This may be the last update on my Boise Basin project. I submitted the 1987 vegetation classification to my cooperator this morning. To make this product, I first created a NDVI change product and found areas of significant change. In areas that didn't fit the "significant" threshold, I used the 2002 veg layer. In areas of significant change, I burned in the 1987 classification. It was an interesting way to do it and probably eliminates a lot of noise you'd otherwise encounter.This project, while frustrating because of having to start over a few times, was a good experience in stats. I got to create regression equations for an image rescaling effort and I also did my first accuracy assessment where I computed map accuracy and kappa.
3 comments:
What's kappa?
kappa - a measure of agreement between two observers taking into account agreement that could occur by chance (expected agreement).
It is used when comparing observations from two sources regarding the same object. In my case, I was comparing the vegetation that my map said existed vs. the vegetation the world actually has. I computed a kappa value that told me the general agreement of the two (in a sense map accuracy) as a percentage of what I'd get by random chance alone.
The last line should actually read: as a percentage above what I'd get by random chance alone.
Post a Comment