new GeoDataset

[DirkEilander]

This new implementation of the GeoDataset allows for any geometry type (not only points) to be combined with multi-dimensional data. My suggestion is to develop it next the current implementation which we will remove in a future release.

TODO:

• parse GeoDataFrames to GeoDataset
• GeoDataset to netcdf with conversion of geometry to wkt
• parse netcdf with wkt to GeoDataset
• write unit tests
• decide on data accessor name (geo instead of vector ?)

Related issue: #177

Goal

• easily work with geometry with associated multidimensional data variables
• compared to the current implementation it should support other than point geometries.

implementation notes:

• use "vector" accessor for data array and dataset
• keep vector accessor backwards compatible
• use geometry property to get a geopandas.geoarray for each of the following types
• expose some of the geoarray methods directly as method of vector.
• staticmethods: from_netcfd / from_gdf / from_wkt
• methods for to_gdf / to_netcdf

Types of dataset on which the accessor should work:

    Dimensions:      (stations: N, time: NT)
    Coordinates:
      * time         (time) [np.datetime64]
      * stations     (stations) [int]
        lon          (stations) [float]
        lat          (stations) [float]
    Data variables:
        waterlevel   (time, stations) [float]

    Dimensions:      (stations: N, time: NT)
    Coordinates:
      * time         (time) [np.datetime64]
      * stations     (stations) [int]
        geometry    (stations) [numpy object with geometries]
    Data variables:
        waterlevel   (time, stations)

    Dimensions:      (stations: N, time: NT)
    Coordinates:
      * time         (time) [np.datetime64]
      * stations     (stations) [int]
        geometry    (stations) [numpy array of wkt str]
    Data variables:
        waterlevel   (time, stations)

Notes on geometry object
Right now the geometries are saved in the geometry object of the GeoDataset as a geopandas.array.GeoArray which is a numpy duck array type with additional geometry methods. This does not yet implement the __array__function__ method and can therefore not yet be integrated into a coordinate or variable of a xarray Dataset. Instead in an xarray Dataset it is reduced to a normal numpy objects array.

[DirkEilander]

@hboisgon Could you already have a look a the implementation and let me know what you agree with the approach?

[dalmijn]

I did write a unit test for an ogr compliant dataset.

Maybe now or in the near future, rename the registration from 'geo' to 'vector'