Core Technology: The Virtual Satellite

The central innovation of AlphaEarth is its ability to synthesize diverse data streams into a single, cohesive model.

1. Multi-Modal Data Fusion

AlphaEarth integrates data from dozens of disparate public sources, including: * Optical Imagery: Standard visual satellite photos (e.g., Landsat, Sentinel-2). * Synthetic Aperture Radar (SAR): Radar data that can "see" through clouds and smoke (e.g., Sentinel-1). * 3D Laser Mapping: LiDAR data for elevation and structure. * Climate Simulations: Meteorological data layers.

2. Geospatial Embeddings

Instead of storing raw pixels (which are heavy and often contain "noise" like clouds or shadows), AlphaEarth converts this data into Embeddings. * Concept: An embedding is a compact numerical representation (a vector of 64 numbers) that mathematically describes the characteristics of a specific location. * Resolution: The model generates these embeddings for every 10 x 10 meter square of the Earth's terrestrial land and coastal waters. * Efficiency: These embeddings require approximately 16 times less storage space than traditional satellite imagery while retaining critical semantic information. This allows computer systems to process global-scale analysis in a fraction of the time usually required.

Products and Services

Satellite Embedding Dataset

The primary commercial and scientific product offered by AlphaEarth Foundations is the Satellite Embedding Dataset. This is one of the largest Earth observation datasets of its kind, hosted on Google Earth Engine. * Coverage: The dataset provides annual embedding layers from 2017 to 2024, covering the entire globe. * Availability: It contains over 1.4 trillion embedding footprints per year. * Analysis-Ready: Unlike raw satellite images that require extensive pre-processing (cloud removal, color correction, atmospheric adjustment), these embeddings are "analysis-ready." Users can feed them directly into machine learning models to perform tasks like classification, regression, or change detection.

Geospatial AI Tools for Enterprise

AlphaEarth technology powers a suite of tools available to enterprise clients and researchers through Google Cloud and Earth Engine.

1. Semantic Similarity Search

This feature allows users to "search" the planet using examples rather than coordinates. * How it works: A user can select a specific feature?for example, a illegal gold mine in the Amazon rainforest?and ask the AI to "find all other locations on Earth that look like this." * Application: This is used for discovering unreported deforestation, identifying specific crop types globally, or locating infrastructure patterns.

2. Generative In-Painting and Cloud Removal

Because the "Virtual Satellite" understands the context of the Earth's surface over time, it can effectively "see through" obstacles. * Cloud Penetration: If a region is covered by clouds in optical imagery, AlphaEarth can infer the surface conditions using radar data and historical context, providing a complete map even during monsoon seasons or in persistently cloudy regions like the Congo Basin.

3. Low-Shot Learning Classifiers

The embeddings are designed to be "data-efficient." * Benefit: Users can train highly accurate custom maps with very sparse training data. For example, a conservation group might only need to label a few dozen examples of a specific tree species to map it across an entire continent, whereas traditional methods would require thousands of labels.

Use Cases and Applications

Environmental Monitoring & Conservation

• Deforestation Tracking: The model is used to detect subtle signs of logging or forest degradation before clear-cutting occurs.
• Biodiversity Mapping: Organizations use the embeddings to map habitat fragmentation and prioritize areas for restoration.

Agriculture and Food Security

• Crop Identification: AlphaEarth allows for the rapid identification of crop types (e.g., distinguishing soy from corn) at a global scale, helping governments and NGOs forecast food supplies.
• Yield Prediction: By analyzing historical embeddings, the system can assess crop health and predict potential yield shortages due to drought or pests.

Urban Planning and Disaster Response

• Urban Expansion: The system tracks the growth of cities and informal settlements (slums) with high precision, aiding in infrastructure planning.
• Disaster Assessment: In the wake of floods or fires, the model helps rapid-response teams identify damaged infrastructure by comparing pre-disaster embeddings with post-disaster data.

Strategic Partnerships

AlphaEarth Foundations collaborates with major global entities to validate and deploy its data: * MapBiomas: Utilizing the dataset to track land use changes and deforestation across Brazil and the Amazon. * United Nations Food and Agriculture Organization (FAO): Leveraging the technology for global food security monitoring. * Harvard Forest: Using the data for ecological research and forest dynamics. * Google Earth Engine Partner Network: A wide ecosystem of commercial and academic partners who build custom applications on top of the AlphaEarth infrastructure.