The last two posts covered flooding. Now, a month later, we are back to covering last year’s most common topic: wildfires. This time, we’ll make a game out of it. Keep in mind that, for many operational fire weather forecasters, this isn’t a game – it is information that could prove useful in saving lives or homes from destruction. If you have read the earlier posts on fire detection and haven’t forgotten what you’ve been told (here’s a good one to go back and read), this should be easy for you.
The following images are the unmapped data from three consecutive VIIRS granules over the Southwest U.S., starting at 20:36 UTC 11 June 2013. The “raw” data has been processed to produce the “True Color”, “Natural Fire Color” and “Fire Temperature” RGB composites. Plus, the brightness temperature data from channel M-13 (4.0 µm) has a color table applied to it to aid in fire detection. Satellite channels near 4 µm are the “industry standard”, so to speak, for detecting fires as they are highly sensitive to sub-pixel heat sources like fires. The “Natural Fire Color” and “Fire Temperature” composites are RGB composites developed just for VIIRS that both had their debut on this very blog.
The question is: how many fires can you see? Remember, you have to allocate resources (firefighters, helicopters, planes, etc.) based on your assessment. The media is hounding you for all the latest statistics on each blaze and they can’t wait until the 5:00 briefing. They need the scoop now to get higher ratings. Plus, the crew is loading fire retardant on the plane as you read this. Where should the pilot fly to? Everyone is counting on you! (Of course, you would never have just satellite data by itself in a real-life scenario – but, do you want to play this game, or just think of flaws?)
I’ll give you a hint: You won’t see any fires unless you view each image at full resolution. Click on the image, then on the “3200×2304″ link below the banner to see the full resolution version. (You could even open each full resolution image in a new tab, and click between the tabs for easy comparison, assuming you’re not using some archaic version of Internet Explorer or another old browser that doesn’t allow tabs. When you would click on the “3200×2304″ link, instead right-click and select “Open in New Tab”. Another option would be to save the images and open them in an image viewing software program that will allow you to zoom in more than 100% but, that is starting to sound like a lot of work and I’m not sure I want to play this game anymore. It’s too complicated. By the way, if that’s the way you feel, don’t become the manager of a fire incident team.)
I’ll give you another hint: Many of the hot spots that indicate fires are only 1-2 pixels in size. Be prepared to look for needles in the haystack, and make sure you have your reading glasses on, if you need them.
So, did you see them all? You should have identified 12 fires. Did you find less than 12? Some of them are hard (or impossible) to see in some of the images. Did you find more than 12? The color scale used on the M-13 image led to false alarms, so you can be forgiven if that’s what caused you count too many.
This example shows some of the complicating factors when trying to identify fires from satellites. It also shows why fire managers never rely on satellite data alone. Now, having said that, VIIRS can and does provide useful information on fires.
First, here’s the answer (link goes to PDF) from the National Interagency Fire Center. They identified 15 active “large incident” fires on 12 June 2013. (They update their maps once per day, so all the fires that started on 11 June make it on the 12 June map.) But, there are differences between their map and what VIIRS saw.
First, the Mail Trail fire (#5 in the PDF) is outside the domain of these three VIIRS granules, so you couldn’t have found that in these images. Fires #3, 4 and 7 (Healy, Porcupine and Ferguson) are obscured by clouds, and/or were mostly contained, transitioning from active to inactive. The Tres Lagunas Fire (#13) started back in May and is undergoing mop up activities. The hot spots from that fire (if there are any left) aren’t visible in the images, but the burn scar is. That leaves the Stockade (#1), Crowley Creek (#2), Hathaway (#6), Fourmile (#8), Silver (#9), Thompson Ridge (#10), Jaroso (#11), Big Meadows (#12), Royal Gorge (#14), and Black Forest (#15) – 10 fires which are all visible in the VIIRS images. Plus, VIIRS saw two more fires that are not included on that list: one in southern California (near the Salton Sea) that I couldn’t find any information on, plus a pellet plant fire in Show Low, Arizona. (Small fires in towns are usually outside the scope of the National Interagency Fire Center, so they don’t bother to list those.)
I would argue that the “Fire Temperature” composite worked the best at identifying each of these fires, but all 4 images have their uses. Here’s the Fire Temperature RGB image with the visible fires identified:
Answer honestly. Which fires did you see, and which fires did you miss?
The Fire Temperature RGB takes advantage of the VIIRS channels in the portion of the electromagnetic spectrum ranging from the near-infrared (NIR) to the shortwave infrared (SWIR). The blue component is M-10 (1.61 µm), the green component is M-11 (2.25 µm) and the red component is M-12 (3.7 µm). As wavelength increases over this range, the contribution of the Earth’s emission sources increases and the contribution from the sun decreases. As a result, only the hottest hot spots show up in M-10, as they have to be seen over the large signal of radiation from the sun reflecting off the Earth’s surface. In M-12 (as in M-13), hot spots from fires produce more radiation at that wavelength than the amount of reflected solar radiation. M-11 is somewhere in the middle. That means relatively cool (e.g. smoldering) or small fires only show up in M-12, which makes those pixels appear red. Pixels containing fires hot enough or large enough to show up in M-11 will take on an orange to yellow color. Pixels containing fires hot enough or large enough to show up in all three channels will appear white.
You have to be careful, though, as some pixels in the Fire Temperature RGB appear red, even though there aren’t any fires in them. A few of these pixels show up red in the M-13 image, and are labelled as “not a fire/false alarm”:
According to the color table used, any pixel with a brightness temperature above 340 K (67 °C) will be colored, with colors ranging from red to orange to pale yellow as temperature increases. Now, look at that area in the True Color image (or on Google Maps):
That area is very dark – almost black – volcanic rock with very little vegetation that has been baking in the sun all day. It has managed to acquire a brightness temperature that is higher than some of the active fire pixels. The Crowley Creek fire doesn’t show up as red in the M-13 image (the Stockade fire is the one with the yellow and orange pixels) and the Fourmile fire is barely visible. (It has two pixels warmer than 340 K, even though 10 pixels appear red in the Fire Temperature RGB). The color scale in the M-13 image could be applied to a different temperature range, but you’ll always have that trade-off: have the colors start at too high a temperature, and you’ll miss some fires; have the colors start at too low a temperature, and you’ll increase the false alarms.
The True Color image should have helped you identify 5 of the fires. The smoke plumes that show up are a dead giveaway. I’m talking about the Big Meadows, Royal Gorge, Jaroso, Thompson Ridge and Silver fires, of course. There may be smoke with the Hathaway fire, but it would be mixed in with the cirrus clouds and hard to see. Not all fires produce a lot of smoke, though. Having information on the ones that do aids in issuing air quality alerts, among other benefits.
Lastly, the Natural Fire Color image highlights most (but not all) of the fires. Look for the red pixels:
The Natural Fire Color doesn’t show active hot spots at Crowley Creek, and the Hathaway and Fourmile fires are difficult to see, because they aren’t quite hot enough. (Generally speaking, any fire that shows up red in the Fire Temperature RGB is too cold to show up as red in the Natural Fire Color.) But, this composite has the advantage of showing burn scars in addition to the active fires. Burn scars appear dark brown. The Fourmile and Crowley Creek burn scars are visible. Plus, burn scars from last year’s fires still show up: The Whitewater-Baldy, High Park and Waldo Canyon scars are identified. The Tres Lagunas was mentioned above, and it’s burn scar is visible. If you look closely, I’m sure you could find more burn scars from last year’s long fire season.
Here are all four images, zoomed in on each fire at 800%, combined into an animation to highlight how each fire appears in each image:
For some reason, you have to click to the full resolution version of the image before the animation will display.
Hopefully, this exercise is useful in demonstrating the complications that arise when trying to detect fires from satellites in space, as well as the strengths and weaknesses of some of the various methods VIIRS has at it’s disposal to aid the fire weather community.