Hi Stephane (and others),
Please, feel free to complain -- the more dialog the better as far as I'm
concerned. Not comfortable when the group gets too quiet
Let me expand
on our "agency" goals a bit, which will perhaps clear up our perspective.
In our perfect world, we'd have a consolidated IOP algorithm to be put into
operational satellite processing, accompanied by a rigorous uncertainty
budget for this algorithm. Currently, a number of approaches exist, but
when you "tear them apart", most are very similar -- differing only by
basis vector parameterization (S, aph*, etc.) or inversion technique. We
have no interest in an algorithm "shoot-out" -- rather, we'd like to get
a group together (yourselves) to study these parameterizations/inversions
to determine their sensitivities in an operational satellite processing
environment. We know each algorithm has been verified using in situ
data (specific manuscripts and the IOCCG report) and some satellite data --
but to our knowledge, most have yet to be rigorously vetted in the
satellite environment (e.g., how/where does the alg perform in l2 space
versus l3 space, does spectral resolution matter (we lose a green band in
VIIRS), what happens when input Lwn are imperfect?) -- nor, has satellite
inversion failure remediation been robustly explored. In other words, we'd
like to better understand why some algs "blow up" when globally applied, and
to determine if/how such events can be avoided. All of you have worked
on this in some capacity, we just want to "get the band back together".
With regards to implementation in msl12 -- well, we need the algs to work
in this processing environment -- otherwise, no operational satellite
products from NASA, right? Our goal is to ensure that msl12 faithfully
replicates the original implementation, however, differences will arise
(we'd like to know where/why) and we'll have to iterate with each PI.
Ultimately, we hope to build a version of msl12 with a generic IOP
algorithm that allows command-line parameterization of S, aph*, etc. and the
inversion method. Not there yet, but it's coming. This will take some
work and much thought as to implementation. As you mention, changing the
inversion (for example) changes the answer -- which is another reason why
we're asking the questions we're asking (why? do we have strong
recommendations as a community? etc.)
I know I didn't answer all of your questions/concerns. We know we'll need
a series of statistical metrics to evaluate the "infinite" series of
approaches, but this is a work in progress and suggestions are welcome.
Again, we don't want to foster an environment of algorithm competition --
one person's approach versus another's -- but, rather work towards
better understanding of how/where everything works (or doesn't) when applied to
satellite Lwn.
Keep the dialog coming!