"Average annual global temperature in a given year" and "temperature in my city tomorrow" are fundamentally very different types of predictions.

Often it is easier to accurately forecast gross dynamics on a long time frame than it is to forecast precise dynamics the exact same process over a short time frame.

You don't even need to understand the math or physics to see why this is intuitively true.

Consider e.g. predicting minutiae about the behavior of a fetus over the next week ("how many fist clenches", "how many kicks") vs. predicting which week the baby will be born -- the latter is substantially easier than the former despite the longer time frame.

Or, more to the point, consider forecasting the position of a particular cloud of molecules in a pot of water being bought to boil vs forecasting the temperature of the water in the pot in 5 minutes. The latter is hilariously trivial -- a small child can be taught how to do this with excellent accuracy. The former is some horrendously difficult phd level fluid mechanics and even then hard/impossible.

In some sense, an educated intuition is exactly the opposite of yours -- it'd be surprising if we were this good at extremely fine-grained weather prediction but couldn't guess the annualized average temperature of the entire system in 50 years. The latter is a much simpler statistic because the timescales and physical scales take a lot of the difficult stochasticity out of the forecasting problem.