# Maldives Ecosystem Maps: Per-Island IUCN Ecosystem Typology Segmentation from Sentinel-2 and PlanetScope (Ai2 Global Ecosystems Atlas)

## Facts

- Provider: Allen Institute for AI (Ai2), Global Ecosystems Atlas Maldives Accelerator
- License: ODC-By-1.0 (https://github.com/allenai/maldives_ecosystem_mapping)
- Format: TIFF (in sampled files)
- Stored at: Google Cloud Storage, gs://ai2-earthsystem-release/Release/GlobalEcosytemAtlas/latest/Maldives (us)
- Access: public, readable without an account
- Category: Earth Observation & Geospatial
- Source: https://github.com/allenai/maldives_ecosystem_mapping
- Page: https://scigantic.com/archive/7a6d23bd-3b05-5dd0-b490-ceb60a9f51f4
- Keywords: Earth Observation & Geospatial, image segmentation, satellite imagery, TIFF, Google Cloud Storage

## About

Predicted ecosystem maps for the islands of the Maldives, released by the Allen Institute for AI (Ai2) for the GEO Global Ecosystems Atlas Maldives Accelerator. A segmentation model trained on vector annotations of island imagery assigns each pixel one of 16 classes from the IUCN Global Ecosystem Typology, including photic coral reefs, seagrass meadows, subtidal sand beds and rocky reefs, intertidal forests and shrublands, coastal saltmarshes, sandy, rocky and artificial shorelines, freshwater lakes, intermittently closed lagoons, permanent marshes, croplands, plantations and urban areas (ecosystem_mapping.json gives the class ids). Per the README, one model run on Sentinel-2 covers 1,247 islands and one on PlanetScope covers 1,419 islands, each written as a single-band GeoTIFF per island named after the Maldives island GeoJSON (or its FCODE). Listed 2026-11-19: the folder holds sentinel2/, planetscope/, maxar/ and labels/ subfolders. Training annotations, the rslearn training dataset and model checkpoints are under Release/GlobalEcosystemSegmenter/202410/. Read with rasterio or QGIS. Ai2 licenses the map predictions under ODC-By 1.0 (annotations CC0); cite the Ai2 maldives_ecosystem_mapping release.

## Load the data

```python
import os
import rasterio
with rasterio.open(os.path.join(os.environ['SCIGANTIC_MOUNT_PATH'], 'labels/baarah.tif')) as src:
    print(src.profile)
    band = src.read(1, window=((0, 512), (0, 512)))
```

Run in a Scigantic notebook on 2026-11-20 against `labels/baarah.tif`: rasterio: 1 band(s) 3116x3136 uint8, crs EPSG:32643, driver GTiff.

From any machine:

```python
import gcsfs
fs = gcsfs.GCSFileSystem(token='anon')
fs.ls('ai2-earthsystem-release/Release/GlobalEcosytemAtlas/latest/Maldives/', detail=False)[:10]
```

Lists the files with anonymous access. Runs on any machine, with no cloud account.

## Data structure

File types in sampled files: TIFF

Example file: `Release/GlobalEcosytemAtlas/latest/Maldives/labels/baarah.tif`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/7a6d23bd-3b05-5dd0-b490-ceb60a9f51f4
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/7a6d23bd-3b05-5dd0-b490-ceb60a9f51f4
- Over MCP: `get_schema_card` with id `7a6d23bd-3b05-5dd0-b490-ceb60a9f51f4` at https://api.scigantic.com/mcp

## Related topics

- [Earth Observation & Geospatial datasets](https://scigantic.com/topics/earth-observation-and-geospatial)
- [Satellite imagery datasets](https://scigantic.com/topics/satellite-imagery)
- [TIFF datasets](https://scigantic.com/topics/tiff)
- [Image segmentation datasets](https://scigantic.com/topics/image-segmentation)
