The number on your quote is a modelled annual yield. The number on your generation meter after twelve months is a measurement. They will not match, and the gap is usually put down to a bad summer or a slack installer.
Neither is normally the explanation. The gap is built into how a panel is rated, and it decomposes into terms that can be listed and quantified.
The rating is a laboratory condition, not a roof
A panel sold as 400 W is rated at Standard Test Conditions: 1,000 watts per square metre of irradiance, a cell temperature of 25°C, and a defined light spectrum. Those three things essentially never occur together on a roof. When irradiance is high enough to approach 1,000 W/m², the cells are far hotter than 25°C, because that much light is landing on a dark surface with poor airflow behind it.
So the nameplate is a benchmark for comparing panels against each other. It is not a prediction of output.
The loss budget, with numbers
Below is the full chain for a specific, stated case: a 4 kWp crystalline silicon array, roof-mounted, facing due south at 35°, in the English Midlands. It is modelled by PVGIS, the European Commission's PV performance tool, using its SARAH2 radiation database averaged over 2005 to 2020.
| Stage | Effect | What it is |
|---|---|---|
| Irradiance on the plane of the array | 1193.1 kWh/m² | Starting point |
| Reflection at oblique sun angles | −3.2% | Light bouncing off the glass rather than entering it |
| Spectral response | +1.8% | A gain: silicon suits a cloudy sky slightly better than the reference spectrum |
| Temperature and low irradiance | −7% | Cells above 25°C, and poor efficiency in dim light |
| Cabling, mismatch, inverter, soiling | −14% | PVGIS's default system loss assumption |
| Delivered | 3,761 kWh/year | −21.2% |
The terms multiply rather than add: 0.968 × 1.018 × 0.930 × 0.860 = 0.788, which is the 21.2% total. Note the third row is positive.
Taking them in turn.
Reflection, 3.2%
Glass transmits light well when the sun is square on and progressively worse as the angle opens up. Early morning and late afternoon sun arrives at a shallow angle to the panel and a measurable fraction of it bounces off. Anti-reflective coatings reduce this. They do not remove it.
Spectrum, a gain of 1.8%
This one runs the other way, and almost nobody mentions it. Crystalline silicon responds slightly better to the bluer, more diffuse light of an overcast sky than to the reference spectrum it is rated against. At British latitudes that is worth just under two per cent, working in your favour. Any loss breakdown that lists spectral response as a deduction has copied it from somewhere sunnier.
Temperature and low light, 7.0%
Silicon loses power as it heats. It also converts poorly in dim conditions, and a British year contains a great deal of dim. PVGIS reports these two together, and 7.0% is the combined annual cost for a roof-mounted array, where heat builds up behind the panels. The same array on a ground-mounted frame with air moving freely behind it loses less. The mechanism, and what it costs on an actual hot day rather than averaged across a year, is in why panels produce less in a heatwave.
Cabling, mismatch, inverter and soiling, 14%
This is PVGIS's default bundle of everything downstream of the cells. Resistive losses in DC and AC cable. Inverter conversion efficiency, typically 96 to 98% at its best operating point and worse at very low input. Mismatch, because no two panels in a string are identical and a series string is held to the behaviour of its weakest member. Dirt on the glass.
Fourteen per cent is an assumption rather than a measurement, and it is the term in the whole table with the most room in it. A good installation with a well-sized inverter and short cable runs does better. A badly strung array with an undersized inverter does worse.
Shading, which is not in the table at all
PVGIS accounts for the far horizon from terrain data. It knows about the hill behind your village. It knows nothing about your neighbour's chimney, the flue on your own roof, or the sycamore that will be four metres taller in a decade.
Object shading is the single largest source of disagreement between a modelled yield and a measured one, and it is the one an installer is supposed to survey. What it costs, and why one shaded cell does not cost you one cell's worth of output, is in do solar panels work in shade.
Degradation, which happens later
The array you measure in year twelve is not the array that was commissioned. Field studies put the median crystalline silicon degradation at 0.5 to 0.6% a year, and the mean higher at 0.8 to 0.9%, because a minority of poor performers drags it. Over a 25-year quote that is the difference between finishing at 88% of original output and finishing at 81%. The detail is in how much output panels actually lose.
Then there is the weather, which is nobody's fault
Even with every loss above accounted for exactly, the same roof produces different amounts in different years. PVGIS reports a year-to-year standard deviation of 130 kWh on the 3,761 kWh reference case. That is 3.5%.
This is the reason every figure on this site is published as a range with a stated floor. The floor is the number worth planning against.
The seasonal spread within a year is far larger than the variation between years.
| Month | Output (kWh) | Irradiation (kWh/m²) |
|---|---|---|
| January | 144 | 42.9 |
| February | 201 | 60.5 |
| March | 331 | 101.4 |
| April | 416 | 131.5 |
| May | 456 | 146.7 |
| June | 450 | 146.9 |
| July | 461 | 152.5 |
| August | 410 | 133.3 |
| September | 352 | 112.5 |
| October | 240 | 74.6 |
| November | 167 | 50.5 |
| December | 134 | 40 |
December delivers about 29% of what July delivers. Any estimate of what solar will do for your bill has to be built on this shape, not on an annual average divided by twelve.
Reading your own quote against this
Three checks are worth doing before you accept a modelled yield.
- 1Ask what irradiance dataset and what system loss percentage were used. If the answer is a number with no source behind it, the yield figure has no source behind it either.
- 2Ask whether object shading was surveyed and what it was measured as. A quote that models the horizon but not the chimney is modelling a different roof.
- 3Divide the quoted annual kWh by the array size in kWp. For a south-facing British roof the result should land somewhere near 900 to 1,000. Materially above that needs explaining.
That last ratio is the quickest single test, and it is also the first thing to look at when reading a solar quote for signs the system has been oversized.
If you want the arithmetic run on your own roof and consumption, our calculator does it and reports a range with the lower bound stated rather than one flattering figure.
Yield, loss breakdown and monthly figures modelled with PVGIS 5.2 (European Commission Joint Research Centre), PVGIS-SARAH2 radiation database, 2005–2020 average. 4 kWp crystalline silicon, roof-mounted, 14% system loss, Birmingham, UK. Retrieved 2026-09-12. Degradation figures from Jordan et al., Compendium of photovoltaic degradation rates, Progress in Photovoltaics 24(7), 2016. Source
4 kWp receiving 1193.1 kWh/m² a year, which is 4,772 kWh of energy on the plane of the array. The terms below multiply rather than add.
- Reflection at oblique angles−3.2% · −153 kWh
Light bouncing off the glass rather than entering it. Running total 4,620 kWh.
- Spectral response+1.8% · +83 kWh
A gain: silicon suits a cloudy sky better than its reference spectrum. Running total 4,703 kWh.
- Temperature and low irradiance−7% · −329 kWh
Cells above 25°C, and poor conversion in dim light. Running total 4,374 kWh.
- Cabling, mismatch, inverter, soiling−14% · −612 kWh
PVGIS's default system loss assumption. Running total 3,761 kWh.
-21.2% total, and a performance ratio of 78.8%. Note that one of the four terms runs the other way: the spectrum gives output back.
PVGIS 5.2 (European Commission JRC), PVGIS-SARAH2, 2005–2020. 4 kWp, roof-mounted, 14% system loss, Birmingham, UK.