Definition
Plain language
A score for how much better a weather prediction is than a simple default guess.
As stated in the literature
A normalized verification metric comparing a forecast's error against a reference baseline (climatology or persistence); higher values indicate greater predictive gain over the baseline.
Also called: forecast-skill, forecast-skill gain
Why it matters: Raw accuracy can look impressive in a place where the weather barely changes, so skill is what tells you whether a model is adding real information.
For example, a forecast that says "rain tomorrow" earns skill only if it beats the lazy guess of "same as today" or "whatever is typical for this date."