Explore multiple datasets together

Current situation¶

The user experience for exploring datasets currently drives many Jupyter users to open QGIS or ArcGIS to explore their data, and they report friction switching between applications.

Some current tools aren’t quite meeting user needs, e.g.:

• my_gdf.explore(): Only shows one dataset at a time

• my_dataarray.plot(): Shows one dataset, non-interactive

• JupyterGIS: Symbology configuration is manual and low-level

• ipyleaflet: Low-level. Can’t render Xarray DataArrays dynamically (see possible solution: https://github.com/davidbrochart/xarray_leaflet)

Proposed improvement¶

Prototype: https://github.com/geojupyter/jupyter-microgis

Add layers to a map for exploration within a Jupyter Notebook:

from ... import explore

explore(
    my_xarray_dataarray,
    {'data': my_geodataframe, 'symbology': 'choropleth'},
)

• Support rioxarray DataArrays

• Support geopandas GeoDataFrames

• Maybe: Support WMTS?

• Support some curated default symbology options

  • Choropleth: number of steps, classification mode, ?

  • Symbol map: shape, min/max size, ...

  • Dot density: ...

  • Cartogram: Maybe?

• Support some symbology customization

  • Each symbology option provides sensible defaults and allows for detailed customization if desired. The point isn’t to help people make maps for publication, though. We’re choosing to trade off user control so we can have a simpler API.

• Use a Jupyter Server extension to tile raster data under the covers, e.g. rioxarray DataArray -> TiTiler

• Send vector data to the renderer as binary (geoarrow)

• Future integration (developed as an independent component): Support a data discovery interface which can help the user find other data they want to integrate with their Notebook analysis

  • Plain language search

  • Produce Python one-liners to bring that dataset into their notebook, e.g. geopandas.read_file(...) and xarray.open_mfdataset(...)

• Display the data on slippy map widget (e.g. DeckGL)

Questions¶

Is this leafmap?¶

Is this idea basically leafmap? If we add support to directly visualize xarray datasets with TiTiler in leafmap, does that meet the goals of this use case?