I have a geodatabase that contains a large number of raster datasets - there are two rasters for each country in the world, and they are very large, in some cases up to 3GB per raster dataset.

I need to convert these to NumPy arrays so as to run an analytical program that I wrote on them, but I'm hitting a memory error when my script tries to convert the first raster dataset. The script creates a list of all of the rasters in the dataset (rasterList), then grabs the country codes at the end of each raster dataset name and populates a new list of all the country codes. I then delete rasterList, and iterate through the country codes, converting rasters to either 'GLUR_'+code or 'Pop00_'+code. I delete each raster from memory after it is converted, but the program doesn't even get that far...

The code is:

import arcpy, os
import numpy as np
from arcpy import env
env.workspace = "J:/Data/CISC.gdb"

filepath = "J:/LIVE/"
rasterList = arcpy.ListRasters("G*")
codeList = []

for raster in rasterList:

del rasterList

for code in codeList:
    print code
    inputGLUR = arcpy.Raster("J:/Data/CISC.gdb/GLUR_"+code)
    lowerLeft = arcpy.Point(inputGLUR.extent.XMin, inputGLUR.extent.YMin)
    cellSize = inputGLUR.meanCellWidth
    print "Converting GLUR_"+code, "to NumPy array"
    arr = arcpy.RasterToNumPyArray(inputGLUR, nodata_to_value=0)
    numpy.save("GLUR_"+code+".npy", arr)
    print "GLUR Converted"
    del inputGLUR
    del arr

    inputPop = arcpy.Raster("J:/Data/CISC.gdb/Pop00_"+code)
    lowerLeft = arcpy.Point(inputPop.extent.XMin, inputPop.extent.YMin)
    cellSize = inputPop.meanCellWidth
    print "Converting Pop00_"+code, "to NumPy array"
    arr = arcpy.RasterToNumPyArray(inputPop, nodata_to_value=0)
    numpy.save("Pop00_"+code, arr)
    print "Population array converted"
    del inputPop
    del arr

and yields this error:

Traceback (most recent call last):
  File "J:\PYTHON\RasterToNumPy.py", line 21, in <module>
    arr = arcpy.RasterToNumPyArray(inputGLUR, nodata_to_value=0)
  File "C:\Program Files (x86)\ArcGIS\Desktop10.3\ArcPy\arcpy\__init__.py", line 2244, in RasterToNumPyArray
    return _RasterToNumPyArray(*args, **kwargs)

This is on a computer with an i7 processor and 16GB RAM. The data is stored on an external disk. Any ideas on how to handle rasters of this size?

  • 2
    You might try to have a look at Numpy memmap which is a memory-mapped file that behaves similar to a Numpy array. Have a look at the memmap doc. You will need to load your files in several steps into a Numpy array with the arcpy.RasterToNumPyArray function in order to avoid the out-of-memory error, and each time update your memmap accordingly
    – chkaiser
    Commented Aug 25, 2015 at 17:37
  • Do you have 64bit background geoprocessing installed? Using that will allow you to use 64bit Python, which would allow more memory usage. Commented Aug 25, 2015 at 17:38
  • @chkaiser Thanks, I'll look into memmap and will update the question if I find my answer there!
    – GabeFS
    Commented Aug 25, 2015 at 18:00
  • @EvilGenius I do have 64bit background geoprocessing installed, and have been running the program in a 64bit python installation - that's where my mind went first, too!
    – GabeFS
    Commented Aug 25, 2015 at 18:01
  • If not memmap, try using a buffer to only read a manageable amount of rasters at a time (and iterate).
    – Clay
    Commented Aug 25, 2015 at 18:28

1 Answer 1


For anyone who finds this and has a similar question, I found a useful answer in this related question.

What worked for me was exporting the large raster datasets from the geodatabase to .tif files, and then using the SciPy image reader to render them as NumPy arrays:

import scipy
from scipy import misc
raster = misc.imread("myraster.tif")
np.save("myarray.npy", raster)

Hope that is helpful to anyone stuck on a similar problem!

  • But it needs PIL to be installed! Also I am facing problem with the method
    – Learner
    Commented Nov 17, 2016 at 13:08

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.