Arcpy RasterToNumPyArray does not read floating point raster

Question

I have below rasters:

1. One with dimension (13646, 7510) where INTEGER pixel values ranges from 0-14.

2. One with dimension (13646, 7510) where DOUBLE (upto 10 decimal places) pixel values ranges from 0-181257.

I can easily read 1 with RasterToNumPyArray but when I try to read second raster exception raised as below:

return _RasterToNumPyArray(*args, **kwargs)

MemoryError:

I tried with scipy.misc it also gives me error for 2 but reads well 1. I have some problem with GDAL since, I think, I have installed OSGEO64 while I am using Python 2.7 32 bit. So I want to avoid GDAL.

Now what is the way to read 2 in arcpy. I am using rasterio in virtual environment now and I want to get rid of the mess of using virtual environment.

My machine details:

Answer 1

Raster pixel size is generally described in bits; "INTEGER" could refer to anything from 1-, 4-, 8- 13-, 16-, 32-, to 64-bit, which makes it difficult to gauge how much RAM the working raster requires. You state you're using a 32-bit Python, which is your issue -- It generally only has 1.2Gb of heap available to reference all objects (including the Python language itself).

An image with dimensions 13646 x 7510 has 102,481,460 pixels. With 64-bit depth (IEEE "double" size), the contiguous memory required is 819,851,680 bytes (~782 Mb). While this is a trivial amount of addressable space in a 64-bit application, it represents 2/3 of the heap in a 32-bit application, so even if you were able to allocate it, it's unlikely you'd be able to manipulate it without generating a memory error.

As a rule, memory errors are encountered due to 32-bit application use (the problem confounds solution in the addressable space). The solution is to use a 64-bit application framework. If you run out of memory with a 64-bit application, then it's a coding issue (somebody hasn't thought through the task well enough).