The Case of the 100-year-old Ash Cloud

Lost in all the commotion caused by Hurricane Sandy, a curious event occurred on the other side of the country on 30 October 2012. A cloud of ash obscured the skies of Kodiak Island, Alaska, diverting flights in the region and forcing the people of Kodiak to stay inside or wear masks. Alaska has quite a few volcanoes, so this may not be a big thing to them except, this was no ordinary volcanic eruption: it was the leftovers of a volcanic eruption from 100 years ago!

The volcano that came to be known as Novarupta erupted on 6 June 1912. It was one of the largest volcanic eruptions of recorded history. It was 10 times more powerful than Mt. St. Helens with 100 times more ash. The explosion was heard more than 1100 km (700 miles) away in Juneau. The force of the eruption caused nearby Mt. Katmai to collapse on itself (10 km away). It formed the Valley of Ten Thousand Smokes and, most importantly for us, covered the surrounding land with 150 m (500 ft) of ash.

This pile of ash – still there today – can be lifted by a stiff breeze (or, more appropriately, “strong breeze” or higher on the Beaufort wind scale), and blown pretty high off the ground (4000 ft according to the news report). This isn’t the first time this has happened. MODIS observed the same thing back in 2003.

So, what did VIIRS see? Here’s the “true color” image, the RGB composite of channels M-03 (0.488 µm, blue), M-04 (0.555 µm, green) and M-05 (0.672 µm, red):

VIIRS "true color" RGB composite of channels M-03, M-04 and M-05, taken 22:23 UTC 30 October 2012

VIIRS "true color" RGB composite of channels M-03, M-04 and M-05, taken 22:23 UTC 30 October 2012

Be sure (as with all the images) to click on the image, then on the link below the banner to see it at full resolution. (The link contains the dimensions of the full size image.)

The ash cloud (blowing right over the center of Kodiak Island) is not as obvious in this image as it was in the MODIS image in the link above, although it is visible. To be fair, the plume was much more optically thick in 2003, and there were fewer clouds and less snow to confuse it with.

Here is the false color (“pseudo-true color” or “natural color”) image, the RGB composite of channels M-05 (0.672 µm, blue), M-07 (0.865 µm, green) and M-10 (1.61 µm, red):

VIIRS false color RGB composite of channels M05, M-07 and M-10, taken 22:23 UTC 30 October 2012

VIIRS false color RGB composite of channels M05, M-07 and M-10, taken 22:23 UTC 30 October 2012

Hmmm. Once again, the ash plume is visible but not particularly noticeable. Is there a way to highlight the ash plume to make it easier to see?

EUMETSAT (the European Organisation for the Exploitation of Meteorological Satellites) has defined an RGB composite for detecting dust. Their product, which was developed primarily to detect dust storms over the Saharan desert, uses channels that are present (or similar to ones that are present) on VIIRS. This means we can apply the dust product for VIIRS as the difference between M-16 and M-15 (red), the difference between M-15 and M-14 (green) and M-15 by itself (blue), all in units of brightness temperature. If you do that, and use the same color scaling they use, you get this image:

The EUMETSAT Dust RGB composite applied to VIIRS for 22:23 UTC 30 October 2012

The EUMETSAT Dust RGB composite applied to VIIRS for 22:23 UTC 30 October 2012

The arrow points to the source region of the ash plume. In this RGB composite, dust shows up as hot pink (magenta), but it’s barely visible here. The reason is that this dust product is primarily useful where there is a large temperature contrast between the dust plume and the background surface, which we don’t have here.

A more common way to detect volcanic ash is to use the “split-window difference”. The “split-window difference” is the difference in brightness temperature between a 10.7-11.0 µm channel and a 12.0 µm channel. This difference is useful because volcanic ash has a difference of opposite sign to most everything else. Here’s what the split window difference (M-15 – M-16) looks like for this case:

VIIRS "Split-window difference" image from 22:23 UTC 30 October 2012

VIIRS "Split-window difference" image from 22:23 UTC 30 October 2012

This image has been scaled so that the colors range from -1 K (black) to +7 K (white). The ash plume stands out a bit more here by being much darker than the background. The only problem is, it isn’t perfect. Large amounts of water vapor, optically thick clouds, desert surfaces and boundary layer temperature inversions can all produce a negative difference (just like volcanic ash does).

These problems can be overcome to a certain extent by combining the “split-window difference” with a Principal Component Image (PCI) analysis technique. (This technique is too complicated to describe here but, if you have access to AMS journals, check out these journal papers.) Now, the ash plume is the only thing that’s black:

VIIRS PCI analysis image from 22:23 UTC 30 October 2012

VIIRS PCI split window analysis image from 22:23 UTC 30 October 2012. Image courtesy Don Hillger. Upside-down text courtesy McIDAS-X.

Notice the smaller plume identified by the orange arrow. This plume is not easy to identify in any of the previous images. The PCI technique works well. But, we’re not going to stop there.

Remember the dust plumes off the Cape Verde islands? They produced a strong signal in the difference between M-12 (3.7 µm) and M-15 (10.7 µm) due to solar reflection. Does a 100-year-old ash plume produce a similarly strong signal? See for yourself:

VIIRS channel difference image between M-12 and M-15 from 22:23 UTC 30 October 2012

VIIRS channel difference image between M-12 and M-15 from 22:23 UTC 30 October 2012

It does produce a signal, but it’s not as bright as the surrounding clouds. The color scale here ranges from -2 K (black) to +90 K (white).

M-06 (0.746 µm) is highly sensitive to anything that reflects solar radiation in the atmosphere or on the surface, which we learned from Hurricane Isaac. Here’s what the M-06 image looks like:

VIIRS channel M-06 image, taken 22:23 UTC 30 October 2012

VIIRS channel M-06 image, taken 22:23 UTC 30 October 2012

“Big deal,” you say. “None of those are better than the PCI analysis.” That may be true, but watch what happens when we combine M-06, the M-12 – M-15 image and the split-window difference image in a single RGB composite:

VIIRS RGB composite of M06 (blue), M12 - M15 (green) and M15 - M16 (red), taken 22:23 UTC 30 October 2012

VIIRS RGB composite of M06 (blue), M12 - M15 (green) and M15 - M16 (red), taken 22:23 UTC 30 October 2012

In this composite, blue values represent the M-06 reflectance scaled from 0 to 1.6, green values represent the brightness temperature difference between M-12 and M-15 scaled from -2 K to +90 K, and red values represent the brightness temperature difference between M-15 and M-16 scaled from -1 K to +7 K.

From a theoretical perspective, this RGB composite does exactly what you want: make the thing you’re trying to detect the only thing that is a certain color. For example, the ash plumes are the only things in this image that are green. From a practical perspective, however, this RGB composite doesn’t work so well. It only works because the ash plume is over water (otherwise M-06 wouldn’t be very useful). It only works during the day, where M-06 is available and the difference between M-12 and M-15 is significant (no solar component to M-12 at night).

Plus, the rainbow of colors is difficult to make sense of: green ash; clouds ranging from light blue to purple to orange (a function of optical thickness, particle size, and phase); bright purple snow; dark purple vegetation; maroon water. It’s not exactly pleasing to the eye. In contrast, the PCI analysis technique that uses the split-window difference works day and night, over ocean and over land. And it isn’t confusing to look at. Maybe we should have stopped when we got to the PCI technique. But then, we wouldn’t have learned anything new.

Hurricane Isaac: Before, During and After

While Hurricane Isaac (then a tropical storm) did not destroy Tampa, Florida as many people feared, it certainly left its mark on the Gulf Coast. With many locations from Florida to Louisiana receiving more than 12″ of rain, and levees unable to keep out the storm surge, flooding was (and still is) a major problem. Look at these aerial photos of Isaac’s aftermath in Louisiana. The Visible Infrared Imaging Radiometer Suite (VIIRS) aboard Suomi NPP saw that flooding, also.

But first, let’s look at the high resolution infrared (IR) window channel (I-05, 11.45 µm) which, at ~375 m resolution, is the highest-resolution IR window channel on a public weather satellite in space today. This image was taken when Isaac was still a tropical storm in the middle of the Gulf of Mexico:

VIIRS I-05 image of Tropical Storm Isaac, taken 18:50 UTC 27 August 2012

VIIRS I-05 image of Tropical Storm Isaac, taken 18:50 UTC 27 August 2012

This image uses a new (to this blog, anyway) color scale, developed by our colleagues at CIMSS, that really highlights the structure of the clouds at the top of Isaac. The color scale is included in the image. For comparison, here’s the GOES Imager IR window channel (channel 4, 10.7 µm) image from roughly the same time:

GOES-13 Imager channel 4 image of Tropical Storm Isaac, taken 18:45 UTC 27 August 2012

GOES-13 Imager channel 4 image of Tropical Storm Isaac, taken 18:45 UTC 27 August 2012

GOES has ~4 km resolution in its IR channels. VIIRS provides amazing details of the structure of tropical cyclones that you just can’t get with current geostationary satellites.

The real story from Isaac, however, is the flooding. It’s hard to capture flooding from a visible and infrared imaging instrument, since flooding usually occurs when it’s cloudy. Clouds block the view of the surface when looking at visible and infrared wavelengths. But, large quantities of water that fail to evaporate or drain into the local rivers after a period of several days can be seen after the skies clear. That’s what happened with Isaac.

Here are before-Isaac and after-Isaac images of the southern tip of the Florida Peninsula. These are false color (“pseudo-true color”) composites of VIIRS channels I-01, I-02 and I-03. These images were taken on the afternoon overpasses of 23 August and 29 August 2012. Many cities on the east coast of Florida got 10-16 inches of rain (250-400 mm for those of you outside the U.S.). See if you can pick out the flooding.

False color RGB composite of VIIRS channels I-01, I-02 and I-03 taken before and after Tropical Storm Isaac (2012)

False color RGB composite of VIIRS channels I-01, I-02 and I-03 taken before and after Tropical Storm Isaac (2012)

If you have been following this blog, you know that, in the “pseudo-true color” RGB composite, water shows up very dark – in most cases, almost black. That’s not always true, of course. You can see sun glint (particularly in the “before” image) that makes water a lighter color and shallow water (where visible radiation [i.e. channel I-01] is able to penetrate to the bottom) shows up as a vivid blue.

Now, notice the Everglades. Many areas of the Everglades, particularly on the east side, appear darker in the “after” image, because those swampy areas have a lot more water in them. Water has a lower reflectivity than vegetation or bare ground at these wavelengths.

The effect of water on the land surface shows up even better in the moderate resolution channel M-06 (0.75 µm). M-06 is a channel not shown before because it is perhaps the worst channel for producing interesting images. M-06 was designed to aid in ocean color retrievals and/or other uses that require atmospheric correction. The M-06 detectors saturate at a low radiance, so any radiation at 0.75 µm that reflects off of clouds, aerosols or the land surface easily show up. About the only things that have low reflectivity in M-06 are atmospheric gases and water surfaces without sun glint. Ocean color retrievals need a very clean atmosphere with no aerosols or clouds and no sun glint to work correctly. You also need to be able to identify what is or is not water, which is what makes M-06 useful for identifying flooding.

Here are the similar before-Isaac and after-Isaac images of Florida from M-06:

VIIRS channel M-06 images of southern Florida taken before and after Tropical Storm Isaac (2012)

VIIRS channel M-06 images of southern Florida taken before and after Tropical Storm Isaac (2012)

Both the land and optically thick clouds saturate M-06, so this channel is useless at identifying clouds over land (except you can see some cloud shadows). Sun glint is saturating the pixels over the Gulf of Mexico in the “before” image, while it is mostly to the east of Florida in the Atlantic Ocean in the “after” image. In the “after image”, reflective cirrus clouds over the Gulf of Mexico show up that are not as easily visible in the RGB composite. Of primary importance here, however, is the dark appearance of the Everglades in the “after” image. All that flood water reduced the reflectivity of the land surface, making it appear darker. That means, if you know where the clouds (and, hence, the cloud shadows) are, it may be possible to use M-06 to identify large flooded areas.

Louisiana and the coast of Mississippi were the hardest hit by Isaac, and the flooding is easily visible here, too. In fact, the massive flooding is easier to see in the RGB composite in this region. Compare the “before” and “after” images, taken on 26 August 2012 and 1 September 2012:

False color RGB composites of VIIRS channels I-01, I-02 and I-03 of southeast Louisiana

False color RGB composites of VIIRS channels I-01, I-02 and I-03 of southeast Louisiana

To make it easier to see, here’s a quick animation of the before and after images. Watch the highlighted areas.

Animated GIF of false color RGB composites taken from VIIRS before and after Hurricane Isaac

Animated GIF of false color RGB composites taken from VIIRS before and after Hurricane Isaac

After the passage of Hurricane Isaac, Lake Maurepas and Lake Pontchartrain almost appear to merge into one big lake! Other flooding is visible near Slidell, Bay St. Louis, Pascagoula Bay, and the heavily hit parishes of Plaquemines, St. Bernard, Lafourche and Terrebonne.

Thin cirrus clouds are visible in the “after” image, which limit the ability of M-06 to detect some of the flooding, but M-06 is still able to see the large areas of flooding highlighted in the animation above. M-06 also detects reflection off of the Twin Spans as well as the Lake Pontchartrain Causeway. And this is at ~750 m resolution!

VIIRS channel M-06 images of southeastern Louisiana taken before and after Hurricane Isaac (2012)

VIIRS channel M-06 images of southeastern Louisiana taken before and after Hurricane Isaac (2012)

So, don’t try to do ocean color retrievals in pixels obscured by big bridges.