There’s an ongoing serious situation in Australia: the bush in New South Wales and Queensland is on fire.
Here’s a look at what the Advanced Himawari Imager (AHI) on Himawari-8 saw on 8 November 2019: click here.
What you see in that loop is the “Natural Fire Color RGB” (known to American forecasters as the “Day Land Cloud Fire RGB”) on the left (link to PDF description here), and the “Fire Temperature RGB” on the right (link to PDF description here). These are precisely the products we debuted on this blog seven years ago when we first looked at fires in Australia. Except, now there is a difference: the “Natural Fire Color RGB” is now made with the 3.7 µm band as the red component (replacing the 2.25 µm band I used originally), since the 3.7 µm channel is even better at detecting fires. This also means that we can produce the VIIRS version using “I-band” resolution (375 m). AHI, used in the loop I linked to above, has 2 km resolution* for the mid- and shortwave infrared (IR) bands.
Along the coast, near the northern edge of the images is Brisbane, the third largest city in Australia. Near the southern edge of those images is Sydney, the largest city in Australia. As you can see from Himawari-8, much of the area between the two is on fire. And, this is not the “Outback” where very few people live. This region contains some of the highest population density in Australia, and it’s also prime habitat for koalas, which don’t live anywhere outside of eastern Australia (except in zoos).
It’s no secret that resolution plays in big role in fire detection from satellites. We’ve covered this many times before. But, to hammer the point home (bit of American slang), here’s the resolution difference between VIIRS and AHI in full view from 3:50 UTC on 7 November 2019:
As always, click on each image to bring up the full resolution version. If you just look at the elephant-thumbnail-sized images above without clicking on them, you might get the impression that fires are easier to spot with AHI than with VIIRS. That’s because AHI makes it appear that the entire 2km-wide pixel* is full of fire, when a fire typically only fills a very small percentage of the total area of the pixel. With 375 m resolution**, VIIRS more accurately pinpoints the locations of fire activity. Although, it should be noted that even this is still a larger scale than most fire fronts. To be really accurate, you need something with the resolution of Landsat’s OLI, or a similar radiometer attached to an aircraft – except these high-resolution instruments don’t provide full global coverage multiple times daily like VIIRS, or hemispheric coverage every 10 minutes like AHI. (*On AHI [and ABI and AMI] pixels may be approximated as square-shaped solid angles that are projected onto the curved surface of the Earth from a point roughly 36,000 km above the Equator. 2 km is the width of an IR pixel at the sub-satellite point [on the Equator], where the resolutions are the highest. **VIIRS pixel resolutions vary across the swath by a factor of 2 between nadir and edge of scan, as we shall see. 375 m is the nadir value.)
For completeness, we can do the same comparison with the Fire Temperature RGB:
This time, we’re comparing 2 km resolution (AHI) against 750 m resolution (VIIRS), so the differences aren’t as stark. But, this is a good opportunity to remind everyone that the Fire Temperature RGB provides information on fire intensity, while the Natural Fire Color (Day Land Cloud Fire) RGB provides information on fire detections (plus smoke and burn scars), and should be used more as a “fire mask”.
There’s another resolution difference that is easy to see from these fires, and it can be quite significant. I first noticed it when looking at this animation I made of the VIIRS Fire Temperature RGB from 1-11 November 2019:
You have to click on the animation to get it to play.
Did you notice the same thing I did? You probably noticed the explosive growth of the fires from 7-9 November, but that’s not what I’m talking about. (Hint: Pay close attention to the nighttime images.) At night, without any sunlight present, you lose information on clouds and the background land surface, and only the fires are visible (unless they are obscured by clouds). That’s where today’s feature of interest resides. I’ll zoom in on some of the fires from 5 November 2019 to make it easier to see:
The image from 14:01 UTC comes from S-NPP, while the image from 14:52 comes from NOAA-20. Is NOAA-20 better than S-NPP at detecting the fires? Well, the reverse happened two nights later:
This time, the fires appear hotter (brighter) in the 15:03 UTC image, which came from S-NPP. The 14:12 UTC image came from NOAA-20. Here’s a sequence of three images from 10 November where the NOAA-20 image is sandwiched by two S-NPP images:
So, why do the fires appear brighter in some images and not others? It’s possible that the fires are becoming more active in the middle image (due to an increase in winds, for example), but it’s more likely that you are seeing the direct result of resolution differences between the various overpasses. “But, I thought both VIIRS instruments had the same resolution,” you might say as though it were a question. And that statement would suggest that you forgot about the “bowtie-effect”. (Not the effect that has anything to do with diamonds, but the effect I wrote a whole chapter about here [PDF].) If you read the **above you would already know that the resolution of VIIRS degrades by a factor of two between nadir and the edge of scan. And, if you didn’t already know, NOAA-20 and S-NPP are positioned in space a half-orbit apart. This means that, in the time it takes between a NOAA-20 overpass and a S-NPP overpass, the Earth has rotated by half the width of the swath (approximately). So, when one VIIRS instrument views something at nadir, it will be close to the edge of scan on the other satellite (and have more coarse resolution as a result).
So, in the last animation, the first image (14:05 UTC) is S-NPP viewing the fires from the east near the edge of scan, the middle image (14:56 UTC) is NOAA-20 viewing the fires near nadir, and the third image is S-NPP viewing the fires from the west – even closer to the edge of scan. (Plus, the terrain is sloping away from S-NPP in the last image as well.)
Those factors contribute to the changing appearance of the fires. They also highlight the value of having two VIIRS instruments in space: if one satellite doesn’t get a good look at a fire, the other one likely will.
By the way, these fires have been producing a lot of smoke. Here is a loop of VIIRS True Color images from 6-11 November:
And the view from the ground is even more apocalyptic:
Australia so far has lost 970,000 hectares due to current bushfires.
“We are no longer the sunburnt country, we are the country on fire 🔥”#AUSTRALIAFIRES pic.twitter.com/NBQBoro1OA
— kel.lunn (@LunnKel) November 12, 2019
Australian bush fires need more media coverage and MORE people spreading the news about it. these fires are becoming much worse than the amazon, yet the rest of the world stays quiet. Please use your voice and help us spread it around!#Australiabushfires #AUSTRALIAFIRES#btsarmy pic.twitter.com/olKtVkKDoB
— 𝖆𝖚𝖘𝖘𝖎𝖊 𝖇𝖑𝖎𝖓𝖐 & 𝖆𝖗𝖒𝖞 (@bangtanxzia) November 12, 2019
Hi Curtis, this is a very interesting post. I think the new recipe of “Natural Fire Color RGB” has also been applied to the https://rammb-slider.cira.colostate.edu/? I tried to replicate this image using AHI data but cannot get the same image. My configurations are:
1. Red: AHI band-7 (reflectance portion only, 0~100, gamma: 1);
2. Green: AHI band-4 (reflectance, SZA-corrected, 0~100, gamma: 1);
3. Blue: AHI band-3 (reflectance, SZA-corrected, 0~100, gamma: 1).
I’ll be very much appreciated if you can point out the difference(s) between my configuration and yours. Thanks!
The differences you are seeing are likely due to the fact that we scale the 3.9 μm bands (e.g. AHI Band 7) based on brightness temperature, not reflectance. We also apply a gamma correction to the red component. Also, note that VIIRS reflectance values are SZA-corrected in the data files. But, we do not apply the SZA correction on AHI or ABI data displayed on SLIDER.
For the full recipe, please refer to the RGB Quick Guide posted on the VISIT training website here: http://rammb.cira.colostate.edu/training/visit/quick_guides/VIIRS_Day_Land_Cloud_Fire_RGB_Quick_Guide_10182018.pdf
Even though this document is specific to VIIRS, we use the same recipe for AHI.
Thanks for your reply, Curtis. I have tried the recipe and couldn’t reproduce the image showing in the SLIDER (neither ABI nor AHI).
For AHI, I applied the following recipe:
1. Red: AHI band-7 (BT, 273~333, gamma: 0.4);
2. Green: AHI band-4 (reflectance, no SZA-corrected, 0~100, gamma: 1);
3. Blue: AHI band-3 (reflectance, no SZA-corrected, 0~100, gamma: 1).
The image shown in the SLIDER is much brighter, greener over land, and bluer over the ocean.
For ABI (GOES-17), I applied the following recipe:
1. Red: ABI band-7 (BT, 273~333, gamma: 0.4);
2. Green: ABI band-3 (reflectance, no SZA-corrected, 0~100, gamma: 1);
3. Blue: ABI band-2 (reflectance, no SZA-corrected, 0~100, gamma: 1).
My image is much redder over CA.
Nevertheless, I can reproduce the fire temperature image for both instruments following the recipe http://rammb.cira.colostate.edu/training/visit/quick_guides/Fire_Temperature_RGB.pdf.
Additionally, I found the natural fire image looks different in full-disk projection comparing to CONUS projection in the SLIDER (GOES-17). Any clue for that? Thanks in advance!