> i kept & extended scat_dataCheck and do the checks for NaN, neg. values
> and Z norm in there.
> cloudbox_checkedCalc calls scat_dataCheck, but has the option to skip it
> completely or to skip the Z norm check therein (oops, haven't updated the
> doc yet. is coming soon.)
>

doc done now, too.

note: the update of cloudbox_checkedCalc might make some of your ARTS scatt
calc setups to crash due to the introduced default check on the scattering
matrix normalization. in this case, either set sca_mat_threshold to a
higher value (default=1e-2) or skip the normalization test by setting
scat_data_check to something else but 'none' and 'all' (generally, we
discourage switching off the test, though).

wishes,
Jana


ps.

> Regarding normalisation, how big difference is there between
>>>>> quadrature rules? 1%, 10% or 100%? Seems reasonable to at least
>>>>> check that normalisation is OK inside a factor of 2. (With an
>>>>> option to deactive this, if you use a solver anyhow checking
>>>>> this.)
>>>>>
>>>>
it not trivial to separate (pure) quadrature issues from grid density (when
grid density is high enough, trapezoidal integration is fine...). in ARTS
we hadn't used anything but trapezoidal integration before (re-)adding
interfaces to DISORT and RT4 and i actually didn't go that far to check how
good their quad methods perform on our standard data with its equidistant
angle grids.

however, what i have seen is that scattering matrix norm deviated by up to
10% from value expected from the ext-abs difference. generally, this tended
to be worse for larger particles, if i remember correctly (which is
straight forward as larger particles exhibit a narrower & stronger forward
peak, which is hard to catch on equidistant grids).

btw, the normalization threshold used by scat_dataCheck (and now also
cloudbox_checkedCalc) is not directly an absolute or relative deviation of
the scatt matrix norm. instead, we use the deviation rescaled by the
extinction, which is effectively the absolute deviation in the scattering
albedo.
This was chosen in order to avoid unnecessarily high demands on
low-scattering particles (when scattering is low, numerical issues might
occur and trigger a relative-norm-deviation threshold, while the scattering
contribution from these particles is low. an absolute-norm-deviation
threshold is hard to determine as its relevance strongly depends on the
total extinction).

i've now set the norm threshold default to 1%.
a part of the single scattering data in arts-xml-data actually does not
pass the check with this threshold. as far as i've tested them, they all
pass with threshold=5%.
i'm not settled yet, whether i just will increase the default threshold. or
whether it's more appropriate to post-process/modify the data (this goes
back to the quadrature/grid density problem. the single scatt data for each
individual angle gridpoint is very likely correct.).




-- 
=====================================================================
Jana Mendrok, Ph.D. (Project Assistent)
Chalmers University of Technology
Earth and Space Sciences
SE-412 96 Gothenburg, Sweden

Phone : +46 (0)31 772 1883
=====================================================================
_______________________________________________
arts_dev.mi mailing list
[email protected]
https://mailman.rrz.uni-hamburg.de/mailman/listinfo/arts_dev.mi

Reply via email to