Stability of forecast model

bluebaer

Hello,

I am a convinced Windy user since several years, but there is one function that I am still missing: The ability to visualize the stability of a forecast model. This is partially give through the comparison of models, but you can see how far the different simulations inside a model differ. There are other providers (weather online) that provide this option by showing the pressure plots of different simulations at defined locations / cities, where you then can see when the simulations converge and provide a map with iso pressure line of the different solutions. Here you can judge how good the pressure converges over location area, e.g. how good the prediction is.

Best regards Thomas

idefix37

@bluebaer

Waiting for a more sophisticated comparison tool, you can already make a comparison between weather models to see which the most relevant at a weather station location. You can do it for 2 parameters, wind and temperature.
Select Reported wind or Reported temperature and choose a weather station near you location (if possible an official one, like wmo or airport).
Then click on Observation vs forecast

In this example you see that the best model at this location for temperature is Meteoblue and for wind (not visible in the screenshot) is AROME.

bluebaer

Hello Idefix37,

Thanks for bringing this up. I have seen the comparison with one model, but not for several. In principle it is this view, but only extended in the future… at the moment this is looking in the rear mirror…

From my perspective this does not answer how good the different simulation runs converge over time. I would, for a first step, just offer this for the long range models (5 days +). There it is relevant to see at what point in time the solutions start to diverge. In the short range (2-5 days) the comparison of the models with their different resolutions gives a good understanding of the stability of the forecasts….

Best regards Thomas

idefix37

@bluebaer said in Stability of forecast model:

I have seen the comparison with one model, but not for several.

Hi Thomas,
You didn't look very well at the example I showed, because the comparison is with 6 models!
Of course the analysis by the Observation vs forecast feature is in the past.
But I am not convinced that a multi-model graphical analysis can really show on the 5-day time frame which is the most stable model. You are talking about apps that show such graphs, you should show screenshots.
Meteoblue offers multi-model graphs, unfortunately which have recently been reduced to 3 days for free users.
https://www.meteoblue.com/en/weather/forecast/multimodel/berlin_germany_2950159

In this example I would not be able to say which is the most relevant model and from when the forecast is more consistent for all models.

bluebaer

Hi,
I saw the comparison of the 6 models. That is good and it is a feature that I had not yet realized.

For sure here is an example from wetteronline:

For the location point analysis. You can see that temperature is going wild after 22.8., so what ever the solution is, it has a high margin of error….

https://www.wetteronline.de/?gid=16522&pcid=pc_modell_expert&pid=p_modell_expert&sid=Ensembles

Or what they have is following:

https://www.wetteronline.de/?gid=16522&pcid=pc_modell_expert&pid=p_modell_expert&sid=Ensembles

If you are located in the area west of the alp the solutions of the models does not converge every well at 42 hours.

And they take it a step further by looking inside a model. Public available is GFS, behind the paywall there is also ECMWF. Here GFS at 120h in goes wild on the iso pressures …

https://www.wetteronline.de/?gid=16522&pcid=pc_modell_expert&pid=p_modell_expert&sid=Ensembles

idefix37

@bluebaer
Thanks for the screenshots. They are almost all from ensemble models, with 20+ members, Windy doesn't seem to be willing to add such models.
Only one of these graphs is a multimodel which could be an idea for Windy adding the different updates to get a better assessment of the forecast stability.
I would rather see a point forecast graph with different models and updates than a map.
And then what parameter would be useful? Pressure is just one.

bluebaer

@idefix37
I am with you on the comparisons of models on the map. The map of a specified model with the variance of the different simulation runs is very specific. This only makes sense form my perspective for the long range models on the iso pressures in order to understand how fare the solutions diverge.

The comparison for a specific location is easier to understand.

What do I look at: This depends on what your after.

Pressure is the crucial parameter generally for stability.

If I’m just looking on the weather for landsports / vacation participation & temperature is a good indicator what variance to expect.

For sailing I also look on the wind.

Surface pressure & goepotential I don’t use.

Feel free to come back and discuss.

Thomas