...

Perhaps the “easiest” way to plot data on the SE grid is to not plot it on the SE grid at all, but to regrid the data first to a regular latitude-longitude grid and use standard tools to plot. This is the approach that E3SM-Diags uses, and once the data is regridded it can be quite quick to plot, outperforming the rest of the options here. It’s also pretty easy to regrid on the fly using scipy.interpolate.griddata. For example, the follow will regrid SE data pretty quickly:

from scipy.interpolate import griddata
import numpy

xi = numpy.linspace(0, 360, 2 * 360)  # to regrid to 1/2 degree
yi = numpy.linspace(-90, 90, 2 * 180)  # to regrid to 1/2 degree
data_regridded = gridddata((x, y), data, (xi[None,:], yi[:,None]), method='linear')

This can then be plotted using the efficient pcolormesh matplotlib (mpl) function (or really any standard contouring/plotting/mapping functions), and the regridding itself is surprisingly fast. Still, we would like some means to visualize the data on the native grid, so skipping this for now…we explore some other options below.

Dual grid

Data on the SE grid is awkward to work with primary because the data on the Gauss-Lobatto-Legendre (GLL) grid exists at cell vertices (i.e., data is point values), while most existing analysis tools expect cell-centered data (i.e., data points represent values over some representative area bounded by each “cell”). We can sort of mimic this cell-centered view by constructing a new mesh that is a “finite volume dual grid” to the SE GLL grid, in which we construct bounding polygons around each GLL node and pretend that the data at the GLL node is representative of the value across the area of the polygon. When adding a new SE grid to E3SM, we construct a SCRIP-formatted file with this information, primarily because some of our remapping tools used to/still do require this cell-centered representation of the data. And example can be found in the inputdata repository for the ne4np4 grid here:

...