bleutner/RStoolbox

Allow fitting models on raster stacks of differing extents

ailich opened this issue · 3 comments

ailich commented

Sometimes I'd like to fit a single model using superClass or unsuperClass on several disjointed areas where I have raster data of the same variables. The way I've done this is the past for supervised classification is to sample each raster at the training points to create a single dataframe, then fit the model in caret, and the use raster::predict on each raster individually. It would be nice however if this could be done directly in RSToolbox and perhaps with the upgrade to terra you could include SpatRasterDatasets as a supported input which is essentially like having a list of rasters/raster stacks that can be of differing extents.

bleutner commented

makes sense, yes. I've been reworking large parts of superClass and will consider this suggestion. Also SpatRasterDatasets seems like a good input structure I wasn't aware of.

  • 👍1
ailich commented

Thanks!

ailich commented

I was actually mistaken about the differing extents. I tried using a SpatRasterDataset instead of a list of rasters and got "Error: [sds] extents do not match ()".