About muda-clima

muda-clima is the working name of the TerraShift platform — a real-time land intelligence layer designed for the two most consequential forested jurisdictions in the world: the Brazilian Amazon and Cerrado, and the Canadian boreal.

Mission

Make the same parcel of land legible to communities, governments, carbon registries, and financial institutions — without any of them having to trust the others to act in good faith. The data layer carries the trust the institutions cannot.

The gap we're closing

Existing tools are fragmented. Planet Labs sells pixels. Global Forest Watch publishes dashboards. Carbon MRV firms run point-in-time audits. None of them integrate satellite change detection, indigenous territorial sovereignty, forward climate risk, and policy/incentive alignment. muda-clima is the fusion layer.

Scope today

This deployment is Phase 0 — a working scaffold and live demo console. It connects to public web services from INPE, FUNAI, Open-Meteo, and Element84 to prove the architecture end-to-end. Production ingest with provenance hashing arrives in Phase 2; community data partitions in Phase 4.

Architecture principles

1. Parcel as the unit of truth — Pixels and grid cells are raw material. Every output is keyed to a stable parcel UUID linked to all external cadastral IDs (CAR, SIGEF, FUNAI, LTSA, ISC) — the audit trail carbon registries and regulators need.
2. Indigenous data held in trust — Three hard data tiers governed by FPIC in Brazil and OCAP in Canada. Communities are partners with revenue share, not subjects of monitoring. Right-to-withdraw is a first-class technical guarantee.
3. Provenance hashed end-to-end — Every record carries a SHA-256 hash over its source dataset versions. Carbon buyers and reinsurers can reproduce a risk score months later and prove which Sentinel scene, CAR version, and CMIP6 run produced it.
4. Permission filter at the data layer — Policy enforcement runs before any record is written to an output queue, not only at the API edge. Defense in depth against accidental community data leakage.
5. Layers stored separately, fused at query time — Satellite, cadastral, indigenous, and climate live in independently governed partitions. They only meet in the fusion engine when a query needs them combined.

Roadmap

1. Phase 0 — Scaffold + demo console FastAPI + PostGIS + parcel CRUD, HTMX/MapLibre demo with live read-through to INPE, FUNAI, Open-Meteo, and Sentinel-2 STAC.
2. Phase 1 — Government API MVP CAR polygon ingest, FUNAI conflict detection (CAR ↔ FUNAI overlap), provenance hashing, single-tenant pilot with one Brazilian state secretariat.
3. Phase 2 — Change detection Sentinel-2 + GLAD/RADD alert ingest stored locally. Alert stream webhook product.
4. Phase 3 — Carbon MRV Project boundary registration, 40-year MapBiomas baseline audit, continuous monitoring reports for Verra/Gold Standard.
5. Phase 4 — Community tier Three-tier permission model in code, community-keyed partitions, Open Policy Agent filter, revenue attribution logging.
6. Phase 5 — Climate risk + financial CMIP6/CanDCS ingest, parcel-keyed risk score, insurer and agricultural lender API surface.

Countries in scope

Brazil — Amazon and Cerrado biomes. Key datasets: INPE PRODES/DETER, MapBiomas land cover (1985–present), CAR rural property cadastre (~6 M polygons), SIGEF certified parcels, FUNAI indigenous territories, IBAMA embargo and fine records.

Canada — boreal forest. Key datasets: NRCan land cover, CPCAD protected areas, ISC First Nations reserve boundaries, CanDCS-U6 downscaled climate at 10 km. Permafrost extent from Arctic-CORDEX ensemble.