The Global Centre for Risk and Innovation (GCRI)
A model result is useful only when its uncertainty is stated and its fitness for a purpose has been tested. Verification checks that the code solves the equations intended; validation checks that the model represents the real system well enough for the decision at hand. This page sets out how uncertainty is quantified and models are verified and validated, and what evidence serves the decision makers who rely on model results.
Uncertainty and validation · Coupled model A · Scenario set 3
nodes: models · edges: exchanged variables
| Hydrology | 1 day | runoff | Yes |
| Power system | 1 h | demand, outages | Yes |
| Economy | 1 quarter | output, prices | Partial |
| Population | 1 year | exposure | Yes |
Fig.Coupled model runner. Design wireframe with placeholder names and illustrative values, not a record or a live system.
Model validation is the assessment of how accurately a model represents the real system for a specific purpose, after verification has shown the code is correct. With uncertainty quantification it states how far results can be trusted, using data not used in calibration and reporting every major source of uncertainty.
For uncertainty and validation, in GCRI's spatial-intelligence reading, the map opens on true-colour imagery, night lights and land cover with relief in 3D and true colour from NASA GIBS, framed on North America; every other NASA layer, live feed and forecast waits in the layer list.
The interactive map needs JavaScript; the layer list and sources are below.
Layers 3
Hazards and exposure, for this page
Keyboard, with the map focused
Hold Ctrl or ⌘ and scroll to zoom
Latest available
NASA GIBS / EOSDIS · ASTER GDEM (METI, NASA) · Blue Marble (NASA) · © OpenStreetMap contributors (ODbL) · Natural Earth
Loading the latest NASA imagery.
Drag to move. Ctrl or ⌘ and scroll, pinch or the buttons to zoom; hover for the values under the pointer; reset returns to the region.
For uncertainty and validation, in GCRI's spatial-intelligence reading, the map opens on true-colour imagery, night lights and land cover with relief in 3D and true colour from NASA GIBS, framed on North America; every other NASA layer, live feed and forecast waits in the layer list.Satellite and model products as NASA publishes them, each with its own date; indicators, not assessments of any place; claim ceiling INTEGRATED.
Layers and dates
Quantifying uncertainty, validation against observations and the limits of calibration. The dossier below sets out how the risk develops, what is measured, where the data come from, how it is modelled and who decides.
Before asking whether a model matches reality, modellers check that the code is correct, through tests, convergence studies and comparison with known solutions.
A model valid for one question may be invalid for another. Validation is judged against the accuracy a specific decision needs.
Input data, parameters, model structure, numerical error and natural variability each contribute, and each needs a different treatment.
Tuning parameters to match past data can make a model fit for the wrong reasons, so validation needs data not used in calibration.
Each term as this page uses it. Where a standard-setting body defines the term, the wording follows that body.
| Term | Definition |
|---|---|
| Model credibility | The degree of trust that a model is fit for a specific use, based on evidence. |
| Calibration | Adjusting model parameters so that model outputs match observed data as closely as possible. |
| Model discrepancy | The systematic difference between a model and reality that remains after calibration. |
Each indicator carries its unit, its cadence, the kind of source behind it and the evidence label a value would carry. No values are shown: a programme supplies them under its own authority.
| Indicator | Unit | Cadence | Source type | Label |
|---|---|---|---|---|
| Verification tests passed | percent of test suite | per release | Model test records | OBSERVED |
| Validation error against independent data | error metric | per validation | Validation studies | ESTIMATED |
| Prediction interval coverage | percent of observations inside stated intervals | per evaluation | Out-of-sample evaluation | OBSERVED |
| Sensitivity indices | share of output variance by input | per analysis | Global sensitivity analysis | MODELLED |
| Models with documented credibility assessment | count | per assessment | Model documentation | DECLARED |
| Independent reviews completed | count | per model | Review records | REPORTED |
The topic as a computational graph: named sources feed the methods, an evidence review gates the outputs, and the decisions stay with the authorities that hold them.
Fig. 02Nodes carry their type, two metric rows and a state; connectors show what feeds what; dashed frames name the owners; the timeline is the cadence. Design of the pipeline, not a record of a run.
Named public datasets and institutional records, each with its licence class. Sources by agreement are used only under their terms and never republished.
| Source | What it gives | Licence class |
|---|---|---|
| ASME V&V 10, V&V 20 and V&V 40 | Verification, validation and credibility standards | Licensed standards, purchase required |
| NASA-STD-7009 | Credibility assessment of models and simulations | Public domain, US government |
| IPCC uncertainty guidance note | Calibrated uncertainty language | Public, IPCC terms |
| SALib and other open sensitivity analysis libraries | Implementations of global sensitivity methods | Open source, MIT licence |
| Published validation studies in each field | Evidence on model performance | Licence per publication |
Systems Science and Modelling prepares evidence. The decisions below are taken by the authorities named, under their own rules, and never by the platform.
| Who decides | The question | Evidence this page serves |
|---|---|---|
| Model owners | Whether a model is fit for a stated purpose | Credibility assessments; validation reports |
| Decision makers | How much weight to give model results | Uncertainty statements |
| Reviewers and auditors | Whether verification and validation were adequate | Protocols and review records |
Modellers distinguish aleatory uncertainty, the inherent variability of a system, from epistemic uncertainty, which comes from incomplete knowledge and can in principle be reduced. Structural uncertainty, about whether the model's equations are right, is often the largest and hardest to quantify. Deep uncertainty arises when experts cannot agree even on probability distributions.
Calibration adjusts parameters so that a model matches observations. A well-calibrated model may still be wrong: many parameter combinations can fit the same data, a problem called equifinality, and a model tuned to past conditions may fail under new ones. Validation therefore requires tests against data not used in calibration, and ideally under different conditions.
Validation is purpose-specific. A model may be adequate for estimating averages but not extremes, or for one region but not another.
Identifying which inputs drive output variation. Global methods are computationally demanding.
Monte Carlo and related methods, using probability and statistics. Results depend on assumed input distributions.
Updating parameter beliefs with data and representing model discrepancy. Priors and discrepancy models involve judgement.
Testing against independent data, held-out periods or different locations. Suitable data are often scarce for extremes.
No validation proves a model correct. It establishes confidence for a purpose, within stated conditions.
The current service level is R0: information and enquiries. An enquiry receives a reference and names the next step; applications, payments, compute allocations and registrations open only when their team, terms and review capacity exist.