1. Introduction
Earth observation data volume is rapidly expanding, posing significant challenges for storage, transmission, and processing, necessitating advanced compression techniques. Existing general-purpose compressors often fail to adequately preserve the unique characteristics and scientific value of these specialized datasets. This paper addresses the critical need for an efficient and domain-aware compression solution for Earth observation imagery and geophysical products. Models used in this article include JPEG 2000, WebP, BPG, and proposed TerraCodec models.
2. Related Work
Traditional image and scientific data compression methods, such as JPEG 2000 and H.265/HEVC, offer acceptable performance for general-purpose imagery but struggle with the multi-spectral, multi-temporal, and high-dynamic-range nature of EO data. Recent advancements in learned compression, often employing autoencoders and neural networks, have shown promise but typically require extensive training on specific data distributions. This section reviews these methods, highlighting their strengths and limitations in the context of Earth observation.
3. Methodology
TerraCodec employs a multi-stage compression pipeline that integrates adaptive spatial-spectral decorrelation with a learned entropy coding module. Initially, a custom wavelet transform is applied to exploit the inherent redundancies across spectral bands and spatial dimensions within EO data. Subsequently, the transformed coefficients are quantized and encoded using a conditional neural network that predicts optimal bit allocation and entropy parameters. This approach allows TerraCodec to adapt its compression strategy based on the specific characteristics of the input Earth observation data.
4. Experimental Results
Our experiments demonstrate that TerraCodec consistently outperforms state-of-the-art compression algorithms across various Earth observation datasets, achieving higher compression ratios at comparable or superior reconstruction quality. Performance was evaluated using metrics such as Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM), alongside visual inspection by domain experts. The table below presents a summary of average performance across a benchmark dataset, showcasing TerraCodec's efficiency.
| Codec | Compression Ratio | PSNR (dB) | SSIM |
|---|---|---|---|
| JPEG 2000 | 12:1 | 35.2 | 0.915 |
| WebP | 15:1 | 36.8 | 0.928 |
| TerraCodec (Proposed) | 28:1 | 37.5 | 0.941 |
5. Discussion
The superior compression performance of TerraCodec has significant implications for the future of Earth observation data management, enabling faster data dissemination and reduced storage costs without compromising scientific utility. While current results are highly promising, further research will focus on extending TerraCodec to handle dynamic, time-series Earth observation products and exploring hardware acceleration for real-time applications. This framework represents a substantial step forward in addressing the grand challenges of big data in remote sensing.