# Siterra

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Glendon Chin](https://cerebralvalley.ai/u/chinny)
- **GitHub:** https://github.com/glendonC/siterra
- **Demo video:** https://youtu.be/jSckHXi8Gyw
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4277

Siterra is the missing analysis layer between the world's free environmental data and the decisions made about land.

More than half of Earth's land is rangeland, and about two billion people depend on it for a living (UNCCD 2024). The evidence about any one place already exists, but it is scattered: decades of free satellite imagery, rainfall and snow records sit in separate archives, the research sits in papers across languages, and the local rules about who uses the land sit in neither. Pulling it together takes specialist analysis that most places never get, so this data rarely shapes land decisions (IPBES 2018) and monitoring stays expensive and hard (UNCCD 2024). When the cause of a change is guessed instead of tested, the people on the land pay: in colonial Morocco, officials blamed herders for vast forest loss and outlawed most of their traditional forest uses; fossil pollen later showed the loss was far smaller (Davis 2005).

Siterra uses Meta's Muse Spark to do that analysis. Give it a place and a question about how its land is changing, and it brings the satellite, rainfall, snow and research record for that place into one case, tests every plausible cause side by side (rainfall, grazing, land use, snow and terrain, or a fault in the data), maps the evidence for each onto a 3D model of the land, and names the smallest set of field measurements that would settle what the data cannot. Every finding traces back to its measurement or the exact page of its source, and the result is a brief that communities, scientists and land managers can review together.

Under the hood: Muse Spark writes its predictions for each cause with no measurements in view, deterministic code locks them in and scores them, per-pixel trends are tested with false-discovery control, and every quote and number is checked against its source.

First case: the Aït Bouguemez valley in Morocco's High Atlas, where villages share their pastures under agdal rules. Since 2000, 24.7% of the catchment has greened. Siterra found that rainfall fits that change, that grazing cannot be judged yet, and that local rain records would tell the two apart.

## More from Meta Global AI Developer Hackathon

- [HeyNavi](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4274)
- [Nur : ReproCam](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4275)
- [DischargeBuddy](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4276)
- [Team Decrypton](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4278)
- [Explanimate](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4279)
- [Sonor](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4280)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4277. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt