0

I am trying to determine the variables that predict forest fire occurrence in a region in Chile.

Firstly, I determined burned areas using LANDSAT data and the dNBR-index. I ultimately made a raster with 30m forest pixels with a value of 1 (burned) or 0 (not burned). I wanted to do a random sampling with minimum distance to reduce the autocorrelation between burned pixels. My problem is that I do not know to what distance my burned LANDSAT pixels are autocorrelated. I tried to make a semivariogram in R but this gave errors, probably due to the binary variable or because there are pixels with no-data values. I made a point shapefile of my raster for which every point has the same binary value of the pixel it is covering. However, if I make a semivariogram, I assume that the autocorrelation of pixels that were not burned isconsidered as well and this is not what I want.

Anyone an idea to investigate the autocorrelation of points or pixels with a specified value (1 in my case, which represents fire).

Improve this question
2
  • Lets focus on the errors you got trying to create a semivariogram in R. Could you post your data (or a picture of it) and the code you used, and the error you got? As a reproducible example? There's no reason why in principle you can't get a variogram from a raster even with missing data.
    – Spacedman
    Commented Apr 8, 2019 at 11:57
  • Maybe you need to rethink what your looking for here. It sounds like you want describe the clustering of observed presence of fire. You could look at nearest neighbour analysis for a start and then do a bit of research on other point pattern methods that might be suitable.
    – Liam G
    Commented Oct 11, 2019 at 22:43

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.