Best posts made by meteoblue

according to our operational verification using over 40 000 stations. Our meteoblue AI still has an error that is about half as large of any other existing model for the variables of Temperature, wind and humidity, which are the ones we can verify easily with stations.This is also true for the US.

it is clear that due to different topographies, resolutions used in the models you can easily find points were models do not agree on wind directions, and that is a good thing…. in less than a minute I e.g. found these:

obviously due to these differences you will find locations where one or the other model has better wind direction, that is why we use over 30 different models (including our own weather forecast models) and all show them on meteoblue in our large model comparison:

you can check those for any location here:

For wind direction we do not yet make an AI forecast (almost no demand) and most measurements (especially) amateur stations cannot be used for this due to wrong placement, (not 10m altitude and many local obstacles close to the stations). But we will probably also make a consensus forecast of wind direction in the future.

@idefix37
The meteoblue AI produces a consensus forecast. That is the forecast you see on our main page, or in every product seen on our page that only shows one forecast. The model average shown in the multimodel is just the average, which is most of the time very different from the AI consensus. Windy shows the AI consensus when our forecast API is used. I believe this is in the timeline local forecasts of windy, but definetly not on the maps and likely also not when you pick data from the maps by mouseclick (but not sure about this last point). The AI takes into consideration many very local factors at very high resolution, so that we do not produce a gridded AI forecast that could be plotted on a map.