You are hung up on the calibrations needed to turn a series of point-location sensor measurements into an estimate of an underlying quantity (temperature in a 5

You are hung up on the calibrations needed to turn a series of point-location sensor measurements into an estimate of an underlying quantity (temperature in a 5x5 degree box over time). This kind of calibration is purely routine in any measurement problem, as I tried to indicate (and as the papers that I linked to demonstrate). Any time you have a lab measurement, you have to try to calibrate it. You learn more about sources of bias over time, and you re-do the calibration, and regenerate the temperature estimate. It seems like you haven't had a chance to work on a long-term measurement system consisting of real-world hardware, and you're proceeding on some kind of common-sense idea that the point measurement is an absolute read on the underlying quantity of interest?