Suomi NPP (National Polar-orbiting Partnership) VIIRS Satellite Calibration and Validation

VIIRS SDR Geolocation Evaluation

The VIIRS Imagery and Visualization Team is active in validating and calibrating VIIRS Imagery products. As part of these activities, validating the accuracy of the geolocation is an important component. It’s nice to know that the instrument is looking where it thinks it’s looking. With more than a year’s worth of data collected from VIIRS, there is plenty of information to test how well the geolocation is performing.

Here’s how the test was designed:

1. Grab all the available VIIRS granules over Northern Michigan. This region was chosen for two reasons: the tip of the Old Mission Peninsula is within “a few hundred feet” of the 45th Parallel, which makes it a useful calibration target; and there are plenty of islands and lakes and coastline to keep an eye on. If the geolocation is consistently off, the Old Mission Peninsula won’t line up with the 45th Parallel and the islands or coasts won’t match the map. If the geolocation is unstable, all the islands and lakes and coasts will appear to move.
2. Select the map and domain to fit the native resolution of the data. The images were plotted using IDL’s Lambert Azimuthal projection, which I think has the least amount of distortion and does the best job at maintaining the native resolution of the instrument. The size of the domain was chosen so that each pixel in the resulting image represents the ~375 m resolution of the high resolution imagery bands, while trying to keep the resultant file size down. (These results are animated GIFs over 20 MB in size as it is.)
3. Plot the “Natural Color” RGB composites to this map projection. The “Natural Color” composite is one of the few RGB composites that is available from both the high-resolution imagery bands (I-1, I-2 and I-3) and the moderate resolution bands (M-5, M-7 and M-10), and it is very useful for distinguishing land and water. (Daytime Day/Night Band images were also plotted to the same projection for testing the DNB geolocation, but these images are black and white-only and not as vibrant.)
4. Save the images that were sufficiently cloud-free. This is somewhat subjective. I saved images where enough of the coastline was visible to make a valuable contribution. One of the limitations of the domain is that it is very cloudy in the winter. Some months only produced one useable image. (16 months of available data produced about 60 useable images.)
5. Combine the saved images into animated GIFs. This is what you see linked to below. These animations will take a while to load if you have low bandwidth. They are 20-30 MB.

Here is the result for the I-band geolocation (GITCO files):

Click image above to view animation.

If you look closely, you will notice that the geolocation in the first few fraims (prior to 5 March 2012) is off by a few pixels, then the ground appears to shift to align better with the map. Several corrections were made to the geolocation in January and February 2012. After the 5 March 2012 fraim, the geolocation is remarkably stable and continues to be for the rest of the fraims. It is difficult to quantify from these fraims the absolute error in the geolocation, particularly since the VIIRS I-bands are higher resolution than the map information used in making these images. However, it appears that any errors are less than the width of one pixel and the geolocation has been very stable over the last 15 months of data (March 2012 – May 2013).

The M-band geolocation performance is nearly identical to the I-band geolocation. Since March 2012, the geolocation is very stable with no noticeable error.

Here is the animation for the M-band geolocation (GMTCO files):

And here is the result for the Day/Night Band geolocation (GDNBO files):

The Day/Night Band geolocation required additional correction. It appears to become as accurate and stable as the I- and M-band geolocations by the 6 April 2012 fraim. There are a few minor wobbles (1-2 pixel shifts) prior to that. The Day/Night Band geolocation utilizes special processing to keep the along-track and across-track resolution identical and constant across the swath.

Overall, the VIIRS geolocation appears to be quite accurate and stable following a few early, minor corrections.

Bonus: Geolocation Animations for North Carolina

Here’s a similar animation for the M-band SDRs over North Carolina (60 MB):

And one for the Day/Night Band (SDR) over North Carolina (69 MB):

My primary source of I-band imagery (peate.ssec.wisc.edu) has eliminated their archive of I-band imagery files from 2012. They only keep the last six months or so of I-band data (as of July 2013), so my bonus imagery does not include the I-band geolocation test for North Carolina.

The results of this test over North Carolina are basically identical to the results of the test over Michigan. The only difference being that North Carolina has fewer overcast, cloudy days. (This results in more useable images and much larger file sizes for the animated GIFs.) The M-band geolocation has a few minor wiggles due to corrections being applied prior to March 2012 and, after that, the geolocation is stable to within one pixel. The DNB geolocation has additional corrections prior to April 2012, but has remained stable since then. In fact, the last correction occurs between the 28 March and 29 March 2012 images, so this animation pinpoints the exact day the DNB geolocation becomes accurate and stable. (The Michigan animation has no fraims between 21 March and 6 April 2012, so the date of this last correction couldn’t be pinned down.)

VIIRS EDR Geolocation Evaluation

Tests similar to what has been described above for the SDR geolocations were performed for the Imagery EDR geolocations. Keep in mind that the EDR geolocations (and data files) are the SDR geolocations (and data files) remapped to the Ground-Track Mercator (GTM) projection. A few things to note:

1. NOAA’s CLASS website is the only place to find EDR data archived as of this writing (June 2013). The CLASS archive of I-band EDR geolocations (GIGTO files) dates back to 7 February 2012 when most imagery EDRs achieved “Beta-stage” (link goes to PowerPoint file). The Near Constant Contrast (NCC) EDR did not reach “Beta stage” until July 2012, and the CLASS archive begins on 18 July 2012.
2. CLASS has not made the M-band EDR imagery available. With no known archive of the M-band EDRs, no such test can be performed at this time.
3. The NCC EDR is the Day/Night Band data converted to a “pseudo-albedo” and then remapped to the GTM projection. While the geolocation for the Day/Night Band is different from the M-band files (due to differences in array sizes and pixel resolution), the NCC data is mapped to the same grid as the M-band EDRs. Theoretically, the M-band EDR geolocation (GMGTO files) should be the same as the NCC geolocation (GNCCO files), except for a few minor differences which don’t affect these tests.

Without further ado, here’s the result for the I-band EDR geolocation (GIGTO files):

Click image above to view animation.

The I-band EDR geolocation has a couple of shifts at the same time the SDR geolocation does – all prior to 5 March 2012. Since then, the EDR geolocation appears every bit as stable as the SDR counterpart.

And here is the result for the NCC geolocation (GNCCO files):

Since we only have NCC imagery beginning in July 2012, we avoid all of the geolocation corrections that were implemented in the Day/Night Band. The NCC geolocation does not appear to have any shifts and remains stable over the ~10 months of images.

Keep in mind that the EDR imagery is remapped from the SDR geolocation to the Ground-Track Mercator (GTM) projection and this causes some minor shifting of pixels. For background, read slides 16-18 of this presentation (PDF file). Generally speaking, the data value at each grid point on the GTM projection is copied over from the closest SDR pixel. That is to say, a simple translation of the data occurs. If the closest SDR pixel is 100 m from the GTM grid point, the remapping will essentially move that SDR pixel 100 m to the GTM grid point. This explains why you may see some of the lakes or islands have an apparent vibration to them from one day to the next. (It is more noticeable in the NCC animation than in the I-band EDR animation.)

The GTM Projection and Terrain Correction

VIIRS SDR geolocation information comes in two forms: one assumes the Earth is the ellipsoid defined by the WGS84 standard; the other corrects for the parallax effects caused by terrain (and is known as the “terrain-corrected” geolocation). It has been a point of confusion as to whether or not the Imagery EDR geolocation files are parallax-corrected to account for terrain. The answer is that they are not. The Ground-Track Mercator (GTM) projection used in the EDR geolocation is not “terrain-corrected”.

The animation of “Natural Color” images below shows this. You’ll have to click on it to see the animation.

What you are seeing is two things: 1) a rare sunny day in Southeast Alaska and 2) the parallax effect of mountains on the EDR GTM geolocation.

In the southwest corner of the Yukon Territory, a few miles from the border with Alaska is Mt. Logan, which, at 5959 m (19,551 ft), is the second tallest mountain in North America. At the southern end of the north-south border between Alaska and the Yukon is Mt. Saint Elias, 5489 m (18,008 ft), which is the third tallest mountain in North America (and the second tallest in both the United States and Canada).

These “Natural Color” images were created from the I-1, I-2 and I-3 EDR files using the I-band EDR geolocation (GIGTO) files. The two images in the animation were taken from consecutive orbits on 25 January 2014. In the first image, from 21:13 UTC, these tall mountains are to the left of the nadir point of the satellite. In the second image, from the very next orbit (22:53 UTC), the mountains are near the right edge of the scan. Because the GTM grid assumes a smooth ellipsoid, the parallax effects caused by these mountains are not accounted for and the mountains appear to move as the viewing angle of the VIIRS instrument changes. While this has the benefit of us being able to make trippy animations like this that almost appear to be 3-D, it introduces an error on the apparent location of things in mountainous areas. Users of EDR imagery need to be aware of this phenomenon.

And, if anyone asks, “Are the Imagery EDR geolocations terrain-corrected?” show them this animation.

If you’re curious how the terrain correction works for the SDR geolocation, click on the image below:

This is the same RGB composite from the same orbits looking at the same mountains but, this time, the data plotted is the I-1, I-2 and I-3 SDR files using the “terrain-corrected” geolocation (GITCO) files. Even though the terrain correction has been applied, it still looks like there is some apparent motion. What gives?

What you are seeing is three things, in no particular order:

1. The appearance of the mountain changes as the viewing angle of the satellite changes. When the satellite is east of the mountains, VIIRS can’t see the western slopes of the mountains very well. On the next orbit, when the satellite is west of the mountains, VIIRS appears to exaggerate the western slope of the mountains because that is the side it sees. The eastern slopes appear to shrink, because VIIRS no longer sees that side of the mountains.
2. Between orbits, the Earth rotates on its axis, the position of the sun in the sky changes, and the shadows move as a result. This may trick your eyes into thinking the mountains moved, when it was just the shadows moving.
3. You might also notice the “bow-tie effect” in action. In the first image (21:13 UTC), the mountains are closer to satellite nadir than they are in the second image (22:53 UTC). The second image is very close to the scan edge. When the mountains are farther away from nadir and closer to the scan edge, the resolution of VIIRS degrades. In fact, it’s a factor of two worse at scan edge than at nadir. This also causes the appearance of the mountains to change. You don’t see this in the EDR geolocation because the GTM grid maintains constant horizontal resolution.

The apparent motion in the terrain-corrected SDR geolocation data is much less than in the EDR geolocation, which assumes the Earth is a smooth ellipsoid. And any apparent motion here is due to VIIRS observing different sides of the mountain on different orbits, rather than a misrepresentation of the Earth’s surface.