A raw power curve tells you what a plant produced, not what it should have produced. Without a physical reference, a dull morning and a soiled array look identical on the dashboard. A clear-sky model fixes that: it estimates the irradiance a site would receive under a cloud-free sky, giving every kilowatt-hour a denominator and every forecast a ceiling to be checked against.
The clear-sky index turns a noisy power curve into a quantity you can compare across hours, sites and seasons.
Normalise with the clear-sky index
A clear-sky model (Haurwitz, Ineichen–Perez or similar) takes solar geometry, site altitude and an atmospheric estimate such as Linke turbidity, and returns the expected global horizontal or plane-of-array irradiance for a cloud-free sky. Divide measured irradiance by that value and you get the clear-sky index, kc — not to be confused with the clearness index kt, which divides by extraterrestrial irradiance. kc sits near 0 to 1, occasionally just above 1 under cloud-edge enhancement, and collapses the strong diurnal and seasonal swing into one comparable quantity.
Once output and irradiance share that normalised scale in your time-series store, cross-hour and cross-site comparison becomes meaningful. A 14:00 reading in December and a 09:00 reading in June stop being apples and oranges; each is simply a fraction of its own physical ceiling.
Detect soiling, shading and drift
Compare measured plane-of-array irradiance against the clear-sky curve on genuinely clear intervals and the residual is mostly the losses you care about. A slow downward drift in performance ratio at high kc points to soiling; a sharp, repeatable dip at the same solar position each day is shading from a structure or an adjacent row. Sudden steps usually mean a sensor, string or inverter fault rather than weather.
The same reference sanity-checks forecasts. If a day-ahead prediction implies output above the clear-sky ceiling, it is wrong before the day starts. Clamping the forecast to the clear-sky envelope is a cheap, physically grounded guard against irradiance inputs that overshoot reality.
A static capacity-factor baseline averages away exactly the structure you need: it cannot separate a cloudy hour from a dirty array, and it flatters underperformance whenever the sun is low. A clear-sky reference is the honest denominator. We build it into the ingestion and time-series layer so normalisation, fault detection and forecast checks share one physical ground truth — see how we approach this.