I have a complex classification problem to solve:

  1. I have multiple shapefiles containing the polygons of all the parcels of different kinds e.g. a shapefile with all polygons with maize, another with rice etc

  2. I have a time series multispectral imagery of those parcels and indexes e.g.NDVI calculated

  3. I have made a shapefile with all the training polygons of the classes needed to describe the parcels according to their description: 10 polygons maize, 10 rice etc.

  4. I have constructed a signature file with the above training polygons merged accordingly to the classes, containing all the different layers' mean values: Layers-> All the NDVI rasters, R,G,B layers from all times

Now with these components -or with others if suggested so-, I need to classify all the parcels to a class according to the signature. For the sake of example lets assume I have 10000 parcels, 20 classes/cultivation types and 30 layers. As a result the Signature file has "size" 20x30.

I need a supervised classification approach, like Maximum Likelihood.

What can I do to assign each parcel to a class?

  • Is your problem simply how to assign a value to a parcel, or actually running the classification? Probably oversimplifying, but it sounds like you've got your classification set up, so that would just leave running a Zonal Stats to identify the majority in each parcel. – phloem Dec 5 '16 at 23:25
  • Thank you for the comment @phloem, however i am afraid it is more complex than that. The classification is needed to assign a value to each parcel. The issue is how. Classification is not Zonal Statistics (although maybe i did not understand your suggestion). If it is supervised for example it needs a training set of polygons etc. Much more complicated than a mean or variance assignment. In my case we have a mathematical problem with 30 dimensions – ODstuck Dec 6 '16 at 8:23
  • My suggestion is to run your classification to classify each pixel, perhaps using Maximum Likelihood as you indicate. Then, use your parcels as zones to generalize those classified pixels into a parcel-level classification. It's just not clear to me which part you're stuck on - you seem to describe exactly how to do the classification. – phloem Dec 6 '16 at 17:23
  • Thank you once more @phloem for the input. I have been successful in this particular endeavor but without ArcGIS. I was asking in particular how to do it with tools provided by Arc. It seems it is not possible. When i have some more time i will close the post with some hints on how one can do that through QGIS nice n easy. – ODstuck Dec 7 '16 at 1:53
  • You might want to edit your question to that effect. Maximum Likelihood: desktop.arcgis.com/en/arcmap/latest/tools/… – phloem Dec 8 '16 at 18:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.