As I introduce T-skin as a control variable, I am re-doing some of the exercise I've done before of looking at each of the main instruments increments and how they compare w/ GSI's. IASI (and CrIS-FSR) has always been in the suspicious list. I want to have a record of this here so we can work together to address whatever issue affects hyperspectral IR.

The temperature increment for microwave instruments such as ATMS (clear sky) and GMI (all sky) are quite consistent between GSI and JEDI. The corresponding T-skin increments are also quite consistent - with still a puzzle over Antartica (but not a show stopper - see e.g., https://github.com/JCSDA-internal/fv3-jedi/pull/1257.

For IASI, the story is quite alarmingly different. For starts, GSI does not fully converge its minimization when given up to 100 iteration to crank; but so be it. JEDI, on the other hand, seems to converge in about 40 something iterations, which is suspicious in itself. But the real issue is how different the increments from both systems look (perhaps not a surprise given the convergence differences - chicken and egg).

Interestingly enough, the increments on specific humidity between the two systems are quite reasonable - here is an example at 850 hPa - GSI (top) JEDI (bottom)

Unfortunately, this is what temperature looks like at the same level:

notice T increments are multiplied by a factor of 2 for plotting enhancement purposes; still -diff in increments are larger than increments themselves.

Now here is what T-skin increments looks like:

It looks as though the weight given to whatever induces the sensitivity of Tb to T-skin is overweighted.