Gridded & climate data

Climate and reanalysis data arrives as a labelled array: a few variables over latitude, longitude and time, in NetCDF, GRIB or Zarr. xarray gives it names, so you subset by coordinate value rather than by position, and reductions carry their dimension names with them.

The traps are quiet ones. Selecting by position when you meant by label returns an array of exactly the right shape. A variable packed into sixteen bits with a scale factor reads correctly through one library and not through another, and a derived field written with its parent’s encoding silently overflows. A mean over latitude and longitude without cos-latitude weighting over-counts the poles, and the answer still looks reasonable.

Below: the concepts cover the formats, the packing and what a regrid actually computes; the how-to guides open, subset, average and regrid; and the fixes are the values that come back wrong and the maps that come out shifted.

Learning paths run through this topic too — see the end of this page.