Tuesday, September 27, 2016

Disturbance Finder Tool

So I mentioned earlier that I was working in Google Earth Engine on some remote sensing stuff, and that my favorite thing about GEE is that it's so easy to share what you're working on with others, for free. I wanted to share some code that I wrote there but I realize that most people can't access GEE without an account (so get an account!*).

The tool I made is specific to looking for burned areas to potential visit on the ground. This is especially hard to do in remote places, because we're usually constrained to the roads network. Sometimes a boat or a float plane can get us to interesting places, and helicopters are even more flexible, but these options are generally quite expensive. So I wanted a tool that could help me look for burned areas close to roads. In North America, it's often fairly easy to find burned areas generally, as these are regularly mapped for the boreal forests in Alaska and Canada. But now I need to look for burned areas in Russia**, and the datasets aren't so comprehensive or well-maintained.

So I decided to use indices from the Landsat data record (right now this only uses Landsat 5, but that's a good 15 years or more of data for Siberia) to look for burned areas. The first step in the tool is to specify the dates you want to look for burned areas in. This is generally only the year that the fire burned or the one immediately following it. You put in the start and end dates for the growing season and center the map where you want to create the image. Then you can poke around looking for burned areas in the image.

The problem you will likely run into is that there are non-burned areas that look like they burned (such as water and cloud shadows, which both appear dark and irregularly-shaped). So you can click once inside the "burn" and once outside it, which will generate a line graph showing you the difference between the burned and unburned areas from the start to the end of the Landsat record. That makes it easier to separate burned areas from other dark objects, since the "signal" of a burn over time is a sudden decrease in reflectance compared to neighboring areas, and (usually) a subsequent recovery over time. If there's no such signal in the data record, it's probably not a fire (or not such a serious one). Happy hunting!

* I'm not trying to shill for Google here, really.

** I created this tool when I was thinking about Siberian forest fires specifically, but it would probably work with little or no tweaking to identify burned areas in other forests or possibly non-forest ecosystems, too. It might also work for finding other types of disturbance that affect forest reflectance properties, but I haven't tested it for that yet.

1 comment :