DimensionalArrayTraits.jl

Abstract base package for dimensional arrays and their specific traits Note this is package currently in the discussion stage.

Requirements on spatial data/domain types from the user's perspective

It is proposed to add behind each point, a possible interface (from the user point of view)

• Return coordinates lon/lat/depth/time (and arbitrary dimension) for a given index i,j,k,...

coord(A,:latitude,i,j,k) # or better coord(A,Lat(),i,j,k)  
lat(A,i,j,k) # for common dimension names
lat(A,j) # if grids are aligned with longitude/latitude/depth/time...; otherwise an error
lon

• Return the value at a given index i,j,k... (the result is no longer georeferenced).

value(A,i,j,k)

• Tell what each coordinate represent (is the 3rd dimension time or depth or something else?).

coordnames(x) # would return :longitude, :latitude, ... or Lon(), Lat()

Use the CF convention names (longitude, latitude, time, …)

• Subsetting based on lon/lat/depth/time (and arbitrary dimension) bounding box
• Irregular grid subsetting returns either: a masked array or a vector of points - need to decide which occurs on getinidex.

A[Lat(x), Lon(y), Time(At(t))]

• Find the k nearest points for a given location
• Integral over a raster (knowing the "volume" of each grid cell)

volume(A,i,j,k)

• Is the cell value for the start, middle or (probably never) the end of the duration/span? Is it an average of the whole period or a specific point in the cell. These specifications should also generalise to any dimension type.
• Categorical dimensions with no size and optional ordered/unordered status
• Arrays can be plotted correctly using the grid information

plot(A[:,:,n])

Out-of-scope (at least initially)

• units and other metadata

Implementation/Performance

• Zero cost indexing (only works on regular grids?)
• Lazy affine transformations on the data

Minimal interface to be implemented for different array types of spatial data

https://gist.github.com/meggart/e29e6381d9400ff789eefbccc109d6f9