The SOA is subject to other thermal considerations, it isn't a graph of expected temperature at different currents, it to limit the allowed current at different die temperatures.
Calibrating temperature sensors, and to some degree the analog measuring circuits is quite easy after the 'raw' voltage has been converted to numbers. The method I use is 'two point compensation', basically you use four values, one pair is the theoretical value and measured value at one end of the temperature range, the other pair is the same at the other end of the range. By range I mean the upper and lower temperatures you are expecting rather than the limits in the data sheet.
The difference between theoretical and actual values give you the absolute error. You get an error value for each end of the range.
The difference between error values gives you the gain error.
Then for each measurement you take, multiply it by the gain error and subtract the absolute error. Fairly simple math.
Brian.