Google DeepMind has launched AlphaEarth Foundations, an AI model designed to produce detailed maps of Earth using large amounts of geospatial data.

The system combines trillions of images from multiple public sources, including satellite images, radar scans, 3D mapping, and climate simulations.

AlphaEarth Foundations can map terrestrial land and coastal waters worldwide, aiming to provide scientists with improved tools to analyze environmental changes.

Google said the model can generate data for areas as small as 10 square metres, and requires less storage space than similar systems.

Initial testing, covering data from 2017 to 2024, showed that AlphaEarth Foundations had a 24% lower average error rate in identifying land use and surface properties, compared to other AI models, according to a DeepMind paper.

The company said the tool could help researchers monitor crop health, deforestation, urban growth, and water resources.

🔗 Source: Euronews

🧠 Food for thought

1️⃣ Multi-source satellite integration solves a decades-old data fragmentation problem

AlphaEarth Foundations addresses a challenge that has persisted since the early days of Earth observation satellites.

When TIROS-1 launched in 1960 as the first successful meteorological satellite, it marked the beginning of space-based Earth monitoring 1. However, over 60 years later, scientists still struggle with data fragmentation from the more than 150 Earth-observing satellites currently in orbit 1.

These satellites collectively generate over 10 terabytes of data daily, but combining information from different sources with varying sensors, resolutions, and timing remains technically difficult 1.

Google’s approach of using AI to synthesize “trillions of images from dozens of public sources” represents a significant leap from the traditional method of relying on individual satellite passes. The Landsat program, which began in 1972, established the foundation for systematic Earth observation, but even today’s satellites face the limitation of capturing specific locations only when their orbital path aligns 2.

AlphaEarth’s “virtual satellite” concept effectively addresses the constraint of waiting for the right satellite to pass overhead, potentially transforming how researchers access and utilize Earth observation data.

2️⃣ Energy-intensive AI tackles environmental monitoring despite its own carbon footprint

AlphaEarth Foundations highlights the complex relationship between AI’s environmental benefits and costs.