MapBiomas Argentina Fire

Annual burned area mapping for Argentina using Landsat satellite imagery, produced by the MapBiomas Argentina team.

---

Collections

Collection	Folder	Status	Coverage	Model
Collection 0	collection-00/	Complete	Patagonia	Logistic regression (GEE JS + R)
Collection 1	collection-01/	In development	All regions	Regularized logistic regression (GEE Python API + R glmnet)

See each collection's README for what is where:

• collection-00/README_00.md — the completed pilot, with its ATBD
• collection-01/README.md — the map of the active collection

Before delving into the repo structure, we suggest reading the Algorithm Theoretical Basis Document of the latest collection (Collection 1).

---

Remotes

This repo has two git remotes:

Remote	URL	Role
origin	barberaivan/mapbiomas-argentina-fire	source of truth — every commit and push lands here, at whatever frequency work happens
mapbiomas	mapbiomas/argentina-fire	the official, public-facing mirror. Updated only explicitly and at low frequency (e.g. at milestones), never as a side effect of a normal commit/push

To publish a snapshot to the official repo: git push mapbiomas main.

---

Getting started (first-time setup)

This repository holds code only. The heavy data — training samples, fitted models, CV metrics, prediction plots — is not in git. It lives in a separate folder called mapbiomas-arg-fire-store (the "store"), and a small script links it into the repo.

Why? Code belongs in git/GitHub (which versions and backs it up); large data files do not. Keeping them apart avoids a bloated, slow git history and the sync conflicts that arise when a cloud-sync tool and git fight over the same files.

Setup is three steps:

1. Get the code

git clone <this-repo-url>
cd mapbiomas-arg-fire

2. Get the data store

Download it once and remember where you put it.

Google Drive (anyone): open the store folder → https://drive.google.com/drive/folders/1hmWZWkX1iHcGAGvqtv-5JSynvLnPx7wX → download it. Your browser saves it as a .zip; unzip it somewhere stable (e.g. your home folder). You should end up with a folder named mapbiomas-arg-fire-store containing collection-00/ and collection-01/ inside.

3. Link the store into the repo

From the repo root, run setup.sh once, giving it the path to the store you just got and — optionally — the path to your GEE Python interpreter:

# store only (Python falls back to system `python3`)
./setup.sh /full/path/to/mapbiomas-arg-fire-store

# store + the GEE venv you run pipeline scripts with (recommended)
./setup.sh /full/path/to/mapbiomas-arg-fire-store /full/path/to/venv/bin/python

That's it. The script creates the symlinks and remembers both paths (in a local, gitignored .local-paths file), so any later re-run is just ./setup.sh with no argument.

Why give it the Python path? Each machine's venv lives somewhere different, so the docs never hardcode it — they say "run with $PYTHON". setup.sh records your interpreter as $PYTHON in .local-paths (for your shell — source .local-paths) and in .claude/settings.local.json (so Claude Code's Bash uses the same one). Both files are gitignored, so two machines can each point $PYTHON at their own venv with nothing committed.

To confirm it worked:

ls collection-01/data          # should list training_observations_*.csv etc.
ls collection-01/models-store  # should list class_*_fit.rds etc.

Heads-up for contributors: because the data is outside git, uncommitted code is backed up nowhere — get in the habit of git commit && git push often. Work on one machine at a time, and git pull before you start. Collaborators without write access to the store upload any exported files to it manually.

Symlinks require Linux or macOS (or Windows with WSL / Developer Mode enabled).

Once set up, collection-01/docs/00-overview.md is where to start reading: the method in one page, and which step is which. Each step's own doc carries the commands to run it.

---

Regions

Code	Region
BA	Bosque Atlántico
CHACO	Chaco
PAMPA	Pampas
CUYO	Monte, Puna y Altos Andes
PAT	Patagonia

---

Documentation

• ATBD (methodology): collection-01/ATBD/build/mapbiomas-argentina-fire-atbd-col01.pdf — the collection 1 document, in English (draft; source in collection-01/ATBD/). The pilot's is collection-00/docs/.../mapbiomas_fire_argentina_atbd_pilot_2025.pdf.