I want to do time series analysis and find change in NDVI. I have rasters where each pixel is represented by some value at a time. Also I have same thing but represented by points.

And I would like to generate new map, with changes. Now Im looking at the transformation matrix and also on this post: What is Change Detection and how I can perform such analysis with open source tools?

Am I thinking right about it? I would really like try to use something in Rcode (How to represent trend over time?), but Ive just started to learn it.

Qgis, landsat 30x30m

  • 1
    Even though the scenes more-or-less overlay they will be slightly different, you will need to georeference one to the other by locations that are clearly distinguishable in both to the subpixel to do change detection, otherwise slight differences in pixel alignment will show 'false change'. If it is within your scope resample both to ~50m to minimize cell alignment problems after georeference. Jul 2, 2015 at 22:27
  • Raster files are downloaded as product 1 from earth explorer so i think they should be georeferenced
    – ABCD
    Jul 2, 2015 at 22:50
  • Yes, they are, but both to their own accuracy. They are not likely to perfectly align to the subpixel and any slight variations in where the cell boundaries fall will give spurious results. They need to be matched exactly to each other. Then solar reflectance, top of atmosphere... Jul 2, 2015 at 22:58
  • OK, I will try resampling them to 50m. Than what? raster1 - raster2, will not give me answers, or?
    – ABCD
    Jul 2, 2015 at 23:07
  • Is this what you are after?: gis.stackexchange.com/q/33402/8104
    – Aaron
    Jul 2, 2015 at 23:10