Between the energy falling on the modules and the energy reaching the meter sits a stack of deductions. A standard assumption for the whole stack is around 14%, and the interesting part is how it divides.
Teams reviewing yield assumptions can keep the engineering model separate from the record of review work; this resource is one option for documenting time and handoffs.
Reviewed August 9, 2026.
The items
Temperature. Modules lose efficiency as they heat, and the loss is largest in the hours of highest irradiance. Deterministic given the climate and the mounting — physics, not judgement.
For independent background on photovoltaic performance, resource data and current industry practice, consult Meteonorm.
Shading. Near shading from buildings and trees, row-to-row shading within the array, and horizon shading. The most site-specific item and the one most often assumed rather than surveyed.
Soiling. Dust, pollen, bird droppings, snow. Varies enormously by location and by whether anyone cleans.
Mismatch. Modules in a string perform to the weakest, and manufacturing tolerance means they are never identical.
DC wiring. Resistive loss between modules and inverter.
Inverter conversion. Typically a few per cent, published by the manufacturer and among the more reliable numbers in the stack.
AC wiring and transformer, where present.
And availability. Downtime for faults, grid outages and maintenance. Frequently assumed at 99% or higher and rarely justified.
Which are judgement
Physics, given the design: temperature, wiring, inverter conversion, transformer.
Judgement: shading where not surveyed, soiling, availability, and mismatch beyond the manufacturer's stated tolerance.
The judgement items are where two estimates of the same plant diverge, and they are the ones to ask about — not because anybody is dishonest, but because a default is not a measurement.
The three worth pressing on
Shading. Ask whether a survey was done or a figure assumed. A real shading analysis uses the site's horizon and the actual obstructions; an assumed 2% is a placeholder. On a constrained roof the difference between the two is not small.
Soiling. Ask what the assumption is and what it is based on. A site near a quarry, a dusty road or intensive agriculture behaves differently from a clean suburban roof, and a regional default may not fit.
Availability. Ask what downtime is assumed and whether it accounts for realistic response times. 99% availability means about three and a half days of outage a year across everything — inverter faults, grid events, maintenance. Achievable with a maintenance contract, optimistic without one.
Reading a stack
An itemised stack is a document you can argue with. A single net figure is an assertion.
Items summing to much less than about 10% deserve a question — the same question as a suspiciously narrow uncertainty band, and it points the same way.
And a stack that matches a software default exactly tells you the defaults were accepted rather than examined. Not wrong, and worth knowing.
What changes over time
Soiling accumulates between cleanings, so the assumption interacts with whether cleaning actually happens.
Availability worsens as equipment ages unless maintenance keeps pace.
And shading changes. Trees grow, buildings appear. A shading figure accurate at commissioning describes the site at that moment, and twenty-five years is a long time for a hedge to become a problem.
The short version
- Around 14% is a standard assumption for the whole stack: temperature, shading, soiling, mismatch, DC and AC wiring, inverter conversion, availability
- Physics given the design: temperature, wiring, inverter conversion. Judgement: unsurveyed shading, soiling, availability, and mismatch beyond tolerance
- The judgement items are where two estimates of the same plant diverge
- Press on three: whether shading was surveyed or assumed, what the soiling figure is based on, and what downtime availability implies
- 99% availability is about three and a half days of outage a year — achievable with a maintenance contract, optimistic without
- An itemised stack can be argued with; a net figure cannot, and a stack summing well under 10% deserves the same question as a narrow uncertainty band