Gaussian Processes with Variogram vs Covariance

[markmbaum]

I noticed that simulating random fields from a GaussianVariogram works but using a GaussianCovariance does not.

For example, after

using Meshes, GeoStatsFunctions, GeoStatsProcesses

The following works fine

field = rand(
    GaussianVariogram() |> GaussianProcess,
    CartesianGrid(10, 10),
    :x => Float64,
)

but using a Covariance does not

field = rand(
    GaussianCovariance() |> GaussianProcess,
    CartesianGrid(10, 10),
    :x => Float64,
)

and produces this error

ERROR: ArgumentError: Cannot construct range from arguments:
start = GaussianCovariance(sill: 1.0, nugget: 0.0, range: 1.0, distance: Euclidean)
step = nothing
stop = nothing
length = nothing
Try specifying more arguments.

This is with

  [aa102bde] GeoStatsProcesses v0.5.8
  [6771c435] GeoStatsFunctions v0.2.1
  [eacbb407] Meshes v0.40.12

[juliohm]

I can reproduce the issue. This is because the Covariance counterparts were added recently for convenience, but no one went ahead and added the necessary methods as in the Variogram types.

Would you like to add the missing methods in GeoStatsFunctions.jl?

[markmbaum]

Yes I can get to it, feel free to assign me.

I've been appreciating the Covariance types, by the way.

[juliohm]

Hi @markmbaum did you have a chance to take a look into this?

[markmbaum]

Yes sorry for the delay, planning to do it soon.