Generate a custom vegetation-index (VI)

Question

I have two land cover types that I need to differentiate for which I want to use a vegetation index. I have data from 4 sensors in the green, red, red-edge and near-infrared spectrum. The land covers mainly differ in the green and red spectrum.
Similar to the NDVI formula, (NIR-RED) / (NIR+RED), I would like to create a new index, that specifically targets the difference in my land covers. I found a lot of literature on different vegetation indices, but couldn't find any specific guidelines for designing my own.
What would be an appropriate formula to differentiate the following data:

radouxju · Answer

Your graph shows a good separability between the 2 classes in green and red, but the separability is less obvious when you look at the differences. If you look at your graph, you can indeed see that the difference between red and green for the class on the left could be either positive or negative (the reflectance values are not clearly separated like between red and NIR). Therefore a difference index would be useless, same with a ratio index. With your 2 clases, the Mahalanobis distance to one of the classes would also do.
Therefore I would rather suggest to use a classification algorithm instead of building a new index (the probability to belong to the class is then the best index).
For a quick and dirty index, use 0.06-Red because this is (visually) the location of the minimum error boundary, and it will increase when you have more vegetation. This will be useful to visualise your 2 classes, but a very poor vegetation index.

Generate a custom vegetation-index (VI)

One Answer

Add your own answers!

Ask a Question