Plantation and Natural Forest Mapping from Satellite and Drone Imagery
Mapping tree crops such as oil palm and separating natural forest from plantations, for yield, deforestation baselines and supply-chain rules such as the EUDR.
Tier Grades
Tier A: Frontier AI labs
feasible
Google DeepMind with WRI built the 10 m natural forest map at 92.2% accuracy.
Tier B: Indonesian AI labs
feasible
Sentinel-2 imagery is open and the 2020 natural forest map is public, so Indonesian labs can build on both.
Tier C: Institution units such as ITB
feasible
Mapping with open Sentinel-2 data and public baselines fits a university lab.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
feasible
EOSDA reports raising its PySawit oil palm model from 78% to up to 90% accuracy, and a public 10 m baseline supports EUDR proofs of value.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeNatural Forests of the World 2020 (MTSViT): compute not stated in the source.
- DataMTSViT: seasonal Sentinel-2 imagery plus elevation and slope over 1280 m patches; 10 m map at 92.2% accuracy against an independent validation set. PySawit: vendor reports accuracy raised from 78% to up to 90% using 90 m weather data.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): Not in the sources read (blog posts give no author list).
- Time horizonMonths.
- Shelf-life signalsThe EUDR 2020 cut-off makes a 2020 baseline map useful for years, but newer forest classes are planned by Google.
- Pipeline roleTier A publishes global baselines; Tier B and Tier C adapt them to Indonesian plantations; Tier D builds EUDR and yield proofs of value.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
Linked models
Links
R&D TopicPlantation and Natural Forest Mapping from Satellite and Drone Imagery
Models
Natural Forests of the World 2020 model (MTSViT)Google DeepMind, geospatial
PySawit (enhanced)EOS Data Analytics (EOSDA), geospatialAlphaEarth Foundations (Satellite Embedding v1 Annual)Google DeepMind, geospatialUAV oil palm (Roboflow)geospatial
Natural Forests of the World 2020 model (MTSViT)Google DeepMind, geospatialPySawit (enhanced)EOS Data Analytics (EOSDA), geospatialAlphaEarth Foundations (Satellite Embedding v1 Annual)Google DeepMind, geospatialUAV oil palm (Roboflow)geospatial
Models 4
Natural Forests of the World 2020 model (MTSViT)Google DeepMind, Google, World Resources Institute (WRI), IIASA, geospatialexemplarPySawit (enhanced)EOS Data Analytics (EOSDA), geospatialexemplarAlphaEarth Foundations (Satellite Embedding v1 Annual)Google DeepMind, geospatialused byUAV oil palm (Roboflow)geospatialproduced byPapers
- A high-resolution canopy height model of the EarthNico Lang et al., arXiv preprint, 2022 founding
- Very high resolution canopy height maps from RGB imagery using self-supervised vision transformer and convolutional decoder trained on Aerial LidarJamie Tolan et al., Remote Sensing of Environment 300, 113888, 2024, 2023 state of the art
- Mapping oil palm density at country scale: An active learning approachAndrés C. Rodríguez et al., Remote Sensing of Environment Volume 261, August 2021, 112479, 2021 state of the art
- Satellite reveals age and extent of oil palm plantations in Southeast AsiaOlha Danylo et al., arXiv preprint, 2020 state of the art
- ForestNet: Classifying Drivers of Deforestation in Indonesia using Deep Learning on Satellite ImageryJeremy Irvin et al., NeurIPS 2020 (per arXiv comment), 2020 state of the art
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- research.google/blog/separating-natural-forests-from-other-tr... https://research.google/blog/separating-natural-forests-from-other-tree-cover-with-ai-for-deforestation-free-supply-chains/
- How EOSDA Enhanced PySawit Model To Help Oil Palm Industry eos.com https://eos.com/blog/eosda-enhanced-pysawit-model-to-improve-its-accuracy/
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Plantation and Natural Forest Mapping from Satellite and Drone Imagery", https://aurion.semai.id/topics/plantation-and-forest-mapping/, records of October 2026.