🌐 GeoJupyter 🌐
@
🌥️ CNG ’26 🌥️

Exploring more approachable geospatial data workflows as a community

Matt Fisher

Schmidt DSE @ UC Berkeley

Guillaume Eynard-Bontemps

Centre National d’Etudes Spatiales (CNES)

👋 Hi, I’m Matt! Website: mattz.cool

💻 Research Software Engineer (RSE)

🧑‍🤝‍🧑 Community builder

👐 Open source maintainer & contributor: jupytergis, jupyter-tiler, earthaccess, several conda-forge packages, more!

🐻 At Schmidt DSE, UC Berkeley

❄️ Previously at National Snow & Ice Data Center (NSIDC)

👐 📖 🧠 🌱 🎶 🥁 🎹 🐶

🐶 🐶 🐶 🐶 🐶🐶 🐶 🐶 🐶 🐶 🐶 🐶

geojupyter.github.io/presentation-cng2026

QR code: These slides! There will be links you may want to follow and bonus slides you may want to see on your own time.

Emojis: Open source, documentation, optimizing for cognitive load, plants and nature, musical instruments, dogs!

👋 Hi, I’m Guillaume! GitHub: guillaumeeb

💻 👾 Satellite image processing and distributed data analysis expert at CNES

❄️ 💾 Maintaining snow detection algorithm (Let it snow), and improving CNES library ecosystem (from the outside mainly)

🌏 Member of Pangeo community for 8 years (missed the start)

🏭 Previously head of the CNES computing center team, specialized in Big Data tools (Spark, Dask)

👐 Open source supporter, funding and small contributor: dask, dask-jobqueue, Let it snow,

👐 📡 🧠 🌏 ⛺ 🏃 🏂 🎸 🍷 🌉

geojupyter.github.io/presentation-cng2026

Emojis: Open source, Satellite, optimizing for cognitive load, earth observation, hiking, running, mountain snow climbing, guitar, party!

🌏 GeoJupyter community overview

⚡ Lightning version!

🔗 geojupyter.org

I’m gonna keep this really high level, but feel free to ask me questions or see bonus slides at the end on your own time :)

GeoJupyter community overview

---

GeoJupyter is an open and community-owned effort to reimagine geospatial interactive computing experiences within the Jupyter architecture to enable more people to confidently engage with geospatial data.

---

Many players!!!

GeoJupyter is not software; it’s a community which will build many things together!

GeoJupyter community overview

Open, participatory development

💞 User-centered & user-led

🤗 Welcoming (like Jupyter)

🔦 Exploring: finding & opening hidden doors

🏯 Data & computational sovereignty

🏗️ Projects in the GeoJupyter community

JupyterGIS

A screenshot of JupyterGIS

…

The development of JupyterGIS has been led by QuantStack.

🤝 Collaborate

Two users collaborating in real-time on a JupyterGIS project (QuantStack)

Thanks again to the modular Jupyter architecture, we can leverage the jupyter-collaboration extension, which enables real-time collaboration on Notebooks, to enable real-time collaboration on a map.

Here you can see two users’ cursors (emphasized for visibility) on the same map.

🤝 Collaborate: Follow a user

One user following another in real-time on a JupyterGIS project (QuantStack)

In this graphic, one user is following another user’s viewport activity in real time.

🤝 Collaborate: Edit together

Two users editing a shared map (QuantStack)

Here you can see collaborators editing a shared map.

🤝 Collaborate: Annotations

Two users having a conversation with spatial context through annotations (QuantStack)

Here you can see collaborators having a spatially aware conversation, like comments in Google Docs.

🔮 Discover data with STAC (WIP)

Browsing a STAC catalog in JupyterGIS (eo science for society blog)

We can discover data using Spatio Temporal Asset Catalogs, or “STAC catalogs”, a modern standard for computer interface with data catalogs.

This is a work in progress that will continue to evolve over time!

📖 Tell a story

Viewing a story map created in JupyterGIS

JupyterGIS also has a story builder interface for building map-driven stories.

Experiment: reproducible viz -> Notebook workflows

Reproducible workflow from viz-land to Notebook-land (👏 Benny & Esha!)

Jupyter Trail (prototyping!)

The 2026 Carto State of Spatial Analytics report:

…the majority use between 3 and 8 tools to get work done

page 35

Cloud-native has become non-negotiable

page 23 (emphasis added)

…

When working across 3 to 8 tools, it’s very difficult to remember or track what you did, and in what order.

Our ability to share, teach about, and reproduce our work all suffer.

…

And because we’re at CNG, I wanted to highlight this other claim from the report :)

Jupyter Trail (prototyping!)

A tool for tracking workflows across tools

A diagram of a “trail”

This diagram shows a hypothetical “trail” through multiple tools.

Here the user has recorded their actions across several tools and curated this history.

Highlight the meaningful part of their path (green line)

De-emphasize their side-quests (faded)

They might use this tool to help them reproduce and/or document their workflow for a publication.

Jupyter Tiler

A building block (on top of TiTiler and xpublish-tiles) for interactively exploring Xarray data.

🎉 Available to try now in JupyterGIS!

A diagram of jupyter-tiler

Mention Development Seed and Earthmover

Jupyter Tiler

Animation of an Xarray-computed layer with jupytergis-tiler

Here, we’re directly rendering from an Xarray dataset based on the requested spatial extent.

This means we can use lazy computation to compute subsets of the dataset to generate tiles. We’ll show more about this in the demo.

Jupyter Tiler

Google Earth Engine UX with open source tools?

Secret sauce is using overviews efficiently

Many researchers use GEE because of its strong and accessible visualization-analysis feedback loop.

GEE is expensive and subjects users to vendor lock-in.

We haven’t quite nailed this in the open source ecosystem yet.

The cloud-native pieces are there, but the user experience isn’t there yet.

…

GEE achieves this through efficient use of overviews – depending on the requested output scale, the appropriate resolution of input data is used.

In other words, if the map is zoomed all the way out, computation is performed on the lowest resolution overview.

🎉 Demo time

🚨 Emergency video, ignore me

Jupyter Tiler - current limitations

Overviews

A visualization of overviews / tile pyramid.

Let’s talk about overviews.

Cloud optimization can mean lots of stuff, and one of those stuff is overviews, or tile pyramids.

• Lower resolution chunks that can be accessed directly
• Critical for cloud-optimizing our data for visualization! When you’re zoomed out, it’s wasteful to read every pixel on your dataset, because your screen probably has far fewer pixels than the full resolution data.
• Zarrs, Cloud Optimized GeoTIFFs are notable formats that support overviews

Jupyter Tiler - current limitations

Overviews

Problem: A computation on an Xarray dataset is still performed at full resolution.

For a tight viz-analysis feedback loop, we might want to compute on overviews.

Jupyter Tiler - current limitations

Overviews

An open Xarray Dataset represents a single resolution – no built-in concept of overviews

A DataTree can represent overviews! But the familiar way of doing computations is with a Dataset or DataArray.

Can technically map a computation across a DataTree!

dt = xr.open_datatree("s3://.../some.zarr", ...)

def add_ndvi(ds):
    ...
    ds["ndvi"] = (ds.nir - ds.red) / (ds.nir + ds.red)
    return ds

dt = dt.map_over_datasets(add_ndvi)

xpublish-tiles can generate tiles from these DataTree objects

Dataset is single-resolution. You can request a specific overview level at open time only.

…

You could build a GeoZarr-style multiscale DataTree for a single object that knows about the overviews.

…

To do a computation, you can map a function to add a task graph to each dataset in the tree. xpublish-tiles can work with this.

Jupyter Tiler - current limitations

Overviews

🫣 That was unfamiliar. A user wouldn’t know they need to do this.

✨ Imagine: Familiar Xarray Datasets (or similar objects) that are aware of overviews and capable of computing on overviews if explicitly requested (e.g. by a tile server!)

🤷 How?

This new complexity and cognitive load is unnecessary friction for something a researcher would expect to “just work” – if my data has overviews, why can’t I visualize with my overviews?

…

Imagine if the existing Dataset API had awareness of the presence of overviews and would allow overview-aware computations. E.g. “compute NDVI for a small coordinate bounding box at max resolution” or “compute NDVI for the whole dataset using the coarsest overview”.

…

How do we build this? Is this a bad idea? Is this feasible? How can we fund the work? Does this fit in Xarray or is it a new thing?

If you saw Joe Hamman’s lightning talk about Zax yesterday, that might be one pathway.

We’re looking to y’all for feedback on this idea. Find me and let’s chat or please reach out post-conference :)

JupyterGIS with OpenEO (WIP)

Another approach: server-side execution!

An OpenEO layer and process graph in JupyterGIS

The approach we just discussed is client-side.

We’re also working on an approach using OpenEO to submit tasks (as graphs) to an OpenEO backend, ideally positioned close to the data.

Guillaume to add more?

How you can participate in GeoJupyter

🗨️ Chat with us, share ideas on the Jupyter Zulip!

🎙️ Sign up for an interview!

📆 Join a hackathon or community meeting!

🧪 Try JupyterGIS! Where does it meet your needs (or not)?

🎁 Report bugs, request features! Open Pull Requests!

😎 Share your rad vibes and leadership!

❤️ Thank you! ❤️

🎉 Bonus slides

GeoJupyter community overview

Our values

👐 Open source & open science - geospatial data is important to everyone!

🤸 Approachability and playfulness, like desktop GIS tools

🪶 Flexibility and reproducibility, like coding methods

🎭 Collaboration and storytelling, like Jupyter Notebooks

GeoJupyter community overview

Geospatial data practice for the modern era

Geospatial data is everywhere and matters for everyone! 🚚🚢🗺️🧪🌏

🤝 Real-time collaboration (like Google Docs)

♻️ Reproducibility (by default!)

🍃 Accessibility (transition to new ways of working, including reproducibility)

☁️ Cloud-native (computing, data formats)

🤖 AI 😱 💥 (risks & opportunities)

GeoJupyter community overview

Partners 🧑‍🔬 🧑‍🏫 🧑‍💻

• QuantStack - Open source for science
• Maryam Hosseini - urban systems, computer vision, & open source
• Clancy Wilmott - Critical Cartography, Geovisualisation and Design
• Qiusheng Wu - Geospatial, AI, & education
• Sarah Chasins & Parker Zeigler - cartography, CS, & open source
• Nancy Thomas & Iryna Dronova - Berkeley Geospatial Innovation Facility
• Carl Boettiger - Geospatial, AI, & education
• Benny Szeghy & Esha Potharaju - GeoJupyter interns
• Friends & neighbors: BIDS, MyST, JupyterHub, earthaccess, DevSeed, Pangeo, 2i2c, Clark University, Stanford, Simula, CNES, ESA
• MANY MORE!!!

Potential future initiatives?

📖 “Scrollytelling” with a Markdown authoring workflow? (initiative)

🤖 GeoAI? 🤔 (initiative, prototype, talk)

🪨 Richer geospatial primitives for Python? (initiative)

⛰️ Reproducible GUI “geoprocessing” (following lessons learned from interns’ exploration)? (initiative)

🎨 Reusable symbology editor component? (initiative)

🧑‍🏫 Example datasets for education? (initiative)