I imagine this inaccuracy is because the nearest stations are over four hours away. I also imagine four hours to the nearest station is common for a large portion of the globe. I suspect this difference in forecasted temperature and the actual temperature is increasingly large the further back you go in time.
It seems like there is a lot of noise and dirty data to be so confident about a couple degree increase.
the confidence on global temperature increase is quite high, you can arrive at the same trend that the serious efforts do with a random selection of a couple of hundred stations and without any corrections. I have done this in the past using raw noaa data.
That was sort of my “aha” moment