I would probably try a low value chip resistor epoxied to a thermistor. The resistor would be in series with the center lead of the coax. With a little thermal isolation. I don't think a resistor in series in the RF circuit would be significantly more lossy than current transformers or other methods of measuring RF current. Maybe even better than other methods.?
The advantage of the chip resistor would be frequency range and true RMS readings.
You could put a second resistor/thermistor that you control a D.C. current into and make a bridge if you wanted to be really accurate. Otherwise just let a microprocessor do a curve fit on a single resistor/thermistor against another ambient temperature reading.
Interesting about the indirect temperature measurement ideas. Here is a link from a quick search. I'm sure there are other, maybe better, choices with a little more searching. There are far infrared sensors with digital interfaces (I2C), too.
The wavelengths used for non-contact temperature measurement seem to be around the?8 ?m to 14 ?m range.
