Democratizing planetary-scale analysis: An ultra-lightweight Earth embedding database for accurate and flexible global land monitoring

Abstract. The rapid evolution of satellite-borne Earth Observation (EO) systems has fundamentally revolutionized terrestrial monitoring, yielding comprehensive petabyte-scale archives. However, the immense computational resources and storage volumes required for global-scale analysis often preclude widespread use by many research teams, hindering broader scientific adoption and the execution of planetary-scale studies. To address these barriers, we present the Embedded Seamless Data (ESD), an ultra-lightweight, 30-m global Earth embedding database spanning the 25-year period from 2000 to 2024. By transforming high-dimensional, multi-sensor observations from the Landsat series (5, 7, 8, and 9) and MODIS Terra into information-dense, quantized latent vectors, ESD distils essential geophysical and semantic features into a unified latent space. Utilizing the ESDNet architecture and Finite Scalar Quantization (FSQ), the dataset achieves a transformative ~340-fold reduction in data volume compared to raw daily archives. This compression allows the entire global land surface for a single year to be encapsulated within approximately 2.4 TB, enabling decadal-scale global analysis on standard local workstations. Rigorous validation demonstrates that ESD maintains high reconstructive fidelity to the original reflectance values across the spectral dimension, achieving a Mean Absolute Error (MAE) of 0.0130 (averaged over six spectral bands, including Blue, Green, Red, NIR, SWIR1, and SWIR2), a Root Mean Square Error (RMSE) of 0.0179, and a Correlation Coefficient (CC) of 0.8543. By condensing the annual phenological cycle into 12 temporal latent steps, the embeddings provide inherent denoising effects and a semantically organized latent space that outperforms raw reflectance data in downstream land-cover classification tasks, achieving a comparable and even higher overall accuracy of 79.74 % than the 76.92 % obtained using raw sensor fusion data on globally distributed land cover sample sets. With robust few-shot learning capabilities and longitudinal consistency across 25 years, the ESD product provides a versatile foundation for democratizing planetary-scale Earth system research and advancing next-generation geospatial artificial intelligence.