And third, we also have some other data sources that help. In certain parts of the world, we've got radars. I believe at least some of the better models these days are using that data... there's plenty of wind data to be had from velocities (though it would seem challenging to use/analyze it given things like range-folded velocities and the ambiguity of tangential winds see the intro to [this article]).

Plus, I believe at least the ECMWF has had real success in ingesting ACARS... atmospheric data that are generally reported by all commercial aircraft.

But... all that said, your point still has some real potency. Especially when we note that, when it comes to our best model, the ECMWF... it still only runs twice a day. Maybe other reasons drive the decision, but you'd think they still weigh how improved such forecasts are. So it seems fairly likely that they've determined that long-range forecasts don't improve much when they are still based heavily upon that older information.

But then, I think everyone would agree that having hourly models available like the HRRR is very useful in important nowcasting situations, such as severe weather awareness. So maybe the overlooked small data changes are (counterintuitively?) more important to long-range forecasting than short-term. Definitely an interesting question you've raised!