TorchGeo embeddings
Notes on
EARTH EMBEDDINGS AS PRODUCTS: TAXONOMY, ECOSYSTEM, AND STANDARDIZED ACCESS
Heng Fang† ∗ Adam J Stewart‡ ∗ Isaac Corley§ * Xiao Xiang Zhu * Hossein Azizpour† † KTH Royal Institute of Technology, Stockholm, Sweden ‡ Chair of Data Science in Earth Observation, Technical University of Munich, Munich, Germany § Wherobots, San Antonio, USA
arXiv:2601.13134v1 19 Jan 2026 [LINK](https://arxiv.org/abs/2601.13134)
Introduction
• A comprehensive survey that organizes existing geospatial embedding products into a structured taxonomy and provides a detailed metadata atlas (resolution, license, etc.). • Unified Integration: implements standardized data loaders for these embeddings in [TorchGeo](https://www.osgeo.org/projects/torchgeo/)
An overview landscape is proposed : a) Analysis Frameworks & Tools; b) Embeddings data artifacts; c) charting downstream application value, specifically mapping tasks and retrieval tasks. Embeddings are differentiated as either location-typed, patch-typed or pixel-typed. Details of existing products are shown.
🌍 1. Foundation Models for Earth Observation (EO)
These are the leading projects that aim to build general-purpose models capable of representing Earth from satellite imagery and other geospatial modalities.
🔧 Projects:
- Clay Foundation Model – [Hugging Face](https://huggingface.co/made-with-clay/Clay) 2024
*A multimodal foundation model for Earth using diverse data sources.*
- Major TOM – [AFrancis IGARSS 2024](https://huggingface.co/Major-TOM)
*Expandable datasets and models for global EO coverage.*
- Earth Index Embeddings – [Earth Genome](https://www.earthgenome.org/earth-index), 2025
*A large-scale embedding system built from Earth observation data.*
- Copernicus-Embed – [Zhu et al., AI4Copernicus Project](https://github.com/zhu-xlab/Copernicus-FM)
*Foundation model leveraging Copernicus Sentinel data.*
- Presto Embeddings – [NASA Harvest](https://nasaharvest.github.io/presto-embeddings/)
*Embedding framework for satellite time series and land use analysis.*
- Tessera Embeddings – [GeoTessera Docs](https://geotessera.readthedocs.io/en/latest/) [REPO](https://github.com/ucam-eo/tessera)
*pixel-based Temporal spectral embeddings for Earth representation.*
- Google Satellite Embedding (AlphaEarth) – [Google Earth Engine](https://developers.google.com/earth-engine/datasets/catalog/GOOGLE_SATELLITE_EMBEDDING_V1_ANNUAL)
*An early-stage embedding model using Google's global satellite data.*
- OlmoEarth – [AllenAI, 2025](https://allenai.org/olmoearth)
*Latent image modeling approach for multimodal Earth observation.*
📚 Key Papers:
- XXZhu 2025 [LINK](https://www.nature.com/articles/s43247-025-03127-x) – “On the Foundations of Earth Foundation Models” - CFBrown 2025 [LINK](https://arxiv.org/abs/2507.22291) – “AlphaEarth Foundations” - KKlemmer 2023 [LINK](https://arxiv.org/abs/2311.17179) – “SatCLIP: Global Location Embeddings with Satellite Imagery”
---
📦 2. **Datasets**
Large-scale, open-access datasets play a central role in training and evaluating Earth foundation models.
🗃 Datasets:
- EuroSAT – [Zenodo](https://zenodo.org/records/7711810)
*Land use classification dataset using Sentinel-2 satellite data.*
- EuroCrops – [PMC 10495462](https://pmc.ncbi.nlm.nih.gov/articles/PMC10495462/)
*Crop type mapping dataset for Europe.*
- National Land Cover Database (NLCD) – Photogrammetric Engineering & Remote Sensing 2001 [LINK](https://www.ingentaconnect.com/content/asprs/pers/2004/00000070/00000007/art00005;jsessionid=2awgc0agiboi4.x-ic-live-02)
*USA land cover classes*
- SSL4EO-S12 – IEEE Geoscience and Remote Sensing 2023 [LINK](https://github.com/zhu-xlab/SSL4EO-S12)
*Multimodal, multitemporal dataset for self-supervised learning.*
- Copernicus-Pretrain – [IEEE Geoscience and Remote Sensing 2023](https://github.com/zhu-xlab/Copernicus-FM)
*an extension of the SSL4EO-S12 dataset to all major Sentinel missions (S1-S5P)*
---
🧠 3. **Models & Methods**
These include both classical and cutting-edge machine learning approaches used in building Earth foundation models.
⚙ Core Methods:
- SatCLIP** – AAAI 2025 etc [LINK](https://arxiv.org/abs/2311.17179)
*Vision-language model for global location representations.*
- MMEarth** – [EU/CV 2024](https://arxiv.org/abs/2405.02771)
*Multimodal pretext tasks for geospatial representation learning.*
- ResNet** – [KHe IEEE/CV 2016](https://viso.ai/deep-learning/resnet-residual-neural-network/)
*Baseline CNN architecture widely used in EO.*
- ConvNeXt V2** – [Woo et al., IEEE/CVF 2023](https://github.com/facebookresearch/ConvNeXt-V2)
*Efficient ConvNet architecture using masked autoencoders (MAE).*
- DINO, DINOv2, DINOv3** – INRIA 2021–2023, META [LINK](https://dinov3.org/)
*Vision transformers with self-supervised learning capabilities.*
- MAE (Masked Autoencoders)** – IEEE/CVF 2021 [LINK](https://arxiv.org/abs/2111.06377)
*Self-supervised learning for vision transformers.*
🧬 Distillation & Advanced Approaches:
- **Distillation methods** – Transfer knowledge from large models. - **Neural plasticity-inspired models** – [ZXiong, arXiv 2024]
*Inspired by biological learning mechanisms.*
- **Multi-label guided soft contrastive learning** – [YWang, IEEE TGRS, 2024] - **Barlow Twins** – [Zbontar et al., arXiv 2021]
*Method for learning representations without contrastive loss.*
- **Continual Barlow Twins** – [IEEE JSTARS, 2023]
*Extends Barlow Twins to continual learning in EO segmentation.*
---
🛠 4. **Tools & Benchmarks**
These are software systems and frameworks that support development, evaluation, or deployment of EO AI models.
🧰 Tools:
- TorchGeo** – [AJStewart ACM 2025](https://www.osgeo.org/projects/torchgeo/)
*PyTorch library for geospatial deep learning.*
- NeuCo-Bench** – [RVinge, arXiv 2025](https://arxiv.org/html/2510.17914v1)
*Benchmarking framework for neural embeddings in Earth observation.*
- GeoINRID** – [GitHub: arjunarao619/GeoINRID](https://github.com/arjunarao619/GeoINRID)
*Geospatial inference and representation learning toolkit.*
🏆 Challenges:
- **Embed2Scale Challenge** – [CVPR CAlbrecht 2025](https://research.ibm.com/publications/the-2025-cvpr-earthvision-data-challenge-by-embed2scale)
*Large-scale Earth vision challenge focused on scale-aware embeddings.*
- TerraMind Blue-Sky Challenge** – [JJakubik, arXiv 2025]
*Generative modeling for Earth observation.*
---
🧭 5. **Key Themes & Trends**
1. Foundation Models**: TorchGeo now includes data loaders designed for search/retrieval (Clay, Major TOM, Earth Index), and for dense prediction tasks like land cover mapping (Copernicus, Presto, Tessera, Google). TorchGeo allows us to enable fair, side-by-side benchmarking of different embedding models on the same downstream tasks, forming the basis for future experiments. Projects are encouraged to strengthen and improve explainability.
1.1 Major TOM Notes** Major TOM embeddings are not (yet) really product-oriented and are aimed with a similar purpose to the MT Core datasets - to make it easier to experiment and benchmark model outputs (hence, unlike TESSERA and AEF which came a few months after, MT embeddings do not have consistent or aggregated temporal scope). We haven't had enough time to finish off the preprint, but my current plan is to provide a simple MT Embedding benchmark at this year's EGU and integrate that into the arxiv pre-print. --Miko
1.2 Earth Index / Earth Genome** Use the Earth Index application (earthindex.ai) for non-technical users to use the embeddings we published on source.coop. Users of the web app (non-technical journalists, indigenous communities/allies, NGOs) have been our main focus. Users of the source.coop embeddings have generally been more technical folks interested in exploring/innovating in what's possible --BenStrong
1.3 Clay** Clay and Presto offer documented tutorials on generating new embeddings with their models. In CLAY, the encoder receives unmasked patches, latitude-longitude data, and timestep information. Notably, the last 2 embeddings from the encoder specifically represent the latitude-longitude and timestep embeddings.
2. Self-Supervised Learning (SSL)**:
3. Multimodal Integration**:
4. Open Data & Tools**: Open-source projects (e.g., TorchGeo, Copernicus-Embed) and public datasets (EuroSAT, EuroCrops) are crucial for reproducibility and democratization of EO AI. Projects are encouraged to increase Input Data Diversity, and to adopt cloud-native data formats for geospatial data.
5. Benchmarking**: Projects are encouraged to standardize in benchmarking. Benchmarks including NeuCo-Bench and Embed2Scale.
---
📌 Research Directions
- Unified Earth Foundation Models**: - Interpretability in EO AI**: Exploring how these embeddings can be interpreted by domain experts. - Ethics and Bias**: Investigating fairness and bias in global EO models trained on unevenly distributed data. - Edge Deployment**: Making these large foundation models deployable on resource-constrained platforms
(e.g., for field use).