TerraCodec: Compressing Earth Observations

2025-10-14

AI Summary

To address the challenges of large-scale Earth observation multitemporal multispectral imagery and the lack of unified, pretrained neural compression models, this paper proposes TerraCodec—a dedicated neural compression framework for remote sensing data. It comprises an efficient static encoder tailored to multispectral characteristics and TEC-TT, a temporal modeling architecture built upon a Temporal Transformer. We introduce Latent Repacking—the first unsupervised, multi-rate adaptive training technique for Transformer-based compression. Notably, TEC-TT achieves zero-shot cloud restoration, attaining state-of-the-art performance on the AllClear benchmark. Evaluated on Sentinel-2 data, TerraCodec improves compression efficiency by 3–10× over conventional codecs, significantly reducing storage and transmission overheads at equivalent reconstruction quality.

Technology Category

• Computer Vision: Remote Sensing / Geospatial AI
• Machine Learning: Learning on the Edge & Model Compression
• Data Mining & Knowledge Management: Data Compression

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs

Abstract

Earth observation (EO) satellites produce massive streams of multispectral image time series, posing pressing challenges for storage and transmission. Yet, learned EO compression remains fragmented, lacking publicly available pretrained models and misaligned with advances in compression for natural imagery. Image codecs overlook temporal redundancy, while video codecs rely on motion priors that fail to capture the radiometric evolution of largely static scenes. We introduce TerraCodec (TEC), a family of learned codecs tailored to EO. TEC includes efficient image-based variants adapted to multispectral inputs, as well as a Temporal Transformer model (TEC-TT) that leverages dependencies across time. To overcome the fixed-rate setting of today's neural codecs, we present Latent Repacking, a novel method for training flexible-rate transformer models that operate on varying rate-distortion settings. Trained on Sentinel-2 data, TerraCodec outperforms classical codecs, achieving 3-10x stronger compression at equivalent image quality. Beyond compression, TEC-TT enables zero-shot cloud inpainting, surpassing state-of-the-art methods on the AllClear benchmark. Our results establish bespoke, learned compression algorithms as a promising direction for Earth observation. Code and model weights will be released under a permissive license.

Problem

• Compressing massive Earth observation satellite image streams
• Overcoming temporal redundancy limitations in existing codecs
• Enabling flexible-rate compression for multispectral time series

Innovation

• Learned codecs tailored to Earth observation data
• Temporal Transformer model leveraging time dependencies
• Latent Repacking method for flexible-rate transformer training