Re: How do you measure the accuracy of weather prediction?
- From: "Ray Koopman" <koopman@xxxxxx>
- Date: 21 Mar 2006 00:06:00 -0800
mpcohen wrote:
Suppose the weatherman says that there is a 30% chance of rain. How do
you measure the accuracy of the prediction? For simplicity sake,
suppose that the weather does not vary by season or from year to year.
If, for example, the overall chance of rain is 10% and on 30% of the
days that the weatherman says that it will rain with a chance of 30%
it actually does rain, that would seem to be fairly good. On the other
hand, if overall it rains 30% of the time and the weatherman says every
day that there is a 30% chance of rain, that would not be very good
predicting. How can we make these ratings quantitative?
Assume you have paired data (p,r) for n days, where p is the asserted
probability of precipitation and r is an indicator of whether there
was precip on that day (1 = yes, 0 = no). Let a = sum r/n; i.e., the
overall proportion of rainy days. Then 1 - sum(p-r)^2/sum(a-r)^2 is a
correlation-like proportionate-reduction-in-error statistic, that says
how much more accurate he was than he would have been if he had just
guessed the overall rate.
.
- References:
- How do you measure the accuracy of weather prediction?
- From: mpcohen
- How do you measure the accuracy of weather prediction?
- Prev by Date: Re: Calculate the Number of Subsets
- Next by Date: Re: Calculus XOR Probability
- Previous by thread: Re: How do you measure the accuracy of weather prediction?
- Next by thread: proof sought
- Index(es):
Relevant Pages
|
