Nrthugu

I have an xarray dataset with three separate 4x4 matrices, currently filled with random values.

I can mask out each 4x4 matrix so that all values which are equal to zero are nan, and I would like to replace those nan values with the value from the next matrix down.

This will eventually be expanded to very large arrays of satellite imagery so I can perform searches and create imagery based off the "last best pixel". Below is the code I'm currently using for reference:

import numpy as np

import xarray as xr



dval = np.random.randint(5,size=[3,4,4])



x = [0,1,2,3]

y = [0,1,2,3]

time = ['2017-10-13','2017-10-12','2017-10-11']



a = xr.DataArray(dval,coords=[time,x,y],dims=['time','x','y'])



a = a.where(a > 0)

b = a.sel(time = time[0]).to_masked_array()

What I'd like to do is have any values masked False in b be replaced with values from the 4x4 matrix corresponding to '2017-10-12'. Any help with this would be greatly appreciated.

You can do forward and backward filling by making using of the ffill() and bfill() methods, e.g.,

import numpy as np

import xarray as xr



dval = np.random.RandomState(0).randint(5,size=[3,4,4])



x = [0,1,2,3]

y = [0,1,2,3]

time = ['2017-10-13','2017-10-12','2017-10-11']



a = xr.DataArray(dval,coords=[time,x,y],dims=['time','x','y'])

a = a.where(a > 0)

filled = a.bfill('time')

Results in:

>>> a

<xarray.DataArray (time: 3, x: 4, y: 4)>

array([[[ 4., nan,  3.,  3.],

        [ 3.,  1.,  3.,  2.],

        [ 4., nan, nan,  4.],

        [ 2.,  1., nan,  1.]],



       [[ 1., nan,  1.,  4.],

        [ 3., nan,  3., nan],

        [ 2.,  3., nan,  1.],

        [ 3.,  3.,  3., nan]],



       [[ 1.,  1.,  1., nan],

        [ 2.,  4.,  3.,  3.],

        [ 2.,  4.,  2., nan],

        [nan,  4., nan,  4.]]])

Coordinates:

  * time     (time) <U10 '2017-10-13' '2017-10-12' '2017-10-11'

  * x        (x) int64 0 1 2 3

  * y        (y) int64 0 1 2 3



>>> filled

<xarray.DataArray (time: 3, x: 4, y: 4)>

array([[[ 4.,  1.,  3.,  3.],

        [ 3.,  1.,  3.,  2.],

        [ 4.,  3.,  2.,  4.],

        [ 2.,  1.,  3.,  1.]],



       [[ 1.,  1.,  1.,  4.],

        [ 3.,  4.,  3.,  3.],

        [ 2.,  3.,  2.,  1.],

        [ 3.,  3.,  3.,  4.]],



       [[ 1.,  1.,  1., nan],

        [ 2.,  4.,  3.,  3.],

        [ 2.,  4.,  2., nan],

        [nan,  4., nan,  4.]]])

Coordinates:

  * time     (time) <U10 '2017-10-13' '2017-10-12' '2017-10-11'

  * x        (x) int64 0 1 2 3

  * y        (y) int64 0 1 2 3

The related interpolate_na() method can also be handy for these situations (but not in this particular case).