Skip to content

How to use this page during the Summit

  • This page is your team’s shared workspace and final report-out page. It captures your group’s process and thinking throughout the Summit and will be used to share your work with others.

  • Use this page as your team’s working record during the Summit and your final report-out.

  • The Summit has several different goals and thus you will use the page differently each day: Day 1 is for alignment, Day 2 is for building one useful thing, and Day 3 is for synthesis and report- out.

  • Look for the green buttons to indicate what you need to edit.

  • Megaphones 📣 indicate which items you will be presenting during the end-of-day report-outs.

  • Only the items with megaphones will be visible when you hit the 'Summit Report Out' button.

  • If you turn off 'Instructions' then you will only see the page content for public display.

Embedders (depth)

Day 1 directions

Change the title to the name of your project.

Edit Day 1 setup in Markdown

For ESIIL staff

Group Number: 5

Breakout Room #: S348

ESIIL staff edit in Markdown

Team hero image

How to replace the image above

Upload an image that represents your project and welcome people to your page.

Upload your own image to docs/assets/hero/ and replace the file named hero.png. Use a wide image if you can, then refresh the site preview to check how it looks. Keep the file path docs/assets/hero/hero.png if you want the Markdown above to keep working.

Open image folder for changing image

See a completed example

People

Day 1 task

Get to know your team: share your cards (5-7 mins). Update your team roster (2-3 min).

Use the in-person name cards to guide quick introductions.

Name card prompts Follow-up notes
Name card prompts for name, institution, area of expertise, research difference, and questions Follow-up notes card with space for names and follow-up ideas

Edit People in Markdown

Name Affiliation Contact Github
Dusty Gannon Oregon State University dustin.gannon@oregonstate.edu Dusty-Gannon
Keqi He Virginia Institute of Marine Science khe@vims.edu hkqcqq
Annie Taylor The Nature Conservancy annie.taylor@tnc.org annietaylor
Levi "Veevee" Cai Univ. of Colorado, Boulder levi.cai@colorado.edu arizonat
Brian Lee Univ. of Colorado, Boulder brian.lee-4@colorado.edu bhyleee
Matthew Helmus Temple University mrhelmus@temple.edu mrhelmus
Ria Gupta Simon Fraser University, Burnaby, Canada riagupta312@gmail.com riaongreatlearning-hub

Team Norms and Decision Making

Gradients of agreement
Love it! ❤️ one finger
Like it 👍 two fingers
Live with it 🤷‍♀️ three fingers
Loathe it 🚫 four fingers 
[Edit Team Norms in Markdown](https://github.com/CU-ESIIL/Summit_group_2026_5/edit/main/docs/index.md?plain=1#L87){ .md-button target="_blank" rel="noopener" }

Our team norms:

  • Single slack channel, communicate over slack by @ing.
  • Some test cases (depth) may illustrate the inventory (breadth) group, but they do not have to be shared in the larger paper (it is okay to keep your case study independent)
  • Step up, step back / even turn taking

AI norms:

  • Research/exploration
  • Code snippets and claude code/codex/copilot ok (don't give sensitive data)
    • Prefer claude, gemini, cyverse and copilot with anthropic activated, if possible
  • Editing/polishing/reviewing work (read after the edit)
  • Create documentation like READMEs
  • Not for first pass writing of papers

Our decision making strategy:

  • Call for vote
  • Hold fingers up to indicate agreement on scale above

Our product(s) 📣

[Edit content below here in Markdown](https://github.com/CU-ESIIL/Summit_group_2026_5/edit/main/docs/index.md?plain=1#L106){ .md-button target="_blank" rel="noopener" }

Short term:

Library of notebooks illustrating different use cases and workflows.

Long term:

  • Comparison of different embeddings across use cases and comparison to traditional ML approaches
  • Test case specific products including:
    • Benchmarking toolkit
    • Site selection tool
    • Domain-specific papers

Day 2 morning whiteboard or notes photo

Morning whiteboard or notes showing the question, hypotheses, and context we used to start Day 2.

Our question(s) 📣

Our working question:

  • Brian:

  • Matthew: Risk mapping of a novel pest

  • Dusty: Counterfactual / control site selection based on similarity in embedding space

  • Annie: Predicting plastic mulch use across space

  • Keqi: "Ghost forest" land cover classification using embedding features

  • Veevee: Tool development for evaluating foundation models for specific tasks

  • Ria: Land classification

What would count as progress:

Why this matters (the “upshot”) 📣

This matters because: - Embeddings could help with generalizability, reproducibility, and simplifying workflows, but we still need to know when and where they work well and when and where they don't.

People who could use this:

  • Land managers
  • Environmental data scientists
  • Embeddings developers

Data sources we’re exploring 📣

Promising data sources:

Methods/technologies we’re testing 📣

methods

Add 2-4 methods/technologies we're testing (stats, models, viz).

View shared code

Methods/technologies we are testing:

Method or technology What we tested Early note
AlphaEarth ... ...
Prithvi ... ...
MOSAIKS ... ...
... ... ...

Challenges identified

  • ...
  • ...

Visuals

Potential "Ghost Forest" locations (Similarity Analysis using Alpha Earth) Method or workflow visual

Next Steps

Short term:

Long term:

Day 3 Tasks

Sythesis: highlight 2-3 visuals that tell the story; keep text crisp. Practice a 6-minute walkthrough of the homepage. Why -> Questions -> Data/Methods -> Findings -> Next

Edit content below here in Markdown

Team Photo

Team photo

Team members and collaborators who contributed to this project.

Findings at a glance 📣

  • Headline 1: Embedding space similarity could be useful for selecting control/comparison sites in natural experiments

dusty image

dusty image

Visuals that tell a story 📣

Visual 1: Similarity of a reference "Ghost Forest" pixel.

Story visual

Visual 2: Spotted Lanternfly data.

Story visual

Visual 3: Spotted Lanternfly risk.

Story visual

Visual 4: Land classification through unsupervised and supervised learning on AlphaEarthEmbeddings.

Ria visual

Ria visual Future: Extend this to wasteland detection research.

Visual 5: Agricultural plastics classification augmented with embeddings generated with the Prithvi-EO-2.0 model - Developed a draft script to generate embeddings with Prithvi-EO-2.0 model - Next steps: get the script working, run a RF classification

Example classification with RS/ML approach (no embeddings): at-figure

Pre-existing geospatial AI benchmarks

Benchmark/pipelines Problem types URL
TorchGeo Overall geospatial pipelines https://github.com/torchgeo/torchgeo
GeoBench Classification, Segmentation https://github.com/servicenow/geo-bench
OceanBench Regression https://github.com/mercator-ocean/oceanbench
SustainBench Classification, Segmentation, Regression https://github.com/sustainlab-group/sustainbench/
AiTlas Classification, Detection https://aitlas.readthedocs.io/en/latest/
EarthNets Classification, Segmentation, Change detection https://earthnets.github.io/

What’s next? 📣

Short term:

  • ...

Long term:

  • ...

Who should see this next

  • ...

Cite & Reuse

If you use these materials, please cite:

Summit Team. (2026). Summit Group 2026 Team 5 — Innovation Summit 2026. https://github.com/CU-ESIIL/Summit_group_2026_5

License: CC-BY-4.0 unless noted.