Styjun

Set up

suppose I have a NetCDF file that stores a number of rasters indexed by date, longitudes and latitudes, loaded in memory with XArray with name "stack":

<xarray.Dataset>
Dimensions: (date: 1, lat: 2000, lon: 7200)
Coordinates:
 * date (date) datetime64[ns] 2000-01-01
 * lat (lat) float64 49.97 49.92 49.87 49.82 ... -49.88 -49.93 -49.98
 * lon (lon) float64 -180.0 -179.9 -179.9 -179.8 ... 179.9 179.9 180.0
Data variables:
 rainfall (date, lat, lon) float64 ...

task

to add a new date to the stack.

method

my approach is to create a Dataset "new" from the raster with the same indices as the NetCDF loaded:

xr.DataArray(
 <some numpy data>,
 dims=['date', 'lat', 'lon'],
 coords=
 'date': [<some datetime64>],
 'lat': <same list of latitudes>,
 'lon': <same list of longitudes>
 ,
 name='rainfall'
).to_dataset()

and then concatenate:

merged = xr.concat([stack, new], dim='date')

This works but is not very elegant and being new to XArray maybe there is a better way to go about this, for example just with some indexing routines say adding a new date and data; something like:

stack[<new_date>] = <some numpy data>

After some work I have a workaround using netCDF4-python which is useful if you are looking for a way not to load the entire netCDF in memory at once.

The original netCDF file is processed with XArray but I fallback to netCDF4 for this particular operation.

An example follows where I want to add a timestep to the unlimited variable date. The other 2 variables are longitudes and latitudes.

First off I open the netCDF using netCDF4 and read the variables date that I will extend as well as the data

d = Dataset('dataset.nc', 'a')
dt = d.variables['date']
data = d.variables['data']

after that I add the numpy array to a slice:

data[len(dt):len(dt)+1, :,:] = <some numpy data>

and finally add the extra time step:

from datetime import datetime
from netCDF4 import date2num
dt[len(dt)-1] = date2num(datetime(<year>, <month>, <day>), dt.units)

Hope for it to be useful to others.

Try to use reindex to extend the original DataArray, and then assign value with indexing.

extra_date = <some datetime64>
date_extended = np.concatenate([stack.date, [extra_date]]
# this will extend the arrays and place NaNs in the new position
stack_extended = stack.reindex('date': date_extended)
# now assign to that position
stack_extended.loc[dict(date=extra_data)] = <some numpy data>