interpolate onto a surface in parallel

[tlsw231]

Are there existing routines that exploit dask functionality to interpolate wrf fields using multiple cores in parallel? I am trying to use interplevel to interpolate a 4d array with dims (t,z,y,x) onto a specific height. Following the wrf-python docs, the following syntax works just fine:

interplevel(var, z, 500.)

where var and z are both 4d arrays with the dimensions indicated above. But the task executes serially with only one active core at a time.

[lpilz]

This behavior is the precise reason we wrote xWRF. WRF-python only wraps fortran-routines, which make an easy parallelization impossible. As far as I know, the PR xgcm#483 should implement the required functionality in xgcm.

For now, it should be possible to use something like the following (untested):

import xarray as xr
import numpy as np
import xwrf
from xgcm.transform import interp_1d_linear

ds = xr.open_dataset(...).xwrf.postprocess()
isolevels = np.array([500.])
<destaggering>
var_interp = interp_1d_linear(ds[var], np.log(ds.air_pressure), np.log(isolevels))

Maybe you need some additional destaggering using xgcm. For that you can consult our docs.

[tlsw231]

I am a little confused by 'interp_1d_linear. Does it accept an xarray dataarray as input but require the second and third arguments to be numpy arrays (as your example suggests)? Loading the values of the depth variable will be quite expensive for me as it is a 4d array.
When I try the following:

# ds has been created using xwrf.postprocess()

grid=Grid(ds)

zf = ds['geopotential_height'].chunk('auto')
zc = grid.interp(zf,'Z')

isolevels= 500.

var_interp = interp_1d_linear(ds['air_potential_temperature'], zc, isolevels)

I get the error:

<ufunc '_interp_1d_linear'> not supported: xarray objects do not directly implement generalized ufuncs. Instead, use xarray.apply_ufunc or explicitly convert to xarray objects to NumPy arrays

Should I be using zc.values instead for the interpolation step?