Assessment of whether a mathematical or computational model is sufficiently accurate and appropriate for its intended purpose by comparing its predictions to independent data, observed behavior, or established benchmarks.

Principle

Use independent evidence and performance criteria relevant to the decision context to judge whether model outputs and behaviors are consistent with reality; emphasize representativeness of test cases and quantification of predictive error.

Demonstration

A rainfall–runoff model calibrated on a subset of gauged basins is evaluated against streamflow observations from withheld basins and independent storm events to determine whether forecast skill meets water-management requirements.

Misapplication

Using the same data for calibration and validation, selecting validation metrics that do not match the decision objective, or equating a large number of passing tests with universal correctness.

Consequence

When done correctly, stakeholders obtain justified confidence in model-based decisions, or conversely identify model limitations that drive revision or additional data collection.

Reversal

Declaring a model invalid by demonstrating systematic predictive failure on independent tests, or confusing validation with verification (the latter checks code correctness rather than predictive adequacy).

Boundary

Validation does not prove truth universally; it is conditional on the tested scenarios, the quality of independent data, and the metrics chosen; it excludes detection of coding bugs addressed by verification.

Semantic Tension

Often conflated with verification and calibration; tension arises when 'validation' is used informally to mean code checking, parameter fitting, or model selection rather than independent predictive assessment.

Synthesis

Model validation is the discipline of testing a model’s fitness-for-purpose by comparing its predictions to independent evidence under conditions relevant to the intended use, thereby informing trust and limitations.