roll setting the longitude values incorrectly

[agstephens]

Hi @ellesmith88,

I think I have found a bug in the roll code. In this example, the resulting Dataset has a set of longitudes that are not the same as the input longitudes. Here are the details:

• input longitudes:

 lon = -179.75, -129.75, -79.75, -29.75, 20.25, 70.25, 120.25, 170.25 ;

• requested bounding box: 1 to 300 degrees longitude

• output longitude array: [ 1., 51., 101., 151., 201., 251.]

I think that the correct roll should return longitudes of value:

20.25, 70.25, 120.25, 170.25, 180.25, 230.25, 280.25

Note that gap between the longitude values is not constant when you wrap it.

Here is a reproducible example of the error:

$ wget "https://github.com/cedadev/mini-ceda-archive/blob/main/archive/badc/cru/data/cru_ts/cru_ts_4.04/data/wet/cru_ts4.04.1901.2019.wet.dat.nc?raw=true" -O cru_ts4.04.1901.2019.wet.dat.nc

$ ncdump -v lat,lon cru_ts4.04.1901.2019.wet.dat.nc | tail -2 | grep =
 lon = -179.75, -129.75, -79.75, -29.75, 20.25, 70.25, 120.25, 170.25 ;

$ python

>>> from clisops.ops.subset import subset
>>> ds = subset('cru_ts4.04.1901.2019.wet.dat.nc', area=['1','1','300','89'], output_type='xarray')
>>> print(ds[0].lon.values)

array([  1.,  51., 101., 151., 201., 251.], dtype=float32)

[ellesmith88]

@agstephens @cehbrecht

I've got this:

            first_element_value = low
            res, offset = calculate_offset(lon, first_element_value)

            # roll the dataset
            ds_roll = ds.roll(shifts={f"{lon.name}": offset}, roll_coords=True)

            # assign longitude to match the roll and copy attrs
            lon_vals = ds_roll.coords[lon.name].values

            if offset < 0:
                lon_vals = np.where(lon_vals > (360 + low), lon_vals, lon_vals % 360)
            else:
                lon_vals = np.where(lon_vals < (360 + low), lon_vals, lon_vals % -360)

            ds_roll.coords[lon.name] = lon_vals
            ds_roll.coords[lon.name].attrs = ds.coords[lon.name].attrs
            return ds_roll

Which seems to be working, but I've only tried a few tests so far. I'm not sure how I got here anymore..! We can look tomorrow morning.