comparing rasters, cells that have changed value

Question

I am attempting to compare two rasters representing a species distribution model for now and the future.

The rasters have the exact same extent and cell sizes.

here is when is gets tricky, they are multi-band rasters, so to create an attribute table I must first convert them to a binary raster (maybe there is another way around this?). Doing this I lose a lot of the information in the rasters.

I want a quick way to compare how many cells have changed value (using python probably) as I have many sets of these to compare.

Answer 1

Try this:

• Use the RasterToNumpyArray function to convert your multi-band raster to n-dimensional array with dimensions (# of bands, rows,columns)
• Use the not_equal function from the numpy module to undertake a cell by cell comparison for each band.
• Use the unique function from the numpy module and list comprehensions to find total number of cells that have changed/not changed.

Example...

Using simple 3 band rasters (species_n and species_f), the sample code below

import arcpy
import numpy as np

sn = arcpy.RasterToNumPyArray(r'D:\species.gdb\species_n',nodata_to_value = -999.999) #species now
print "\nSpecies Now"
print "Data Type: {0}".format(sn.dtype)
print "Bands, Rows, Columns: {0}\n".format(sn.shape)


sf = arcpy.RasterToNumPyArray(r'D:\species.gdb\species_f',nodata_to_value = -999.999) #species future
print "\nSpecies Future"
print "Data Type: {0}".format(sn.dtype)
print "Bands, Rows, Columns: {0}\n".format(sn.shape)


#Band comparison
#cell by cell comparison - returns true if cell values not equal, else false.
band_comparison = np.not_equal(sn[0],sf[0])

u, indices = np.unique(band_comparison, return_inverse = True)

#Count #number of changed cells
#this step can be replaced with a far more elegant approach
cell_counts = zip(u,[u[indices].tolist().count(x) for x in u])
print "\nChanged Cells:\n{0}".format(cell_counts)

produces the results below:

Note: in this example, band 1 in species_n is exactly the same as band_1 in species_f. The results above show that there are no changed cells, as expected.

species_f(bands 2 and 3) are multiples of species_n(bands 2 and 3)

Band 2 comparison:

band_comparison = np.not_equal(sn[1],sf[1])

produces:

10 cells have not changed (in this example, these are just no data cells) 26 cells have changed.