OBIA introduction
OBIA stands for Object-based Image Analysis, and the steps are pretty straightforward. First, we find/acquire an aerial or satellite image of our study area. Here's an example from near Babb, MT- just outside of the Many Glacier entrance to Glacier National Park. We are going to try to map our target species, juniper, here. Thinking about homebrewing some gin? Take a walk near Duck Lake (maybe take your flyrod along)!
Next we create "objects" from groups of similar pixels- the segmentation step. Now, depending on the scale you're interested in, this can identify large agricultural fields, chunks of similar looking forest, or if you get way down into the nitty gritty- single plants and trees. That's what we're after.
After the initial shock that you've just split up your study area into 100 million groups of pixels, you can start to see why:
It looks a little bit like I've done a fantastic job of painting those details, but we've just summarized what's going on within each of those polygons ("groups of pixels"). So now for each of those ~100 million polygons, we have 33 different bits of information to help us decipher between them and figure out which are the ones we're interested in- thanks computers!. Here are a couple close ups:
If I had the time, or if all of the juniper in the study area was located on the shores of the lakes and rivers I was hiking by or casting into, I could map these by hand and maybe by my 90th birthday I'd have some useful information. Alternatively, I'd love to watch my children grow up and again let the computer do some work. So far, testing on one tile of our study area (~approximately .06 million polygons) using a random forest model is promising- about 90% overall accuracy.
Not perfect, but if it saves me a few decades and teaches us more about mapping wildlife foods from space, well worth it!
Until next time- Nate
P.S. If anyone has a wild gin recipe, these next few days are going to be awfully hot...
0 comments