Your raster data is large, oddly shaped, and has areas where data is not available. You have a nifty function that describes or analyzes the data - but how do you apply this function to a sample of the raster in a way that a.) is faithful to the scope of the full dataset b.) accounts for spatial variability and bias c.) is not confounded by "no data", and d.) is reproducible
Generate random subsamples of a raster dataset that in ensemble ensure a robust spatial sampling of the raster and meet the above conditions.
Geotiff raster data.