FAQ

yt and xarray have many similarities in how they handle their Datasets, but there are also many aspects that differ to varying degree. This page describes some of the difficulties you may encounter while using yt_xarray to communicate between the two and how to solve those issues.

xarray datasets with a mix of dimensionality

It is common for xarray datasets to have variables that have a mix of dimensionality. Some variables may be 3D in space (x, y, z), 2D (x, y), 3D in space and time (x, y, t), etc. But with the present implementation, the fields that are loaded into yt must have the same dimensionality. So keep in mind that for most of the functionality here, you will have to provide subsets of fields that you want to work with. For variables with time dimensions, you will generally have to provide a timestep to load.

See Loading data with yt_xarray and Loading 2D fields into yt for examples.

unknown coordinate names

yt datasets have a fixed expectation for coordinate names. In cartesian, these coordinate names are 'x', 'y', 'z' while for geographic coordinate systems the coordinate names are 'latitude', 'longtiude' and then either 'altitude' or 'depth'. To work with xarray variables defined with coordinate names that differ from these, yt_xarray provides some coordinate aliasing, which in part relies on cf_xarray (if it is installed) for additional conversion to standard names.

See Coordinate aliasing for an example.

unknown coordinate systems

yt is designed for volumetric data: 3D (or 2D) data defined in a coordinate system with simple transformations to length axes. The simplest case is cartesian coordinates where 'x', 'y' and 'z' are already in units of length but simple non-cartesian coordinate systems are also implemented, including spherical or geographic coordinate systems. Time-dependence is generally handled by subsequent loading of single time steps.

In xarray, there are no restrictions on coordinate systems: as long as your dimensions are self-described in the dataset, you can extract your arrays. For example, it is common for atmospheric and oceanographic data to use a pressure-related coordinate systems (such as a sigma coordinate system).

While these more general coordinate systems can be loaded into yt via yt_xarray coordinate aliasing (see above), it is important to recognize that while most of yt’s routines will successfully run, the results may be unexpected. Many of yt’s analysis routines involve projecting or volume-averaging, which would yield incorrect results when treating, for example, a pressure-coordinate as a length. Generally, the functionality that will work is that related to data selection such as plotting slices (yt.SlicePlot()), subselecting data or finding extrema. Some methods, like yt.PhasePlot() may work if care is taken to set the averaging keyword arguments appropriately.

The solution for full functionality is to interpolate your data onto a volumetric grid (e.g., if using sigma-pressure coordinates, first calculate altitudes then interpolate your data onto a regular altitude-latitude-longitude grid).

Irregular Grids

The present implementation works with regularly gridded data. Grids may be uniform (equal spacing in each dimension) or stretched (variable spacing in each dimension) and spacing need to be equal between dimensions. But hierarchical or adaptive grids are not presently supported: though it may be possible to load your data in yt, it just might take a bit more manual work with yt’s generic data loaders (NOTE: if you do go this route, consider submitting a PR to yt_xarray to expand the dataset types it can handle!).