Primarily for precipitation where models can’t accurately predict the location of showers and the graphs are very sensitive to pin location, perhaps something like the following would be useful:
define a radius (eg. 10km) around the pin location
if >1 forecast data nodes are within this radius, use their mean (possibly with inverse distance weighting)
otherwise use the current interpolation method at the pin
It’s an interesting concept and one I’ve thought about many times. I always come back to the same conclusion “why would you blur (average) best data you have”?
This concept is very similar to Probability of Precipitation (PoP6 and PoP12) - check it out on wikipedia.
The other thought I have is that I’m not in the business of tinkering with forecasts. I just show forecasts in their original released values. Tinkering with data can also led to support problems - e.g., “The average precipitation is not accurate. It says it’s supposed to rain but it’s not.”
In any case, there is a better approach, which is to use ensemble models.
I think I am more interested in the expected amount of precipitation (QPF) per hour than the probability of any precipitation (PoP).
Ensemble models sound like a good solution, but are any available? Arome & Icon-D2 are not? Presumably an amalgamated ensemble would give a blurred precipitation forecast with no showers evident? It might be good for the graph, but less so for the map.
You can look at the forecast graph and it may look dry for the next few days, but unless you scroll through time on the map you don’t know how likely precipitation is. ie. the reason to blur the data is to give this information at a glance and make the graph less misleading, when a better solution (ensemble) may not be available. Another option might be to have multiple pins and multiple graphs?
I understand the issue with support enquiries and explaining what the options mean, and having too many confusing options. It is a difficult balance to make but I think if the information is not otherwise available it may be justified?
There are many ensemble models available but are generally lower (half) resolution. I have a few ideas on how to present the data in useful ways. I certainly favor using ensemble models over averaging normal models over space and time.
It’s not really a “difficult balance” - I have very little time so I steer clear of anything that may add more support requirements.
Thanks, I think it would be very useful and interesting. Could you ‘simply’ make ensembles available in the list of providers? I hope you find time at some point…