News
European technologies in security and space
Commissioner Andrius Kubilius' visit to CloudFerro highlights the importance of European technologies for Europe's security and operational readiness
AI for Earth Observation
From pixels to intelligence. On-platform.
Build and deploy AI models on satellite data at scale with embeddings, training datasets, and anomaly alerts processed on CREODIAS. No data transfer, faster time-to-model, and lower expertise barrier.
- 40 billion embeddings published
- Zero data egress
- Open models, or bring your own
Solution Overview
AI models on satellite data at scale
AI in Earth Observation is an AI layer built on CREODIAS that covers analysis-ready data, precomputed embeddings, anomaly monitoring and training-dataset tooling, running on CloudFerro GPUs next to the EODATA archive so the pipeline never leaves one platform.
Embeddings from Sentinel-1 GRD and Sentinel-2 L1C/L2A, delivered as GeoParquet, COG or ZARR
Anomaly detection on embedding time series, set up per area of interest as a subscription
AI-ready training data: ready to use training datasets in EODATA archive
Analysis-ready inputs: Sentinel-2 cloud masks, Sentinel-1 RTC and ZARR v3 data cubes
Your benefits
Spend your time on the model, not the pipeline
The hard part of AI for Earth observation is rarely the model. It's everything that comes before it. Here's what changes when that part is done for you.
Your data never leaves the platform
Data access, embedding generation, training and inference all run inside CREODIAS. No egress bills, no transfer delays, no copies to track across providers.
Embeddings in minutes
Precomputed embeddings and ready-made datasets remove the slowest steps in an AI-for-EO workflow. Your team works on the model instead of wrangling data.
Useful without an ML team
Precomputed embeddings, freemium datasets and no-code dataset generation give analysts and domain experts a way in, without a machine learning background.
One pilot or a whole country
The same processors run a single-scene test and a country-scale operational pipeline. Start small on trial credits, then scale on the same stack.
European data, European rules
Built on CREODIAS, in EU data centres, run by an EU company. No transatlantic transfers which matters when a tender makes sovereignty a hard requirement.
Use Cases
Where satellite AI is applied
The scenarios built where satellite data becomes a decision quickly, repeatedly and at scale.
Early warning for environmental change
Subscribe to anomaly detection over your areas of interest and get flagged when the signal shifts — deforestation, flooding, burn scars and land change — without building a monitoring system.
Infrastructure and border monitoring
Track change at the sites that matter — infrastructure, borders, critical assets — across Sentinel-1 and Sentinel-2 time series, with processing kept on EU infrastructure.
Crop, yield and land cover modelling
Use embeddings as ready-made features for crop-type mapping, yield prediction, land cover change and agricultural risk. Train on signals already extracted, not raw scenes.
Finding images by description
Search large EO archives with plain text or a reference image, using multimodal models and precomputed FAISS indices. Available now as a research demo.
Architecture
Four layers, from archive to insight
Each layer feeds the next. Raw Copernicus data goes in at the bottom; embeddings, alerts and training datasets come out at the top.
Start from the archive
Sentinel-1, Sentinel-2, VHR, DEM and land cover data are read straight from the CREODIAS EODATA repository. No download, no staging copy.
Make it analysis-ready
Core processors mask clouds in Sentinel-2, terrain-correct Sentinel-1 on demand and build ZARR v3 data cubes — the clean inputs reliable AI depends on.
Generate embeddings
Pick a model from the catalogue (TorchGeo and TerraTorch backends) or bring your own. Set the area and time range; embeddings arrive as GeoParquet, COG or ZARR.
Monitor, search, inferface
Anomaly detection and semantic search run on the precomputed embeddings. Optional: an inference API, Sentinel-2 super-resolution and VHR object detection.
All you need included
Standard
- Embeddings Processor — patch-based and TESSERA pixel-based, Sentinel-2 and Sentinel-1 GRD
- Precomputed generic embeddings
- Anomaly Detection Service (subscription, AOI-based)
- AI Training Datasets Portfolio
- MajorTOM-as-a-Service (beta)
- Cloud Mask Processor (Sentinel-2, SEnSeIv2)
- Data Cubes Processor (S1, S2, ESA Land Cover, DEM — ZARR v3)
Optional add-ons
- Bring-Your-Own-Model (BYOM) embedding computation
- Super-Resolution Processor (Sentinel-2)
- VHR/UAV embeddings and object detection
- Inference Models API
- S1 RTC/ARD/NRB Processor
Components
Built on CloudFerro products
Earth Observation data catalogue
Access and process extensive EO datasets directly in the cloud, with fast discovery, flexible interfaces and no need to download.
Storage
Object Storage
Manage object storage containers and buckets for S3/Swift-compatible data access.
Compute
GPU Services
Run AI, ML, and HPC workloads on high-performance GPU resources in CloudFerro cloud.
EO Products & Services
Managed Notebooks
Start working in a managed JupyterLab environment without setup. Use preconfigured ML and data science libraries, choose CPU or GPU-backed notebooks.
Customer stories
Already proven at planetary scale
The embeddings technology behind this solution has already processed the Copernicus archive at global scale — openly, and together with ESA.
First global AI embeddings dataset with ESA Φ-lab
With ESA Φ-lab, we turned 147 TB of Copernicus data into over 40 billion AI embeddings. It’s the first open, global dataset of its kind, free on CREODIAS and Hugging Face.
Application anomaly watch
Anomaly Watch flags land-surface change, from fires and floods to new construction, by comparing each location against its own seasonal history.
Application similarity search
Type “flooded fields” or “solar farm” and get matching Sentinel-2 imagery from across Europe in seconds. You don’t need coordinates or filters.
Compliance
Certified at every level
ISO/IEC 27001/27017/27018
Standards for security, privacy and data protection in cloud environments.
Attestation C5
Independently audited cloud security aligned with BSI requirements.
BSI-200-1
Information security based on the established German BSI IT-Grundschutz standard.
EU Facility Security Clearance – Level III
Authorisation to protect EU classified information up to EU SECRET level.
Frequently Asked Questions
AI in Earth Observation
What ML engineers, architects and project leads want to know before they run a pilot. If your question isn't here, our data scientists will answer it.
Other solutions
Payload Data Ground Segment as a Service
CloudFerro PDGS as a Service enables the full lifecycle of satellite payload data, from ingestion and processing to archiving and user delivery.
EO Data Platform
Accelerate cloud innovation with automated, secure, and cost-controlled sandbox environments.
Area Monitoring System as a Service
A next step after the Sen4CAP solution and the product features were developed based on it.