How solar production forecasts work

"Look how Sun Today!" is what Solar Forecast promises, but under the joke sits a genuinely interesting chain of physics: from a weather model's cloud prediction to the number of kilowatt-hours your inverter will report tomorrow evening. Here is that chain, link by link.

Step 1: How much sunlight reaches the ground?

Weather models predict irradiance — solar power per square metre. The key quantity is GHI (global horizontal irradiance), which combines direct beam sunlight with diffuse light scattered by the sky. On a clear summer day in Belgium GHI peaks around 900 W/m²; under a thick December overcast it can stay below 50 W/m². Cloud cover is by far the biggest and hardest part of the forecast — everything else is deterministic.

Step 2: How much of it hits your panels?

Your panels are not horizontal. The forecast must project the irradiance onto the plane of your roof, which depends on:

• Orientation (azimuth): south-facing catches the most annual energy; east/west spreads it over morning/evening.
• Tilt: ±35° is optimal at Benelux latitudes for annual yield; flatter favours summer, steeper favours winter.
• Sun position: computed exactly from date, time and location — the one part of the chain with zero uncertainty.

Step 3: From sunlight to electricity

A "6.4 kWp" installation produces 6.4 kW only in lab conditions. Real output applies a stack of losses:

Loss	Typical
Temperature (panels lose ±0.3–0.4%/°C above 25 °C)	0–10%
Inverter efficiency	2–4%
Cabling, mismatch, dirt	2–5%
Shading	site-specific

Counter-intuitively, a bright cool spring day often beats a scorching July day: at 60 °C cell temperature a panel gives up more than 10% of its rated power.

Why would I want a forecast at all?

• Shift consumption: run the dishwasher, washing machine or car charger when production peaks — especially valuable with a capacity tariff or dynamic prices.
• Battery strategy: if tomorrow is sunny, let a home battery run empty tonight instead of charging it from the grid.
• Sanity checks: if the forecast said 25 kWh and your inverter logged 6, something is wrong — a tripped string, a very dirty panel, or new shading.

How accurate are these forecasts?

Day-ahead forecasts are typically within 10–20% on clear or fully overcast days, and worst on partly-cloudy days where a single cloud bank's timing decides everything. Aggregated over a week they're remarkably good; hour-by-hour on a cumulus afternoon, treat them as a vibe.

See it live

Solar Forecast puts this chain to work: it fetches today's hourly production forecast from forecast.solar for a multi-string PV setup and draws it as a lightweight chart, right in your browser. Free, no cookies, no account — look how Sun Today.