// the find
developmentseed/lonboard
Fast, interactive geospatial data visualization in Jupyter.
Lonboard is a Jupyter widget for rendering large geospatial datasets interactively using deck.gl on the GPU, with data transferred as Arrow/Parquet rather than GeoJSON. It's aimed at geospatial analysts and data scientists who've hit the wall with folium or ipyleaflet on datasets with millions of features.
Apache Arrow as the wire format is the right call — it avoids JSON serialization overhead and lets the GPU-side consume data directly without re-parsing. The anywidget architecture means it works across JupyterLab, Jupyter Notebook, Colab, and Panel without separate widget registrations. CRS reprojection is handled transparently on the Python side before data hits the frontend, which prevents the classic 'why is my map in the ocean' debug session. The ecosystem integration is solid — DuckDB, MovingPandas, GeoParquet, H3, S2, and raster COG all have first-class support rather than being bolted on as afterthoughts.
Under 1,000 stars and 52 forks for something this technically capable suggests limited community traction — if developmentseed deprioritizes it, you're on your own. The frontend is a TypeScript/pnpm build that you have to maintain separately from the Python package, which raises the contribution barrier and has bitten other anywidget projects when deck.gl has breaking major versions (there's already a deck.gl 9.3 upgrade spec sitting in dev-docs). Export to standalone HTML exists but the output size for large datasets is unwieldy since it has to embed the Arrow buffers. No server-side rendering path — everything assumes a live Python kernel, so static notebook sharing on nbviewer or GitHub shows nothing.