We’re using TileMill more and more as an analytical tool, especially as we explore infrared bands. Infrared is not visible to the human eye, but it carries a wealth of environmental information, and many satellites capture it as well as – or even instead of – visible light. Since I’ve been working with a lot of visible data from the MODIS satellites for the Cloudless Atlas lately, I got curious about what could be seen in the infrared bands once they too were seamless and cloudless.

Here’s a look at the aftermath of the 2010 Russian forest fires using satellite IR data. You can scroll around and look at 2011’s EVI2 (a measurement of vegetation – read on for details), and to check whether a dark feature is a fire scar, you can pull across the red-green overlay showing year-on-year change. Or, with the overlay on, you can get a regional look at where vegetation was in midsummer 2011 as compared with a year earlier:

Update: The demo that supported this post is no longer available.

The north and east of this view is covered in taiga – dense, cold-hardy forest that shows up brightly in EVI2 because of its heavy foliage. Across the middle and west is mixed forest with large areas cleared for cities and farms. If you zoom out, you might notice that this intermediate zone is mostly green, suggesting that it consistently improved from 2010 to 2011 – and you can clearly see the bright, taiga-covered southern edge of the Ural mountains interrupting it in the east. The southern edge of our view is mostly steppe, much of it converted to grain fields, verging into scrubland and desert (nearly black in EVI2) in Kazakhstan.

Using TileMill for analysis

I started with a raster layer of EVI2, an index that estimates plant density from the ratio of red and near-infrared light reflected to space. (Growing plants absorb red light for photosynthesis, while water in the leaves reflects infrared. When they die, their chlorophyll breaks down and they dry out, and the ratio tilts the other way.) I used this MODIS subset from NASA, pulling red data from the visible image and near-infrared data from channel 2 of the 7-2-1 false-color image. This layer used images from June and July 2011, around the time of peak growth in the year following the fires:

Western Russia’s EVI2, midsummer 2010. Which features are fire scars?

When I looked at the EVI2 alone, I saw marks that might be fire scars, but it was hard to say for sure. The dark patches could be naturally barren areas like dry lakebeds or scrublands – or even urban areas. To find out what was normal and what wasn’t, I needed to look at change. I made a second composite covering the same June-July period in 2010, a few weeks before the worst of the fires, and created a difference layer – a raster with values showing how each pixel changed between 2010 and 2011. In the difference layer, neutral gray meant EVI2 was unchanged year-on-year, dark gray meant it decreased in 2011, and light gray meant it increased. It was hard to interpret at a glance, so I used TileMill’s colorizing tools to show decreased growth as red and increased growth as green:

Year-on-year change in EVI2 (same view as above).