Some useful datasets that I have curated and quality-controlled:
A 190-station dataset for the US at hourly resolution, spanning May-October from 1981 to 2015. This data is originally from the NCDC and has been extensively quality-controlled. Variables contained are temperature, dewpoint temperature, wet-bulb temperature, and specific humidity, as well as metadata. Github link: https://github.com/cr2630git/finalhourlystationdataset
A 7877-station dataset for the world at subdaily resolution, spanning all months from 1973 to 2016. This data is originally from HadISD and has had only light quality control beyond that implemented by the Hadley Centre. Variables contained are temperature, dewpoint temperature, and wet-bulb temperature, as well as metadata. Files are too large to place on Github (~500 MB), so are available upon request for transfer via Globus.