What does Tessera's AI foundation model actually do?

It processes optical and SAR data from ESA's Sentinel satellites to generate embeddings — high-dimensional numerical vectors that help AI models efficiently identify surface features like crop changes, urban growth, and flood extents. Once pre-trained, it can be fine-tuned for specific downstream tasks without building a model from scratch.

Why does Tessera use both Sentinel-1 and Sentinel-2 data?

Sentinel-2 provides high-resolution multispectral optical imagery ideal for land-use and agricultural monitoring, while Sentinel-1 uses SAR radar that can penetrate cloud cover for all-weather observation. Combining both sources improves model accuracy under complex weather conditions.

Who benefits from this kind of Earth observation foundation model?

Government agencies, academic institutions, and private companies working in geospatial analysis all stand to benefit. They can fine-tune the pre-trained model for their specific applications — such as disaster monitoring or vegetation tracking — saving significant time and computing resources compared to training from scratch.

Tessera AI Foundation Model Takes on the Earth…

Tessera AI Foundation Model Takes on the Earth Observation Sector

Tessera has launched an AI foundation model purpose-built for Earth observation, integrating optical imagery and Synthetic Aperture Radar (SAR) data from ESA's Sentinel satellite constellation to generate high-fidelity embeddings. The model aims to lower barriers for geospatial analysis across government, academic, and commercial applications.

20 days ago

Highlights

Tessera has launched an AI foundation model for Earth observation that integrates ESA Sentinel-1 SAR and Sentinel-2 optical satellite data.

The model generates embeddings — high-dimensional vectors — enabling efficient AI recognition of surface features such as floods, urban expansion, and crop changes.

By fusing radar and optical data sources, the model maintains high accuracy even under cloudy or complex weather conditions.

As a foundation model, it can be fine-tuned for diverse downstream geospatial tasks, reducing training costs and deployment time for end users.

Government agencies, academic researchers, and commercial operators can leverage the model without building remote sensing AI systems from scratch.

Tessera has unveiled an AI foundation model designed specifically for Earth observation, combining optical imagery and Synthetic Aperture Radar (SAR) data from the European Space Agency's (ESA) Sentinel satellite series to generate highly detailed training data in the form of embeddings. The development marks a new technical milestone for the remote sensing and geospatial analysis sectors.

What Are Embeddings?

Embeddings are a foundational concept in machine learning, referring to the process of converting raw data — such as imagery or text — into high-dimensional numerical vectors. This representation allows AI models to learn and identify surface features more efficiently, including changes in farmland, urban expansion, flood conditions, and vegetation cover.

The Advantage of Sentinel Data

The Sentinel satellite series forms the backbone of the European Union's Copernicus Programme, providing freely accessible global Earth observation data that includes:

Optical imagery (Sentinel-2): High-resolution multispectral imagery suited for land-use and agricultural monitoring.
Radar data (Sentinel-1): SAR imagery capable of penetrating cloud cover, enabling all-weather surface monitoring.

Tessera's foundation model fuses both data sources, allowing the model to maintain high accuracy even under complex weather conditions.

Industry Significance of the Foundation Model

The foundation model concept — popularized by large language models such as the GPT series — offers a key advantage: once pre-trained, the model can be fine-tuned for a wide range of downstream tasks, significantly reducing the time and computational cost required for individual application development.

For the Earth observation industry, Tessera's approach offers several concrete benefits:

Government agencies, academic institutions, and private enterprises can deploy remote sensing analysis applications more rapidly.
Organizations no longer need to train models from scratch, saving substantial computing resources.
Universal cross-task embeddings facilitate data sharing and comparison across different application scenarios.

Outlook

Tessera's Earth observation AI foundation model represents a new direction in the deep integration of remote sensing technology and artificial intelligence. As satellite data volumes continue to grow exponentially, efficiently extracting actionable insights from vast image archives will be a central challenge for the geospatial industry — and Tessera's technical approach may well be part of the answer.

原文來源： 查看原文