Earth & environment · OECD FoS 1.5 · China MoE 0705 地理学 / 0816 测绘科学与技术
Geography & remote sensing
地理与遥感
L2Local breakthroughs ↑ Rising Score 2.10 · Last reviewed 2026-10-21 · Next review 2027-01-19
BIGGEST BREAKTHROUGH 2025-07-30 AI · majorField · major
AlphaEarth Foundations global embeddings ↗
10 m global annual embeddings
Why this level
Geospatial foundation models are replacing per-task models: NASA/IBM open-sourced Prithvi; DeepMind released AlphaEarth Foundations global embeddings.
S = 0.30·method + 0.35·discovery + 0.25·adoption + 0.10·education (education scores 0 when there is no evidence; if empty, the other weights are renormalised)
Sources
- huggingface.co/ibm-nasa-geospatial
- deepmind.google/discover/blog/alphaearth-foundations-helps-map-our-planet-in-unprecedented-detail
Composite 1.5–2.49: at least one peer-reviewed/officially verified AI-key result, or one dimension at 3; routine workflow largely unchanged. All levels →
Research directions
Geospatial foundation modelsLand-cover changeDisaster remote sensing
Papers
- TerraMind: Large-Scale Generative Multimodality for Earth Observation
EO foundation models go multimodal-generative
2025-04-15 · arXiv
- SatMAE: Pre-training Transformers for Temporal and Multi-Spectral Satellite Imagery
Early remote-sensing foundation model
2022-07-17 · arXiv
- Dynamic World, Near real-time global 10 m land use land cover mapping
Land-cover mapping moves from annual to per-overpass
2022-06-09 · Scientific Data
- Continental-Scale Building Detection from High Resolution Satellite Imagery
Building maps for under-mapped regions
2021-07-26 · arXiv
- Towards Consistent Geospatial Representations: Aligning One Encoder to Many Foundation Models
Shows geospatial foundation models are complementary and aligns one encoder to many, matching the best average rank.
2026-10-20 · arXiv · Arthur Ouaknine, David Rolnick
- MatchCityLoc: Aerial Localization in LoD City Models via Surface Normal Matching
Matches surface normals from a zero-shot model to LoD city models for UAV localisation without GNSS.
2026-10-19 · arXiv · Oussema Dhaouadi, Zuria Bauer, Johannes Michael Meier, Marc Pollefeys
- PiCo: Physics-Informed Concept Probing for Geo-Foundation Models
Uses synthetic radiative-transfer imagery to find physical concept vectors that explain geo-foundation-model predictions.
2026-10-14 · arXiv · Yihang She, Andrew Blake, Jianbo Qi, Srinivasan Keshav
- Revisiting the Task of Building Damage Assessment by Satellite Imaging
Revisits satellite building-damage assessment, arguing the field lacks a shared foundation despite a decade of work.
2026-10-09 · arXiv · Anthony Cioppa, Virgile Blua, Sébastien Piérard, Pietro Ceccato
- Shortcut Learning in Geospatial Foundation Model Embeddings for Cross-Disaster Damage Assessment
Holding out whole disasters cuts balanced accuracy of geo-foundation-model embeddings from 0.807 to 0.638 (xView2) and 0.818 to 0.568 (BRIGHT).
2026-10-09 · arXiv · Ali Shibli, Andrea Nascetti, Yifang Ban
- FairRSFM: A Biome-Aware Benchmark and Debiasing Framework for Remote Sensing Foundation Models
A biome-aware benchmark exposing performance disparities of remote-sensing foundation models across ecological regions.
2026-10-05 · arXiv · Md Aminur Hossain, Omkumar Vaghasiya, Rajeev Ranjan Dwivedi, Vinod Kurmi
- T-JEPA: A Temporal Joint-Embedding Predictive Architecture for Learning Better Remote Sensing Representations
T-JEPA learns time-gap-conditioned latent transitions from irregular Earth-observation sequences.
2026-10-05 · arXiv · Bowen Peng, Li Liu, Yongxiang Liu, Weijie Li
- HyperSAM: A Promptable Foundation Model for Hyperspectral Remote Sensing
A promptable foundation model for hyperspectral remote sensing.
2026-09-29 · arXiv · Li Pang, Xinqiao Wu, Jing Yao, Pedram Ghamisi
- From arXiv, bioRxiv, medRxiv and OpenAlex (last ~30 days); summaries written by AIMap from the abstract.
Milestones 10
Big glowing nodes are leaps: AI leap cool · Field leap warm · double ring = both
-
Model
KAIST GeoCore-9B: 9B-parameter geo-aware generative EO foundation model (NeurIPS 2026)AI: trained from scratch on 10M satellite image–text pairs with location and GSD conditioning; Field: beats task-specific models on cloud removal and SAR-to-optical
KAIST / EurekAlert!Preprint -
Model
Tessera open EO foundation model peer-reviewed at CVPR 2026; ESA announces global 10 m embeddings 2017–2025AI: open fused SAR/optical time-series embeddings, an open alternative to AlphaEarth; Field: global-scale mapping on a laptop, already used for UK habitat monitoring
ESA -
Deployment
NASA: Prithvi becomes the first AI geospatial foundation model in orbitEO foundation models move to on-board inference
NASA Science -
Model
ESA & IBM TerraMind: open any-to-any generative EO foundation model (ICCV 2025)EO foundation models go multimodal-generative
arXiv / ESAPreprint -
System
IBM and NASA release global Prithvi-EO-2.0 (600M parameters)Open geospatial model goes global
IBM Research -
System
NASA & IBM open-source Prithvi geospatial modelLargest open geospatial foundation model (date confirmed by IBM newsroom)
Hugging FaceTo be verified -
Paper
SatMAE: pre-training transformers for temporal and multi-spectral satellite imageryEarly remote-sensing foundation model
arXivPreprint -
Dataset
Dynamic World: near-real-time global 10 m land cover with deep learning (Scientific Data)Land-cover mapping moves from annual to per-overpass
Scientific Data -
Dataset
Open Buildings: continental-scale building detection from satellite imagery (arXiv 2107.12283)Building maps for under-mapped regions
arXivPreprint
No milestones match this filter.
More papers & repos 5
| Type | Title | Date |
|---|---|---|
| Paper | TerraMind: Large-Scale Generative Multimodality for Earth Observation Preprint | 2025-04-15 |
| Paper | SatMAE: Pre-training Transformers for Temporal and Multi-Spectral Satellite Imagery Preprint | 2022-07-17 |
| Paper | Dynamic World, Near real-time global 10 m land use land cover mapping | 2022-06-09 |
| Paper | Continental-Scale Building Detection from High Resolution Satellite Imagery Preprint | 2021-07-26 |
| Repo | ibm-nasa-geospatial (Hugging Face) To be verified | — |
Latest news 5
- Putting Earth AI to work: Geospatial agents for our planet's biggest challengesGoogle Research Blogen
- Unlocking Earth AI’s planetary geospatial foundation models for global public healthGoogle Research Blogen