But, I have a subtler question about it. Essentially, “what’s the unit of analysis?”
When I put my zip-code into a weather site, and it tells me “40% chance of rain”, that seems to match the prediction for – in my case – “Baltimore City”.
I can’t really reconcile the fact that there’s a 40% chance of rain in Baltimore City, and still a 40% chance of rain in my fraction of it.
Similarly, when you look at at forecast broken down hourly. . .it might say
1-2 40%
2-3 30%
3-4 30%
But if you look at the forecast for the afternoon. . .it’s maybe 40%. You follow? If there’s a 40% chance of rain between 1-2. A 30% chance between 2-3. And a 30% chance between 3-4, there’s a much larger chance of rain between 1-4 than 40%. Specifically, about a 70% chance of rain ((1-.6)(1-.7)(1-.7)).
I guess the real question is this: is predicting rain fall on a local level, and over short periods of time complete bullshit?
If so, what’s the real geographic area, and time-span of the predictions?
As I understand it, it’s calculated using the following witchcraft and voodoo.
They expect the conditions of the air (temp, humidity, breeze, particulate matter, magical forces) to be XXX at AA:00 (AM/PM). Based on past observations, when the conditions to be present, have existed in the past, it precipitated YY% of the time.
It’s the only job that you can be wrong more than 50% of the time, and not get fired. (Public sector workers excluded, there, the threshold is nearly 99%)
You can only multiply probabilities like that if the events are independent, i.e., event A has no connection with event B. However, if it’s raining at some time in a city, then it’s very much more likely to be raining an hour later in the same city, and similarly if it’s not raining there.
I may be missing your main point, but the weather in this part of the country the past month or two has been dry. There have been plenty of fronts and systems moving through this area from the rain-soaked Plains but by the time they get here they may or may not produce any real rain.
Percentages are quoted for the various regions of the Midstate and for differing time frames. Rarely a day goes by when some portions of the Midstate will see rain, sometimes quite a bit, while others remain rain-free.
I choose to regard the percentages as nothing more than guesses that “somewhere in this area, at some part of the forecast period, there will be some rain.” Hitting the time and the place together is as random as it can be. Even with air mass movements and temperatures, pressure gradations, moisture content, dew point, etc., the microgeography of counties, towns and neighborhoods makes any greater precision impossible.
Bullshit is a bit heavy. But it’s a good substitute on any accuracy scale.