I honestly appreciate the question. 6 hours from now, the prog or TAF (and similarly wind aloft, etc) may indicate a forecast of a wide area of scattered or widespread precipitation. Sometimes the coverage, intensity, and confidence are very high. You see this in more stable-type weather - large areas of rain, snow, etc. which is broadly easier to predict. For unstable, convective weather, common to the US in the every afternoon from May to September, it’s not as specifically predictable until it’s closer to the forecast period.
For example, if you look at a 6 or 12 hour forecast, you’ll notice it’s giving that prediction in a wide time range - often a 2-6 hour block of probability (maybe a 30-40% chance). Often times this is because the precipitation has not yet developed and taken its course with regard to diurnal variation, specifically what happens to the atmosphere when the sun comes up or goes down. During the day there’s more vertical mixing, surface heating, etc. So forecasters know where the stuff might occur that far out, on what’s called the synoptic scale, but they don’t know exactly where or when on the smaller meso or storm scales. Thus, the 6-hour forecast might indicate strong storms over a 1,000sq mile area sometime between, say, 5 and 9PM. That’s a really broad area (and trust me, it gets even broader than that).
But those storms have not yet initiated and that’s the key. Convection is predictable several hours (even days) out in a broad sense, but those predictions are not accurate enough to say “I think we can make Airport x” until the storms actually fire up. Once that happens, usually in the mid to late afternoon, then it becomes a lot more predictable. A tornado watch may not go up until a hour or two prior to initiation of anyof the associated storm cells. Like, during clear skies, maybe an agitated altocumulus field.
Now, once the storms initiate and stabilize, larger, commercial aircraft may either delay or pick their way around that stuff, but it’s a timing game. The forecasts get way more accurate the closer to the time period. A forecast that indicates a period 1-2 hours away is almost treated as Gospel. For instance, during last night’s stream we watched several flights funnel through a couple gaps in the line between Omaha and Kansas City, one big gap near Oklahoma City. Sometimes those gaps are very narrow and hundreds of planes will fly through it (with careful coordination of ATC), otherwise you continue to delay or take the really long way around the entire storm system, potentially hundreds of miles out of the way. We watched a flight do that last night - from Guadalajara to Chicago - flew all the way over West Texas to stay on the back side of a major squall line.
For a different perspective, small aircraft will generally avoid flying near that stuff altogether, however, about 30-60 minutes after the line passes, watch all the planes launch to go do training, fly home, etc. I do this frequently.
Embedded thunderstorms are even more insidious - you have to know where they are and again, they often don’t build up until mid to late day. The only effective way to see these is radar (or lightning detection), maybe a stray METAR will catch one, but generally these are not something you want to mess with unless you have a really good picture.
Using just a 6 or 12 hour forecast model would pretty much ground all that activity because it would generate way too wide and non-specific of an area of weather. But then, say we do launch, and there is weather out there, we need the specifics of where that is - usually by use of radar, often supplemented by METAR, either from a data/network link or with the help of ATC. Of course, this is with the caveat that there is a delay when using data versus, say, an onboard radar. The forecast will not provide those specifics - we need, again, radar, METAR, PIREP, satellite, etc.
The main takeaway is that storm systems aren’t this congruous thing that travels from one side of the world to another and maintains its shape the entire way. They’re constantly changing shape, intensifying, diminishing, re-developing, skipping, jumping, in broadly, but not specifically, predictable ways until the forecast window gets closer. And that is made possible by observations and computer-model-driven calculations of diurnal changes and interactions with other weather. It’s generally a giant mess that has had tremendous improvement in accuracy over the last century, but it’s not and may never be accurate enough in most cases to say, at 6:00AM, “there will be a line of thunderstorms passing over O’Hare airport at 12:00PM (or 6:00PM).”
Thus, when it’s about launch time, we go from solely forecast-driven to observational-driven (and still looking at forecasts if it’s a longer flight).
For the sake of education and earnest learning, later today or tomorrow I will post some forecasts and we can see what actually happens 6 and 12 hours later, and compare notes. The big caveat to all of this is I live in a country that has a ton of microclimates, and one big party in the middle where it all comes together, so as pilots we have to pay attention to how this works.
tl;dr summary to your question - yes, the farther out the forecast, the more it’s making it up.
I just finished writing a treatise on why it is bad! 