MET Norway, negative values for Global Tilted Radiation

[Hukkenhei]

The API call can sometimes return a negative value for Global Tilted Radiation (GTI), which should not be possible. Sometimes the values are small (-1.0, -1.6), but other times the values are very significant, like in the below example.

Example:
Lat: 60.5162811940834
Lon: 10.296247765242
Panel Tilt: 45
Panel Azimuth: 35

Timestamp: 2025-05-26 21:00 (UTC) -84 155.9 W/m²

Screenshot:

Link:
https://open-meteo.com/en/docs/metno-api?longitude=10.296247765242&latitude=60.5162811940834&past_days=2&hourly=global_tilted_irradiance&tilt=45&azimuth=35&time_mode=time_interval&start_date=2025-05-26&end_date=2025-05-27

[omcandido]

I observed a similar thing with the direct and diffuse radiations of the KNMI model. open-meteo uses a model to calculate them, and possibly reuses those values to calculate the tilted radiation. I see that for your example, indeed the direct and diffuse radiations also "explode" on May 26.

[patrick-zippenfenig]

Thanks for the report. I will investigate in the next days. In the past months, some solar algorithms have been modified for the satellite radiation API. Maybe those changes negatively effected some models

[patrick-zippenfenig]

The issue was due to rounding and minor floating point differences around sunrise and sunset. For example, MetNo shows 0.6W, while the solar model gives 0.4W. After rounding, MetNo becomes 1W and the model 0W, causing large numbers in the calculation. I’ve now capped the calculation at 5W of extraterrestrial radiation, treating anything below ~3W GHI as diffuse.

An additional upper limit also solves small negative values.