0

Is it possible, with Python + netCDF4, to open an existing NetCDF file and change one of the dimensions from fixed size to an unlimited dimension, such that I can append data to it?

I found this question/answer, which lists several options for doing this with NCO/xarray, but I'm specifically looking for a method using the netCDF4 package.

Below is a minimal example which creates a NetCDF file with a fixed dimension (this part, of course, does not exist in reality, otherwise I could simply create the file with an unlimited dimension...), and then re-opens it in an attempt to modify the time dimension. The netCDF4.Dimension time_dim has a function/method isunlimited() to test whether the dimension is unlimited or not, but nothing like e.g. make_unlimited(), which I was hoping for. 

import netCDF4 as nc4
import numpy as np

# Create dataset for testing
nc1 = nc4.Dataset('test.nc', 'w')
dim = nc1.createDimension('time', 10)
var = nc1.createVariable('time', 'f8', 'time')
var[:] = np.arange(10)
nc1.close()

# Read it back
nc2 = nc4.Dataset('test.nc', 'a')
time_dim = nc2.dimensions['time']

# ...
# (1) Make time_dim unlimited
# (2) Append some data
# ...

nc2.close()
Bart
  • 9,825
  • 5
  • 47
  • 73
  • I think you need to create a new unlimited time dimension and a variable based on this and copy all data over. – Mike Müller Dec 01 '17 at 08:25
  • But then the dimension has a different name (unless you make a temporary copy of everything, remove the old dimensions/variables, create the unlimited dim with its old name and copy everything back, which would be a pain in the ....) – Bart Dec 01 '17 at 13:39
  • 1
    Well, you can rename a dimension with `nc2.renameDimension('time', 'tmp_time)` before making a new one named `time`. – Mike Müller Dec 01 '17 at 15:14
  • Thanks Mike, didn't know that! I'll see what I can do with that – Bart Dec 01 '17 at 15:21

1 Answers1

0

Something like this...

# rename the old dimension
data.renameDimension('time_dim', 'tmp_dim')
# create a new unlimited dimension with the old fixed size dimension's name
data.createDimension('time_dim', None)
# copy the data from the renamed dimension
time_dim[:] = tmp_dim[:]
# delete the temporary dimension
del data.dimensions['tmp_dim']
botheredbybees
  • 582
  • 6
  • 14