contiguous time axis

[danhamill]

What is your issue?

I have a dataset I created using xarray. When I am using netcdf-java to read the time axis, I am getting edges that aren't the same values as the time axis coordinates.

The xarray-developed dataset looks like:

From netcdf-java I see the CoordinateAxis1DTime axis with the following attributes:

The edge array is offset by -0.5 days. After a little digging through the java docs, I found the CoordinateAxis1DTime.edge array will only be the same as the CoordinateAxis1DTime.coords array when the ContiniousAxis1DTime.isContiguous() evaluates as True. But the java docs indicate:

How can I write a netcdf file using xarray where the time axis would be considered contiguous?

[dcherian]

Presumably there is some attribute that netcdf-Java is interpreting. It's not clear to me what that is.

Perhaps @dopplershift has an idea.

[dopplershift]

@ethanrd @haileyajohnson @tdrwenski any thoughts on the above question?

[ethanrd]

If cell bounds/edges are not provided with a coordinate variable, the coordinate values are assumed to represent the center points of a set of contiguous cells. In this default case, the cell edges can be represented as an (n+1) sized array with all but the first and last values shared between neighboring cells. In the non-contiguous case, the bounds/edges must be represented as an (n,2) sized array.

Looks like there's already an xarray issue (#1475 ) on handling cell bounds.

[ethanrd]

So, I believe the results you are seeing above are the expected results with the edge value half way between neighboring coordinate values.

[dcherian]

Thanks @dopplershift and @ethanrd . It seems like it's interpreting the absence of an attribute (!).

[danhamill]

Hey!

Thanks for the thoughts. Took me a wile to realize the edge array and the time coordinates aren't the same thing. The time coordinates are really labels and the edge array is calculated from it.

Not sure I agree with the default calculation as the mid point of the time coordinate labels, but I think that is a battle I won't take up.

+1 for supporting this kind of coordinate attribute in xarray. Based on the other issues linked here, it seems it still has not been implemented?