Assign new dimension to xarray.Dataset using existing coordinates

Question

I have an xarray.Dataset that looks like this, which is available here: https://psl.noaa.gov/thredds/dodsC/Datasets/NARR/Dailies/monolevel/acpcp.1979.nc:

 <xarray.Dataset>
Dimensions: (time: 365, y: 277, x: 349, nbnds: 2)
Coordinates:
 * time (time) datetime64[ns] 1979年01月01日 ... 1979年12月31日
 * y (y) float32 0.0 3.246e+04 ... 8.927e+06 8.96e+06
 * x (x) float32 0.0 3.246e+04 ... 1.126e+07 1.13e+07
 lat (y, x) float32 1.0 1.104 1.208 ... 46.93 46.64 46.35
 lon (y, x) float32 -145.5 -145.3 -145.1 ... -2.644 -2.57
Dimensions without coordinates: nbnds
Data variables:
 Lambert_Conformal int32 ...
 time_bnds (time, nbnds) float64 ...
 acpcp (time, y, x) float32 ...
Attributes: (12/17)
 Conventions: CF-1.2
 centerlat: 50.0
 centerlon: -107.0
 comments: 
 institution: National Centers for Environmental Predi...
 latcorners: [ 1.000001 0.897945 46.3544 46.63433 ]
 ... ...
 history: created Mon Mar 21 17:56:47 MDT 2016 by ...
 dataset_title: NCEP North American Regional Reanalysis ...
 references: https://www.esrl.noaa.gov/psd/data/gridd...
 source: http://www.emc.ncep.noaa.gov/mmb/rreanl/...
 _NCProperties: version=2,netcdf=4.6.3,hdf5=1.10.5
 DODS_EXTRA.Unlimited_Dimension: time
 

And I want to be able to select a given point using .sel based on the lat and lon coordinates like:

ds.sel(lat=point_lat, lon=point_lon, method='nearest')

But I get KeyError: 'no index found for coordinate lat'. I think this makes sense because lat is a coordinate but not a dimension.

So, what would be the best way to be able to use .sel in this case - is there a way to simply add lat and lon to the set of dimensions?

• Patrick: I am pretty sure there must be a global attribute by the name of "featureType" with a value of "timeSeries". Can you please confirm?
• Patrick: Do you have a link to the file?
• pbreach: @Patrick Updated example for a case that isn't a single point with a link to the file

Answer 1

This netCDF file has only a single latitude-longitude point. lat and lon coordinates use the station dimension which has a length of 1. lat and lon are thus also not coordinate variables (in the sense of the CF Metadata Conventions) and I am a bit surprised that xarray would list them as coordinates but then not allow you to select on them (which indeed does not make sense). I see a lat value of 45.16, which would place this in Lake Michigan or Lake Huron, most likely.

What you have here is most likely a netCDF file with discrete sampling geometries (more CF-speak), in this case a timeSeries of 1-minute interval (time, 1 million+ length over nearly 2 years) of temperature measured at 18 depths (z). You can select on either of these two dimensions as usual.

In short, this is data for a single location. If you want to select data for another location, you should get a gridded data product.

• pbreach: You're right, this isn't the best example with only a single lat-lon point. I've updated for a case where the question still applies and included a link to the file.