Thermal Imaging - How Not to Measure Temperature

Thermal imaging cameras offer amazing perfomance - The one used here is accurate to +-2K and will discriminate temperatrue differences of as low as 0.08K.

However it seems that not many people realise they have limitations:
You cannot just point one at a subject and say what the temperature is. In some cases it is not possible to even guess the temperature of the object (reflective surface).

Here are some indications of what can go wrong.
A copper heatsink 3mm thick with various surface finishes is used to show the pitfalls:

One end of the matt tinplated copper block was polished until pure copper was exposed
The centre of the block was polished untill the tin plate was still present.
The plate was then spray painted down one half with matt grey paint.

First a video showing reflection from the unpainted side of a hot object moving
A still from the video
Now a video of the WHOLE of the plate heating :

Is the temperature of the block 23 26 or 41C?

 Note that the painted area shows insignificant change to the temperature whereas the unpainted side shows a reflection causing the reported temperature to change by approximately 20C

Next observe the plate being heated by electronics attached to the other side.
You will see the painted area slowly heating whils the bare metal changes very little: 

Finally Just to prove that the shiny side has not been masked Plate is hot and shiny side reflects hot object

How good is the grey undercoat at normalising the emissivity These 2 plots show how the temperature changes along the line LI01 placed first in the unpainted area and then in the painted area.:

With no paint this is the response of the plate

Painted - not perfect but a lot better.

So just what is the temperature of the plate? The answer is I do not know - it is approximately the temperature shown on the grey painted area... but since I have not calibrated the emissivity of the paint I do not know! And since I have not measured the humidity I do not know. And since I did not measure the distance from sensor to object I do not know. And since I did not measure the air temperature I do not know.

How about outside - NOTE these are not calibrated images no emissivity/atmospheric corrections applied.
Time taken 2012-11-14 17:45  Ambient temperature 7.0C Camera range -40 to 120C hence some of the temperatures measured are outside the camera range (-45C seems to be the saturation level of measurement). Humidity ? high

So it seems that clear sky no sun has a "temperature" of about -44C and seems to decrease linearly with angle above the horizon.
Take a picture of a MMT thermometer and claim that the temperature of the case is X is wrong unless you have eliminated reflections and calibrated the emissivity.
Take temperatures of houses and you will show heat leaks and hot spots but you cannot claim that the temperature has much accuracy (it will be more accurate than the shiny MMT surface)

What is interesting is of course that you can take thermal images through the atmosphere. This proves there is little emissions in the thermal IR band from resident gases.

The IR spectrum (from FLIR documentation):
 FLIR S45 camera Spectral Range 7.5 to 13um

Germanium used for lens has this transmission vs wavelength property: