one of the cornerstones of turbulence theory (see fig. The parameter K is a dimensionless constant.e. the dimension T −2 on the left-hand-side can only be balanced by ε2/3 since k has no time dependence. The energy spectrum must. It is known as Kolmogorov’s constant and experimentally is found to be approximately 1. way to derive this is to note that we may define an eddy turn-over time τ (k). if the assumptions are right. this intermediate range is known as the inertial range. THE KOLMOGOROV THEORY Table 2.2.5. (2. be a function only of the energy flux ε and the wavenumber itself. ε Dimension 1/L U 2 ∼ L2 /T 2 EL ∼ L3 /T 2 E/T ∼ L2 /T 3 31 sufficiently local (in spectral space). i.1) . τ (k) = (k 3 E(k))−1/2 (2. then if we are in a steady state there must be a flux of energy from large scales to small also equal to ε.20) . If the rate of energy input per unit volume by stirring is equal to ε. also ε. E(k) Energy Flux.18) We see that g(k) must have dimensions L5/3 and the functional relationship we must have. 2. Given this. In the inertial range. therefore. by assumption.1: Quantity Wavenumber. Thus. because the inertial terms and not the forcing of dissipation must dominate in the momentum balance here. k Energy per unit mass. k).2. That is.

22) . Kolmogorov’s assumptions are then equivalent to setting ∼ kE(k) τ (k) (2. the first time using some very high Reynolds number oceanographic observations [17].23) where h is an unknown function and ko the wavenumber at the forcing scale.3: The energy spectrum in three-dimensional turbulence. TURBULENCE. we could conceivably have E(k) = Cε 2/3 k −5/3 h(k/ko ) (2. the assumptions of homogeneity and isotropy are really ansatzes — we make them because we want to have a tractable model of turbulence. This is just as dimensionally consistent as (2.32 CHAPTER 2. again yields: E(k) = K 2/3 −5/3 k . it is transferred (‘cascaded’) to small scales. It requires that the energy be cascaded from large to small scales in a series of steps. in the theory of Kolmogorov. Certainly we can conceive of a thought experiment which is homogeneous and isotropic. This spectral form has been verified many times observationally. Energy is supplied at some rate . where it is ultimately dissipated by viscosity. BASIC THEORY ε ⇑ ε 2/ 3 K -5/ 3 stirring Energy transfer ε dissipation Wavenumber Figure 2. Lacking a more comprehensive or fundamental theory we can test it only through experiment. The essential physical assumption is that there exists an inertial range in which the energy flux is constant.21) which. We essentially postulate that at some wavenumber much smaller than ⇑ ⇑ Kυ ε (2. since we demand that be constant. Now. for then the energy spectra will be determined by spectrally local quantities.19). Without this.

yielding the dissipation wavenumber. In the atmosphere ε is ultimately determined by how much energy is received by the sun. τi << τd and inertial effects dominate. = ν ∂t and so a viscous or dissipation timescale is τd ∼ 1 k2ν (2.26) is inaccurate. These are very sensible formulae.27) ν or the associated length-scale 1/4 ν3 Ld ∼ (2. For L >> Ld . In fact in high Reynolds turbulence. THE KOLMOGOROV THEORY 33 the forcing scale there is no functional dependence of the energy spectra on the forcing scale. It is the only quantity which can be created from the quantities ν and ε with the dimensions of length.26) The wavenumber at which dissipation becomes important is given by equating these.. the energy spectrum falls off more rapidly than k −5/3 and the ‘inertial’ timescale falls off less rapidly than (2.) How big is Ld in the atmosphere? It is rather complicated because the largest scales of motion in the atmosphere are two-dimensional. The Kolmogorov theory allows us to determine this scale. In the inertial range friction is unimportant because the timescales on which it acts are too long for it be important—dynamical effects dominate instead.26) implies. and h(k/ko ) = 1. and (2. For L << Ld .24) The eddy turnover time in the Kolmogorov spectrum is τ (k) = ε −1/3 k −2/3 (2. Further. ε 1/4 kd ∼ 3 (2. (In fact for length-scales smaller than the dissipation scale.2. at smaller timescales the viscous timescale decreases. τd << τi and frictional effects dominate. as the stirring rate (ε) increases or ν decreases the frictional scale decreases.19) does not apply. not all of this goes to stir the atmosphere—some of it is simply radiated back . however.28) ε Ld is sometimes called the Kolmogorov scale. In the momentum equation we have ∂u + . (2. The viscous scale At some small length-scale we should expect viscosity to become important and the scaling theory we have just set up will not work.. and dissipation dominates even more. the −5/3 spectra is observed to a fairly high degree of accuracy — it is perhaps better observed than one might expect. we need to estimate the value of ε.25) 2 u (2.2. However.

Using (2. as well as some other useful scaling relationships. How should we model it? This.29) where r is an arbitrary scaling exponent. This is likely to be strictly valid only in the limit of infinite Reynolds number. (It is important to note that the infinite Reynolds number limit is a limit. but essentially equivalent.e. can be obtained in a slightly different.31) and he velocity gradient scales as v ∼ ε k . is the (unsolved) problem of turbulence.29) holds. Now make the following physical assumptions: First we make the locality hypothesis. (2. 2. way as follows. in the intermediate scales between the forcing scales and dissipation scales. A very crude estimate of ε. likely to be wrong by a factor of ten. 1/3 2/3 . If we for the moment ignore viscosity. This has a number of interesting conse∼ v∼ 1/3 −1/3 from which the assumed constancy of quences. become infinite) at very small scales. we would end up with about 1027 degrees of freedom—a number greater than Avogadro’s number. this term contributes even in the zero-viscosity limit. namely that the energy flux through a wavenumber k depends only on local quantities. So far there is no physics. If we try to simulate the atmosphere on a computer by resolving all scales from the global to the Kolmogorov scale. So dissipation becomes important at millimetre scales in the atmosphere. on a time-average. as we shall see.2.) The time average in practice need be no longer than a few longest eddy turn-over times. which gives the so-called Euler equations. This is because. TURBULENCE. giving ε ≈ 10−8 m2 s−3 . BASIC THEORY to space as infra-red radiation. but this is in fact avoided in any physical situation by the presence of viscosity.30) ∼ λ3r−1 l gives r = 1/3..34 CHAPTER 2. and depending on how local the energy transfer actually is we do not need an infinite Reynolds number for the scaling to be valid in the inertial range.3 Scaling arguments for inertial ranges Kolmogorov’s spectrum. Thus trying to model turbulence is akin to trying to model an ideal gas by following the motion of each individual molecule.28) we find the dissipation scale to be Ld ≈ 1 − 10 mm. These quantities thus blow up (i. but for finite Reynolds number it is made plausible by the locality hypothesis. At length-scale of 100m in the atmospheric boundary layer we estimate velocity fluctuations of order 1 cm/s. and is different from simply neglecting the viscous term in (1. Dimensional analysis then tells us that the energy flux scales as v3 (2. comes from noting that ε has units of U 3 /L. in a nutshell. as does the vorticity ω = × v. namely the wavenumber itself and the energy E(k) or velocity v(k). the Euler equations are invariant under the following scaling transformation: x ⇒ xλ v ⇒ vλr t ⇒ tλ1−r (2. Third we assume that the scale invariance (2. the flux of energy from large to small scales is assumed finite and constant. The velocity then scales as k . Second.1).

3 Two-Dimensional Turbulence In two dimensions the situation is complicated by another quadratic invariant. Energy is input at some large scales.2. using (2. 35 (2.37) + u · ζ = ν 2ζ ∂t where u = ui + vj and ζ = k · ×u.7) conserves not only 1 the energy but also the enstrophy Z = 2 ζ 2 dx = k 2 E(k)dk. how does the distribution of energy and enstrophy change in a turbulent flow? . In two dimensions.22) easily since dimensionally E ∼ v 2 k −1 ∼ ε 2/3 k −2/3 k −1 ∼ ε 2/3 k −5/3 . In the limit of viscosity tending to zero. scale as (δvl )m ∼ m/3 l m/3 ∼ m/3 k −m/3 . the enstrophy.33) kν ∼ 1/4 v dx (2.36) kν with k = kν . this scale becomes smaller and smaller in such a way as to preserve the constancy of the energy dissipation. This result is demanded by the phenomenology. The viscous effects become important at a range given by equating the viscous and inertial terms in the momentum equation. and the magnitude of the stirring largely determines the energy input and cascade rate. TWO-DIMENSIONAL TURBULENCE We can now recover (2.3.34). If viscosity tends to zero. which are the average of the m th power of the velocity difference over distances l ∼ 1/k. The energy dissipation is given by ˙ E= νv · 2 (2.34) ν3 −1 The scale lν ∼ kν is called the Kolmogorov scale. 2.32) The structure functions Sm of order m. that is when νk 2 v ∼ kv 2 which yields . but the energy dissipation does not. lν tends to zero. (2.31) this expression scales as (for a box of unit volume) ε 2/3 2 ˙ E ∼ νkν v 2 ∼ ν 2/3 kn u2 ∼ ε (2. which is the Fourier transform of the energy spectra. It is easily verified that when ν = 0 (4. Finally. the vorticity equation is: ∂ζ (2. We now ask. The scale at which viscous effects become important is determined by the value of the molecular viscosity by (2. the time scales as t ∼ l 2/3 implying that for smaller scales the ‘eddy-turnover time’ on which structures at that scale deform becomes smaller and smaller. scales as S2 ∼ 2/3 k −2/3 .35) Since the length at which dissipation acts is the Kolmogorov scale. In particular the second-order structure function.

. in a nearly inviscid fluid. The vorticity of each element of fluid is conserved as the fluid moves. we should expect that quasi-random motion of the fluid will act to elongate the band but. Initial Later CHAPTER 2. vorticity gradients must increase and the enstrophy distribution is thereby moved to smaller scales. a band of fluid (hatched region) will generically be elongated.36 Figure 2. I. the energy in the fluid is 1 E=− ψζ dx (2. However.4: In incompressible two-dimensional flow.    . but its area (proportional to the fluid mass) will be preserved. vorticity gradients will increase. Since vorticity is tied to fluid parcels.       . . TURBULENCE..  . BASIC THEORY 2. Now. Vorticity Elongation Consider a band or a patch of vorticity.. The following three arguments illustrate why this should be so.4. the vorticity over the hatched area is also preserved.4 Energy and Enstrophy Transfer in Two-Dimensional Turbulence Two-dimensional turbulence differs from three-dimensional turbulence in that energy is expected to be transferred to larger scale.    .. 2. This behaviour arises from the twin integral constraints of energy and enstrophy. as in fig. and if the vorticity is elongated primarily only in one direction (as it must be to preserve area) the integration involved in solving the Poisson equation is such that scale of the streamfunction will normally become larger ..38) 2 Now the streamfunction is obtained by solving the Poisson equation 2 ψ = ζ . thus. since its area must be preserved. This is equivalent to the enstrophy moving to smaller scales.

with wavenumber playing the role of distance from the fulcrum. Suppose we start with some given distribution of energy in wavenumber space and by way of nonlinear interactions we redistribute the energy. and which causes the distribution to spread out in wavenumber space. eddying flow in all directions. II.4.39) (2.40) Any rearrangement which preserves both quantities. respectively E= Z= E(k) dk k 2 E(k) dk (2. as follows. An energy-enstrophy conservation argument A moments thought will reveal that the distribution of energy and enstrophy in wavenumber space are are analogous to the distribution of mass and moment of inertia of a lever respectively. We may formalize this argument. As a consequence. ENERGY AND ENSTROPHY TRANSFER IN TWO-DIMENSIONAL TURBULENCE 37 K1 K0 Initial Spectrum Energy Later Spectrum Wavenumber Figure 2. Thus. Any rearrangement of mass such that its distribution also becomes wider must be such that the center of mass moves toward the fulcrum.2. which can be expected in a nonlinear.5: In two-dimensional flow. the cascade of enstrophy to small scales is accompanied by a transfer of energy to large scales. the ‘center of gravity’ of the energy spectrum will move to large scales (smaller wavenumber) provided that the width of the distribution increases. The total energy and enstrophy are. will tend to move energy to small wavenumbers and enstrophy to large. . energy would move to smaller wavenumbers and enstrophy to larger.

47) (2. An appropriately defined measure of the center of the enstrophy distribution.43) E(k) dk.38 Let ke = CHAPTER 2. 2. moves to higher wavenumber. we normalize units so that the denominator is unity. so that dJ > 0. which is a direct measure of length. The spreading out of the energy distribution is formalized by setting dI d = dt dt Expanding out the integral gives I = = k 2 E(k) dk − 2ke 2 k 2 E(k) dk − ke 2 kE(k) dk + ke (k − ke )2 E(k) dk > 0.43) gives 2 1 dke dI =− < 0. the centroid of the distribution moves to smaller wavenumber and to larger scale (see fig. where the last equation follows because ke = kE(k)dk is. TURBULENCE.44) Thus. Because both energy and enstrophy are conserved. (2. dt E(k) dk dt (2. except that we work with the inverse wavenumber.46) using (2. (2. (2. from (2.41) and. (2. on the other hand.46) where Z(q) is the enstrophy distribution and qe = qZ(q) dq . the energy-weighted centroid. for simplicity. Z(q) dq (2.47) gives J = 2 q 2 Z(q) dq − qe Z(q) dq. Let q = 1/k and assume that the enstrophy distribution spreads out by nonlinear interactions.41).48) . BASIC THEORY kE(k) dk E(k) dk (2.5).45) Expanding the integrand in (2. dt where J = (q − qe )2 Z(q) dq. The proof of this3 is analogous.42) E(k) dk (2.

t ) (2. and the corresponding wavenumber gets larger. and L is a length-scale. ENERGY AND ENSTROPHY TRANSFER IN TWO-DIMENSIONAL TURBULENCE 39 But q 2 Z(q)dq is conserved. are nondimensional quantities. because this is the energy. solely on dimensional considerations.. Thus. It is the first of these that suggests a similarity argument be used.49) 2 dqe 1 dJ =− <0 dt Z(q) dq dt (2. the wavenumber.4.52) ˆ where E.51) where E(k) is the energy spectrum — i. 2. t. t) = U 2 LE(k. L (2. Thus. E(k)δk is the energy in the small wavenumber interval δk. the only parameters available to determine the energy spectrum are U .e. the (square root of the) total energy.54) . and so E should not depend explicitly on L. the length scale characterizing the enstrophy distribution gets smaller.53) (2. That the energy is conserved. That there is no externally imposed length-scale in the problem. There is a priori no ‘external’ length-scale. Write the total energy (per unit mass) of a fluid as U2 = E(k) dk (2. and k. However. dJ d =− dt dt whence 2 qe Z(q) dq (2. we can write ˆ ˆ ˆ E(k.2.50) Thus. III A similarity argument This argument is based on two physical requirements: 1. To make this so. the time. t) = t g(k t ) where g is an arbitrary function of its arguments. Then ˆ E = = Ut U g(kLt ) L L Ut g(U kt). on physical considerations. and its arguments. let ˆ ˆˆ E(k.

and kp must diminish with time. .59) l dt The quantity 1/U k is the ‘eddy-turnover time’ of an eddy of characteristic size 1/k. the eddy turnover time is also the typical time it takes for the eddy containing scales to double in size. (2. CHAPTER 2.57) (The constant may be chosen to be one. the mean wavenumber decreases with time. Conservation of energy now implies that the integral I= ∞ 0 tg(U kt) dk (2. ∞) and A = ϑg(ϑ) dϑ/ g(ϑ) dϑ is a constant. If we were to choose (say) E = t g(k/t ) then we would not obtain a form for (2. the energy weighted mean wavenumber (say k) moves to smaller wavenumber. or larger scale. Defining a characteristic scale l = 1/k then a measure of the rate of increase of scale is given by k = 1 dl = U k(B/A). (2.55) which is independent of L. Defining ϑ = U kt then this requirement is met if g(ϑ)dϑ = Constant. If the wavenumber of the energy peak is denoted kp .56) not be a function of time. BASIC THEORY ˆ E(k. and is due to Batchelor (1969). A little thought will show that it is the most general solution for the energy spectrum that does not depend on the parameter L. Similarly. To see this explicitly. and the energy move to larger scales. TURBULENCE. modulo the constant A.55) that is independent of L xxxx. t) = U 2 LE = U 3 tg(U kt). Suppose that the energy is initially peaked at some wavenumber.58) Ut where all the integrals are over the interval (0. as time proceeds features in the spectrum moves to smaller k. (2. Thus. Thus. ˆ ˆ ˆ ˆˆ ˆ ˆ ˆ It is the choice E = t g(k t ) is crucial here.) The spectrum is a function of k only through the combination ϑ = U kt. Thus. then tkp is preserved. The total enstrophy in this similarity argument is not conserved. This choice is a similarity solution. we have kE dk E dk ktg(U kt) dk = tg(U kt) dk 1 ϑg(ϑ) dϑ = U t g(ϑ) dϑ 1 = A (2. that is the time taken for a parcel of fluid to travel a distance 1/k.40 Therefore.

2. The total dissipation of energy is.37). 2.6: The putative energy spectrum in two-dimensional turbulence. Using the dimensionally correct scaling k 3 E(k) η∼ (2. These arguments lead one to propose the following scenario in two-dimensional turbulence (fig.60) Since vorticity itself is bounded from above we see that energy dissipation goes to zero as viscosity goes to zero. Energy supplied at some rate is transferred to large scales.6). whereas enstrophy is transferred to small scales at a rate η where it may be 2 dissipated by viscosity.61) τ (k) ⇑ ⇑ kυ η . η η transfer η 2 / 3 k -3 dissipation Wavenumber Figure 2. If the forcing is localized at a scale k)f the η ≈ kf η.4. dE = −ν dt ζ 2 dx (2. On the other hand.4.2. Energy and enstrophy are input at some scale LI and energy is transferred to larger scales (toward the fulcrum) and enstrophy is cascaded to small scales where ultimately it is dissipated. In the enstrophy inertial range the enstrophy cascade rate η is assumed constant.1 Inertial ranges in 2D turbulence We now consider how energy and enstrophy might be distributed in forced-dissipative twodimensional flow. It is easy to show that energy dissipation goes to zero as Reynolds number rises. there is no mechanism for the dissipation of energy at small scales in high Reynolds number two-dimensional turbulence. unlike the three dimensional case. we do expect enstrophy to be dissipated at large wavenumbers. Thus. and hence also in the infinite Reynolds number (but finite energy) limit. ENERGY AND ENSTROPHY TRANSFER IN TWO-DIMENSIONAL TURBULENCE 41 ε2/ 3 k -5/ 3 transfer ε ⇑ Energy stirring ⇑ ε. from (2.

(2.4. If the spectra were any steeper then turnover times would actually increase with wavenumber.67) where k0 is a lower wavenumber cut-off. note that timescale (2. but we will mention only the most egregious.61) yields the log-corrected range E(k) = K η2/3 (log(k/k0 ))−1/3 k −3 .34) we assume that the enstrophy flux is constant with wavenumber. We will not delve into these issues. (2. The scaling transformation (2. constant.64) is apparently independent of scale. The exponent n determining the slope of the inertial range is given. Using this in (2.62) where K is also. analogously to (2. The velocity now thus scales as v ∼ η1/3 k −1 .66) 2.37).12).29).1) or (2.36) goes to a finite limit given by ˙ Z = d 1 2 ζ dx = ν dt 2 4 ∼ νkν v 2 ∼ η kν 0 ζ 2 ζ (2.34) kν ∼ η1/3 ν 1/2 (2. which gives. it is length-scale invariant. It is of course also quite possible to obtain (2. it is supposed.68) . This formula recognizes the straining effects of all velocity scales larger than the scale of interest.65) The energy dissipation is easily calculated to got to zero as ν → 0. A useful refinement of the estimate of the inertial timescale is: τ= k k0 (p 2 E(p))dp −1/2 . order one.64) Thus. The appropriate Kolmogorov scale is given by equating the inertial and viscous term in (1. a universal. BASIC THEORY E(k) = K η2/3 k −3 . The enstrophy dissipation. (2.62) from scaling arguments identical to those following equation (2.42 yields the prediction CHAPTER 2.2 Difficulties with the phenomenology The phenomenology of two-dimensional turbulence is unfortunately not quite as straightforward as it might seem. by n = −(2r + 1) yielding the −3 spectra of (4. which seems unphysical. as before.29) still holds. TURBULENCE. and the time scales with distance as t ∼ l/v ∼ η−1/3 .63) which gives λ = 1. analogously to (2. but now instead of (2. (2. Dimensionally we have η∼ v3 ∼ λ3r−3 l3 (2. To begin.

It takes the fluid an infinite time to come to equilibrium with an infinitely long inertial range. are not at this time fully understood. say k. in two-dimensions each octave makes the same contribution. In fact numerical simulations do reveal a slope steeper than k −3 . being a maximum close to k. which is as follows. However. and their precise relationship with the enstrophy cascade. There is one other aspect of the phenomenology which is superficially a problem. implying locality and a posteriori being consistent with the locality hypothesis. it has been shown rigorously that the inviscid equations — (2. a useful measure of this locality is given by the straining at a particular wavenumber. which is not an inconsistency with the rigorous results which only prohibit singularities forming in finite time.70) In three dimensions.69) The contributions to the integrand from each octave are given by E(p)p3 log p (2.2.70) the strain is then local and the shallow slope is forbidden by the Kolmogorovian scaling results. firstly because we are concerned with the zero viscosity limit in (2. The strain. The total strain T (k) at k is given by T (k) = 0 k E(p)p 3 d log p 1/2 (2. The inertial range predictions are based on the assumption of locality. of energy and enstrophy transfers. and there is no inconsistency. the dynamical processes leading to their formation.37) with the right-hand-side set to zero — have no singularities and enstrophy dissipation remains zero.66) implies that enstrophy dissipation remains constant.4. are hardly local after all! This very heuristic result implies that the two-dimensional phenomenology is on the verge of not being selfconsistent. in spectral space. then the enstrophy inertial range (assuming it exists) would slowly spread to larger and larger wavenumbers. ENERGY AND ENSTROPHY TRANSFER IN TWO-DIMENSIONAL TURBULENCE 43 This spectrum is nevertheless likely to be observationally indistinguishable from the uncorrected range. during this period of adjustment the fluid has indeed zero enstrophy dissipation. Now. This is not really a contradiction. this has not fixed the underlying problem with the two-dimensional phenomenology. use of the −5/3 spectra indicates that the contributions from each octave below k increase with wavenumber. often dominated by isolated vortices. and therefore that palinstrophy (mean square curl of the vorticity) is infinite somewhere. However. and suggests that the −3 spectral slope is the shallowest limit that is likely to be actually achieved in nature or in any particular computer simulation. Even if we were to suddenly ‘turn off’ the viscosity in an infinitely high resolution simulation of (2. Why? Well. . from other wavenumbers. rather than a very robust result. However. using (2. However. suppose the detailed dynamics attempt in some way to produce a shallower slope. The fundamental assumption of Kolmogorov scaling is not satisfied.37). if the dynamics organizes itself into structures with a steeper slope (say k −4 ) the strain is quite nonlocal.66). and possibly the enstrophy transfer. However. Only then is the enstrophy dissipation non-zero. In the limit of zero viscosity (2.

The reasons for the loss of predictability were probably only properly understood when it was realized that even systems with a small number of degrees of freedom could be unpredictable. Some authors will draw a distinction between ‘sensitive dependence on initial conditions’ and ‘unpredictability’. for ˆ ˆ example ζ = ζk exp(ix · k) or ζ = ζk exp(ix · k)dk. The former’s meaning is unambiguous. but in our knowledge of them. Even prior to the classical paper of Lorenz (1963) and Ruelle and Takens(1971) it was believed that turbulence was truly unpredictable (see e. In most cases the difference between stochasticity and chaos is merely a difference in our knowledge of the dynamics. the algorithm producing the random numbers is completely deterministic. and the subsequent motion would then be described by an . The former is appropriate in a bounded domain (where the wavenumbers are quantized). At any given instant the equation of motion may be linearized about its current state. and Novikov 1961) notwithstanding the picture of Landau of turbulence as a large collection of periodic. when the equations of motion are not known. Whereas chaos is essentially but a word for deterministic unpredictability and ‘randomness’. We are usually concerned with a finite domain. Thompson 1959. we have chaos. had at this time much impact on theories of.71) where akpq are geometrical coupling coefficients which arise when (2.. here we take them to be synonymous.44 CHAPTER 2. as for example in Brownian motion. The physical space fields ζ (x) may be expressed as an infinite Fourier sum or integral. motions. interestingly enough. not stochasticity. and use the latter (since it is but one word) to mean unpredictability arising from chaos. and if we regard that algorithm as part of the system.g. or ideas of how to cope with. Actually. The latter is sometimes applied only to indeterminism arising out of stochasticity. the latter in an infinite domain. However. and it is normally applied to deterministic systems. most computers have ‘random number generators’ built in.37) is Fourier transformed. For example. stochasticity describes randomness arising from incomplete knowledge of the system. but will nevertheless often replace sums by integrals where it will simplify things. The modern ideas of nonlinear dynamics and chaos have not. Assuming that the dynamical systems arguments applicable to weak turbulence apply to strong turbulence. The unpredictability was thought to arise from the utter complexity of the flow. and these are often used in the simulation of stochastic systems. and hence that a turbulent fluid is in fact unpredictable. However. TURBULENCE.5 Predictability of Turbulence One of the central properties of turbulence is its unpredictability due to nonlinear interactions. then just using the scaling laws we can heuristically obtain estimates of the predictability time for a turbulent fluid. and presumably predictable. strong turbulence. In two-dimensions (for simplicity) the inviscid vorticity equation may be written. in spectral form ∂ζk + ∂t akpq ζp ζq = −νk 2 ζk (2. The hats over transformed quantities have been dropped. BASIC THEORY 2. the difference between chaos and stochasticity lies not so much in the underlying dynamics.

on the other hand. This of course is just the inverse of the eddy turnover time (4. which is kuk where uk is just a typical velocity at scale k. errors will have grown sufficiently that a linear approximation is no longer valid. if we can confine the initial error to smaller and smaller scales of motion. errors initially confined to a scale k at t = 0 will contaminate the scale 2k after a time τk . for then the matrix Akq is dominated by terms close to its diagonal. After a time τk . it becomes meaningful to inquire as to the growth of errors at any particular scale k. It is just that we can put off that time indefinitely if we know the initial conditions well enough.73) If the energy spectrum is a power law of the form E = C k −n this becomes T = [C k (n−3)/2 ]k0 k 2 . namely that they exhibit ‘global regularity’.2). PREDICTABILITY OF TURBULENCE equation. Because the system is chaotic. the payoff is that the ‘predictability time’ (the time taken for errors to propagate to all scales of motion) can be made longer and longer. at that scale errors will saturate but at the same time will begin to contaminate the ‘next larger’ (in a logarithmic sense) scale. meaning they stay analytic for all time provided the initial conditions are sufficiently smooth. and so on. of the form: ∂ζk + Akq ζq = 0 ∂t 45 (2. This does not mean that two-dimensional flow is in practice necessarily predictable.72) and the eigenvalues of the matrix Akq (whose explicit form does not concern us here) determine the short term growth of errors in the system. Two-dimensional turbulence is almost certainly chaotic and an arbitrarily small amount of noise will render a flow truly unpredictable sometime in the future. The total time taken for errors to contaminate all scales from k to the largest scale k0 is then given by. In particular. In three dimensions. but converges if n < 3. If spectral interaction in the inertial range are sufficiently local.5. treating the wavenumber spectrum as continuous. This is consistent with what has been rigorously proven about the two-dimensional Navier-Stokes equations. What does these heuristic results mean? Taken them at face value they imply that two-dimensional turbulence is indefinitely predictable. with or without viscosity.74) converges as k → ∞ So that even if we push our initial error out to smaller . and can reduce the amount of external noise sufficiently.74) As k → ∞ the estimate diverges for n > 3. (n − 3) (2. Thus.2. The predictability time estimate from (2. things are more worrisome. valid for short times. Akq has positive (growing) eigenvalues. T = = k k0 k k0 τk d(lnk) d(lnk) k 3 E(k) (2. the rate of error growth at any particular scale is then given by the size of the appropriate coefficient of Akq . indeed infinite.

we could make forecasts better if we could observe the atmosphere down to smaller scales of motion. (Actually. as it were. the atmosphere is not especially energetic at these scales and we could estimate a typical velocity of about 1 m s−1 giving an eddy turnover time of about 2 days. Here. but we may do the sum manually. this also gives a 2 day eddy turnover time. Observing down to 100. So in 2 days motion at 400 km scales is unpredictable. and may not exist. say U ∼ 2 m s−1 . If we knew the enstrophy cascade rate through the atmosphere we could evaluate the predictability time using the formulae derived above. the atmosphere starts to behave three-dimensionally. the theoretical limit to predictability is governed by the scale at which the atmosphere turns three-dimensional.) In principle. as well as computer simulations. the predictability time does not keep on increasing. but such a proof is lacking. do give similar results though. probably about 100 km. If the Navier-Stokes equations themselves were shown to have finite-time singularities. BASIC THEORY and smaller scales. if not zero. TURBULENCE. since because the system is classical we do not expect such finite time catastrophes. The dynamics at these scales is a little more intense. Suppose then we have no knowledge of the dynamical fields at scales smaller than 200km. This is an indicator that something is awry. I’ve fudged the numbers so they come out reasonable. the eddy turnover times decrease rapidly with scale and the predictability time is largely governed by the predictability time of the largest scale of motion. then at some small wavenumber the local Reynolds number will be small and viscous effects will start to dominate over inertial effects. Now the current atmospheric observing system is such that over continents the atmosphere is fairly well observed down to scales of a couple of hundred kilometers. after about 12 days motion at 6000 km is completely unpredictable.46 CHAPTER 2. However. either with our methodology or with the Euler equations. Fourier transforming in our heads. Beyond the viscous wavenumber. than a three-dimensional fluid. or at least a quasi-geostrophic fluid. However. Continuing the process. Thus. 50 and 25 kilometres would (if the atmosphere remained two-dimensional) each add about a couple of days to our forecast times. Coincidentally (?). The phenomenology thus suggests that the three dimensional Euler equations are not well-posed. and our weather forecasts are essentially useless. If the Euler equations were ill-posed. at small scales of motion the atmosphere starts behaving three-dimensionally. If one were able to prove global regularity for the three-dimensional Euler equations then we would know our analysis were wrong. What does this mean about the weather? In the troposphere the large scale flow behaves more like a two-dimensional fluid. but of course our models of the atmosphere are certainly not perfect. the energy spectrum gets steeper. and as soon as the asymptotic spectra is steeper than −3 we are again assured of indefinite predictability. more careful calculations. So we see that we can’t increase the length of time we can make good weather forecasts for longer than . no classical flow is inviscid. This is probably a little better than our experience suggests as to how good weather forecasts are in practice. it would be a more serious matter. Aside from certain rather intense small scale phenomena. No matter how small viscosity. so after 4 days motion at 800 km is unpredictable. The time it takes for errors initially confined to small scales to spread to the largest scales is simply a few large eddy turnover times (because the eddy turnover times of the small scales are so small). with viscosity. and the correct equations for a classical fluid are the Navier-Stokes equations. At scales smaller than about 100km. it would mean that they do not correctly describe a physically realizable system.

we have P (k)k χ∝ (2. SPECTRUM OF A PASSIVE TRACER 47 about two weeks. and κ is its diffusivity. then τ (k) = [k 3 E(k)]−1/2 Suppose that the turbulent spectrum is given by E(k) = Ak −n (2.2. One other point may be apposite. are determined by slower. In a turbulent system the largest Lyapunov exponent is likely be associated with the smallest scales of motion. that obeys the equation Dφ = F + κ 2 φ. If the dye is injected at a rate χ then. able to change the course of the weather a week or so later.4 applies just as well to a passive tracer in either two or three dimensions as it does to vorticity in two dimensions. 2. If the spectral slope of the turbulence is −3 or shallower. This is the theoretical predictability limit of the atmosphere. no matter how good our models and no matter how good our observing system. (Note that dissipation only reduces the tracer variance.78) (2. Thus we expect a transfer of tracer variance from large-scales to small.76) τ (k) where P (k) is the spectrum of the tracer. The timescales of error growth affecting the larger scales.) The turbulent flow will generically tend to stretch patches of dye into elongated filaments. larger scale processes whether or not the cascade-like growth of error described above is correct. by analogy with our treatment of the cascade of energy.6. in much the same way as vorticity in two-dimensional turbulence is filamented — note that fig. and the error growth associated with this effectively saturates at small scales. The so-called butterfly effect has its origins in this argument: a butterfly flapping its wings over the Amazon is. not the amount of tracer itself. The predictability of a system is sometimes characterized by its spectrum of Lyapunov exponents. This means that the largest Lyapunov exponents probably have nothing whatever to do with the growth of error at the larger scales in a turbulent fluid. k is the wavenumber and τ (k) is the timescale of the turbulent flow. phenomenologically. their ratio is the Prandtl number σ = ν/κ. We assume that the tracer is injected as some well-defined scale k0 .6 Spectrum of a Passive Tracer Let us now consider. such as a dye. the spectrum of a passive tracer.77) . so it goes.75) Dt where F is the ‘forcing’ or injection of the dye. (2. which are the timescales of most interest. and that κ is sufficiently small that dissipation only occurs at very small scales. 2. In general κ differs from the kinematic molecular viscosity ν.

Using (2.77).80) A = Cε 2/3 where ε is the rate of energy transfer to small scales and C the Kolmogorov constant. this is the only spectral slope for which this occurs. If the spectrum is steeper than −3 then the integrand is dominated by contributions from 3 low wavenumbers.77) should be replaced by τ (k) = k k0 −1/2 p E(p)dp 2 (2. The scale at which diffusion becomes important (the diffusive microscale) is given by equating the turbulent time-scale τ (k) to the diffusive time-scale (κk 2 )−1 .82) where C is a constant. χ .1 Examples Consider a range of wavenumbers over which neither viscosity nor diffusivity directly influence the turbulent motion and the tracer. or the eddy-turnover time at large scales. BASIC THEORY P (k)k [Ak 3−n ]−1/2 (2. Inertial range flow in three dimensions . independently. TURBULENCE.83) (This result dates to Oboukhov (1949) and.80) where B is a constant.77).48 then χ∝ and CHAPTER 2.6. This is independent of the flux of tracer. at least approximately. Determination of expressions for these in two and three dimensions are left as problems for the reader. The tracer spectrum becomes P (k) = Dε−1/3 χk −5/3 . If the spectrum is shallower than −3 then the integrand is dominated by the contributions from high wavenumbers and (2. Indeed for k k0 then we can approximate the integral by [k0 E(k0 )]−1/2 . Note that the steeper the energy spectrum the shallower the tracer spectrum.) It is interesting that the −5/3 exponent appears in both the energy spectrum and the passive tracer spectrum.81) where k0 is the low-wavenumber limit of the spectrum. (2. as may happen in two-dimensional turbulence. Then in (2. Experiments show that this range does. and n = 5/3. 1. energy is cascading to larger scales. In all these cases the tracer cascade is to smaller scales even if. The energy spectrum then becomes P (k) = Cχτ (k0 )−1 k −1 (2.5. exist with a value of D of about 0.81) effectively reduces to (2. τ (k0 ). 2. essentially because the equation for the tracer is linear.79) P (k) = BA−1/2 χk (n−5)/2 (2. Corrsin (1951). If the energy spectrum is steeper than −3 then the estimate (2.

2. The appropriate low wavenumber is the dissipation wavenumber. However.85) The passive tracer spectrum now has the same slope as the spectrum of vorticity variance (i. Enstrophy range in two-dimensional turbulence In the forward (enstrophy) inertial range the timescale is just τ (kν ) = η−1/3 (2. The energy spectrum is then very steep.86) the same as (2.88) but by (2.84).e. the tracer spectrum is given by (2. Inverse energy-cascade range in two-dimensional turbulence Now we are supposing that the energy injection occurs at a smaller scale than the tracer injection. Even in the viscous range then. (2. ε This is sometimes called the Batchelor spectrum. although ε is now a cascade to larger scales and D does not necessarily equal D. SPECTRUM OF A PASSIVE TRACER 49 2.83). 3.6. Directly from (2. which is perhaps comforting since the tracer and vorticity obey the same equation in two dimensions.88) ν 1/2 −1 χk .89) In two dimensions the spectral slope in the corresponding (high wavenumber) viscous but non-dissipative range is the same.84) (assuming of course that classical phenomenology holds).85) and the phenomenology predicts that there is no break in the spectral slope . the appropriate timescale at the dissipation scale is given not by (2.87) 1/2 (2.82) will apply. the enstrophy spectrum). The tracer spectrum is then P (k) = D ε−1/3 χk −5/3 . 4. The viscous range of large Prandtl number flow If σ = ν/κ 1 then there may exist a range of wavenumbers in which viscosity is important but not tracer diffusion. (2. and (2. so that there exists a range of wavenumbers over which energy is cascading to larger scales while tracer variance is simultaneously cascading to smaller scales. (2.76) the corresponding tracer spectrum is then P (k) = Bη−1/3 χk −1 .. so that in three dimensions k0 = kν = and τ (kν ) = The tracer spectrum is then P (k) = D ε ν3 ν ε 1/4 (2.

92) In the energy inertial range of three dimensional flow we have τ (k) = (k 2 ε)−1/3 so that P (k) = χ(k)ε−1/3 k −5/3 and (2.91) dk However. 2 (2. Small Prandtl number flow For small Prandtl number the energy inertial range may co-exist with a range over which tracer variance is being dissipated. τ (k) (2.94) (2. The tracer spectrum is then 3 P (k) = χ0 ε−1/3 k −5/3 exp(− κε −1/3 k 2/3 ).90) ν 5. TURBULENCE. (2.97) . η 1/6 kν = 3 (2.91) becomes dχ = −2κχ ε −1/3 k 1/3 .93) where χ0 is the tracer flux at the beginning (low wavenumber end) of the tracer dissipation range. rather it diminishes according to dχ = −2κk 2 P (k). (2. The spectra in a number of these cases are illustrated in figure (xxx) (not yet available).50 CHAPTER 2. The flux of the tracer is now no longer constant. BASIC THEORY of a passive tracer in two dimensions when the viscous range is reached at the dissipation wavenumber. 2 The tracer spectrum thus falls exponentially in this range.95) (2.96) (2. dk Solving this gives 3 χ = χ0 exp(− κε −1/3 k 2/3 ). we may still assume that χ and P (k) are related by χ= P (k)k .

(2.104) .7.98) 2 We can proceed analogously to the case in wavenumber space. such that the total energy (per unit mass) is given by 1 E = u2 = E(ω) dω.2. if we suppose that the angular frequency corresponding to k is given by 1/2 ω = E(k)k 3 = K1/2 ε1/3 k 2/3 (2. in either two or three dimensions. dimensional analysis then yields E(ω) ∼ So the spectrum is L2 ∼ εω−2 T (2.101) cannot really be justified. 2.99) ˜ E(ω) = Cεω−2 (2. supposing that the energy spectrum is a function of ε. Then E(ω) = E(k)∂k/∂ω and we find that 3 ˜ E(ω) = K3/2 εω−2 2 (2. For if we suppose that the energy spectrum is a function of the enstrophy cascade rate η and the frequency ω we have that L2 ˜ E(ω) ∼ ∼ ηa ωb ∼ T −3a T −b T (2. We can also obtain this result.101) ˜ where K is Kolmogorov’s constant. the equality in (2. In the enstrophy inertial range of two dimensional turbulence such dimensional analysis is of little use.102) However. (Note that the dimensions of ε.100) where C is a non-dimensional constant. this is not possible. SPECTRA IN THE TIME DOMAIN 51 2.103) where a and b are putative powers to make the equation dimensionally consistent.1 The space-time spectrum ˆ E = E(k. ω) dω dk ˆ We might suppose that there exists the general space-time spectrum E such that (2.) For the energy inertial range. Physically. makes no specific reference to it being an energy transfer through wavenumber space. the energy transfer rate. the problem arises because each wavenumber in the enstrophy inertial ranges has the same time-scale. along with an estimate for C in terms of the Kolmogorov constant.7. L2 /T 3 . Clearly.7 Spectra in the Time Domain ˜ Let E(ω) be the frequency spectrum in the time-domain.

52 If we write CHAPTER 2. for the energy inertial range tk = (k 2 ε)−1/3 .107) which is satisfied if 0 ∞ g(α) dα = 1. then by simple similarity (i.7.111) 2. so that dtk /dk = −2/3k −5/3 ε−1/3 and 1 dα = −2/3k −5/3 ε−1/3 (2. ε 2/3 k −5/3 g(α)tk dk 31 εg(α)tk dα 2ω 31 α − εg(α) dα 2ω ω − εω−2 h(α)dα (2. BASIC THEORY ˆ E(ω..2 Eulerian Spectra Suppose that a probe that measures velocity fluctuations is put in a turbulent fluid.105) where f (ω. dimensional) arguments we must have f (ω. ˜ E(ω) = = = = = where h(α) is an undetermined function. (2. ˜ The time-spectrum E is given by ˜ E(ω) = ε 2/3 k −5/3 g(α)tk dk. We further require that ∞ 0 g(ωtt )tk dω = 1 (2. k) = g(ωtk )tk (2.110) ω dk Thus. because the velocity variations at a fixed point are due to the sweeping past the probe of small eddies by large . (2. What energy spectrum would result from its measurements? It would not be the ω−2 derived above.106) where tk is the time-scale at wavenumebr k. TURBULENCE.108) α(= ωtk ) is just the argument of g. k) (2.e.109) Using turbulence phenomenology. k) is some function of its arguments. k) = ε 2/3 k −5/3 f (ω.

113) which is dominated by contributions to the integrand from near k itself.’ This is a little unsatisfactory.115) is the spectrum measured at a point.114) which is dominated by contributions from near k0 .112) is equivalent to Taylor’s frozen field hypothesis. The difference is between the contribution to the integral of the shear spectrum Shear2 ∝ l k0 ε 2/3 p 1/3 dp (2. SPECTRA IN THE TIME DOMAIN 53 eddies. To directly measure the spectrum in wavenumber space requires measurements of the velocity correlation between two points. the probe serves to map the spatial frequency of the turbulence into the time-frequency by the relationship ω = Uk (2.7. Note that this does not violate the locality hypothesis. Here we are saying that small eddies are swept by eddies at the energy containing scales.115) Using these transformations is often the simplest way to measure the energy spectrum in a fluid. since one is supposing that the small-scale turbulent structure is frozen as it is swept past the probe. what does the ω−2 spectrum physically represent? Tennekes and Lumley argue that this is a ‘lagrangian’ time spectrum. Using (2.112) where U is the rms velocity of the energy containing scales.19) gives ˜ E(ω) = Kε 2/3 U 2/3 ω−5/3 In the enstrophy inertial range of two-dimensional turbulence a similar argument gives ˜ E(ω) = Kη2/3 U 2/3 ω−3 (2. which says that small eddies are torn apart by eddies of comparable scale. Using it in the Kolmogorov −5/3 spectrum (2.116) (2. a more difficult measurement. and a reader who gives a better explanation will win a small prize. . related to the temporal evolution ‘seen by an observer moving with the turbulent velocity fluctuations. That is.2. If (2. and the velocity spectrum Velocity2 ∝ l k0 ε2/3 p −5/3 dp (2.

and classical mechanics through the well-known but less-understood Kolmogorov-Arnol’d-Moser theorem. A First Course in Turbulence. This is modern account of turbulence. he also made contributions to the theory of war. instead of an electronic computer performing the calculations. N. It is the ultimate reference on the subject. Doering and Gibbon. TURBULENCE. This form of the proof arose in discussion with Isaac Held. performing calculations in unison all directed by a conductor at the front. but easy to recognize when we see it! 2. Turbulence: the legacy of A. This two volume book is encyclopaedic in content and contains a wealth of information. Further Reading Tennekes and Lumley 1972. He was a known as a pacifist. Frisch 1995. written in a slightly personal but readable style. BASIC THEORY Notes 1.’ Remarkably.54 CHAPTER 2. . However. statistics. Monin and Yaglom 1966. Lewis Fry Richardson (1881–1953) was a British scientist best known as the person who envisioned weather forecasting in its current form — that is. 3. This is perhaps as readable account as one can get on the mathematics of regularity and well-posedness of the Navier-Stokes and Euler equations. Kolmogorov was a Russian theoretical physicist. He is also well-known for the ‘Richardson number. A. he envisioned a hall full of people. N. Statistical Fluid mechanics. and in that sense turbulence is like pornography: hard to define. This book remains the classical introduction to the subject. numerical weather prediction. Kolmogorov. A precise definition of turbulence is hard to come by. who made seminal contributions to turbulence (in three papers in 1941 and another in 1962).

suppose that the Coriolis parameter is constant. The two physical effects that pervade these notes. namely rotation and stratification. Further. the basic problem rears its head immediately. rotation and stratification give one something else to hang to. the quasi-geostrophic equations have been the home for most theoretical and numerical studies. then stratification.1 Rotational Effects in Two-Dimensional Turbulence We have seen that one of the effects of rapid rotation on a fluid is its ‘two dimensionalisation. and it becomes possible to address geophysically interesting phenomena without having to solve the whole problem. On the other hand. perhaps ironically. Our plan is consider the effects of rotation first. Then (3. The effects of rotation are of course already playing a role in enabling us to reduce a complex three-dimensional flow to two-dimensional flow. such motions are described by the quasi-geostrophic equations but this is not strictly necessary.Chapter 3 Geostrophic Turbulence Geostrophic turbulence is.’ captured by the Taylor-Proudman theorem. This is the simplest equation with which to study the effects of rotation. there is nothing else to understand other than the problem of turbulence itself. and unless one can ‘solve’ that problem there is little else to say. Nevertheless. 3. loosely speaking. Typically.1) where q = ζ +f . indeed their effects are so pervasive that. the quasi-geostrophic potential vorticity equation reduces to the two-dimensional equation.1) 55 . continue to provide basic constraints on the flow. In the limit of motion of a scale much shorter than the deformation radius. turbulent flow that exists in flows that are in neargeostrophic balance. Dq =0 Dt (3. we can envision highly nonlinear flow in the frontal geostrophic equations and perhaps even in the planetary geostrophic equations. it becomes easier to say something interesting about geostrophic turbulence than about incompressible twoor three-dimensional turbulence. In those problems. and we focus on them.

for example—is unaffected by rotation. as in f = f0 + βy. The flow. A ring of fluid encircling the earth at a velocity u has an angular momentum per unit mass a cos θ(u + cos θ) where θ is the latitude. one might have posited two-dimensionality..3).4) If the dominant term in this equation is the one involving β. 3. one assumes that variations in Coriolis parameter are small. constant rotation has no effect on purely two-dimensional motion. though. as discussed further in chapter (xx).1. that the Coriolis parameter is variable. Dt (3.. Re-write (3.1). then we obtain βv = 0. so long as it is not supposed to be 1/O(Ro ) bigger than the other terms. will be organized into flow in the zonal direction. Then we have D (ζ + βy) = 0 Dt or Dζ + βv = 0 Dt (3. That is. ab initio in which there is no asymptotic restriction on the size of f . and (3. Flow which is already two-dimensional—flow on a soap film.1 Organization of turbulence into zonal flow Heuristics Let us now consider how flow can become organized into zonal bands.e. i. ∂t If ζ ∼ U/L and if t ∼ T then the terms in this equation scale as U LT : U2 : βU L2 (3.] This constraint may be interpreted as a consequence of angular momentum and energy conservation. However. this does not preclude the β term being the dominant one in any subsequent equation.5) . there is no flow in the meridional direction. giving it a meridional velocity) while conserving its angular momentum requires its velocity and hence energy to increase. that βy = O(Ro )f . presuming it exists. from the perspective of two-dimensional turbulence.2) Thus. ironically.56 CHAPTER 3.6) (3. Unless there is a source for that energy the flow is constrained to remain zonal.3) (3. [An aside: From the point of view of the quasi-geostrophic asymptotics used in deriving (3.e. GEOSTROPHIC TURBULENCE becomes simply the two-dimensional vorticity equation Dζ = 0. Suppose.1) in full as ∂ζ + u · ζ + βv = 0. Moving this ring of air polewards (i. Alternatively.

Turbulence Heuristics Can we be more precise about the scaling using the phenomenology of turbulence? Let us suppose that the fluid is stirred at some well-defined scale kf .. The eddy-turnover time is τ (k) = ε −1/3 k −2/3 (3. but note that factors of order unity. and at smaller scales the advective term is dominant. where k is the wavenumber. For large scales the β-term will be dominant. By analogy with the procedure for finding the dissipation scale in turbulence.9) both indicate that at some large scale the vorticity equation Rossby waves are likely to dominate whereas at small scales advection.11) . The cross-over scale. the β term in the vorticity equation will start to make its presence felt.3.7).1. and this in turn is scale dependent. since we have chosen the same length scale to connect vorticity to velocity and to be the β-scale. ζ ∼ Z).e.7) and (3. dominates. we can find the scale at which linear Rossby waves dominate by equating the inverse of the turbulent eddy turnover time to the Rossby wave frequency. 3. Equating these two gives the well-known equation for the β-wavenumber kβ = β U (3. as indicated in fig.7) is by a direct consideration of timescales.10) This equation is the inverse of (3.. Another heuristic way to derive (3.7) This is not a unique definition of the cross-over scale.1. ROTATIONAL EFFECTS IN TWO-DIMENSIONAL TURBULENCE 57 How time scales (i. energy cascades to large scales at that same rate. (3. advectively or with a Rossby wave frequency scaling) is determined by which of the other two terms dominates. β (3. is given by Lβ = U . and even π . β (3. and turbulence.e. If the scale is different. and it is by no means a priori clear that this should be so. the various terms scale as U LT : UZ : βU L (3.1 The cross-over is reasonably sharp.8) where Z is the scaling for vorticity (i. or the ‘β-scale’ Lβ . Then LβZ = Z .9) In any case. producing an energy input ε. At some scale. cannot be revealed by simple scaling arguments such as these. The Rossby wave frequency is β/k and an inverse advective timescale is U k. Then.

energy transfer will be relatively inefficient at those scales where linear Rossby waves dominate. or ‘turbulence frequency. and the mean vorticity. the simple estimate U k where U is an rms velocity.12) From a practical perspective this is less useful than (3.58 CHAPTER 3. i.1: The β–turbulence cross-over. Generation of anisotropy None of the measures discussed so far take into account the anisotropy inherent in Rossby waves.e.. constant) Inverse Timescale Wavenumber Figure 3.e. But the wave-turbulence boundary is not isotropic. or even vorticity. then Rossby waves (turbulence) dominates the dynamics. 2 + ky (3. Equating this to the inverse Rossby wave frequency k/β gives the β-scale kβ = β3 ε 1/5 . nor do they suggest how the flow might organize itself into zonal structures. Where the Rossby wave frequency is larger (smaller) than the turbulent frequency.13) . which is constant. GEOSTROPHIC TURBULENCE (a) Rossby Wave (1/k) (b) Energy transfer rate (k^2/3) (c) Uk (d) Inverse Mean Vorticity (i. The other curves are various estimates of the inverse turbulence timescale. Now.’ These are the turbulent eddy transfer rate. at large (small) scales. U ) at the energy containing scales is likely to be a function of β. whereas the magnitude of the eddies (i. (3. proportional to k 2/3 in a k −5/3 spectrum. it is a little more fundamental from the point of view of turbulence since one can often imagine that ε is determined by processes largely independent of the β. proportional to 1/k. The thick solid curve is the frequency of Rossby waves. the Rossby wave frequency is quite anisotropic.e.10). being given by ωβ = − 2 kx βkx . since it is generally much easier to measure velocities than energy transfer rates. Nonetheless.

which is constant.14) k2 where k is the isotropic wavenumber.2 0 x Kβ / 1. All produce qualitatively similar shapes.2 – 1.2 1.2 1. ROTATIONAL EFFECTS IN TWO-DIMENSIONAL TURBULENCE K y 59 y / Kβ (b) K y / Kβ (c) K 1.2 Figure 3.3. 3.2 – 1. Within the ‘dumb-bells’ Rossby waves dominate and enstrophy transfer is inhibited.and ywavenumber components of the wave-turbulence boundary.2 / Kβ (a) K 0 – 1.17) . we suppose that the turbulent part of the flow remains isotropic.2. If the ‘turbulence frequency’ is parameterized by the simple expression U k then the wave turbulence boundary is given from Uk = − which has solutions kxβ kyβ = = β U β U 1/2 βkx .2: The anisotropic wave-turbulence boundary kβ . The inverse cascade thus leads to a predominance of zonal flow.1. (c) the mean vorticity. proportional to k 2/3 in a k −5/3 spectrum. in wave-vector space calculated three different estimates of the turbulent frequency: (a) The turbulent eddy transfer rate.16) 1/2 sin θ cos1/2 θ (3.15) where the polar co-ordinate is parameterized by the angle θ = tan−1 (ky /kx ).2 0 1.2 – 1. k2 cos3/2 θ (3. If. the wave turbulence boundary is given from the solution of βkx (3. This rather uninformative-looking formula is illustrated in fig. namely ε 1/3 k 2/3 = − kxβ kyβ = = β3 β3 1/5 cos8/5 θ 1/5 sin θ cos3/5 θ (3. as a first approximation.2 – 1. Solving this gives differing expressions for the x.2 1. (b) the simple estimate U k where U is an rms velocity.

The energy cascades to larger scales. In this way a turbulent flow can produce zonally elongated structures. inviscid) energy spectrum in a simulation on a β–plane. Panels (b) and (c) show the spectrum at later times. 3. forced-dissipative simulations show a robust tendency to produce zonally-elongated structures and zonal jets (fig. Does this putative mechanism actually work? Fig.1 Stratified Geostrophic Turbulence Quasi-geostrophic flow as an analogue to two-dimensional flow Now let us consider stratified effects in a simple setting. and it may do this by cascading toward the kx = 0 axis. 3. 3. there is still a natural tendency of the energy to seek the gravest mode.2 3. inviscid two-dimensional simulation on the β-plane.3 shows the freely evolving (unforced. The initial spectrum (a) is isotropic.3: Evolution of the energy spectrum in an unforced. Nevertheless. ky ) space. The panels show contours of energy in wavenumber (kx . but it will be unable to efficiently excite modes in the dumb-bell.2. with an initially isotropic spectrum. where the natural frequency of the oscillation is higher than the turbulent frequency. ‘avoiding’ the region inside the dumb-bell and piling up at kx = 0.2 is dominated by Rossby waves.. Consistently. GEOSTROPHIC TURBULENCE Figure 3. 3. i.60 CHAPTER 3.4). The region inside the dumb-bell shapes in fig.e. toward zonal flow. energy cascades to larger scales. If the flow is stirred at a wavenumber higher than this. The (dimensional) unforced and inviscid . namely the quasi-geostrophic equations with constant Coriolis parameter and constant stratification.

4: (a) Gray-scale image of zonally average zonal velocity√ a function of time and latitude. governing equation is then Dq Dt = 0 (3. In particular. which are obtained by multiplying (3.18) q = 2 ψ + λ2 ∂ ψ ∂z2 2 where λ = f0 /N and D/Dt = ∂/∂t + u · is the two-dimensional material derivative.2.18) by ψ and q and integrating over the domain.3. with appropriate boundary conditions there are two quadratic invariants of the motion. STRATIFIED GEOSTROPHIC TURBULENCE 61 Figure 3. (b) Values of ∂ 2 U/∂y 2 as a function of latitude. where U is the zonally averaged zonal velocity. the energy and the enstrophy.19) . If we rescale the vertical co-ordinate by H . These equations are strongly analogous to the equations of motion for purely two-dimensional flow. The conserved quantities are E= ( ψ)2 + λ2 ∂ψ ∂z 2 dV (3. then λ = (f0 /NH ) is the inverse first deformation radius. as produced in a simulation forced around wavenumber 80 and with kβ = β/U ≈ 10. the height of the domain.

the dynamics of quasi-geostrophic turbulence cannot in general be expected to be isotropic in three-dimensional wavenumber. i. The invariants have exactly the same form as the two-dimensional invariants.23) . we should expect that any dynamical behaviour that occurs in the two-dimensional equations. However. (However. Thus. The energy integral in addition requires a horizontal boundary condition at the top and bottom of the domain. The analogy with two-dimensional flow is even more transparent if we further rescale the vertical co-ordinate by 1/λ. the energy cascade to larger horizontal scales is accompanied by a cascade to larger vertical scales—a barotropization of the flow. GEOSTROPHIC TURBULENCE ∂ 2ψ dV (3. Given that. and enstrophy to small scale. Then the energy and enstrophy invariants are: Z= q 2 dV = 2 2 ψ + λ2 E= and Z= ( 3 ψ) 2 dV 2 3 ψdV (3. In two-dimensional turbulence the equation of motion is isotropic in those two-dimensions. Thus. appropriately scaled by the deformation radius in the vertical. because potential vorticity in the quasi-geostrophic equations is conserved when advected by the horizontal flow.20) ∂z2 where the integral is over a three-dimensional domain.. let z → z/λ.e. the wavenumber is the now three-dimensional wavenumber.21) q 2 dV = (3. The arguments surrounding the transfer of energy to large-scale. Very small horizontal scales Consider very small horizontal scales. Horizontal boundary conditions of either no-normal (ψ = constant) or periodic flow will suffice for this. classical quasigeostrophic turbulence will be characterized by a cascade of energy to large-scale with a k −5/3 spectrum and a cascade of enstrophy to small-scales with a k −3 spectrum. The enstrophy integral in fact holds for each horizontal layer.62 and CHAPTER 3. the governing equation (3. for it is an important and robust process in geostrophic turbulence. should have an analogy in quasi-geostrophic flow. and that depends solely on the energy/enstrophy constraints. Interestingly.22) where the subscript ‘3’ makes explicit the three-dimensional nature of the derivative.18) is decidedly non-isotropic even though some of invariants—the energy and enstrophy—are. In quasi-geostrophic turbulence. we continue to use the convention that ψ = i∂ψ/∂x + j∂ψ/∂y and 2 ψ = ∂ 2 ψ/∂x 2 + ∂ 2 ψ/∂y 2 ). We will come across this again. such that L2 H L2 N 2 Z f02 (3. and setting ∂ψ/∂z = 0 will suffice. are based on the existence of such constraints.

respectively...25) Dt ∂z Temperature advection at the boundary. they may be considered a physical model of two immiscible fluid layers of differing density. Very large horizontal scales Now consider scales for which L2 H The evolution equation is L2 N 2 Z . then this criterion is satisfied for scales much smaller than the deformation radius. that of two-layer flow. 2) (3. 3. qi ) = 0 ∂t We will consider flow in two layers. In either case the parameters λi are given by the expressions λ2 = i f02 g Di (i = 1. which is perhaps a better interpretation for the ocean where the thermocline separates the two fluids.3.2. If the vertical scale of motion remains that of the domain size. governed by the quasi-geostrophic equations: (1 = 1. STRATIFIED GEOSTROPHIC TURBULENCE 63 where LH and LZ are the horizontal and vertical scales of motion. f02 (3. More here. To examine the detailed dynamical behaviour of quasi-geostrophic turbulence. This is probably the most appropriate interpretation for the atmosphere. we turn to a simpler model.2 Two-layer quasi-geostrophic flow ∂qi + J (ψi .18).27) ψ2 + λ2 (ψ1 2 − ψ2 ) These equations may be considered as either a crude finite difference approximation to the continuous equations (3.28) . Alternately. where the sharp increase in stratification at the tropopause inhibits vertical motion.26) where J (a. since the depth of the thermocline is rarely greater than 1 km. In this case the two layers should be on unequal thickness.2.24) D ∂ 2ψ λ2 2 = 0 (3. b) = q1 = q2 = ∂a ∂b ∂b ∂a − ∂x ∂y ∂y ∂x 2 ψ1 + λ2 (ψ2 − ψ1 ) 1 2 (3. The governing equation now reduces to the two-dimensional vorticity equation Dζ /Dt = 0 where ζ = 2 ψ.. 2) (3. but this cannot be guaranteed. especially given the notion of a forward cascade of enstrophy to smaller horizontal and vertical scales. with a rigid-lid boundary condition (w = 0) at the top and bottom.

29) ψ1 + λ(ψ1 − ψ2 ) dA.30) both of which may be thought of as finite difference analogs of the continuous invariants.31) Then the potential vorticities for each layer may be written: q1 = q2 = 2 2 ψ +( ψ −( 2 2 − λ2 )τ − λ2 )τ (3. The enstrophy is conserved layer-wise. of one.33) and the equations of motion may be rewritten as evolution equations for ψ and τ as follows: ∂ ∂t 2 2 ψ + J (ψ. . proportional to the variance of temperature.34) (3. The equations then conserve the total energy. Let us further simplify to the case of two equal layer depths. Then λ1 = λ2 = λ = f0 / g D = (f0 /ND). 2 (3. Thus. λ−1 is simply related to the first baroclinic or internal radius of deformation. ( 2 − λ2 )τ ) = 0 − λ2 )τ ) = 0 (3.64 CHAPTER 3. D. as well as volume integrated. 2 ψ) + J (ψ. ( 2 We immediately note the following: 1. E= and the enstrophy Z= 2 ( ψ1 )2 + ( ψ2 )2 + λ2 (ψ1 − ψ2 )2 dA (3. Baroclinic and barotropic decomposition Define the barotropic and barotropic streamfunctions by ψ τ := := 1 (ψ1 + ψ2 ) 2 1 (ψ1 − ψ2 ) 2 (3. The first two terms in the energy expression represent the kinetic energy. ψ and τ are vertical modes.’ kz . of zero. and the last term is the potential energy. GEOSTROPHIC TURBULENCE where Di is the layer depth and g is the reduced gravity. and τ a baroclinic mode with a ‘vertical wavenumber.32) (3.35) ∂ ( ∂t − λ2 )τ + J (τ. possibly thereby foregoing for √ the moment an easy oceanic relevance. ψ is the barotropic mode with a ‘vertical wavenumber.’ kz . 2 ψ) + J (τ.

geostrophic turbulence may be considered to be similarly comprised of a sum of interacting triads.3. ψ) → ψ (τ. ( 2 − λ2 )τ )dA − λ2 )τ )dA (3.2.43) d d E = (T + C) = 0 dt dt (3. That is. whose vertical wavenumbers must also sum to zero. An easy integration by parts shows that ψJ (τ. assuming once again that the domain is either periodic or has solid walls. gives dT = dt dC = dt where T = ( ψ)2 dA (3.36) (3. τ ) → τ (3. Conservation properties Multiply (3. whose two-dimensional vector wavenumbers sum to zero. τ ) → ψ (ψ.40) is the energy associated with the barotropic flow (the ‘barotropic energy’) C= is the baroclinic energy.44) or total energy is conserved. Just as purely two dimensional turbulence can be considered to be a plethora of interacting triads. The types of triad interaction are: (ψ.38) 3. ( and therefore 2 ( τ )2 dA (3.42) − λ2 )τ )dA = − τ J (ψ. Wherever the Laplacian operator acts on τ it is accompanied by −λ2 . . STRATIFIED GEOSTROPHIC TURBULENCE 65 2. it is as if the effective horizontal wavenumber (squared) of τ is shifted.41) ψJ (τ.35) by ψ and (3. ( 2 − λ2 )τ )dA (3.39) 2 (3. so that k 2 → k 2 + λ2 . ( τ J (ψ.37) (3.35) by τ and horizontally integrating over the domain.

that of two-dimensional turbulence. Barotropic triads: An interaction that is purely barotropic (i. but perhaps richer than. we begin to suspect that the phenomenology to two-layer turbulence is somehow related to.45) dZ = 0. and the associated enstrophy k 2 T (k)dk. Thus. (3. Baroclinic triads: Baroclinic triads involve two baroclinic wavenumbers (say p. purely barotropic flow is exactly the same as purely two-dimensional flow. Phenomenological analysis Two types of triad interactions are possible: I. the relationship between energy and enstrophy in two-dimensional flow. The energy and enstrophy conservation laws for this triad are d (3. Their vector sum is zero.e. Thus spectrally as: T = T (k)dk (3.48) which is analogous to.47) and similarly for C(k). The enstrophy spectrum Z(k) is then related to the energy spectra by Z= Z(k)dk = k 2 T (k) + (k 2 + λ2 )C(k)dk. the barotropic energy. but because of the presence of λ2 not exactly the same as. we may define the spectra of the energy and enstrophy. the conserved quantities are Energy: Enstrophy: II. as if τ = 0) conserves T . GEOSTROPHIC TURBULENCE The enstrophy invariant may be expressed in terms of ψ and τ as: Z= ( 2 ψ)2 + ( 2 − λ2 )τ 2 dA (3. (3..50) . Explicitly.51) (T (k) + C(p) + C(q)) = 0 dt d (T (k) + T (p) + T (q)) = 0 dt d 2 k T (k) + p2 T (p) + q 2 T (q) = 0 dt (3.66 CHAPTER 3. Thus.49) (3. q) interacting with a barotropic wavenumber (say k).46) dt Just as for two-dimensional turbulence.

2. and no production of barotropic energy at k λ. Note that the transfer of enstrophy to small scales in a purely two-dimensional fashion depends on the two-layer nature of the flow.55) That is to say. energy is conserved in the baroclinic field.q. as the horizontal scales become smaller. (p.53) ψ1 + λ2 (ψj − ψi ) ψi [i = (1. 3.51) and (3. with the barotropic mode k acting the mediate the interaction.52). Enstrophy is cascaded to small scales and energy is transferred to larger scales.54) In this case. Then the energy and enstrophy conservation laws collapse to: d (C(p) + C(q)) = 0 dt (3.3. Suppose that we define the quasi-wavenumber k by k 2 := k 2 + λ2 for a baroclinic mode and k 2 = k 2 for a barotropic mode.q. and similarly for p and q . and baroclinic and barotropic modes are both important. A baroclinic triad then behaves as if it were a barotropic triad.56) d k 2 E(k) + p 2 E(p) + q 2 E(q) dt . STRATIFIED GEOSTROPHIC TURBULENCE d 2 k T (k) + (p 2 + λ2 )C(p) + (q 2 + λ2 )C(q) = 0 dt Consider the following cases of baroclinic triad: 67 (3. j = 3 − 1] (3. so does the vertical scale and the first internal deformation radius is not the relevant parameter. In reality.52) 1. k) λ. (p.q. Then neglect λ2 in (3. Thus. qi ) = 0 ∂t where qi = ≈ 2 2 (3. the small scales of a continuously stratified flow may not be representable by a two-layer model: remember that in a continuously stratified quasi-geostrophic model the enstrophy cascade occurs in three-dimensional wavenumber.2. Alternatively. Then energy and enstrophy conservation can be written d (E(k) + E(p) + E(q)) = 0 dt = 0 (3. reconsider the layer form of the equations. each layer is decoupled from the other. This is the most general case. There is no constraint on the transfer of baroclinic energy to smaller scales. 2). until such a time as the scale becomes comparable with the deformation scale at which time the dynamics are no longer quasi-barotropic. ∂qi + J (ψi . k) λ. (p. k) ∼ λ.

58) (3.59) ˙ ˙ Baroclinic instability requires that both C(q) and T (k) be positive.68 wind or solar net input CHAPTER 3. This are formally identical with the conservation laws for purely two-dimensional flow. 4. Since the gravest mode has λ = 0 this implies a barotropization of the flow.60) .5: Schema of two-layer baroclinic turbulence (after Salmon 1980) where E(k) is the energy (barotropic or baroclinic) of the particular mode.57) d 2 k T (k) + λ2 C(p) + (k 2 + λ2 )E(q) dt Then ˙ ˙ ˙ C(p) = − C(q) + T (k) 2 2 ˙ ˙ λ C(p) = − (k + λ )C(q) + k 2 T (k) ˙2 whence ˙ ˙ C(q) = (λ2 − k 2 )T (k) (3. Baroclinic Instability Baroclinic instability in the classic Phillips problem concerns the instability of a flow with vertical but no horizontal shear. Then k 2 ≈ q 2 and the conservation laws are: d (T (k) + C(p) + C(q)) = 0 dt = 0 (3. we expect energy to seek the gravest (smallest quasi-wavenumber) mode. This can only occur if k 2 < λ2 . We can approximate this in a triad interaction for which p (k. (3. GEOSTROPHIC TURBULENCE baroclinic energy scattering into 3-d turbulence barotropic energy K0 KR KD Net loss to boundary layer friction Figure 3. λ). By analogy with twodimensional flow. q.

where both baroclinic and barotropic modes are excited. STRATIFIED GEOSTROPHIC TURBULENCE K vertical ω INITIAL STATE "FINAL" STATE 69 d c e baroclinic ω LINEAR DISPERSION CURVES b c barotropic ^1 2 a F K horizontal Figure 3. At large horizontal scale we imagine some source of baroclinic energy. there is a high wave-number cut-off for baroclinic instability.3. From here there is an enstrophy cascade in each layer to smaller and smaller scales.5 and fig. or in the ocean might be the wind.6).2. At scales larger than the deformation radius. 3. Baroclinic instability effects a nonlocal transfer of energy to the deformation scale. and is not dependent on linearizing the equations and looking for normal mode instabilities (although of course it is consistent with such a calculation (section ??). the energy at large scales is dissipated by boundary layer effects: Ekman drag. there is an inverse barotropic cascade of energy to larger scales. . which in the atmosphere might be the differential heating between pole and equator. This cut-off arises solely from considerations of energy and enstrophy conservation.6: Energy flow in two-layer baroclinic turbulence (after Rhines 1977) Thus. until eventually the scale is small enough so that non-geostrophic effects become important and enstrophy is scattered by three-dimensional effects. which causes the excitation of large-scale barotropic modes. for example. Phenomenology of Baroclinic Turbulence Putting together considerations above leads to the following picture of baroclinic turbulence in a two-layer system (see fig. is a scale independent dissipation mechanism. 3.

(3. Large Scales For large scales. This takes the form of a self-consistency argument. and that the terms on the left-hand-sideof (3. (3. theory of two-layer geostrophic turbulence. in the energy containing scales.e. The wavenumber k0 is the halting scale of the inverse cascade. 2 τ) − U ∂ ∂x 2 τ + Dψ (3. We may suppose that this cascade holds for wavenumbers k0 < k λ. or the domain size. GEOSTROPHIC TURBULENCE 3. energy extracted from the mean flow is essentially trapped at scales larger than the deformation scale.62). Suppose that the main effect of the baroclinic terms.61) 2 2 Thus. τ ) = −U ψ + Dτ . the potential vorticity in each layer is. each layer is decoupled from the other. including the forcing terms involving the mean shear. Then at these large scales the barotropic streamfunction obeys the two-dimensional vorticity equation. 2 ψ) = −J (τ. ψ2 . we shall further argue that. k 2 λ2 . with β = 0. the β–effect.70 CHAPTER 3.2 Small Scales For small scales. q1 = q2 = 2 ψ1 + λ2 (ψ2 − ψ1 ) ≈ ψ2 + λ (ψ1 − ψ2 ) ≈ 2 2 ψ1 . . The baroclinic and barotropic equations respectively become ∂ ∂t and 2 ψ + J (ψ. i. However. enstrophy will cascade to smaller scales and. k 2 λ2 we eliminate terms involving k 2 if they appear along with terms involving 2 λ . and we may expect an energy cascade to large scales with energy spectrum: Eψ = C1 ε2/3 k −5/3 . act to supply energy to the barotropic mode.62) indeed dominate at large scales.62) ∂τ ∂ (3. should there be an energy source at scales smaller than the deformation scale it will cascade to larger scales. Thus.64) where C1 is the Kolmogorov-Kraichnan constant appropriate for the inverse cascade and ε is the as yet undetermined energy flux through the system. determined by one or more of frictional effects.63) + J (ψ. Thus.3 †A Scaling Theory for Geostrophic Turbulence We now construct a phenomenological. but quantitative. ∂t ∂x In the barotropic equation.. |ψ| |τ | and that we can then ignore the nonlinear term on the right-hand-side of (3. baroclinic instability (of the mean flow) occurs at scales larger than the deformation radius.

86)) Eτ = C2 ετ ε−1/3 k −5/3 (3. Scaling arguments would −1 suggest that at the scale k0 the magnitude of τ is given by τ∼ U k0 (3. the energy in the barotropic and baroclinic modes are comparable at sufficiently large scales. Thus. (3.3. 2 ψ) = −U ∂ ∂x 2 τ + Dψ .70) .1 Scaling properties The baroclinic equation may be written as ∂τ + J (ψ. Thus.69) with associated velocity (proportional to the vertical shear of the eddies) at this scale being vτ ∼ U.66) ψ + J (ψ. Specifically.f. (2.67) 3. we expect the baroclinic energy spectrum to be that of a forward cascade a passive tracer in a −5/3 spectrum. given by (c. k0 (3.3. at the energy containing wavenumber k0 λ we expect |ψ| ∼ The barotropic equation (3. 2 k0 (3.68) That is.. Thus.71) (3.64).65) where C2 is the Kolmogorov constant appropriate for a forward passive tracer cascade. we have that ετ = ε. ∂t (3.62) then becomes ∂ ∂t 2 λ|τ | . since energy is not lost to small scales. Now. any energy that is put in at large scales by the interaction with the mean flow (via the term proportional to U ψx ) will be cascaded to smaller scales. τ is stirred by ψ in a mean gradient provided by the shear U .3. from the baroclinic streamfunction is being advected as a passive tracer — it is being stirred by ψ.66) the magnitude of the barotropic streamfunction at this scale is given by ψ∼ λU . Using (3. τ − Uy) = Dτ . εt au is the transfer rate of baroclinic energy and ε is the same quantity appearing in (3. Since the energy density in the former is ( ψ)2 and in the latter ( τ )2 + λ2 τ 2 ∼ λ2 τ 2 the magnitude of ψ must then be much larger than that of τ . †A SCALING THEORY FOR GEOSTROPHIC TURBULENCE 71 Now.

the energy input to the system is given by the polewards heat flux. (Note also that the eddy diffusivity is just the magnitude of the barotropic streamfunction at the energy containing scales. GEOSTROPHIC TURBULENCE with asscoiated barotropic velocity at the energy containing scale being given by vψ ∼ λU . That is. Using (3. implies a mixing length of k0 .77) . However.73) and (3. 2 k0 (3. and in doing so become anisotropic.75) −1 which. Finally. at the scale kβ ∼ β3 ε 1/5 (3.69) and (3.3.73) The correlation between ψx and τ cannot be determined by this argument. if the mixing velocity is the barotropic stirring velocity.63) by τ and integrating over the domain.76) we obtain kβ = β . k0 (3.2 The β–effect As discussed in section 3. because in the absence of frictional processes energy will still seek to cascade to larger scales.72 CHAPTER 3.71) gives κ∼ vb ∂b ∂y = ψx τ ∂τ ∂y (3.76) The energy containing scale is not a priori the same as kβ .) 3. we calculate the ’eddy diffusivity’ defined by κ≡ Using (3.74) λU 2 k0 (3.72) Multiplying (3. Nevertheless. ε = U λ2 ψx τ ∼ U 3 λ3 . we have produced a physically based ‘closure’ for the flux of energy through the system in terms only of the mean shear and other external (at least to quasi-geostrophic theory) parameters such as the deformation scale and the halting scale k0 . the cascade becomes much less efficient and friction does have more time to act to halt the cascade. Uλ (3.1 the β–effect provides a soft barrier for the inverse cascade.

and ψ∼ vψ ∼ U 3 λ3 . 3.3. The theory and related numerical simulations was expounded in a pair of papers by V. Larichev and I.77)). as U increases not only does the eddy amplitude increase as a direct consequence (as is (3.79) (3. namely τ∼ U 2λ . κ∼ λ3 U 3 2 kβ 73 (3. the inverse cascade can extend to larger scales.71) and (3. Similarly. The reason for these is that as βdecreases. thereby increasing the overall energy of the flow. Clearly. the magnitude of the barotropic streamfunction is equal to the eddy diffusivity. Furthermore. †A SCALING THEORY FOR GEOSTROPHIC TURBULENCE Now using (3. Held (see bibliography).71)) but ssalso kβ falls (see (3. The ‘wave-turbulence’ boundary harks back to Rhines (1975) in GFD. in this model. 2.75) and (3. the eddy amplitudes increase more rapidly with the mean shear than previously. . β (3.80) vτ ∼ U. giving rise to a superlinear increase of the eddy magnitudes with U . and reduce the growth rate of baroclinic instability.69) and (3.3.3. β (3. β2 U 2 λ2 .82) Again.3 Notes Discussion 1. the eddies become less energetic with increasing β.77) we obtain for the eddy diffusivity. although the cause is slightly different from the linear case in which the presence of a mean gradient of planetary vorticity usually acts to stabilize a flow.81) (3.69).78) The magnitudes of the eddies themselves are easily given using (3.

Sign up to vote on this title
UsefulNot useful