SpaceTimeVariogram for unstructured spatio-temporal data

[johanneskopton]

Thanks for this very useful package!

As far as I understand, the SpaceTimeVariogram function expects a full space-time grid, so m locations and n timesteps, leading to m*n observations.

I need something different: n points in space and time with one observation each.

What do you think about adding this feature? I need to implement this anyway, so very happy to contribute. However, the API would need to be changed for this.

[mmaelicke]

Hi @johanneskopton,

I had exactly this in mind for ages. The space-time grid should just be a shortcut and special case in the future, as it is not flexible enough. I didn't find the time so far, as I am currently not using space-time geostatistics, but contributions are more than welcome.

I am happy to discuss future API, implementation details and help wherever I can. We can also have a zoom call or something to get you started as quickly as possible.

Best,
Mirko

[johanneskopton]

Hi @mmaelicke,

this is very good news!

The problem is, I checked the timeline of my current (space-time geostatistics related) research project and figured that I unfortunately won't have the time, to first integrate the methods into scikit-gstat properly and then use them for the project to generate results.

So my plan is, to first hack the whole spacetime-kriging thing together kind of quick and dirty to get the research going and then get back to you on how to properly implement it in scikit-gstat for others (and my future-self) to use.

I hope I can then also contribute to #151.

Cheers,
Johannes