issue_comments

Ah thanks, there is a bug here.

xarray is trying to overwrite indexes for the bnds dimension but there are no indexes associated with that dimension. As a work around, you can assign values for bnds: ds1["bnds"] = [0,1] and ds2["bnds"] = [0,1] and then use join="override"

Since lat is a dimension coordinate or "index variable" you want join="override" instead of compat="override".

A PR to make the docs clearer on this would be appreciated!

Looks like instead of using compat you can tell concat to just use the indexers (so lat, lon etc.) from the leftmost object using join='left', so it will ignore the (small) differences and return you python ds3 = xr.concat([ds1,ds2], dim='time', join='left')

<xarray.Dataset> Dimensions: (bnds: 2, lat: 96, lon: 144, time: 1152) Coordinates: height float64 2.0 * lon (lon) float64 0.0 2.5 5.0 7.5 10.0 ... 350.0 352.5 355.0 357.5 * lat (lat) float64 -90.0 -88.11 -86.21 -84.32 ... 86.21 88.11 90.0 * time (time) object 2006-01-16 12:00:00 ... 2101-12-16 12:00:00 Dimensions without coordinates: bnds Data variables: time_bnds (time, bnds) object 2006-01-01 00:00:00 ... 2102-01-01 00:00:00 lat_bnds (time, lat, bnds) float64 -90.0 -89.05 -89.05 ... nan nan nan lon_bnds (time, lon, bnds) float64 -1.25 1.25 1.25 ... 356.2 356.2 358.8 tas (time, lat, lon) float32 244.75986 244.83588 ... nan nan Attributes: institution: Norwegian Climate Centre institute_id: NCC experiment_id: rcp26 source: NorESM1-ME 2011 atmosphere: CAM-Oslo (CAM4-Oslo-... model_id: NorESM1-ME forcing: GHG, SA, Oz, Sl, BC, OC parent_experiment_id: historical parent_experiment_rip: r1i1p1 branch_time: 56940.0 contact: Please send any requests or bug reports to noresm... initialization_method: 1 physics_version: 1 tracking_id: fd266701-a253-4b74-91e1-5c0213483ba2 product: output experiment: RCP2.6 frequency: mon creation_date: 2012-05-16T19:40:05Z history: 2012-05-16T19:40:05Z CMOR rewrote data to comply ... Conventions: CF-1.4 project_id: CMIP5 table_id: Table Amon (01 February 2012) 81f919710c21dca8a17... title: NorESM1-ME model output prepared for CMIP5 RCP2.6 parent_experiment: historical modeling_realm: atmos realization: 1 cmor_version: 2.7.1 Is that what you're after? Best to check through the data explicitly

even though the latitude arrays are completely identical AFAIK.

Are they completely identical? I see you've used np.allclose to assert that they are identical rather than np.array_equal. You might want to try concat with compat='override'?