GraphComp: Extreme Error-bounded Compression of Scientific Data via Temporal Graph Autoencoders

TL;DR

GraphComp addresses the challenge of compressing voluminous scientific data under strict per-point error bounds by transforming grid data into irregular temporal graphs and learning compact representations with a temporal graph autoencoder. The method combines Felzenszwalb-based segmentation, meta-timestamp selection, graph convolutional encoding, and a latent temporal model, followed by a grid-reconstruction step and an error-bounded residual correction to guarantee $\epsilon$-bounded decompression. Empirical results show that GraphComp achieves substantially higher compression ratios than state-of-the-art error-bounded compressors (by 22%–50% on real and synthetic data) while maintaining competitive speed, with a demonstrated transferability of learned models and a WGAN-based synthetic data generator to support reproducibility. The work advances graph-based representations for scientific data and provides practical open-source tooling for researchers needing high-ratio compression with guaranteed accuracy.

Abstract

The generation of voluminous scientific data poses significant challenges for efficient storage, transfer, and analysis. Recently, error-bounded lossy compression methods emerged due to their ability to achieve high compression ratios while controlling data distortion. However, they often overlook the inherent spatial and temporal correlations within scientific data, thus missing opportunities for higher compression. In this paper we propose GRAPHCOMP, a novel graph-based method for error-bounded lossy compression of scientific data. We perform irregular segmentation of the original grid data and generate a graph representation that preserves the spatial and temporal correlations. Inspired by Graph Neural Networks (GNNs), we then propose a temporal graph autoencoder to learn latent representations that significantly reduce the size of the graph, effectively compressing the original data. Decompression reverses the process and utilizes the learnt graph model together with the latent representation to reconstruct an approximation of the original data. The decompressed data are guaranteed to satisfy a user-defined point-wise error bound. We compare our method against the state-of-the-art error-bounded lossy methods (i.e., HPEZ, SZ3.1, SPERR, and ZFP) on large-scale real and synthetic data. GRAPHCOMP consistently achieves the highest compression ratio across most datasets, outperforming the second-best method by margins ranging from 22% to 50%.

Figures (19)

• Figure 1: The RedSea dataset: a visualization of Red Sea reanalysis temperature data hoteit-RSRA2018 covering Eastern Africa, the Arabian Peninsula, the Indian Ocean and neighboring regions.
• Figure 2: Overview of GraphComp, which includes three primary steps: ① graph initialization, ② graph representation learning, and ③ grid reconstruction. To guarantee a point-wise error bound, we further integrate an error-bounded module.
• Figure 3: Overview of our temporal graph autoencoder T-AutoG for latent representation, which learns spatial and temporal patterns; the decoder is omitted.
• Figure 4: The overview of grid reconstruction from latent representation to reconstructed grid data.
• Figure 5: Compression ratios for the RedSea500 dataset for different $R_{\max}$ values.

Theorems & Definitions (1)

• Definition 1: Graph set $\mathcal{G}$