Global forest typology at 10-meter resolution for forest and land-use monitoring.

Abstract

Distinguishing forest types-primary, naturally regenerating, planted, and plantation forests-from agricultural tree crops and other land uses is essential for carbon accounting, biodiversity assessment, conservation planning, and supply-chain regulation. However, no existing global dataset resolves this typology at high spatial resolution. We present the Forest Typology (ForTy) v1 dataset, a global 10-meter resolution baseline map for 2020 that classifies all land into six categories aligned with FAO and EU Deforestation Regulation (EUDR) definitions: Primary Forest, Naturally Regenerating Forest, Planted Forest, Plantation Forest, Tree Crops and Agroforestry, and Other Land. A cascaded deep learning pipeline, trained on 1.7 million globally distributed samples, generates per-class relative confidence maps from geospatial satellite embeddings by combining weakly supervised learning with active learning. Independent validation against 8,190 expert-labeled stratified random sites yields an overall accuracy of 90.2% for the six-class scheme, 94.8% for natural forest classification, and 95.5% for forest/non-forest classification.