You can become remarkably good at estimating cloud heights at varying layers when you have to do it every 30 minutes throughout your working day. Local knowledge also provides valuable help on days when the clouds are being a bit difficult.
How do they know if they're any good without actually going up to check if their observations are correct?
If they put out a METAR saying cloud is 1500' when it's actually 2000', do any pilots actually radio this info back to help the ground observor improve their judgement?
From my visits to towers on reasonably nice days, they seem to pluck the figure literally from the air, putting SCT 030, when i would've said FEW 040. Obviously it's always best to air on the pessimistic side.