problem with Modis simulator:

[Jean-Louis DUFRESNE]

Dear all,

We have a problem with the Modis simulator, subprogram
modis_simulator.F90. We use Cospv.1.4.1.

After computing the various cloud fractions, the low, mid and high cloud
fraction is set to undef when the cloud fractions is 0 (lines 563 to 565
copied below):
where (Cloud_Fraction_High_Mean == 0) Cloud_Fraction_High_Mean =
R_UNDEF
where (Cloud_Fraction_Mid_Mean == 0) Cloud_Fraction_Mid_Mean =
R_UNDEF
where (Cloud_Fraction_Low_Mean == 0) Cloud_Fraction_Low_Mean =
R_UNDEF

A direct consequence is that when the cloud fraction is actually zero,
the cloud fraction is set to undef. Then when you computed the mean, the
undef values are excluded, i.e. the cases where the fraction is null are
excluded, and therefore the mean cloud fraction is strongly
overestimated. In our models, where the cloud fraction of mid level
clouds is frequently zero, the consequence is that the Modis simulator
shows a high value of the mid level cloud fraction, although its actual
value is very low.

Is our understanding of what is happening correct?
Does other people have the same problem?
Why these tests have been included? If it's because of the surface
pressure (for instance over mountains), it's probably better to make
directly the test with this variable.

Best regards,
Abderrahmane Idelkadi & Jean-Louis Dufresne

--
Jean-Louis DUFRESNE

Lab. de Météo. Dynamique (LMD) et Institut Pierre Simon Laplace (IPSL)
http://www.lmd.jussieu.fr/~jldufres tel : +33 1 44 27 50 14
Campus Jussieu; Tour 45, Aile 45-55, 3e etage, bureau 311
UPMC Boite 99 ; 4 place Jussieu ; 75252 Paris Cedex 05 - France

[dustin...@noaa.gov]

Hi Jean-Louis and Abderrahmane,

Thank you for bringing this to our attention.

The issue you mention is in fact an error, undefined values are being assigned to clear scenes, which should be zero. We haven't had any other modeling centers bring this bug to our attention. However, Levi Silvers at GFDL recently discovered an issue in the MODIS simulator where there was a divide by zero when computing the gridmean diagnostics. Both of these errors have been corrected in the recently released COSP2, but I didn't propagate all of these changes back into COSP1. I'm working on a patch for COSP1 that will correct both of these errors.

On a brighter note, I would like to highlight that both of these errors do not impact any of the MODIS diagnostics requested for CFMIP3/CMIP6 (http://clipc-services.ceda.ac.uk/dreq/mipVars.html).

Cheers,

Dustin Swales