Earth Observation Foundation Models
Models pretrained on large satellite and aerial archives that are fine-tuned for land cover, crops, floods and other mapping tasks.
Tier Grades
Tier A: Frontier AI labs
feasible
IBM with NASA, Google DeepMind and Ant Group pretrain EO models on archives such as 4.2 million HLS samples.
Tier B: Indonesian AI labs
feasible
Open Prithvi-EO-2.0 weights and the open HLS archive allow fine-tuning on Indonesian land cover.
Tier C: Institution units such as ITB
feasible
Fine-tuning a model of up to 600M parameters on labelled local tiles fits a university remote sensing lab.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
feasible
AlphaEarth annual embeddings and open Prithvi weights fit plantation and land monitoring proofs of value with few labels.
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
- ComputePrithvi-EO-2.0: up to 600M parameters; training compute not stated in the abstract.
- DataPrithvi-EO-2.0: 4.2 million global time-series samples from NASA Harmonized Landsat and Sentinel-2 at 30 m; AlphaEarth Foundations releases annual embedding layers for 2017 to 2024.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 36 authors on the Prithvi-EO-2.0 paper (arXiv 2412.02732); 19 on AlphaEarth Foundations.
- Time horizonYears.
- Shelf-life signalsPrithvi-EO-2.0 beats Prithvi-EO-1.0 by 8% on GEO-Bench and six other geospatial foundation models.
- Pipeline roleTier A pretrains and releases EO models and embeddings; Tier B and Tier C fine-tune them for Indonesian land; Tier D builds monitoring proofs of value on them.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Exemplars
Observed actors
IBM
NASA
Google DeepMind
Ant Group
Aerospace Information Research Institute, Chinese Academy of Sciences (AIRCAS)
Clay Foundation (made-with-clay)
European Space Agency (ESA)
Meta
Stanford UniversityLinked models
AlphaEarth Foundations (Satellite Embedding v1 Annual)geospatialexemplarPrithvi-EO-2.0geospatialexemplarSkySensegeospatialexemplarDINOv3 SAT-493M ViT-7Bgeospatialproduced byDINOv3 SAT-493M ViT-L (distilled)geospatialproduced byRingMo-Agentgeospatialproduced bySkySense V2geospatialproduced byRingMoEgeospatialproduced byTerraMindgeospatialproduced byTerraCodecgeospatialproduced byPrithvi-EO-2.0 100M-TLgeospatialproduced byPrithvi-EO-2.0 300Mgeospatialproduced byPrithvi-EO-2.0 600Mgeospatialproduced byPrithvi-EO-2.0-600M-TLgeospatialproduced byPrithvi-EO-2.0 TINY-TLgeospatialproduced byRingMo-Aerialgeospatialproduced byClaygeospatialproduced byRingMo-litegeospatialproduced byPrithvi-EO 1.0geospatialproduced bySatMAEgeospatialproduced byRingMo familygeospatialproduced bySkySense-Ogeospatialproduced byGranite-Geospatial-Oceangeospatialproduced byPrithvi v1geospatialproduced byLinks
R&D TopicEarth Observation Foundation Models
ModelsPrithvi-EO-2.0IBM, geospatialAlphaEarth Foundations (Satellite Embedding v1 Annual)Google DeepMind, geospatialSkySenseAnt Group, geospatialRingMo familyAerospace Information Research Institute, Chinese Academy of Sciences (AIRCAS), geospatial+20 more
Prithvi-EO-2.0IBM, geospatialAlphaEarth Foundations (Satellite Embedding v1 Annual)Google DeepMind, geospatialSkySenseAnt Group, geospatialRingMo familyAerospace Information Research Institute, Chinese Academy of Sciences (AIRCAS), geospatial+20 more
Models 24
Prithvi-EO-2.0IBM, NASA, Julich (Forschungszentrum Julich), geospatialexemplarPrithvi-EO-2.0 600MIBM, NASA, geospatialproduced byRingMo familyAerospace Information Research Institute, Chinese Academy of Sciences (AIRCAS), geospatialproduced byPrithvi-EO-2.0 300MIBM, NASA, geospatialproduced byPrithvi-EO-2.0-600M-TLIBM, NASA, geospatialproduced bySkySenseAnt Group, Wuhan University, MYBank, geospatialexemplarDINOv3 SAT-493M ViT-7BMeta, geospatialproduced byDINOv3 SAT-493M ViT-L (distilled)Meta, geospatialproduced byPrithvi v1IBM, NASA, geospatialproduced byPrithvi-EO 1.0IBM, NASA, geospatialproduced byPrithvi-EO-2.0 100M-TLIBM, NASA, geospatialproduced byPrithvi-EO-2.0 TINY-TLIBM, NASA, geospatialproduced by+12
TerraMindIBM, European Space Agency (ESA), Julich (Forschungszentrum Julich), geospatialproduced byAlphaEarth Foundations (Satellite Embedding v1 Annual)Google DeepMind, geospatialexemplarGranite-Geospatial-OceanIBM, geospatialproduced byRingMo-AerialAerospace Information Research Institute, Chinese Academy of Sciences (AIRCAS), geospatialproduced byRingMo-AgentAerospace Information Research Institute, Chinese Academy of Sciences (AIRCAS), geospatialproduced byRingMo-liteAerospace Information Research Institute, Chinese Academy of Sciences (AIRCAS), geospatialproduced byRingMoEAerospace Information Research Institute, Chinese Academy of Sciences (AIRCAS), geospatialproduced bySkySense V2Ant Group, geospatialproduced byClayClay Foundation (made-with-clay), geospatialproduced bySatMAEStanford University, geospatialproduced bySkySense-OAnt Group, geospatialproduced byTerraCodecIBM, geospatialproduced byPapers
- SatMAE: Pre-training Transformers for Temporal and Multi-Spectral Satellite ImageryYezhen Cong et al., NeurIPS 2022 (per arXiv comment), 2022 founding
- AlphaEarth Foundations: An embedding field model for accurate and efficient global mapping from sparse label dataChristopher F. Brown et al., arXiv preprint, 2025 state of the art
- Neural Plasticity-Inspired Multimodal Foundation Model for Earth ObservationZhitong Xiong et al., arXiv preprint, 2024 state of the art
- Lightweight, Pre-trained Transformers for Remote Sensing TimeseriesGabriel Tseng et al., arXiv preprint, 2023 state of the art
- On the Foundations of Earth and Climate Foundation ModelsXiao Xiang Zhu et al., arXiv preprint, 2024 survey
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- arxiv.org/abs/2412.02732 https://arxiv.org/abs/2412.02732
- arxiv.org/abs/2507.22291 https://arxiv.org/abs/2507.22291
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Earth Observation Foundation Models", https://aurion.semai.id/topics/earth-observation-foundation-models/, records of October 2026.