Weather forecast

How to interpret reliability

October 30, 2019

I am almost sure that you are familiar with this image. It is taken from my favourite weather forecast portal, nut there are many others. I highlighted the reliability of the forecast, that is 50%. You may think: “Oh, not bad, after all”. Indeed, it is not the case: 50% means absolute uncertainty; it is pure case, like playing heads or tails with a coin.
In information theory this is called “entropy”, so the situation of maximum uncertainty occurring when two events have the same frequency; there is no information.
It is not that meteorological models do not work, but they simply return the probability of a forecast on the basis of all the data they have inside, and the calculation algorithm they use.

Th weather forecast, with the reliability estimation highlighted

Consider, that if they stated 30%, this would be a more reliable forecast that 50%: it would in fact mean that, at 70%, the weather would not be what forecasted. But indeed, you will probably never read 30% or 40% reliability. Or, in case this happens, it would be a nonsense. Maximum uncertainty is indeed 50%.
So ? We must still wait a while to know how will be the weather on Sunday: so that, the model could better orient itself. It is evident that the tendency is of bad weather, but it not for sure that it could rain.
Be aware: the same considerations hold also when the forecast says: sun.