Building the Atlas

A brief overview on the process of developing the Global Ecosystems Atlas

The Global Ecosystems Atlas synthesis map is built using a dual approach described in the dedicated technical paper:

  1. Ingesting existing geospatial data from various sources.

  2. Targeted spatial modelling to fill in gaps where existing data is lacking.

The map is classified according to the first three levels of the IUCN Global Ecosystem Typology: realm, biome and ecosystem functional group. The process is managed and driven by the science team at the Global Ecology Lab at James Cook University in close collaboration with key technical partners, including Esri, Google, and AI2.

Ingesting existing data

Existing ecosystem, vegetation, and land-cover datasets are prioritised and compiled into a central data catalogue through active searches and outreach. Datasets are evaluated for open-access licensing, recency, supporting documentation, validation, and spatial resolution. Eligible map classes are then cross-referenced to the IUCN typology following the guidelines developed by the IUCN.

A geospatial processing pipeline, developed by Esri, drives the integration of these validated source maps into a standardised 100m grid, establishing a unified synthesis product. Where multiple sources exist on the same pixel, an overlap index based on the evaluation criteria determines the stacking order of source layers, with the most representative and authoritative source prioritised in the synthesis map.

Filling data gaps

For areas unlikely to be mapped in the near future, new ecosystem maps are generated to fill these gaps using geospatial foundation models provided by Google (AlphaEarth) and Ai2 (OlmoEarth). These models are trained on EcoTrain, a stratified dataset of ecosystem occurrence records annotated by experts in the Global Ecosystem Typology. Predictions from these models are generated at 100m resolution and directly ingested into the synthesis pipeline to achieve contiguous global coverage.

The combined outputs are delivered as cloud-optimised GeoTIFFs containing multi-class ecosystem layers, overlaying and underlaying "standalone" ecosystems (e.g., pelagic and subterranean ecosystems), data masks, agreement layers, source metadata as well as dataset attribution information.