interp over time coordinate

Before I start, I'm very excited about the interp addition in 0.10.7. Great addition and thanks to @fujiisoup and @shoyer.

I see there was a bit of a discussion in the interp pull request, #2104, about interpolating over times and that it was suggested to wait for use cases. I can think of an immediate use case in my line of work.
I frequently use regular gridded geophysical data (time, lat, lon), not unlike the sample tutorial air_temperature data, and the data must be interpolated to line up with corresponding satellite measurements that are irregularly spaced in lat, lon and time.

Being able to interpolate in one quick step would be fantastic. For example:

ds = xr.tutorial.load_dataset('air_temperature')
ds.interp(lat=60.5, lon=211, time='2013-01-01T03:14:37')

Problem description

Currently issues TypeError: cannot perform reduce with flexible type.

Desired Output

<xarray.Dataset>
Dimensions:  ()
Coordinates:
    lat      float64 60.5
    lon      int64 211
    time    datetime64[ns] 2013-01-01T03:14:37
Data variables:
    air      float64 273.5

[shoyer]

Thanks for bringing this up.

@fujiisoup actually implemented interpolation for datetime64 as part of writing interp_like() , which will be in the next release of xarray (0.10.8). With the current development version of xarray:

In [6]: import numpy as np

In [7]: ds.interp(lat=60.5, lon=211, time=np.datetime64('2013-01-01T03:14:37'))
Out[7]:
<xarray.Dataset>
Dimensions:  ()
Coordinates:
    lat      float64 60.5
    lon      int64 211
    time     datetime64[ns] 2013-01-01T03:14:37
Data variables:
    air      float64 273.5
Attributes:
    Conventions:  COARDS
    title:        4x daily NMC reanalysis (1948)
    description:  Data is from NMC initialized reanalysis\n(4x/day).  These a...
    platform:     Model
    references:   http://www.esrl.noaa.gov/psd/data/gridded/data.ncep.reanaly...

However, you do currently need to wrap your indexer explicitly in a np.datetime64 object. It would be nice to support casting strings to dates, somewhat like what we do in .sel(). This would be a good candidate for a pull request.

(I'll try to release 0.10.8 soon, maybe within the next week)

Thanks, I will gladly exercise patience until then.