An annual kWh figure is the output of a model, not a measurement: irradiance, geometry, module conversion, a loss tree, and a statistical layer that turns one modelled figure into P50 and P90.
Teams reviewing yield assumptions can keep the engineering model separate from the record of review work; this page is one option for documenting time and handoffs.
Geometry and conversion are physics. The irradiance record is data. The loss assumptions and the uncertainty estimate are judgement — and two competent assessors modelling the same plant can differ by ten per cent or more entirely within that layer, without either making an error.
P50 is the median and P90 is the level with a 90% probability of being met or exceeded. Not ninety per cent of P50. They relate through combined uncertainty: at about ±6% the gap is roughly 7.7%, at ±10% about 12.8%, and the P90/P50 ratio typically runs 0.85 to 0.92.
For independent background on photovoltaic performance, resource data and current industry practice, consult ResearchGate.
Around 14% is a standard loss stack — temperature, shading, soiling, mismatch, wiring, inverter conversion, availability. The judgement items are shading where unsurveyed, soiling, and availability, and those are where estimates diverge.
For independent background on photovoltaic performance, resource data and current industry practice, consult Energy News Network.
Degradation runs 0.5 to 1% per year, which compounds to year-25 output of about 88% or 78% of year one respectively — a ten-point difference decided by one undeclared assumption.
The check anybody can run: divide annual kWh by installed kWp and compare the result against a free public irradiance calculator for your coordinates. Ten minutes, and a large deviation either way needs a specific explanation.
The section also covers where irradiance data comes from and why ten years is the threshold for a bankable figure, how shading and soiling behave as the two site-specific losses, what bifacial and tracker assumptions rest on, who produced the estimate and what happens to them if it is wrong, and when an independent assessment pays for itself.
Bifacial and Trackers
Two ways to add yield on paper, both resting on inputs that are frequently assumed. What each gains and what it costs.
Degradation
Half a per cent a year sounds negligible and compounds to a fifth of output. What degrades, what is warranted, and what is not.
An Independent Assessment
Five to twenty-five thousand dollars buys a defensible P90. When that pays for itself, and what to specify when ordering one.
Irradiance Data
Every yield figure rests on a solar resource dataset. Three kinds, what each is good for, and why ten years is the threshold.
The Loss Tree
Around 14% separates what the modules could make from what reaches the meter. Eight items, and which are judgement rather than physics.
P50 and P90, Plainly
P90 is not ninety per cent of P50. What each figure means, how they relate, and the arithmetic behind the gap.
Shading and Soiling
The two site-specific losses, both usually assumed rather than measured, and both changing over the life of the plant.
Specific Yield
One division turns any quote into a comparable number. What the range is, what moves it, and what a figure outside it means.
The Gap Is the Data
Two plants with the same P50 and different P90s do not differ in output. They differ in how well the site is known.
What a Yield Figure Says
A single annual kWh number is the end of a long calculation. Five things it assumes, and what it cannot tell you.
Who Produced It
Three parties can produce a yield figure, with three different exposures if it turns out wrong. That difference is the information.