Differences in GOES12 IR3 detectors show up in image processing.

In the text below, the GOES images being discussed have been extracted from GVAR format data using the "gvar" program with options set such that the 10 bit counts were extracted with no scaling into 16 bit tiff image files (the options were "-w ushort -u counts"). Though the values extracted from the GVAR are called "counts", they are not actually raw instrument counts, but are values that have been radiometrically calibrated and corrected at Wallops before being retransmitted in the GVAR format.

Recently, while coding an adaptive image enhancement algorithm, I noticed some differences in the image data from the two IR3 detectors on GOES-12. I had needed to make histograms of sub regions of the images, and noticed that in many of the histograms every other bin was empty. When the highest order bit from these images was extracted and displayed; it had obvious non-random patterns. Here is an example of a sub region of one of the IR3 images along with an image showing the value of the highest order bit and a histogram of the data values:

The left image here is a typical IR3 image, lots of blurry water vapor and a few colder clouds. The right image is the highest order bit of the left image; displayed such that black indicates the bit is zero (i.e the data count is an even number) and white indicates the bit is one (the data count is an odd number). The histogram shows the distribution of counts for the small sub region; it has no y labels since the feature of interest is simply that every other bin is nearly empty.

There are a few distinctive features in the IR3 highest order bit image. First there are large areas where all the data counts are even numbers. The actual data values in those areas are slowly varying, such that a progression of interger values (e.g. 200,200,201,201,202) would be expected instead of a progression of even values (e.g. 200,200,202,202,202) That indicates that the IR3 data has less inherent radiometric precision that the data counts that are delivered in the GVAR format. Inspecting the actual values in the data counts in several regions indicates that precision is about 2 data counts. Second, the striping on the lower third or so of the image is a nice visual evidence of the two IR3 sensors, which have slightly differing responses and/or calibrations, resulting in one sensor having all even values while the other has all odd.

For comparison, here is a similar set of images from channel IR2, where the highest order bit looks somewhat random:

The random appearance of the highest order bit is the result of several factors, but combined with the histogram showing the more even distribution of counts, it indicates that the precision of the IR2 channel is much closer to the counts extracted from the GVAR file.

If the IR3 data is sorted into separate images for each detector, another difference shows up.

"even" detector highest order bits	"odd" detector highest order bits

Here, since the data itself is varying slowly, the spatial rate at which the highest order bit changes indicates that the two IR3 detectors have different precisions.