EART60071 Chap.3 Advection, the CFL Number and Numerical Diffusion
Advection, the CFL Number and Numerical Diffusion
Advection, the CFL Number and Numerical Diffusion treats the transport equation that every grid-based model in the module relies on. A conserved quantity carried by a steady wind should simply slide along unchanged, yet the first-order upwind scheme in the practical makes a triangular plume lower and wider as it travels.
The chapter explains that numerical diffusion, defines the Courant number as the distance moved per step divided by the grid spacing, and shows why the scheme becomes unstable once that number exceeds one. Hand calculations on a single spike show the three behaviours side by side: smearing below the limit, a clean shift exactly at it and growing oscillation above it.
The same ideas return in the column model, where rain and cloud are carried between levels, and in the shallow-water model, so a clear grasp here pays off across three weeks of the module and in every question that shows transport output.
What this chapter covers
- 01
The advection equation as transport without mixing
- 02
The first-order upwind update
- 03
The Courant number and its meaning
- 04
Stability needs C no greater than one
- 05
Numerical diffusion below the limit
- 06
Recognising instability in a plot
Worked example · free
Choose a stable step for a 15 m/s transport problem
- 1C = |v|Δt/Δx = 15 × 12/300 = 0.6, so the plume moves 0.6 of a cell per step.
- 1Stability needs C ≤ 1, so Δt ≤ 300/15 = 20 s.
- 1A 25 s step gives C = 1.25, above the limit, so growing oscillations should be expected.
- 2Ten steps of 12 s at 15 m s−1 move the exact shape 1800 m, six cells, while upwind keeps the right position but lowers and widens the peak.
Key terms
- Advection equation
- A partial differential equation describing the transport of a conserved quantity by a flow without mixing.
- Upwind scheme
- A difference method that takes the spatial gradient towards the side from which the flow arrives.
- Numerical instability
- Growth of spurious oscillations caused by the numerical method rather than by the process being modelled.
- Partial differential equation
- An equation involving derivatives with respect to more than one variable, such as time and position.
- Courant number
- The dimensionless ratio of the distance a feature moves in one time-step to the grid spacing.
Advection, the CFL Number and Numerical Diffusion FAQ
What is the CFL number and why does it matter?
It is the speed times the time-step divided by the grid spacing, the fraction of a cell a feature moves per step. For the first-order upwind scheme it must not exceed one: above that, the solution develops growing oscillations that come from the method. In the practical the default gives 0.5, a 5 s step gives 1.0 and a 6 s step gives 1.2.
Why does the plume spread when the equation has no diffusion term?
The scheme stores one value per cell. When the plume moves only part of a cell in a step, upwind blends the incoming and resident amounts across the whole cell, and repeating that blend every step behaves like mixing. The peak falls and the base widens even though the total amount is conserved.
Why is a Courant number of exactly one special in the practical?
For constant-velocity, one-dimensional upwind transport, C equal to one moves every cell's contents exactly one cell downstream per step, so the profile shifts without any blending and the numerical diffusion disappears. It is a property of this simple problem, not a general recipe, and any larger value is unstable.
Does a stable transport scheme give the right answer?
Not necessarily. Meeting the stability limit only prevents the solution from blowing up. A first-order upwind result at a Courant number of one half is stable yet still lowers and widens a transported peak with every step. The column-model lecture notes that the scheme used there is better than upwind, which shows that accuracy and stability are separate properties of a method.
Exam move
Revise this chapter by doing one update by hand until it is automatic. Take four cells holding 0, 4, 0 and 0, and apply the upwind update at Courant numbers of 0.5, 1.0 and 1.2, checking that the total stays at four. Note which result smears, which shifts cleanly and which produces a negative value, and attach the words numerical diffusion and numerical instability to the right cases.
Then practise the inverse question: given a speed and grid spacing, find the longest stable step, and given a desired finer grid, say what must happen to the step. Finally, connect the idea forward: rain falling through a column, cloud water lifted by an updraft and fluid moving in the shallow-water model are all advection, so the same limit and the same artefacts apply whenever you interpret their output.