observation vs. forecast
-
I'd like to know, what timestamp is the reference for the forecasts?
I.e. is it the forecast 24h before observation time, or 12h, or the forecast closest to the time of observation?
To me, it's important to really weigh the fit of the forecast to the actual weather situation.
Couldn't find this information anywhere.
Thanks in advance! -
Is there nobody from the developer team, who can answer this question?
@St-Patrick said in observation vs. forecast:
I'd like to know, what timestamp is the reference for the forecasts?
I.e. is it the forecast 24h before observation time, or 12h, or the forecast closest to the time of observation?
To me, it's important to really weigh the fit of the forecast to the actual weather situation.
Couldn't find this information anywhere.
Thanks in advance! -
@St-Patrick Hi, you can get further info about model info updates here:
-
This post is deleted! -
@Suty this doesn't answer the question asked. So I'll ask it differently. How are the forecast plot curves determined? A single forecast for a specific lead time? Or is a combination of forecasts over multiple lead times?
-
@Bruce-Gunn-2
It is nowhere stated that the forecast curves of the different models are established with a forecast lead time of 12 hours or 24 hours. They are approximately synchronous with the observations.
This makes it possible to detect the bias of the different models compared to reality. -
@idefix37 are you saying that its a combination (median error?) of ALL forecast lead times for each model for the time in question?
-
@Bruce-Gunn-2
No, the forecast curves are plotted based on the forecast wind or temperature in the same way that the weather station curve is plotted every 3 hours.
On the right side of the panel, the median error is displayed for each model. It is calculated over the period of the observation days displayed, which is approximately one week.
See what is the median error:
https://insidelearningmachines.com/median_absolute_error/ -
@idefix37 thanks for the reply.
Still struggling to understand this, but I think you are saying that the model forecasts plotted are ones that were initialised around a week ago (at the time the model plot curve starts).
Correct? -
@Bruce-Gunn-2
No, they come from the latest model update. If the weather station reports 10 knots at 10 a.m. and the model shows 12 knots at the same time and location, these are the values used to plot the curves. These values (reported and forecast) will be used every 3 hours.Right now you can see at this weather station 3 knots and the model predicts 5 knots.
These values are used to plot the curves.
-
This post is deleted! -
@idefix37 thanks for being patient with me.
Am I right to say that the curves show the most recently available model run at the time of the observation?If that's the case then the curves show a comparison with the first several hours of the model runs only. :(
-
@Bruce-Gunn-2
Yes. But this is useful. It shows if the model has a biais at a given location.
As example, users are worrying because the default model don’t provide accurate temperature in their mountain location. The ECMWF has a 9km resolution and so it has a coarse topography model. It is not able to “see” the right elevation of a location and then it provides a wrong temperature forecast. GFS is worst : At Mont Blanc summit (4810m) GFS provides the temperature 2000m lower. These comparison curves show these discrepancies.
Then to know the accuracy of a model prediction 10 days ahead the Observation vs forecast does not provide an answer. -
@idefix37
Ok, I can see that could be useful in the case you describe.But if I want to know which model that on average most accurately forecasts 1, 2, or 3 etc days ahead, that display doesn't tell me that.
Would be worth Windy writing an FAQ on this perhaps.
-
@Bruce-Gunn-2
Which models do you want to compare? With what parameters and where?
