Solar output comes down to one relation: Energy = Array size × Efficiency × Sunlight × Performance ratio. The array and panel efficiency are yours to set; sunlight (irradiation) comes from long-run climate averages for the chosen location; the performance ratio folds in the everyday losses — heat, wiring, inverter conversion, dust — that keep a real system below its theoretical maximum.
A 6 kW residential array in a sunny climate might see 5.5 peak-sun-hours a day. At a typical 80% performance ratio, that's roughly 6 × 5.5 × 0.8 ≈ 26 kWh a day — about 9,600 kWh a year, before accounting for winter/summer swing or shading.
Climate data gives global horizontal irradiation — what a flat surface receives. A panel is neither flat nor pointed at the sun, so the number has to be moved onto its plane, and that step is where estimators go wrong.
It is done in three parts. First the horizontal total is split into beam (straight from the sun) and diffuse (scattered by the sky) using the Erbs correlation on the clearness index — how much of the theoretical top-of-atmosphere sunlight actually reaches the ground. London comes out at roughly half diffuse, which is why British roofs are more forgiving about orientation than Spanish ones. Then the beam is projected onto the tilted plane by a ratio integrated across the whole day. Finally the sky is treated as uniformly bright — the isotropic assumption — so a tilted panel sees (1 + cos β)/2 of it, plus a little reflected off the ground.
This page used to get that badly wrong. It compared the sun’s angle at solar noon only and scaled the result by orientation — using an expression that was identical for every direction from east round to north. A north-facing array came out at 91% of south and above a flat roof, which is impossible: pointing a panel away from the sun cannot beat pointing it at the sky. Integrating across the day fixes it, because a surface facing the wrong way spends most of the day with the sun behind it, and noon is the one moment that hides this.
Isotropic sky is the conservative choice: it under-reads a clear-sky site by a few per cent, because real skies are brighter near the sun. For a tool someone is using to decide whether to spend money, that is the right direction to be wrong in.
Enormously, and more than the old version of this page suggested. Facing the equator at a good tilt, London collects about 19% more than a flat roof. Turn that same array to face the pole and it collects 59% of the south-facing figure — and less than if you had left it flat, because it loses the direct beam and half the sky at once. East or west costs about 18%, which is often worth accepting for a flatter output curve across the day.
On tilt, the familiar rule is “set it to your latitude”. That is too steep. Diffuse light arrives from the whole sky rather than from the sun’s direction, and the more of it there is the flatter you want the panel: the optimum here comes out at 41° for London at 51.5°N and 31° for Phoenix at 33.4°N. Below latitude in every location on the list.
The fraction of a system's theoretical output it actually delivers once every real-world loss is accounted for — heat derating, wiring resistance, inverter conversion, soiling, and small mismatches between panels. A well-installed system typically lands around 75–85%.
This tool uses baked-in long-run climate averages, not live weather, satellite imagery, or site-specific shading. A tree, a chimney, or an unusually cloudy year can shift real output meaningfully above or below the estimate — which is exactly why installers run a proper site survey before quoting.
Mostly through irradiation — how much sunlight a location receives on average — and secondarily through temperature, since panel efficiency drops slightly as they heat up. Two identically sized systems in different climates can have meaningfully different annual yields.
Releases in which this page changed, newest last. Derived from the archived copy of every release, not from notes written afterwards — so it reflects what actually shipped. Site-wide passes are left out; they are in the full changelog.