Changes in water vapor can cause changes in the derived GPS-RO temperature retrieval even when there are no changes in temperature. This effect needs to be quantified to give confidence in its usefulness over time (potentially many decades into the future) and to help determine how low in the atmosphere the temperature retrievals are accurate. Using reanalyses of the recent past and model output to 2100, GPS RO bending angles can be calculated. The GPS RO temperature retrievals can then be calculated using those bending angles. The difference between the derived and the model temperatures would then demonstrate the potential reliability of GPS RO for long time series applications and would also show how low in the atmosphere the retrievals are accurate with regard to potential changes in water vapor. Similar analysis using potential changes in the ionosphere must be made as well to show that, when based on GPS RO temperature retrievals, a potential future change in the ionosphere won't be falsely construed as a change in temperature of the stratosphere.
Response from Kevin Trenberth:
Indeed the potential contamination of lack of knowledge of water vapor can be quantified. An estimate of water vapor from climatology would be one estimate, but even use of estimates from NWP that take advantage of microwave water vapor channels should produce much more reliable results and keep water vapor uncertainties small from a certain range (my guess is above 6 km). We should not think of GPSRO alone but in combination with other (cruder) estimates.