Wild Week of Wildfires, Part II

Last time on “Wild Week of Wildfires“, we looked at the Little Bear Fire and High Park Fire, two lightning-ignited fires burning out west that were so hot they caused saturation in the two 3.7 µm channels on VIIRS (I-04 and M-12). There was mention of the Duck Lake Fire, a lightning-ignited fire in northern Michigan, which VIIRS also saw, and I couldn’t resist showing some more images.

On 9 June 2012, the same day the High Park Fire exploded (figuratively speaking), the Duck Lake Fire finally reached 100% containment after burning over 21,000 acres. The next day (10 June 2012), Suomi NPP passed over the Upper Peninsula of Michigan, and it was actually a clear day. (This joke comes courtesy of 20+ years experience of living in Michigan.) Even with 100% containment, the hot spot of the fire was still clearly visible in VIIRS channel I-04 (3.7 µm) that afternoon:

Channel I-04 image of the Duck Lake Fire from VIIRS, taken 18:18 UTC 10 June 2012

Channel I-04 image of the Duck Lake Fire from VIIRS, taken 18:18 UTC 10 June 2012

The highest brightness temperature in the burn area in this channel at this time was    ~331 K. As we saw before with the Lower North Fork Fire, the high resolution false color composite of channels I-01, I-02 and I-03 is useful in highlighting the burn area:

False color RGB composite of VIIRS channels I-01 (blue), I-02 (green) and I-03 (red), taken 18:18 UTC 10 June 2012

False color RGB composite of VIIRS channels I-01 (blue), I-02 (green) and I-03 (red), taken 18:18 UTC 10 June 2012

Notice the large, brown area that coincides with the hot spot in the I-04 image. The combination of wavelengths used in this composite (0.64 µm [blue], 0.865 µm [green] and 1.61 µm [red]) is quite sensitive to the amount (and health) of the vegetation.

You might have also noticed several other interesting features in the image that show up better when you zoom in:

False color composite of VIIRS channels I-01, I-02, and I-03 from 18:18 UTC 10 June 2012

False color composite of VIIRS channels I-01, I-02, and I-03 from 18:18 UTC 10 June 2012

The Upper Peninsula of Michigan was based on mining for most of its history, and several large mines and quarries still exist, which VIIRS can easily see.

If you didn’t know any better, you might confuse the iron mine southwest of Marquette, Michigan with a frozen lake, or miraculously un-melted snow leftover from winter, since that is just what snow and ice look like in this kind of RGB composite. Compare that with the true color view of the same area:

True color RGB composite of VIIRS channels M-3, M-4 and M-5, taken 18:18 UTC 10 June 2012

True color RGB composite of VIIRS channels M-3, M-4 and M-5, taken 18:18 UTC 10 June 2012

In this case, the iron mine stands out as a bright red. Why?

The true color composite uses wavelengths at 0.48 µm (blue), 0.55 µm (green) and 0.67 µm (red). The red channel in the true color composite is actually in the red portion of the visible spectrum. The blue channel in the false color composite (0.64 µm) is also in the red portion of the visible spectrum.

This example shows that the iron oxide (rust) produced at the iron mine is highly reflective in the red portion of the visible spectrum. That’s what gives it the characteristic rust color. Iron oxide is not nearly as reflective at shorter or longer wavelengths, so it shows up blue when red wavelengths are used as the blue channel (as in the false color composite) and red when they are used as the red channel (as in the true color composite).

Let this be a lesson to anyone who uses the false color composite as part of a snow and ice detection algorithm. Snow and ice are not the only things to show up that color. You may be looking at a really large iron mine.

A Wild Week of Wildfires

The last few weeks have been filled with lightning-ignited wildfires across the United States. The County Line Fire, along the Florida-Georgia border was caused by lightning on 5 April 2012 and burned ~35,000 acres. The Whitewater-Baldy Complex (began 16 May 2012) – the largest wildfire in New Mexico history – started as two different fires (both caused by lightning) that merged together. It’s over 280,000 acres (that’s not a typo) and continues to burn (as of 13 June 2012). The Duck Lake Fire (began 24 May 2012) burned 21,000 acres of Michigan’s Upper Peninsula and was caused by lightning. The Little Bear Fire (began 4 June 2012), also in New Mexico, was caused by lightning and has burned ~37,000 acres.  Much closer to home, the High Park Fire (began 9 June 2012) is already the largest wildfire in Larimer County history and the third largest fire in Colorado history. It has burned ~46,000 acres and I bet you can guess what caused it.

It’s not clear who is to blame here – there is a long list of suspects – but I bet it was Thor. Even though the U.S. is generally the domain of the Thunderbird, Thor has a mountain-crushing hammer called Mjöllnir, which makes him as good a suspect as any. He may have been in cahoots with Indra or Marduk who are the bringers of rain, and have been holding back on us. Look at how dry it has been across the majority of the country.

With all of these fires, it’s hard to know where to begin. We’re going to ignore the County Line Fire as it was put out over a month ago. We’re also going to ignore the Whitewater-Baldy Complex, as it is so big, it can be seen by GOES. (Kidding! We kid because we love.) Plus, it’s been done before. The VIIRS view of the High Park Fire has also been looked at by CIMSS, with an interesting comparison between VIIRS and MODIS.

What we are going to do is show off interesting features of some of these fires that haven’t been shown or discussed before (as far as we know). We begin with “saturation”. Both the High Park Fire and Little Bear Fire saturated the VIIRS 3.7 µm channels (I-04 and M-12):

Channel I-04 image of the Little Bear Fire from VIIRS taken 20:16 UTC 9 June 2012

Channel I-04 (3.7 µm) image of the Little Bear Fire from VIIRS taken 20:16 UTC 9 June 2012

Channel M-12 image of the Little Bear Fire from VIIRS taken 20:16 UTC 9 June 2012

Channel M-12 (3.7 µm) image of the Little Bear Fire from VIIRS taken 20:16 UTC 9 June 2012

Channel I-04 image of the High Park Fire from VIIRS taken 19:59 UTC 10 June 2012

Channel I-04 (3.7 µm) image of the High Park Fire from VIIRS taken 19:59 UTC 10 June 2012

Channel M-12 image of the High Park Fire from VIIRS taken 19:59 UTC 10 June 2012

Channel M-12 (3.7 µm) image of the High Park Fire from VIIRS taken 19:59 UTC 10 June 2012

The top two images are of the Little Bear Fire, which formed near the border of Lincoln and Otero counties in New Mexico. The bottom two images are of the High Park Fire in Larimer County, Colorado. For each fire, the high resolution 3.7 µm channel (I-04) is compared with the moderate resolution 3.7 µm channel (M-12). The colors range from white (cold) to black (hot). But, wait a minute! If white is cold, why are there white pixels mixed in with the black ones that indicate the hot spots? That’s because these channels are saturating and experiencing “fold-over”. The peak brightness temperatures these channels can measure is ~ 367 – 368 K. Anything warmer than that won’t be detected, so the channel is said to be saturated. When it really gets above that limit you can have “fold-over”, where not only are you not observing the higher, correct temperature, the detectors actually report a lower temperature or radiance. In these fires, the fold-over is resulting in brightness temperatures down to 203 K for M-12 and 208 K for I-04, which is about 90-100 K colder than even the area surrounding the fires!

Luckily, VIIRS has a 4.0 µm channel (M-13) that was designed to not saturate at the temperature of typical wildfires. Compare the hottest pixels in the M-13 images below with the fold-over pixels from M-12 and I-04 above:

Channel M-13 image of the Little Bear Fire from VIIRS taken 20:16 UTC 9 June 2012

Channel M-13 (4.0 µm) image of the Little Bear Fire from VIIRS taken 20:16 UTC 9 June 2012

Channel M-13 image of the High Park Fire from VIIRS taken 19:59 UTC 10 June 2012

Channel M-13 (4.0 µm) image of the High Park Fire from VIIRS taken 19:59 UTC 10 June 2012

The hottest pixel in M-13 reached a temperature of 588 K for the Little Bear Fire and 570 K for the High Park Fire – over 200 K warmer than the saturation points of M-12 and I-04!

These fires were so hot, they appeared in channels that don’t usually show a fire signal. Limiting our attention to the High Park Fire (which was almost literally in our back yard), here’s the I-05 (11.5 µm) image from 10 June 2012:

Channel I-05 image of the High Park Fire from VIIRS taken 19:59 UTC 10 June 2012

Channel I-05 (11.5 µm) image of the High Park Fire from VIIRS taken 19:59 UTC 10 June 2012

The highest temperature observed in I-05 was 380 K. Longer wavelength channels, such as in I-05 are less sensitive to sub-pixel hot spots than channels in the 3.7 – 4.0 µm range, so fires don’t often show up. For pixels to have a 380 K brightness temperature in I-05, it means that the average temperature over the entire pixel had to be above +100 °C – hot enough to boil water!

Fires don’t often show up at shorter wavelengths, either, because the amount of solar radiation usually dwarfs any signal from the Earth’s surface. But, the High Park Fire did reach saturation at 2.25 µm (M-11):

Channel M-11 image of the High Park Fire from VIIRS taken 19:59 UTC 10 June 2012

Channel M-11 (2.25 µm) image of the High Park Fire from VIIRS taken 19:59 UTC 10 June 2012

The color scale has been reversed so that it is more inline with visible imagery. The white pixels represent saturation in M-11 at a radiance of 38 W m-2 µm-1 sr-1. The reflectance of these pixels saturated at a value of 1.6, which means that the amount of radiation detected in this channel was more than 1.6 times the amount you would expect to see if the surface was a perfect mirror reflecting all the solar radiation back to the satellite. Thus, the fire’s contribution to the total radiance was significant in this channel.

The contribution from the surface (i.e., the fire) was also visible in the 1.6 µm channel (M-10), but it isn’t exciting enough to show. One channel shorter down on VIIRS (M-9, 1.38 µm) and the signal disappears against the high reflectivity of the smoke plume.

It’s impossible to leave out the Day/Night Band, which shows just how large and how close the High Park Fire got to Fort Collins:

Day/Night Band image of the High Park Fire from VIIRS taken 09:58 UTC 11 June 2012

Day/Night Band image of the High Park Fire from VIIRS taken 09:58 UTC 11 June 2012. Image courtesy Dan Lindsey.

The smoke plume, while not exactly visible, is affecting the view of the east side of the fire and Fort Collins, making them appear more blurry than they would if the sky were completely clear. You can also see that, overnight on 11 June 2012, the fire covered an area larger than any of the cities visible in the image (except for Denver, which is mostly cropped off the bottom of the image).

Hopefully, Marduk will start doing his job and bring us some rain and these will be the last fires for a while.

The Hewlett Fire

According to reports, a man camping along the Hewlett Gulch trail in Roosevelt National Forest on 14 May 2012 had his camping stove knocked over in a gust of wind. One week (and $2.9 million) later, the Hewlett Fire scorched more than 7600 acres before fire crews could gain the upper hand. At one point 80 homes were evacuated but, thankfully, none of them were damaged. The smoke plume could be seen as far away as Laramie, Wyoming. Less than 20 miles away from the Cooperative Institute for Research in the Atmosphere, our home, it certainly caught our attention.

VIIRS aboard Suomi NPP monitored the fire day and night. About an hour after the fire was first reported, VIIRS captured the hot spot in channel I-04 (3.7 µm):

Image of the Hewlett Fire from VIIRS channel I-04, 20:05 UTC 14 May 2012

Image of the Hewlett Fire from VIIRS channel I-04, 20:05 UTC 14 May 2012

In the above image, the warmest (darkest) pixel had a brightness temperature of 350 K.  A simple RGB composite of channels I-01 (0.64 µm), I-02 (0.87 µm) and I-03 (1.61 µm), with no other manipulation, from the same time as the I-04 image above, produces a red spot right over the I-04 hot spot:

False color RGB composite of VIIRS channels I-01, I-02 and I-03, 20:05 UTC 14 May 2012

False color RGB composite of VIIRS channels I-01, I-02 and I-03, 20:05 UTC 14 May 2012

Perhaps more amazing (but less useful from a firefighting perspective) is that, if you look closely (and you know the geography of the area), you can make out the locations of the following highways: I-25, I-76 and I-80, plus the main Union Pacific railroad tracks that more-or-less parallel I-80 in southern Wyoming. The high resolution imagery bands on VIIRS have enough resolution to identify interstate highways!

Suomi NPP passed over the area that night (15 May 2012) and the Day/Night Band (DNB) captured the fire burning brightly:

Day/Night Band image of the Hewlett Fire, 08:25 UTC 15 May 2012

Day/Night Band image of the Hewlett Fire, 08:25 UTC 15 May 2012. Image courtesy Dan Lindsey.

By the time of the 17 May 2012 nighttime overpass – two days later – the fire had grown significantly. With no clouds around, the DNB easily saw the Hewlett Fire, as it was the brightest thing in the area. The image below has been enhanced to make the nearby city lights easier to see relative to the fire.

Day/Night Band image of the Hewlett Fire, 09:26 UTC 17 May 2012

Day/Night Band image of the Hewlett Fire, 09:26 UTC 17 May 2012

In the above image, lights from various cities have been identified. The red arrow indicates the Hewlett Fire, which was bright enough and large enough to be confused for a city. The yellow arrow indicates what might be oil and/or gas flares burning in rural Weld County, which you can also see in the 15 May 2012 DNB image. Weld County is home to a third of all the oil and gas wells in Colorado.

In this zoomed-in image, you can see that the light from the fire covered an area approximately one third the size of Fort Collins:

Zoomed Day/Night Band image of the Hewlett Fire, 09:26 UTC 17 May 2012

Zoomed Day/Night Band image of the Hewlett Fire, 09:26 UTC 17 May 2012. Image courtesy Dan Lindsey.

This image was taken before the burn area even reached its maximum size. At the same time, channel I-04 also saw this ring of fire (not to be confused with the “ring of fire” caused by the recent annular eclipse):

VIIRS channel I-04 image of the Hewlett Fire, 09:26 UTC 17 May 2012

VIIRS channel I-04 image of the Hewlett Fire, 09:26 UTC 17 May 2012

Once again, darker colors indicate higher brightness temperatures. The peak temperature in channel I-04 at this time was 356 K.

Even though it caused no damage to homes or structures, it was a little too close for comfort for many people.

As a final note, our partners up the hill in the Department of Atmospheric Science have taken an interest in the Hewlett Fire. If you are interested in the non-satellite side of the research into this fire, research groups led by Professors Rutledge, Kreidenweis and Collett have collected radar observations and in situ aerosol samples of the smoke plume. Contact them for more information.

Time-lapse of the Lower North Fork Fire

On 26 March 2012, strong winds, high temperatures and low humidities re-ignited embers from a controlled burn that took place the previous week near Conifer, CO. The Lower North Fork fire quickly spread in the high winds, eventually burning more than 4000 acres and damaging or destroying 27 homes. Three people were killed, presumably because they were unable to evacuate before their homes were engulfed in flame. One family’s daring escape from the fire was caught on a cell phone camera and made national news (CAUTION: strong language has not been edited out). Many interesting pictures of the fire may be found here, here, and here.

Channel I-4 of VIIRS (centered at 3.74 µm) captured the hot spot from the Lower North Fork fire on each of Suomi-NPP’s afternoon (ascending) overpasses last week. These images make up the loop shown below.

5-day loop of I-4 images of the Lower North Fork fire

5-day loop of afternoon I-4 images of the Lower North Fork fire

In this image loop, the color scale represents observed brightness temperature such that warmer pixels appear darker and cooler pixels appear lighter. Pixels warmer than 330 K appear black, and pixels colder than 250 K appear white. The time between each image in the loop is approximately 24 hours.

The first image in the loop, taken at 20:24 UTC on the 26th, captured the hot spot shortly after the fire was first reported. The hot spot as seen by I-4 expanded significantly during the first 24 hours, before lighter winds and firefighting efforts greatly limited the growth of the fire. Over the last three frames, the hot spot can be seen to cool and shrink slightly.

Low (liquid) clouds can be seen as dark splotches on the images from the 28th and 29th of March, which should not be confused with fires. This is due to the fact that liquid clouds are highly reflective at 3.7 µm, and the reflection of solar radiation during the day increases the observed brightness temperature, so they appear darker. The persistently bright sideways “C” shape to the northeast of the fire is Chatfield Reservoir, which has a low brightness temperature due to the low water temperature in the reservoir and the relatively low emissivity of liquid water at this wavelength. Cherry Creek Reservoir (to the northeast of Chatfield Reservoir) and Marston Lake (to the north of Chatfield Reservoir) can also be seen.

With clear skies, the burn area shows up quite clearly in the I-band false color RGB composite of I-1, I-2 and I-3, taken at 20:06 UTC 27 March 2012 – the same time as the second frame of the loop above.

RGB composite of VIIRS channels I-1, I-2 and I-3 of the Lower North Fork fire, 20:06 UTC 27 March 2012

RGB composite of VIIRS channels I-1, I-2 and I-3 of the Lower North Fork fire, 20:06 UTC 27 March 2012

The burn area shows up as a sizeable dark brown spot in the forests (which show up as green) southwest of Denver.

After the driest and warmest March on record in Denver, hopefully this is not the start of a long, devastating fire season (link goes to PDF file).

I- and M- Band Views of the Heartstrong Fire

The Heartstrong Fire in Yuma County, Colorado, 18 March 2012

The Heartstrong Fire in Yuma County, Colorado, 18 March 2012 (uncredited photo)

On 18 March 2012, very warm, very dry and very windy conditions existed throughout eastern Colorado. Surface observations showed temperatures in the 70s and 80s, dew points in the teens and 20s, and sustained winds at 20-30 knots (gusting over 40 knots). Wind gusts up to 60 knots (~70 mph) were reported.

Surface observations, 19:00 UTC 18 March 2012

Surface observations, 19:00 UTC 18 March 2012 (courtesy UCAR)

A red flag warning was issued for nearly all of eastern Colorado. And with good reason! A grass fire started in Yuma County, CO (which borders Nebraska and Kansas) in the early afternoon, and quickly grew out of control. The media dubbed it the Heartstrong Fire. An area 14 x 16 miles had to be evacuated, although only 2400 acres actually burned. The smoke plume was easily visible from the Goodland, KS, National Weather Service radar. Two homes were destroyed, and three firefighters were injured battling the blaze.

Radar image of smoke from the Heartstrong Fire, 21:17 UTC 18 March 2012

Radar image of smoke from the Heartstrong Fire seen by the Goodland, KS, NWS radar, 21:17 UTC 18 March 2012 (courtesy UCAR)

"True Color" image of the Heartstrong Fire, 19:34 UTC 18 March 2012

"True Color" image (RGB composite of VIIRS channels M3, M4 and M5) of the Heartstrong Fire, 19:34 UTC 18 March 2012

Even though cirrus clouds covered the area (as seen in the true color image above), VIIRS observed the fire in its two 3.7 µm channels. The VIIRS images shown here, from 19:34 UTC, were taken roughly 20 minutes after the fire was first reported. The moderate resolution band M-12 (centered at 3.7 µm) identifies a hot spot (which shows up as black in the image below) that is approximately 6 pixels by 3 pixels. With ~750 m resolution at nadir in this band, that corresponds to a total area of 10.2 km² of pixels that contain a fire signal.

Image of the Heartstrong Fire from VIIRS channel M-12, 19:34 UTC 18 March 2012

Image of the Heartstrong Fire from VIIRS channel M-12, 19:34 UTC 18 March 2012

The high resolution imagery band I-4 (centered at 3.74 µm) also identifies the hot spot. In this case it is approximately 11 pixels by 5 pixels in size. At ~375 m resolution at nadir, this corresponds to an area of 7.7 km² of pixels that contain a fire signal.

Image of the Heartstrong Fire  from VIIRS channel I-4, 19:34 UTC 18 March 2012

Image of the Heartstrong Fire (indicated by the red arrow) from VIIRS channel I-4, 19:34 UTC 18 March 2012

Thus, the difference in resolution between these two channels leads to a difference in the apparent size of the hot spot as seen by satellites. However, it should be noted that this apparent size is only an estimate of the size of the hot spot visible in the satellite image, not the actual size of the fire. Fires move in narrow flame fronts that cover only a small percentage of the pixel area. From a firefighting perspective, detecting which pixels actually contain fire and where the actual burning occurs within those pixels are two different things.

Of additional interest is the difference in observed brightness temperatures between these two channels. The warmest pixel in M-12 was 327 K, while the warmest pixel in I-4 was 342 K. As the observed brightness temperature is related to the fraction of each pixel covered by fire, the higher resolution images produce higher brightness temperatures in the hot spot.

This means that, to a human observer, the hot spot appears larger in the M-band image, while, from an automated algorithm point-of-view, the I-band image has a larger number of pixels within the hot spot, and higher brightness temperatures. The difference in the appearance of the hot spot between these channels is more clearly seen in the figure below. Be sure to click on the image, and then look for the “1700×702” link above the image title and click on that to see the comparison in its highest quality.

Comparison between the I-4 and M-12 views of the Heartstrong Fire

Comparison between the I-4 and M-12 views of the Heartstrong Fire. The previous I-4 and M-12 images (taken at 19:34 UTC, 18 March 2012) have been zoomed in for additional clarity.

As an additional note, band M-13 (centered at 4.05 µm) is the primary band used in active fire detection. This band was designed specifically to measure the radiative signal of hot spots without sensor saturation. The M-13 image of the fire is shown below.

Image of the Heartstrong Fire taken by VIIRS band M-13, 19:34 UTC 18 March 2012

Image of the Heartstrong Fire from VIIRS channel M-13, 19:34 UTC 18 March 2012

There is a dedicated team of researchers actively exploring fire detection from VIIRS. You can learn more about fire detection and the status of their current fire detection products by visiting viirsfire.geog.umd.edu.