Julia package for high-level loading and manipulating multidimensional labeled arrays. Mostly designed for working with Earth System data cubes...but should hopefully be extensible to other scientific fields.

The package essentially tries to combine the functionality of the python package xarray with the reduction/transformation macros from DataFramesMeta.jl. In contrast to AxisArrays.jl and YAXArrays.jl, this package focuses on loading existing data cubes (preferably on the cloud) and getting to training/analysis stacks as cleanly as possible with the use of macros and Chain.jl.

Current support for zarr and NetCDFs files. Macro naming mirror DataFramesMeta.jl with a preceding A e.g. @subset => @Asubset.

• support for common satellite data types
• integrate support for Dagger.jl for out-of-core operations
• Unitful.jl integration

You can install MacroArrays by typing in the repl:

] add MacroArrays

...for getting publicly-available CMIP6 climate model data on google cloud

using MacroArrays

A = @zread "gs:https://cmip6/CMIP6/ScenarioMIP/NOAA-GFDL/GFDL-CM4/ssp585/r1i1p1f1/day/pr/gr1/v20180701/"

...perform some simple operations. For example, get monthly precipitation anomalies

A_anom = @chain A begin
  @Aselect :pr        #only necessary if you have multiple vars in the dataset
  @Atransform :month = Month.(:time)
  @Aby :month :monthly_mean = mean.(:pr)
  @Atransform :month_ano = :pr .- :monthly_mean
end

...or subset and reduce the data

A_reduced = @chain A begin
  @Atransform :month = Month.(:time)
  @Arsubset :month in Set([1,2,3])
  @Asubset :lat.<45 :lon.<10
end

...pipe to training device. Should convert to ::cuArray ok.

using CUDA, Flux
...
A_anom |> gpu

... in place operators

@Arsubset! A :lon in Set(30:0.5:80)