Page 1 of 1
Posted: Mon Jan 13, 2020 2:36 pm America/New_York
There seems to be something off in the adg_443_giop data, when going from 2019329 to 2019337. It's a global loss of value, and stays that way in the next available 8day periods (up through 353). The offsets do not appear in the VIIRS version of the same. Thoughts?
Posted: Mon Jan 13, 2020 4:54 pm America/New_York
I just computed the difference in the adg_443 values from these two files.A20193292019336.L3m_8D_IOP_adg_443_giop_4km.ncA20193372019344.L3m_8D_IOP_adg_443_giop_4km.nc
I got a global field that looks like the attached image where red regions have higher
adg in the first 8-day period and blue regions have higher adg in the second.
Can you be more specific about which files you are reading and about what you mean
by global loss of value?
Posted: Mon Jan 13, 2020 5:06 pm America/New_York
Um, strike that image. I just realized I didn't leave out the "no data" areas.
I'll have another look at this tomorrow, but if you could provide any additional
details in the mean time, that might help make sure I am looking in the
Posted: Mon Jan 13, 2020 5:50 pm America/New_York
The only difference in the files was that I started with the 9km data. I was about to upload gifs on this, when realized it's even visible in the thumbnails on the level-3 browser (before = Nov 5 - Dec 2; after = Dec 3 - Dec 10 images).
Thanks for looking into it. Robert
Posted: Tue Jan 14, 2020 10:14 am America/New_York
I was a bit more careful this time, and I do see what you are seeing.
The following three images show differences between subsequent
8-day 9km composites of adg_443_giop. Red indicates that adg is
greater in the minuend, while blue shows areas where adg is greater
in the subtrahend. White reflects no change.
The first two are what one might expect, with some regions decreasing
and some increasing from one eight-day period to the next. The third
difference image indicates that most of the world decreased in adg
between the Nov. 25-Dec.2 composite and the Dec.3-Dec.10 composite.
This does seem suspicious. I will put this before the local brain trust
to see if some processing parameter may have changed.
Posted: Tue Jan 14, 2020 1:38 pm America/New_York
I've been having some hallway conversations with folks who
know a lot more about how our sausage is made than I do.
The general thinking is that MODIS instrument calibration
is a never-ending process as the aging spacecraft is exposed
to the ever changing space environment. As one gets closer
to real time, one moves more and more from interpolation
of instrument characterization data to extrapolation of how
we guess the instrument is currently changing.
If you check the metadata of the files we have been discussing,
you will find that they all have a processing_version that ends
with "QL" which stands for quick-look.
for f in *.nc;do echo -n $f;ncdump -h $f|grep processing_version;done
A20192892019296.L3m_8D_IOP_adg_443_giop_9km.nc :processing_version = "2018.0" ;
A20192972019304.L3m_8D_IOP_adg_443_giop_9km.nc :processing_version = "2018.0" ;
A20193052019312.L3m_8D_IOP_adg_443_giop_9km.nc :processing_version = "2018.0QL" ;
A20193132019320.L3m_8D_IOP_adg_443_giop_9km.nc :processing_version = "2018.0QL" ;
A20193212019328.L3m_8D_IOP_adg_443_giop_9km.nc :processing_version = "2018.0QL" ;
A20193292019336.L3m_8D_IOP_adg_443_giop_9km.nc :processing_version = "2018.0QL" ;
A20193372019344.L3m_8D_IOP_adg_443_giop_9km.nc :processing_version = "2018.0QL" ;
As you can see from the above, you have to go back to the
24-31 Oct. 2019 composite before you lose the "QL" and have
what we call our refined products which are old enough to
benefit from improved instrument calibration models.
The upshot here is that quick-look products are more likely to be subject
to these sorts of issues. We make them available for those needing
initial qualitative estimates closer to real time. If you require a more
rigorous, quantitative estimate, you are better off waiting until the
refined versions of the products become available.
All that said, we still appreciate folks, such as you, in the ocean color
community keeping vigilant and reporting back issues that they
notice. This is especially true if you notice odd step functions in
our refined products.
Posted: Tue Jan 14, 2020 2:00 pm America/New_York
I had no idea about the processing version indicating QL (!) I will take that into account in future use of the data.
Thanks for checking on this and letting me know the status.