Tessera AI Model Promises Enhanced Accessibility to Earth Observation Data

On 23 October 2023, the Tessera AI model launched, aiming to democratize access to Earth observation data. Developed by a consortium including the European Space Agency (ESA), Tessera converts raw satellite imagery into analyzable data without requiring extensive technical expertise.

Earth observation systems generate terabytes of data daily, mainly through ESA's Copernicus programme. Much of this data requires specialized knowledge to interpret. Tessera integrates multimodal AI architectures similar to CLIP to connect raw imagery with actionable insights. "Tessera is structured to allow users to query satellite datasets using natural language prompts," explained Dr. Amira Vetland, a senior researcher at the ESA AI Laboratory.

The model is trained on high-resolution satellite data from Sentinel-2 and Landsat, combined with climate models and socioeconomic indices. It generates automated reports on applications like land use changes and atmospheric pollution. A demonstration earlier this month showed Tessera identifying deforestation in the Amazon over three years with 89% accuracy, comparable to manual GIS workflows.

Developers emphasize that Tessera is an augmentative tool. "Tessera is not attempting to replace fine-grained analyses performed by remote sensing specialists," said Vetland. "Instead, it aims to expand the accessibility of these datasets to address global challenges like climate change and food security."

The practical implications are significant. Environmental NGOs and local governments could use Tessera to monitor compliance with policies or assess disaster impacts. In Kenya, a pilot programme used the AI model to estimate water reservoir levels, aiding planning in drought-prone areas.

However, concerns about the reproducibility and transparency of Tessera’s results persist. AI-generated insights often face scrutiny over the "black box" problem, where decision-making processes remain unclear. Vetland noted that developers included interpretability features to log predictions, but independent audits and peer-reviewed validations are vital for adoption.

The project has received substantial funding, with €12 million ($12.6 million USD) allocated through the Horizon Europe programme. ESA sees Tessera as part of its larger push toward an open science framework for satellite data.

Experts outside the consortium express cautious optimism. Dr. Mikko Haapala from the Finnish Meteorological Institute called Tessera’s approach "ambitious but needed," adding, "If such models can reliably deliver insights at scale, they can fundamentally transform how we integrate environmental monitoring into policy." He cautioned, "The scalability of machine learning approaches in global contexts is not yet a solved problem. Data sparsity and biases in training datasets could limit the model’s applicability."

The development of Tessera reflects a growing recognition of the need for accessible Earth observation tools to combat climate change. Large-scale AI models adapted for environmental contexts are on the rise. Tessera’s developers argue that its open-access approach and focus on public sector use cases provide a distinct advantage.

Currently, Tessera is in beta testing, with plans for general availability by mid-2024. The next few months will be crucial for assessing its utility among users. Whether the AI model can empower researchers and policymakers remains uncertain, but its launch signifies progress in leveraging AI to tackle environmental challenges.