The TACO Specification
A FAIR-compliant, cloud-native specification that defines a formal and scalable format for packaging and sharing AI-ready Earth Observation datasets.
Skip to content
Scroll Down
The Structural Bottleneck
There is no standard for AI-ready Earth Observation datasets. The data remains fragmented, siloed, and difficult to unify, holding back the full potential of planetary-scale insight.
Ad Hoc File Structures
Data producers adopt inconsistent organizational patterns, making datasets difficult to interpret, compare, and integrate across different projects.
Loosely Defined Formats
Without standardized schemas, data interpretation becomes subjective and error-prone, hindering automated processing.
Inconsistent Semantic Encodings
Varying metadata standards prevent meaningful data integration and limit the development of interoperable AI tools.
Limited Cloud Optimization
Most datasets lack efficient partial read support and cloud-native access patterns, resulting in poor performance and high costs.
Scroll Down
Introducing TACO
Built on GDAL & Apache Parquet • Multi-language Support • Cloud-Optimized • FAIR-compliant
TACO is a unified specification that brings structure, clarity, and interoperability to AI-ready Earth Observation data. To achieve this, TACO defines a complete, interoperable specification:
Formal Data Model
A structured data model built on SAMPLEs 🌽, TORTILLA 🫓, and rich, standardized metadata schemas, all coming together to generate a TACO 🌮.
- Self-contained
SAMPLEunits - Hierarchical
TORTILLAcontainers - Standardized metadata conventions
Binary File Format
Efficient binary serialization using Apache Parquet metadata and GDAL VFS for fast, scalable access to geospatial data.
- Partial read optimization
- Cloud-native access patterns
- Self-contained & portable
Unified API
Consistent interfaces across Python, R, and Julia for creating and accessing datasets.
- Toolbox for dataset creation
- Reader for efficient access
- DataFrame integration
Cloud-Optimized Architecture
Built-in support for major cloud storage platforms (S3, Azure, GCS) with efficient partial reads and minimal HTTP requests, reducing both latency and operational costs.
Foundation Model Ready
Designed specifically for the era of foundation models, enabling seamless combination of diverse datasets for effective training and performance evaluation.
Scroll Down
Why Choose TACO?
Accelerated Development
Reduce time-to-science with standardized, ready-to-use datasets
Data Integrity
Built-in validation and consistency checks ensure data quality
Enhanced Reproducibility
Standardized formats enable reliable scientific reproducibility
Community-Driven
Developed with input from leading EO and AI researchers
Scalable Architecture
Designed to handle datasets from gigabytes to petabytes
Foundation Model Ready
Optimized for training and evaluating large-scale AI models
Scroll Down
Shape the Future of AI-ready EO Datasets
Be part of the movement to standardize AI-ready Earth Observation data. Collaborate, learn, and contribute to tools that accelerate global insight and innovation.
Back to Top