Working group number: 9
Coordination: Carlos Jimenez
Topic: UTH retrievals from AMSU data

AMSU data can be used to derive a long, continuous and global dataset
of upper tropospheric humidity, but so far these data has only been
marginally used for a direct retrieval of UTH. Bremen and Chalmers are
currently working on exploiting the data by different approaches.

Three of them were discussed in the working group, due to the
participants involved, a fourth one was reported the day before as a
presentation. The approaches discussed can be summarised as:

- simple linear regression with 1 AMSU-B channel. 

- more complicated neural networks regression with AMSU-A and AMSU-B
  channels even more

- complicated multi-physical linear regression with AMSU-A and AMSU-B
  channels

Some retrieval precision figures were reported, with numbers between 3
to 9 % in relative humidity. But the comparison was difficult. First,
different humidity parameters were retrieved, namely weighting
function averaged relative humidity and layer averaged relative
humidity, depending on the approaches. Second, different datasets were
used for the regressions.

Some issues were discussed during the session. For instance:

- what training dataset is better? Simulated radiances or co-located
  measured radiances with radiosondes? There was agreement here for
  the later option.

- how good are our present simulated datasets? How can we improve
  them? Including surface emissivity cases? Clouds?

- why present regression seems to over-estimate low UTH values and
  under-estimate high UTH values? This was still and opened question.

- How can we test a study case common for all regressions so we get a
  good grip on the relative performance form all methods?

At the end some action were planned for the near future, namely:

- submitting three articles describing the three present approaches
  with the present datasets.

- testing all regressions with an ECMWF simulated database to evaluate
  relative performance between approaches.

and more like a long term objective:

- to set a project to build an AMSU synthetic dataset addressing more
  realistic simulations, including scattering and liquid absorption
  from clouds and surface emissivity changes.



Written by Carlos Jiminez.