I don't quite get why you're thinking about measuring the returned frequency. Unless you're getting a bounce off a moving object, sure the frequency you get back will be the same as you sent. What you're looking for is the round-trip time, and I would have thought that the way to time that would be the same as any other 'ping' based sensing.
Have you calculated the effect of the temperature change you're trying to measure to see what sort of timing resolution you need to achieve? Gut feeling is that this sort of method would be difficult to apply over long distances due to attenuation, but the shorter the distance the less effect temperature will have on the round-trip time which would result in poorer resolution on the calculated temperature for a given timing resolution.