Lab · Exploration 3
Many Clocks, One Earth
How can a model ignore the details without losing their effects?
A cloud may live for an hour. A storm may last for days. The ocean can retain memory for decades. Earth-system models must connect processes that operate on radically different clocks.
How can a model follow the slow story without losing the influence of everything happening too quickly or too finely to resolve?Enter the multiscale lab ↓
01 · Relative time
Choose the question
FastComparableSlow
relative to: Tomorrow’s rainfall
secondsminuteshoursdaysmonthsyearsdecades
fast
Cloud life cycle
A cloud can form, precipitate, and disappear within a small fraction of a climate simulation.
Cloud and storm evolution are central. Turbulent motions are fast but still influence rain formation.
A process is not fast or slow by itself. It is fast or slow relative to the question being asked.
The vertical positions separate labels only; they do not represent another physical quantity.
Watch first · short video
How We Simplify Reality Without Losing It
Start with the central modeling choice: which details can be removed while their effects are retained?
Watch video ↗Applied Mathematics in Geosciences · Episode 7
Related: The Hidden Tradeoff: Detail, Accuracy, and Insight ↗Then explore it yourself ↓
Experiment one
What survives when we average?
A model often replaces rapid fluctuations with a slower description. Try doing that yourself.
Averaging changes the question. It may preserve the slow tendency while discarding the exact path of every rapid fluctuation. But the averaging window matters. Too little averaging leaves fast variability. Too much can delay or distort the signal we wanted to recover.
What assumption did averaging use?
Averaging works best when the fast process changes many times before the slow process changes substantially. This separation of time scalesA large difference between the characteristic times of two interacting processes. gives the fast behavior enough time to reveal a stable net effect.
Experiment two
Can fluctuations matter when their average is zero?
The fast fluctuation has zero mean by construction. The question is whether its effect must also average to zero.
⟨Y⟩ = 0
The fast fluctuation has zero mean by construction. The question is whether its effect must also average to zero.
Two equally frequent values
Y = −1Y = +1
⟨Y⟩ = 0
F(Y) = Y ⇒ ⟨F(Y)⟩ = 0
F(Y) = Y2 ⇒ ⟨F(Y)⟩ = 1
⟨F(Y)⟩ ≠ F(⟨Y⟩)
Zero mean of a fluctuation does not imply zero mean of its nonlinear effect.
A two-value example
If Y alternates equally between −1 and +1, its average is 0. But Y² is always 1. The fluctuations disappear from the mean of Y, but not from the mean of their effect.
Clouds, radiation, chemical reactions, biological growth, and melting all contain nonlinear responses. Replacing a fluctuating condition with its mean can therefore change the predicted average effect.
Experiment three
A finer grid can change the model
Processes smaller than a numerical grid must be summarized through a parameterization, a rule that estimates their net influence from variables the model does resolve.
total effect = resolved contribution + parameterized contribution
Total effectBalanced representation
Resolved 0.23
Parameterized 0.77
Total 1.00
Unresolved fraction: 77%
Increasing resolution does not simply draw the same model more sharply. It changes which physics the model calculates directly and which physics must be summarized.
Because the resolved–unresolved boundary moves, the parameterization must change with resolution. This rule is often called a closureThe rule used to represent how unresolved processes influence resolved variables. : it connects the missing scales back to the variables the model retains.
A finer model is always the same model with more detail.
A finer grid may change which processes are represented directly, which are parameterized, and what the model’s variables and parameters mean.
Break the assumption
When does the reduced model fail?
Begin with a successful reduction, then break one assumption at a time and watch why the slow prediction fails.
Baseline
Reduction works
- clear time-scale separation
- little unresolved memory
- approximately stable fast-process statistics
Overlap the clocks
Averaging assumes that the fast process explores many states while the slow state changes very little. When the clocks overlap, that assumption fails.
Add memory
The present slow state is no longer enough. Recent history carries information about what happens next.
X(t0) = X*
Change the environment
The unresolved physics can respond to the resolved state. A single fixed average is therefore not valid everywhere.
Fixed parameterization
= constant
State-dependent
= (X)
The clocks overlap.The unresolved process remembers its past.The unresolved statistics change with the environment.
Explore the mathematics
dXdt = f(X,Y)
dYdt = 1ε g(X,Y)
0 < ε ≪ 1
- X is the slow variable.
- Y is the fast variable.
- A small ε means Y changes much more rapidly than X.
- Deleting Y is not valid because the slow equation still depends on it.
Averaging seeks the typical influence of Y while X is nearly fixed.
slow change
=
average fast influence
+
remaining fluctuations
dX = (X) dt + Σ(X) dWt
Here (X) is the average tendency. Σ(X) sets the size of the remaining variability, and dWt is an idealized random increment. The random term summarizes unresolved variability; it does not claim that the underlying physics has no structure.
Final synthesis01 02 03 04
What did the missing scales leave behind?
Fast is relative.
A process is fast only relative to the question and observation window.
Averaging is a model.
It preserves selected effects rather than simply deleting detail.
Zero mean does not mean zero influence.
Nonlinearity and correlated fluctuations can create systematic effects.
Resolution changes the scientific description.
A parameterization belongs to a particular resolved–unresolved boundary.
The Earth runs on many clocks. A useful model decides which clocks to follow directly, and how the others continue to shape the story.
Sources and further reading
- A. J. Majda and B. Gershgorin, “Quantifying uncertainty in climate change science through empirical information theory,” Proceedings of the National Academy of Sciences, 2010.
- R. Klein, “Scale-dependent models for atmospheric flows,” Annual Review of Fluid Mechanics, 2010.
- D. Stensrud, Parameterization Schemes: Keys to Understanding Numerical Weather Prediction Models, Cambridge University Press.