Do you have time series temperature data at multiple heights? An interpolation of the high frequency time series by elevation (by determining a temperature lapse rate between model heights) and then binning by month should work. If you only have monthly data, perhaps doing the same if you have access to monthly temperature at multiple model heights. Lastly, if only one model height is available, perhaps use the standard atmospheric lapse rate as the most simple (but highest uncertainty) method: -6.5 K/km.