PV forecasts in fog: why weather models get winter wrong
· VictronLab
A battery controller is only as good as its PV forecast. In valleys with winter fog all the major weather models are regularly wrong. We measured by how much, and what helps.
The test
We compared the archived forecasts of four weather models (ECMWF IFS, ICON-EU, GFS and Météo-France) for the winter of November 2025 to February 2026 with the actual daily yield of a 26 kWp system in a Slovak valley.
The result: no model sees fog
All models forecast sun on 16 to 20 winter days when the valley lay in fog: around 22 kWh forecast on average, 3 kWh actually produced. The mean daily error of every model was about half of the mean yield.
ECMWF was the most accurate, the others were close behind. The model values for low cloud and humidity did not help either: if a model does not foresee the fog, it is not in its cloud data either.
What helps: your own system as a sensor
Winter fog often lasts several days, as long as an inversion persists. That makes the ratio of yesterday's measured yield to yesterday's forecast a good hint for today. Blending 70 % of this ratio into the model forecast cuts the winter forecast error by 29 %.
Gridzilla uses the system itself as a sensor in exactly this way and corrects the forecast continuously with the measured values. On top of that, an AI advisor assesses weather situations like fog and decides how much reserve the battery should hold.
What this means for battery control
If you charge the battery from the grid at night in winter, do not take the next day's PV forecast on trust. A wrong sun forecast means the battery goes into a foggy day empty. The better strategy: plan cautiously and correct the forecast with your own measurements.
- No global weather model detects valley fog reliably.
- ECMWF was the most accurate in the test.
- Using your own system as a sensor improves the winter forecast by around 30 %.