Acquire 2007 radiometric calibration

[mggr]

This has been an ongoing task since around April.

[mggr]

As of last week, we finally have Specim calibration thanks to Andrew Wilson.

As of this week, we have CASI calibration thanks to Andrew Wilson.

ATM calibration is not yet available - Andrew needs to have the lightbench data (may already have) and then needs to derive the calibration from that. He has explained the process to me (need to document) but thinks that he should do it and we should attempt to replicate. In the meantime, the 2006 ATM seems to work for processing purposes.

[mggr]

5/Oct: Wrote up the ATM calibration procedure as discussed on the phone (see Processing/RadiometricCalibration), asked Andrew to check it and reminded him we still need the ATM calibration.

[mggr]

20/Nov: still waiting for datasets, now from ARSF (though AKW would do too).

[mggr]

Kidlington sent us copies of the post-season ATM radiometric calibration (done on 15/Oct/2007). I also have a logsheet describing what each raw file is.

I've bumped AKW with some more questions on the procedure, but will go ahead and start on it asap.

[mggr]

Acquired copies of the pre-season ATM calibration during a visit to Kidlington (start of Dec) and acquired the ATM light bench N-cal values (calibrated light output) from AKW shortly after the steering committee meeting (17 Dec). The light bench calibration is dated 2003 - have asked FSF to update this on the next convenient opportunity.

[mggr]

Sending AKW a package of stuff to look at (raw data plus results) to compare results on.

[mggr]

AKW returned some results last week. They compare fairly well with the automatic method but a few issues remain. Copy of return email summarising:

Hi Andrew,

Pretty lengthy email here, I'm afraid.  Short summary for everyone else:
 * 2008 is as dark as late 2007
 * automatic method seems close to Andrew's manual method but needs tweaks here and there.

Andrew Wilson wrote (reordered):
> I've spent a couple of half-days processing the ATM data fom
> pre-season (Day047) and post-season (Day 278) using my original visual
> method as a comparison, as requested.

Much obliged for you looking at this :)  I've looked over the results and made some comparisons (see attached akw-comparison.xls).  There are two significant points.

The first one is that I tend to get numbers that are lower for the dark values and higher for the light values.  While this doesn't make much difference for most channels (~1% of dynamic range), it makes a big difference on the channels with low signal (i.e. bands 1 & 2).

What do you think of the differences?

> the documents - column 476 for Day 047 and column 509 for Day 278.

The second point was to do with column numbers - my column numbers and yours broadly agree, but there was quite a lot of variability between bands.  For example, my 278 dataset for 12.5 scanspeed (BB1/2 at 5/30), had the following output:

----------------------------
dark file: output/azhba025.278.bsq
light file: output/azhba002.278.bsq
Size is  938 x 539 x 11
----------results-----------
band  1 - max at index = 455 ; Dark is 24521, light is 24531
band  2 - max at index = 513 ; Dark is 24463, light is 24546
band  3 - max at index = 519 ; Dark is 24424, light is 24837
band  4 - max at index = 520 ; Dark is 24493, light is 24760
band  5 - max at index = 520 ; Dark is 24515, light is 25379
band  6 - max at index = 517 ; Dark is 24699, light is 25753
band  7 - max at index = 507 ; Dark is 24662, light is 27058
band  8 - max at index = 499 ; Dark is 24293, light is 25361
band  9 - max at index = 484 ; Dark is 24491, light is 27571
band 10 - max at index = 480 ; Dark is 24605, light is 29809
band 11 - max at index = 452 ; Dark is 24787, light is 26529

You can see the peak column is all over the place.  I used qcdisplay to look at the raw data file (002) and the overall brightness does seem to move around.  I've plotted spatial profiles of bands 1-10 in qcdisplay and curves do seem to bend over a bit (see attached screenshot, note the column numbers appear to be reversed).

I also did a quick manual scan with ENVI on one dataset and got numbers in the same ball park as my automatic results (band 1 is badly wrong in the automatic method due to the noise level):
band  1 = 551
band  2 = 517
band  3 = 517
band  4 = 518
band  5 = 519
band  6 = 520
band  7 = 512
band  8 = ~500
band  9 = 492
band 10 = 488

Currently I'm calculating the peak value for each band individually.  I guess this may not be a good idea ;)  I'm not sure why the bands would have different peaks though.

Are you considering only particular bands when you're picking where the peak is?

> Do you have the pre-season calibration for 2008 to see if any winter
> maintenance (if there was any) has returned the sensitivity back to
> more normal levels.

I've run the 2008 pre-season with the same code as the 2007 calibrations (ie. prior to any fixes we make) and it looks pretty similar to the 2007 post-season.  If anything the sensitivity is only marginally improved and even a little worse in bands 9 & 10.

I've attached a spreadsheet with the 2008 results added.

> As I mentioned, with a major change in gains between calibrations, it
> is now difficult to assign which calibration should be used for the
> which datasets spread throughout the whole flying season.

Agreed :/  We'll go through all the ATM flights and make an estimate of the brightness (maybe picking max value from azatm output).  While this won't be perfect, it should give us an idea if it was a step change or a decline.  If it's a smooth decline, do you think interpolating the calibration would be acceptable?

> Call if you wish to discuss further on the results or the statistical
> method that you intend to use from now on. How close is it to my
> manual method ?

Not close enough yet ;)  I think it'd be good to iron out the differences until we're happy we're getting something of reasonable quality.

I'll also do a comparison against CASI spectra to see how close things are (might take a while).

[mggr]

Anthony did some comparisons against CASI spectra, but the data was too sparse, ditto for Eagle. Visual or numerical examination is useless without a ground truth, so comparison is the only option. The current plan is, for each flight, do a comparison against whatever data is available and select the closest calibration. Users should be informed of the uncertainty.

Furthermore, the ATM has died in Ethiopia, which may render this work non-useful!

[anee]

Scan speed 12.5:

Band 9 min value 24360, offset 127
Band 10 min value 24470, offset 128

Scan speed 25:

Band 9 min value 24310, offset 177
Band 10 min value 24470, offset 128

Scan speed 50:

Band 9 min value 24200, offset 287
Band 10 min value 24300, offset 298

[anee]

Eagle vs. 2007 ATM calibration

[anee]

Hawk vs. 2007 ATM calibration

[anee]

vs. Eagle:

vs. Hawk:

[mggr]

Finalised versions of the ATM calibration (with dark correction as shown above) are now in the ~arsf/calibration/ structure.

[mggr]

Alternative view of similar data with atm bands as bars.