This project compares read times of netCDF-3, netCDF-4, netCDF-4 Classic, Zarr, and Zarr being read with Xarray.

The data used in this example is publicly avaliable from Unidata's example files. The file is from the Community Climate System Model (CCSM) with one time step of precipitation flux, air temperature, and eastward wind. The data was converted from netCDF-3 to Zarr using this repository. The netCDF-4 files were created from the original example file using Unidata's nccopy utility. Zarr directory stores were chunked using Rechunker.

The following chunk sizes applied to latitude and longitude (lat/lon) were compared:

• 128 x 256 (original)
• 64 x 64
• 16 x 16
• 4 x 4

Run times are recorded in milliseconds using timeit with the average taken of 100 excecutions.

Dependencies & Versions

• python=3
• netcdf4=1.5.7
• hdf5=1.8.20
• jupyter=6.1.3
• zarr=2.8.3
• xarray = 0.11.3
• matplotlib=3.4.2