Wow, this discussion really took off. This reply is a little late so go back to the bottom of the first page for context.
I've plotted the test data from JCA433 (attached). The upper plot is all three temperature columns and the bottom one is the last two columns relative to the first column.
It is fairly obvious in the bottom plot that the nearby FAWN data is not well correlated with the two sets of shield data. I'm guessing that the FAWN station is too far away to be meaningful for a test like this and/or sited differently.
Taking kcidwx's comments on calibration a little further: The data from the two shields is showing a difference of roughly 0.5F to 1.0F. If the sensors in question are calibrated to +-0.2F accuracy (not impossible but a tall order) then consider this. If the true temperature is 70F, one sensor could read 69.8F and the other could read 70.2F and they would both be within the +-0.2F spec limit. However, comparing the readings, they are 0.4F apart. So, if they are reading different by 0.5F then that is just barely enough to be a significant difference in light of the calibrated accuracy.
I am guessing that the sensors used by JCA433 are the Ambient F007TH units (right?). If so, they are specified by Ambient to be accurate to +-1.0F. I don't know if Ambient claims NIST traceability or over what time period the calibration would be valid (one year is typical). Anyway, my point is that these sensors would need to be further calibrated for the data from JCA433 to be meaningful. He mentioned having calibrated the sensors but now I am curious as to how they were calibrated.