@idefix37 Do you mean, for each hour, when an observation was made, they put the most actual forecast for that hour from the whether models? To me it would make more sense when they take the forecast i. e. 24h before observation. Otherwise the quality of prediction can't be weighed very good, because only a few hours before observation, they all merge very close to the actual whether conditions. 1 or 2 days before, the models often have quite big differences.

I definitely would prefer a comparison with a forecast at least 12h before observation time. I don't need a forecast for the now time. Then I can watch out of my window as well. To measure the quality of a prediction, it should be a real prediction for at least 12h, better 24h. Best option would be, if I can decide this in my personal settings. The actual mode of comparing the closest prediction to observations almost taking place at the same time is meaningless. Please take this into account when discussing new features with the developer team.