I don't have any experience with this myself, but my first thought is just get a dozen or so sensors (ideally different brands and manual as well) and average them together, then use that as your reference point. Any individual sensor may be off, but on average they should be accurate. Might want to exclude any extreme outliers first though in case you have one or two particularly bad sensors.
It's either that or you acquire a calibration source (may not be applicable depending on the kind of sensor) or pay a lot of money for a lab grade pre-calibrated sensor.