Module IV · Exploration 12
The Model Ladder
No model is the Earth. Each one keeps some processes, leaves out others, and becomes useful only relative to a question.
Same event. Different questions. Different instruments.Playing
Conceptual ↕ Reduced ↕ ComprehensiveEmulator: fast side branch
Familiar responses agree; new-condition responses separate.
Watch first · short video
How We Simplify Reality Without Losing It
See why models form a hierarchy of purposeful simplifications rather than a ranking from wrong to right.
Watch video ↗Applied Mathematics in Geosciences · Episode 7
Related: How Simple Models Explain Complex Earth Systems ↗Then explore it yourself ↓
Four Screens, One Ocean
Select a screen. The shared ENSO display changes because each instrument makes different features visible.
- Retains
- feedback, surface temperature, ocean heat memory
- Omits
- spatial structure, weather, remote interactions
- Helps answer
- What minimal mechanism can produce growth and reversal?
A Hierarchy of Instruments
Move through the physical-model hierarchy. Detail, cost, transparency, and possible behavior change together, but not as a ranking from bad to good.
ConceptualVariables retainedfew Processes representedfew Computational costlow Mechanism visibilityclear
feedback and memory
This level retains the features needed for this question.
Moving downward asks what appears when processes and spatial detail are added. Moving upward asks what can be removed without losing the conclusion. Neither direction is automatically progress.
Build a Model for the Question
Choose a family, then inspect the mathematical object it can represent. No family is automatically appropriate for every question.
Same History, Different Futures
First hold the familiar surface history fixed. Then apply one identical new forcing and watch retained pathways separate the responses.
Held fixedhistorical surface index · axes · intervention timeChanginghidden representation
Matching a familiar curve is calibration. It does not make hidden states or mechanisms identical.
Several models can reproduce the same familiar behavior for different reasons.
Model Detective
All four systems begin with nearly identical familiar output curves. Diagnose them by probing capabilities and responses, not by visual appearance alone.
Choose diagnostic tests
Run at least two tests before identifying the systems.What Was Removed Does Not Disappear
Adjust how the omitted influence returns to a reduced model. The reference, no-closure model, and selected closure update together.
Turn on state dependence to test whether the same unresolved process changes with the large-scale condition. The same unresolved process need not have the same effect under every large-scale condition.
Same Equations, Different Computation
Model error and numerical error are not the same. After equations are chosen, a computer still approximates them with a grid and time steps.
A model can calculate the wrong equations accurately, or calculate appropriate equations poorly. Numerical convergence and physical adequacy must be tested separately.
One Phenomenon, Several Useful Models
ENSO questions should travel through the hierarchy. When an answer changes, investigate the change rather than hide it inside one skill score.
A recharge oscillator isolates feedback, growth, reversal, and ocean heat memory.
A spatial model adds equatorial waves, thermocline structure, and the location of warming.
A coupled global model adds weather, clouds, remote basins, land, and changing forcing.
A learned approximation accelerates forecasts, calibration, or large ensembles.
Which feedback initiates growth?Where is memory stored?What reverses the event?What creates event diversity?Which uncertainty limits prediction?
Explore ENSO case study →The Models Disagree. What Should We Test Next?
Select one or two actions. The goal is not to collect the most data, but to choose evidence that can separate the competing explanations.
A useful model hierarchy does not merely produce several answers. It helps identify the observation or simulation that can distinguish among them.
What This Studio Model Leaves Out
These browser models are deliberately simplified and do not reproduce the full equations, resolution, parameterizations, or cost of operational weather and climate models.
The “comprehensive” representation is an educational synthetic model, not a global Earth-system simulation. Emulator extrapolation patterns are illustrative, and real emulators can fail in many other ways.
Real model families branch, overlap, and use different variables, resolutions, closures, and evidence. The purpose here is not to rank models, but to expose the questions that must be answered before an output deserves trust.
Sources, methods, and synthetic-data note
- Isaac Held on the value of model hierarchies in climate science.
- Literature on conceptual, reduced, comprehensive, and learned Earth-system models.
- Research on deterministic, stochastic, memory-aware, and learned closures.
Every response on this page is generated locally from deterministic, seeded teaching models. The comprehensive representation is an educational proxy, not a general circulation model; none of the outputs is a climate forecast.